diff options
| author | Błażej Szczygieł <spaz16@wp.pl> | 2016-09-14 22:10:03 +0200 |
|---|---|---|
| committer | Błażej Szczygieł <spaz16@wp.pl> | 2016-10-19 13:34:28 +0200 |
| commit | 5268443fdfd6f9f8172cede1140810ae21f7990a (patch) | |
| tree | 1bfb077f7557a4ee18734c8bb38ec0ddc22c8418 | |
| parent | 2d77a6f5d3beae3b341e4a7f331202bd1a010508 (diff) | |
Add libvpx thirdparty library
Only necessary files
275 files changed, 83150 insertions, 0 deletions
diff --git a/thirdparty/README.md b/thirdparty/README.md index 1de2949d44..0b0917c1e9 100644 --- a/thirdparty/README.md +++ b/thirdparty/README.md @@ -114,6 +114,13 @@ Files extracted from upstream source: - COPYING +## libvpx + +- Upstream: http://www.webmproject.org/code/ +- Version: 1.6.0 +- License: BSD-3-Clause + + ## libwebp - Upstream: https://chromium.googlesource.com/webm/libwebp/ diff --git a/thirdparty/libvpx/AUTHORS b/thirdparty/libvpx/AUTHORS new file mode 100644 index 0000000000..fcd5c534a8 --- /dev/null +++ b/thirdparty/libvpx/AUTHORS @@ -0,0 +1,142 @@ +# This file is automatically generated from the git commit history +# by tools/gen_authors.sh. + +Aaron Watry <awatry@gmail.com> +Abo Talib Mahfoodh <ab.mahfoodh@gmail.com> +Adam Xu <adam@xuyaowu.com> +Adrian Grange <agrange@google.com> +Aℓex Converse <aconverse@google.com> +Ahmad Sharif <asharif@google.com> +Alexander Voronov <avoronov@graphics.cs.msu.ru> +Alexis Ballier <aballier@gentoo.org> +Alok Ahuja <waveletcoeff@gmail.com> +Alpha Lam <hclam@google.com> +A.Mahfoodh <ab.mahfoodh@gmail.com> +Ami Fischman <fischman@chromium.org> +Andoni Morales Alastruey <ylatuya@gmail.com> +Andres Mejia <mcitadel@gmail.com> +Andrew Russell <anrussell@google.com> +Angie Chiang <angiebird@google.com> +Aron Rosenberg <arosenberg@logitech.com> +Attila Nagy <attilanagy@google.com> +Brion Vibber <bvibber@wikimedia.org> +changjun.yang <changjun.yang@intel.com> +Charles 'Buck' Krasic <ckrasic@google.com> +chm <chm@rock-chips.com> +Christian Duvivier <cduvivier@google.com> +Daniele Castagna <dcastagna@chromium.org> +Daniel Kang <ddkang@google.com> +Deb Mukherjee <debargha@google.com> +Dim Temp <dimtemp0@gmail.com> +Dmitry Kovalev <dkovalev@google.com> +Dragan Mrdjan <dmrdjan@mips.com> +Ed Baker <edward.baker@intel.com> +Ehsan Akhgari <ehsan.akhgari@gmail.com> +Erik Niemeyer <erik.a.niemeyer@intel.com> +Fabio Pedretti <fabio.ped@libero.it> +Frank Galligan <fgalligan@google.com> +Fredrik Söderquist <fs@opera.com> +Fritz Koenig <frkoenig@google.com> +Gaute Strokkenes <gaute.strokkenes@broadcom.com> +Geza Lore <gezalore@gmail.com> +Ghislain MARY <ghislainmary2@gmail.com> +Giuseppe Scrivano <gscrivano@gnu.org> +Gordana Cmiljanovic <gordana.cmiljanovic@imgtec.com> +Guillaume Martres <gmartres@google.com> +Guillermo Ballester Valor <gbvalor@gmail.com> +Hangyu Kuang <hkuang@google.com> +Hanno Böck <hanno@hboeck.de> +Henrik Lundin <hlundin@google.com> +Hui Su <huisu@google.com> +Ivan Maltz <ivanmaltz@google.com> +Jacek Caban <cjacek@gmail.com> +Jacky Chen <jackychen@google.com> +James Berry <jamesberry@google.com> +James Yu <james.yu@linaro.org> +James Zern <jzern@google.com> +Jan Gerber <j@mailb.org> +Jan Kratochvil <jan.kratochvil@redhat.com> +Janne Salonen <jsalonen@google.com> +Jean-Yves Avenard <jyavenard@mozilla.com> +Jeff Faust <jfaust@google.com> +Jeff Muizelaar <jmuizelaar@mozilla.com> +Jeff Petkau <jpet@chromium.org> +Jia Jia <jia.jia@linaro.org> +Jian Zhou <zhoujian@google.com> +Jim Bankoski <jimbankoski@google.com> +Jingning Han <jingning@google.com> +Joey Parrish <joeyparrish@google.com> +Johann Koenig <johannkoenig@google.com> +John Koleszar <jkoleszar@google.com> +Johnny Klonaris <google@jawknee.com> +John Stark <jhnstrk@gmail.com> +Joshua Bleecher Snyder <josh@treelinelabs.com> +Joshua Litt <joshualitt@google.com> +Julia Robson <juliamrobson@gmail.com> +Justin Clift <justin@salasaga.org> +Justin Lebar <justin.lebar@gmail.com> +KO Myung-Hun <komh@chollian.net> +Lawrence Velázquez <larryv@macports.org> +Linfeng Zhang <linfengz@google.com> +Lou Quillio <louquillio@google.com> +Luca Barbato <lu_zero@gentoo.org> +Makoto Kato <makoto.kt@gmail.com> +Mans Rullgard <mans@mansr.com> +Marco Paniconi <marpan@google.com> +Mark Mentovai <mark@chromium.org> +Martin Ettl <ettl.martin78@googlemail.com> +Martin Storsjo <martin@martin.st> +Matthew Heaney <matthewjheaney@chromium.org> +Michael Kohler <michaelkohler@live.com> +Mike Frysinger <vapier@chromium.org> +Mike Hommey <mhommey@mozilla.com> +Mikhal Shemer <mikhal@google.com> +Minghai Shang <minghai@google.com> +Morton Jonuschat <yabawock@gmail.com> +Nico Weber <thakis@chromium.org> +Parag Salasakar <img.mips1@gmail.com> +Pascal Massimino <pascal.massimino@gmail.com> +Patrik Westin <patrik.westin@gmail.com> +Paul Wilkins <paulwilkins@google.com> +Pavol Rusnak <stick@gk2.sk> +Paweł Hajdan <phajdan@google.com> +Pengchong Jin <pengchong@google.com> +Peter de Rivaz <peter.derivaz@gmail.com> +Philip Jägenstedt <philipj@opera.com> +Priit Laes <plaes@plaes.org> +Rafael Ávila de Espíndola <rafael.espindola@gmail.com> +Rafaël Carré <funman@videolan.org> +Ralph Giles <giles@xiph.org> +Rob Bradford <rob@linux.intel.com> +Ronald S. Bultje <rsbultje@gmail.com> +Rui Ueyama <ruiu@google.com> +Sami Pietilä <samipietila@google.com> +Sasi Inguva <isasi@google.com> +Scott Graham <scottmg@chromium.org> +Scott LaVarnway <slavarnway@google.com> +Sean McGovern <gseanmcg@gmail.com> +Sergey Kolomenkin <kolomenkin@gmail.com> +Sergey Ulanov <sergeyu@chromium.org> +Shimon Doodkin <helpmepro1@gmail.com> +Shunyao Li <shunyaoli@google.com> +Stefan Holmer <holmer@google.com> +Suman Sunkara <sunkaras@google.com> +Taekhyun Kim <takim@nvidia.com> +Takanori MATSUURA <t.matsuu@gmail.com> +Tamar Levy <tamar.levy@intel.com> +Tao Bai <michaelbai@chromium.org> +Tero Rintaluoma <teror@google.com> +Thijs Vermeir <thijsvermeir@gmail.com> +Tim Kopp <tkopp@google.com> +Timothy B. Terriberry <tterribe@xiph.org> +Tom Finegan <tomfinegan@google.com> +Vignesh Venkatasubramanian <vigneshv@google.com> +Yaowu Xu <yaowu@google.com> +Yi Luo <luoyi@google.com> +Yongzhe Wang <yongzhe@google.com> +Yunqing Wang <yunqingwang@google.com> +Yury Gitman <yuryg@google.com> +Zoe Liu <zoeliu@google.com> +Google Inc. +The Mozilla Foundation +The Xiph.Org Foundation diff --git a/thirdparty/libvpx/CHANGELOG b/thirdparty/libvpx/CHANGELOG new file mode 100644 index 0000000000..795d395f96 --- /dev/null +++ b/thirdparty/libvpx/CHANGELOG @@ -0,0 +1,654 @@ +2016-07-20 v1.6.0 "Khaki Campbell Duck" + This release improves upon the VP9 encoder and speeds up the encoding and + decoding processes. + + - Upgrading: + This release is ABI incompatible with 1.5.0 due to a new 'color_range' enum + in vpx_image and some minor changes to the VP8_COMP structure. + + The default key frame interval for VP9 has changed from 128 to 9999. + + - Enhancement: + A core focus has been performance for low end Intel processors. SSSE3 + instructions such as 'pshufb' have been avoided and instructions have been + reordered to better accommodate the more constrained pipelines. + + As a result, devices based on Celeron processors have seen substantial + decoding improvements. From Indian Runner Duck to Javan Whistling Duck, + decoding speed improved between 10 and 30%. Between Javan Whistling Duck + and Khaki Campbell Duck, it improved another 10 to 15%. + + While Celeron benefited most, Core-i5 also improved 5% and 10% between the + respective releases. + + Realtime performance for WebRTC for both speed and quality has received a + lot of attention. + + - Bug Fixes: + A number of fuzzing issues, found variously by Mozilla, Chromium and others, + have been fixed and we strongly recommend updating. + +2015-11-09 v1.5.0 "Javan Whistling Duck" + This release improves upon the VP9 encoder and speeds up the encoding and + decoding processes. + + - Upgrading: + This release is ABI incompatible with 1.4.0. It drops deprecated VP8 + controls and adds a variety of VP9 controls for testing. + + The vpxenc utility now prefers VP9 by default. + + - Enhancements: + Faster VP9 encoding and decoding + Smaller library size by combining functions used by VP8 and VP9 + + - Bug Fixes: + A variety of fuzzing issues + +2015-04-03 v1.4.0 "Indian Runner Duck" + This release includes significant improvements to the VP9 codec. + + - Upgrading: + This release is ABI incompatible with 1.3.0. It drops the compatibility + layer, requiring VPX_IMG_FMT_* instead of IMG_FMT_*, and adds several codec + controls for VP9. + + - Enhancements: + Faster VP9 encoding and decoding + Multithreaded VP9 decoding (tile and frame-based) + Multithreaded VP9 encoding - on by default + YUV 4:2:2 and 4:4:4 support in VP9 + 10 and 12bit support in VP9 + 64bit ARM support by replacing ARM assembly with intrinsics + + - Bug Fixes: + Fixes a VP9 bitstream issue in Profile 1. This only affected non-YUV 4:2:0 + files. + + - Known Issues: + Frame Parallel decoding fails for segmented and non-420 files. + +2013-11-15 v1.3.0 "Forest" + This release introduces the VP9 codec in a backward-compatible way. + All existing users of VP8 can continue to use the library without + modification. However, some VP8 options do not map to VP9 in the same manner. + + The VP9 encoder in this release is not feature complete. Users interested in + the encoder are advised to use the git master branch and discuss issues on + libvpx mailing lists. + + - Upgrading: + This release is ABI and API compatible with Duclair (v1.0.0). Users + of older releases should refer to the Upgrading notes in this document + for that release. + + - Enhancements: + Get rid of bashisms in the main build scripts + Added usage info on command line options + Add lossless compression mode + Dll build of libvpx + Add additional Mac OS X targets: 10.7, 10.8 and 10.9 (darwin11-13) + Add option to disable documentation + configure: add --enable-external-build support + make: support V=1 as short form of verbose=yes + configure: support mingw-w64 + configure: support hardfloat armv7 CHOSTS + configure: add support for android x86 + Add estimated completion time to vpxenc + Don't exit on decode errors in vpxenc + vpxenc: support scaling prior to encoding + vpxdec: support scaling output + vpxenc: improve progress indicators with --skip + msvs: Don't link to winmm.lib + Add a new script for producing vcxproj files + Produce Visual Studio 10 and 11 project files + Produce Windows Phone project files + msvs-build: use msbuild for vs >= 2005 + configure: default configure log to config.log + Add encoding option --static-thresh + + - Speed: + Miscellaneous speed optimizations for VP8 and VP9. + + - Quality: + In general, quality is consistent with the Eider release. + + - Bug Fixes: + This release represents approximately a year of engineering effort, + and contains multiple bug fixes. Please refer to git history for details. + + +2012-12-21 v1.2.0 + This release acts as a checkpoint for a large amount of internal refactoring + and testing. It also contains a number of small bugfixes, so all users are + encouraged to upgrade. + + - Upgrading: + This release is ABI and API compatible with Duclair (v1.0.0). Users + of older releases should refer to the Upgrading notes in this + document for that release. + + - Enhancements: + VP8 optimizations for MIPS dspr2 + vpxenc: add -quiet option + + - Speed: + Encoder and decoder speed is consistent with the Eider release. + + - Quality: + In general, quality is consistent with the Eider release. + + Minor tweaks to ARNR filtering + Minor improvements to real time encoding with multiple temporal layers + + - Bug Fixes: + Fixes multithreaded encoder race condition in loopfilter + Fixes multi-resolution threaded encoding + Fix potential encoder dead-lock after picture resize + + +2012-05-09 v1.1.0 "Eider" + This introduces a number of enhancements, mostly focused on real-time + encoding. In addition, it fixes a decoder bug (first introduced in + Duclair) so all users of that release are encouraged to upgrade. + + - Upgrading: + This release is ABI and API compatible with Duclair (v1.0.0). Users + of older releases should refer to the Upgrading notes in this + document for that release. + + This release introduces a new temporal denoiser, controlled by the + VP8E_SET_NOISE_SENSITIVITY control. The temporal denoiser does not + currently take a strength parameter, so the control is effectively + a boolean - zero (off) or non-zero (on). For compatibility with + existing applications, the values accepted are the same as those + for the spatial denoiser (0-6). The temporal denoiser is enabled + by default, and the older spatial denoiser may be restored by + configuring with --disable-temporal-denoising. The temporal denoiser + is more computationally intensive than the spatial one. + + This release removes support for a legacy, decode only API that was + supported, but deprecated, at the initial release of libvpx + (v0.9.0). This is not expected to have any impact. If you are + impacted, you can apply a reversion to commit 2bf8fb58 locally. + Please update to the latest libvpx API if you are affected. + + - Enhancements: + Adds a motion compensated temporal denoiser to the encoder, which + gives higher quality than the older spatial denoiser. (See above + for notes on upgrading). + + In addition, support for new compilers and platforms were added, + including: + improved support for XCode + Android x86 NDK build + OS/2 support + SunCC support + + Changing resolution with vpx_codec_enc_config_set() is now + supported. Previously, reinitializing the codec was required to + change the input resolution. + + The vpxenc application has initial support for producing multiple + encodes from the same input in one call. Resizing is not yet + supported, but varying other codec parameters is. Use -- to + delineate output streams. Options persist from one stream to the + next. + + Also, the vpxenc application will now use a keyframe interval of + 5 seconds by default. Use the --kf-max-dist option to override. + + - Speed: + Decoder performance improved 2.5% versus Duclair. Encoder speed is + consistent with Duclair for most material. Two pass encoding of + slideshow-like material will see significant improvements. + + Large realtime encoding speed gains at a small quality expense are + possible by configuring the on-the-fly bitpacking experiment with + --enable-onthefly-bitpacking. Realtime encoder can be up to 13% + faster (ARM) depending on the number of threads and bitrate + settings. This technique sees constant gain over the 5-16 speed + range. For VC style input the loss seen is up to 0.2dB. See commit + 52cf4dca for further details. + + - Quality: + On the whole, quality is consistent with the Duclair release. Some + tweaks: + + Reduced blockiness in easy sections by applying a penalty to + intra modes. + + Improved quality of static sections (like slideshows) with + two pass encoding. + + Improved keyframe sizing with multiple temporal layers + + - Bug Fixes: + Corrected alt-ref contribution to frame rate for visible updates + to the alt-ref buffer. This affected applications making manual + usage of the frame reference flags, or temporal layers. + + Additional constraints were added to disable multi-frame quality + enhancement (MFQE) in sections of the frame where there is motion. + (#392) + + Fixed corruption issues when vpx_codec_enc_config_set() was called + with spatial resampling enabled. + + Fixed a decoder error introduced in Duclair where the segmentation + map was not being reinitialized on keyframes (#378) + + +2012-01-27 v1.0.0 "Duclair" + Our fourth named release, focused on performance and features related to + real-time encoding. It also fixes a decoder crash bug introduced in + v0.9.7, so all users of that release are encouraged to upgrade. + + - Upgrading: + This release is ABI incompatible with prior releases of libvpx, so the + "major" version number has been bumped to 1. You must recompile your + applications against the latest version of the libvpx headers. The + API remains compatible, and this should not require code changes in most + applications. + + - Enhancements: + This release introduces several substantial new features to the encoder, + of particular interest to real time streaming applications. + + Temporal scalability allows the encoder to produce a stream that can + be decimated to different frame rates, with independent rate targetting + for each substream. + + Multiframe quality enhancement postprocessing can make visual quality + more consistent in the presence of frames that are substantially + different quality than the surrounding frames, as in the temporal + scalability case and in some forced keyframe scenarios. + + Multiple-resolution encoding support allows the encoding of the + same content at different resolutions faster than encoding them + separately. + + - Speed: + Optimization targets for this release included the decoder and the real- + time modes of the encoder. Decoder speed on x86 has improved 10.5% with + this release. Encoder improvements followed a curve where speeds 1-3 + improved 4.0%-1.5%, speeds 4-8 improved <1%, and speeds 9-16 improved + 1.5% to 10.5%, respectively. "Best" mode speed is consistent with the + Cayuga release. + + - Quality: + Encoder quality in the single stream case is consistent with the Cayuga + release. + + - Bug Fixes: + This release fixes an OOB read decoder crash bug present in v0.9.7 + related to the clamping of motion vectors in SPLITMV blocks. This + behavior could be triggered by corrupt input or by starting + decoding from a P-frame. + + +2011-08-15 v0.9.7-p1 "Cayuga" patch 1 + This is an incremental bugfix release against Cayuga. All users of that + release are strongly encouraged to upgrade. + + - Fix potential OOB reads (cdae03a) + + An unbounded out of bounds read was discovered when the + decoder was requested to perform error concealment (new in + Cayuga) given a frame with corrupt partition sizes. + + A bounded out of bounds read was discovered affecting all + versions of libvpx. Given an multipartition input frame that + is truncated between the mode/mv partition and the first + residiual paritition (in the block of partition offsets), up + to 3 extra bytes could have been read from the source buffer. + The code will not take any action regardless of the contents + of these undefined bytes, as the truncated buffer is detected + immediately following the read based on the calculated + starting position of the coefficient partition. + + - Fix potential error concealment crash when the very first frame + is missing or corrupt (a609be5) + + - Fix significant artifacts in error concealment (a4c2211, 99d870a) + + - Revert 1-pass CBR rate control changes (e961317) + Further testing showed this change produced undesirable visual + artifacts, rolling back for now. + + +2011-08-02 v0.9.7 "Cayuga" + Our third named release, focused on a faster, higher quality, encoder. + + - Upgrading: + This release is backwards compatible with Aylesbury (v0.9.5) and + Bali (v0.9.6). Users of older releases should refer to the Upgrading + notes in this document for that release. + + - Enhancements: + Stereo 3D format support for vpxenc + Runtime detection of available processor cores. + Allow specifying --end-usage by enum name + vpxdec: test for frame corruption + vpxenc: add quantizer histogram display + vpxenc: add rate histogram display + Set VPX_FRAME_IS_DROPPABLE + update configure for ios sdk 4.3 + Avoid text relocations in ARM vp8 decoder + Generate a vpx.pc file for pkg-config. + New ways of passing encoded data between encoder and decoder. + + - Speed: + This release includes across-the-board speed improvements to the + encoder. On x86, these measure at approximately 11.5% in Best mode, + 21.5% in Good mode (speed 0), and 22.5% in Realtime mode (speed 6). + On ARM Cortex A9 with Neon extensions, real-time encoding of video + telephony content is 35% faster than Bali on single core and 48% + faster on multi-core. On the NVidia Tegra2 platform, real time + encoding is 40% faster than Bali. + + Decoder speed was not a priority for this release, but improved + approximately 8.4% on x86. + + Reduce motion vector search on alt-ref frame. + Encoder loopfilter running in its own thread + Reworked loopfilter to precalculate more parameters + SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). + Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3. + Removed redundant checks + Reduced structure sizes + utilize preload in ARMv6 MC/LPF/Copy routines + ARM optimized quantization, dfct, variance, subtract + Increase chrow row alignment to 16 bytes. + disable trellis optimization for first pass + Write SSSE3 sub-pixel filter function + Improve SSE2 half-pixel filter funtions + Add vp8_sub_pixel_variance16x8_ssse3 function + Reduce unnecessary distortion computation + Use diamond search to replace full search + Preload reference area in sub-pixel motion search (real-time mode) + + - Quality: + This release focused primarily on one-pass use cases, including + video conferencing. Low latency data rate control was significantly + improved, improving streamability over bandwidth constrained links. + Added support for error concealment, allowing frames to maintain + visual quality in the presence of substantial packet loss. + + Add rc_max_intra_bitrate_pct control + Limit size of initial keyframe in one-pass. + Improve framerate adaptation + Improved 1-pass CBR rate control + Improved KF insertion after fades to still. + Improved key frame detection. + Improved activity masking (lower PSNR impact for same SSIM boost) + Improved interaction between GF and ARFs + Adding error-concealment to the decoder. + Adding support for independent partitions + Adjusted rate-distortion constants + + + - Bug Fixes: + Removed firstpass motion map + Fix parallel make install + Fix multithreaded encoding for 1 MB wide frame + Fixed iwalsh_neon build problems with RVDS4.1 + Fix semaphore emulation, spin-wait intrinsics on Windows + Fix build with xcode4 and simplify GLOBAL. + Mark ARM asm objects as allowing a non-executable stack. + Fix vpxenc encoding incorrect webm file header on big endian + + +2011-03-07 v0.9.6 "Bali" + Our second named release, focused on a faster, higher quality, encoder. + + - Upgrading: + This release is backwards compatible with Aylesbury (v0.9.5). Users + of older releases should refer to the Upgrading notes in this + document for that release. + + - Enhancements: + vpxenc --psnr shows a summary when encode completes + --tune=ssim option to enable activity masking + improved postproc visualizations for development + updated support for Apple iOS to SDK 4.2 + query decoder to determine which reference frames were updated + implemented error tracking in the decoder + fix pipe support on windows + + - Speed: + Primary focus was on good quality mode, speed 0. Average improvement + on x86 about 40%, up to 100% on user-generated content at that speed. + Best quality mode speed improved 35%, and realtime speed 10-20%. This + release also saw significant improvement in realtime encoding speed + on ARM platforms. + + Improved encoder threading + Dont pick encoder filter level when loopfilter is disabled. + Avoid double copying of key frames into alt and golden buffer + FDCT optimizations. + x86 sse2 temporal filter + SSSE3 version of fast quantizer + vp8_rd_pick_best_mbsegmentation code restructure + Adjusted breakout RD for SPLITMV + Changed segmentation check order + Improved rd_pick_intra4x4block + Adds armv6 optimized variance calculation + ARMv6 optimized sad16x16 + ARMv6 optimized half pixel variance calculations + Full search SAD function optimization in SSE4.1 + Improve MV prediction accuracy to achieve performance gain + Improve MV prediction in vp8_pick_inter_mode() for speed>3 + + - Quality: + Best quality mode improved PSNR 6.3%, and SSIM 6.1%. This release + also includes support for "activity masking," which greatly improves + SSIM at the expense of PSNR. For now, this feature is available with + the --tune=ssim option. Further experimentation in this area + is ongoing. This release also introduces a new rate control mode + called "CQ," which changes the allocation of bits within a clip to + the sections where they will have the most visual impact. + + Tuning for the more exact quantizer. + Relax rate control for last few frames + CQ Mode + Limit key frame quantizer for forced key frames. + KF/GF Pulsing + Add simple version of activity masking. + make rdmult adaptive for intra in quantizer RDO + cap the best quantizer for 2nd order DC + change the threshold of DC check for encode breakout + + - Bug Fixes: + Fix crash on Sparc Solaris. + Fix counter of fixed keyframe distance + ARNR filter pointer update bug fix + Fixed use of motion percentage in KF/GF group calc + Changed condition for using RD in Intra Mode + Fix encoder real-time only configuration. + Fix ARM encoder crash with multiple token partitions + Fixed bug first cluster timecode of webm file is wrong. + Fixed various encoder bugs with odd-sized images + vp8e_get_preview fixed when spatial resampling enabled + quantizer: fix assertion in fast quantizer path + Allocate source buffers to be multiples of 16 + Fix for manual Golden frame frequency + Fix drastic undershoot in long form content + + +2010-10-28 v0.9.5 "Aylesbury" + Our first named release, focused on a faster decoder, and a better encoder. + + - Upgrading: + This release incorporates backwards-incompatible changes to the + ivfenc and ivfdec tools. These tools are now called vpxenc and vpxdec. + + vpxdec + * the -q (quiet) option has been removed, and replaced with + -v (verbose). the output is quiet by default. Use -v to see + the version number of the binary. + + * The default behavior is now to write output to a single file + instead of individual frames. The -y option has been removed. + Y4M output is the default. + + * For raw I420/YV12 output instead of Y4M, the --i420 or --yv12 + options must be specified. + + $ ivfdec -o OUTPUT INPUT + $ vpxdec --i420 -o OUTPUT INPUT + + * If an output file is not specified, the default is to write + Y4M to stdout. This makes piping more natural. + + $ ivfdec -y -o - INPUT | ... + $ vpxdec INPUT | ... + + * The output file has additional flexibility for formatting the + filename. It supports escape characters for constructing a + filename from the width, height, and sequence number. This + replaces the -p option. To get the equivalent: + + $ ivfdec -p frame INPUT + $ vpxdec --i420 -o frame-%wx%h-%4.i420 INPUT + + vpxenc + * The output file must be specified with -o, rather than as the + last argument. + + $ ivfenc <options> INPUT OUTPUT + $ vpxenc <options> -o OUTPUT INPUT + + * The output defaults to webm. To get IVF output, use the --ivf + option. + + $ ivfenc <options> INPUT OUTPUT.ivf + $ vpxenc <options> -o OUTPUT.ivf --ivf INPUT + + + - Enhancements: + ivfenc and ivfdec have been renamed to vpxenc, vpxdec. + vpxdec supports .webm input + vpxdec writes .y4m by default + vpxenc writes .webm output by default + vpxenc --psnr now shows the average/overall PSNR at the end + ARM platforms now support runtime cpu detection + vpxdec visualizations added for motion vectors, block modes, references + vpxdec now silent by default + vpxdec --progress shows frame-by-frame timing information + vpxenc supports the distinction between --fps and --timebase + NASM is now a supported assembler + configure: enable PIC for shared libs by default + configure: add --enable-small + configure: support for ppc32-linux-gcc + configure: support for sparc-solaris-gcc + + - Bugs: + Improve handling of invalid frames + Fix valgrind errors in the NEON loop filters. + Fix loopfilter delta zero transitions + Fix valgrind errors in vp8_sixtap_predict8x4_armv6(). + Build fixes for darwin-icc + + - Speed: + 20-40% (average 28%) improvement in libvpx decoder speed, + including: + Rewrite vp8_short_walsh4x4_sse2() + Optimizations on the loopfilters. + Miscellaneous improvements for Atom + Add 4-tap version of 2nd-pass ARMv6 MC filter. + Improved multithread utilization + Better instruction choices on x86 + reorder data to use wider instructions + Update NEON wide idcts + Make block access to frame buffer sequential + Improved subset block search + Bilinear subpixel optimizations for ssse3. + Decrease memory footprint + + Encoder speed improvements (percentage gain not measured): + Skip unnecessary search of identical frames + Add SSE2 subtract functions + Improve bounds checking in vp8_diamond_search_sadx4() + Added vp8_fast_quantize_b_sse2 + + - Quality: + Over 7% overall PSNR improvement (6.3% SSIM) in "best" quality + encoding mode, and up to 60% improvement on very noisy, still + or slow moving source video + + Motion compensated temporal filter for Alt-Ref Noise Reduction + Improved use of trellis quantization on 2nd order Y blocks + Tune effect of motion on KF/GF boost in two pass + Allow coefficient optimization for good quality speed 0. + Improved control of active min quantizer for two pass. + Enable ARFs for non-lagged compress + +2010-09-02 v0.9.2 + - Enhancements: + Disable frame dropping by default + Improved multithreaded performance + Improved Force Key Frame Behaviour + Increased rate control buffer level precision + Fix bug in 1st pass motion compensation + ivfenc: correct fixed kf interval, --disable-kf + - Speed: + Changed above and left context data layout + Rework idct calling structure. + Removed unnecessary MB_MODE_INFO copies + x86: SSSE3 sixtap prediction + Reworked IDCT to include reconstruction (add) step + Swap alt/gold/new/last frame buffer ptrs instead of copying. + Improve SSE2 loopfilter functions + Change bitreader to use a larger window. + Avoid loopfilter reinitialization when possible + - Quality: + Normalize quantizer's zero bin and rounding factors + Add trellis quantization. + Make the quantizer exact. + Updates to ARNR filtering algorithm + Fix breakout thresh computation for golden & AltRef frames + Redo the forward 4x4 dct + Improve the accuracy of forward walsh-hadamard transform + Further adjustment of RD behaviour with Q and Zbin. + - Build System: + Allow linking of libs built with MinGW to MSVC + Fix target auto-detection on mingw32 + Allow --cpu= to work for x86. + configure: pass original arguments through to make dist + Fix builds without runtime CPU detection + msvs: fix install of codec sources + msvs: Change devenv.com command line for better msys support + msvs: Add vs9 targets. + Add x86_64-linux-icc target + - Bugs: + Potential crashes on older MinGW builds + Fix two-pass framrate for Y4M input. + Fixed simple loop filter, other crashes on ARM v6 + arm: fix missing dependency with --enable-shared + configure: support directories containing .o + Replace pinsrw (SSE) with MMX instructions + apple: include proper mach primatives + Fixed rate control bug with long key frame interval. + Fix DSO link errors on x86-64 when not using a version script + Fixed buffer selection for UV in AltRef filtering + + +2010-06-17 v0.9.1 + - Enhancements: + * ivfenc/ivfdec now support YUV4MPEG2 input and pipe I/O + * Speed optimizations + - Bugfixes: + * Rate control + * Prevent out-of-bounds accesses on invalid data + - Build system updates: + * Detect toolchain to be used automatically for native builds + * Support building shared libraries + * Better autotools emulation (--prefix, --libdir, DESTDIR) + - Updated LICENSE + * http://webmproject.blogspot.com/2010/06/changes-to-webm-open-source-license.html + + +2010-05-18 v0.9.0 + - Initial open source release. Welcome to WebM and VP8! + diff --git a/thirdparty/libvpx/LICENSE b/thirdparty/libvpx/LICENSE new file mode 100644 index 0000000000..1ce44343c4 --- /dev/null +++ b/thirdparty/libvpx/LICENSE @@ -0,0 +1,31 @@ +Copyright (c) 2010, The WebM Project authors. All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are +met: + + * Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + * Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in + the documentation and/or other materials provided with the + distribution. + + * Neither the name of Google, nor the WebM Project, nor the names + of its contributors may be used to endorse or promote products + derived from this software without specific prior written + permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + diff --git a/thirdparty/libvpx/PATENTS b/thirdparty/libvpx/PATENTS new file mode 100644 index 0000000000..caedf607e9 --- /dev/null +++ b/thirdparty/libvpx/PATENTS @@ -0,0 +1,23 @@ +Additional IP Rights Grant (Patents) +------------------------------------ + +"These implementations" means the copyrightable works that implement the WebM +codecs distributed by Google as part of the WebM Project. + +Google hereby grants to you a perpetual, worldwide, non-exclusive, no-charge, +royalty-free, irrevocable (except as stated in this section) patent license to +make, have made, use, offer to sell, sell, import, transfer, and otherwise +run, modify and propagate the contents of these implementations of WebM, where +such license applies only to those patent claims, both currently owned by +Google and acquired in the future, licensable by Google that are necessarily +infringed by these implementations of WebM. This grant does not include claims +that would be infringed only as a consequence of further modification of these +implementations. If you or your agent or exclusive licensee institute or order +or agree to the institution of patent litigation or any other patent +enforcement activity against any entity (including a cross-claim or +counterclaim in a lawsuit) alleging that any of these implementations of WebM +or any code incorporated within any of these implementations of WebM +constitute direct or contributory patent infringement, or inducement of +patent infringement, then any patent rights granted to you under this License +for these implementations of WebM shall terminate as of the date such +litigation is filed. diff --git a/thirdparty/libvpx/third_party/x86inc/LICENSE b/thirdparty/libvpx/third_party/x86inc/LICENSE new file mode 100644 index 0000000000..7d07645a17 --- /dev/null +++ b/thirdparty/libvpx/third_party/x86inc/LICENSE @@ -0,0 +1,18 @@ +Copyright (C) 2005-2012 x264 project + +Authors: Loren Merritt <lorenm@u.washington.edu> + Anton Mitrofanov <BugMaster@narod.ru> + Jason Garrett-Glaser <darkshikari@gmail.com> + Henrik Gramner <hengar-6@student.ltu.se> + +Permission to use, copy, modify, and/or distribute this software for any +purpose with or without fee is hereby granted, provided that the above +copyright notice and this permission notice appear in all copies. + +THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES +WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF +MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR +ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES +WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN +ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF +OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. diff --git a/thirdparty/libvpx/third_party/x86inc/README.libvpx b/thirdparty/libvpx/third_party/x86inc/README.libvpx new file mode 100644 index 0000000000..8d3cd966da --- /dev/null +++ b/thirdparty/libvpx/third_party/x86inc/README.libvpx @@ -0,0 +1,20 @@ +URL: https://git.videolan.org/git/x264.git +Version: d23d18655249944c1ca894b451e2c82c7a584c62 +License: ISC +License File: LICENSE + +Description: +x264/libav's framework for x86 assembly. Contains a variety of macros and +defines that help automatically allow assembly to work cross-platform. + +Local Modifications: +Get configuration from vpx_config.asm. +Prefix functions with vpx by default. +Manage name mangling (prefixing with '_') manually because 'PREFIX' does not + exist in libvpx. +Expand PIC default to macho64 and respect CONFIG_PIC from libvpx +Set 'private_extern' visibility for macho targets. +Copy PIC 'GLOBAL' macros from x86_abi_support.asm +Use .text instead of .rodata on macho to avoid broken tables in PIC mode. +Use .text with no alignment for aout +Only use 'hidden' visibility with Chromium diff --git a/thirdparty/libvpx/third_party/x86inc/x86inc.asm b/thirdparty/libvpx/third_party/x86inc/x86inc.asm new file mode 100644 index 0000000000..b647dff2f8 --- /dev/null +++ b/thirdparty/libvpx/third_party/x86inc/x86inc.asm @@ -0,0 +1,1649 @@ +;***************************************************************************** +;* x86inc.asm: x264asm abstraction layer +;***************************************************************************** +;* Copyright (C) 2005-2016 x264 project +;* +;* Authors: Loren Merritt <lorenm@u.washington.edu> +;* Anton Mitrofanov <BugMaster@narod.ru> +;* Fiona Glaser <fiona@x264.com> +;* Henrik Gramner <henrik@gramner.com> +;* +;* Permission to use, copy, modify, and/or distribute this software for any +;* purpose with or without fee is hereby granted, provided that the above +;* copyright notice and this permission notice appear in all copies. +;* +;* THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES +;* WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF +;* MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR +;* ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES +;* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN +;* ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF +;* OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. +;***************************************************************************** + +; This is a header file for the x264ASM assembly language, which uses +; NASM/YASM syntax combined with a large number of macros to provide easy +; abstraction between different calling conventions (x86_32, win64, linux64). +; It also has various other useful features to simplify writing the kind of +; DSP functions that are most often used in x264. + +; Unlike the rest of x264, this file is available under an ISC license, as it +; has significant usefulness outside of x264 and we want it to be available +; to the largest audience possible. Of course, if you modify it for your own +; purposes to add a new feature, we strongly encourage contributing a patch +; as this feature might be useful for others as well. Send patches or ideas +; to x264-devel@videolan.org . + +%include "vpx_config.asm" + +%ifndef private_prefix + %define private_prefix vpx +%endif + +%ifndef public_prefix + %define public_prefix private_prefix +%endif + +%ifndef STACK_ALIGNMENT + %if ARCH_X86_64 + %define STACK_ALIGNMENT 16 + %else + %define STACK_ALIGNMENT 4 + %endif +%endif + +%define WIN64 0 +%define UNIX64 0 +%if ARCH_X86_64 + %ifidn __OUTPUT_FORMAT__,win32 + %define WIN64 1 + %elifidn __OUTPUT_FORMAT__,win64 + %define WIN64 1 + %elifidn __OUTPUT_FORMAT__,x64 + %define WIN64 1 + %else + %define UNIX64 1 + %endif +%endif + +%define FORMAT_ELF 0 +%ifidn __OUTPUT_FORMAT__,elf + %define FORMAT_ELF 1 +%elifidn __OUTPUT_FORMAT__,elf32 + %define FORMAT_ELF 1 +%elifidn __OUTPUT_FORMAT__,elf64 + %define FORMAT_ELF 1 +%endif + +%define FORMAT_MACHO 0 +%ifidn __OUTPUT_FORMAT__,macho32 + %define FORMAT_MACHO 1 +%elifidn __OUTPUT_FORMAT__,macho64 + %define FORMAT_MACHO 1 +%endif + +; Set PREFIX for libvpx builds. +%if FORMAT_ELF + %undef PREFIX +%elif WIN64 + %undef PREFIX +%else + %define PREFIX +%endif + +%ifdef PREFIX + %define mangle(x) _ %+ x +%else + %define mangle(x) x +%endif + +; In some instances macho32 tables get misaligned when using .rodata. +; When looking at the disassembly it appears that the offset is either +; correct or consistently off by 90. Placing them in the .text section +; works around the issue. It appears to be specific to the way libvpx +; handles the tables. +%macro SECTION_RODATA 0-1 16 + %ifidn __OUTPUT_FORMAT__,macho32 + SECTION .text align=%1 + fakegot: + %elifidn __OUTPUT_FORMAT__,aout + SECTION .text + %else + SECTION .rodata align=%1 + %endif +%endmacro + +; PIC macros are copied from vpx_ports/x86_abi_support.asm. The "define PIC" +; from original code is added in for 64bit. +%ifidn __OUTPUT_FORMAT__,elf32 +%define ABI_IS_32BIT 1 +%elifidn __OUTPUT_FORMAT__,macho32 +%define ABI_IS_32BIT 1 +%elifidn __OUTPUT_FORMAT__,win32 +%define ABI_IS_32BIT 1 +%elifidn __OUTPUT_FORMAT__,aout +%define ABI_IS_32BIT 1 +%else +%define ABI_IS_32BIT 0 +%endif + +%if ABI_IS_32BIT + %if CONFIG_PIC=1 + %ifidn __OUTPUT_FORMAT__,elf32 + %define GET_GOT_DEFINED 1 + %define WRT_PLT wrt ..plt + %macro GET_GOT 1 + extern _GLOBAL_OFFSET_TABLE_ + push %1 + call %%get_got + %%sub_offset: + jmp %%exitGG + %%get_got: + mov %1, [esp] + add %1, _GLOBAL_OFFSET_TABLE_ + $$ - %%sub_offset wrt ..gotpc + ret + %%exitGG: + %undef GLOBAL + %define GLOBAL(x) x + %1 wrt ..gotoff + %undef RESTORE_GOT + %define RESTORE_GOT pop %1 + %endmacro + %elifidn __OUTPUT_FORMAT__,macho32 + %define GET_GOT_DEFINED 1 + %macro GET_GOT 1 + push %1 + call %%get_got + %%get_got: + pop %1 + %undef GLOBAL + %define GLOBAL(x) x + %1 - %%get_got + %undef RESTORE_GOT + %define RESTORE_GOT pop %1 + %endmacro + %else + %define GET_GOT_DEFINED 0 + %endif + %endif + + %if ARCH_X86_64 == 0 + %undef PIC + %endif + +%else + %macro GET_GOT 1 + %endmacro + %define GLOBAL(x) rel x + %define WRT_PLT wrt ..plt + + %if WIN64 + %define PIC + %elifidn __OUTPUT_FORMAT__,macho64 + %define PIC + %elif CONFIG_PIC + %define PIC + %endif +%endif + +%ifnmacro GET_GOT + %macro GET_GOT 1 + %endmacro + %define GLOBAL(x) x +%endif +%ifndef RESTORE_GOT + %define RESTORE_GOT +%endif +%ifndef WRT_PLT + %define WRT_PLT +%endif + +%ifdef PIC + default rel +%endif + +%ifndef GET_GOT_DEFINED + %define GET_GOT_DEFINED 0 +%endif +; Done with PIC macros + +%ifdef __NASM_VER__ + %use smartalign +%endif + +; Macros to eliminate most code duplication between x86_32 and x86_64: +; Currently this works only for leaf functions which load all their arguments +; into registers at the start, and make no other use of the stack. Luckily that +; covers most of x264's asm. + +; PROLOGUE: +; %1 = number of arguments. loads them from stack if needed. +; %2 = number of registers used. pushes callee-saved regs if needed. +; %3 = number of xmm registers used. pushes callee-saved xmm regs if needed. +; %4 = (optional) stack size to be allocated. The stack will be aligned before +; allocating the specified stack size. If the required stack alignment is +; larger than the known stack alignment the stack will be manually aligned +; and an extra register will be allocated to hold the original stack +; pointer (to not invalidate r0m etc.). To prevent the use of an extra +; register as stack pointer, request a negative stack size. +; %4+/%5+ = list of names to define to registers +; PROLOGUE can also be invoked by adding the same options to cglobal + +; e.g. +; cglobal foo, 2,3,7,0x40, dst, src, tmp +; declares a function (foo) that automatically loads two arguments (dst and +; src) into registers, uses one additional register (tmp) plus 7 vector +; registers (m0-m6) and allocates 0x40 bytes of stack space. + +; TODO Some functions can use some args directly from the stack. If they're the +; last args then you can just not declare them, but if they're in the middle +; we need more flexible macro. + +; RET: +; Pops anything that was pushed by PROLOGUE, and returns. + +; REP_RET: +; Use this instead of RET if it's a branch target. + +; registers: +; rN and rNq are the native-size register holding function argument N +; rNd, rNw, rNb are dword, word, and byte size +; rNh is the high 8 bits of the word size +; rNm is the original location of arg N (a register or on the stack), dword +; rNmp is native size + +%macro DECLARE_REG 2-3 + %define r%1q %2 + %define r%1d %2d + %define r%1w %2w + %define r%1b %2b + %define r%1h %2h + %define %2q %2 + %if %0 == 2 + %define r%1m %2d + %define r%1mp %2 + %elif ARCH_X86_64 ; memory + %define r%1m [rstk + stack_offset + %3] + %define r%1mp qword r %+ %1 %+ m + %else + %define r%1m [rstk + stack_offset + %3] + %define r%1mp dword r %+ %1 %+ m + %endif + %define r%1 %2 +%endmacro + +%macro DECLARE_REG_SIZE 3 + %define r%1q r%1 + %define e%1q r%1 + %define r%1d e%1 + %define e%1d e%1 + %define r%1w %1 + %define e%1w %1 + %define r%1h %3 + %define e%1h %3 + %define r%1b %2 + %define e%1b %2 + %if ARCH_X86_64 == 0 + %define r%1 e%1 + %endif +%endmacro + +DECLARE_REG_SIZE ax, al, ah +DECLARE_REG_SIZE bx, bl, bh +DECLARE_REG_SIZE cx, cl, ch +DECLARE_REG_SIZE dx, dl, dh +DECLARE_REG_SIZE si, sil, null +DECLARE_REG_SIZE di, dil, null +DECLARE_REG_SIZE bp, bpl, null + +; t# defines for when per-arch register allocation is more complex than just function arguments + +%macro DECLARE_REG_TMP 1-* + %assign %%i 0 + %rep %0 + CAT_XDEFINE t, %%i, r%1 + %assign %%i %%i+1 + %rotate 1 + %endrep +%endmacro + +%macro DECLARE_REG_TMP_SIZE 0-* + %rep %0 + %define t%1q t%1 %+ q + %define t%1d t%1 %+ d + %define t%1w t%1 %+ w + %define t%1h t%1 %+ h + %define t%1b t%1 %+ b + %rotate 1 + %endrep +%endmacro + +DECLARE_REG_TMP_SIZE 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14 + +%if ARCH_X86_64 + %define gprsize 8 +%else + %define gprsize 4 +%endif + +%macro PUSH 1 + push %1 + %ifidn rstk, rsp + %assign stack_offset stack_offset+gprsize + %endif +%endmacro + +%macro POP 1 + pop %1 + %ifidn rstk, rsp + %assign stack_offset stack_offset-gprsize + %endif +%endmacro + +%macro PUSH_IF_USED 1-* + %rep %0 + %if %1 < regs_used + PUSH r%1 + %endif + %rotate 1 + %endrep +%endmacro + +%macro POP_IF_USED 1-* + %rep %0 + %if %1 < regs_used + pop r%1 + %endif + %rotate 1 + %endrep +%endmacro + +%macro LOAD_IF_USED 1-* + %rep %0 + %if %1 < num_args + mov r%1, r %+ %1 %+ mp + %endif + %rotate 1 + %endrep +%endmacro + +%macro SUB 2 + sub %1, %2 + %ifidn %1, rstk + %assign stack_offset stack_offset+(%2) + %endif +%endmacro + +%macro ADD 2 + add %1, %2 + %ifidn %1, rstk + %assign stack_offset stack_offset-(%2) + %endif +%endmacro + +%macro movifnidn 2 + %ifnidn %1, %2 + mov %1, %2 + %endif +%endmacro + +%macro movsxdifnidn 2 + %ifnidn %1, %2 + movsxd %1, %2 + %endif +%endmacro + +%macro ASSERT 1 + %if (%1) == 0 + %error assertion ``%1'' failed + %endif +%endmacro + +%macro DEFINE_ARGS 0-* + %ifdef n_arg_names + %assign %%i 0 + %rep n_arg_names + CAT_UNDEF arg_name %+ %%i, q + CAT_UNDEF arg_name %+ %%i, d + CAT_UNDEF arg_name %+ %%i, w + CAT_UNDEF arg_name %+ %%i, h + CAT_UNDEF arg_name %+ %%i, b + CAT_UNDEF arg_name %+ %%i, m + CAT_UNDEF arg_name %+ %%i, mp + CAT_UNDEF arg_name, %%i + %assign %%i %%i+1 + %endrep + %endif + + %xdefine %%stack_offset stack_offset + %undef stack_offset ; so that the current value of stack_offset doesn't get baked in by xdefine + %assign %%i 0 + %rep %0 + %xdefine %1q r %+ %%i %+ q + %xdefine %1d r %+ %%i %+ d + %xdefine %1w r %+ %%i %+ w + %xdefine %1h r %+ %%i %+ h + %xdefine %1b r %+ %%i %+ b + %xdefine %1m r %+ %%i %+ m + %xdefine %1mp r %+ %%i %+ mp + CAT_XDEFINE arg_name, %%i, %1 + %assign %%i %%i+1 + %rotate 1 + %endrep + %xdefine stack_offset %%stack_offset + %assign n_arg_names %0 +%endmacro + +%define required_stack_alignment ((mmsize + 15) & ~15) + +%macro ALLOC_STACK 1-2 0 ; stack_size, n_xmm_regs (for win64 only) + %ifnum %1 + %if %1 != 0 + %assign %%pad 0 + %assign stack_size %1 + %if stack_size < 0 + %assign stack_size -stack_size + %endif + %if WIN64 + %assign %%pad %%pad + 32 ; shadow space + %if mmsize != 8 + %assign xmm_regs_used %2 + %if xmm_regs_used > 8 + %assign %%pad %%pad + (xmm_regs_used-8)*16 ; callee-saved xmm registers + %endif + %endif + %endif + %if required_stack_alignment <= STACK_ALIGNMENT + ; maintain the current stack alignment + %assign stack_size_padded stack_size + %%pad + ((-%%pad-stack_offset-gprsize) & (STACK_ALIGNMENT-1)) + SUB rsp, stack_size_padded + %else + %assign %%reg_num (regs_used - 1) + %xdefine rstk r %+ %%reg_num + ; align stack, and save original stack location directly above + ; it, i.e. in [rsp+stack_size_padded], so we can restore the + ; stack in a single instruction (i.e. mov rsp, rstk or mov + ; rsp, [rsp+stack_size_padded]) + %if %1 < 0 ; need to store rsp on stack + %xdefine rstkm [rsp + stack_size + %%pad] + %assign %%pad %%pad + gprsize + %else ; can keep rsp in rstk during whole function + %xdefine rstkm rstk + %endif + %assign stack_size_padded stack_size + ((%%pad + required_stack_alignment-1) & ~(required_stack_alignment-1)) + mov rstk, rsp + and rsp, ~(required_stack_alignment-1) + sub rsp, stack_size_padded + movifnidn rstkm, rstk + %endif + WIN64_PUSH_XMM + %endif + %endif +%endmacro + +%macro SETUP_STACK_POINTER 1 + %ifnum %1 + %if %1 != 0 && required_stack_alignment > STACK_ALIGNMENT + %if %1 > 0 + %assign regs_used (regs_used + 1) + %endif + %if ARCH_X86_64 && regs_used < 5 + UNIX64 * 3 + ; Ensure that we don't clobber any registers containing arguments + %assign regs_used 5 + UNIX64 * 3 + %endif + %endif + %endif +%endmacro + +%macro DEFINE_ARGS_INTERNAL 3+ + %ifnum %2 + DEFINE_ARGS %3 + %elif %1 == 4 + DEFINE_ARGS %2 + %elif %1 > 4 + DEFINE_ARGS %2, %3 + %endif +%endmacro + +%if WIN64 ; Windows x64 ;================================================= + +DECLARE_REG 0, rcx +DECLARE_REG 1, rdx +DECLARE_REG 2, R8 +DECLARE_REG 3, R9 +DECLARE_REG 4, R10, 40 +DECLARE_REG 5, R11, 48 +DECLARE_REG 6, rax, 56 +DECLARE_REG 7, rdi, 64 +DECLARE_REG 8, rsi, 72 +DECLARE_REG 9, rbx, 80 +DECLARE_REG 10, rbp, 88 +DECLARE_REG 11, R12, 96 +DECLARE_REG 12, R13, 104 +DECLARE_REG 13, R14, 112 +DECLARE_REG 14, R15, 120 + +%macro PROLOGUE 2-5+ 0 ; #args, #regs, #xmm_regs, [stack_size,] arg_names... + %assign num_args %1 + %assign regs_used %2 + ASSERT regs_used >= num_args + SETUP_STACK_POINTER %4 + ASSERT regs_used <= 15 + PUSH_IF_USED 7, 8, 9, 10, 11, 12, 13, 14 + ALLOC_STACK %4, %3 + %if mmsize != 8 && stack_size == 0 + WIN64_SPILL_XMM %3 + %endif + LOAD_IF_USED 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 + DEFINE_ARGS_INTERNAL %0, %4, %5 +%endmacro + +%macro WIN64_PUSH_XMM 0 + ; Use the shadow space to store XMM6 and XMM7, the rest needs stack space allocated. + %if xmm_regs_used > 6 + movaps [rstk + stack_offset + 8], xmm6 + %endif + %if xmm_regs_used > 7 + movaps [rstk + stack_offset + 24], xmm7 + %endif + %if xmm_regs_used > 8 + %assign %%i 8 + %rep xmm_regs_used-8 + movaps [rsp + (%%i-8)*16 + stack_size + 32], xmm %+ %%i + %assign %%i %%i+1 + %endrep + %endif +%endmacro + +%macro WIN64_SPILL_XMM 1 + %assign xmm_regs_used %1 + ASSERT xmm_regs_used <= 16 + %if xmm_regs_used > 8 + ; Allocate stack space for callee-saved xmm registers plus shadow space and align the stack. + %assign %%pad (xmm_regs_used-8)*16 + 32 + %assign stack_size_padded %%pad + ((-%%pad-stack_offset-gprsize) & (STACK_ALIGNMENT-1)) + SUB rsp, stack_size_padded + %endif + WIN64_PUSH_XMM +%endmacro + +%macro WIN64_RESTORE_XMM_INTERNAL 1 + %assign %%pad_size 0 + %if xmm_regs_used > 8 + %assign %%i xmm_regs_used + %rep xmm_regs_used-8 + %assign %%i %%i-1 + movaps xmm %+ %%i, [%1 + (%%i-8)*16 + stack_size + 32] + %endrep + %endif + %if stack_size_padded > 0 + %if stack_size > 0 && required_stack_alignment > STACK_ALIGNMENT + mov rsp, rstkm + %else + add %1, stack_size_padded + %assign %%pad_size stack_size_padded + %endif + %endif + %if xmm_regs_used > 7 + movaps xmm7, [%1 + stack_offset - %%pad_size + 24] + %endif + %if xmm_regs_used > 6 + movaps xmm6, [%1 + stack_offset - %%pad_size + 8] + %endif +%endmacro + +%macro WIN64_RESTORE_XMM 1 + WIN64_RESTORE_XMM_INTERNAL %1 + %assign stack_offset (stack_offset-stack_size_padded) + %assign xmm_regs_used 0 +%endmacro + +%define has_epilogue regs_used > 7 || xmm_regs_used > 6 || mmsize == 32 || stack_size > 0 + +%macro RET 0 + WIN64_RESTORE_XMM_INTERNAL rsp + POP_IF_USED 14, 13, 12, 11, 10, 9, 8, 7 + %if mmsize == 32 + vzeroupper + %endif + AUTO_REP_RET +%endmacro + +%elif ARCH_X86_64 ; *nix x64 ;============================================= + +DECLARE_REG 0, rdi +DECLARE_REG 1, rsi +DECLARE_REG 2, rdx +DECLARE_REG 3, rcx +DECLARE_REG 4, R8 +DECLARE_REG 5, R9 +DECLARE_REG 6, rax, 8 +DECLARE_REG 7, R10, 16 +DECLARE_REG 8, R11, 24 +DECLARE_REG 9, rbx, 32 +DECLARE_REG 10, rbp, 40 +DECLARE_REG 11, R12, 48 +DECLARE_REG 12, R13, 56 +DECLARE_REG 13, R14, 64 +DECLARE_REG 14, R15, 72 + +%macro PROLOGUE 2-5+ ; #args, #regs, #xmm_regs, [stack_size,] arg_names... + %assign num_args %1 + %assign regs_used %2 + ASSERT regs_used >= num_args + SETUP_STACK_POINTER %4 + ASSERT regs_used <= 15 + PUSH_IF_USED 9, 10, 11, 12, 13, 14 + ALLOC_STACK %4 + LOAD_IF_USED 6, 7, 8, 9, 10, 11, 12, 13, 14 + DEFINE_ARGS_INTERNAL %0, %4, %5 +%endmacro + +%define has_epilogue regs_used > 9 || mmsize == 32 || stack_size > 0 + +%macro RET 0 + %if stack_size_padded > 0 + %if required_stack_alignment > STACK_ALIGNMENT + mov rsp, rstkm + %else + add rsp, stack_size_padded + %endif + %endif + POP_IF_USED 14, 13, 12, 11, 10, 9 + %if mmsize == 32 + vzeroupper + %endif + AUTO_REP_RET +%endmacro + +%else ; X86_32 ;============================================================== + +DECLARE_REG 0, eax, 4 +DECLARE_REG 1, ecx, 8 +DECLARE_REG 2, edx, 12 +DECLARE_REG 3, ebx, 16 +DECLARE_REG 4, esi, 20 +DECLARE_REG 5, edi, 24 +DECLARE_REG 6, ebp, 28 +%define rsp esp + +%macro DECLARE_ARG 1-* + %rep %0 + %define r%1m [rstk + stack_offset + 4*%1 + 4] + %define r%1mp dword r%1m + %rotate 1 + %endrep +%endmacro + +DECLARE_ARG 7, 8, 9, 10, 11, 12, 13, 14 + +%macro PROLOGUE 2-5+ ; #args, #regs, #xmm_regs, [stack_size,] arg_names... + %assign num_args %1 + %assign regs_used %2 + ASSERT regs_used >= num_args + %if num_args > 7 + %assign num_args 7 + %endif + %if regs_used > 7 + %assign regs_used 7 + %endif + SETUP_STACK_POINTER %4 + ASSERT regs_used <= 7 + PUSH_IF_USED 3, 4, 5, 6 + ALLOC_STACK %4 + LOAD_IF_USED 0, 1, 2, 3, 4, 5, 6 + DEFINE_ARGS_INTERNAL %0, %4, %5 +%endmacro + +%define has_epilogue regs_used > 3 || mmsize == 32 || stack_size > 0 + +%macro RET 0 + %if stack_size_padded > 0 + %if required_stack_alignment > STACK_ALIGNMENT + mov rsp, rstkm + %else + add rsp, stack_size_padded + %endif + %endif + POP_IF_USED 6, 5, 4, 3 + %if mmsize == 32 + vzeroupper + %endif + AUTO_REP_RET +%endmacro + +%endif ;====================================================================== + +%if WIN64 == 0 + %macro WIN64_SPILL_XMM 1 + %endmacro + %macro WIN64_RESTORE_XMM 1 + %endmacro + %macro WIN64_PUSH_XMM 0 + %endmacro +%endif + +; On AMD cpus <=K10, an ordinary ret is slow if it immediately follows either +; a branch or a branch target. So switch to a 2-byte form of ret in that case. +; We can automatically detect "follows a branch", but not a branch target. +; (SSSE3 is a sufficient condition to know that your cpu doesn't have this problem.) +%macro REP_RET 0 + %if has_epilogue + RET + %else + rep ret + %endif + annotate_function_size +%endmacro + +%define last_branch_adr $$ +%macro AUTO_REP_RET 0 + %if notcpuflag(ssse3) + times ((last_branch_adr-$)>>31)+1 rep ; times 1 iff $ == last_branch_adr. + %endif + ret + annotate_function_size +%endmacro + +%macro BRANCH_INSTR 0-* + %rep %0 + %macro %1 1-2 %1 + %2 %1 + %if notcpuflag(ssse3) + %%branch_instr equ $ + %xdefine last_branch_adr %%branch_instr + %endif + %endmacro + %rotate 1 + %endrep +%endmacro + +BRANCH_INSTR jz, je, jnz, jne, jl, jle, jnl, jnle, jg, jge, jng, jnge, ja, jae, jna, jnae, jb, jbe, jnb, jnbe, jc, jnc, js, jns, jo, jno, jp, jnp + +%macro TAIL_CALL 2 ; callee, is_nonadjacent + %if has_epilogue + call %1 + RET + %elif %2 + jmp %1 + %endif + annotate_function_size +%endmacro + +;============================================================================= +; arch-independent part +;============================================================================= + +%assign function_align 16 + +; Begin a function. +; Applies any symbol mangling needed for C linkage, and sets up a define such that +; subsequent uses of the function name automatically refer to the mangled version. +; Appends cpuflags to the function name if cpuflags has been specified. +; The "" empty default parameter is a workaround for nasm, which fails if SUFFIX +; is empty and we call cglobal_internal with just %1 %+ SUFFIX (without %2). +%macro cglobal 1-2+ "" ; name, [PROLOGUE args] + cglobal_internal 1, %1 %+ SUFFIX, %2 +%endmacro +%macro cvisible 1-2+ "" ; name, [PROLOGUE args] + cglobal_internal 0, %1 %+ SUFFIX, %2 +%endmacro +%macro cglobal_internal 2-3+ + annotate_function_size + %if %1 + %xdefine %%FUNCTION_PREFIX private_prefix + ; libvpx explicitly sets visibility in shared object builds. Avoid + ; setting visibility to hidden as it may break builds that split + ; sources on e.g., directory boundaries. + %ifdef CHROMIUM + %xdefine %%VISIBILITY hidden + %else + %xdefine %%VISIBILITY + %endif + %else + %xdefine %%FUNCTION_PREFIX public_prefix + %xdefine %%VISIBILITY + %endif + %ifndef cglobaled_%2 + %xdefine %2 mangle(%%FUNCTION_PREFIX %+ _ %+ %2) + %xdefine %2.skip_prologue %2 %+ .skip_prologue + CAT_XDEFINE cglobaled_, %2, 1 + %endif + %xdefine current_function %2 + %xdefine current_function_section __SECT__ + %if FORMAT_ELF + global %2:function %%VISIBILITY + %elif FORMAT_MACHO + %ifdef __NASM_VER__ + global %2 + %else + global %2:private_extern + %endif + %else + global %2 + %endif + align function_align + %2: + RESET_MM_PERMUTATION ; needed for x86-64, also makes disassembly somewhat nicer + %xdefine rstk rsp ; copy of the original stack pointer, used when greater alignment than the known stack alignment is required + %assign stack_offset 0 ; stack pointer offset relative to the return address + %assign stack_size 0 ; amount of stack space that can be freely used inside a function + %assign stack_size_padded 0 ; total amount of allocated stack space, including space for callee-saved xmm registers on WIN64 and alignment padding + %assign xmm_regs_used 0 ; number of XMM registers requested, used for dealing with callee-saved registers on WIN64 + %ifnidn %3, "" + PROLOGUE %3 + %endif +%endmacro + +%macro cextern 1 + %xdefine %1 mangle(private_prefix %+ _ %+ %1) + CAT_XDEFINE cglobaled_, %1, 1 + extern %1 +%endmacro + +; like cextern, but without the prefix +%macro cextern_naked 1 + %ifdef PREFIX + %xdefine %1 mangle(%1) + %endif + CAT_XDEFINE cglobaled_, %1, 1 + extern %1 +%endmacro + +%macro const 1-2+ + %xdefine %1 mangle(private_prefix %+ _ %+ %1) + %if FORMAT_ELF + global %1:data hidden + %else + global %1 + %endif + %1: %2 +%endmacro + +; This is needed for ELF, otherwise the GNU linker assumes the stack is executable by default. +%if FORMAT_ELF + [SECTION .note.GNU-stack noalloc noexec nowrite progbits] +%endif + +; Tell debuggers how large the function was. +; This may be invoked multiple times per function; we rely on later instances overriding earlier ones. +; This is invoked by RET and similar macros, and also cglobal does it for the previous function, +; but if the last function in a source file doesn't use any of the standard macros for its epilogue, +; then its size might be unspecified. +%macro annotate_function_size 0 + %ifdef __YASM_VER__ + %ifdef current_function + %if FORMAT_ELF + current_function_section + %%ecf equ $ + size current_function %%ecf - current_function + __SECT__ + %endif + %endif + %endif +%endmacro + +; cpuflags + +%assign cpuflags_mmx (1<<0) +%assign cpuflags_mmx2 (1<<1) | cpuflags_mmx +%assign cpuflags_3dnow (1<<2) | cpuflags_mmx +%assign cpuflags_3dnowext (1<<3) | cpuflags_3dnow +%assign cpuflags_sse (1<<4) | cpuflags_mmx2 +%assign cpuflags_sse2 (1<<5) | cpuflags_sse +%assign cpuflags_sse2slow (1<<6) | cpuflags_sse2 +%assign cpuflags_sse3 (1<<7) | cpuflags_sse2 +%assign cpuflags_ssse3 (1<<8) | cpuflags_sse3 +%assign cpuflags_sse4 (1<<9) | cpuflags_ssse3 +%assign cpuflags_sse42 (1<<10)| cpuflags_sse4 +%assign cpuflags_avx (1<<11)| cpuflags_sse42 +%assign cpuflags_xop (1<<12)| cpuflags_avx +%assign cpuflags_fma4 (1<<13)| cpuflags_avx +%assign cpuflags_fma3 (1<<14)| cpuflags_avx +%assign cpuflags_avx2 (1<<15)| cpuflags_fma3 + +%assign cpuflags_cache32 (1<<16) +%assign cpuflags_cache64 (1<<17) +%assign cpuflags_slowctz (1<<18) +%assign cpuflags_lzcnt (1<<19) +%assign cpuflags_aligned (1<<20) ; not a cpu feature, but a function variant +%assign cpuflags_atom (1<<21) +%assign cpuflags_bmi1 (1<<22)|cpuflags_lzcnt +%assign cpuflags_bmi2 (1<<23)|cpuflags_bmi1 + +; Returns a boolean value expressing whether or not the specified cpuflag is enabled. +%define cpuflag(x) (((((cpuflags & (cpuflags_ %+ x)) ^ (cpuflags_ %+ x)) - 1) >> 31) & 1) +%define notcpuflag(x) (cpuflag(x) ^ 1) + +; Takes an arbitrary number of cpuflags from the above list. +; All subsequent functions (up to the next INIT_CPUFLAGS) is built for the specified cpu. +; You shouldn't need to invoke this macro directly, it's a subroutine for INIT_MMX &co. +%macro INIT_CPUFLAGS 0-* + %xdefine SUFFIX + %undef cpuname + %assign cpuflags 0 + + %if %0 >= 1 + %rep %0 + %ifdef cpuname + %xdefine cpuname cpuname %+ _%1 + %else + %xdefine cpuname %1 + %endif + %assign cpuflags cpuflags | cpuflags_%1 + %rotate 1 + %endrep + %xdefine SUFFIX _ %+ cpuname + + %if cpuflag(avx) + %assign avx_enabled 1 + %endif + %if (mmsize == 16 && notcpuflag(sse2)) || (mmsize == 32 && notcpuflag(avx2)) + %define mova movaps + %define movu movups + %define movnta movntps + %endif + %if cpuflag(aligned) + %define movu mova + %elif cpuflag(sse3) && notcpuflag(ssse3) + %define movu lddqu + %endif + %endif + + %if ARCH_X86_64 || cpuflag(sse2) + %ifdef __NASM_VER__ + ALIGNMODE k8 + %else + CPU amdnop + %endif + %else + %ifdef __NASM_VER__ + ALIGNMODE nop + %else + CPU basicnop + %endif + %endif +%endmacro + +; Merge mmx and sse* +; m# is a simd register of the currently selected size +; xm# is the corresponding xmm register if mmsize >= 16, otherwise the same as m# +; ym# is the corresponding ymm register if mmsize >= 32, otherwise the same as m# +; (All 3 remain in sync through SWAP.) + +%macro CAT_XDEFINE 3 + %xdefine %1%2 %3 +%endmacro + +%macro CAT_UNDEF 2 + %undef %1%2 +%endmacro + +%macro INIT_MMX 0-1+ + %assign avx_enabled 0 + %define RESET_MM_PERMUTATION INIT_MMX %1 + %define mmsize 8 + %define num_mmregs 8 + %define mova movq + %define movu movq + %define movh movd + %define movnta movntq + %assign %%i 0 + %rep 8 + CAT_XDEFINE m, %%i, mm %+ %%i + CAT_XDEFINE nnmm, %%i, %%i + %assign %%i %%i+1 + %endrep + %rep 8 + CAT_UNDEF m, %%i + CAT_UNDEF nnmm, %%i + %assign %%i %%i+1 + %endrep + INIT_CPUFLAGS %1 +%endmacro + +%macro INIT_XMM 0-1+ + %assign avx_enabled 0 + %define RESET_MM_PERMUTATION INIT_XMM %1 + %define mmsize 16 + %define num_mmregs 8 + %if ARCH_X86_64 + %define num_mmregs 16 + %endif + %define mova movdqa + %define movu movdqu + %define movh movq + %define movnta movntdq + %assign %%i 0 + %rep num_mmregs + CAT_XDEFINE m, %%i, xmm %+ %%i + CAT_XDEFINE nnxmm, %%i, %%i + %assign %%i %%i+1 + %endrep + INIT_CPUFLAGS %1 +%endmacro + +%macro INIT_YMM 0-1+ + %assign avx_enabled 1 + %define RESET_MM_PERMUTATION INIT_YMM %1 + %define mmsize 32 + %define num_mmregs 8 + %if ARCH_X86_64 + %define num_mmregs 16 + %endif + %define mova movdqa + %define movu movdqu + %undef movh + %define movnta movntdq + %assign %%i 0 + %rep num_mmregs + CAT_XDEFINE m, %%i, ymm %+ %%i + CAT_XDEFINE nnymm, %%i, %%i + %assign %%i %%i+1 + %endrep + INIT_CPUFLAGS %1 +%endmacro + +INIT_XMM + +%macro DECLARE_MMCAST 1 + %define mmmm%1 mm%1 + %define mmxmm%1 mm%1 + %define mmymm%1 mm%1 + %define xmmmm%1 mm%1 + %define xmmxmm%1 xmm%1 + %define xmmymm%1 xmm%1 + %define ymmmm%1 mm%1 + %define ymmxmm%1 xmm%1 + %define ymmymm%1 ymm%1 + %define xm%1 xmm %+ m%1 + %define ym%1 ymm %+ m%1 +%endmacro + +%assign i 0 +%rep 16 + DECLARE_MMCAST i + %assign i i+1 +%endrep + +; I often want to use macros that permute their arguments. e.g. there's no +; efficient way to implement butterfly or transpose or dct without swapping some +; arguments. +; +; I would like to not have to manually keep track of the permutations: +; If I insert a permutation in the middle of a function, it should automatically +; change everything that follows. For more complex macros I may also have multiple +; implementations, e.g. the SSE2 and SSSE3 versions may have different permutations. +; +; Hence these macros. Insert a PERMUTE or some SWAPs at the end of a macro that +; permutes its arguments. It's equivalent to exchanging the contents of the +; registers, except that this way you exchange the register names instead, so it +; doesn't cost any cycles. + +%macro PERMUTE 2-* ; takes a list of pairs to swap + %rep %0/2 + %xdefine %%tmp%2 m%2 + %rotate 2 + %endrep + %rep %0/2 + %xdefine m%1 %%tmp%2 + CAT_XDEFINE nn, m%1, %1 + %rotate 2 + %endrep +%endmacro + +%macro SWAP 2+ ; swaps a single chain (sometimes more concise than pairs) + %ifnum %1 ; SWAP 0, 1, ... + SWAP_INTERNAL_NUM %1, %2 + %else ; SWAP m0, m1, ... + SWAP_INTERNAL_NAME %1, %2 + %endif +%endmacro + +%macro SWAP_INTERNAL_NUM 2-* + %rep %0-1 + %xdefine %%tmp m%1 + %xdefine m%1 m%2 + %xdefine m%2 %%tmp + CAT_XDEFINE nn, m%1, %1 + CAT_XDEFINE nn, m%2, %2 + %rotate 1 + %endrep +%endmacro + +%macro SWAP_INTERNAL_NAME 2-* + %xdefine %%args nn %+ %1 + %rep %0-1 + %xdefine %%args %%args, nn %+ %2 + %rotate 1 + %endrep + SWAP_INTERNAL_NUM %%args +%endmacro + +; If SAVE_MM_PERMUTATION is placed at the end of a function, then any later +; calls to that function will automatically load the permutation, so values can +; be returned in mmregs. +%macro SAVE_MM_PERMUTATION 0-1 + %if %0 + %xdefine %%f %1_m + %else + %xdefine %%f current_function %+ _m + %endif + %assign %%i 0 + %rep num_mmregs + CAT_XDEFINE %%f, %%i, m %+ %%i + %assign %%i %%i+1 + %endrep +%endmacro + +%macro LOAD_MM_PERMUTATION 1 ; name to load from + %ifdef %1_m0 + %assign %%i 0 + %rep num_mmregs + CAT_XDEFINE m, %%i, %1_m %+ %%i + CAT_XDEFINE nn, m %+ %%i, %%i + %assign %%i %%i+1 + %endrep + %endif +%endmacro + +; Append cpuflags to the callee's name iff the appended name is known and the plain name isn't +%macro call 1 + call_internal %1 %+ SUFFIX, %1 +%endmacro +%macro call_internal 2 + %xdefine %%i %2 + %ifndef cglobaled_%2 + %ifdef cglobaled_%1 + %xdefine %%i %1 + %endif + %endif + call %%i + LOAD_MM_PERMUTATION %%i +%endmacro + +; Substitutions that reduce instruction size but are functionally equivalent +%macro add 2 + %ifnum %2 + %if %2==128 + sub %1, -128 + %else + add %1, %2 + %endif + %else + add %1, %2 + %endif +%endmacro + +%macro sub 2 + %ifnum %2 + %if %2==128 + add %1, -128 + %else + sub %1, %2 + %endif + %else + sub %1, %2 + %endif +%endmacro + +;============================================================================= +; AVX abstraction layer +;============================================================================= + +%assign i 0 +%rep 16 + %if i < 8 + CAT_XDEFINE sizeofmm, i, 8 + %endif + CAT_XDEFINE sizeofxmm, i, 16 + CAT_XDEFINE sizeofymm, i, 32 + %assign i i+1 +%endrep +%undef i + +%macro CHECK_AVX_INSTR_EMU 3-* + %xdefine %%opcode %1 + %xdefine %%dst %2 + %rep %0-2 + %ifidn %%dst, %3 + %error non-avx emulation of ``%%opcode'' is not supported + %endif + %rotate 1 + %endrep +%endmacro + +;%1 == instruction +;%2 == minimal instruction set +;%3 == 1 if float, 0 if int +;%4 == 1 if non-destructive or 4-operand (xmm, xmm, xmm, imm), 0 otherwise +;%5 == 1 if commutative (i.e. doesn't matter which src arg is which), 0 if not +;%6+: operands +%macro RUN_AVX_INSTR 6-9+ + %ifnum sizeof%7 + %assign __sizeofreg sizeof%7 + %elifnum sizeof%6 + %assign __sizeofreg sizeof%6 + %else + %assign __sizeofreg mmsize + %endif + %assign __emulate_avx 0 + %if avx_enabled && __sizeofreg >= 16 + %xdefine __instr v%1 + %else + %xdefine __instr %1 + %if %0 >= 8+%4 + %assign __emulate_avx 1 + %endif + %endif + %ifnidn %2, fnord + %ifdef cpuname + %if notcpuflag(%2) + %error use of ``%1'' %2 instruction in cpuname function: current_function + %elif cpuflags_%2 < cpuflags_sse && notcpuflag(sse2) && __sizeofreg > 8 + %error use of ``%1'' sse2 instruction in cpuname function: current_function + %endif + %endif + %endif + + %if __emulate_avx + %xdefine __src1 %7 + %xdefine __src2 %8 + %ifnidn %6, %7 + %if %0 >= 9 + CHECK_AVX_INSTR_EMU {%1 %6, %7, %8, %9}, %6, %8, %9 + %else + CHECK_AVX_INSTR_EMU {%1 %6, %7, %8}, %6, %8 + %endif + %if %5 && %4 == 0 + %ifnid %8 + ; 3-operand AVX instructions with a memory arg can only have it in src2, + ; whereas SSE emulation prefers to have it in src1 (i.e. the mov). + ; So, if the instruction is commutative with a memory arg, swap them. + %xdefine __src1 %8 + %xdefine __src2 %7 + %endif + %endif + %if __sizeofreg == 8 + MOVQ %6, __src1 + %elif %3 + MOVAPS %6, __src1 + %else + MOVDQA %6, __src1 + %endif + %endif + %if %0 >= 9 + %1 %6, __src2, %9 + %else + %1 %6, __src2 + %endif + %elif %0 >= 9 + __instr %6, %7, %8, %9 + %elif %0 == 8 + __instr %6, %7, %8 + %elif %0 == 7 + __instr %6, %7 + %else + __instr %6 + %endif +%endmacro + +;%1 == instruction +;%2 == minimal instruction set +;%3 == 1 if float, 0 if int +;%4 == 1 if non-destructive or 4-operand (xmm, xmm, xmm, imm), 0 otherwise +;%5 == 1 if commutative (i.e. doesn't matter which src arg is which), 0 if not +%macro AVX_INSTR 1-5 fnord, 0, 1, 0 + %macro %1 1-10 fnord, fnord, fnord, fnord, %1, %2, %3, %4, %5 + %ifidn %2, fnord + RUN_AVX_INSTR %6, %7, %8, %9, %10, %1 + %elifidn %3, fnord + RUN_AVX_INSTR %6, %7, %8, %9, %10, %1, %2 + %elifidn %4, fnord + RUN_AVX_INSTR %6, %7, %8, %9, %10, %1, %2, %3 + %elifidn %5, fnord + RUN_AVX_INSTR %6, %7, %8, %9, %10, %1, %2, %3, %4 + %else + RUN_AVX_INSTR %6, %7, %8, %9, %10, %1, %2, %3, %4, %5 + %endif + %endmacro +%endmacro + +; Instructions with both VEX and non-VEX encodings +; Non-destructive instructions are written without parameters +AVX_INSTR addpd, sse2, 1, 0, 1 +AVX_INSTR addps, sse, 1, 0, 1 +AVX_INSTR addsd, sse2, 1, 0, 1 +AVX_INSTR addss, sse, 1, 0, 1 +AVX_INSTR addsubpd, sse3, 1, 0, 0 +AVX_INSTR addsubps, sse3, 1, 0, 0 +AVX_INSTR aesdec, fnord, 0, 0, 0 +AVX_INSTR aesdeclast, fnord, 0, 0, 0 +AVX_INSTR aesenc, fnord, 0, 0, 0 +AVX_INSTR aesenclast, fnord, 0, 0, 0 +AVX_INSTR aesimc +AVX_INSTR aeskeygenassist +AVX_INSTR andnpd, sse2, 1, 0, 0 +AVX_INSTR andnps, sse, 1, 0, 0 +AVX_INSTR andpd, sse2, 1, 0, 1 +AVX_INSTR andps, sse, 1, 0, 1 +AVX_INSTR blendpd, sse4, 1, 0, 0 +AVX_INSTR blendps, sse4, 1, 0, 0 +AVX_INSTR blendvpd, sse4, 1, 0, 0 +AVX_INSTR blendvps, sse4, 1, 0, 0 +AVX_INSTR cmppd, sse2, 1, 1, 0 +AVX_INSTR cmpps, sse, 1, 1, 0 +AVX_INSTR cmpsd, sse2, 1, 1, 0 +AVX_INSTR cmpss, sse, 1, 1, 0 +AVX_INSTR comisd, sse2 +AVX_INSTR comiss, sse +AVX_INSTR cvtdq2pd, sse2 +AVX_INSTR cvtdq2ps, sse2 +AVX_INSTR cvtpd2dq, sse2 +AVX_INSTR cvtpd2ps, sse2 +AVX_INSTR cvtps2dq, sse2 +AVX_INSTR cvtps2pd, sse2 +AVX_INSTR cvtsd2si, sse2 +AVX_INSTR cvtsd2ss, sse2 +AVX_INSTR cvtsi2sd, sse2 +AVX_INSTR cvtsi2ss, sse +AVX_INSTR cvtss2sd, sse2 +AVX_INSTR cvtss2si, sse +AVX_INSTR cvttpd2dq, sse2 +AVX_INSTR cvttps2dq, sse2 +AVX_INSTR cvttsd2si, sse2 +AVX_INSTR cvttss2si, sse +AVX_INSTR divpd, sse2, 1, 0, 0 +AVX_INSTR divps, sse, 1, 0, 0 +AVX_INSTR divsd, sse2, 1, 0, 0 +AVX_INSTR divss, sse, 1, 0, 0 +AVX_INSTR dppd, sse4, 1, 1, 0 +AVX_INSTR dpps, sse4, 1, 1, 0 +AVX_INSTR extractps, sse4 +AVX_INSTR haddpd, sse3, 1, 0, 0 +AVX_INSTR haddps, sse3, 1, 0, 0 +AVX_INSTR hsubpd, sse3, 1, 0, 0 +AVX_INSTR hsubps, sse3, 1, 0, 0 +AVX_INSTR insertps, sse4, 1, 1, 0 +AVX_INSTR lddqu, sse3 +AVX_INSTR ldmxcsr, sse +AVX_INSTR maskmovdqu, sse2 +AVX_INSTR maxpd, sse2, 1, 0, 1 +AVX_INSTR maxps, sse, 1, 0, 1 +AVX_INSTR maxsd, sse2, 1, 0, 1 +AVX_INSTR maxss, sse, 1, 0, 1 +AVX_INSTR minpd, sse2, 1, 0, 1 +AVX_INSTR minps, sse, 1, 0, 1 +AVX_INSTR minsd, sse2, 1, 0, 1 +AVX_INSTR minss, sse, 1, 0, 1 +AVX_INSTR movapd, sse2 +AVX_INSTR movaps, sse +AVX_INSTR movd, mmx +AVX_INSTR movddup, sse3 +AVX_INSTR movdqa, sse2 +AVX_INSTR movdqu, sse2 +AVX_INSTR movhlps, sse, 1, 0, 0 +AVX_INSTR movhpd, sse2, 1, 0, 0 +AVX_INSTR movhps, sse, 1, 0, 0 +AVX_INSTR movlhps, sse, 1, 0, 0 +AVX_INSTR movlpd, sse2, 1, 0, 0 +AVX_INSTR movlps, sse, 1, 0, 0 +AVX_INSTR movmskpd, sse2 +AVX_INSTR movmskps, sse +AVX_INSTR movntdq, sse2 +AVX_INSTR movntdqa, sse4 +AVX_INSTR movntpd, sse2 +AVX_INSTR movntps, sse +AVX_INSTR movq, mmx +AVX_INSTR movsd, sse2, 1, 0, 0 +AVX_INSTR movshdup, sse3 +AVX_INSTR movsldup, sse3 +AVX_INSTR movss, sse, 1, 0, 0 +AVX_INSTR movupd, sse2 +AVX_INSTR movups, sse +AVX_INSTR mpsadbw, sse4 +AVX_INSTR mulpd, sse2, 1, 0, 1 +AVX_INSTR mulps, sse, 1, 0, 1 +AVX_INSTR mulsd, sse2, 1, 0, 1 +AVX_INSTR mulss, sse, 1, 0, 1 +AVX_INSTR orpd, sse2, 1, 0, 1 +AVX_INSTR orps, sse, 1, 0, 1 +AVX_INSTR pabsb, ssse3 +AVX_INSTR pabsd, ssse3 +AVX_INSTR pabsw, ssse3 +AVX_INSTR packsswb, mmx, 0, 0, 0 +AVX_INSTR packssdw, mmx, 0, 0, 0 +AVX_INSTR packuswb, mmx, 0, 0, 0 +AVX_INSTR packusdw, sse4, 0, 0, 0 +AVX_INSTR paddb, mmx, 0, 0, 1 +AVX_INSTR paddw, mmx, 0, 0, 1 +AVX_INSTR paddd, mmx, 0, 0, 1 +AVX_INSTR paddq, sse2, 0, 0, 1 +AVX_INSTR paddsb, mmx, 0, 0, 1 +AVX_INSTR paddsw, mmx, 0, 0, 1 +AVX_INSTR paddusb, mmx, 0, 0, 1 +AVX_INSTR paddusw, mmx, 0, 0, 1 +AVX_INSTR palignr, ssse3 +AVX_INSTR pand, mmx, 0, 0, 1 +AVX_INSTR pandn, mmx, 0, 0, 0 +AVX_INSTR pavgb, mmx2, 0, 0, 1 +AVX_INSTR pavgw, mmx2, 0, 0, 1 +AVX_INSTR pblendvb, sse4, 0, 0, 0 +AVX_INSTR pblendw, sse4 +AVX_INSTR pclmulqdq +AVX_INSTR pcmpestri, sse42 +AVX_INSTR pcmpestrm, sse42 +AVX_INSTR pcmpistri, sse42 +AVX_INSTR pcmpistrm, sse42 +AVX_INSTR pcmpeqb, mmx, 0, 0, 1 +AVX_INSTR pcmpeqw, mmx, 0, 0, 1 +AVX_INSTR pcmpeqd, mmx, 0, 0, 1 +AVX_INSTR pcmpeqq, sse4, 0, 0, 1 +AVX_INSTR pcmpgtb, mmx, 0, 0, 0 +AVX_INSTR pcmpgtw, mmx, 0, 0, 0 +AVX_INSTR pcmpgtd, mmx, 0, 0, 0 +AVX_INSTR pcmpgtq, sse42, 0, 0, 0 +AVX_INSTR pextrb, sse4 +AVX_INSTR pextrd, sse4 +AVX_INSTR pextrq, sse4 +AVX_INSTR pextrw, mmx2 +AVX_INSTR phaddw, ssse3, 0, 0, 0 +AVX_INSTR phaddd, ssse3, 0, 0, 0 +AVX_INSTR phaddsw, ssse3, 0, 0, 0 +AVX_INSTR phminposuw, sse4 +AVX_INSTR phsubw, ssse3, 0, 0, 0 +AVX_INSTR phsubd, ssse3, 0, 0, 0 +AVX_INSTR phsubsw, ssse3, 0, 0, 0 +AVX_INSTR pinsrb, sse4 +AVX_INSTR pinsrd, sse4 +AVX_INSTR pinsrq, sse4 +AVX_INSTR pinsrw, mmx2 +AVX_INSTR pmaddwd, mmx, 0, 0, 1 +AVX_INSTR pmaddubsw, ssse3, 0, 0, 0 +AVX_INSTR pmaxsb, sse4, 0, 0, 1 +AVX_INSTR pmaxsw, mmx2, 0, 0, 1 +AVX_INSTR pmaxsd, sse4, 0, 0, 1 +AVX_INSTR pmaxub, mmx2, 0, 0, 1 +AVX_INSTR pmaxuw, sse4, 0, 0, 1 +AVX_INSTR pmaxud, sse4, 0, 0, 1 +AVX_INSTR pminsb, sse4, 0, 0, 1 +AVX_INSTR pminsw, mmx2, 0, 0, 1 +AVX_INSTR pminsd, sse4, 0, 0, 1 +AVX_INSTR pminub, mmx2, 0, 0, 1 +AVX_INSTR pminuw, sse4, 0, 0, 1 +AVX_INSTR pminud, sse4, 0, 0, 1 +AVX_INSTR pmovmskb, mmx2 +AVX_INSTR pmovsxbw, sse4 +AVX_INSTR pmovsxbd, sse4 +AVX_INSTR pmovsxbq, sse4 +AVX_INSTR pmovsxwd, sse4 +AVX_INSTR pmovsxwq, sse4 +AVX_INSTR pmovsxdq, sse4 +AVX_INSTR pmovzxbw, sse4 +AVX_INSTR pmovzxbd, sse4 +AVX_INSTR pmovzxbq, sse4 +AVX_INSTR pmovzxwd, sse4 +AVX_INSTR pmovzxwq, sse4 +AVX_INSTR pmovzxdq, sse4 +AVX_INSTR pmuldq, sse4, 0, 0, 1 +AVX_INSTR pmulhrsw, ssse3, 0, 0, 1 +AVX_INSTR pmulhuw, mmx2, 0, 0, 1 +AVX_INSTR pmulhw, mmx, 0, 0, 1 +AVX_INSTR pmullw, mmx, 0, 0, 1 +AVX_INSTR pmulld, sse4, 0, 0, 1 +AVX_INSTR pmuludq, sse2, 0, 0, 1 +AVX_INSTR por, mmx, 0, 0, 1 +AVX_INSTR psadbw, mmx2, 0, 0, 1 +AVX_INSTR pshufb, ssse3, 0, 0, 0 +AVX_INSTR pshufd, sse2 +AVX_INSTR pshufhw, sse2 +AVX_INSTR pshuflw, sse2 +AVX_INSTR psignb, ssse3, 0, 0, 0 +AVX_INSTR psignw, ssse3, 0, 0, 0 +AVX_INSTR psignd, ssse3, 0, 0, 0 +AVX_INSTR psllw, mmx, 0, 0, 0 +AVX_INSTR pslld, mmx, 0, 0, 0 +AVX_INSTR psllq, mmx, 0, 0, 0 +AVX_INSTR pslldq, sse2, 0, 0, 0 +AVX_INSTR psraw, mmx, 0, 0, 0 +AVX_INSTR psrad, mmx, 0, 0, 0 +AVX_INSTR psrlw, mmx, 0, 0, 0 +AVX_INSTR psrld, mmx, 0, 0, 0 +AVX_INSTR psrlq, mmx, 0, 0, 0 +AVX_INSTR psrldq, sse2, 0, 0, 0 +AVX_INSTR psubb, mmx, 0, 0, 0 +AVX_INSTR psubw, mmx, 0, 0, 0 +AVX_INSTR psubd, mmx, 0, 0, 0 +AVX_INSTR psubq, sse2, 0, 0, 0 +AVX_INSTR psubsb, mmx, 0, 0, 0 +AVX_INSTR psubsw, mmx, 0, 0, 0 +AVX_INSTR psubusb, mmx, 0, 0, 0 +AVX_INSTR psubusw, mmx, 0, 0, 0 +AVX_INSTR ptest, sse4 +AVX_INSTR punpckhbw, mmx, 0, 0, 0 +AVX_INSTR punpckhwd, mmx, 0, 0, 0 +AVX_INSTR punpckhdq, mmx, 0, 0, 0 +AVX_INSTR punpckhqdq, sse2, 0, 0, 0 +AVX_INSTR punpcklbw, mmx, 0, 0, 0 +AVX_INSTR punpcklwd, mmx, 0, 0, 0 +AVX_INSTR punpckldq, mmx, 0, 0, 0 +AVX_INSTR punpcklqdq, sse2, 0, 0, 0 +AVX_INSTR pxor, mmx, 0, 0, 1 +AVX_INSTR rcpps, sse, 1, 0, 0 +AVX_INSTR rcpss, sse, 1, 0, 0 +AVX_INSTR roundpd, sse4 +AVX_INSTR roundps, sse4 +AVX_INSTR roundsd, sse4 +AVX_INSTR roundss, sse4 +AVX_INSTR rsqrtps, sse, 1, 0, 0 +AVX_INSTR rsqrtss, sse, 1, 0, 0 +AVX_INSTR shufpd, sse2, 1, 1, 0 +AVX_INSTR shufps, sse, 1, 1, 0 +AVX_INSTR sqrtpd, sse2, 1, 0, 0 +AVX_INSTR sqrtps, sse, 1, 0, 0 +AVX_INSTR sqrtsd, sse2, 1, 0, 0 +AVX_INSTR sqrtss, sse, 1, 0, 0 +AVX_INSTR stmxcsr, sse +AVX_INSTR subpd, sse2, 1, 0, 0 +AVX_INSTR subps, sse, 1, 0, 0 +AVX_INSTR subsd, sse2, 1, 0, 0 +AVX_INSTR subss, sse, 1, 0, 0 +AVX_INSTR ucomisd, sse2 +AVX_INSTR ucomiss, sse +AVX_INSTR unpckhpd, sse2, 1, 0, 0 +AVX_INSTR unpckhps, sse, 1, 0, 0 +AVX_INSTR unpcklpd, sse2, 1, 0, 0 +AVX_INSTR unpcklps, sse, 1, 0, 0 +AVX_INSTR xorpd, sse2, 1, 0, 1 +AVX_INSTR xorps, sse, 1, 0, 1 + +; 3DNow instructions, for sharing code between AVX, SSE and 3DN +AVX_INSTR pfadd, 3dnow, 1, 0, 1 +AVX_INSTR pfsub, 3dnow, 1, 0, 0 +AVX_INSTR pfmul, 3dnow, 1, 0, 1 + +; base-4 constants for shuffles +%assign i 0 +%rep 256 + %assign j ((i>>6)&3)*1000 + ((i>>4)&3)*100 + ((i>>2)&3)*10 + (i&3) + %if j < 10 + CAT_XDEFINE q000, j, i + %elif j < 100 + CAT_XDEFINE q00, j, i + %elif j < 1000 + CAT_XDEFINE q0, j, i + %else + CAT_XDEFINE q, j, i + %endif + %assign i i+1 +%endrep +%undef i +%undef j + +%macro FMA_INSTR 3 + %macro %1 4-7 %1, %2, %3 + %if cpuflag(xop) + v%5 %1, %2, %3, %4 + %elifnidn %1, %4 + %6 %1, %2, %3 + %7 %1, %4 + %else + %error non-xop emulation of ``%5 %1, %2, %3, %4'' is not supported + %endif + %endmacro +%endmacro + +FMA_INSTR pmacsww, pmullw, paddw +FMA_INSTR pmacsdd, pmulld, paddd ; sse4 emulation +FMA_INSTR pmacsdql, pmuldq, paddq ; sse4 emulation +FMA_INSTR pmadcswd, pmaddwd, paddd + +; Macros for consolidating FMA3 and FMA4 using 4-operand (dst, src1, src2, src3) syntax. +; FMA3 is only possible if dst is the same as one of the src registers. +; Either src2 or src3 can be a memory operand. +%macro FMA4_INSTR 2-* + %push fma4_instr + %xdefine %$prefix %1 + %rep %0 - 1 + %macro %$prefix%2 4-6 %$prefix, %2 + %if notcpuflag(fma3) && notcpuflag(fma4) + %error use of ``%5%6'' fma instruction in cpuname function: current_function + %elif cpuflag(fma4) + v%5%6 %1, %2, %3, %4 + %elifidn %1, %2 + ; If %3 or %4 is a memory operand it needs to be encoded as the last operand. + %ifid %3 + v%{5}213%6 %2, %3, %4 + %else + v%{5}132%6 %2, %4, %3 + %endif + %elifidn %1, %3 + v%{5}213%6 %3, %2, %4 + %elifidn %1, %4 + v%{5}231%6 %4, %2, %3 + %else + %error fma3 emulation of ``%5%6 %1, %2, %3, %4'' is not supported + %endif + %endmacro + %rotate 1 + %endrep + %pop +%endmacro + +FMA4_INSTR fmadd, pd, ps, sd, ss +FMA4_INSTR fmaddsub, pd, ps +FMA4_INSTR fmsub, pd, ps, sd, ss +FMA4_INSTR fmsubadd, pd, ps +FMA4_INSTR fnmadd, pd, ps, sd, ss +FMA4_INSTR fnmsub, pd, ps, sd, ss + +; workaround: vpbroadcastq is broken in x86_32 due to a yasm bug (fixed in 1.3.0) +%ifdef __YASM_VER__ + %if __YASM_VERSION_ID__ < 0x01030000 && ARCH_X86_64 == 0 + %macro vpbroadcastq 2 + %if sizeof%1 == 16 + movddup %1, %2 + %else + vbroadcastsd %1, %2 + %endif + %endmacro + %endif +%endif diff --git a/thirdparty/libvpx/vp8/common/alloccommon.c b/thirdparty/libvpx/vp8/common/alloccommon.c new file mode 100644 index 0000000000..8dfd4ce203 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/alloccommon.c @@ -0,0 +1,190 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "vpx_config.h" +#include "alloccommon.h" +#include "blockd.h" +#include "vpx_mem/vpx_mem.h" +#include "onyxc_int.h" +#include "findnearmv.h" +#include "entropymode.h" +#include "systemdependent.h" + +void vp8_de_alloc_frame_buffers(VP8_COMMON *oci) +{ + int i; + for (i = 0; i < NUM_YV12_BUFFERS; i++) + vp8_yv12_de_alloc_frame_buffer(&oci->yv12_fb[i]); + + vp8_yv12_de_alloc_frame_buffer(&oci->temp_scale_frame); +#if CONFIG_POSTPROC + vp8_yv12_de_alloc_frame_buffer(&oci->post_proc_buffer); + if (oci->post_proc_buffer_int_used) + vp8_yv12_de_alloc_frame_buffer(&oci->post_proc_buffer_int); + + vpx_free(oci->pp_limits_buffer); + oci->pp_limits_buffer = NULL; +#endif + + vpx_free(oci->above_context); + vpx_free(oci->mip); +#if CONFIG_ERROR_CONCEALMENT + vpx_free(oci->prev_mip); + oci->prev_mip = NULL; +#endif + + oci->above_context = NULL; + oci->mip = NULL; +} + +int vp8_alloc_frame_buffers(VP8_COMMON *oci, int width, int height) +{ + int i; + + vp8_de_alloc_frame_buffers(oci); + + /* our internal buffers are always multiples of 16 */ + if ((width & 0xf) != 0) + width += 16 - (width & 0xf); + + if ((height & 0xf) != 0) + height += 16 - (height & 0xf); + + + for (i = 0; i < NUM_YV12_BUFFERS; i++) + { + oci->fb_idx_ref_cnt[i] = 0; + oci->yv12_fb[i].flags = 0; + if (vp8_yv12_alloc_frame_buffer(&oci->yv12_fb[i], width, height, VP8BORDERINPIXELS) < 0) + goto allocation_fail; + } + + oci->new_fb_idx = 0; + oci->lst_fb_idx = 1; + oci->gld_fb_idx = 2; + oci->alt_fb_idx = 3; + + oci->fb_idx_ref_cnt[0] = 1; + oci->fb_idx_ref_cnt[1] = 1; + oci->fb_idx_ref_cnt[2] = 1; + oci->fb_idx_ref_cnt[3] = 1; + + if (vp8_yv12_alloc_frame_buffer(&oci->temp_scale_frame, width, 16, VP8BORDERINPIXELS) < 0) + goto allocation_fail; + + oci->mb_rows = height >> 4; + oci->mb_cols = width >> 4; + oci->MBs = oci->mb_rows * oci->mb_cols; + oci->mode_info_stride = oci->mb_cols + 1; + oci->mip = vpx_calloc((oci->mb_cols + 1) * (oci->mb_rows + 1), sizeof(MODE_INFO)); + + if (!oci->mip) + goto allocation_fail; + + oci->mi = oci->mip + oci->mode_info_stride + 1; + + /* Allocation of previous mode info will be done in vp8_decode_frame() + * as it is a decoder only data */ + + oci->above_context = vpx_calloc(sizeof(ENTROPY_CONTEXT_PLANES) * oci->mb_cols, 1); + + if (!oci->above_context) + goto allocation_fail; + +#if CONFIG_POSTPROC + if (vp8_yv12_alloc_frame_buffer(&oci->post_proc_buffer, width, height, VP8BORDERINPIXELS) < 0) + goto allocation_fail; + + oci->post_proc_buffer_int_used = 0; + memset(&oci->postproc_state, 0, sizeof(oci->postproc_state)); + memset(oci->post_proc_buffer.buffer_alloc, 128, + oci->post_proc_buffer.frame_size); + + /* Allocate buffer to store post-processing filter coefficients. + * + * Note: Round up mb_cols to support SIMD reads + */ + oci->pp_limits_buffer = vpx_memalign(16, 24 * ((oci->mb_cols + 1) & ~1)); + if (!oci->pp_limits_buffer) + goto allocation_fail; +#endif + + return 0; + +allocation_fail: + vp8_de_alloc_frame_buffers(oci); + return 1; +} + +void vp8_setup_version(VP8_COMMON *cm) +{ + switch (cm->version) + { + case 0: + cm->no_lpf = 0; + cm->filter_type = NORMAL_LOOPFILTER; + cm->use_bilinear_mc_filter = 0; + cm->full_pixel = 0; + break; + case 1: + cm->no_lpf = 0; + cm->filter_type = SIMPLE_LOOPFILTER; + cm->use_bilinear_mc_filter = 1; + cm->full_pixel = 0; + break; + case 2: + cm->no_lpf = 1; + cm->filter_type = NORMAL_LOOPFILTER; + cm->use_bilinear_mc_filter = 1; + cm->full_pixel = 0; + break; + case 3: + cm->no_lpf = 1; + cm->filter_type = SIMPLE_LOOPFILTER; + cm->use_bilinear_mc_filter = 1; + cm->full_pixel = 1; + break; + default: + /*4,5,6,7 are reserved for future use*/ + cm->no_lpf = 0; + cm->filter_type = NORMAL_LOOPFILTER; + cm->use_bilinear_mc_filter = 0; + cm->full_pixel = 0; + break; + } +} +void vp8_create_common(VP8_COMMON *oci) +{ + vp8_machine_specific_config(oci); + + vp8_init_mbmode_probs(oci); + vp8_default_bmode_probs(oci->fc.bmode_prob); + + oci->mb_no_coeff_skip = 1; + oci->no_lpf = 0; + oci->filter_type = NORMAL_LOOPFILTER; + oci->use_bilinear_mc_filter = 0; + oci->full_pixel = 0; + oci->multi_token_partition = ONE_PARTITION; + oci->clamp_type = RECON_CLAMP_REQUIRED; + + /* Initialize reference frame sign bias structure to defaults */ + memset(oci->ref_frame_sign_bias, 0, sizeof(oci->ref_frame_sign_bias)); + + /* Default disable buffer to buffer copying */ + oci->copy_buffer_to_gf = 0; + oci->copy_buffer_to_arf = 0; +} + +void vp8_remove_common(VP8_COMMON *oci) +{ + vp8_de_alloc_frame_buffers(oci); +} diff --git a/thirdparty/libvpx/vp8/common/alloccommon.h b/thirdparty/libvpx/vp8/common/alloccommon.h new file mode 100644 index 0000000000..93e99d76b1 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/alloccommon.h @@ -0,0 +1,31 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_ALLOCCOMMON_H_ +#define VP8_COMMON_ALLOCCOMMON_H_ + +#include "onyxc_int.h" + +#ifdef __cplusplus +extern "C" { +#endif + +void vp8_create_common(VP8_COMMON *oci); +void vp8_remove_common(VP8_COMMON *oci); +void vp8_de_alloc_frame_buffers(VP8_COMMON *oci); +int vp8_alloc_frame_buffers(VP8_COMMON *oci, int width, int height); +void vp8_setup_version(VP8_COMMON *oci); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_ALLOCCOMMON_H_ diff --git a/thirdparty/libvpx/vp8/common/arm/loopfilter_arm.c b/thirdparty/libvpx/vp8/common/arm/loopfilter_arm.c new file mode 100644 index 0000000000..5840c2bbaa --- /dev/null +++ b/thirdparty/libvpx/vp8/common/arm/loopfilter_arm.c @@ -0,0 +1,181 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "vpx_config.h" +#include "vp8_rtcd.h" +#include "vp8/common/loopfilter.h" +#include "vp8/common/onyxc_int.h" + +#define prototype_loopfilter(sym) \ + void sym(unsigned char *src, int pitch, const unsigned char *blimit,\ + const unsigned char *limit, const unsigned char *thresh, int count) + +#if HAVE_MEDIA +extern prototype_loopfilter(vp8_loop_filter_horizontal_edge_armv6); +extern prototype_loopfilter(vp8_loop_filter_vertical_edge_armv6); +extern prototype_loopfilter(vp8_mbloop_filter_horizontal_edge_armv6); +extern prototype_loopfilter(vp8_mbloop_filter_vertical_edge_armv6); +#endif + +#if HAVE_NEON +typedef void loopfilter_y_neon(unsigned char *src, int pitch, + unsigned char blimit, unsigned char limit, unsigned char thresh); +typedef void loopfilter_uv_neon(unsigned char *u, int pitch, + unsigned char blimit, unsigned char limit, unsigned char thresh, + unsigned char *v); + +extern loopfilter_y_neon vp8_loop_filter_horizontal_edge_y_neon; +extern loopfilter_y_neon vp8_loop_filter_vertical_edge_y_neon; +extern loopfilter_uv_neon vp8_loop_filter_horizontal_edge_uv_neon; +extern loopfilter_uv_neon vp8_loop_filter_vertical_edge_uv_neon; + +extern loopfilter_y_neon vp8_mbloop_filter_horizontal_edge_y_neon; +extern loopfilter_y_neon vp8_mbloop_filter_vertical_edge_y_neon; +extern loopfilter_uv_neon vp8_mbloop_filter_horizontal_edge_uv_neon; +extern loopfilter_uv_neon vp8_mbloop_filter_vertical_edge_uv_neon; +#endif + +#if HAVE_MEDIA +/* ARMV6/MEDIA loopfilter functions*/ +/* Horizontal MB filtering */ +void vp8_loop_filter_mbh_armv6(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr, + int y_stride, int uv_stride, loop_filter_info *lfi) +{ + vp8_mbloop_filter_horizontal_edge_armv6(y_ptr, y_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 2); + + if (u_ptr) + vp8_mbloop_filter_horizontal_edge_armv6(u_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 1); + + if (v_ptr) + vp8_mbloop_filter_horizontal_edge_armv6(v_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 1); +} + +/* Vertical MB Filtering */ +void vp8_loop_filter_mbv_armv6(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr, + int y_stride, int uv_stride, loop_filter_info *lfi) +{ + vp8_mbloop_filter_vertical_edge_armv6(y_ptr, y_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 2); + + if (u_ptr) + vp8_mbloop_filter_vertical_edge_armv6(u_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 1); + + if (v_ptr) + vp8_mbloop_filter_vertical_edge_armv6(v_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 1); +} + +/* Horizontal B Filtering */ +void vp8_loop_filter_bh_armv6(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr, + int y_stride, int uv_stride, loop_filter_info *lfi) +{ + vp8_loop_filter_horizontal_edge_armv6(y_ptr + 4 * y_stride, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2); + vp8_loop_filter_horizontal_edge_armv6(y_ptr + 8 * y_stride, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2); + vp8_loop_filter_horizontal_edge_armv6(y_ptr + 12 * y_stride, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2); + + if (u_ptr) + vp8_loop_filter_horizontal_edge_armv6(u_ptr + 4 * uv_stride, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, 1); + + if (v_ptr) + vp8_loop_filter_horizontal_edge_armv6(v_ptr + 4 * uv_stride, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, 1); +} + +void vp8_loop_filter_bhs_armv6(unsigned char *y_ptr, int y_stride, + const unsigned char *blimit) +{ + vp8_loop_filter_simple_horizontal_edge_armv6(y_ptr + 4 * y_stride, y_stride, blimit); + vp8_loop_filter_simple_horizontal_edge_armv6(y_ptr + 8 * y_stride, y_stride, blimit); + vp8_loop_filter_simple_horizontal_edge_armv6(y_ptr + 12 * y_stride, y_stride, blimit); +} + +/* Vertical B Filtering */ +void vp8_loop_filter_bv_armv6(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr, + int y_stride, int uv_stride, loop_filter_info *lfi) +{ + vp8_loop_filter_vertical_edge_armv6(y_ptr + 4, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2); + vp8_loop_filter_vertical_edge_armv6(y_ptr + 8, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2); + vp8_loop_filter_vertical_edge_armv6(y_ptr + 12, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2); + + if (u_ptr) + vp8_loop_filter_vertical_edge_armv6(u_ptr + 4, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, 1); + + if (v_ptr) + vp8_loop_filter_vertical_edge_armv6(v_ptr + 4, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, 1); +} + +void vp8_loop_filter_bvs_armv6(unsigned char *y_ptr, int y_stride, + const unsigned char *blimit) +{ + vp8_loop_filter_simple_vertical_edge_armv6(y_ptr + 4, y_stride, blimit); + vp8_loop_filter_simple_vertical_edge_armv6(y_ptr + 8, y_stride, blimit); + vp8_loop_filter_simple_vertical_edge_armv6(y_ptr + 12, y_stride, blimit); +} +#endif + +#if HAVE_NEON +/* NEON loopfilter functions */ +/* Horizontal MB filtering */ +void vp8_loop_filter_mbh_neon(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr, + int y_stride, int uv_stride, loop_filter_info *lfi) +{ + unsigned char mblim = *lfi->mblim; + unsigned char lim = *lfi->lim; + unsigned char hev_thr = *lfi->hev_thr; + vp8_mbloop_filter_horizontal_edge_y_neon(y_ptr, y_stride, mblim, lim, hev_thr); + + if (u_ptr) + vp8_mbloop_filter_horizontal_edge_uv_neon(u_ptr, uv_stride, mblim, lim, hev_thr, v_ptr); +} + +/* Vertical MB Filtering */ +void vp8_loop_filter_mbv_neon(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr, + int y_stride, int uv_stride, loop_filter_info *lfi) +{ + unsigned char mblim = *lfi->mblim; + unsigned char lim = *lfi->lim; + unsigned char hev_thr = *lfi->hev_thr; + + vp8_mbloop_filter_vertical_edge_y_neon(y_ptr, y_stride, mblim, lim, hev_thr); + + if (u_ptr) + vp8_mbloop_filter_vertical_edge_uv_neon(u_ptr, uv_stride, mblim, lim, hev_thr, v_ptr); +} + +/* Horizontal B Filtering */ +void vp8_loop_filter_bh_neon(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr, + int y_stride, int uv_stride, loop_filter_info *lfi) +{ + unsigned char blim = *lfi->blim; + unsigned char lim = *lfi->lim; + unsigned char hev_thr = *lfi->hev_thr; + + vp8_loop_filter_horizontal_edge_y_neon(y_ptr + 4 * y_stride, y_stride, blim, lim, hev_thr); + vp8_loop_filter_horizontal_edge_y_neon(y_ptr + 8 * y_stride, y_stride, blim, lim, hev_thr); + vp8_loop_filter_horizontal_edge_y_neon(y_ptr + 12 * y_stride, y_stride, blim, lim, hev_thr); + + if (u_ptr) + vp8_loop_filter_horizontal_edge_uv_neon(u_ptr + 4 * uv_stride, uv_stride, blim, lim, hev_thr, v_ptr + 4 * uv_stride); +} + +/* Vertical B Filtering */ +void vp8_loop_filter_bv_neon(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr, + int y_stride, int uv_stride, loop_filter_info *lfi) +{ + unsigned char blim = *lfi->blim; + unsigned char lim = *lfi->lim; + unsigned char hev_thr = *lfi->hev_thr; + + vp8_loop_filter_vertical_edge_y_neon(y_ptr + 4, y_stride, blim, lim, hev_thr); + vp8_loop_filter_vertical_edge_y_neon(y_ptr + 8, y_stride, blim, lim, hev_thr); + vp8_loop_filter_vertical_edge_y_neon(y_ptr + 12, y_stride, blim, lim, hev_thr); + + if (u_ptr) + vp8_loop_filter_vertical_edge_uv_neon(u_ptr + 4, uv_stride, blim, lim, hev_thr, v_ptr + 4); +} +#endif diff --git a/thirdparty/libvpx/vp8/common/arm/neon/bilinearpredict_neon.c b/thirdparty/libvpx/vp8/common/arm/neon/bilinearpredict_neon.c new file mode 100644 index 0000000000..bb6ea76ba4 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/arm/neon/bilinearpredict_neon.c @@ -0,0 +1,591 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +static const uint8_t bifilter4_coeff[8][2] = { + {128, 0}, + {112, 16}, + { 96, 32}, + { 80, 48}, + { 64, 64}, + { 48, 80}, + { 32, 96}, + { 16, 112} +}; + +void vp8_bilinear_predict8x4_neon( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch) { + uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8; + uint8x8_t d7u8, d9u8, d11u8, d22u8, d23u8, d24u8, d25u8, d26u8; + uint8x16_t q1u8, q2u8, q3u8, q4u8, q5u8; + uint16x8_t q1u16, q2u16, q3u16, q4u16; + uint16x8_t q6u16, q7u16, q8u16, q9u16, q10u16; + + if (xoffset == 0) { // skip_1stpass_filter + d22u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d23u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d24u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d25u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d26u8 = vld1_u8(src_ptr); + } else { + q1u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q2u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q3u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q4u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q5u8 = vld1q_u8(src_ptr); + + d0u8 = vdup_n_u8(bifilter4_coeff[xoffset][0]); + d1u8 = vdup_n_u8(bifilter4_coeff[xoffset][1]); + + q6u16 = vmull_u8(vget_low_u8(q1u8), d0u8); + q7u16 = vmull_u8(vget_low_u8(q2u8), d0u8); + q8u16 = vmull_u8(vget_low_u8(q3u8), d0u8); + q9u16 = vmull_u8(vget_low_u8(q4u8), d0u8); + q10u16 = vmull_u8(vget_low_u8(q5u8), d0u8); + + d3u8 = vext_u8(vget_low_u8(q1u8), vget_high_u8(q1u8), 1); + d5u8 = vext_u8(vget_low_u8(q2u8), vget_high_u8(q2u8), 1); + d7u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 1); + d9u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 1); + d11u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 1); + + q6u16 = vmlal_u8(q6u16, d3u8, d1u8); + q7u16 = vmlal_u8(q7u16, d5u8, d1u8); + q8u16 = vmlal_u8(q8u16, d7u8, d1u8); + q9u16 = vmlal_u8(q9u16, d9u8, d1u8); + q10u16 = vmlal_u8(q10u16, d11u8, d1u8); + + d22u8 = vqrshrn_n_u16(q6u16, 7); + d23u8 = vqrshrn_n_u16(q7u16, 7); + d24u8 = vqrshrn_n_u16(q8u16, 7); + d25u8 = vqrshrn_n_u16(q9u16, 7); + d26u8 = vqrshrn_n_u16(q10u16, 7); + } + + // secondpass_filter + if (yoffset == 0) { // skip_2ndpass_filter + vst1_u8((uint8_t *)dst_ptr, d22u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d23u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d24u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d25u8); + } else { + d0u8 = vdup_n_u8(bifilter4_coeff[yoffset][0]); + d1u8 = vdup_n_u8(bifilter4_coeff[yoffset][1]); + + q1u16 = vmull_u8(d22u8, d0u8); + q2u16 = vmull_u8(d23u8, d0u8); + q3u16 = vmull_u8(d24u8, d0u8); + q4u16 = vmull_u8(d25u8, d0u8); + + q1u16 = vmlal_u8(q1u16, d23u8, d1u8); + q2u16 = vmlal_u8(q2u16, d24u8, d1u8); + q3u16 = vmlal_u8(q3u16, d25u8, d1u8); + q4u16 = vmlal_u8(q4u16, d26u8, d1u8); + + d2u8 = vqrshrn_n_u16(q1u16, 7); + d3u8 = vqrshrn_n_u16(q2u16, 7); + d4u8 = vqrshrn_n_u16(q3u16, 7); + d5u8 = vqrshrn_n_u16(q4u16, 7); + + vst1_u8((uint8_t *)dst_ptr, d2u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d3u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d4u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d5u8); + } + return; +} + +void vp8_bilinear_predict8x8_neon( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch) { + uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8, d7u8, d8u8, d9u8, d11u8; + uint8x8_t d22u8, d23u8, d24u8, d25u8, d26u8, d27u8, d28u8, d29u8, d30u8; + uint8x16_t q1u8, q2u8, q3u8, q4u8, q5u8; + uint16x8_t q1u16, q2u16, q3u16, q4u16, q5u16; + uint16x8_t q6u16, q7u16, q8u16, q9u16, q10u16; + + if (xoffset == 0) { // skip_1stpass_filter + d22u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d23u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d24u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d25u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d26u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d27u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d28u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d29u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d30u8 = vld1_u8(src_ptr); + } else { + q1u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q2u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q3u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q4u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + + d0u8 = vdup_n_u8(bifilter4_coeff[xoffset][0]); + d1u8 = vdup_n_u8(bifilter4_coeff[xoffset][1]); + + q6u16 = vmull_u8(vget_low_u8(q1u8), d0u8); + q7u16 = vmull_u8(vget_low_u8(q2u8), d0u8); + q8u16 = vmull_u8(vget_low_u8(q3u8), d0u8); + q9u16 = vmull_u8(vget_low_u8(q4u8), d0u8); + + d3u8 = vext_u8(vget_low_u8(q1u8), vget_high_u8(q1u8), 1); + d5u8 = vext_u8(vget_low_u8(q2u8), vget_high_u8(q2u8), 1); + d7u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 1); + d9u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 1); + + q6u16 = vmlal_u8(q6u16, d3u8, d1u8); + q7u16 = vmlal_u8(q7u16, d5u8, d1u8); + q8u16 = vmlal_u8(q8u16, d7u8, d1u8); + q9u16 = vmlal_u8(q9u16, d9u8, d1u8); + + d22u8 = vqrshrn_n_u16(q6u16, 7); + d23u8 = vqrshrn_n_u16(q7u16, 7); + d24u8 = vqrshrn_n_u16(q8u16, 7); + d25u8 = vqrshrn_n_u16(q9u16, 7); + + // first_pass filtering on the rest 5-line data + q1u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q2u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q3u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q4u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q5u8 = vld1q_u8(src_ptr); + + q6u16 = vmull_u8(vget_low_u8(q1u8), d0u8); + q7u16 = vmull_u8(vget_low_u8(q2u8), d0u8); + q8u16 = vmull_u8(vget_low_u8(q3u8), d0u8); + q9u16 = vmull_u8(vget_low_u8(q4u8), d0u8); + q10u16 = vmull_u8(vget_low_u8(q5u8), d0u8); + + d3u8 = vext_u8(vget_low_u8(q1u8), vget_high_u8(q1u8), 1); + d5u8 = vext_u8(vget_low_u8(q2u8), vget_high_u8(q2u8), 1); + d7u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 1); + d9u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 1); + d11u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 1); + + q6u16 = vmlal_u8(q6u16, d3u8, d1u8); + q7u16 = vmlal_u8(q7u16, d5u8, d1u8); + q8u16 = vmlal_u8(q8u16, d7u8, d1u8); + q9u16 = vmlal_u8(q9u16, d9u8, d1u8); + q10u16 = vmlal_u8(q10u16, d11u8, d1u8); + + d26u8 = vqrshrn_n_u16(q6u16, 7); + d27u8 = vqrshrn_n_u16(q7u16, 7); + d28u8 = vqrshrn_n_u16(q8u16, 7); + d29u8 = vqrshrn_n_u16(q9u16, 7); + d30u8 = vqrshrn_n_u16(q10u16, 7); + } + + // secondpass_filter + if (yoffset == 0) { // skip_2ndpass_filter + vst1_u8((uint8_t *)dst_ptr, d22u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d23u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d24u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d25u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d26u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d27u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d28u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d29u8); + } else { + d0u8 = vdup_n_u8(bifilter4_coeff[yoffset][0]); + d1u8 = vdup_n_u8(bifilter4_coeff[yoffset][1]); + + q1u16 = vmull_u8(d22u8, d0u8); + q2u16 = vmull_u8(d23u8, d0u8); + q3u16 = vmull_u8(d24u8, d0u8); + q4u16 = vmull_u8(d25u8, d0u8); + q5u16 = vmull_u8(d26u8, d0u8); + q6u16 = vmull_u8(d27u8, d0u8); + q7u16 = vmull_u8(d28u8, d0u8); + q8u16 = vmull_u8(d29u8, d0u8); + + q1u16 = vmlal_u8(q1u16, d23u8, d1u8); + q2u16 = vmlal_u8(q2u16, d24u8, d1u8); + q3u16 = vmlal_u8(q3u16, d25u8, d1u8); + q4u16 = vmlal_u8(q4u16, d26u8, d1u8); + q5u16 = vmlal_u8(q5u16, d27u8, d1u8); + q6u16 = vmlal_u8(q6u16, d28u8, d1u8); + q7u16 = vmlal_u8(q7u16, d29u8, d1u8); + q8u16 = vmlal_u8(q8u16, d30u8, d1u8); + + d2u8 = vqrshrn_n_u16(q1u16, 7); + d3u8 = vqrshrn_n_u16(q2u16, 7); + d4u8 = vqrshrn_n_u16(q3u16, 7); + d5u8 = vqrshrn_n_u16(q4u16, 7); + d6u8 = vqrshrn_n_u16(q5u16, 7); + d7u8 = vqrshrn_n_u16(q6u16, 7); + d8u8 = vqrshrn_n_u16(q7u16, 7); + d9u8 = vqrshrn_n_u16(q8u16, 7); + + vst1_u8((uint8_t *)dst_ptr, d2u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d3u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d4u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d5u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d6u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d7u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d8u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d9u8); + } + return; +} + +void vp8_bilinear_predict16x16_neon( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch) { + int i; + unsigned char tmp[272]; + unsigned char *tmpp; + uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8, d7u8, d8u8, d9u8; + uint8x8_t d10u8, d11u8, d12u8, d13u8, d14u8, d15u8, d16u8, d17u8, d18u8; + uint8x8_t d19u8, d20u8, d21u8; + uint8x16_t q1u8, q2u8, q3u8, q4u8, q5u8, q6u8, q7u8, q8u8, q9u8, q10u8; + uint8x16_t q11u8, q12u8, q13u8, q14u8, q15u8; + uint16x8_t q1u16, q2u16, q3u16, q4u16, q5u16, q6u16, q7u16, q8u16; + uint16x8_t q9u16, q10u16, q11u16, q12u16, q13u16, q14u16; + + if (xoffset == 0) { // secondpass_bfilter16x16_only + d0u8 = vdup_n_u8(bifilter4_coeff[yoffset][0]); + d1u8 = vdup_n_u8(bifilter4_coeff[yoffset][1]); + + q11u8 = vld1q_u8(src_ptr); + src_ptr += src_pixels_per_line; + for (i = 4; i > 0; i--) { + q12u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q13u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q14u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q15u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + + q1u16 = vmull_u8(vget_low_u8(q11u8), d0u8); + q2u16 = vmull_u8(vget_high_u8(q11u8), d0u8); + q3u16 = vmull_u8(vget_low_u8(q12u8), d0u8); + q4u16 = vmull_u8(vget_high_u8(q12u8), d0u8); + q5u16 = vmull_u8(vget_low_u8(q13u8), d0u8); + q6u16 = vmull_u8(vget_high_u8(q13u8), d0u8); + q7u16 = vmull_u8(vget_low_u8(q14u8), d0u8); + q8u16 = vmull_u8(vget_high_u8(q14u8), d0u8); + + q1u16 = vmlal_u8(q1u16, vget_low_u8(q12u8), d1u8); + q2u16 = vmlal_u8(q2u16, vget_high_u8(q12u8), d1u8); + q3u16 = vmlal_u8(q3u16, vget_low_u8(q13u8), d1u8); + q4u16 = vmlal_u8(q4u16, vget_high_u8(q13u8), d1u8); + q5u16 = vmlal_u8(q5u16, vget_low_u8(q14u8), d1u8); + q6u16 = vmlal_u8(q6u16, vget_high_u8(q14u8), d1u8); + q7u16 = vmlal_u8(q7u16, vget_low_u8(q15u8), d1u8); + q8u16 = vmlal_u8(q8u16, vget_high_u8(q15u8), d1u8); + + d2u8 = vqrshrn_n_u16(q1u16, 7); + d3u8 = vqrshrn_n_u16(q2u16, 7); + d4u8 = vqrshrn_n_u16(q3u16, 7); + d5u8 = vqrshrn_n_u16(q4u16, 7); + d6u8 = vqrshrn_n_u16(q5u16, 7); + d7u8 = vqrshrn_n_u16(q6u16, 7); + d8u8 = vqrshrn_n_u16(q7u16, 7); + d9u8 = vqrshrn_n_u16(q8u16, 7); + + q1u8 = vcombine_u8(d2u8, d3u8); + q2u8 = vcombine_u8(d4u8, d5u8); + q3u8 = vcombine_u8(d6u8, d7u8); + q4u8 = vcombine_u8(d8u8, d9u8); + + q11u8 = q15u8; + + vst1q_u8((uint8_t *)dst_ptr, q1u8); dst_ptr += dst_pitch; + vst1q_u8((uint8_t *)dst_ptr, q2u8); dst_ptr += dst_pitch; + vst1q_u8((uint8_t *)dst_ptr, q3u8); dst_ptr += dst_pitch; + vst1q_u8((uint8_t *)dst_ptr, q4u8); dst_ptr += dst_pitch; + } + return; + } + + if (yoffset == 0) { // firstpass_bfilter16x16_only + d0u8 = vdup_n_u8(bifilter4_coeff[xoffset][0]); + d1u8 = vdup_n_u8(bifilter4_coeff[xoffset][1]); + + for (i = 4; i > 0 ; i--) { + d2u8 = vld1_u8(src_ptr); + d3u8 = vld1_u8(src_ptr + 8); + d4u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + d5u8 = vld1_u8(src_ptr); + d6u8 = vld1_u8(src_ptr + 8); + d7u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + d8u8 = vld1_u8(src_ptr); + d9u8 = vld1_u8(src_ptr + 8); + d10u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + d11u8 = vld1_u8(src_ptr); + d12u8 = vld1_u8(src_ptr + 8); + d13u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + + q7u16 = vmull_u8(d2u8, d0u8); + q8u16 = vmull_u8(d3u8, d0u8); + q9u16 = vmull_u8(d5u8, d0u8); + q10u16 = vmull_u8(d6u8, d0u8); + q11u16 = vmull_u8(d8u8, d0u8); + q12u16 = vmull_u8(d9u8, d0u8); + q13u16 = vmull_u8(d11u8, d0u8); + q14u16 = vmull_u8(d12u8, d0u8); + + d2u8 = vext_u8(d2u8, d3u8, 1); + d5u8 = vext_u8(d5u8, d6u8, 1); + d8u8 = vext_u8(d8u8, d9u8, 1); + d11u8 = vext_u8(d11u8, d12u8, 1); + + q7u16 = vmlal_u8(q7u16, d2u8, d1u8); + q9u16 = vmlal_u8(q9u16, d5u8, d1u8); + q11u16 = vmlal_u8(q11u16, d8u8, d1u8); + q13u16 = vmlal_u8(q13u16, d11u8, d1u8); + + d3u8 = vext_u8(d3u8, d4u8, 1); + d6u8 = vext_u8(d6u8, d7u8, 1); + d9u8 = vext_u8(d9u8, d10u8, 1); + d12u8 = vext_u8(d12u8, d13u8, 1); + + q8u16 = vmlal_u8(q8u16, d3u8, d1u8); + q10u16 = vmlal_u8(q10u16, d6u8, d1u8); + q12u16 = vmlal_u8(q12u16, d9u8, d1u8); + q14u16 = vmlal_u8(q14u16, d12u8, d1u8); + + d14u8 = vqrshrn_n_u16(q7u16, 7); + d15u8 = vqrshrn_n_u16(q8u16, 7); + d16u8 = vqrshrn_n_u16(q9u16, 7); + d17u8 = vqrshrn_n_u16(q10u16, 7); + d18u8 = vqrshrn_n_u16(q11u16, 7); + d19u8 = vqrshrn_n_u16(q12u16, 7); + d20u8 = vqrshrn_n_u16(q13u16, 7); + d21u8 = vqrshrn_n_u16(q14u16, 7); + + q7u8 = vcombine_u8(d14u8, d15u8); + q8u8 = vcombine_u8(d16u8, d17u8); + q9u8 = vcombine_u8(d18u8, d19u8); + q10u8 =vcombine_u8(d20u8, d21u8); + + vst1q_u8((uint8_t *)dst_ptr, q7u8); dst_ptr += dst_pitch; + vst1q_u8((uint8_t *)dst_ptr, q8u8); dst_ptr += dst_pitch; + vst1q_u8((uint8_t *)dst_ptr, q9u8); dst_ptr += dst_pitch; + vst1q_u8((uint8_t *)dst_ptr, q10u8); dst_ptr += dst_pitch; + } + return; + } + + d0u8 = vdup_n_u8(bifilter4_coeff[xoffset][0]); + d1u8 = vdup_n_u8(bifilter4_coeff[xoffset][1]); + + d2u8 = vld1_u8(src_ptr); + d3u8 = vld1_u8(src_ptr + 8); + d4u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + d5u8 = vld1_u8(src_ptr); + d6u8 = vld1_u8(src_ptr + 8); + d7u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + d8u8 = vld1_u8(src_ptr); + d9u8 = vld1_u8(src_ptr + 8); + d10u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + d11u8 = vld1_u8(src_ptr); + d12u8 = vld1_u8(src_ptr + 8); + d13u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + + // First Pass: output_height lines x output_width columns (17x16) + tmpp = tmp; + for (i = 3; i > 0; i--) { + q7u16 = vmull_u8(d2u8, d0u8); + q8u16 = vmull_u8(d3u8, d0u8); + q9u16 = vmull_u8(d5u8, d0u8); + q10u16 = vmull_u8(d6u8, d0u8); + q11u16 = vmull_u8(d8u8, d0u8); + q12u16 = vmull_u8(d9u8, d0u8); + q13u16 = vmull_u8(d11u8, d0u8); + q14u16 = vmull_u8(d12u8, d0u8); + + d2u8 = vext_u8(d2u8, d3u8, 1); + d5u8 = vext_u8(d5u8, d6u8, 1); + d8u8 = vext_u8(d8u8, d9u8, 1); + d11u8 = vext_u8(d11u8, d12u8, 1); + + q7u16 = vmlal_u8(q7u16, d2u8, d1u8); + q9u16 = vmlal_u8(q9u16, d5u8, d1u8); + q11u16 = vmlal_u8(q11u16, d8u8, d1u8); + q13u16 = vmlal_u8(q13u16, d11u8, d1u8); + + d3u8 = vext_u8(d3u8, d4u8, 1); + d6u8 = vext_u8(d6u8, d7u8, 1); + d9u8 = vext_u8(d9u8, d10u8, 1); + d12u8 = vext_u8(d12u8, d13u8, 1); + + q8u16 = vmlal_u8(q8u16, d3u8, d1u8); + q10u16 = vmlal_u8(q10u16, d6u8, d1u8); + q12u16 = vmlal_u8(q12u16, d9u8, d1u8); + q14u16 = vmlal_u8(q14u16, d12u8, d1u8); + + d14u8 = vqrshrn_n_u16(q7u16, 7); + d15u8 = vqrshrn_n_u16(q8u16, 7); + d16u8 = vqrshrn_n_u16(q9u16, 7); + d17u8 = vqrshrn_n_u16(q10u16, 7); + d18u8 = vqrshrn_n_u16(q11u16, 7); + d19u8 = vqrshrn_n_u16(q12u16, 7); + d20u8 = vqrshrn_n_u16(q13u16, 7); + d21u8 = vqrshrn_n_u16(q14u16, 7); + + d2u8 = vld1_u8(src_ptr); + d3u8 = vld1_u8(src_ptr + 8); + d4u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + d5u8 = vld1_u8(src_ptr); + d6u8 = vld1_u8(src_ptr + 8); + d7u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + d8u8 = vld1_u8(src_ptr); + d9u8 = vld1_u8(src_ptr + 8); + d10u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + d11u8 = vld1_u8(src_ptr); + d12u8 = vld1_u8(src_ptr + 8); + d13u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + + q7u8 = vcombine_u8(d14u8, d15u8); + q8u8 = vcombine_u8(d16u8, d17u8); + q9u8 = vcombine_u8(d18u8, d19u8); + q10u8 = vcombine_u8(d20u8, d21u8); + + vst1q_u8((uint8_t *)tmpp, q7u8); tmpp += 16; + vst1q_u8((uint8_t *)tmpp, q8u8); tmpp += 16; + vst1q_u8((uint8_t *)tmpp, q9u8); tmpp += 16; + vst1q_u8((uint8_t *)tmpp, q10u8); tmpp += 16; + } + + // First-pass filtering for rest 5 lines + d14u8 = vld1_u8(src_ptr); + d15u8 = vld1_u8(src_ptr + 8); + d16u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + + q9u16 = vmull_u8(d2u8, d0u8); + q10u16 = vmull_u8(d3u8, d0u8); + q11u16 = vmull_u8(d5u8, d0u8); + q12u16 = vmull_u8(d6u8, d0u8); + q13u16 = vmull_u8(d8u8, d0u8); + q14u16 = vmull_u8(d9u8, d0u8); + + d2u8 = vext_u8(d2u8, d3u8, 1); + d5u8 = vext_u8(d5u8, d6u8, 1); + d8u8 = vext_u8(d8u8, d9u8, 1); + + q9u16 = vmlal_u8(q9u16, d2u8, d1u8); + q11u16 = vmlal_u8(q11u16, d5u8, d1u8); + q13u16 = vmlal_u8(q13u16, d8u8, d1u8); + + d3u8 = vext_u8(d3u8, d4u8, 1); + d6u8 = vext_u8(d6u8, d7u8, 1); + d9u8 = vext_u8(d9u8, d10u8, 1); + + q10u16 = vmlal_u8(q10u16, d3u8, d1u8); + q12u16 = vmlal_u8(q12u16, d6u8, d1u8); + q14u16 = vmlal_u8(q14u16, d9u8, d1u8); + + q1u16 = vmull_u8(d11u8, d0u8); + q2u16 = vmull_u8(d12u8, d0u8); + q3u16 = vmull_u8(d14u8, d0u8); + q4u16 = vmull_u8(d15u8, d0u8); + + d11u8 = vext_u8(d11u8, d12u8, 1); + d14u8 = vext_u8(d14u8, d15u8, 1); + + q1u16 = vmlal_u8(q1u16, d11u8, d1u8); + q3u16 = vmlal_u8(q3u16, d14u8, d1u8); + + d12u8 = vext_u8(d12u8, d13u8, 1); + d15u8 = vext_u8(d15u8, d16u8, 1); + + q2u16 = vmlal_u8(q2u16, d12u8, d1u8); + q4u16 = vmlal_u8(q4u16, d15u8, d1u8); + + d10u8 = vqrshrn_n_u16(q9u16, 7); + d11u8 = vqrshrn_n_u16(q10u16, 7); + d12u8 = vqrshrn_n_u16(q11u16, 7); + d13u8 = vqrshrn_n_u16(q12u16, 7); + d14u8 = vqrshrn_n_u16(q13u16, 7); + d15u8 = vqrshrn_n_u16(q14u16, 7); + d16u8 = vqrshrn_n_u16(q1u16, 7); + d17u8 = vqrshrn_n_u16(q2u16, 7); + d18u8 = vqrshrn_n_u16(q3u16, 7); + d19u8 = vqrshrn_n_u16(q4u16, 7); + + q5u8 = vcombine_u8(d10u8, d11u8); + q6u8 = vcombine_u8(d12u8, d13u8); + q7u8 = vcombine_u8(d14u8, d15u8); + q8u8 = vcombine_u8(d16u8, d17u8); + q9u8 = vcombine_u8(d18u8, d19u8); + + vst1q_u8((uint8_t *)tmpp, q5u8); tmpp += 16; + vst1q_u8((uint8_t *)tmpp, q6u8); tmpp += 16; + vst1q_u8((uint8_t *)tmpp, q7u8); tmpp += 16; + vst1q_u8((uint8_t *)tmpp, q8u8); tmpp += 16; + vst1q_u8((uint8_t *)tmpp, q9u8); + + // secondpass_filter + d0u8 = vdup_n_u8(bifilter4_coeff[yoffset][0]); + d1u8 = vdup_n_u8(bifilter4_coeff[yoffset][1]); + + tmpp = tmp; + q11u8 = vld1q_u8(tmpp); + tmpp += 16; + for (i = 4; i > 0; i--) { + q12u8 = vld1q_u8(tmpp); tmpp += 16; + q13u8 = vld1q_u8(tmpp); tmpp += 16; + q14u8 = vld1q_u8(tmpp); tmpp += 16; + q15u8 = vld1q_u8(tmpp); tmpp += 16; + + q1u16 = vmull_u8(vget_low_u8(q11u8), d0u8); + q2u16 = vmull_u8(vget_high_u8(q11u8), d0u8); + q3u16 = vmull_u8(vget_low_u8(q12u8), d0u8); + q4u16 = vmull_u8(vget_high_u8(q12u8), d0u8); + q5u16 = vmull_u8(vget_low_u8(q13u8), d0u8); + q6u16 = vmull_u8(vget_high_u8(q13u8), d0u8); + q7u16 = vmull_u8(vget_low_u8(q14u8), d0u8); + q8u16 = vmull_u8(vget_high_u8(q14u8), d0u8); + + q1u16 = vmlal_u8(q1u16, vget_low_u8(q12u8), d1u8); + q2u16 = vmlal_u8(q2u16, vget_high_u8(q12u8), d1u8); + q3u16 = vmlal_u8(q3u16, vget_low_u8(q13u8), d1u8); + q4u16 = vmlal_u8(q4u16, vget_high_u8(q13u8), d1u8); + q5u16 = vmlal_u8(q5u16, vget_low_u8(q14u8), d1u8); + q6u16 = vmlal_u8(q6u16, vget_high_u8(q14u8), d1u8); + q7u16 = vmlal_u8(q7u16, vget_low_u8(q15u8), d1u8); + q8u16 = vmlal_u8(q8u16, vget_high_u8(q15u8), d1u8); + + d2u8 = vqrshrn_n_u16(q1u16, 7); + d3u8 = vqrshrn_n_u16(q2u16, 7); + d4u8 = vqrshrn_n_u16(q3u16, 7); + d5u8 = vqrshrn_n_u16(q4u16, 7); + d6u8 = vqrshrn_n_u16(q5u16, 7); + d7u8 = vqrshrn_n_u16(q6u16, 7); + d8u8 = vqrshrn_n_u16(q7u16, 7); + d9u8 = vqrshrn_n_u16(q8u16, 7); + + q1u8 = vcombine_u8(d2u8, d3u8); + q2u8 = vcombine_u8(d4u8, d5u8); + q3u8 = vcombine_u8(d6u8, d7u8); + q4u8 = vcombine_u8(d8u8, d9u8); + + q11u8 = q15u8; + + vst1q_u8((uint8_t *)dst_ptr, q1u8); dst_ptr += dst_pitch; + vst1q_u8((uint8_t *)dst_ptr, q2u8); dst_ptr += dst_pitch; + vst1q_u8((uint8_t *)dst_ptr, q3u8); dst_ptr += dst_pitch; + vst1q_u8((uint8_t *)dst_ptr, q4u8); dst_ptr += dst_pitch; + } + return; +} diff --git a/thirdparty/libvpx/vp8/common/arm/neon/copymem_neon.c b/thirdparty/libvpx/vp8/common/arm/neon/copymem_neon.c new file mode 100644 index 0000000000..deced115c1 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/arm/neon/copymem_neon.c @@ -0,0 +1,59 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +void vp8_copy_mem8x4_neon( + unsigned char *src, + int src_stride, + unsigned char *dst, + int dst_stride) { + uint8x8_t vtmp; + int r; + + for (r = 0; r < 4; r++) { + vtmp = vld1_u8(src); + vst1_u8(dst, vtmp); + src += src_stride; + dst += dst_stride; + } +} + +void vp8_copy_mem8x8_neon( + unsigned char *src, + int src_stride, + unsigned char *dst, + int dst_stride) { + uint8x8_t vtmp; + int r; + + for (r = 0; r < 8; r++) { + vtmp = vld1_u8(src); + vst1_u8(dst, vtmp); + src += src_stride; + dst += dst_stride; + } +} + +void vp8_copy_mem16x16_neon( + unsigned char *src, + int src_stride, + unsigned char *dst, + int dst_stride) { + int r; + uint8x16_t qtmp; + + for (r = 0; r < 16; r++) { + qtmp = vld1q_u8(src); + vst1q_u8(dst, qtmp); + src += src_stride; + dst += dst_stride; + } +} diff --git a/thirdparty/libvpx/vp8/common/arm/neon/dc_only_idct_add_neon.c b/thirdparty/libvpx/vp8/common/arm/neon/dc_only_idct_add_neon.c new file mode 100644 index 0000000000..ad5f41d7de --- /dev/null +++ b/thirdparty/libvpx/vp8/common/arm/neon/dc_only_idct_add_neon.c @@ -0,0 +1,42 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +void vp8_dc_only_idct_add_neon( + int16_t input_dc, + unsigned char *pred_ptr, + int pred_stride, + unsigned char *dst_ptr, + int dst_stride) { + int i; + uint16_t a1 = ((input_dc + 4) >> 3); + uint32x2_t d2u32 = vdup_n_u32(0); + uint8x8_t d2u8; + uint16x8_t q1u16; + uint16x8_t qAdd; + + qAdd = vdupq_n_u16(a1); + + for (i = 0; i < 2; i++) { + d2u32 = vld1_lane_u32((const uint32_t *)pred_ptr, d2u32, 0); + pred_ptr += pred_stride; + d2u32 = vld1_lane_u32((const uint32_t *)pred_ptr, d2u32, 1); + pred_ptr += pred_stride; + + q1u16 = vaddw_u8(qAdd, vreinterpret_u8_u32(d2u32)); + d2u8 = vqmovun_s16(vreinterpretq_s16_u16(q1u16)); + + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d2u8), 0); + dst_ptr += dst_stride; + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d2u8), 1); + dst_ptr += dst_stride; + } +} diff --git a/thirdparty/libvpx/vp8/common/arm/neon/dequant_idct_neon.c b/thirdparty/libvpx/vp8/common/arm/neon/dequant_idct_neon.c new file mode 100644 index 0000000000..58e11922c7 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/arm/neon/dequant_idct_neon.c @@ -0,0 +1,142 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +static const int16_t cospi8sqrt2minus1 = 20091; +static const int16_t sinpi8sqrt2 = 35468; + +void vp8_dequant_idct_add_neon( + int16_t *input, + int16_t *dq, + unsigned char *dst, + int stride) { + unsigned char *dst0; + int32x2_t d14, d15; + int16x4_t d2, d3, d4, d5, d10, d11, d12, d13; + int16x8_t q1, q2, q3, q4, q5, q6; + int16x8_t qEmpty = vdupq_n_s16(0); + int32x2x2_t d2tmp0, d2tmp1; + int16x4x2_t d2tmp2, d2tmp3; + + d14 = d15 = vdup_n_s32(0); + + // load input + q3 = vld1q_s16(input); + vst1q_s16(input, qEmpty); + input += 8; + q4 = vld1q_s16(input); + vst1q_s16(input, qEmpty); + + // load dq + q5 = vld1q_s16(dq); + dq += 8; + q6 = vld1q_s16(dq); + + // load src from dst + dst0 = dst; + d14 = vld1_lane_s32((const int32_t *)dst0, d14, 0); + dst0 += stride; + d14 = vld1_lane_s32((const int32_t *)dst0, d14, 1); + dst0 += stride; + d15 = vld1_lane_s32((const int32_t *)dst0, d15, 0); + dst0 += stride; + d15 = vld1_lane_s32((const int32_t *)dst0, d15, 1); + + q1 = vreinterpretq_s16_u16(vmulq_u16(vreinterpretq_u16_s16(q3), + vreinterpretq_u16_s16(q5))); + q2 = vreinterpretq_s16_u16(vmulq_u16(vreinterpretq_u16_s16(q4), + vreinterpretq_u16_s16(q6))); + + d12 = vqadd_s16(vget_low_s16(q1), vget_low_s16(q2)); + d13 = vqsub_s16(vget_low_s16(q1), vget_low_s16(q2)); + + q2 = vcombine_s16(vget_high_s16(q1), vget_high_s16(q2)); + + q3 = vqdmulhq_n_s16(q2, sinpi8sqrt2); + q4 = vqdmulhq_n_s16(q2, cospi8sqrt2minus1); + + q3 = vshrq_n_s16(q3, 1); + q4 = vshrq_n_s16(q4, 1); + + q3 = vqaddq_s16(q3, q2); + q4 = vqaddq_s16(q4, q2); + + d10 = vqsub_s16(vget_low_s16(q3), vget_high_s16(q4)); + d11 = vqadd_s16(vget_high_s16(q3), vget_low_s16(q4)); + + d2 = vqadd_s16(d12, d11); + d3 = vqadd_s16(d13, d10); + d4 = vqsub_s16(d13, d10); + d5 = vqsub_s16(d12, d11); + + d2tmp0 = vtrn_s32(vreinterpret_s32_s16(d2), vreinterpret_s32_s16(d4)); + d2tmp1 = vtrn_s32(vreinterpret_s32_s16(d3), vreinterpret_s32_s16(d5)); + d2tmp2 = vtrn_s16(vreinterpret_s16_s32(d2tmp0.val[0]), + vreinterpret_s16_s32(d2tmp1.val[0])); + d2tmp3 = vtrn_s16(vreinterpret_s16_s32(d2tmp0.val[1]), + vreinterpret_s16_s32(d2tmp1.val[1])); + + // loop 2 + q2 = vcombine_s16(d2tmp2.val[1], d2tmp3.val[1]); + + q3 = vqdmulhq_n_s16(q2, sinpi8sqrt2); + q4 = vqdmulhq_n_s16(q2, cospi8sqrt2minus1); + + d12 = vqadd_s16(d2tmp2.val[0], d2tmp3.val[0]); + d13 = vqsub_s16(d2tmp2.val[0], d2tmp3.val[0]); + + q3 = vshrq_n_s16(q3, 1); + q4 = vshrq_n_s16(q4, 1); + + q3 = vqaddq_s16(q3, q2); + q4 = vqaddq_s16(q4, q2); + + d10 = vqsub_s16(vget_low_s16(q3), vget_high_s16(q4)); + d11 = vqadd_s16(vget_high_s16(q3), vget_low_s16(q4)); + + d2 = vqadd_s16(d12, d11); + d3 = vqadd_s16(d13, d10); + d4 = vqsub_s16(d13, d10); + d5 = vqsub_s16(d12, d11); + + d2 = vrshr_n_s16(d2, 3); + d3 = vrshr_n_s16(d3, 3); + d4 = vrshr_n_s16(d4, 3); + d5 = vrshr_n_s16(d5, 3); + + d2tmp0 = vtrn_s32(vreinterpret_s32_s16(d2), vreinterpret_s32_s16(d4)); + d2tmp1 = vtrn_s32(vreinterpret_s32_s16(d3), vreinterpret_s32_s16(d5)); + d2tmp2 = vtrn_s16(vreinterpret_s16_s32(d2tmp0.val[0]), + vreinterpret_s16_s32(d2tmp1.val[0])); + d2tmp3 = vtrn_s16(vreinterpret_s16_s32(d2tmp0.val[1]), + vreinterpret_s16_s32(d2tmp1.val[1])); + + q1 = vcombine_s16(d2tmp2.val[0], d2tmp2.val[1]); + q2 = vcombine_s16(d2tmp3.val[0], d2tmp3.val[1]); + + q1 = vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(q1), + vreinterpret_u8_s32(d14))); + q2 = vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(q2), + vreinterpret_u8_s32(d15))); + + d14 = vreinterpret_s32_u8(vqmovun_s16(q1)); + d15 = vreinterpret_s32_u8(vqmovun_s16(q2)); + + dst0 = dst; + vst1_lane_s32((int32_t *)dst0, d14, 0); + dst0 += stride; + vst1_lane_s32((int32_t *)dst0, d14, 1); + dst0 += stride; + vst1_lane_s32((int32_t *)dst0, d15, 0); + dst0 += stride; + vst1_lane_s32((int32_t *)dst0, d15, 1); + return; +} diff --git a/thirdparty/libvpx/vp8/common/arm/neon/dequantizeb_neon.c b/thirdparty/libvpx/vp8/common/arm/neon/dequantizeb_neon.c new file mode 100644 index 0000000000..54e709dd3c --- /dev/null +++ b/thirdparty/libvpx/vp8/common/arm/neon/dequantizeb_neon.c @@ -0,0 +1,25 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +#include "vp8/common/blockd.h" + +void vp8_dequantize_b_neon(BLOCKD *d, short *DQC) { + int16x8x2_t qQ, qDQC, qDQ; + + qQ = vld2q_s16(d->qcoeff); + qDQC = vld2q_s16(DQC); + + qDQ.val[0] = vmulq_s16(qQ.val[0], qDQC.val[0]); + qDQ.val[1] = vmulq_s16(qQ.val[1], qDQC.val[1]); + + vst2q_s16(d->dqcoeff, qDQ); +} diff --git a/thirdparty/libvpx/vp8/common/arm/neon/idct_blk_neon.c b/thirdparty/libvpx/vp8/common/arm/neon/idct_blk_neon.c new file mode 100644 index 0000000000..fb327a7260 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/arm/neon/idct_blk_neon.c @@ -0,0 +1,96 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "vpx_config.h" +#include "vp8_rtcd.h" + +/* place these declarations here because we don't want to maintain them + * outside of this scope + */ +void idct_dequant_full_2x_neon(short *q, short *dq, + unsigned char *dst, int stride); +void idct_dequant_0_2x_neon(short *q, short dq, + unsigned char *dst, int stride); + + +void vp8_dequant_idct_add_y_block_neon(short *q, short *dq, + unsigned char *dst, + int stride, char *eobs) +{ + int i; + + for (i = 0; i < 4; i++) + { + if (((short *)(eobs))[0]) + { + if (((short *)eobs)[0] & 0xfefe) + idct_dequant_full_2x_neon (q, dq, dst, stride); + else + idct_dequant_0_2x_neon (q, dq[0], dst, stride); + } + + if (((short *)(eobs))[1]) + { + if (((short *)eobs)[1] & 0xfefe) + idct_dequant_full_2x_neon (q+32, dq, dst+8, stride); + else + idct_dequant_0_2x_neon (q+32, dq[0], dst+8, stride); + } + q += 64; + dst += 4*stride; + eobs += 4; + } +} + +void vp8_dequant_idct_add_uv_block_neon(short *q, short *dq, + unsigned char *dstu, + unsigned char *dstv, + int stride, char *eobs) +{ + if (((short *)(eobs))[0]) + { + if (((short *)eobs)[0] & 0xfefe) + idct_dequant_full_2x_neon (q, dq, dstu, stride); + else + idct_dequant_0_2x_neon (q, dq[0], dstu, stride); + } + + q += 32; + dstu += 4*stride; + + if (((short *)(eobs))[1]) + { + if (((short *)eobs)[1] & 0xfefe) + idct_dequant_full_2x_neon (q, dq, dstu, stride); + else + idct_dequant_0_2x_neon (q, dq[0], dstu, stride); + } + + q += 32; + + if (((short *)(eobs))[2]) + { + if (((short *)eobs)[2] & 0xfefe) + idct_dequant_full_2x_neon (q, dq, dstv, stride); + else + idct_dequant_0_2x_neon (q, dq[0], dstv, stride); + } + + q += 32; + dstv += 4*stride; + + if (((short *)(eobs))[3]) + { + if (((short *)eobs)[3] & 0xfefe) + idct_dequant_full_2x_neon (q, dq, dstv, stride); + else + idct_dequant_0_2x_neon (q, dq[0], dstv, stride); + } +} diff --git a/thirdparty/libvpx/vp8/common/arm/neon/idct_dequant_0_2x_neon.c b/thirdparty/libvpx/vp8/common/arm/neon/idct_dequant_0_2x_neon.c new file mode 100644 index 0000000000..e6f862fa89 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/arm/neon/idct_dequant_0_2x_neon.c @@ -0,0 +1,63 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +void idct_dequant_0_2x_neon( + int16_t *q, + int16_t dq, + unsigned char *dst, + int stride) { + unsigned char *dst0; + int i, a0, a1; + int16x8x2_t q2Add; + int32x2_t d2s32 = vdup_n_s32(0), + d4s32 = vdup_n_s32(0); + uint8x8_t d2u8, d4u8; + uint16x8_t q1u16, q2u16; + + a0 = ((q[0] * dq) + 4) >> 3; + a1 = ((q[16] * dq) + 4) >> 3; + q[0] = q[16] = 0; + q2Add.val[0] = vdupq_n_s16((int16_t)a0); + q2Add.val[1] = vdupq_n_s16((int16_t)a1); + + for (i = 0; i < 2; i++, dst += 4) { + dst0 = dst; + d2s32 = vld1_lane_s32((const int32_t *)dst0, d2s32, 0); + dst0 += stride; + d2s32 = vld1_lane_s32((const int32_t *)dst0, d2s32, 1); + dst0 += stride; + d4s32 = vld1_lane_s32((const int32_t *)dst0, d4s32, 0); + dst0 += stride; + d4s32 = vld1_lane_s32((const int32_t *)dst0, d4s32, 1); + + q1u16 = vaddw_u8(vreinterpretq_u16_s16(q2Add.val[i]), + vreinterpret_u8_s32(d2s32)); + q2u16 = vaddw_u8(vreinterpretq_u16_s16(q2Add.val[i]), + vreinterpret_u8_s32(d4s32)); + + d2u8 = vqmovun_s16(vreinterpretq_s16_u16(q1u16)); + d4u8 = vqmovun_s16(vreinterpretq_s16_u16(q2u16)); + + d2s32 = vreinterpret_s32_u8(d2u8); + d4s32 = vreinterpret_s32_u8(d4u8); + + dst0 = dst; + vst1_lane_s32((int32_t *)dst0, d2s32, 0); + dst0 += stride; + vst1_lane_s32((int32_t *)dst0, d2s32, 1); + dst0 += stride; + vst1_lane_s32((int32_t *)dst0, d4s32, 0); + dst0 += stride; + vst1_lane_s32((int32_t *)dst0, d4s32, 1); + } + return; +} diff --git a/thirdparty/libvpx/vp8/common/arm/neon/idct_dequant_full_2x_neon.c b/thirdparty/libvpx/vp8/common/arm/neon/idct_dequant_full_2x_neon.c new file mode 100644 index 0000000000..a60ed46b76 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/arm/neon/idct_dequant_full_2x_neon.c @@ -0,0 +1,185 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +static const int16_t cospi8sqrt2minus1 = 20091; +static const int16_t sinpi8sqrt2 = 17734; +// because the lowest bit in 0x8a8c is 0, we can pre-shift this + +void idct_dequant_full_2x_neon( + int16_t *q, + int16_t *dq, + unsigned char *dst, + int stride) { + unsigned char *dst0, *dst1; + int32x2_t d28, d29, d30, d31; + int16x8_t q0, q1, q2, q3, q4, q5, q6, q7, q8, q9, q10, q11; + int16x8_t qEmpty = vdupq_n_s16(0); + int32x4x2_t q2tmp0, q2tmp1; + int16x8x2_t q2tmp2, q2tmp3; + int16x4_t dLow0, dLow1, dHigh0, dHigh1; + + d28 = d29 = d30 = d31 = vdup_n_s32(0); + + // load dq + q0 = vld1q_s16(dq); + dq += 8; + q1 = vld1q_s16(dq); + + // load q + q2 = vld1q_s16(q); + vst1q_s16(q, qEmpty); + q += 8; + q3 = vld1q_s16(q); + vst1q_s16(q, qEmpty); + q += 8; + q4 = vld1q_s16(q); + vst1q_s16(q, qEmpty); + q += 8; + q5 = vld1q_s16(q); + vst1q_s16(q, qEmpty); + + // load src from dst + dst0 = dst; + dst1 = dst + 4; + d28 = vld1_lane_s32((const int32_t *)dst0, d28, 0); + dst0 += stride; + d28 = vld1_lane_s32((const int32_t *)dst1, d28, 1); + dst1 += stride; + d29 = vld1_lane_s32((const int32_t *)dst0, d29, 0); + dst0 += stride; + d29 = vld1_lane_s32((const int32_t *)dst1, d29, 1); + dst1 += stride; + + d30 = vld1_lane_s32((const int32_t *)dst0, d30, 0); + dst0 += stride; + d30 = vld1_lane_s32((const int32_t *)dst1, d30, 1); + dst1 += stride; + d31 = vld1_lane_s32((const int32_t *)dst0, d31, 0); + d31 = vld1_lane_s32((const int32_t *)dst1, d31, 1); + + q2 = vmulq_s16(q2, q0); + q3 = vmulq_s16(q3, q1); + q4 = vmulq_s16(q4, q0); + q5 = vmulq_s16(q5, q1); + + // vswp + dLow0 = vget_low_s16(q2); + dHigh0 = vget_high_s16(q2); + dLow1 = vget_low_s16(q4); + dHigh1 = vget_high_s16(q4); + q2 = vcombine_s16(dLow0, dLow1); + q4 = vcombine_s16(dHigh0, dHigh1); + + dLow0 = vget_low_s16(q3); + dHigh0 = vget_high_s16(q3); + dLow1 = vget_low_s16(q5); + dHigh1 = vget_high_s16(q5); + q3 = vcombine_s16(dLow0, dLow1); + q5 = vcombine_s16(dHigh0, dHigh1); + + q6 = vqdmulhq_n_s16(q4, sinpi8sqrt2); + q7 = vqdmulhq_n_s16(q5, sinpi8sqrt2); + q8 = vqdmulhq_n_s16(q4, cospi8sqrt2minus1); + q9 = vqdmulhq_n_s16(q5, cospi8sqrt2minus1); + + q10 = vqaddq_s16(q2, q3); + q11 = vqsubq_s16(q2, q3); + + q8 = vshrq_n_s16(q8, 1); + q9 = vshrq_n_s16(q9, 1); + + q4 = vqaddq_s16(q4, q8); + q5 = vqaddq_s16(q5, q9); + + q2 = vqsubq_s16(q6, q5); + q3 = vqaddq_s16(q7, q4); + + q4 = vqaddq_s16(q10, q3); + q5 = vqaddq_s16(q11, q2); + q6 = vqsubq_s16(q11, q2); + q7 = vqsubq_s16(q10, q3); + + q2tmp0 = vtrnq_s32(vreinterpretq_s32_s16(q4), vreinterpretq_s32_s16(q6)); + q2tmp1 = vtrnq_s32(vreinterpretq_s32_s16(q5), vreinterpretq_s32_s16(q7)); + q2tmp2 = vtrnq_s16(vreinterpretq_s16_s32(q2tmp0.val[0]), + vreinterpretq_s16_s32(q2tmp1.val[0])); + q2tmp3 = vtrnq_s16(vreinterpretq_s16_s32(q2tmp0.val[1]), + vreinterpretq_s16_s32(q2tmp1.val[1])); + + // loop 2 + q8 = vqdmulhq_n_s16(q2tmp2.val[1], sinpi8sqrt2); + q9 = vqdmulhq_n_s16(q2tmp3.val[1], sinpi8sqrt2); + q10 = vqdmulhq_n_s16(q2tmp2.val[1], cospi8sqrt2minus1); + q11 = vqdmulhq_n_s16(q2tmp3.val[1], cospi8sqrt2minus1); + + q2 = vqaddq_s16(q2tmp2.val[0], q2tmp3.val[0]); + q3 = vqsubq_s16(q2tmp2.val[0], q2tmp3.val[0]); + + q10 = vshrq_n_s16(q10, 1); + q11 = vshrq_n_s16(q11, 1); + + q10 = vqaddq_s16(q2tmp2.val[1], q10); + q11 = vqaddq_s16(q2tmp3.val[1], q11); + + q8 = vqsubq_s16(q8, q11); + q9 = vqaddq_s16(q9, q10); + + q4 = vqaddq_s16(q2, q9); + q5 = vqaddq_s16(q3, q8); + q6 = vqsubq_s16(q3, q8); + q7 = vqsubq_s16(q2, q9); + + q4 = vrshrq_n_s16(q4, 3); + q5 = vrshrq_n_s16(q5, 3); + q6 = vrshrq_n_s16(q6, 3); + q7 = vrshrq_n_s16(q7, 3); + + q2tmp0 = vtrnq_s32(vreinterpretq_s32_s16(q4), vreinterpretq_s32_s16(q6)); + q2tmp1 = vtrnq_s32(vreinterpretq_s32_s16(q5), vreinterpretq_s32_s16(q7)); + q2tmp2 = vtrnq_s16(vreinterpretq_s16_s32(q2tmp0.val[0]), + vreinterpretq_s16_s32(q2tmp1.val[0])); + q2tmp3 = vtrnq_s16(vreinterpretq_s16_s32(q2tmp0.val[1]), + vreinterpretq_s16_s32(q2tmp1.val[1])); + + q4 = vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(q2tmp2.val[0]), + vreinterpret_u8_s32(d28))); + q5 = vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(q2tmp2.val[1]), + vreinterpret_u8_s32(d29))); + q6 = vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(q2tmp3.val[0]), + vreinterpret_u8_s32(d30))); + q7 = vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(q2tmp3.val[1]), + vreinterpret_u8_s32(d31))); + + d28 = vreinterpret_s32_u8(vqmovun_s16(q4)); + d29 = vreinterpret_s32_u8(vqmovun_s16(q5)); + d30 = vreinterpret_s32_u8(vqmovun_s16(q6)); + d31 = vreinterpret_s32_u8(vqmovun_s16(q7)); + + dst0 = dst; + dst1 = dst + 4; + vst1_lane_s32((int32_t *)dst0, d28, 0); + dst0 += stride; + vst1_lane_s32((int32_t *)dst1, d28, 1); + dst1 += stride; + vst1_lane_s32((int32_t *)dst0, d29, 0); + dst0 += stride; + vst1_lane_s32((int32_t *)dst1, d29, 1); + dst1 += stride; + + vst1_lane_s32((int32_t *)dst0, d30, 0); + dst0 += stride; + vst1_lane_s32((int32_t *)dst1, d30, 1); + dst1 += stride; + vst1_lane_s32((int32_t *)dst0, d31, 0); + vst1_lane_s32((int32_t *)dst1, d31, 1); + return; +} diff --git a/thirdparty/libvpx/vp8/common/arm/neon/iwalsh_neon.c b/thirdparty/libvpx/vp8/common/arm/neon/iwalsh_neon.c new file mode 100644 index 0000000000..6ea9dd712a --- /dev/null +++ b/thirdparty/libvpx/vp8/common/arm/neon/iwalsh_neon.c @@ -0,0 +1,102 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +void vp8_short_inv_walsh4x4_neon( + int16_t *input, + int16_t *mb_dqcoeff) { + int16x8_t q0s16, q1s16, q2s16, q3s16; + int16x4_t d4s16, d5s16, d6s16, d7s16; + int16x4x2_t v2tmp0, v2tmp1; + int32x2x2_t v2tmp2, v2tmp3; + int16x8_t qAdd3; + + q0s16 = vld1q_s16(input); + q1s16 = vld1q_s16(input + 8); + + // 1st for loop + d4s16 = vadd_s16(vget_low_s16(q0s16), vget_high_s16(q1s16)); + d6s16 = vadd_s16(vget_high_s16(q0s16), vget_low_s16(q1s16)); + d5s16 = vsub_s16(vget_low_s16(q0s16), vget_high_s16(q1s16)); + d7s16 = vsub_s16(vget_high_s16(q0s16), vget_low_s16(q1s16)); + + q2s16 = vcombine_s16(d4s16, d5s16); + q3s16 = vcombine_s16(d6s16, d7s16); + + q0s16 = vaddq_s16(q2s16, q3s16); + q1s16 = vsubq_s16(q2s16, q3s16); + + v2tmp2 = vtrn_s32(vreinterpret_s32_s16(vget_low_s16(q0s16)), + vreinterpret_s32_s16(vget_low_s16(q1s16))); + v2tmp3 = vtrn_s32(vreinterpret_s32_s16(vget_high_s16(q0s16)), + vreinterpret_s32_s16(vget_high_s16(q1s16))); + v2tmp0 = vtrn_s16(vreinterpret_s16_s32(v2tmp2.val[0]), + vreinterpret_s16_s32(v2tmp3.val[0])); + v2tmp1 = vtrn_s16(vreinterpret_s16_s32(v2tmp2.val[1]), + vreinterpret_s16_s32(v2tmp3.val[1])); + + // 2nd for loop + d4s16 = vadd_s16(v2tmp0.val[0], v2tmp1.val[1]); + d6s16 = vadd_s16(v2tmp0.val[1], v2tmp1.val[0]); + d5s16 = vsub_s16(v2tmp0.val[0], v2tmp1.val[1]); + d7s16 = vsub_s16(v2tmp0.val[1], v2tmp1.val[0]); + q2s16 = vcombine_s16(d4s16, d5s16); + q3s16 = vcombine_s16(d6s16, d7s16); + + qAdd3 = vdupq_n_s16(3); + + q0s16 = vaddq_s16(q2s16, q3s16); + q1s16 = vsubq_s16(q2s16, q3s16); + + q0s16 = vaddq_s16(q0s16, qAdd3); + q1s16 = vaddq_s16(q1s16, qAdd3); + + q0s16 = vshrq_n_s16(q0s16, 3); + q1s16 = vshrq_n_s16(q1s16, 3); + + // store + vst1_lane_s16(mb_dqcoeff, vget_low_s16(q0s16), 0); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_high_s16(q0s16), 0); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_low_s16(q1s16), 0); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_high_s16(q1s16), 0); + mb_dqcoeff += 16; + + vst1_lane_s16(mb_dqcoeff, vget_low_s16(q0s16), 1); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_high_s16(q0s16), 1); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_low_s16(q1s16), 1); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_high_s16(q1s16), 1); + mb_dqcoeff += 16; + + vst1_lane_s16(mb_dqcoeff, vget_low_s16(q0s16), 2); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_high_s16(q0s16), 2); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_low_s16(q1s16), 2); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_high_s16(q1s16), 2); + mb_dqcoeff += 16; + + vst1_lane_s16(mb_dqcoeff, vget_low_s16(q0s16), 3); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_high_s16(q0s16), 3); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_low_s16(q1s16), 3); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_high_s16(q1s16), 3); + mb_dqcoeff += 16; + return; +} diff --git a/thirdparty/libvpx/vp8/common/arm/neon/loopfiltersimplehorizontaledge_neon.c b/thirdparty/libvpx/vp8/common/arm/neon/loopfiltersimplehorizontaledge_neon.c new file mode 100644 index 0000000000..b25686ffb8 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/arm/neon/loopfiltersimplehorizontaledge_neon.c @@ -0,0 +1,111 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> +#include "./vpx_config.h" + +static INLINE void vp8_loop_filter_simple_horizontal_edge_neon( + unsigned char *s, + int p, + const unsigned char *blimit) { + uint8_t *sp; + uint8x16_t qblimit, q0u8; + uint8x16_t q5u8, q6u8, q7u8, q8u8, q9u8, q10u8, q14u8, q15u8; + int16x8_t q2s16, q3s16, q13s16; + int8x8_t d8s8, d9s8; + int8x16_t q2s8, q3s8, q4s8, q10s8, q11s8, q14s8; + + qblimit = vdupq_n_u8(*blimit); + + sp = s - (p << 1); + q5u8 = vld1q_u8(sp); + sp += p; + q6u8 = vld1q_u8(sp); + sp += p; + q7u8 = vld1q_u8(sp); + sp += p; + q8u8 = vld1q_u8(sp); + + q15u8 = vabdq_u8(q6u8, q7u8); + q14u8 = vabdq_u8(q5u8, q8u8); + + q15u8 = vqaddq_u8(q15u8, q15u8); + q14u8 = vshrq_n_u8(q14u8, 1); + q0u8 = vdupq_n_u8(0x80); + q13s16 = vdupq_n_s16(3); + q15u8 = vqaddq_u8(q15u8, q14u8); + + q5u8 = veorq_u8(q5u8, q0u8); + q6u8 = veorq_u8(q6u8, q0u8); + q7u8 = veorq_u8(q7u8, q0u8); + q8u8 = veorq_u8(q8u8, q0u8); + + q15u8 = vcgeq_u8(qblimit, q15u8); + + q2s16 = vsubl_s8(vget_low_s8(vreinterpretq_s8_u8(q7u8)), + vget_low_s8(vreinterpretq_s8_u8(q6u8))); + q3s16 = vsubl_s8(vget_high_s8(vreinterpretq_s8_u8(q7u8)), + vget_high_s8(vreinterpretq_s8_u8(q6u8))); + + q4s8 = vqsubq_s8(vreinterpretq_s8_u8(q5u8), + vreinterpretq_s8_u8(q8u8)); + + q2s16 = vmulq_s16(q2s16, q13s16); + q3s16 = vmulq_s16(q3s16, q13s16); + + q10u8 = vdupq_n_u8(3); + q9u8 = vdupq_n_u8(4); + + q2s16 = vaddw_s8(q2s16, vget_low_s8(q4s8)); + q3s16 = vaddw_s8(q3s16, vget_high_s8(q4s8)); + + d8s8 = vqmovn_s16(q2s16); + d9s8 = vqmovn_s16(q3s16); + q4s8 = vcombine_s8(d8s8, d9s8); + + q14s8 = vandq_s8(q4s8, vreinterpretq_s8_u8(q15u8)); + + q2s8 = vqaddq_s8(q14s8, vreinterpretq_s8_u8(q10u8)); + q3s8 = vqaddq_s8(q14s8, vreinterpretq_s8_u8(q9u8)); + q2s8 = vshrq_n_s8(q2s8, 3); + q3s8 = vshrq_n_s8(q3s8, 3); + + q11s8 = vqaddq_s8(vreinterpretq_s8_u8(q6u8), q2s8); + q10s8 = vqsubq_s8(vreinterpretq_s8_u8(q7u8), q3s8); + + q6u8 = veorq_u8(vreinterpretq_u8_s8(q11s8), q0u8); + q7u8 = veorq_u8(vreinterpretq_u8_s8(q10s8), q0u8); + + vst1q_u8(s, q7u8); + s -= p; + vst1q_u8(s, q6u8); + return; +} + +void vp8_loop_filter_bhs_neon( + unsigned char *y_ptr, + int y_stride, + const unsigned char *blimit) { + y_ptr += y_stride * 4; + vp8_loop_filter_simple_horizontal_edge_neon(y_ptr, y_stride, blimit); + y_ptr += y_stride * 4; + vp8_loop_filter_simple_horizontal_edge_neon(y_ptr, y_stride, blimit); + y_ptr += y_stride * 4; + vp8_loop_filter_simple_horizontal_edge_neon(y_ptr, y_stride, blimit); + return; +} + +void vp8_loop_filter_mbhs_neon( + unsigned char *y_ptr, + int y_stride, + const unsigned char *blimit) { + vp8_loop_filter_simple_horizontal_edge_neon(y_ptr, y_stride, blimit); + return; +} diff --git a/thirdparty/libvpx/vp8/common/arm/neon/loopfiltersimpleverticaledge_neon.c b/thirdparty/libvpx/vp8/common/arm/neon/loopfiltersimpleverticaledge_neon.c new file mode 100644 index 0000000000..921bcad698 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/arm/neon/loopfiltersimpleverticaledge_neon.c @@ -0,0 +1,283 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> +#include "./vpx_config.h" +#include "vpx_ports/arm.h" + +#ifdef VPX_INCOMPATIBLE_GCC +static INLINE void write_2x4(unsigned char *dst, int pitch, + const uint8x8x2_t result) { + /* + * uint8x8x2_t result + 00 01 02 03 | 04 05 06 07 + 10 11 12 13 | 14 15 16 17 + --- + * after vtrn_u8 + 00 10 02 12 | 04 14 06 16 + 01 11 03 13 | 05 15 07 17 + */ + const uint8x8x2_t r01_u8 = vtrn_u8(result.val[0], + result.val[1]); + const uint16x4_t x_0_4 = vreinterpret_u16_u8(r01_u8.val[0]); + const uint16x4_t x_1_5 = vreinterpret_u16_u8(r01_u8.val[1]); + vst1_lane_u16((uint16_t *)dst, x_0_4, 0); + dst += pitch; + vst1_lane_u16((uint16_t *)dst, x_1_5, 0); + dst += pitch; + vst1_lane_u16((uint16_t *)dst, x_0_4, 1); + dst += pitch; + vst1_lane_u16((uint16_t *)dst, x_1_5, 1); + dst += pitch; + vst1_lane_u16((uint16_t *)dst, x_0_4, 2); + dst += pitch; + vst1_lane_u16((uint16_t *)dst, x_1_5, 2); + dst += pitch; + vst1_lane_u16((uint16_t *)dst, x_0_4, 3); + dst += pitch; + vst1_lane_u16((uint16_t *)dst, x_1_5, 3); +} + +static INLINE void write_2x8(unsigned char *dst, int pitch, + const uint8x8x2_t result, + const uint8x8x2_t result2) { + write_2x4(dst, pitch, result); + dst += pitch * 8; + write_2x4(dst, pitch, result2); +} +#else +static INLINE void write_2x8(unsigned char *dst, int pitch, + const uint8x8x2_t result, + const uint8x8x2_t result2) { + vst2_lane_u8(dst, result, 0); + dst += pitch; + vst2_lane_u8(dst, result, 1); + dst += pitch; + vst2_lane_u8(dst, result, 2); + dst += pitch; + vst2_lane_u8(dst, result, 3); + dst += pitch; + vst2_lane_u8(dst, result, 4); + dst += pitch; + vst2_lane_u8(dst, result, 5); + dst += pitch; + vst2_lane_u8(dst, result, 6); + dst += pitch; + vst2_lane_u8(dst, result, 7); + dst += pitch; + + vst2_lane_u8(dst, result2, 0); + dst += pitch; + vst2_lane_u8(dst, result2, 1); + dst += pitch; + vst2_lane_u8(dst, result2, 2); + dst += pitch; + vst2_lane_u8(dst, result2, 3); + dst += pitch; + vst2_lane_u8(dst, result2, 4); + dst += pitch; + vst2_lane_u8(dst, result2, 5); + dst += pitch; + vst2_lane_u8(dst, result2, 6); + dst += pitch; + vst2_lane_u8(dst, result2, 7); +} +#endif // VPX_INCOMPATIBLE_GCC + + +#ifdef VPX_INCOMPATIBLE_GCC +static INLINE +uint8x8x4_t read_4x8(unsigned char *src, int pitch) { + uint8x8x4_t x; + const uint8x8_t a = vld1_u8(src); + const uint8x8_t b = vld1_u8(src + pitch * 1); + const uint8x8_t c = vld1_u8(src + pitch * 2); + const uint8x8_t d = vld1_u8(src + pitch * 3); + const uint8x8_t e = vld1_u8(src + pitch * 4); + const uint8x8_t f = vld1_u8(src + pitch * 5); + const uint8x8_t g = vld1_u8(src + pitch * 6); + const uint8x8_t h = vld1_u8(src + pitch * 7); + const uint32x2x2_t r04_u32 = vtrn_u32(vreinterpret_u32_u8(a), + vreinterpret_u32_u8(e)); + const uint32x2x2_t r15_u32 = vtrn_u32(vreinterpret_u32_u8(b), + vreinterpret_u32_u8(f)); + const uint32x2x2_t r26_u32 = vtrn_u32(vreinterpret_u32_u8(c), + vreinterpret_u32_u8(g)); + const uint32x2x2_t r37_u32 = vtrn_u32(vreinterpret_u32_u8(d), + vreinterpret_u32_u8(h)); + const uint16x4x2_t r02_u16 = vtrn_u16(vreinterpret_u16_u32(r04_u32.val[0]), + vreinterpret_u16_u32(r26_u32.val[0])); + const uint16x4x2_t r13_u16 = vtrn_u16(vreinterpret_u16_u32(r15_u32.val[0]), + vreinterpret_u16_u32(r37_u32.val[0])); + const uint8x8x2_t r01_u8 = vtrn_u8(vreinterpret_u8_u16(r02_u16.val[0]), + vreinterpret_u8_u16(r13_u16.val[0])); + const uint8x8x2_t r23_u8 = vtrn_u8(vreinterpret_u8_u16(r02_u16.val[1]), + vreinterpret_u8_u16(r13_u16.val[1])); + /* + * after vtrn_u32 + 00 01 02 03 | 40 41 42 43 + 10 11 12 13 | 50 51 52 53 + 20 21 22 23 | 60 61 62 63 + 30 31 32 33 | 70 71 72 73 + --- + * after vtrn_u16 + 00 01 20 21 | 40 41 60 61 + 02 03 22 23 | 42 43 62 63 + 10 11 30 31 | 50 51 70 71 + 12 13 32 33 | 52 52 72 73 + + 00 01 20 21 | 40 41 60 61 + 10 11 30 31 | 50 51 70 71 + 02 03 22 23 | 42 43 62 63 + 12 13 32 33 | 52 52 72 73 + --- + * after vtrn_u8 + 00 10 20 30 | 40 50 60 70 + 01 11 21 31 | 41 51 61 71 + 02 12 22 32 | 42 52 62 72 + 03 13 23 33 | 43 53 63 73 + */ + x.val[0] = r01_u8.val[0]; + x.val[1] = r01_u8.val[1]; + x.val[2] = r23_u8.val[0]; + x.val[3] = r23_u8.val[1]; + + return x; +} +#else +static INLINE +uint8x8x4_t read_4x8(unsigned char *src, int pitch) { + uint8x8x4_t x; + x.val[0] = x.val[1] = x.val[2] = x.val[3] = vdup_n_u8(0); + x = vld4_lane_u8(src, x, 0); + src += pitch; + x = vld4_lane_u8(src, x, 1); + src += pitch; + x = vld4_lane_u8(src, x, 2); + src += pitch; + x = vld4_lane_u8(src, x, 3); + src += pitch; + x = vld4_lane_u8(src, x, 4); + src += pitch; + x = vld4_lane_u8(src, x, 5); + src += pitch; + x = vld4_lane_u8(src, x, 6); + src += pitch; + x = vld4_lane_u8(src, x, 7); + return x; +} +#endif // VPX_INCOMPATIBLE_GCC + +static INLINE void vp8_loop_filter_simple_vertical_edge_neon( + unsigned char *s, + int p, + const unsigned char *blimit) { + unsigned char *src1; + uint8x16_t qblimit, q0u8; + uint8x16_t q3u8, q4u8, q5u8, q6u8, q7u8, q11u8, q12u8, q14u8, q15u8; + int16x8_t q2s16, q13s16, q11s16; + int8x8_t d28s8, d29s8; + int8x16_t q2s8, q3s8, q10s8, q11s8, q14s8; + uint8x8x4_t d0u8x4; // d6, d7, d8, d9 + uint8x8x4_t d1u8x4; // d10, d11, d12, d13 + uint8x8x2_t d2u8x2; // d12, d13 + uint8x8x2_t d3u8x2; // d14, d15 + + qblimit = vdupq_n_u8(*blimit); + + src1 = s - 2; + d0u8x4 = read_4x8(src1, p); + src1 += p * 8; + d1u8x4 = read_4x8(src1, p); + + q3u8 = vcombine_u8(d0u8x4.val[0], d1u8x4.val[0]); // d6 d10 + q4u8 = vcombine_u8(d0u8x4.val[2], d1u8x4.val[2]); // d8 d12 + q5u8 = vcombine_u8(d0u8x4.val[1], d1u8x4.val[1]); // d7 d11 + q6u8 = vcombine_u8(d0u8x4.val[3], d1u8x4.val[3]); // d9 d13 + + q15u8 = vabdq_u8(q5u8, q4u8); + q14u8 = vabdq_u8(q3u8, q6u8); + + q15u8 = vqaddq_u8(q15u8, q15u8); + q14u8 = vshrq_n_u8(q14u8, 1); + q0u8 = vdupq_n_u8(0x80); + q11s16 = vdupq_n_s16(3); + q15u8 = vqaddq_u8(q15u8, q14u8); + + q3u8 = veorq_u8(q3u8, q0u8); + q4u8 = veorq_u8(q4u8, q0u8); + q5u8 = veorq_u8(q5u8, q0u8); + q6u8 = veorq_u8(q6u8, q0u8); + + q15u8 = vcgeq_u8(qblimit, q15u8); + + q2s16 = vsubl_s8(vget_low_s8(vreinterpretq_s8_u8(q4u8)), + vget_low_s8(vreinterpretq_s8_u8(q5u8))); + q13s16 = vsubl_s8(vget_high_s8(vreinterpretq_s8_u8(q4u8)), + vget_high_s8(vreinterpretq_s8_u8(q5u8))); + + q14s8 = vqsubq_s8(vreinterpretq_s8_u8(q3u8), + vreinterpretq_s8_u8(q6u8)); + + q2s16 = vmulq_s16(q2s16, q11s16); + q13s16 = vmulq_s16(q13s16, q11s16); + + q11u8 = vdupq_n_u8(3); + q12u8 = vdupq_n_u8(4); + + q2s16 = vaddw_s8(q2s16, vget_low_s8(q14s8)); + q13s16 = vaddw_s8(q13s16, vget_high_s8(q14s8)); + + d28s8 = vqmovn_s16(q2s16); + d29s8 = vqmovn_s16(q13s16); + q14s8 = vcombine_s8(d28s8, d29s8); + + q14s8 = vandq_s8(q14s8, vreinterpretq_s8_u8(q15u8)); + + q2s8 = vqaddq_s8(q14s8, vreinterpretq_s8_u8(q11u8)); + q3s8 = vqaddq_s8(q14s8, vreinterpretq_s8_u8(q12u8)); + q2s8 = vshrq_n_s8(q2s8, 3); + q14s8 = vshrq_n_s8(q3s8, 3); + + q11s8 = vqaddq_s8(vreinterpretq_s8_u8(q5u8), q2s8); + q10s8 = vqsubq_s8(vreinterpretq_s8_u8(q4u8), q14s8); + + q6u8 = veorq_u8(vreinterpretq_u8_s8(q11s8), q0u8); + q7u8 = veorq_u8(vreinterpretq_u8_s8(q10s8), q0u8); + + d2u8x2.val[0] = vget_low_u8(q6u8); // d12 + d2u8x2.val[1] = vget_low_u8(q7u8); // d14 + d3u8x2.val[0] = vget_high_u8(q6u8); // d13 + d3u8x2.val[1] = vget_high_u8(q7u8); // d15 + + src1 = s - 1; + write_2x8(src1, p, d2u8x2, d3u8x2); +} + +void vp8_loop_filter_bvs_neon( + unsigned char *y_ptr, + int y_stride, + const unsigned char *blimit) { + y_ptr += 4; + vp8_loop_filter_simple_vertical_edge_neon(y_ptr, y_stride, blimit); + y_ptr += 4; + vp8_loop_filter_simple_vertical_edge_neon(y_ptr, y_stride, blimit); + y_ptr += 4; + vp8_loop_filter_simple_vertical_edge_neon(y_ptr, y_stride, blimit); + return; +} + +void vp8_loop_filter_mbvs_neon( + unsigned char *y_ptr, + int y_stride, + const unsigned char *blimit) { + vp8_loop_filter_simple_vertical_edge_neon(y_ptr, y_stride, blimit); + return; +} diff --git a/thirdparty/libvpx/vp8/common/arm/neon/mbloopfilter_neon.c b/thirdparty/libvpx/vp8/common/arm/neon/mbloopfilter_neon.c new file mode 100644 index 0000000000..5351f4be66 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/arm/neon/mbloopfilter_neon.c @@ -0,0 +1,625 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> +#include "./vpx_config.h" + +static INLINE void vp8_mbloop_filter_neon( + uint8x16_t qblimit, // mblimit + uint8x16_t qlimit, // limit + uint8x16_t qthresh, // thresh + uint8x16_t q3, // p2 + uint8x16_t q4, // p2 + uint8x16_t q5, // p1 + uint8x16_t q6, // p0 + uint8x16_t q7, // q0 + uint8x16_t q8, // q1 + uint8x16_t q9, // q2 + uint8x16_t q10, // q3 + uint8x16_t *q4r, // p1 + uint8x16_t *q5r, // p1 + uint8x16_t *q6r, // p0 + uint8x16_t *q7r, // q0 + uint8x16_t *q8r, // q1 + uint8x16_t *q9r) { // q1 + uint8x16_t q0u8, q1u8, q11u8, q12u8, q13u8, q14u8, q15u8; + int16x8_t q0s16, q2s16, q11s16, q12s16, q13s16, q14s16, q15s16; + int8x16_t q1s8, q6s8, q7s8, q2s8, q11s8, q13s8; + uint16x8_t q0u16, q11u16, q12u16, q13u16, q14u16, q15u16; + int8x16_t q0s8, q12s8, q14s8, q15s8; + int8x8_t d0, d1, d2, d3, d4, d5, d24, d25, d28, d29; + + q11u8 = vabdq_u8(q3, q4); + q12u8 = vabdq_u8(q4, q5); + q13u8 = vabdq_u8(q5, q6); + q14u8 = vabdq_u8(q8, q7); + q1u8 = vabdq_u8(q9, q8); + q0u8 = vabdq_u8(q10, q9); + + q11u8 = vmaxq_u8(q11u8, q12u8); + q12u8 = vmaxq_u8(q13u8, q14u8); + q1u8 = vmaxq_u8(q1u8, q0u8); + q15u8 = vmaxq_u8(q11u8, q12u8); + + q12u8 = vabdq_u8(q6, q7); + + // vp8_hevmask + q13u8 = vcgtq_u8(q13u8, qthresh); + q14u8 = vcgtq_u8(q14u8, qthresh); + q15u8 = vmaxq_u8(q15u8, q1u8); + + q15u8 = vcgeq_u8(qlimit, q15u8); + + q1u8 = vabdq_u8(q5, q8); + q12u8 = vqaddq_u8(q12u8, q12u8); + + // vp8_filter() function + // convert to signed + q0u8 = vdupq_n_u8(0x80); + q9 = veorq_u8(q9, q0u8); + q8 = veorq_u8(q8, q0u8); + q7 = veorq_u8(q7, q0u8); + q6 = veorq_u8(q6, q0u8); + q5 = veorq_u8(q5, q0u8); + q4 = veorq_u8(q4, q0u8); + + q1u8 = vshrq_n_u8(q1u8, 1); + q12u8 = vqaddq_u8(q12u8, q1u8); + + q14u8 = vorrq_u8(q13u8, q14u8); + q12u8 = vcgeq_u8(qblimit, q12u8); + + q2s16 = vsubl_s8(vget_low_s8(vreinterpretq_s8_u8(q7)), + vget_low_s8(vreinterpretq_s8_u8(q6))); + q13s16 = vsubl_s8(vget_high_s8(vreinterpretq_s8_u8(q7)), + vget_high_s8(vreinterpretq_s8_u8(q6))); + + q1s8 = vqsubq_s8(vreinterpretq_s8_u8(q5), + vreinterpretq_s8_u8(q8)); + + q11s16 = vdupq_n_s16(3); + q2s16 = vmulq_s16(q2s16, q11s16); + q13s16 = vmulq_s16(q13s16, q11s16); + + q15u8 = vandq_u8(q15u8, q12u8); + + q2s16 = vaddw_s8(q2s16, vget_low_s8(q1s8)); + q13s16 = vaddw_s8(q13s16, vget_high_s8(q1s8)); + + q12u8 = vdupq_n_u8(3); + q11u8 = vdupq_n_u8(4); + // vp8_filter = clamp(vp8_filter + 3 * ( qs0 - ps0)) + d2 = vqmovn_s16(q2s16); + d3 = vqmovn_s16(q13s16); + q1s8 = vcombine_s8(d2, d3); + q1s8 = vandq_s8(q1s8, vreinterpretq_s8_u8(q15u8)); + q13s8 = vandq_s8(q1s8, vreinterpretq_s8_u8(q14u8)); + + q2s8 = vqaddq_s8(q13s8, vreinterpretq_s8_u8(q11u8)); + q13s8 = vqaddq_s8(q13s8, vreinterpretq_s8_u8(q12u8)); + q2s8 = vshrq_n_s8(q2s8, 3); + q13s8 = vshrq_n_s8(q13s8, 3); + + q7s8 = vqsubq_s8(vreinterpretq_s8_u8(q7), q2s8); + q6s8 = vqaddq_s8(vreinterpretq_s8_u8(q6), q13s8); + + q1s8 = vbicq_s8(q1s8, vreinterpretq_s8_u8(q14u8)); + + q0u16 = q11u16 = q12u16 = q13u16 = q14u16 = q15u16 = vdupq_n_u16(63); + d5 = vdup_n_s8(9); + d4 = vdup_n_s8(18); + + q0s16 = vmlal_s8(vreinterpretq_s16_u16(q0u16), vget_low_s8(q1s8), d5); + q11s16 = vmlal_s8(vreinterpretq_s16_u16(q11u16), vget_high_s8(q1s8), d5); + d5 = vdup_n_s8(27); + q12s16 = vmlal_s8(vreinterpretq_s16_u16(q12u16), vget_low_s8(q1s8), d4); + q13s16 = vmlal_s8(vreinterpretq_s16_u16(q13u16), vget_high_s8(q1s8), d4); + q14s16 = vmlal_s8(vreinterpretq_s16_u16(q14u16), vget_low_s8(q1s8), d5); + q15s16 = vmlal_s8(vreinterpretq_s16_u16(q15u16), vget_high_s8(q1s8), d5); + + d0 = vqshrn_n_s16(q0s16 , 7); + d1 = vqshrn_n_s16(q11s16, 7); + d24 = vqshrn_n_s16(q12s16, 7); + d25 = vqshrn_n_s16(q13s16, 7); + d28 = vqshrn_n_s16(q14s16, 7); + d29 = vqshrn_n_s16(q15s16, 7); + + q0s8 = vcombine_s8(d0, d1); + q12s8 = vcombine_s8(d24, d25); + q14s8 = vcombine_s8(d28, d29); + + q11s8 = vqsubq_s8(vreinterpretq_s8_u8(q9), q0s8); + q0s8 = vqaddq_s8(vreinterpretq_s8_u8(q4), q0s8); + q13s8 = vqsubq_s8(vreinterpretq_s8_u8(q8), q12s8); + q12s8 = vqaddq_s8(vreinterpretq_s8_u8(q5), q12s8); + q15s8 = vqsubq_s8((q7s8), q14s8); + q14s8 = vqaddq_s8((q6s8), q14s8); + + q1u8 = vdupq_n_u8(0x80); + *q9r = veorq_u8(vreinterpretq_u8_s8(q11s8), q1u8); + *q8r = veorq_u8(vreinterpretq_u8_s8(q13s8), q1u8); + *q7r = veorq_u8(vreinterpretq_u8_s8(q15s8), q1u8); + *q6r = veorq_u8(vreinterpretq_u8_s8(q14s8), q1u8); + *q5r = veorq_u8(vreinterpretq_u8_s8(q12s8), q1u8); + *q4r = veorq_u8(vreinterpretq_u8_s8(q0s8), q1u8); + return; +} + +void vp8_mbloop_filter_horizontal_edge_y_neon( + unsigned char *src, + int pitch, + unsigned char blimit, + unsigned char limit, + unsigned char thresh) { + uint8x16_t qblimit, qlimit, qthresh, q3, q4; + uint8x16_t q5, q6, q7, q8, q9, q10; + + qblimit = vdupq_n_u8(blimit); + qlimit = vdupq_n_u8(limit); + qthresh = vdupq_n_u8(thresh); + + src -= (pitch << 2); + + q3 = vld1q_u8(src); + src += pitch; + q4 = vld1q_u8(src); + src += pitch; + q5 = vld1q_u8(src); + src += pitch; + q6 = vld1q_u8(src); + src += pitch; + q7 = vld1q_u8(src); + src += pitch; + q8 = vld1q_u8(src); + src += pitch; + q9 = vld1q_u8(src); + src += pitch; + q10 = vld1q_u8(src); + + vp8_mbloop_filter_neon(qblimit, qlimit, qthresh, q3, q4, + q5, q6, q7, q8, q9, q10, + &q4, &q5, &q6, &q7, &q8, &q9); + + src -= (pitch * 6); + vst1q_u8(src, q4); + src += pitch; + vst1q_u8(src, q5); + src += pitch; + vst1q_u8(src, q6); + src += pitch; + vst1q_u8(src, q7); + src += pitch; + vst1q_u8(src, q8); + src += pitch; + vst1q_u8(src, q9); + return; +} + +void vp8_mbloop_filter_horizontal_edge_uv_neon( + unsigned char *u, + int pitch, + unsigned char blimit, + unsigned char limit, + unsigned char thresh, + unsigned char *v) { + uint8x16_t qblimit, qlimit, qthresh, q3, q4; + uint8x16_t q5, q6, q7, q8, q9, q10; + uint8x8_t d6, d7, d8, d9, d10, d11, d12, d13, d14; + uint8x8_t d15, d16, d17, d18, d19, d20, d21; + + qblimit = vdupq_n_u8(blimit); + qlimit = vdupq_n_u8(limit); + qthresh = vdupq_n_u8(thresh); + + u -= (pitch << 2); + v -= (pitch << 2); + + d6 = vld1_u8(u); + u += pitch; + d7 = vld1_u8(v); + v += pitch; + d8 = vld1_u8(u); + u += pitch; + d9 = vld1_u8(v); + v += pitch; + d10 = vld1_u8(u); + u += pitch; + d11 = vld1_u8(v); + v += pitch; + d12 = vld1_u8(u); + u += pitch; + d13 = vld1_u8(v); + v += pitch; + d14 = vld1_u8(u); + u += pitch; + d15 = vld1_u8(v); + v += pitch; + d16 = vld1_u8(u); + u += pitch; + d17 = vld1_u8(v); + v += pitch; + d18 = vld1_u8(u); + u += pitch; + d19 = vld1_u8(v); + v += pitch; + d20 = vld1_u8(u); + d21 = vld1_u8(v); + + q3 = vcombine_u8(d6, d7); + q4 = vcombine_u8(d8, d9); + q5 = vcombine_u8(d10, d11); + q6 = vcombine_u8(d12, d13); + q7 = vcombine_u8(d14, d15); + q8 = vcombine_u8(d16, d17); + q9 = vcombine_u8(d18, d19); + q10 = vcombine_u8(d20, d21); + + vp8_mbloop_filter_neon(qblimit, qlimit, qthresh, q3, q4, + q5, q6, q7, q8, q9, q10, + &q4, &q5, &q6, &q7, &q8, &q9); + + u -= (pitch * 6); + v -= (pitch * 6); + vst1_u8(u, vget_low_u8(q4)); + u += pitch; + vst1_u8(v, vget_high_u8(q4)); + v += pitch; + vst1_u8(u, vget_low_u8(q5)); + u += pitch; + vst1_u8(v, vget_high_u8(q5)); + v += pitch; + vst1_u8(u, vget_low_u8(q6)); + u += pitch; + vst1_u8(v, vget_high_u8(q6)); + v += pitch; + vst1_u8(u, vget_low_u8(q7)); + u += pitch; + vst1_u8(v, vget_high_u8(q7)); + v += pitch; + vst1_u8(u, vget_low_u8(q8)); + u += pitch; + vst1_u8(v, vget_high_u8(q8)); + v += pitch; + vst1_u8(u, vget_low_u8(q9)); + vst1_u8(v, vget_high_u8(q9)); + return; +} + +void vp8_mbloop_filter_vertical_edge_y_neon( + unsigned char *src, + int pitch, + unsigned char blimit, + unsigned char limit, + unsigned char thresh) { + unsigned char *s1, *s2; + uint8x16_t qblimit, qlimit, qthresh, q3, q4; + uint8x16_t q5, q6, q7, q8, q9, q10; + uint8x8_t d6, d7, d8, d9, d10, d11, d12, d13, d14; + uint8x8_t d15, d16, d17, d18, d19, d20, d21; + uint32x4x2_t q2tmp0, q2tmp1, q2tmp2, q2tmp3; + uint16x8x2_t q2tmp4, q2tmp5, q2tmp6, q2tmp7; + uint8x16x2_t q2tmp8, q2tmp9, q2tmp10, q2tmp11; + + qblimit = vdupq_n_u8(blimit); + qlimit = vdupq_n_u8(limit); + qthresh = vdupq_n_u8(thresh); + + s1 = src - 4; + s2 = s1 + 8 * pitch; + d6 = vld1_u8(s1); + s1 += pitch; + d7 = vld1_u8(s2); + s2 += pitch; + d8 = vld1_u8(s1); + s1 += pitch; + d9 = vld1_u8(s2); + s2 += pitch; + d10 = vld1_u8(s1); + s1 += pitch; + d11 = vld1_u8(s2); + s2 += pitch; + d12 = vld1_u8(s1); + s1 += pitch; + d13 = vld1_u8(s2); + s2 += pitch; + d14 = vld1_u8(s1); + s1 += pitch; + d15 = vld1_u8(s2); + s2 += pitch; + d16 = vld1_u8(s1); + s1 += pitch; + d17 = vld1_u8(s2); + s2 += pitch; + d18 = vld1_u8(s1); + s1 += pitch; + d19 = vld1_u8(s2); + s2 += pitch; + d20 = vld1_u8(s1); + d21 = vld1_u8(s2); + + q3 = vcombine_u8(d6, d7); + q4 = vcombine_u8(d8, d9); + q5 = vcombine_u8(d10, d11); + q6 = vcombine_u8(d12, d13); + q7 = vcombine_u8(d14, d15); + q8 = vcombine_u8(d16, d17); + q9 = vcombine_u8(d18, d19); + q10 = vcombine_u8(d20, d21); + + q2tmp0 = vtrnq_u32(vreinterpretq_u32_u8(q3), vreinterpretq_u32_u8(q7)); + q2tmp1 = vtrnq_u32(vreinterpretq_u32_u8(q4), vreinterpretq_u32_u8(q8)); + q2tmp2 = vtrnq_u32(vreinterpretq_u32_u8(q5), vreinterpretq_u32_u8(q9)); + q2tmp3 = vtrnq_u32(vreinterpretq_u32_u8(q6), vreinterpretq_u32_u8(q10)); + + q2tmp4 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[0]), + vreinterpretq_u16_u32(q2tmp2.val[0])); + q2tmp5 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[0]), + vreinterpretq_u16_u32(q2tmp3.val[0])); + q2tmp6 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[1]), + vreinterpretq_u16_u32(q2tmp2.val[1])); + q2tmp7 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[1]), + vreinterpretq_u16_u32(q2tmp3.val[1])); + + q2tmp8 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[0]), + vreinterpretq_u8_u16(q2tmp5.val[0])); + q2tmp9 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[1]), + vreinterpretq_u8_u16(q2tmp5.val[1])); + q2tmp10 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[0]), + vreinterpretq_u8_u16(q2tmp7.val[0])); + q2tmp11 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[1]), + vreinterpretq_u8_u16(q2tmp7.val[1])); + + q3 = q2tmp8.val[0]; + q4 = q2tmp8.val[1]; + q5 = q2tmp9.val[0]; + q6 = q2tmp9.val[1]; + q7 = q2tmp10.val[0]; + q8 = q2tmp10.val[1]; + q9 = q2tmp11.val[0]; + q10 = q2tmp11.val[1]; + + vp8_mbloop_filter_neon(qblimit, qlimit, qthresh, q3, q4, + q5, q6, q7, q8, q9, q10, + &q4, &q5, &q6, &q7, &q8, &q9); + + q2tmp0 = vtrnq_u32(vreinterpretq_u32_u8(q3), vreinterpretq_u32_u8(q7)); + q2tmp1 = vtrnq_u32(vreinterpretq_u32_u8(q4), vreinterpretq_u32_u8(q8)); + q2tmp2 = vtrnq_u32(vreinterpretq_u32_u8(q5), vreinterpretq_u32_u8(q9)); + q2tmp3 = vtrnq_u32(vreinterpretq_u32_u8(q6), vreinterpretq_u32_u8(q10)); + + q2tmp4 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[0]), + vreinterpretq_u16_u32(q2tmp2.val[0])); + q2tmp5 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[0]), + vreinterpretq_u16_u32(q2tmp3.val[0])); + q2tmp6 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[1]), + vreinterpretq_u16_u32(q2tmp2.val[1])); + q2tmp7 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[1]), + vreinterpretq_u16_u32(q2tmp3.val[1])); + + q2tmp8 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[0]), + vreinterpretq_u8_u16(q2tmp5.val[0])); + q2tmp9 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[1]), + vreinterpretq_u8_u16(q2tmp5.val[1])); + q2tmp10 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[0]), + vreinterpretq_u8_u16(q2tmp7.val[0])); + q2tmp11 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[1]), + vreinterpretq_u8_u16(q2tmp7.val[1])); + + q3 = q2tmp8.val[0]; + q4 = q2tmp8.val[1]; + q5 = q2tmp9.val[0]; + q6 = q2tmp9.val[1]; + q7 = q2tmp10.val[0]; + q8 = q2tmp10.val[1]; + q9 = q2tmp11.val[0]; + q10 = q2tmp11.val[1]; + + s1 -= 7 * pitch; + s2 -= 7 * pitch; + + vst1_u8(s1, vget_low_u8(q3)); + s1 += pitch; + vst1_u8(s2, vget_high_u8(q3)); + s2 += pitch; + vst1_u8(s1, vget_low_u8(q4)); + s1 += pitch; + vst1_u8(s2, vget_high_u8(q4)); + s2 += pitch; + vst1_u8(s1, vget_low_u8(q5)); + s1 += pitch; + vst1_u8(s2, vget_high_u8(q5)); + s2 += pitch; + vst1_u8(s1, vget_low_u8(q6)); + s1 += pitch; + vst1_u8(s2, vget_high_u8(q6)); + s2 += pitch; + vst1_u8(s1, vget_low_u8(q7)); + s1 += pitch; + vst1_u8(s2, vget_high_u8(q7)); + s2 += pitch; + vst1_u8(s1, vget_low_u8(q8)); + s1 += pitch; + vst1_u8(s2, vget_high_u8(q8)); + s2 += pitch; + vst1_u8(s1, vget_low_u8(q9)); + s1 += pitch; + vst1_u8(s2, vget_high_u8(q9)); + s2 += pitch; + vst1_u8(s1, vget_low_u8(q10)); + vst1_u8(s2, vget_high_u8(q10)); + return; +} + +void vp8_mbloop_filter_vertical_edge_uv_neon( + unsigned char *u, + int pitch, + unsigned char blimit, + unsigned char limit, + unsigned char thresh, + unsigned char *v) { + unsigned char *us, *ud; + unsigned char *vs, *vd; + uint8x16_t qblimit, qlimit, qthresh, q3, q4; + uint8x16_t q5, q6, q7, q8, q9, q10; + uint8x8_t d6, d7, d8, d9, d10, d11, d12, d13, d14; + uint8x8_t d15, d16, d17, d18, d19, d20, d21; + uint32x4x2_t q2tmp0, q2tmp1, q2tmp2, q2tmp3; + uint16x8x2_t q2tmp4, q2tmp5, q2tmp6, q2tmp7; + uint8x16x2_t q2tmp8, q2tmp9, q2tmp10, q2tmp11; + + qblimit = vdupq_n_u8(blimit); + qlimit = vdupq_n_u8(limit); + qthresh = vdupq_n_u8(thresh); + + us = u - 4; + vs = v - 4; + d6 = vld1_u8(us); + us += pitch; + d7 = vld1_u8(vs); + vs += pitch; + d8 = vld1_u8(us); + us += pitch; + d9 = vld1_u8(vs); + vs += pitch; + d10 = vld1_u8(us); + us += pitch; + d11 = vld1_u8(vs); + vs += pitch; + d12 = vld1_u8(us); + us += pitch; + d13 = vld1_u8(vs); + vs += pitch; + d14 = vld1_u8(us); + us += pitch; + d15 = vld1_u8(vs); + vs += pitch; + d16 = vld1_u8(us); + us += pitch; + d17 = vld1_u8(vs); + vs += pitch; + d18 = vld1_u8(us); + us += pitch; + d19 = vld1_u8(vs); + vs += pitch; + d20 = vld1_u8(us); + d21 = vld1_u8(vs); + + q3 = vcombine_u8(d6, d7); + q4 = vcombine_u8(d8, d9); + q5 = vcombine_u8(d10, d11); + q6 = vcombine_u8(d12, d13); + q7 = vcombine_u8(d14, d15); + q8 = vcombine_u8(d16, d17); + q9 = vcombine_u8(d18, d19); + q10 = vcombine_u8(d20, d21); + + q2tmp0 = vtrnq_u32(vreinterpretq_u32_u8(q3), vreinterpretq_u32_u8(q7)); + q2tmp1 = vtrnq_u32(vreinterpretq_u32_u8(q4), vreinterpretq_u32_u8(q8)); + q2tmp2 = vtrnq_u32(vreinterpretq_u32_u8(q5), vreinterpretq_u32_u8(q9)); + q2tmp3 = vtrnq_u32(vreinterpretq_u32_u8(q6), vreinterpretq_u32_u8(q10)); + + q2tmp4 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[0]), + vreinterpretq_u16_u32(q2tmp2.val[0])); + q2tmp5 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[0]), + vreinterpretq_u16_u32(q2tmp3.val[0])); + q2tmp6 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[1]), + vreinterpretq_u16_u32(q2tmp2.val[1])); + q2tmp7 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[1]), + vreinterpretq_u16_u32(q2tmp3.val[1])); + + q2tmp8 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[0]), + vreinterpretq_u8_u16(q2tmp5.val[0])); + q2tmp9 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[1]), + vreinterpretq_u8_u16(q2tmp5.val[1])); + q2tmp10 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[0]), + vreinterpretq_u8_u16(q2tmp7.val[0])); + q2tmp11 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[1]), + vreinterpretq_u8_u16(q2tmp7.val[1])); + + q3 = q2tmp8.val[0]; + q4 = q2tmp8.val[1]; + q5 = q2tmp9.val[0]; + q6 = q2tmp9.val[1]; + q7 = q2tmp10.val[0]; + q8 = q2tmp10.val[1]; + q9 = q2tmp11.val[0]; + q10 = q2tmp11.val[1]; + + vp8_mbloop_filter_neon(qblimit, qlimit, qthresh, q3, q4, + q5, q6, q7, q8, q9, q10, + &q4, &q5, &q6, &q7, &q8, &q9); + + q2tmp0 = vtrnq_u32(vreinterpretq_u32_u8(q3), vreinterpretq_u32_u8(q7)); + q2tmp1 = vtrnq_u32(vreinterpretq_u32_u8(q4), vreinterpretq_u32_u8(q8)); + q2tmp2 = vtrnq_u32(vreinterpretq_u32_u8(q5), vreinterpretq_u32_u8(q9)); + q2tmp3 = vtrnq_u32(vreinterpretq_u32_u8(q6), vreinterpretq_u32_u8(q10)); + + q2tmp4 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[0]), + vreinterpretq_u16_u32(q2tmp2.val[0])); + q2tmp5 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[0]), + vreinterpretq_u16_u32(q2tmp3.val[0])); + q2tmp6 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[1]), + vreinterpretq_u16_u32(q2tmp2.val[1])); + q2tmp7 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[1]), + vreinterpretq_u16_u32(q2tmp3.val[1])); + + q2tmp8 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[0]), + vreinterpretq_u8_u16(q2tmp5.val[0])); + q2tmp9 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[1]), + vreinterpretq_u8_u16(q2tmp5.val[1])); + q2tmp10 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[0]), + vreinterpretq_u8_u16(q2tmp7.val[0])); + q2tmp11 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[1]), + vreinterpretq_u8_u16(q2tmp7.val[1])); + + q3 = q2tmp8.val[0]; + q4 = q2tmp8.val[1]; + q5 = q2tmp9.val[0]; + q6 = q2tmp9.val[1]; + q7 = q2tmp10.val[0]; + q8 = q2tmp10.val[1]; + q9 = q2tmp11.val[0]; + q10 = q2tmp11.val[1]; + + ud = u - 4; + vst1_u8(ud, vget_low_u8(q3)); + ud += pitch; + vst1_u8(ud, vget_low_u8(q4)); + ud += pitch; + vst1_u8(ud, vget_low_u8(q5)); + ud += pitch; + vst1_u8(ud, vget_low_u8(q6)); + ud += pitch; + vst1_u8(ud, vget_low_u8(q7)); + ud += pitch; + vst1_u8(ud, vget_low_u8(q8)); + ud += pitch; + vst1_u8(ud, vget_low_u8(q9)); + ud += pitch; + vst1_u8(ud, vget_low_u8(q10)); + + vd = v - 4; + vst1_u8(vd, vget_high_u8(q3)); + vd += pitch; + vst1_u8(vd, vget_high_u8(q4)); + vd += pitch; + vst1_u8(vd, vget_high_u8(q5)); + vd += pitch; + vst1_u8(vd, vget_high_u8(q6)); + vd += pitch; + vst1_u8(vd, vget_high_u8(q7)); + vd += pitch; + vst1_u8(vd, vget_high_u8(q8)); + vd += pitch; + vst1_u8(vd, vget_high_u8(q9)); + vd += pitch; + vst1_u8(vd, vget_high_u8(q10)); + return; +} diff --git a/thirdparty/libvpx/vp8/common/arm/neon/shortidct4x4llm_neon.c b/thirdparty/libvpx/vp8/common/arm/neon/shortidct4x4llm_neon.c new file mode 100644 index 0000000000..373afa6ed3 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/arm/neon/shortidct4x4llm_neon.c @@ -0,0 +1,123 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +static const int16_t cospi8sqrt2minus1 = 20091; +static const int16_t sinpi8sqrt2 = 35468; + +void vp8_short_idct4x4llm_neon( + int16_t *input, + unsigned char *pred_ptr, + int pred_stride, + unsigned char *dst_ptr, + int dst_stride) { + int i; + uint32x2_t d6u32 = vdup_n_u32(0); + uint8x8_t d1u8; + int16x4_t d2, d3, d4, d5, d10, d11, d12, d13; + uint16x8_t q1u16; + int16x8_t q1s16, q2s16, q3s16, q4s16; + int32x2x2_t v2tmp0, v2tmp1; + int16x4x2_t v2tmp2, v2tmp3; + + d2 = vld1_s16(input); + d3 = vld1_s16(input + 4); + d4 = vld1_s16(input + 8); + d5 = vld1_s16(input + 12); + + // 1st for loop + q1s16 = vcombine_s16(d2, d4); // Swap d3 d4 here + q2s16 = vcombine_s16(d3, d5); + + q3s16 = vqdmulhq_n_s16(q2s16, sinpi8sqrt2); + q4s16 = vqdmulhq_n_s16(q2s16, cospi8sqrt2minus1); + + d12 = vqadd_s16(vget_low_s16(q1s16), vget_high_s16(q1s16)); // a1 + d13 = vqsub_s16(vget_low_s16(q1s16), vget_high_s16(q1s16)); // b1 + + q3s16 = vshrq_n_s16(q3s16, 1); + q4s16 = vshrq_n_s16(q4s16, 1); + + q3s16 = vqaddq_s16(q3s16, q2s16); + q4s16 = vqaddq_s16(q4s16, q2s16); + + d10 = vqsub_s16(vget_low_s16(q3s16), vget_high_s16(q4s16)); // c1 + d11 = vqadd_s16(vget_high_s16(q3s16), vget_low_s16(q4s16)); // d1 + + d2 = vqadd_s16(d12, d11); + d3 = vqadd_s16(d13, d10); + d4 = vqsub_s16(d13, d10); + d5 = vqsub_s16(d12, d11); + + v2tmp0 = vtrn_s32(vreinterpret_s32_s16(d2), vreinterpret_s32_s16(d4)); + v2tmp1 = vtrn_s32(vreinterpret_s32_s16(d3), vreinterpret_s32_s16(d5)); + v2tmp2 = vtrn_s16(vreinterpret_s16_s32(v2tmp0.val[0]), + vreinterpret_s16_s32(v2tmp1.val[0])); + v2tmp3 = vtrn_s16(vreinterpret_s16_s32(v2tmp0.val[1]), + vreinterpret_s16_s32(v2tmp1.val[1])); + + // 2nd for loop + q1s16 = vcombine_s16(v2tmp2.val[0], v2tmp3.val[0]); + q2s16 = vcombine_s16(v2tmp2.val[1], v2tmp3.val[1]); + + q3s16 = vqdmulhq_n_s16(q2s16, sinpi8sqrt2); + q4s16 = vqdmulhq_n_s16(q2s16, cospi8sqrt2minus1); + + d12 = vqadd_s16(vget_low_s16(q1s16), vget_high_s16(q1s16)); // a1 + d13 = vqsub_s16(vget_low_s16(q1s16), vget_high_s16(q1s16)); // b1 + + q3s16 = vshrq_n_s16(q3s16, 1); + q4s16 = vshrq_n_s16(q4s16, 1); + + q3s16 = vqaddq_s16(q3s16, q2s16); + q4s16 = vqaddq_s16(q4s16, q2s16); + + d10 = vqsub_s16(vget_low_s16(q3s16), vget_high_s16(q4s16)); // c1 + d11 = vqadd_s16(vget_high_s16(q3s16), vget_low_s16(q4s16)); // d1 + + d2 = vqadd_s16(d12, d11); + d3 = vqadd_s16(d13, d10); + d4 = vqsub_s16(d13, d10); + d5 = vqsub_s16(d12, d11); + + d2 = vrshr_n_s16(d2, 3); + d3 = vrshr_n_s16(d3, 3); + d4 = vrshr_n_s16(d4, 3); + d5 = vrshr_n_s16(d5, 3); + + v2tmp0 = vtrn_s32(vreinterpret_s32_s16(d2), vreinterpret_s32_s16(d4)); + v2tmp1 = vtrn_s32(vreinterpret_s32_s16(d3), vreinterpret_s32_s16(d5)); + v2tmp2 = vtrn_s16(vreinterpret_s16_s32(v2tmp0.val[0]), + vreinterpret_s16_s32(v2tmp1.val[0])); + v2tmp3 = vtrn_s16(vreinterpret_s16_s32(v2tmp0.val[1]), + vreinterpret_s16_s32(v2tmp1.val[1])); + + q1s16 = vcombine_s16(v2tmp2.val[0], v2tmp2.val[1]); + q2s16 = vcombine_s16(v2tmp3.val[0], v2tmp3.val[1]); + + // dc_only_idct_add + for (i = 0; i < 2; i++, q1s16 = q2s16) { + d6u32 = vld1_lane_u32((const uint32_t *)pred_ptr, d6u32, 0); + pred_ptr += pred_stride; + d6u32 = vld1_lane_u32((const uint32_t *)pred_ptr, d6u32, 1); + pred_ptr += pred_stride; + + q1u16 = vaddw_u8(vreinterpretq_u16_s16(q1s16), + vreinterpret_u8_u32(d6u32)); + d1u8 = vqmovun_s16(vreinterpretq_s16_u16(q1u16)); + + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d1u8), 0); + dst_ptr += dst_stride; + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d1u8), 1); + dst_ptr += dst_stride; + } + return; +} diff --git a/thirdparty/libvpx/vp8/common/arm/neon/sixtappredict_neon.c b/thirdparty/libvpx/vp8/common/arm/neon/sixtappredict_neon.c new file mode 100644 index 0000000000..49d8d221fc --- /dev/null +++ b/thirdparty/libvpx/vp8/common/arm/neon/sixtappredict_neon.c @@ -0,0 +1,1377 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> +#include "vpx_ports/mem.h" + +static const int8_t vp8_sub_pel_filters[8][8] = { + {0, 0, 128, 0, 0, 0, 0, 0}, /* note that 1/8 pel positionyys are */ + {0, -6, 123, 12, -1, 0, 0, 0}, /* just as per alpha -0.5 bicubic */ + {2, -11, 108, 36, -8, 1, 0, 0}, /* New 1/4 pel 6 tap filter */ + {0, -9, 93, 50, -6, 0, 0, 0}, + {3, -16, 77, 77, -16, 3, 0, 0}, /* New 1/2 pel 6 tap filter */ + {0, -6, 50, 93, -9, 0, 0, 0}, + {1, -8, 36, 108, -11, 2, 0, 0}, /* New 1/4 pel 6 tap filter */ + {0, -1, 12, 123, -6, 0, 0, 0}, +}; + +void vp8_sixtap_predict8x4_neon( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch) { + unsigned char *src; + uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8, d7u8, d8u8, d9u8; + uint8x8_t d22u8, d23u8, d24u8, d25u8, d26u8; + uint8x8_t d27u8, d28u8, d29u8, d30u8, d31u8; + int8x8_t dtmps8, d0s8, d1s8, d2s8, d3s8, d4s8, d5s8; + uint16x8_t q3u16, q4u16, q5u16, q6u16, q7u16; + uint16x8_t q8u16, q9u16, q10u16, q11u16, q12u16; + int16x8_t q3s16, q4s16, q5s16, q6s16, q7s16; + int16x8_t q8s16, q9s16, q10s16, q11s16, q12s16; + uint8x16_t q3u8, q4u8, q5u8, q6u8, q7u8; + + if (xoffset == 0) { // secondpass_filter8x4_only + // load second_pass filter + dtmps8 = vld1_s8(vp8_sub_pel_filters[yoffset]); + d0s8 = vdup_lane_s8(dtmps8, 0); + d1s8 = vdup_lane_s8(dtmps8, 1); + d2s8 = vdup_lane_s8(dtmps8, 2); + d3s8 = vdup_lane_s8(dtmps8, 3); + d4s8 = vdup_lane_s8(dtmps8, 4); + d5s8 = vdup_lane_s8(dtmps8, 5); + d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8)); + d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8)); + d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8)); + d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8)); + d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8)); + d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8)); + + // load src data + src = src_ptr - src_pixels_per_line * 2; + d22u8 = vld1_u8(src); + src += src_pixels_per_line; + d23u8 = vld1_u8(src); + src += src_pixels_per_line; + d24u8 = vld1_u8(src); + src += src_pixels_per_line; + d25u8 = vld1_u8(src); + src += src_pixels_per_line; + d26u8 = vld1_u8(src); + src += src_pixels_per_line; + d27u8 = vld1_u8(src); + src += src_pixels_per_line; + d28u8 = vld1_u8(src); + src += src_pixels_per_line; + d29u8 = vld1_u8(src); + src += src_pixels_per_line; + d30u8 = vld1_u8(src); + + q3u16 = vmull_u8(d22u8, d0u8); + q4u16 = vmull_u8(d23u8, d0u8); + q5u16 = vmull_u8(d24u8, d0u8); + q6u16 = vmull_u8(d25u8, d0u8); + + q3u16 = vmlsl_u8(q3u16, d23u8, d1u8); + q4u16 = vmlsl_u8(q4u16, d24u8, d1u8); + q5u16 = vmlsl_u8(q5u16, d25u8, d1u8); + q6u16 = vmlsl_u8(q6u16, d26u8, d1u8); + + q3u16 = vmlsl_u8(q3u16, d26u8, d4u8); + q4u16 = vmlsl_u8(q4u16, d27u8, d4u8); + q5u16 = vmlsl_u8(q5u16, d28u8, d4u8); + q6u16 = vmlsl_u8(q6u16, d29u8, d4u8); + + q3u16 = vmlal_u8(q3u16, d24u8, d2u8); + q4u16 = vmlal_u8(q4u16, d25u8, d2u8); + q5u16 = vmlal_u8(q5u16, d26u8, d2u8); + q6u16 = vmlal_u8(q6u16, d27u8, d2u8); + + q3u16 = vmlal_u8(q3u16, d27u8, d5u8); + q4u16 = vmlal_u8(q4u16, d28u8, d5u8); + q5u16 = vmlal_u8(q5u16, d29u8, d5u8); + q6u16 = vmlal_u8(q6u16, d30u8, d5u8); + + q7u16 = vmull_u8(d25u8, d3u8); + q8u16 = vmull_u8(d26u8, d3u8); + q9u16 = vmull_u8(d27u8, d3u8); + q10u16 = vmull_u8(d28u8, d3u8); + + q3s16 = vreinterpretq_s16_u16(q3u16); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + + q7s16 = vqaddq_s16(q7s16, q3s16); + q8s16 = vqaddq_s16(q8s16, q4s16); + q9s16 = vqaddq_s16(q9s16, q5s16); + q10s16 = vqaddq_s16(q10s16, q6s16); + + d6u8 = vqrshrun_n_s16(q7s16, 7); + d7u8 = vqrshrun_n_s16(q8s16, 7); + d8u8 = vqrshrun_n_s16(q9s16, 7); + d9u8 = vqrshrun_n_s16(q10s16, 7); + + vst1_u8(dst_ptr, d6u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d7u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d8u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d9u8); + return; + } + + // load first_pass filter + dtmps8 = vld1_s8(vp8_sub_pel_filters[xoffset]); + d0s8 = vdup_lane_s8(dtmps8, 0); + d1s8 = vdup_lane_s8(dtmps8, 1); + d2s8 = vdup_lane_s8(dtmps8, 2); + d3s8 = vdup_lane_s8(dtmps8, 3); + d4s8 = vdup_lane_s8(dtmps8, 4); + d5s8 = vdup_lane_s8(dtmps8, 5); + d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8)); + d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8)); + d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8)); + d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8)); + d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8)); + d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8)); + + // First pass: output_height lines x output_width columns (9x4) + if (yoffset == 0) // firstpass_filter4x4_only + src = src_ptr - 2; + else + src = src_ptr - 2 - (src_pixels_per_line * 2); + q3u8 = vld1q_u8(src); + src += src_pixels_per_line; + q4u8 = vld1q_u8(src); + src += src_pixels_per_line; + q5u8 = vld1q_u8(src); + src += src_pixels_per_line; + q6u8 = vld1q_u8(src); + + q7u16 = vmull_u8(vget_low_u8(q3u8), d0u8); + q8u16 = vmull_u8(vget_low_u8(q4u8), d0u8); + q9u16 = vmull_u8(vget_low_u8(q5u8), d0u8); + q10u16 = vmull_u8(vget_low_u8(q6u8), d0u8); + + d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 1); + d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 1); + d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 1); + d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 1); + + q7u16 = vmlsl_u8(q7u16, d28u8, d1u8); + q8u16 = vmlsl_u8(q8u16, d29u8, d1u8); + q9u16 = vmlsl_u8(q9u16, d30u8, d1u8); + q10u16 = vmlsl_u8(q10u16, d31u8, d1u8); + + d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 4); + d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 4); + d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 4); + d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 4); + + q7u16 = vmlsl_u8(q7u16, d28u8, d4u8); + q8u16 = vmlsl_u8(q8u16, d29u8, d4u8); + q9u16 = vmlsl_u8(q9u16, d30u8, d4u8); + q10u16 = vmlsl_u8(q10u16, d31u8, d4u8); + + d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 2); + d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 2); + d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 2); + d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 2); + + q7u16 = vmlal_u8(q7u16, d28u8, d2u8); + q8u16 = vmlal_u8(q8u16, d29u8, d2u8); + q9u16 = vmlal_u8(q9u16, d30u8, d2u8); + q10u16 = vmlal_u8(q10u16, d31u8, d2u8); + + d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 5); + d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 5); + d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 5); + d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 5); + + q7u16 = vmlal_u8(q7u16, d28u8, d5u8); + q8u16 = vmlal_u8(q8u16, d29u8, d5u8); + q9u16 = vmlal_u8(q9u16, d30u8, d5u8); + q10u16 = vmlal_u8(q10u16, d31u8, d5u8); + + d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 3); + d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 3); + d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 3); + d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 3); + + q3u16 = vmull_u8(d28u8, d3u8); + q4u16 = vmull_u8(d29u8, d3u8); + q5u16 = vmull_u8(d30u8, d3u8); + q6u16 = vmull_u8(d31u8, d3u8); + + q3s16 = vreinterpretq_s16_u16(q3u16); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + + q7s16 = vqaddq_s16(q7s16, q3s16); + q8s16 = vqaddq_s16(q8s16, q4s16); + q9s16 = vqaddq_s16(q9s16, q5s16); + q10s16 = vqaddq_s16(q10s16, q6s16); + + d22u8 = vqrshrun_n_s16(q7s16, 7); + d23u8 = vqrshrun_n_s16(q8s16, 7); + d24u8 = vqrshrun_n_s16(q9s16, 7); + d25u8 = vqrshrun_n_s16(q10s16, 7); + + if (yoffset == 0) { // firstpass_filter8x4_only + vst1_u8(dst_ptr, d22u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d23u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d24u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d25u8); + return; + } + + // First Pass on rest 5-line data + src += src_pixels_per_line; + q3u8 = vld1q_u8(src); + src += src_pixels_per_line; + q4u8 = vld1q_u8(src); + src += src_pixels_per_line; + q5u8 = vld1q_u8(src); + src += src_pixels_per_line; + q6u8 = vld1q_u8(src); + src += src_pixels_per_line; + q7u8 = vld1q_u8(src); + + q8u16 = vmull_u8(vget_low_u8(q3u8), d0u8); + q9u16 = vmull_u8(vget_low_u8(q4u8), d0u8); + q10u16 = vmull_u8(vget_low_u8(q5u8), d0u8); + q11u16 = vmull_u8(vget_low_u8(q6u8), d0u8); + q12u16 = vmull_u8(vget_low_u8(q7u8), d0u8); + + d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 1); + d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 1); + d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 1); + d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 1); + d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 1); + + q8u16 = vmlsl_u8(q8u16, d27u8, d1u8); + q9u16 = vmlsl_u8(q9u16, d28u8, d1u8); + q10u16 = vmlsl_u8(q10u16, d29u8, d1u8); + q11u16 = vmlsl_u8(q11u16, d30u8, d1u8); + q12u16 = vmlsl_u8(q12u16, d31u8, d1u8); + + d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 4); + d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 4); + d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 4); + d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 4); + d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 4); + + q8u16 = vmlsl_u8(q8u16, d27u8, d4u8); + q9u16 = vmlsl_u8(q9u16, d28u8, d4u8); + q10u16 = vmlsl_u8(q10u16, d29u8, d4u8); + q11u16 = vmlsl_u8(q11u16, d30u8, d4u8); + q12u16 = vmlsl_u8(q12u16, d31u8, d4u8); + + d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 2); + d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 2); + d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 2); + d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 2); + d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 2); + + q8u16 = vmlal_u8(q8u16, d27u8, d2u8); + q9u16 = vmlal_u8(q9u16, d28u8, d2u8); + q10u16 = vmlal_u8(q10u16, d29u8, d2u8); + q11u16 = vmlal_u8(q11u16, d30u8, d2u8); + q12u16 = vmlal_u8(q12u16, d31u8, d2u8); + + d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 5); + d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 5); + d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 5); + d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 5); + d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 5); + + q8u16 = vmlal_u8(q8u16, d27u8, d5u8); + q9u16 = vmlal_u8(q9u16, d28u8, d5u8); + q10u16 = vmlal_u8(q10u16, d29u8, d5u8); + q11u16 = vmlal_u8(q11u16, d30u8, d5u8); + q12u16 = vmlal_u8(q12u16, d31u8, d5u8); + + d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 3); + d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 3); + d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 3); + d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 3); + d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 3); + + q3u16 = vmull_u8(d27u8, d3u8); + q4u16 = vmull_u8(d28u8, d3u8); + q5u16 = vmull_u8(d29u8, d3u8); + q6u16 = vmull_u8(d30u8, d3u8); + q7u16 = vmull_u8(d31u8, d3u8); + + q3s16 = vreinterpretq_s16_u16(q3u16); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + q11s16 = vreinterpretq_s16_u16(q11u16); + q12s16 = vreinterpretq_s16_u16(q12u16); + + q8s16 = vqaddq_s16(q8s16, q3s16); + q9s16 = vqaddq_s16(q9s16, q4s16); + q10s16 = vqaddq_s16(q10s16, q5s16); + q11s16 = vqaddq_s16(q11s16, q6s16); + q12s16 = vqaddq_s16(q12s16, q7s16); + + d26u8 = vqrshrun_n_s16(q8s16, 7); + d27u8 = vqrshrun_n_s16(q9s16, 7); + d28u8 = vqrshrun_n_s16(q10s16, 7); + d29u8 = vqrshrun_n_s16(q11s16, 7); + d30u8 = vqrshrun_n_s16(q12s16, 7); + + // Second pass: 8x4 + dtmps8 = vld1_s8(vp8_sub_pel_filters[yoffset]); + d0s8 = vdup_lane_s8(dtmps8, 0); + d1s8 = vdup_lane_s8(dtmps8, 1); + d2s8 = vdup_lane_s8(dtmps8, 2); + d3s8 = vdup_lane_s8(dtmps8, 3); + d4s8 = vdup_lane_s8(dtmps8, 4); + d5s8 = vdup_lane_s8(dtmps8, 5); + d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8)); + d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8)); + d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8)); + d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8)); + d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8)); + d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8)); + + q3u16 = vmull_u8(d22u8, d0u8); + q4u16 = vmull_u8(d23u8, d0u8); + q5u16 = vmull_u8(d24u8, d0u8); + q6u16 = vmull_u8(d25u8, d0u8); + + q3u16 = vmlsl_u8(q3u16, d23u8, d1u8); + q4u16 = vmlsl_u8(q4u16, d24u8, d1u8); + q5u16 = vmlsl_u8(q5u16, d25u8, d1u8); + q6u16 = vmlsl_u8(q6u16, d26u8, d1u8); + + q3u16 = vmlsl_u8(q3u16, d26u8, d4u8); + q4u16 = vmlsl_u8(q4u16, d27u8, d4u8); + q5u16 = vmlsl_u8(q5u16, d28u8, d4u8); + q6u16 = vmlsl_u8(q6u16, d29u8, d4u8); + + q3u16 = vmlal_u8(q3u16, d24u8, d2u8); + q4u16 = vmlal_u8(q4u16, d25u8, d2u8); + q5u16 = vmlal_u8(q5u16, d26u8, d2u8); + q6u16 = vmlal_u8(q6u16, d27u8, d2u8); + + q3u16 = vmlal_u8(q3u16, d27u8, d5u8); + q4u16 = vmlal_u8(q4u16, d28u8, d5u8); + q5u16 = vmlal_u8(q5u16, d29u8, d5u8); + q6u16 = vmlal_u8(q6u16, d30u8, d5u8); + + q7u16 = vmull_u8(d25u8, d3u8); + q8u16 = vmull_u8(d26u8, d3u8); + q9u16 = vmull_u8(d27u8, d3u8); + q10u16 = vmull_u8(d28u8, d3u8); + + q3s16 = vreinterpretq_s16_u16(q3u16); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + + q7s16 = vqaddq_s16(q7s16, q3s16); + q8s16 = vqaddq_s16(q8s16, q4s16); + q9s16 = vqaddq_s16(q9s16, q5s16); + q10s16 = vqaddq_s16(q10s16, q6s16); + + d6u8 = vqrshrun_n_s16(q7s16, 7); + d7u8 = vqrshrun_n_s16(q8s16, 7); + d8u8 = vqrshrun_n_s16(q9s16, 7); + d9u8 = vqrshrun_n_s16(q10s16, 7); + + vst1_u8(dst_ptr, d6u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d7u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d8u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d9u8); + return; +} + +void vp8_sixtap_predict8x8_neon( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch) { + unsigned char *src, *tmpp; + unsigned char tmp[64]; + int i; + uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8, d7u8, d8u8, d9u8; + uint8x8_t d18u8, d19u8, d20u8, d21u8, d22u8, d23u8, d24u8, d25u8; + uint8x8_t d26u8, d27u8, d28u8, d29u8, d30u8, d31u8; + int8x8_t dtmps8, d0s8, d1s8, d2s8, d3s8, d4s8, d5s8; + uint16x8_t q3u16, q4u16, q5u16, q6u16, q7u16; + uint16x8_t q8u16, q9u16, q10u16, q11u16, q12u16; + int16x8_t q3s16, q4s16, q5s16, q6s16, q7s16; + int16x8_t q8s16, q9s16, q10s16, q11s16, q12s16; + uint8x16_t q3u8, q4u8, q5u8, q6u8, q7u8, q9u8, q10u8, q11u8, q12u8; + + if (xoffset == 0) { // secondpass_filter8x8_only + // load second_pass filter + dtmps8 = vld1_s8(vp8_sub_pel_filters[yoffset]); + d0s8 = vdup_lane_s8(dtmps8, 0); + d1s8 = vdup_lane_s8(dtmps8, 1); + d2s8 = vdup_lane_s8(dtmps8, 2); + d3s8 = vdup_lane_s8(dtmps8, 3); + d4s8 = vdup_lane_s8(dtmps8, 4); + d5s8 = vdup_lane_s8(dtmps8, 5); + d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8)); + d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8)); + d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8)); + d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8)); + d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8)); + d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8)); + + // load src data + src = src_ptr - src_pixels_per_line * 2; + d18u8 = vld1_u8(src); + src += src_pixels_per_line; + d19u8 = vld1_u8(src); + src += src_pixels_per_line; + d20u8 = vld1_u8(src); + src += src_pixels_per_line; + d21u8 = vld1_u8(src); + src += src_pixels_per_line; + d22u8 = vld1_u8(src); + src += src_pixels_per_line; + d23u8 = vld1_u8(src); + src += src_pixels_per_line; + d24u8 = vld1_u8(src); + src += src_pixels_per_line; + d25u8 = vld1_u8(src); + src += src_pixels_per_line; + d26u8 = vld1_u8(src); + src += src_pixels_per_line; + d27u8 = vld1_u8(src); + src += src_pixels_per_line; + d28u8 = vld1_u8(src); + src += src_pixels_per_line; + d29u8 = vld1_u8(src); + src += src_pixels_per_line; + d30u8 = vld1_u8(src); + + for (i = 2; i > 0; i--) { + q3u16 = vmull_u8(d18u8, d0u8); + q4u16 = vmull_u8(d19u8, d0u8); + q5u16 = vmull_u8(d20u8, d0u8); + q6u16 = vmull_u8(d21u8, d0u8); + + q3u16 = vmlsl_u8(q3u16, d19u8, d1u8); + q4u16 = vmlsl_u8(q4u16, d20u8, d1u8); + q5u16 = vmlsl_u8(q5u16, d21u8, d1u8); + q6u16 = vmlsl_u8(q6u16, d22u8, d1u8); + + q3u16 = vmlsl_u8(q3u16, d22u8, d4u8); + q4u16 = vmlsl_u8(q4u16, d23u8, d4u8); + q5u16 = vmlsl_u8(q5u16, d24u8, d4u8); + q6u16 = vmlsl_u8(q6u16, d25u8, d4u8); + + q3u16 = vmlal_u8(q3u16, d20u8, d2u8); + q4u16 = vmlal_u8(q4u16, d21u8, d2u8); + q5u16 = vmlal_u8(q5u16, d22u8, d2u8); + q6u16 = vmlal_u8(q6u16, d23u8, d2u8); + + q3u16 = vmlal_u8(q3u16, d23u8, d5u8); + q4u16 = vmlal_u8(q4u16, d24u8, d5u8); + q5u16 = vmlal_u8(q5u16, d25u8, d5u8); + q6u16 = vmlal_u8(q6u16, d26u8, d5u8); + + q7u16 = vmull_u8(d21u8, d3u8); + q8u16 = vmull_u8(d22u8, d3u8); + q9u16 = vmull_u8(d23u8, d3u8); + q10u16 = vmull_u8(d24u8, d3u8); + + q3s16 = vreinterpretq_s16_u16(q3u16); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + + q7s16 = vqaddq_s16(q7s16, q3s16); + q8s16 = vqaddq_s16(q8s16, q4s16); + q9s16 = vqaddq_s16(q9s16, q5s16); + q10s16 = vqaddq_s16(q10s16, q6s16); + + d6u8 = vqrshrun_n_s16(q7s16, 7); + d7u8 = vqrshrun_n_s16(q8s16, 7); + d8u8 = vqrshrun_n_s16(q9s16, 7); + d9u8 = vqrshrun_n_s16(q10s16, 7); + + d18u8 = d22u8; + d19u8 = d23u8; + d20u8 = d24u8; + d21u8 = d25u8; + d22u8 = d26u8; + d23u8 = d27u8; + d24u8 = d28u8; + d25u8 = d29u8; + d26u8 = d30u8; + + vst1_u8(dst_ptr, d6u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d7u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d8u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d9u8); + dst_ptr += dst_pitch; + } + return; + } + + // load first_pass filter + dtmps8 = vld1_s8(vp8_sub_pel_filters[xoffset]); + d0s8 = vdup_lane_s8(dtmps8, 0); + d1s8 = vdup_lane_s8(dtmps8, 1); + d2s8 = vdup_lane_s8(dtmps8, 2); + d3s8 = vdup_lane_s8(dtmps8, 3); + d4s8 = vdup_lane_s8(dtmps8, 4); + d5s8 = vdup_lane_s8(dtmps8, 5); + d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8)); + d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8)); + d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8)); + d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8)); + d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8)); + d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8)); + + // First pass: output_height lines x output_width columns (9x4) + if (yoffset == 0) // firstpass_filter4x4_only + src = src_ptr - 2; + else + src = src_ptr - 2 - (src_pixels_per_line * 2); + + tmpp = tmp; + for (i = 2; i > 0; i--) { + q3u8 = vld1q_u8(src); + src += src_pixels_per_line; + q4u8 = vld1q_u8(src); + src += src_pixels_per_line; + q5u8 = vld1q_u8(src); + src += src_pixels_per_line; + q6u8 = vld1q_u8(src); + src += src_pixels_per_line; + + __builtin_prefetch(src); + __builtin_prefetch(src + src_pixels_per_line); + __builtin_prefetch(src + src_pixels_per_line * 2); + + q7u16 = vmull_u8(vget_low_u8(q3u8), d0u8); + q8u16 = vmull_u8(vget_low_u8(q4u8), d0u8); + q9u16 = vmull_u8(vget_low_u8(q5u8), d0u8); + q10u16 = vmull_u8(vget_low_u8(q6u8), d0u8); + + d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 1); + d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 1); + d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 1); + d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 1); + + q7u16 = vmlsl_u8(q7u16, d28u8, d1u8); + q8u16 = vmlsl_u8(q8u16, d29u8, d1u8); + q9u16 = vmlsl_u8(q9u16, d30u8, d1u8); + q10u16 = vmlsl_u8(q10u16, d31u8, d1u8); + + d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 4); + d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 4); + d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 4); + d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 4); + + q7u16 = vmlsl_u8(q7u16, d28u8, d4u8); + q8u16 = vmlsl_u8(q8u16, d29u8, d4u8); + q9u16 = vmlsl_u8(q9u16, d30u8, d4u8); + q10u16 = vmlsl_u8(q10u16, d31u8, d4u8); + + d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 2); + d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 2); + d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 2); + d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 2); + + q7u16 = vmlal_u8(q7u16, d28u8, d2u8); + q8u16 = vmlal_u8(q8u16, d29u8, d2u8); + q9u16 = vmlal_u8(q9u16, d30u8, d2u8); + q10u16 = vmlal_u8(q10u16, d31u8, d2u8); + + d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 5); + d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 5); + d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 5); + d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 5); + + q7u16 = vmlal_u8(q7u16, d28u8, d5u8); + q8u16 = vmlal_u8(q8u16, d29u8, d5u8); + q9u16 = vmlal_u8(q9u16, d30u8, d5u8); + q10u16 = vmlal_u8(q10u16, d31u8, d5u8); + + d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 3); + d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 3); + d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 3); + d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 3); + + q3u16 = vmull_u8(d28u8, d3u8); + q4u16 = vmull_u8(d29u8, d3u8); + q5u16 = vmull_u8(d30u8, d3u8); + q6u16 = vmull_u8(d31u8, d3u8); + + q3s16 = vreinterpretq_s16_u16(q3u16); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + + q7s16 = vqaddq_s16(q7s16, q3s16); + q8s16 = vqaddq_s16(q8s16, q4s16); + q9s16 = vqaddq_s16(q9s16, q5s16); + q10s16 = vqaddq_s16(q10s16, q6s16); + + d22u8 = vqrshrun_n_s16(q7s16, 7); + d23u8 = vqrshrun_n_s16(q8s16, 7); + d24u8 = vqrshrun_n_s16(q9s16, 7); + d25u8 = vqrshrun_n_s16(q10s16, 7); + + if (yoffset == 0) { // firstpass_filter8x4_only + vst1_u8(dst_ptr, d22u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d23u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d24u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d25u8); + dst_ptr += dst_pitch; + } else { + vst1_u8(tmpp, d22u8); + tmpp += 8; + vst1_u8(tmpp, d23u8); + tmpp += 8; + vst1_u8(tmpp, d24u8); + tmpp += 8; + vst1_u8(tmpp, d25u8); + tmpp += 8; + } + } + if (yoffset == 0) + return; + + // First Pass on rest 5-line data + q3u8 = vld1q_u8(src); + src += src_pixels_per_line; + q4u8 = vld1q_u8(src); + src += src_pixels_per_line; + q5u8 = vld1q_u8(src); + src += src_pixels_per_line; + q6u8 = vld1q_u8(src); + src += src_pixels_per_line; + q7u8 = vld1q_u8(src); + + q8u16 = vmull_u8(vget_low_u8(q3u8), d0u8); + q9u16 = vmull_u8(vget_low_u8(q4u8), d0u8); + q10u16 = vmull_u8(vget_low_u8(q5u8), d0u8); + q11u16 = vmull_u8(vget_low_u8(q6u8), d0u8); + q12u16 = vmull_u8(vget_low_u8(q7u8), d0u8); + + d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 1); + d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 1); + d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 1); + d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 1); + d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 1); + + q8u16 = vmlsl_u8(q8u16, d27u8, d1u8); + q9u16 = vmlsl_u8(q9u16, d28u8, d1u8); + q10u16 = vmlsl_u8(q10u16, d29u8, d1u8); + q11u16 = vmlsl_u8(q11u16, d30u8, d1u8); + q12u16 = vmlsl_u8(q12u16, d31u8, d1u8); + + d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 4); + d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 4); + d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 4); + d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 4); + d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 4); + + q8u16 = vmlsl_u8(q8u16, d27u8, d4u8); + q9u16 = vmlsl_u8(q9u16, d28u8, d4u8); + q10u16 = vmlsl_u8(q10u16, d29u8, d4u8); + q11u16 = vmlsl_u8(q11u16, d30u8, d4u8); + q12u16 = vmlsl_u8(q12u16, d31u8, d4u8); + + d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 2); + d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 2); + d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 2); + d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 2); + d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 2); + + q8u16 = vmlal_u8(q8u16, d27u8, d2u8); + q9u16 = vmlal_u8(q9u16, d28u8, d2u8); + q10u16 = vmlal_u8(q10u16, d29u8, d2u8); + q11u16 = vmlal_u8(q11u16, d30u8, d2u8); + q12u16 = vmlal_u8(q12u16, d31u8, d2u8); + + d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 5); + d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 5); + d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 5); + d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 5); + d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 5); + + q8u16 = vmlal_u8(q8u16, d27u8, d5u8); + q9u16 = vmlal_u8(q9u16, d28u8, d5u8); + q10u16 = vmlal_u8(q10u16, d29u8, d5u8); + q11u16 = vmlal_u8(q11u16, d30u8, d5u8); + q12u16 = vmlal_u8(q12u16, d31u8, d5u8); + + d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 3); + d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 3); + d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 3); + d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 3); + d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 3); + + q3u16 = vmull_u8(d27u8, d3u8); + q4u16 = vmull_u8(d28u8, d3u8); + q5u16 = vmull_u8(d29u8, d3u8); + q6u16 = vmull_u8(d30u8, d3u8); + q7u16 = vmull_u8(d31u8, d3u8); + + q3s16 = vreinterpretq_s16_u16(q3u16); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + q11s16 = vreinterpretq_s16_u16(q11u16); + q12s16 = vreinterpretq_s16_u16(q12u16); + + q8s16 = vqaddq_s16(q8s16, q3s16); + q9s16 = vqaddq_s16(q9s16, q4s16); + q10s16 = vqaddq_s16(q10s16, q5s16); + q11s16 = vqaddq_s16(q11s16, q6s16); + q12s16 = vqaddq_s16(q12s16, q7s16); + + d26u8 = vqrshrun_n_s16(q8s16, 7); + d27u8 = vqrshrun_n_s16(q9s16, 7); + d28u8 = vqrshrun_n_s16(q10s16, 7); + d29u8 = vqrshrun_n_s16(q11s16, 7); + d30u8 = vqrshrun_n_s16(q12s16, 7); + + // Second pass: 8x8 + dtmps8 = vld1_s8(vp8_sub_pel_filters[yoffset]); + d0s8 = vdup_lane_s8(dtmps8, 0); + d1s8 = vdup_lane_s8(dtmps8, 1); + d2s8 = vdup_lane_s8(dtmps8, 2); + d3s8 = vdup_lane_s8(dtmps8, 3); + d4s8 = vdup_lane_s8(dtmps8, 4); + d5s8 = vdup_lane_s8(dtmps8, 5); + d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8)); + d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8)); + d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8)); + d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8)); + d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8)); + d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8)); + + tmpp = tmp; + q9u8 = vld1q_u8(tmpp); + tmpp += 16; + q10u8 = vld1q_u8(tmpp); + tmpp += 16; + q11u8 = vld1q_u8(tmpp); + tmpp += 16; + q12u8 = vld1q_u8(tmpp); + + d18u8 = vget_low_u8(q9u8); + d19u8 = vget_high_u8(q9u8); + d20u8 = vget_low_u8(q10u8); + d21u8 = vget_high_u8(q10u8); + d22u8 = vget_low_u8(q11u8); + d23u8 = vget_high_u8(q11u8); + d24u8 = vget_low_u8(q12u8); + d25u8 = vget_high_u8(q12u8); + + for (i = 2; i > 0; i--) { + q3u16 = vmull_u8(d18u8, d0u8); + q4u16 = vmull_u8(d19u8, d0u8); + q5u16 = vmull_u8(d20u8, d0u8); + q6u16 = vmull_u8(d21u8, d0u8); + + q3u16 = vmlsl_u8(q3u16, d19u8, d1u8); + q4u16 = vmlsl_u8(q4u16, d20u8, d1u8); + q5u16 = vmlsl_u8(q5u16, d21u8, d1u8); + q6u16 = vmlsl_u8(q6u16, d22u8, d1u8); + + q3u16 = vmlsl_u8(q3u16, d22u8, d4u8); + q4u16 = vmlsl_u8(q4u16, d23u8, d4u8); + q5u16 = vmlsl_u8(q5u16, d24u8, d4u8); + q6u16 = vmlsl_u8(q6u16, d25u8, d4u8); + + q3u16 = vmlal_u8(q3u16, d20u8, d2u8); + q4u16 = vmlal_u8(q4u16, d21u8, d2u8); + q5u16 = vmlal_u8(q5u16, d22u8, d2u8); + q6u16 = vmlal_u8(q6u16, d23u8, d2u8); + + q3u16 = vmlal_u8(q3u16, d23u8, d5u8); + q4u16 = vmlal_u8(q4u16, d24u8, d5u8); + q5u16 = vmlal_u8(q5u16, d25u8, d5u8); + q6u16 = vmlal_u8(q6u16, d26u8, d5u8); + + q7u16 = vmull_u8(d21u8, d3u8); + q8u16 = vmull_u8(d22u8, d3u8); + q9u16 = vmull_u8(d23u8, d3u8); + q10u16 = vmull_u8(d24u8, d3u8); + + q3s16 = vreinterpretq_s16_u16(q3u16); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + + q7s16 = vqaddq_s16(q7s16, q3s16); + q8s16 = vqaddq_s16(q8s16, q4s16); + q9s16 = vqaddq_s16(q9s16, q5s16); + q10s16 = vqaddq_s16(q10s16, q6s16); + + d6u8 = vqrshrun_n_s16(q7s16, 7); + d7u8 = vqrshrun_n_s16(q8s16, 7); + d8u8 = vqrshrun_n_s16(q9s16, 7); + d9u8 = vqrshrun_n_s16(q10s16, 7); + + d18u8 = d22u8; + d19u8 = d23u8; + d20u8 = d24u8; + d21u8 = d25u8; + d22u8 = d26u8; + d23u8 = d27u8; + d24u8 = d28u8; + d25u8 = d29u8; + d26u8 = d30u8; + + vst1_u8(dst_ptr, d6u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d7u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d8u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d9u8); + dst_ptr += dst_pitch; + } + return; +} + +void vp8_sixtap_predict16x16_neon( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch) { + unsigned char *src, *src_tmp, *dst, *tmpp; + unsigned char tmp[336]; + int i, j; + uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8, d7u8, d8u8, d9u8; + uint8x8_t d10u8, d11u8, d12u8, d13u8, d14u8, d15u8, d18u8, d19u8; + uint8x8_t d20u8, d21u8, d22u8, d23u8, d24u8, d25u8, d26u8, d27u8; + uint8x8_t d28u8, d29u8, d30u8, d31u8; + int8x8_t dtmps8, d0s8, d1s8, d2s8, d3s8, d4s8, d5s8; + uint8x16_t q3u8, q4u8; + uint16x8_t q3u16, q4u16, q5u16, q6u16, q7u16, q8u16, q9u16, q10u16; + uint16x8_t q11u16, q12u16, q13u16, q15u16; + int16x8_t q3s16, q4s16, q5s16, q6s16, q7s16, q8s16, q9s16, q10s16; + int16x8_t q11s16, q12s16, q13s16, q15s16; + + if (xoffset == 0) { // secondpass_filter8x8_only + // load second_pass filter + dtmps8 = vld1_s8(vp8_sub_pel_filters[yoffset]); + d0s8 = vdup_lane_s8(dtmps8, 0); + d1s8 = vdup_lane_s8(dtmps8, 1); + d2s8 = vdup_lane_s8(dtmps8, 2); + d3s8 = vdup_lane_s8(dtmps8, 3); + d4s8 = vdup_lane_s8(dtmps8, 4); + d5s8 = vdup_lane_s8(dtmps8, 5); + d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8)); + d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8)); + d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8)); + d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8)); + d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8)); + d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8)); + + // load src data + src_tmp = src_ptr - src_pixels_per_line * 2; + for (i = 0; i < 2; i++) { + src = src_tmp + i * 8; + dst = dst_ptr + i * 8; + d18u8 = vld1_u8(src); + src += src_pixels_per_line; + d19u8 = vld1_u8(src); + src += src_pixels_per_line; + d20u8 = vld1_u8(src); + src += src_pixels_per_line; + d21u8 = vld1_u8(src); + src += src_pixels_per_line; + d22u8 = vld1_u8(src); + src += src_pixels_per_line; + for (j = 0; j < 4; j++) { + d23u8 = vld1_u8(src); + src += src_pixels_per_line; + d24u8 = vld1_u8(src); + src += src_pixels_per_line; + d25u8 = vld1_u8(src); + src += src_pixels_per_line; + d26u8 = vld1_u8(src); + src += src_pixels_per_line; + + q3u16 = vmull_u8(d18u8, d0u8); + q4u16 = vmull_u8(d19u8, d0u8); + q5u16 = vmull_u8(d20u8, d0u8); + q6u16 = vmull_u8(d21u8, d0u8); + + q3u16 = vmlsl_u8(q3u16, d19u8, d1u8); + q4u16 = vmlsl_u8(q4u16, d20u8, d1u8); + q5u16 = vmlsl_u8(q5u16, d21u8, d1u8); + q6u16 = vmlsl_u8(q6u16, d22u8, d1u8); + + q3u16 = vmlsl_u8(q3u16, d22u8, d4u8); + q4u16 = vmlsl_u8(q4u16, d23u8, d4u8); + q5u16 = vmlsl_u8(q5u16, d24u8, d4u8); + q6u16 = vmlsl_u8(q6u16, d25u8, d4u8); + + q3u16 = vmlal_u8(q3u16, d20u8, d2u8); + q4u16 = vmlal_u8(q4u16, d21u8, d2u8); + q5u16 = vmlal_u8(q5u16, d22u8, d2u8); + q6u16 = vmlal_u8(q6u16, d23u8, d2u8); + + q3u16 = vmlal_u8(q3u16, d23u8, d5u8); + q4u16 = vmlal_u8(q4u16, d24u8, d5u8); + q5u16 = vmlal_u8(q5u16, d25u8, d5u8); + q6u16 = vmlal_u8(q6u16, d26u8, d5u8); + + q7u16 = vmull_u8(d21u8, d3u8); + q8u16 = vmull_u8(d22u8, d3u8); + q9u16 = vmull_u8(d23u8, d3u8); + q10u16 = vmull_u8(d24u8, d3u8); + + q3s16 = vreinterpretq_s16_u16(q3u16); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + + q7s16 = vqaddq_s16(q7s16, q3s16); + q8s16 = vqaddq_s16(q8s16, q4s16); + q9s16 = vqaddq_s16(q9s16, q5s16); + q10s16 = vqaddq_s16(q10s16, q6s16); + + d6u8 = vqrshrun_n_s16(q7s16, 7); + d7u8 = vqrshrun_n_s16(q8s16, 7); + d8u8 = vqrshrun_n_s16(q9s16, 7); + d9u8 = vqrshrun_n_s16(q10s16, 7); + + d18u8 = d22u8; + d19u8 = d23u8; + d20u8 = d24u8; + d21u8 = d25u8; + d22u8 = d26u8; + + vst1_u8(dst, d6u8); + dst += dst_pitch; + vst1_u8(dst, d7u8); + dst += dst_pitch; + vst1_u8(dst, d8u8); + dst += dst_pitch; + vst1_u8(dst, d9u8); + dst += dst_pitch; + } + } + return; + } + + // load first_pass filter + dtmps8 = vld1_s8(vp8_sub_pel_filters[xoffset]); + d0s8 = vdup_lane_s8(dtmps8, 0); + d1s8 = vdup_lane_s8(dtmps8, 1); + d2s8 = vdup_lane_s8(dtmps8, 2); + d3s8 = vdup_lane_s8(dtmps8, 3); + d4s8 = vdup_lane_s8(dtmps8, 4); + d5s8 = vdup_lane_s8(dtmps8, 5); + d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8)); + d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8)); + d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8)); + d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8)); + d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8)); + d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8)); + + // First pass: output_height lines x output_width columns (9x4) + if (yoffset == 0) { // firstpass_filter4x4_only + src = src_ptr - 2; + dst = dst_ptr; + for (i = 0; i < 8; i++) { + d6u8 = vld1_u8(src); + d7u8 = vld1_u8(src + 8); + d8u8 = vld1_u8(src + 16); + src += src_pixels_per_line; + d9u8 = vld1_u8(src); + d10u8 = vld1_u8(src + 8); + d11u8 = vld1_u8(src + 16); + src += src_pixels_per_line; + + __builtin_prefetch(src); + __builtin_prefetch(src + src_pixels_per_line); + + q6u16 = vmull_u8(d6u8, d0u8); + q7u16 = vmull_u8(d7u8, d0u8); + q8u16 = vmull_u8(d9u8, d0u8); + q9u16 = vmull_u8(d10u8, d0u8); + + d20u8 = vext_u8(d6u8, d7u8, 1); + d21u8 = vext_u8(d9u8, d10u8, 1); + d22u8 = vext_u8(d7u8, d8u8, 1); + d23u8 = vext_u8(d10u8, d11u8, 1); + d24u8 = vext_u8(d6u8, d7u8, 4); + d25u8 = vext_u8(d9u8, d10u8, 4); + d26u8 = vext_u8(d7u8, d8u8, 4); + d27u8 = vext_u8(d10u8, d11u8, 4); + d28u8 = vext_u8(d6u8, d7u8, 5); + d29u8 = vext_u8(d9u8, d10u8, 5); + + q6u16 = vmlsl_u8(q6u16, d20u8, d1u8); + q8u16 = vmlsl_u8(q8u16, d21u8, d1u8); + q7u16 = vmlsl_u8(q7u16, d22u8, d1u8); + q9u16 = vmlsl_u8(q9u16, d23u8, d1u8); + q6u16 = vmlsl_u8(q6u16, d24u8, d4u8); + q8u16 = vmlsl_u8(q8u16, d25u8, d4u8); + q7u16 = vmlsl_u8(q7u16, d26u8, d4u8); + q9u16 = vmlsl_u8(q9u16, d27u8, d4u8); + q6u16 = vmlal_u8(q6u16, d28u8, d5u8); + q8u16 = vmlal_u8(q8u16, d29u8, d5u8); + + d20u8 = vext_u8(d7u8, d8u8, 5); + d21u8 = vext_u8(d10u8, d11u8, 5); + d22u8 = vext_u8(d6u8, d7u8, 2); + d23u8 = vext_u8(d9u8, d10u8, 2); + d24u8 = vext_u8(d7u8, d8u8, 2); + d25u8 = vext_u8(d10u8, d11u8, 2); + d26u8 = vext_u8(d6u8, d7u8, 3); + d27u8 = vext_u8(d9u8, d10u8, 3); + d28u8 = vext_u8(d7u8, d8u8, 3); + d29u8 = vext_u8(d10u8, d11u8, 3); + + q7u16 = vmlal_u8(q7u16, d20u8, d5u8); + q9u16 = vmlal_u8(q9u16, d21u8, d5u8); + q6u16 = vmlal_u8(q6u16, d22u8, d2u8); + q8u16 = vmlal_u8(q8u16, d23u8, d2u8); + q7u16 = vmlal_u8(q7u16, d24u8, d2u8); + q9u16 = vmlal_u8(q9u16, d25u8, d2u8); + + q10u16 = vmull_u8(d26u8, d3u8); + q11u16 = vmull_u8(d27u8, d3u8); + q12u16 = vmull_u8(d28u8, d3u8); + q15u16 = vmull_u8(d29u8, d3u8); + + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + q11s16 = vreinterpretq_s16_u16(q11u16); + q12s16 = vreinterpretq_s16_u16(q12u16); + q15s16 = vreinterpretq_s16_u16(q15u16); + + q6s16 = vqaddq_s16(q6s16, q10s16); + q8s16 = vqaddq_s16(q8s16, q11s16); + q7s16 = vqaddq_s16(q7s16, q12s16); + q9s16 = vqaddq_s16(q9s16, q15s16); + + d6u8 = vqrshrun_n_s16(q6s16, 7); + d7u8 = vqrshrun_n_s16(q7s16, 7); + d8u8 = vqrshrun_n_s16(q8s16, 7); + d9u8 = vqrshrun_n_s16(q9s16, 7); + + q3u8 = vcombine_u8(d6u8, d7u8); + q4u8 = vcombine_u8(d8u8, d9u8); + vst1q_u8(dst, q3u8); + dst += dst_pitch; + vst1q_u8(dst, q4u8); + dst += dst_pitch; + } + return; + } + + src = src_ptr - 2 - src_pixels_per_line * 2; + tmpp = tmp; + for (i = 0; i < 7; i++) { + d6u8 = vld1_u8(src); + d7u8 = vld1_u8(src + 8); + d8u8 = vld1_u8(src + 16); + src += src_pixels_per_line; + d9u8 = vld1_u8(src); + d10u8 = vld1_u8(src + 8); + d11u8 = vld1_u8(src + 16); + src += src_pixels_per_line; + d12u8 = vld1_u8(src); + d13u8 = vld1_u8(src + 8); + d14u8 = vld1_u8(src + 16); + src += src_pixels_per_line; + + __builtin_prefetch(src); + __builtin_prefetch(src + src_pixels_per_line); + __builtin_prefetch(src + src_pixels_per_line * 2); + + q8u16 = vmull_u8(d6u8, d0u8); + q9u16 = vmull_u8(d7u8, d0u8); + q10u16 = vmull_u8(d9u8, d0u8); + q11u16 = vmull_u8(d10u8, d0u8); + q12u16 = vmull_u8(d12u8, d0u8); + q13u16 = vmull_u8(d13u8, d0u8); + + d28u8 = vext_u8(d6u8, d7u8, 1); + d29u8 = vext_u8(d9u8, d10u8, 1); + d30u8 = vext_u8(d12u8, d13u8, 1); + q8u16 = vmlsl_u8(q8u16, d28u8, d1u8); + q10u16 = vmlsl_u8(q10u16, d29u8, d1u8); + q12u16 = vmlsl_u8(q12u16, d30u8, d1u8); + d28u8 = vext_u8(d7u8, d8u8, 1); + d29u8 = vext_u8(d10u8, d11u8, 1); + d30u8 = vext_u8(d13u8, d14u8, 1); + q9u16 = vmlsl_u8(q9u16, d28u8, d1u8); + q11u16 = vmlsl_u8(q11u16, d29u8, d1u8); + q13u16 = vmlsl_u8(q13u16, d30u8, d1u8); + + d28u8 = vext_u8(d6u8, d7u8, 4); + d29u8 = vext_u8(d9u8, d10u8, 4); + d30u8 = vext_u8(d12u8, d13u8, 4); + q8u16 = vmlsl_u8(q8u16, d28u8, d4u8); + q10u16 = vmlsl_u8(q10u16, d29u8, d4u8); + q12u16 = vmlsl_u8(q12u16, d30u8, d4u8); + d28u8 = vext_u8(d7u8, d8u8, 4); + d29u8 = vext_u8(d10u8, d11u8, 4); + d30u8 = vext_u8(d13u8, d14u8, 4); + q9u16 = vmlsl_u8(q9u16, d28u8, d4u8); + q11u16 = vmlsl_u8(q11u16, d29u8, d4u8); + q13u16 = vmlsl_u8(q13u16, d30u8, d4u8); + + d28u8 = vext_u8(d6u8, d7u8, 5); + d29u8 = vext_u8(d9u8, d10u8, 5); + d30u8 = vext_u8(d12u8, d13u8, 5); + q8u16 = vmlal_u8(q8u16, d28u8, d5u8); + q10u16 = vmlal_u8(q10u16, d29u8, d5u8); + q12u16 = vmlal_u8(q12u16, d30u8, d5u8); + d28u8 = vext_u8(d7u8, d8u8, 5); + d29u8 = vext_u8(d10u8, d11u8, 5); + d30u8 = vext_u8(d13u8, d14u8, 5); + q9u16 = vmlal_u8(q9u16, d28u8, d5u8); + q11u16 = vmlal_u8(q11u16, d29u8, d5u8); + q13u16 = vmlal_u8(q13u16, d30u8, d5u8); + + d28u8 = vext_u8(d6u8, d7u8, 2); + d29u8 = vext_u8(d9u8, d10u8, 2); + d30u8 = vext_u8(d12u8, d13u8, 2); + q8u16 = vmlal_u8(q8u16, d28u8, d2u8); + q10u16 = vmlal_u8(q10u16, d29u8, d2u8); + q12u16 = vmlal_u8(q12u16, d30u8, d2u8); + d28u8 = vext_u8(d7u8, d8u8, 2); + d29u8 = vext_u8(d10u8, d11u8, 2); + d30u8 = vext_u8(d13u8, d14u8, 2); + q9u16 = vmlal_u8(q9u16, d28u8, d2u8); + q11u16 = vmlal_u8(q11u16, d29u8, d2u8); + q13u16 = vmlal_u8(q13u16, d30u8, d2u8); + + d28u8 = vext_u8(d6u8, d7u8, 3); + d29u8 = vext_u8(d9u8, d10u8, 3); + d30u8 = vext_u8(d12u8, d13u8, 3); + d15u8 = vext_u8(d7u8, d8u8, 3); + d31u8 = vext_u8(d10u8, d11u8, 3); + d6u8 = vext_u8(d13u8, d14u8, 3); + q4u16 = vmull_u8(d28u8, d3u8); + q5u16 = vmull_u8(d29u8, d3u8); + q6u16 = vmull_u8(d30u8, d3u8); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + q12s16 = vreinterpretq_s16_u16(q12u16); + q8s16 = vqaddq_s16(q8s16, q4s16); + q10s16 = vqaddq_s16(q10s16, q5s16); + q12s16 = vqaddq_s16(q12s16, q6s16); + + q6u16 = vmull_u8(d15u8, d3u8); + q7u16 = vmull_u8(d31u8, d3u8); + q3u16 = vmull_u8(d6u8, d3u8); + q3s16 = vreinterpretq_s16_u16(q3u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q11s16 = vreinterpretq_s16_u16(q11u16); + q13s16 = vreinterpretq_s16_u16(q13u16); + q9s16 = vqaddq_s16(q9s16, q6s16); + q11s16 = vqaddq_s16(q11s16, q7s16); + q13s16 = vqaddq_s16(q13s16, q3s16); + + d6u8 = vqrshrun_n_s16(q8s16, 7); + d7u8 = vqrshrun_n_s16(q9s16, 7); + d8u8 = vqrshrun_n_s16(q10s16, 7); + d9u8 = vqrshrun_n_s16(q11s16, 7); + d10u8 = vqrshrun_n_s16(q12s16, 7); + d11u8 = vqrshrun_n_s16(q13s16, 7); + + vst1_u8(tmpp, d6u8); + tmpp += 8; + vst1_u8(tmpp, d7u8); + tmpp += 8; + vst1_u8(tmpp, d8u8); + tmpp += 8; + vst1_u8(tmpp, d9u8); + tmpp += 8; + vst1_u8(tmpp, d10u8); + tmpp += 8; + vst1_u8(tmpp, d11u8); + tmpp += 8; + } + + // Second pass: 16x16 + dtmps8 = vld1_s8(vp8_sub_pel_filters[yoffset]); + d0s8 = vdup_lane_s8(dtmps8, 0); + d1s8 = vdup_lane_s8(dtmps8, 1); + d2s8 = vdup_lane_s8(dtmps8, 2); + d3s8 = vdup_lane_s8(dtmps8, 3); + d4s8 = vdup_lane_s8(dtmps8, 4); + d5s8 = vdup_lane_s8(dtmps8, 5); + d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8)); + d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8)); + d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8)); + d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8)); + d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8)); + d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8)); + + for (i = 0; i < 2; i++) { + dst = dst_ptr + 8 * i; + tmpp = tmp + 8 * i; + d18u8 = vld1_u8(tmpp); + tmpp += 16; + d19u8 = vld1_u8(tmpp); + tmpp += 16; + d20u8 = vld1_u8(tmpp); + tmpp += 16; + d21u8 = vld1_u8(tmpp); + tmpp += 16; + d22u8 = vld1_u8(tmpp); + tmpp += 16; + for (j = 0; j < 4; j++) { + d23u8 = vld1_u8(tmpp); + tmpp += 16; + d24u8 = vld1_u8(tmpp); + tmpp += 16; + d25u8 = vld1_u8(tmpp); + tmpp += 16; + d26u8 = vld1_u8(tmpp); + tmpp += 16; + + q3u16 = vmull_u8(d18u8, d0u8); + q4u16 = vmull_u8(d19u8, d0u8); + q5u16 = vmull_u8(d20u8, d0u8); + q6u16 = vmull_u8(d21u8, d0u8); + + q3u16 = vmlsl_u8(q3u16, d19u8, d1u8); + q4u16 = vmlsl_u8(q4u16, d20u8, d1u8); + q5u16 = vmlsl_u8(q5u16, d21u8, d1u8); + q6u16 = vmlsl_u8(q6u16, d22u8, d1u8); + + q3u16 = vmlsl_u8(q3u16, d22u8, d4u8); + q4u16 = vmlsl_u8(q4u16, d23u8, d4u8); + q5u16 = vmlsl_u8(q5u16, d24u8, d4u8); + q6u16 = vmlsl_u8(q6u16, d25u8, d4u8); + + q3u16 = vmlal_u8(q3u16, d20u8, d2u8); + q4u16 = vmlal_u8(q4u16, d21u8, d2u8); + q5u16 = vmlal_u8(q5u16, d22u8, d2u8); + q6u16 = vmlal_u8(q6u16, d23u8, d2u8); + + q3u16 = vmlal_u8(q3u16, d23u8, d5u8); + q4u16 = vmlal_u8(q4u16, d24u8, d5u8); + q5u16 = vmlal_u8(q5u16, d25u8, d5u8); + q6u16 = vmlal_u8(q6u16, d26u8, d5u8); + + q7u16 = vmull_u8(d21u8, d3u8); + q8u16 = vmull_u8(d22u8, d3u8); + q9u16 = vmull_u8(d23u8, d3u8); + q10u16 = vmull_u8(d24u8, d3u8); + + q3s16 = vreinterpretq_s16_u16(q3u16); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + + q7s16 = vqaddq_s16(q7s16, q3s16); + q8s16 = vqaddq_s16(q8s16, q4s16); + q9s16 = vqaddq_s16(q9s16, q5s16); + q10s16 = vqaddq_s16(q10s16, q6s16); + + d6u8 = vqrshrun_n_s16(q7s16, 7); + d7u8 = vqrshrun_n_s16(q8s16, 7); + d8u8 = vqrshrun_n_s16(q9s16, 7); + d9u8 = vqrshrun_n_s16(q10s16, 7); + + d18u8 = d22u8; + d19u8 = d23u8; + d20u8 = d24u8; + d21u8 = d25u8; + d22u8 = d26u8; + + vst1_u8(dst, d6u8); + dst += dst_pitch; + vst1_u8(dst, d7u8); + dst += dst_pitch; + vst1_u8(dst, d8u8); + dst += dst_pitch; + vst1_u8(dst, d9u8); + dst += dst_pitch; + } + } + return; +} diff --git a/thirdparty/libvpx/vp8/common/arm/neon/vp8_loopfilter_neon.c b/thirdparty/libvpx/vp8/common/arm/neon/vp8_loopfilter_neon.c new file mode 100644 index 0000000000..9d6807af71 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/arm/neon/vp8_loopfilter_neon.c @@ -0,0 +1,550 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> +#include "./vpx_config.h" +#include "vpx_ports/arm.h" + +static INLINE void vp8_loop_filter_neon( + uint8x16_t qblimit, // flimit + uint8x16_t qlimit, // limit + uint8x16_t qthresh, // thresh + uint8x16_t q3, // p3 + uint8x16_t q4, // p2 + uint8x16_t q5, // p1 + uint8x16_t q6, // p0 + uint8x16_t q7, // q0 + uint8x16_t q8, // q1 + uint8x16_t q9, // q2 + uint8x16_t q10, // q3 + uint8x16_t *q5r, // p1 + uint8x16_t *q6r, // p0 + uint8x16_t *q7r, // q0 + uint8x16_t *q8r) { // q1 + uint8x16_t q0u8, q1u8, q2u8, q11u8, q12u8, q13u8, q14u8, q15u8; + int16x8_t q2s16, q11s16; + uint16x8_t q4u16; + int8x16_t q1s8, q2s8, q10s8, q11s8, q12s8, q13s8; + int8x8_t d2s8, d3s8; + + q11u8 = vabdq_u8(q3, q4); + q12u8 = vabdq_u8(q4, q5); + q13u8 = vabdq_u8(q5, q6); + q14u8 = vabdq_u8(q8, q7); + q3 = vabdq_u8(q9, q8); + q4 = vabdq_u8(q10, q9); + + q11u8 = vmaxq_u8(q11u8, q12u8); + q12u8 = vmaxq_u8(q13u8, q14u8); + q3 = vmaxq_u8(q3, q4); + q15u8 = vmaxq_u8(q11u8, q12u8); + + q9 = vabdq_u8(q6, q7); + + // vp8_hevmask + q13u8 = vcgtq_u8(q13u8, qthresh); + q14u8 = vcgtq_u8(q14u8, qthresh); + q15u8 = vmaxq_u8(q15u8, q3); + + q2u8 = vabdq_u8(q5, q8); + q9 = vqaddq_u8(q9, q9); + + q15u8 = vcgeq_u8(qlimit, q15u8); + + // vp8_filter() function + // convert to signed + q10 = vdupq_n_u8(0x80); + q8 = veorq_u8(q8, q10); + q7 = veorq_u8(q7, q10); + q6 = veorq_u8(q6, q10); + q5 = veorq_u8(q5, q10); + + q2u8 = vshrq_n_u8(q2u8, 1); + q9 = vqaddq_u8(q9, q2u8); + + q10 = vdupq_n_u8(3); + + q2s16 = vsubl_s8(vget_low_s8(vreinterpretq_s8_u8(q7)), + vget_low_s8(vreinterpretq_s8_u8(q6))); + q11s16 = vsubl_s8(vget_high_s8(vreinterpretq_s8_u8(q7)), + vget_high_s8(vreinterpretq_s8_u8(q6))); + + q9 = vcgeq_u8(qblimit, q9); + + q1s8 = vqsubq_s8(vreinterpretq_s8_u8(q5), + vreinterpretq_s8_u8(q8)); + + q14u8 = vorrq_u8(q13u8, q14u8); + + q4u16 = vmovl_u8(vget_low_u8(q10)); + q2s16 = vmulq_s16(q2s16, vreinterpretq_s16_u16(q4u16)); + q11s16 = vmulq_s16(q11s16, vreinterpretq_s16_u16(q4u16)); + + q1u8 = vandq_u8(vreinterpretq_u8_s8(q1s8), q14u8); + q15u8 = vandq_u8(q15u8, q9); + + q1s8 = vreinterpretq_s8_u8(q1u8); + q2s16 = vaddw_s8(q2s16, vget_low_s8(q1s8)); + q11s16 = vaddw_s8(q11s16, vget_high_s8(q1s8)); + + q9 = vdupq_n_u8(4); + // vp8_filter = clamp(vp8_filter + 3 * ( qs0 - ps0)) + d2s8 = vqmovn_s16(q2s16); + d3s8 = vqmovn_s16(q11s16); + q1s8 = vcombine_s8(d2s8, d3s8); + q1u8 = vandq_u8(vreinterpretq_u8_s8(q1s8), q15u8); + q1s8 = vreinterpretq_s8_u8(q1u8); + + q2s8 = vqaddq_s8(q1s8, vreinterpretq_s8_u8(q10)); + q1s8 = vqaddq_s8(q1s8, vreinterpretq_s8_u8(q9)); + q2s8 = vshrq_n_s8(q2s8, 3); + q1s8 = vshrq_n_s8(q1s8, 3); + + q11s8 = vqaddq_s8(vreinterpretq_s8_u8(q6), q2s8); + q10s8 = vqsubq_s8(vreinterpretq_s8_u8(q7), q1s8); + + q1s8 = vrshrq_n_s8(q1s8, 1); + q1s8 = vbicq_s8(q1s8, vreinterpretq_s8_u8(q14u8)); + + q13s8 = vqaddq_s8(vreinterpretq_s8_u8(q5), q1s8); + q12s8 = vqsubq_s8(vreinterpretq_s8_u8(q8), q1s8); + + q0u8 = vdupq_n_u8(0x80); + *q8r = veorq_u8(vreinterpretq_u8_s8(q12s8), q0u8); + *q7r = veorq_u8(vreinterpretq_u8_s8(q10s8), q0u8); + *q6r = veorq_u8(vreinterpretq_u8_s8(q11s8), q0u8); + *q5r = veorq_u8(vreinterpretq_u8_s8(q13s8), q0u8); + return; +} + +void vp8_loop_filter_horizontal_edge_y_neon( + unsigned char *src, + int pitch, + unsigned char blimit, + unsigned char limit, + unsigned char thresh) { + uint8x16_t qblimit, qlimit, qthresh, q3, q4; + uint8x16_t q5, q6, q7, q8, q9, q10; + + qblimit = vdupq_n_u8(blimit); + qlimit = vdupq_n_u8(limit); + qthresh = vdupq_n_u8(thresh); + src -= (pitch << 2); + + q3 = vld1q_u8(src); + src += pitch; + q4 = vld1q_u8(src); + src += pitch; + q5 = vld1q_u8(src); + src += pitch; + q6 = vld1q_u8(src); + src += pitch; + q7 = vld1q_u8(src); + src += pitch; + q8 = vld1q_u8(src); + src += pitch; + q9 = vld1q_u8(src); + src += pitch; + q10 = vld1q_u8(src); + + vp8_loop_filter_neon(qblimit, qlimit, qthresh, q3, q4, + q5, q6, q7, q8, q9, q10, + &q5, &q6, &q7, &q8); + + src -= (pitch * 5); + vst1q_u8(src, q5); + src += pitch; + vst1q_u8(src, q6); + src += pitch; + vst1q_u8(src, q7); + src += pitch; + vst1q_u8(src, q8); + return; +} + +void vp8_loop_filter_horizontal_edge_uv_neon( + unsigned char *u, + int pitch, + unsigned char blimit, + unsigned char limit, + unsigned char thresh, + unsigned char *v) { + uint8x16_t qblimit, qlimit, qthresh, q3, q4; + uint8x16_t q5, q6, q7, q8, q9, q10; + uint8x8_t d6, d7, d8, d9, d10, d11, d12, d13, d14; + uint8x8_t d15, d16, d17, d18, d19, d20, d21; + + qblimit = vdupq_n_u8(blimit); + qlimit = vdupq_n_u8(limit); + qthresh = vdupq_n_u8(thresh); + + u -= (pitch << 2); + v -= (pitch << 2); + + d6 = vld1_u8(u); + u += pitch; + d7 = vld1_u8(v); + v += pitch; + d8 = vld1_u8(u); + u += pitch; + d9 = vld1_u8(v); + v += pitch; + d10 = vld1_u8(u); + u += pitch; + d11 = vld1_u8(v); + v += pitch; + d12 = vld1_u8(u); + u += pitch; + d13 = vld1_u8(v); + v += pitch; + d14 = vld1_u8(u); + u += pitch; + d15 = vld1_u8(v); + v += pitch; + d16 = vld1_u8(u); + u += pitch; + d17 = vld1_u8(v); + v += pitch; + d18 = vld1_u8(u); + u += pitch; + d19 = vld1_u8(v); + v += pitch; + d20 = vld1_u8(u); + d21 = vld1_u8(v); + + q3 = vcombine_u8(d6, d7); + q4 = vcombine_u8(d8, d9); + q5 = vcombine_u8(d10, d11); + q6 = vcombine_u8(d12, d13); + q7 = vcombine_u8(d14, d15); + q8 = vcombine_u8(d16, d17); + q9 = vcombine_u8(d18, d19); + q10 = vcombine_u8(d20, d21); + + vp8_loop_filter_neon(qblimit, qlimit, qthresh, q3, q4, + q5, q6, q7, q8, q9, q10, + &q5, &q6, &q7, &q8); + + u -= (pitch * 5); + vst1_u8(u, vget_low_u8(q5)); + u += pitch; + vst1_u8(u, vget_low_u8(q6)); + u += pitch; + vst1_u8(u, vget_low_u8(q7)); + u += pitch; + vst1_u8(u, vget_low_u8(q8)); + + v -= (pitch * 5); + vst1_u8(v, vget_high_u8(q5)); + v += pitch; + vst1_u8(v, vget_high_u8(q6)); + v += pitch; + vst1_u8(v, vget_high_u8(q7)); + v += pitch; + vst1_u8(v, vget_high_u8(q8)); + return; +} + +static INLINE void write_4x8(unsigned char *dst, int pitch, + const uint8x8x4_t result) { +#ifdef VPX_INCOMPATIBLE_GCC + /* + * uint8x8x4_t result + 00 01 02 03 | 04 05 06 07 + 10 11 12 13 | 14 15 16 17 + 20 21 22 23 | 24 25 26 27 + 30 31 32 33 | 34 35 36 37 + --- + * after vtrn_u16 + 00 01 20 21 | 04 05 24 25 + 02 03 22 23 | 06 07 26 27 + 10 11 30 31 | 14 15 34 35 + 12 13 32 33 | 16 17 36 37 + --- + * after vtrn_u8 + 00 10 20 30 | 04 14 24 34 + 01 11 21 31 | 05 15 25 35 + 02 12 22 32 | 06 16 26 36 + 03 13 23 33 | 07 17 27 37 + */ + const uint16x4x2_t r02_u16 = vtrn_u16(vreinterpret_u16_u8(result.val[0]), + vreinterpret_u16_u8(result.val[2])); + const uint16x4x2_t r13_u16 = vtrn_u16(vreinterpret_u16_u8(result.val[1]), + vreinterpret_u16_u8(result.val[3])); + const uint8x8x2_t r01_u8 = vtrn_u8(vreinterpret_u8_u16(r02_u16.val[0]), + vreinterpret_u8_u16(r13_u16.val[0])); + const uint8x8x2_t r23_u8 = vtrn_u8(vreinterpret_u8_u16(r02_u16.val[1]), + vreinterpret_u8_u16(r13_u16.val[1])); + const uint32x2_t x_0_4 = vreinterpret_u32_u8(r01_u8.val[0]); + const uint32x2_t x_1_5 = vreinterpret_u32_u8(r01_u8.val[1]); + const uint32x2_t x_2_6 = vreinterpret_u32_u8(r23_u8.val[0]); + const uint32x2_t x_3_7 = vreinterpret_u32_u8(r23_u8.val[1]); + vst1_lane_u32((uint32_t *)dst, x_0_4, 0); + dst += pitch; + vst1_lane_u32((uint32_t *)dst, x_1_5, 0); + dst += pitch; + vst1_lane_u32((uint32_t *)dst, x_2_6, 0); + dst += pitch; + vst1_lane_u32((uint32_t *)dst, x_3_7, 0); + dst += pitch; + vst1_lane_u32((uint32_t *)dst, x_0_4, 1); + dst += pitch; + vst1_lane_u32((uint32_t *)dst, x_1_5, 1); + dst += pitch; + vst1_lane_u32((uint32_t *)dst, x_2_6, 1); + dst += pitch; + vst1_lane_u32((uint32_t *)dst, x_3_7, 1); +#else + vst4_lane_u8(dst, result, 0); + dst += pitch; + vst4_lane_u8(dst, result, 1); + dst += pitch; + vst4_lane_u8(dst, result, 2); + dst += pitch; + vst4_lane_u8(dst, result, 3); + dst += pitch; + vst4_lane_u8(dst, result, 4); + dst += pitch; + vst4_lane_u8(dst, result, 5); + dst += pitch; + vst4_lane_u8(dst, result, 6); + dst += pitch; + vst4_lane_u8(dst, result, 7); +#endif // VPX_INCOMPATIBLE_GCC +} + +void vp8_loop_filter_vertical_edge_y_neon( + unsigned char *src, + int pitch, + unsigned char blimit, + unsigned char limit, + unsigned char thresh) { + unsigned char *s, *d; + uint8x16_t qblimit, qlimit, qthresh, q3, q4; + uint8x16_t q5, q6, q7, q8, q9, q10; + uint8x8_t d6, d7, d8, d9, d10, d11, d12, d13, d14; + uint8x8_t d15, d16, d17, d18, d19, d20, d21; + uint32x4x2_t q2tmp0, q2tmp1, q2tmp2, q2tmp3; + uint16x8x2_t q2tmp4, q2tmp5, q2tmp6, q2tmp7; + uint8x16x2_t q2tmp8, q2tmp9, q2tmp10, q2tmp11; + uint8x8x4_t q4ResultH, q4ResultL; + + qblimit = vdupq_n_u8(blimit); + qlimit = vdupq_n_u8(limit); + qthresh = vdupq_n_u8(thresh); + + s = src - 4; + d6 = vld1_u8(s); + s += pitch; + d8 = vld1_u8(s); + s += pitch; + d10 = vld1_u8(s); + s += pitch; + d12 = vld1_u8(s); + s += pitch; + d14 = vld1_u8(s); + s += pitch; + d16 = vld1_u8(s); + s += pitch; + d18 = vld1_u8(s); + s += pitch; + d20 = vld1_u8(s); + s += pitch; + d7 = vld1_u8(s); + s += pitch; + d9 = vld1_u8(s); + s += pitch; + d11 = vld1_u8(s); + s += pitch; + d13 = vld1_u8(s); + s += pitch; + d15 = vld1_u8(s); + s += pitch; + d17 = vld1_u8(s); + s += pitch; + d19 = vld1_u8(s); + s += pitch; + d21 = vld1_u8(s); + + q3 = vcombine_u8(d6, d7); + q4 = vcombine_u8(d8, d9); + q5 = vcombine_u8(d10, d11); + q6 = vcombine_u8(d12, d13); + q7 = vcombine_u8(d14, d15); + q8 = vcombine_u8(d16, d17); + q9 = vcombine_u8(d18, d19); + q10 = vcombine_u8(d20, d21); + + q2tmp0 = vtrnq_u32(vreinterpretq_u32_u8(q3), vreinterpretq_u32_u8(q7)); + q2tmp1 = vtrnq_u32(vreinterpretq_u32_u8(q4), vreinterpretq_u32_u8(q8)); + q2tmp2 = vtrnq_u32(vreinterpretq_u32_u8(q5), vreinterpretq_u32_u8(q9)); + q2tmp3 = vtrnq_u32(vreinterpretq_u32_u8(q6), vreinterpretq_u32_u8(q10)); + + q2tmp4 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[0]), + vreinterpretq_u16_u32(q2tmp2.val[0])); + q2tmp5 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[0]), + vreinterpretq_u16_u32(q2tmp3.val[0])); + q2tmp6 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[1]), + vreinterpretq_u16_u32(q2tmp2.val[1])); + q2tmp7 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[1]), + vreinterpretq_u16_u32(q2tmp3.val[1])); + + q2tmp8 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[0]), + vreinterpretq_u8_u16(q2tmp5.val[0])); + q2tmp9 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[1]), + vreinterpretq_u8_u16(q2tmp5.val[1])); + q2tmp10 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[0]), + vreinterpretq_u8_u16(q2tmp7.val[0])); + q2tmp11 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[1]), + vreinterpretq_u8_u16(q2tmp7.val[1])); + + q3 = q2tmp8.val[0]; + q4 = q2tmp8.val[1]; + q5 = q2tmp9.val[0]; + q6 = q2tmp9.val[1]; + q7 = q2tmp10.val[0]; + q8 = q2tmp10.val[1]; + q9 = q2tmp11.val[0]; + q10 = q2tmp11.val[1]; + + vp8_loop_filter_neon(qblimit, qlimit, qthresh, q3, q4, + q5, q6, q7, q8, q9, q10, + &q5, &q6, &q7, &q8); + + q4ResultL.val[0] = vget_low_u8(q5); // d10 + q4ResultL.val[1] = vget_low_u8(q6); // d12 + q4ResultL.val[2] = vget_low_u8(q7); // d14 + q4ResultL.val[3] = vget_low_u8(q8); // d16 + q4ResultH.val[0] = vget_high_u8(q5); // d11 + q4ResultH.val[1] = vget_high_u8(q6); // d13 + q4ResultH.val[2] = vget_high_u8(q7); // d15 + q4ResultH.val[3] = vget_high_u8(q8); // d17 + + d = src - 2; + write_4x8(d, pitch, q4ResultL); + d += pitch * 8; + write_4x8(d, pitch, q4ResultH); +} + +void vp8_loop_filter_vertical_edge_uv_neon( + unsigned char *u, + int pitch, + unsigned char blimit, + unsigned char limit, + unsigned char thresh, + unsigned char *v) { + unsigned char *us, *ud; + unsigned char *vs, *vd; + uint8x16_t qblimit, qlimit, qthresh, q3, q4; + uint8x16_t q5, q6, q7, q8, q9, q10; + uint8x8_t d6, d7, d8, d9, d10, d11, d12, d13, d14; + uint8x8_t d15, d16, d17, d18, d19, d20, d21; + uint32x4x2_t q2tmp0, q2tmp1, q2tmp2, q2tmp3; + uint16x8x2_t q2tmp4, q2tmp5, q2tmp6, q2tmp7; + uint8x16x2_t q2tmp8, q2tmp9, q2tmp10, q2tmp11; + uint8x8x4_t q4ResultH, q4ResultL; + + qblimit = vdupq_n_u8(blimit); + qlimit = vdupq_n_u8(limit); + qthresh = vdupq_n_u8(thresh); + + us = u - 4; + d6 = vld1_u8(us); + us += pitch; + d8 = vld1_u8(us); + us += pitch; + d10 = vld1_u8(us); + us += pitch; + d12 = vld1_u8(us); + us += pitch; + d14 = vld1_u8(us); + us += pitch; + d16 = vld1_u8(us); + us += pitch; + d18 = vld1_u8(us); + us += pitch; + d20 = vld1_u8(us); + + vs = v - 4; + d7 = vld1_u8(vs); + vs += pitch; + d9 = vld1_u8(vs); + vs += pitch; + d11 = vld1_u8(vs); + vs += pitch; + d13 = vld1_u8(vs); + vs += pitch; + d15 = vld1_u8(vs); + vs += pitch; + d17 = vld1_u8(vs); + vs += pitch; + d19 = vld1_u8(vs); + vs += pitch; + d21 = vld1_u8(vs); + + q3 = vcombine_u8(d6, d7); + q4 = vcombine_u8(d8, d9); + q5 = vcombine_u8(d10, d11); + q6 = vcombine_u8(d12, d13); + q7 = vcombine_u8(d14, d15); + q8 = vcombine_u8(d16, d17); + q9 = vcombine_u8(d18, d19); + q10 = vcombine_u8(d20, d21); + + q2tmp0 = vtrnq_u32(vreinterpretq_u32_u8(q3), vreinterpretq_u32_u8(q7)); + q2tmp1 = vtrnq_u32(vreinterpretq_u32_u8(q4), vreinterpretq_u32_u8(q8)); + q2tmp2 = vtrnq_u32(vreinterpretq_u32_u8(q5), vreinterpretq_u32_u8(q9)); + q2tmp3 = vtrnq_u32(vreinterpretq_u32_u8(q6), vreinterpretq_u32_u8(q10)); + + q2tmp4 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[0]), + vreinterpretq_u16_u32(q2tmp2.val[0])); + q2tmp5 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[0]), + vreinterpretq_u16_u32(q2tmp3.val[0])); + q2tmp6 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[1]), + vreinterpretq_u16_u32(q2tmp2.val[1])); + q2tmp7 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[1]), + vreinterpretq_u16_u32(q2tmp3.val[1])); + + q2tmp8 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[0]), + vreinterpretq_u8_u16(q2tmp5.val[0])); + q2tmp9 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[1]), + vreinterpretq_u8_u16(q2tmp5.val[1])); + q2tmp10 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[0]), + vreinterpretq_u8_u16(q2tmp7.val[0])); + q2tmp11 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[1]), + vreinterpretq_u8_u16(q2tmp7.val[1])); + + q3 = q2tmp8.val[0]; + q4 = q2tmp8.val[1]; + q5 = q2tmp9.val[0]; + q6 = q2tmp9.val[1]; + q7 = q2tmp10.val[0]; + q8 = q2tmp10.val[1]; + q9 = q2tmp11.val[0]; + q10 = q2tmp11.val[1]; + + vp8_loop_filter_neon(qblimit, qlimit, qthresh, q3, q4, + q5, q6, q7, q8, q9, q10, + &q5, &q6, &q7, &q8); + + q4ResultL.val[0] = vget_low_u8(q5); // d10 + q4ResultL.val[1] = vget_low_u8(q6); // d12 + q4ResultL.val[2] = vget_low_u8(q7); // d14 + q4ResultL.val[3] = vget_low_u8(q8); // d16 + ud = u - 2; + write_4x8(ud, pitch, q4ResultL); + + q4ResultH.val[0] = vget_high_u8(q5); // d11 + q4ResultH.val[1] = vget_high_u8(q6); // d13 + q4ResultH.val[2] = vget_high_u8(q7); // d15 + q4ResultH.val[3] = vget_high_u8(q8); // d17 + vd = v - 2; + write_4x8(vd, pitch, q4ResultH); +} diff --git a/thirdparty/libvpx/vp8/common/blockd.c b/thirdparty/libvpx/vp8/common/blockd.c new file mode 100644 index 0000000000..1fc3cd0ca7 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/blockd.c @@ -0,0 +1,22 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "blockd.h" +#include "vpx_mem/vpx_mem.h" + +const unsigned char vp8_block2left[25] = +{ + 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 +}; +const unsigned char vp8_block2above[25] = +{ + 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 4, 5, 4, 5, 6, 7, 6, 7, 8 +}; diff --git a/thirdparty/libvpx/vp8/common/blockd.h b/thirdparty/libvpx/vp8/common/blockd.h new file mode 100644 index 0000000000..192108a06d --- /dev/null +++ b/thirdparty/libvpx/vp8/common/blockd.h @@ -0,0 +1,312 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_BLOCKD_H_ +#define VP8_COMMON_BLOCKD_H_ + +void vpx_log(const char *format, ...); + +#include "vpx_config.h" +#include "vpx_scale/yv12config.h" +#include "mv.h" +#include "treecoder.h" +#include "vpx_ports/mem.h" + +#ifdef __cplusplus +extern "C" { +#endif + +/*#define DCPRED 1*/ +#define DCPREDSIMTHRESH 0 +#define DCPREDCNTTHRESH 3 + +#define MB_FEATURE_TREE_PROBS 3 +#define MAX_MB_SEGMENTS 4 + +#define MAX_REF_LF_DELTAS 4 +#define MAX_MODE_LF_DELTAS 4 + +/* Segment Feature Masks */ +#define SEGMENT_DELTADATA 0 +#define SEGMENT_ABSDATA 1 + +typedef struct +{ + int r, c; +} POS; + +#define PLANE_TYPE_Y_NO_DC 0 +#define PLANE_TYPE_Y2 1 +#define PLANE_TYPE_UV 2 +#define PLANE_TYPE_Y_WITH_DC 3 + + +typedef char ENTROPY_CONTEXT; +typedef struct +{ + ENTROPY_CONTEXT y1[4]; + ENTROPY_CONTEXT u[2]; + ENTROPY_CONTEXT v[2]; + ENTROPY_CONTEXT y2; +} ENTROPY_CONTEXT_PLANES; + +extern const unsigned char vp8_block2left[25]; +extern const unsigned char vp8_block2above[25]; + +#define VP8_COMBINEENTROPYCONTEXTS( Dest, A, B) \ + Dest = (A)+(B); + + +typedef enum +{ + KEY_FRAME = 0, + INTER_FRAME = 1 +} FRAME_TYPE; + +typedef enum +{ + DC_PRED, /* average of above and left pixels */ + V_PRED, /* vertical prediction */ + H_PRED, /* horizontal prediction */ + TM_PRED, /* Truemotion prediction */ + B_PRED, /* block based prediction, each block has its own prediction mode */ + + NEARESTMV, + NEARMV, + ZEROMV, + NEWMV, + SPLITMV, + + MB_MODE_COUNT +} MB_PREDICTION_MODE; + +/* Macroblock level features */ +typedef enum +{ + MB_LVL_ALT_Q = 0, /* Use alternate Quantizer .... */ + MB_LVL_ALT_LF = 1, /* Use alternate loop filter value... */ + MB_LVL_MAX = 2 /* Number of MB level features supported */ + +} MB_LVL_FEATURES; + +/* Segment Feature Masks */ +#define SEGMENT_ALTQ 0x01 +#define SEGMENT_ALT_LF 0x02 + +#define VP8_YMODES (B_PRED + 1) +#define VP8_UV_MODES (TM_PRED + 1) + +#define VP8_MVREFS (1 + SPLITMV - NEARESTMV) + +typedef enum +{ + B_DC_PRED, /* average of above and left pixels */ + B_TM_PRED, + + B_VE_PRED, /* vertical prediction */ + B_HE_PRED, /* horizontal prediction */ + + B_LD_PRED, + B_RD_PRED, + + B_VR_PRED, + B_VL_PRED, + B_HD_PRED, + B_HU_PRED, + + LEFT4X4, + ABOVE4X4, + ZERO4X4, + NEW4X4, + + B_MODE_COUNT +} B_PREDICTION_MODE; + +#define VP8_BINTRAMODES (B_HU_PRED + 1) /* 10 */ +#define VP8_SUBMVREFS (1 + NEW4X4 - LEFT4X4) + +/* For keyframes, intra block modes are predicted by the (already decoded) + modes for the Y blocks to the left and above us; for interframes, there + is a single probability table. */ + +union b_mode_info +{ + B_PREDICTION_MODE as_mode; + int_mv mv; +}; + +typedef enum +{ + INTRA_FRAME = 0, + LAST_FRAME = 1, + GOLDEN_FRAME = 2, + ALTREF_FRAME = 3, + MAX_REF_FRAMES = 4 +} MV_REFERENCE_FRAME; + +typedef struct +{ + uint8_t mode, uv_mode; + uint8_t ref_frame; + uint8_t is_4x4; + int_mv mv; + + uint8_t partitioning; + uint8_t mb_skip_coeff; /* does this mb has coefficients at all, 1=no coefficients, 0=need decode tokens */ + uint8_t need_to_clamp_mvs; + uint8_t segment_id; /* Which set of segmentation parameters should be used for this MB */ +} MB_MODE_INFO; + +typedef struct modeinfo +{ + MB_MODE_INFO mbmi; + union b_mode_info bmi[16]; +} MODE_INFO; + +#if CONFIG_MULTI_RES_ENCODING +/* The mb-level information needed to be stored for higher-resolution encoder */ +typedef struct +{ + MB_PREDICTION_MODE mode; + MV_REFERENCE_FRAME ref_frame; + int_mv mv; + int dissim; /* dissimilarity level of the macroblock */ +} LOWER_RES_MB_INFO; + +/* The frame-level information needed to be stored for higher-resolution + * encoder */ +typedef struct +{ + FRAME_TYPE frame_type; + int is_frame_dropped; + // The frame rate for the lowest resolution. + double low_res_framerate; + /* The frame number of each reference frames */ + unsigned int low_res_ref_frames[MAX_REF_FRAMES]; + // The video frame counter value for the key frame, for lowest resolution. + unsigned int key_frame_counter_value; + LOWER_RES_MB_INFO *mb_info; +} LOWER_RES_FRAME_INFO; +#endif + +typedef struct blockd +{ + short *qcoeff; + short *dqcoeff; + unsigned char *predictor; + short *dequant; + + int offset; + char *eob; + + union b_mode_info bmi; +} BLOCKD; + +typedef void (*vp8_subpix_fn_t)(unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch); + +typedef struct macroblockd +{ + DECLARE_ALIGNED(16, unsigned char, predictor[384]); + DECLARE_ALIGNED(16, short, qcoeff[400]); + DECLARE_ALIGNED(16, short, dqcoeff[400]); + DECLARE_ALIGNED(16, char, eobs[25]); + + DECLARE_ALIGNED(16, short, dequant_y1[16]); + DECLARE_ALIGNED(16, short, dequant_y1_dc[16]); + DECLARE_ALIGNED(16, short, dequant_y2[16]); + DECLARE_ALIGNED(16, short, dequant_uv[16]); + + /* 16 Y blocks, 4 U, 4 V, 1 DC 2nd order block, each with 16 entries. */ + BLOCKD block[25]; + int fullpixel_mask; + + YV12_BUFFER_CONFIG pre; /* Filtered copy of previous frame reconstruction */ + YV12_BUFFER_CONFIG dst; + + MODE_INFO *mode_info_context; + int mode_info_stride; + + FRAME_TYPE frame_type; + + int up_available; + int left_available; + + unsigned char *recon_above[3]; + unsigned char *recon_left[3]; + int recon_left_stride[2]; + + /* Y,U,V,Y2 */ + ENTROPY_CONTEXT_PLANES *above_context; + ENTROPY_CONTEXT_PLANES *left_context; + + /* 0 indicates segmentation at MB level is not enabled. Otherwise the individual bits indicate which features are active. */ + unsigned char segmentation_enabled; + + /* 0 (do not update) 1 (update) the macroblock segmentation map. */ + unsigned char update_mb_segmentation_map; + + /* 0 (do not update) 1 (update) the macroblock segmentation feature data. */ + unsigned char update_mb_segmentation_data; + + /* 0 (do not update) 1 (update) the macroblock segmentation feature data. */ + unsigned char mb_segement_abs_delta; + + /* Per frame flags that define which MB level features (such as quantizer or loop filter level) */ + /* are enabled and when enabled the proabilities used to decode the per MB flags in MB_MODE_INFO */ + vp8_prob mb_segment_tree_probs[MB_FEATURE_TREE_PROBS]; /* Probability Tree used to code Segment number */ + + signed char segment_feature_data[MB_LVL_MAX][MAX_MB_SEGMENTS]; /* Segment parameters */ + + /* mode_based Loop filter adjustment */ + unsigned char mode_ref_lf_delta_enabled; + unsigned char mode_ref_lf_delta_update; + + /* Delta values have the range +/- MAX_LOOP_FILTER */ + signed char last_ref_lf_deltas[MAX_REF_LF_DELTAS]; /* 0 = Intra, Last, GF, ARF */ + signed char ref_lf_deltas[MAX_REF_LF_DELTAS]; /* 0 = Intra, Last, GF, ARF */ + signed char last_mode_lf_deltas[MAX_MODE_LF_DELTAS]; /* 0 = BPRED, ZERO_MV, MV, SPLIT */ + signed char mode_lf_deltas[MAX_MODE_LF_DELTAS]; /* 0 = BPRED, ZERO_MV, MV, SPLIT */ + + /* Distance of MB away from frame edges */ + int mb_to_left_edge; + int mb_to_right_edge; + int mb_to_top_edge; + int mb_to_bottom_edge; + + + + vp8_subpix_fn_t subpixel_predict; + vp8_subpix_fn_t subpixel_predict8x4; + vp8_subpix_fn_t subpixel_predict8x8; + vp8_subpix_fn_t subpixel_predict16x16; + + void *current_bc; + + int corrupted; + +#if ARCH_X86 || ARCH_X86_64 + /* This is an intermediate buffer currently used in sub-pixel motion search + * to keep a copy of the reference area. This buffer can be used for other + * purpose. + */ + DECLARE_ALIGNED(32, unsigned char, y_buf[22*32]); +#endif +} MACROBLOCKD; + + +extern void vp8_build_block_doffsets(MACROBLOCKD *x); +extern void vp8_setup_block_dptrs(MACROBLOCKD *x); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_BLOCKD_H_ diff --git a/thirdparty/libvpx/vp8/common/coefupdateprobs.h b/thirdparty/libvpx/vp8/common/coefupdateprobs.h new file mode 100644 index 0000000000..d96a19e747 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/coefupdateprobs.h @@ -0,0 +1,197 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP8_COMMON_COEFUPDATEPROBS_H_ +#define VP8_COMMON_COEFUPDATEPROBS_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +/* Update probabilities for the nodes in the token entropy tree. + Generated file included by entropy.c */ + +const vp8_prob vp8_coef_update_probs [BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [ENTROPY_NODES] = +{ + { + { + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {176, 246, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {223, 241, 252, 255, 255, 255, 255, 255, 255, 255, 255, }, + {249, 253, 253, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 244, 252, 255, 255, 255, 255, 255, 255, 255, 255, }, + {234, 254, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + {253, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 246, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + {239, 253, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + {254, 255, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 248, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + {251, 255, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 253, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + {251, 254, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + {254, 255, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 254, 253, 255, 254, 255, 255, 255, 255, 255, 255, }, + {250, 255, 254, 255, 254, 255, 255, 255, 255, 255, 255, }, + {254, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + }, + { + { + {217, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {225, 252, 241, 253, 255, 255, 254, 255, 255, 255, 255, }, + {234, 250, 241, 250, 253, 255, 253, 254, 255, 255, 255, }, + }, + { + {255, 254, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {223, 254, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + {238, 253, 254, 254, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 248, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + {249, 254, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 253, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {247, 254, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 253, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + {252, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 254, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + {253, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 254, 253, 255, 255, 255, 255, 255, 255, 255, 255, }, + {250, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {254, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + }, + { + { + {186, 251, 250, 255, 255, 255, 255, 255, 255, 255, 255, }, + {234, 251, 244, 254, 255, 255, 255, 255, 255, 255, 255, }, + {251, 251, 243, 253, 254, 255, 254, 255, 255, 255, 255, }, + }, + { + {255, 253, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + {236, 253, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + {251, 253, 253, 254, 254, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 254, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + {254, 254, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 254, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {254, 254, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {254, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {254, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + }, + { + { + {248, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {250, 254, 252, 254, 255, 255, 255, 255, 255, 255, 255, }, + {248, 254, 249, 253, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 253, 253, 255, 255, 255, 255, 255, 255, 255, 255, }, + {246, 253, 253, 255, 255, 255, 255, 255, 255, 255, 255, }, + {252, 254, 251, 254, 254, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 254, 252, 255, 255, 255, 255, 255, 255, 255, 255, }, + {248, 254, 253, 255, 255, 255, 255, 255, 255, 255, 255, }, + {253, 255, 254, 254, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 251, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + {245, 251, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + {253, 253, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 251, 253, 255, 255, 255, 255, 255, 255, 255, 255, }, + {252, 253, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 254, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 252, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {249, 255, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 255, 254, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 255, 253, 255, 255, 255, 255, 255, 255, 255, 255, }, + {250, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + { + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {254, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + {255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, }, + }, + }, +}; + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_COEFUPDATEPROBS_H_ diff --git a/thirdparty/libvpx/vp8/common/common.h b/thirdparty/libvpx/vp8/common/common.h new file mode 100644 index 0000000000..e58a9cc23b --- /dev/null +++ b/thirdparty/libvpx/vp8/common/common.h @@ -0,0 +1,48 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_COMMON_H_ +#define VP8_COMMON_COMMON_H_ + +#include <assert.h> + +/* Interface header for common constant data structures and lookup tables */ + +#include "vpx_mem/vpx_mem.h" + +#ifdef __cplusplus +extern "C" { +#endif + +/* Only need this for fixed-size arrays, for structs just assign. */ + +#define vp8_copy( Dest, Src) { \ + assert( sizeof( Dest) == sizeof( Src)); \ + memcpy( Dest, Src, sizeof( Src)); \ + } + +/* Use this for variably-sized arrays. */ + +#define vp8_copy_array( Dest, Src, N) { \ + assert( sizeof( *Dest) == sizeof( *Src)); \ + memcpy( Dest, Src, N * sizeof( *Src)); \ + } + +#define vp8_zero( Dest) memset( &Dest, 0, sizeof( Dest)); + +#define vp8_zero_array( Dest, N) memset( Dest, 0, N * sizeof( *Dest)); + + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_COMMON_H_ diff --git a/thirdparty/libvpx/vp8/common/copy_c.c b/thirdparty/libvpx/vp8/common/copy_c.c new file mode 100644 index 0000000000..e3392913f6 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/copy_c.c @@ -0,0 +1,32 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include <string.h> + +#include "./vp8_rtcd.h" +#include "vpx/vpx_integer.h" + +/* Copy 2 macroblocks to a buffer */ +void vp8_copy32xn_c(const unsigned char *src_ptr, int src_stride, + unsigned char *dst_ptr, int dst_stride, + int height) +{ + int r; + + for (r = 0; r < height; r++) + { + memcpy(dst_ptr, src_ptr, 32); + + src_ptr += src_stride; + dst_ptr += dst_stride; + + } +} diff --git a/thirdparty/libvpx/vp8/common/debugmodes.c b/thirdparty/libvpx/vp8/common/debugmodes.c new file mode 100644 index 0000000000..159fddc6a7 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/debugmodes.c @@ -0,0 +1,155 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include <stdio.h> +#include "blockd.h" + + +void vp8_print_modes_and_motion_vectors(MODE_INFO *mi, int rows, int cols, int frame) +{ + + int mb_row; + int mb_col; + int mb_index = 0; + FILE *mvs = fopen("mvs.stt", "a"); + + /* print out the macroblock Y modes */ + mb_index = 0; + fprintf(mvs, "Mb Modes for Frame %d\n", frame); + + for (mb_row = 0; mb_row < rows; mb_row++) + { + for (mb_col = 0; mb_col < cols; mb_col++) + { + + fprintf(mvs, "%2d ", mi[mb_index].mbmi.mode); + + mb_index++; + } + + fprintf(mvs, "\n"); + mb_index++; + } + + fprintf(mvs, "\n"); + + mb_index = 0; + fprintf(mvs, "Mb mv ref for Frame %d\n", frame); + + for (mb_row = 0; mb_row < rows; mb_row++) + { + for (mb_col = 0; mb_col < cols; mb_col++) + { + + fprintf(mvs, "%2d ", mi[mb_index].mbmi.ref_frame); + + mb_index++; + } + + fprintf(mvs, "\n"); + mb_index++; + } + + fprintf(mvs, "\n"); + + /* print out the macroblock UV modes */ + mb_index = 0; + fprintf(mvs, "UV Modes for Frame %d\n", frame); + + for (mb_row = 0; mb_row < rows; mb_row++) + { + for (mb_col = 0; mb_col < cols; mb_col++) + { + + fprintf(mvs, "%2d ", mi[mb_index].mbmi.uv_mode); + + mb_index++; + } + + mb_index++; + fprintf(mvs, "\n"); + } + + fprintf(mvs, "\n"); + + /* print out the block modes */ + fprintf(mvs, "Mbs for Frame %d\n", frame); + { + int b_row; + + for (b_row = 0; b_row < 4 * rows; b_row++) + { + int b_col; + int bindex; + + for (b_col = 0; b_col < 4 * cols; b_col++) + { + mb_index = (b_row >> 2) * (cols + 1) + (b_col >> 2); + bindex = (b_row & 3) * 4 + (b_col & 3); + + if (mi[mb_index].mbmi.mode == B_PRED) + fprintf(mvs, "%2d ", mi[mb_index].bmi[bindex].as_mode); + else + fprintf(mvs, "xx "); + + } + + fprintf(mvs, "\n"); + } + } + fprintf(mvs, "\n"); + + /* print out the macroblock mvs */ + mb_index = 0; + fprintf(mvs, "MVs for Frame %d\n", frame); + + for (mb_row = 0; mb_row < rows; mb_row++) + { + for (mb_col = 0; mb_col < cols; mb_col++) + { + fprintf(mvs, "%5d:%-5d", mi[mb_index].mbmi.mv.as_mv.row / 2, mi[mb_index].mbmi.mv.as_mv.col / 2); + + mb_index++; + } + + mb_index++; + fprintf(mvs, "\n"); + } + + fprintf(mvs, "\n"); + + + /* print out the block modes */ + fprintf(mvs, "MVs for Frame %d\n", frame); + { + int b_row; + + for (b_row = 0; b_row < 4 * rows; b_row++) + { + int b_col; + int bindex; + + for (b_col = 0; b_col < 4 * cols; b_col++) + { + mb_index = (b_row >> 2) * (cols + 1) + (b_col >> 2); + bindex = (b_row & 3) * 4 + (b_col & 3); + fprintf(mvs, "%3d:%-3d ", mi[mb_index].bmi[bindex].mv.as_mv.row, mi[mb_index].bmi[bindex].mv.as_mv.col); + + } + + fprintf(mvs, "\n"); + } + } + fprintf(mvs, "\n"); + + + fclose(mvs); +} diff --git a/thirdparty/libvpx/vp8/common/default_coef_probs.h b/thirdparty/libvpx/vp8/common/default_coef_probs.h new file mode 100644 index 0000000000..4d69e4be66 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/default_coef_probs.h @@ -0,0 +1,200 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. +*/ + +#ifndef VP8_COMMON_DEFAULT_COEF_PROBS_H_ +#define VP8_COMMON_DEFAULT_COEF_PROBS_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +/*Generated file, included by entropy.c*/ + + +static const vp8_prob default_coef_probs [BLOCK_TYPES] + [COEF_BANDS] + [PREV_COEF_CONTEXTS] + [ENTROPY_NODES] = +{ + { /* Block Type ( 0 ) */ + { /* Coeff Band ( 0 )*/ + { 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 }, + { 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 }, + { 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 } + }, + { /* Coeff Band ( 1 )*/ + { 253, 136, 254, 255, 228, 219, 128, 128, 128, 128, 128 }, + { 189, 129, 242, 255, 227, 213, 255, 219, 128, 128, 128 }, + { 106, 126, 227, 252, 214, 209, 255, 255, 128, 128, 128 } + }, + { /* Coeff Band ( 2 )*/ + { 1, 98, 248, 255, 236, 226, 255, 255, 128, 128, 128 }, + { 181, 133, 238, 254, 221, 234, 255, 154, 128, 128, 128 }, + { 78, 134, 202, 247, 198, 180, 255, 219, 128, 128, 128 } + }, + { /* Coeff Band ( 3 )*/ + { 1, 185, 249, 255, 243, 255, 128, 128, 128, 128, 128 }, + { 184, 150, 247, 255, 236, 224, 128, 128, 128, 128, 128 }, + { 77, 110, 216, 255, 236, 230, 128, 128, 128, 128, 128 } + }, + { /* Coeff Band ( 4 )*/ + { 1, 101, 251, 255, 241, 255, 128, 128, 128, 128, 128 }, + { 170, 139, 241, 252, 236, 209, 255, 255, 128, 128, 128 }, + { 37, 116, 196, 243, 228, 255, 255, 255, 128, 128, 128 } + }, + { /* Coeff Band ( 5 )*/ + { 1, 204, 254, 255, 245, 255, 128, 128, 128, 128, 128 }, + { 207, 160, 250, 255, 238, 128, 128, 128, 128, 128, 128 }, + { 102, 103, 231, 255, 211, 171, 128, 128, 128, 128, 128 } + }, + { /* Coeff Band ( 6 )*/ + { 1, 152, 252, 255, 240, 255, 128, 128, 128, 128, 128 }, + { 177, 135, 243, 255, 234, 225, 128, 128, 128, 128, 128 }, + { 80, 129, 211, 255, 194, 224, 128, 128, 128, 128, 128 } + }, + { /* Coeff Band ( 7 )*/ + { 1, 1, 255, 128, 128, 128, 128, 128, 128, 128, 128 }, + { 246, 1, 255, 128, 128, 128, 128, 128, 128, 128, 128 }, + { 255, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 } + } + }, + { /* Block Type ( 1 ) */ + { /* Coeff Band ( 0 )*/ + { 198, 35, 237, 223, 193, 187, 162, 160, 145, 155, 62 }, + { 131, 45, 198, 221, 172, 176, 220, 157, 252, 221, 1 }, + { 68, 47, 146, 208, 149, 167, 221, 162, 255, 223, 128 } + }, + { /* Coeff Band ( 1 )*/ + { 1, 149, 241, 255, 221, 224, 255, 255, 128, 128, 128 }, + { 184, 141, 234, 253, 222, 220, 255, 199, 128, 128, 128 }, + { 81, 99, 181, 242, 176, 190, 249, 202, 255, 255, 128 } + }, + { /* Coeff Band ( 2 )*/ + { 1, 129, 232, 253, 214, 197, 242, 196, 255, 255, 128 }, + { 99, 121, 210, 250, 201, 198, 255, 202, 128, 128, 128 }, + { 23, 91, 163, 242, 170, 187, 247, 210, 255, 255, 128 } + }, + { /* Coeff Band ( 3 )*/ + { 1, 200, 246, 255, 234, 255, 128, 128, 128, 128, 128 }, + { 109, 178, 241, 255, 231, 245, 255, 255, 128, 128, 128 }, + { 44, 130, 201, 253, 205, 192, 255, 255, 128, 128, 128 } + }, + { /* Coeff Band ( 4 )*/ + { 1, 132, 239, 251, 219, 209, 255, 165, 128, 128, 128 }, + { 94, 136, 225, 251, 218, 190, 255, 255, 128, 128, 128 }, + { 22, 100, 174, 245, 186, 161, 255, 199, 128, 128, 128 } + }, + { /* Coeff Band ( 5 )*/ + { 1, 182, 249, 255, 232, 235, 128, 128, 128, 128, 128 }, + { 124, 143, 241, 255, 227, 234, 128, 128, 128, 128, 128 }, + { 35, 77, 181, 251, 193, 211, 255, 205, 128, 128, 128 } + }, + { /* Coeff Band ( 6 )*/ + { 1, 157, 247, 255, 236, 231, 255, 255, 128, 128, 128 }, + { 121, 141, 235, 255, 225, 227, 255, 255, 128, 128, 128 }, + { 45, 99, 188, 251, 195, 217, 255, 224, 128, 128, 128 } + }, + { /* Coeff Band ( 7 )*/ + { 1, 1, 251, 255, 213, 255, 128, 128, 128, 128, 128 }, + { 203, 1, 248, 255, 255, 128, 128, 128, 128, 128, 128 }, + { 137, 1, 177, 255, 224, 255, 128, 128, 128, 128, 128 } + } + }, + { /* Block Type ( 2 ) */ + { /* Coeff Band ( 0 )*/ + { 253, 9, 248, 251, 207, 208, 255, 192, 128, 128, 128 }, + { 175, 13, 224, 243, 193, 185, 249, 198, 255, 255, 128 }, + { 73, 17, 171, 221, 161, 179, 236, 167, 255, 234, 128 } + }, + { /* Coeff Band ( 1 )*/ + { 1, 95, 247, 253, 212, 183, 255, 255, 128, 128, 128 }, + { 239, 90, 244, 250, 211, 209, 255, 255, 128, 128, 128 }, + { 155, 77, 195, 248, 188, 195, 255, 255, 128, 128, 128 } + }, + { /* Coeff Band ( 2 )*/ + { 1, 24, 239, 251, 218, 219, 255, 205, 128, 128, 128 }, + { 201, 51, 219, 255, 196, 186, 128, 128, 128, 128, 128 }, + { 69, 46, 190, 239, 201, 218, 255, 228, 128, 128, 128 } + }, + { /* Coeff Band ( 3 )*/ + { 1, 191, 251, 255, 255, 128, 128, 128, 128, 128, 128 }, + { 223, 165, 249, 255, 213, 255, 128, 128, 128, 128, 128 }, + { 141, 124, 248, 255, 255, 128, 128, 128, 128, 128, 128 } + }, + { /* Coeff Band ( 4 )*/ + { 1, 16, 248, 255, 255, 128, 128, 128, 128, 128, 128 }, + { 190, 36, 230, 255, 236, 255, 128, 128, 128, 128, 128 }, + { 149, 1, 255, 128, 128, 128, 128, 128, 128, 128, 128 } + }, + { /* Coeff Band ( 5 )*/ + { 1, 226, 255, 128, 128, 128, 128, 128, 128, 128, 128 }, + { 247, 192, 255, 128, 128, 128, 128, 128, 128, 128, 128 }, + { 240, 128, 255, 128, 128, 128, 128, 128, 128, 128, 128 } + }, + { /* Coeff Band ( 6 )*/ + { 1, 134, 252, 255, 255, 128, 128, 128, 128, 128, 128 }, + { 213, 62, 250, 255, 255, 128, 128, 128, 128, 128, 128 }, + { 55, 93, 255, 128, 128, 128, 128, 128, 128, 128, 128 } + }, + { /* Coeff Band ( 7 )*/ + { 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 }, + { 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 }, + { 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 } + } + }, + { /* Block Type ( 3 ) */ + { /* Coeff Band ( 0 )*/ + { 202, 24, 213, 235, 186, 191, 220, 160, 240, 175, 255 }, + { 126, 38, 182, 232, 169, 184, 228, 174, 255, 187, 128 }, + { 61, 46, 138, 219, 151, 178, 240, 170, 255, 216, 128 } + }, + { /* Coeff Band ( 1 )*/ + { 1, 112, 230, 250, 199, 191, 247, 159, 255, 255, 128 }, + { 166, 109, 228, 252, 211, 215, 255, 174, 128, 128, 128 }, + { 39, 77, 162, 232, 172, 180, 245, 178, 255, 255, 128 } + }, + { /* Coeff Band ( 2 )*/ + { 1, 52, 220, 246, 198, 199, 249, 220, 255, 255, 128 }, + { 124, 74, 191, 243, 183, 193, 250, 221, 255, 255, 128 }, + { 24, 71, 130, 219, 154, 170, 243, 182, 255, 255, 128 } + }, + { /* Coeff Band ( 3 )*/ + { 1, 182, 225, 249, 219, 240, 255, 224, 128, 128, 128 }, + { 149, 150, 226, 252, 216, 205, 255, 171, 128, 128, 128 }, + { 28, 108, 170, 242, 183, 194, 254, 223, 255, 255, 128 } + }, + { /* Coeff Band ( 4 )*/ + { 1, 81, 230, 252, 204, 203, 255, 192, 128, 128, 128 }, + { 123, 102, 209, 247, 188, 196, 255, 233, 128, 128, 128 }, + { 20, 95, 153, 243, 164, 173, 255, 203, 128, 128, 128 } + }, + { /* Coeff Band ( 5 )*/ + { 1, 222, 248, 255, 216, 213, 128, 128, 128, 128, 128 }, + { 168, 175, 246, 252, 235, 205, 255, 255, 128, 128, 128 }, + { 47, 116, 215, 255, 211, 212, 255, 255, 128, 128, 128 } + }, + { /* Coeff Band ( 6 )*/ + { 1, 121, 236, 253, 212, 214, 255, 255, 128, 128, 128 }, + { 141, 84, 213, 252, 201, 202, 255, 219, 128, 128, 128 }, + { 42, 80, 160, 240, 162, 185, 255, 205, 128, 128, 128 } + }, + { /* Coeff Band ( 7 )*/ + { 1, 1, 255, 128, 128, 128, 128, 128, 128, 128, 128 }, + { 244, 1, 255, 128, 128, 128, 128, 128, 128, 128, 128 }, + { 238, 1, 255, 128, 128, 128, 128, 128, 128, 128, 128 } + } + } +}; + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_DEFAULT_COEF_PROBS_H_ diff --git a/thirdparty/libvpx/vp8/common/dequantize.c b/thirdparty/libvpx/vp8/common/dequantize.c new file mode 100644 index 0000000000..f8b04fa4ee --- /dev/null +++ b/thirdparty/libvpx/vp8/common/dequantize.c @@ -0,0 +1,43 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "vpx_config.h" +#include "vp8_rtcd.h" +#include "vp8/common/blockd.h" +#include "vpx_mem/vpx_mem.h" + +void vp8_dequantize_b_c(BLOCKD *d, short *DQC) +{ + int i; + short *DQ = d->dqcoeff; + short *Q = d->qcoeff; + + for (i = 0; i < 16; i++) + { + DQ[i] = Q[i] * DQC[i]; + } +} + +void vp8_dequant_idct_add_c(short *input, short *dq, + unsigned char *dest, int stride) +{ + int i; + + for (i = 0; i < 16; i++) + { + input[i] = dq[i] * input[i]; + } + + vp8_short_idct4x4llm_c(input, dest, stride, dest, stride); + + memset(input, 0, 32); + +} diff --git a/thirdparty/libvpx/vp8/common/entropy.c b/thirdparty/libvpx/vp8/common/entropy.c new file mode 100644 index 0000000000..c00e565f06 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/entropy.c @@ -0,0 +1,188 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "entropy.h" +#include "blockd.h" +#include "onyxc_int.h" +#include "vpx_mem/vpx_mem.h" + +#include "coefupdateprobs.h" + +DECLARE_ALIGNED(16, const unsigned char, vp8_norm[256]) = +{ + 0, 7, 6, 6, 5, 5, 5, 5, 4, 4, 4, 4, 4, 4, 4, 4, + 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, + 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, + 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 +}; + +DECLARE_ALIGNED(16, const unsigned char, vp8_coef_bands[16]) = +{ 0, 1, 2, 3, 6, 4, 5, 6, 6, 6, 6, 6, 6, 6, 6, 7}; + +DECLARE_ALIGNED(16, const unsigned char, + vp8_prev_token_class[MAX_ENTROPY_TOKENS]) = +{ 0, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 0}; + +DECLARE_ALIGNED(16, const int, vp8_default_zig_zag1d[16]) = +{ + 0, 1, 4, 8, + 5, 2, 3, 6, + 9, 12, 13, 10, + 7, 11, 14, 15, +}; + +DECLARE_ALIGNED(16, const short, vp8_default_inv_zig_zag[16]) = +{ + 1, 2, 6, 7, + 3, 5, 8, 13, + 4, 9, 12, 14, + 10, 11, 15, 16 +}; + +/* vp8_default_zig_zag_mask generated with: + + void vp8_init_scan_order_mask() + { + int i; + + for (i = 0; i < 16; i++) + { + vp8_default_zig_zag_mask[vp8_default_zig_zag1d[i]] = 1 << i; + } + + } +*/ +DECLARE_ALIGNED(16, const short, vp8_default_zig_zag_mask[16]) = +{ + 1, 2, 32, 64, + 4, 16, 128, 4096, + 8, 256, 2048, 8192, + 512, 1024, 16384, -32768 +}; + +const int vp8_mb_feature_data_bits[MB_LVL_MAX] = {7, 6}; + +/* Array indices are identical to previously-existing CONTEXT_NODE indices */ + +const vp8_tree_index vp8_coef_tree[ 22] = /* corresponding _CONTEXT_NODEs */ +{ + -DCT_EOB_TOKEN, 2, /* 0 = EOB */ + -ZERO_TOKEN, 4, /* 1 = ZERO */ + -ONE_TOKEN, 6, /* 2 = ONE */ + 8, 12, /* 3 = LOW_VAL */ + -TWO_TOKEN, 10, /* 4 = TWO */ + -THREE_TOKEN, -FOUR_TOKEN, /* 5 = THREE */ + 14, 16, /* 6 = HIGH_LOW */ + -DCT_VAL_CATEGORY1, -DCT_VAL_CATEGORY2, /* 7 = CAT_ONE */ + 18, 20, /* 8 = CAT_THREEFOUR */ + -DCT_VAL_CATEGORY3, -DCT_VAL_CATEGORY4, /* 9 = CAT_THREE */ + -DCT_VAL_CATEGORY5, -DCT_VAL_CATEGORY6 /* 10 = CAT_FIVE */ +}; + +/* vp8_coef_encodings generated with: + vp8_tokens_from_tree(vp8_coef_encodings, vp8_coef_tree); +*/ +vp8_token vp8_coef_encodings[MAX_ENTROPY_TOKENS] = +{ + {2, 2}, + {6, 3}, + {28, 5}, + {58, 6}, + {59, 6}, + {60, 6}, + {61, 6}, + {124, 7}, + {125, 7}, + {126, 7}, + {127, 7}, + {0, 1} +}; + +/* Trees for extra bits. Probabilities are constant and + do not depend on previously encoded bits */ + +static const vp8_prob Pcat1[] = { 159}; +static const vp8_prob Pcat2[] = { 165, 145}; +static const vp8_prob Pcat3[] = { 173, 148, 140}; +static const vp8_prob Pcat4[] = { 176, 155, 140, 135}; +static const vp8_prob Pcat5[] = { 180, 157, 141, 134, 130}; +static const vp8_prob Pcat6[] = +{ 254, 254, 243, 230, 196, 177, 153, 140, 133, 130, 129}; + + +/* tree index tables generated with: + + void init_bit_tree(vp8_tree_index *p, int n) + { + int i = 0; + + while (++i < n) + { + p[0] = p[1] = i << 1; + p += 2; + } + + p[0] = p[1] = 0; + } + + void init_bit_trees() + { + init_bit_tree(cat1, 1); + init_bit_tree(cat2, 2); + init_bit_tree(cat3, 3); + init_bit_tree(cat4, 4); + init_bit_tree(cat5, 5); + init_bit_tree(cat6, 11); + } +*/ + +static const vp8_tree_index cat1[2] = { 0, 0 }; +static const vp8_tree_index cat2[4] = { 2, 2, 0, 0 }; +static const vp8_tree_index cat3[6] = { 2, 2, 4, 4, 0, 0 }; +static const vp8_tree_index cat4[8] = { 2, 2, 4, 4, 6, 6, 0, 0 }; +static const vp8_tree_index cat5[10] = { 2, 2, 4, 4, 6, 6, 8, 8, 0, 0 }; +static const vp8_tree_index cat6[22] = { 2, 2, 4, 4, 6, 6, 8, 8, 10, 10, 12, 12, + 14, 14, 16, 16, 18, 18, 20, 20, 0, 0 }; + +const vp8_extra_bit_struct vp8_extra_bits[12] = +{ + { 0, 0, 0, 0}, + { 0, 0, 0, 1}, + { 0, 0, 0, 2}, + { 0, 0, 0, 3}, + { 0, 0, 0, 4}, + { cat1, Pcat1, 1, 5}, + { cat2, Pcat2, 2, 7}, + { cat3, Pcat3, 3, 11}, + { cat4, Pcat4, 4, 19}, + { cat5, Pcat5, 5, 35}, + { cat6, Pcat6, 11, 67}, + { 0, 0, 0, 0} +}; + +#include "default_coef_probs.h" + +void vp8_default_coef_probs(VP8_COMMON *pc) +{ + memcpy(pc->fc.coef_probs, default_coef_probs, sizeof(default_coef_probs)); +} + diff --git a/thirdparty/libvpx/vp8/common/entropy.h b/thirdparty/libvpx/vp8/common/entropy.h new file mode 100644 index 0000000000..a90bab4bac --- /dev/null +++ b/thirdparty/libvpx/vp8/common/entropy.h @@ -0,0 +1,109 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_ENTROPY_H_ +#define VP8_COMMON_ENTROPY_H_ + +#include "treecoder.h" +#include "blockd.h" + +#ifdef __cplusplus +extern "C" { +#endif + +/* Coefficient token alphabet */ + +#define ZERO_TOKEN 0 /* 0 Extra Bits 0+0 */ +#define ONE_TOKEN 1 /* 1 Extra Bits 0+1 */ +#define TWO_TOKEN 2 /* 2 Extra Bits 0+1 */ +#define THREE_TOKEN 3 /* 3 Extra Bits 0+1 */ +#define FOUR_TOKEN 4 /* 4 Extra Bits 0+1 */ +#define DCT_VAL_CATEGORY1 5 /* 5-6 Extra Bits 1+1 */ +#define DCT_VAL_CATEGORY2 6 /* 7-10 Extra Bits 2+1 */ +#define DCT_VAL_CATEGORY3 7 /* 11-18 Extra Bits 3+1 */ +#define DCT_VAL_CATEGORY4 8 /* 19-34 Extra Bits 4+1 */ +#define DCT_VAL_CATEGORY5 9 /* 35-66 Extra Bits 5+1 */ +#define DCT_VAL_CATEGORY6 10 /* 67+ Extra Bits 11+1 */ +#define DCT_EOB_TOKEN 11 /* EOB Extra Bits 0+0 */ + +#define MAX_ENTROPY_TOKENS 12 +#define ENTROPY_NODES 11 + +extern const vp8_tree_index vp8_coef_tree[]; + +extern const struct vp8_token_struct vp8_coef_encodings[MAX_ENTROPY_TOKENS]; + +typedef struct +{ + vp8_tree_p tree; + const vp8_prob *prob; + int Len; + int base_val; +} vp8_extra_bit_struct; + +extern const vp8_extra_bit_struct vp8_extra_bits[12]; /* indexed by token value */ + +#define PROB_UPDATE_BASELINE_COST 7 + +#define MAX_PROB 255 +#define DCT_MAX_VALUE 2048 + + +/* Coefficients are predicted via a 3-dimensional probability table. */ + +/* Outside dimension. 0 = Y no DC, 1 = Y2, 2 = UV, 3 = Y with DC */ + +#define BLOCK_TYPES 4 + +/* Middle dimension is a coarsening of the coefficient's + position within the 4x4 DCT. */ + +#define COEF_BANDS 8 +extern DECLARE_ALIGNED(16, const unsigned char, vp8_coef_bands[16]); + +/* Inside dimension is 3-valued measure of nearby complexity, that is, + the extent to which nearby coefficients are nonzero. For the first + coefficient (DC, unless block type is 0), we look at the (already encoded) + blocks above and to the left of the current block. The context index is + then the number (0,1,or 2) of these blocks having nonzero coefficients. + After decoding a coefficient, the measure is roughly the size of the + most recently decoded coefficient (0 for 0, 1 for 1, 2 for >1). + Note that the intuitive meaning of this measure changes as coefficients + are decoded, e.g., prior to the first token, a zero means that my neighbors + are empty while, after the first token, because of the use of end-of-block, + a zero means we just decoded a zero and hence guarantees that a non-zero + coefficient will appear later in this block. However, this shift + in meaning is perfectly OK because our context depends also on the + coefficient band (and since zigzag positions 0, 1, and 2 are in + distinct bands). */ + +/*# define DC_TOKEN_CONTEXTS 3*/ /* 00, 0!0, !0!0 */ +# define PREV_COEF_CONTEXTS 3 + +extern DECLARE_ALIGNED(16, const unsigned char, vp8_prev_token_class[MAX_ENTROPY_TOKENS]); + +extern const vp8_prob vp8_coef_update_probs [BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [ENTROPY_NODES]; + + +struct VP8Common; +void vp8_default_coef_probs(struct VP8Common *); + +extern DECLARE_ALIGNED(16, const int, vp8_default_zig_zag1d[16]); +extern DECLARE_ALIGNED(16, const short, vp8_default_inv_zig_zag[16]); +extern DECLARE_ALIGNED(16, const short, vp8_default_zig_zag_mask[16]); +extern const int vp8_mb_feature_data_bits[MB_LVL_MAX]; + +void vp8_coef_tree_initialize(void); +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_ENTROPY_H_ diff --git a/thirdparty/libvpx/vp8/common/entropymode.c b/thirdparty/libvpx/vp8/common/entropymode.c new file mode 100644 index 0000000000..8981a8d3c2 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/entropymode.c @@ -0,0 +1,171 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#define USE_PREBUILT_TABLES + +#include "entropymode.h" +#include "entropy.h" +#include "vpx_mem/vpx_mem.h" + +#include "vp8_entropymodedata.h" + +int vp8_mv_cont(const int_mv *l, const int_mv *a) +{ + int lez = (l->as_int == 0); + int aez = (a->as_int == 0); + int lea = (l->as_int == a->as_int); + + if (lea && lez) + return SUBMVREF_LEFT_ABOVE_ZED; + + if (lea) + return SUBMVREF_LEFT_ABOVE_SAME; + + if (aez) + return SUBMVREF_ABOVE_ZED; + + if (lez) + return SUBMVREF_LEFT_ZED; + + return SUBMVREF_NORMAL; +} + +static const vp8_prob sub_mv_ref_prob [VP8_SUBMVREFS-1] = { 180, 162, 25}; + +const vp8_prob vp8_sub_mv_ref_prob2 [SUBMVREF_COUNT][VP8_SUBMVREFS-1] = +{ + { 147, 136, 18 }, + { 106, 145, 1 }, + { 179, 121, 1 }, + { 223, 1 , 34 }, + { 208, 1 , 1 } +}; + + + +const vp8_mbsplit vp8_mbsplits [VP8_NUMMBSPLITS] = +{ + { + 0, 0, 0, 0, + 0, 0, 0, 0, + 1, 1, 1, 1, + 1, 1, 1, 1, + }, + { + 0, 0, 1, 1, + 0, 0, 1, 1, + 0, 0, 1, 1, + 0, 0, 1, 1, + }, + { + 0, 0, 1, 1, + 0, 0, 1, 1, + 2, 2, 3, 3, + 2, 2, 3, 3, + }, + { + 0, 1, 2, 3, + 4, 5, 6, 7, + 8, 9, 10, 11, + 12, 13, 14, 15, + } +}; + +const int vp8_mbsplit_count [VP8_NUMMBSPLITS] = { 2, 2, 4, 16}; + +const vp8_prob vp8_mbsplit_probs [VP8_NUMMBSPLITS-1] = { 110, 111, 150}; + + +/* Array indices are identical to previously-existing INTRAMODECONTEXTNODES. */ + +const vp8_tree_index vp8_bmode_tree[18] = /* INTRAMODECONTEXTNODE value */ +{ + -B_DC_PRED, 2, /* 0 = DC_NODE */ + -B_TM_PRED, 4, /* 1 = TM_NODE */ + -B_VE_PRED, 6, /* 2 = VE_NODE */ + 8, 12, /* 3 = COM_NODE */ + -B_HE_PRED, 10, /* 4 = HE_NODE */ + -B_RD_PRED, -B_VR_PRED, /* 5 = RD_NODE */ + -B_LD_PRED, 14, /* 6 = LD_NODE */ + -B_VL_PRED, 16, /* 7 = VL_NODE */ + -B_HD_PRED, -B_HU_PRED /* 8 = HD_NODE */ +}; + +/* Again, these trees use the same probability indices as their + explicitly-programmed predecessors. */ + +const vp8_tree_index vp8_ymode_tree[8] = +{ + -DC_PRED, 2, + 4, 6, + -V_PRED, -H_PRED, + -TM_PRED, -B_PRED +}; + +const vp8_tree_index vp8_kf_ymode_tree[8] = +{ + -B_PRED, 2, + 4, 6, + -DC_PRED, -V_PRED, + -H_PRED, -TM_PRED +}; + +const vp8_tree_index vp8_uv_mode_tree[6] = +{ + -DC_PRED, 2, + -V_PRED, 4, + -H_PRED, -TM_PRED +}; + +const vp8_tree_index vp8_mbsplit_tree[6] = +{ + -3, 2, + -2, 4, + -0, -1 +}; + +const vp8_tree_index vp8_mv_ref_tree[8] = +{ + -ZEROMV, 2, + -NEARESTMV, 4, + -NEARMV, 6, + -NEWMV, -SPLITMV +}; + +const vp8_tree_index vp8_sub_mv_ref_tree[6] = +{ + -LEFT4X4, 2, + -ABOVE4X4, 4, + -ZERO4X4, -NEW4X4 +}; + +const vp8_tree_index vp8_small_mvtree [14] = +{ + 2, 8, + 4, 6, + -0, -1, + -2, -3, + 10, 12, + -4, -5, + -6, -7 +}; + +void vp8_init_mbmode_probs(VP8_COMMON *x) +{ + memcpy(x->fc.ymode_prob, vp8_ymode_prob, sizeof(vp8_ymode_prob)); + memcpy(x->fc.uv_mode_prob, vp8_uv_mode_prob, sizeof(vp8_uv_mode_prob)); + memcpy(x->fc.sub_mv_ref_prob, sub_mv_ref_prob, sizeof(sub_mv_ref_prob)); +} + +void vp8_default_bmode_probs(vp8_prob p [VP8_BINTRAMODES-1]) +{ + memcpy(p, vp8_bmode_prob, sizeof(vp8_bmode_prob)); +} + diff --git a/thirdparty/libvpx/vp8/common/entropymode.h b/thirdparty/libvpx/vp8/common/entropymode.h new file mode 100644 index 0000000000..81bdfc4b8b --- /dev/null +++ b/thirdparty/libvpx/vp8/common/entropymode.h @@ -0,0 +1,88 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_ENTROPYMODE_H_ +#define VP8_COMMON_ENTROPYMODE_H_ + +#include "onyxc_int.h" +#include "treecoder.h" + +#ifdef __cplusplus +extern "C" { +#endif + +typedef enum +{ + SUBMVREF_NORMAL, + SUBMVREF_LEFT_ZED, + SUBMVREF_ABOVE_ZED, + SUBMVREF_LEFT_ABOVE_SAME, + SUBMVREF_LEFT_ABOVE_ZED +} sumvfref_t; + +typedef int vp8_mbsplit[16]; + +#define VP8_NUMMBSPLITS 4 + +extern const vp8_mbsplit vp8_mbsplits [VP8_NUMMBSPLITS]; + +extern const int vp8_mbsplit_count [VP8_NUMMBSPLITS]; /* # of subsets */ + +extern const vp8_prob vp8_mbsplit_probs [VP8_NUMMBSPLITS-1]; + +extern int vp8_mv_cont(const int_mv *l, const int_mv *a); +#define SUBMVREF_COUNT 5 +extern const vp8_prob vp8_sub_mv_ref_prob2 [SUBMVREF_COUNT][VP8_SUBMVREFS-1]; + + +extern const unsigned int vp8_kf_default_bmode_counts [VP8_BINTRAMODES] [VP8_BINTRAMODES] [VP8_BINTRAMODES]; + + +extern const vp8_tree_index vp8_bmode_tree[]; + +extern const vp8_tree_index vp8_ymode_tree[]; +extern const vp8_tree_index vp8_kf_ymode_tree[]; +extern const vp8_tree_index vp8_uv_mode_tree[]; + +extern const vp8_tree_index vp8_mbsplit_tree[]; +extern const vp8_tree_index vp8_mv_ref_tree[]; +extern const vp8_tree_index vp8_sub_mv_ref_tree[]; + +extern const struct vp8_token_struct vp8_bmode_encodings[VP8_BINTRAMODES]; +extern const struct vp8_token_struct vp8_ymode_encodings[VP8_YMODES]; +extern const struct vp8_token_struct vp8_kf_ymode_encodings[VP8_YMODES]; +extern const struct vp8_token_struct vp8_uv_mode_encodings[VP8_UV_MODES]; +extern const struct vp8_token_struct vp8_mbsplit_encodings[VP8_NUMMBSPLITS]; + +/* Inter mode values do not start at zero */ + +extern const struct vp8_token_struct vp8_mv_ref_encoding_array[VP8_MVREFS]; +extern const struct vp8_token_struct vp8_sub_mv_ref_encoding_array[VP8_SUBMVREFS]; + +extern const vp8_tree_index vp8_small_mvtree[]; + +extern const struct vp8_token_struct vp8_small_mvencodings[8]; + +/* Key frame default mode probs */ +extern const vp8_prob vp8_kf_bmode_prob[VP8_BINTRAMODES][VP8_BINTRAMODES] +[VP8_BINTRAMODES-1]; +extern const vp8_prob vp8_kf_uv_mode_prob[VP8_UV_MODES-1]; +extern const vp8_prob vp8_kf_ymode_prob[VP8_YMODES-1]; + +void vp8_init_mbmode_probs(VP8_COMMON *x); +void vp8_default_bmode_probs(vp8_prob dest [VP8_BINTRAMODES-1]); +void vp8_kf_default_bmode_probs(vp8_prob dest [VP8_BINTRAMODES] [VP8_BINTRAMODES] [VP8_BINTRAMODES-1]); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_ENTROPYMODE_H_ diff --git a/thirdparty/libvpx/vp8/common/entropymv.c b/thirdparty/libvpx/vp8/common/entropymv.c new file mode 100644 index 0000000000..e5df1f0955 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/entropymv.c @@ -0,0 +1,49 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "entropymv.h" + +const MV_CONTEXT vp8_mv_update_probs[2] = +{ + {{ + 237, + 246, + 253, 253, 254, 254, 254, 254, 254, + 254, 254, 254, 254, 254, 250, 250, 252, 254, 254 + }}, + {{ + 231, + 243, + 245, 253, 254, 254, 254, 254, 254, + 254, 254, 254, 254, 254, 251, 251, 254, 254, 254 + }} +}; +const MV_CONTEXT vp8_default_mv_context[2] = +{ + {{ + /* row */ + 162, /* is short */ + 128, /* sign */ + 225, 146, 172, 147, 214, 39, 156, /* short tree */ + 128, 129, 132, 75, 145, 178, 206, 239, 254, 254 /* long bits */ + }}, + + + + {{ + /* same for column */ + 164, /* is short */ + 128, + 204, 170, 119, 235, 140, 230, 228, + 128, 130, 130, 74, 148, 180, 203, 236, 254, 254 /* long bits */ + + }} +}; diff --git a/thirdparty/libvpx/vp8/common/entropymv.h b/thirdparty/libvpx/vp8/common/entropymv.h new file mode 100644 index 0000000000..42840d58ad --- /dev/null +++ b/thirdparty/libvpx/vp8/common/entropymv.h @@ -0,0 +1,52 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_ENTROPYMV_H_ +#define VP8_COMMON_ENTROPYMV_H_ + +#include "treecoder.h" + +#ifdef __cplusplus +extern "C" { +#endif + +enum +{ + mv_max = 1023, /* max absolute value of a MV component */ + MVvals = (2 * mv_max) + 1, /* # possible values "" */ + mvfp_max = 255, /* max absolute value of a full pixel MV component */ + MVfpvals = (2 * mvfp_max) +1, /* # possible full pixel MV values */ + + mvlong_width = 10, /* Large MVs have 9 bit magnitudes */ + mvnum_short = 8, /* magnitudes 0 through 7 */ + + /* probability offsets for coding each MV component */ + + mvpis_short = 0, /* short (<= 7) vs long (>= 8) */ + MVPsign, /* sign for non-zero */ + MVPshort, /* 8 short values = 7-position tree */ + + MVPbits = MVPshort + mvnum_short - 1, /* mvlong_width long value bits */ + MVPcount = MVPbits + mvlong_width /* (with independent probabilities) */ +}; + +typedef struct mv_context +{ + vp8_prob prob[MVPcount]; /* often come in row, col pairs */ +} MV_CONTEXT; + +extern const MV_CONTEXT vp8_mv_update_probs[2], vp8_default_mv_context[2]; + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_ENTROPYMV_H_ diff --git a/thirdparty/libvpx/vp8/common/extend.c b/thirdparty/libvpx/vp8/common/extend.c new file mode 100644 index 0000000000..2d938ad782 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/extend.c @@ -0,0 +1,188 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "extend.h" +#include "vpx_mem/vpx_mem.h" + + +static void copy_and_extend_plane +( + unsigned char *s, /* source */ + int sp, /* source pitch */ + unsigned char *d, /* destination */ + int dp, /* destination pitch */ + int h, /* height */ + int w, /* width */ + int et, /* extend top border */ + int el, /* extend left border */ + int eb, /* extend bottom border */ + int er /* extend right border */ +) +{ + int i; + unsigned char *src_ptr1, *src_ptr2; + unsigned char *dest_ptr1, *dest_ptr2; + int linesize; + + /* copy the left and right most columns out */ + src_ptr1 = s; + src_ptr2 = s + w - 1; + dest_ptr1 = d - el; + dest_ptr2 = d + w; + + for (i = 0; i < h; i++) + { + memset(dest_ptr1, src_ptr1[0], el); + memcpy(dest_ptr1 + el, src_ptr1, w); + memset(dest_ptr2, src_ptr2[0], er); + src_ptr1 += sp; + src_ptr2 += sp; + dest_ptr1 += dp; + dest_ptr2 += dp; + } + + /* Now copy the top and bottom lines into each line of the respective + * borders + */ + src_ptr1 = d - el; + src_ptr2 = d + dp * (h - 1) - el; + dest_ptr1 = d + dp * (-et) - el; + dest_ptr2 = d + dp * (h) - el; + linesize = el + er + w; + + for (i = 0; i < et; i++) + { + memcpy(dest_ptr1, src_ptr1, linesize); + dest_ptr1 += dp; + } + + for (i = 0; i < eb; i++) + { + memcpy(dest_ptr2, src_ptr2, linesize); + dest_ptr2 += dp; + } +} + + +void vp8_copy_and_extend_frame(YV12_BUFFER_CONFIG *src, + YV12_BUFFER_CONFIG *dst) +{ + int et = dst->border; + int el = dst->border; + int eb = dst->border + dst->y_height - src->y_height; + int er = dst->border + dst->y_width - src->y_width; + + copy_and_extend_plane(src->y_buffer, src->y_stride, + dst->y_buffer, dst->y_stride, + src->y_height, src->y_width, + et, el, eb, er); + + et = dst->border >> 1; + el = dst->border >> 1; + eb = (dst->border >> 1) + dst->uv_height - src->uv_height; + er = (dst->border >> 1) + dst->uv_width - src->uv_width; + + copy_and_extend_plane(src->u_buffer, src->uv_stride, + dst->u_buffer, dst->uv_stride, + src->uv_height, src->uv_width, + et, el, eb, er); + + copy_and_extend_plane(src->v_buffer, src->uv_stride, + dst->v_buffer, dst->uv_stride, + src->uv_height, src->uv_width, + et, el, eb, er); +} + + +void vp8_copy_and_extend_frame_with_rect(YV12_BUFFER_CONFIG *src, + YV12_BUFFER_CONFIG *dst, + int srcy, int srcx, + int srch, int srcw) +{ + int et = dst->border; + int el = dst->border; + int eb = dst->border + dst->y_height - src->y_height; + int er = dst->border + dst->y_width - src->y_width; + int src_y_offset = srcy * src->y_stride + srcx; + int dst_y_offset = srcy * dst->y_stride + srcx; + int src_uv_offset = ((srcy * src->uv_stride) >> 1) + (srcx >> 1); + int dst_uv_offset = ((srcy * dst->uv_stride) >> 1) + (srcx >> 1); + + /* If the side is not touching the bounder then don't extend. */ + if (srcy) + et = 0; + if (srcx) + el = 0; + if (srcy + srch != src->y_height) + eb = 0; + if (srcx + srcw != src->y_width) + er = 0; + + copy_and_extend_plane(src->y_buffer + src_y_offset, + src->y_stride, + dst->y_buffer + dst_y_offset, + dst->y_stride, + srch, srcw, + et, el, eb, er); + + et = (et + 1) >> 1; + el = (el + 1) >> 1; + eb = (eb + 1) >> 1; + er = (er + 1) >> 1; + srch = (srch + 1) >> 1; + srcw = (srcw + 1) >> 1; + + copy_and_extend_plane(src->u_buffer + src_uv_offset, + src->uv_stride, + dst->u_buffer + dst_uv_offset, + dst->uv_stride, + srch, srcw, + et, el, eb, er); + + copy_and_extend_plane(src->v_buffer + src_uv_offset, + src->uv_stride, + dst->v_buffer + dst_uv_offset, + dst->uv_stride, + srch, srcw, + et, el, eb, er); +} + + +/* note the extension is only for the last row, for intra prediction purpose */ +void vp8_extend_mb_row(YV12_BUFFER_CONFIG *ybf, + unsigned char *YPtr, + unsigned char *UPtr, + unsigned char *VPtr) +{ + int i; + + YPtr += ybf->y_stride * 14; + UPtr += ybf->uv_stride * 6; + VPtr += ybf->uv_stride * 6; + + for (i = 0; i < 4; i++) + { + YPtr[i] = YPtr[-1]; + UPtr[i] = UPtr[-1]; + VPtr[i] = VPtr[-1]; + } + + YPtr += ybf->y_stride; + UPtr += ybf->uv_stride; + VPtr += ybf->uv_stride; + + for (i = 0; i < 4; i++) + { + YPtr[i] = YPtr[-1]; + UPtr[i] = UPtr[-1]; + VPtr[i] = VPtr[-1]; + } +} diff --git a/thirdparty/libvpx/vp8/common/extend.h b/thirdparty/libvpx/vp8/common/extend.h new file mode 100644 index 0000000000..068f4ac523 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/extend.h @@ -0,0 +1,33 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_EXTEND_H_ +#define VP8_COMMON_EXTEND_H_ + +#include "vpx_scale/yv12config.h" + +#ifdef __cplusplus +extern "C" { +#endif + +void vp8_extend_mb_row(YV12_BUFFER_CONFIG *ybf, unsigned char *YPtr, unsigned char *UPtr, unsigned char *VPtr); +void vp8_copy_and_extend_frame(YV12_BUFFER_CONFIG *src, + YV12_BUFFER_CONFIG *dst); +void vp8_copy_and_extend_frame_with_rect(YV12_BUFFER_CONFIG *src, + YV12_BUFFER_CONFIG *dst, + int srcy, int srcx, + int srch, int srcw); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_EXTEND_H_ diff --git a/thirdparty/libvpx/vp8/common/filter.c b/thirdparty/libvpx/vp8/common/filter.c new file mode 100644 index 0000000000..84c608effa --- /dev/null +++ b/thirdparty/libvpx/vp8/common/filter.c @@ -0,0 +1,493 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "filter.h" +#include "./vp8_rtcd.h" + +DECLARE_ALIGNED(16, const short, vp8_bilinear_filters[8][2]) = +{ + { 128, 0 }, + { 112, 16 }, + { 96, 32 }, + { 80, 48 }, + { 64, 64 }, + { 48, 80 }, + { 32, 96 }, + { 16, 112 } +}; + +DECLARE_ALIGNED(16, const short, vp8_sub_pel_filters[8][6]) = +{ + + { 0, 0, 128, 0, 0, 0 }, /* note that 1/8 pel positions are just as per alpha -0.5 bicubic */ + { 0, -6, 123, 12, -1, 0 }, + { 2, -11, 108, 36, -8, 1 }, /* New 1/4 pel 6 tap filter */ + { 0, -9, 93, 50, -6, 0 }, + { 3, -16, 77, 77, -16, 3 }, /* New 1/2 pel 6 tap filter */ + { 0, -6, 50, 93, -9, 0 }, + { 1, -8, 36, 108, -11, 2 }, /* New 1/4 pel 6 tap filter */ + { 0, -1, 12, 123, -6, 0 }, +}; + +static void filter_block2d_first_pass +( + unsigned char *src_ptr, + int *output_ptr, + unsigned int src_pixels_per_line, + unsigned int pixel_step, + unsigned int output_height, + unsigned int output_width, + const short *vp8_filter +) +{ + unsigned int i, j; + int Temp; + + for (i = 0; i < output_height; i++) + { + for (j = 0; j < output_width; j++) + { + Temp = ((int)src_ptr[-2 * (int)pixel_step] * vp8_filter[0]) + + ((int)src_ptr[-1 * (int)pixel_step] * vp8_filter[1]) + + ((int)src_ptr[0] * vp8_filter[2]) + + ((int)src_ptr[pixel_step] * vp8_filter[3]) + + ((int)src_ptr[2*pixel_step] * vp8_filter[4]) + + ((int)src_ptr[3*pixel_step] * vp8_filter[5]) + + (VP8_FILTER_WEIGHT >> 1); /* Rounding */ + + /* Normalize back to 0-255 */ + Temp = Temp >> VP8_FILTER_SHIFT; + + if (Temp < 0) + Temp = 0; + else if (Temp > 255) + Temp = 255; + + output_ptr[j] = Temp; + src_ptr++; + } + + /* Next row... */ + src_ptr += src_pixels_per_line - output_width; + output_ptr += output_width; + } +} + +static void filter_block2d_second_pass +( + int *src_ptr, + unsigned char *output_ptr, + int output_pitch, + unsigned int src_pixels_per_line, + unsigned int pixel_step, + unsigned int output_height, + unsigned int output_width, + const short *vp8_filter +) +{ + unsigned int i, j; + int Temp; + + for (i = 0; i < output_height; i++) + { + for (j = 0; j < output_width; j++) + { + /* Apply filter */ + Temp = ((int)src_ptr[-2 * (int)pixel_step] * vp8_filter[0]) + + ((int)src_ptr[-1 * (int)pixel_step] * vp8_filter[1]) + + ((int)src_ptr[0] * vp8_filter[2]) + + ((int)src_ptr[pixel_step] * vp8_filter[3]) + + ((int)src_ptr[2*pixel_step] * vp8_filter[4]) + + ((int)src_ptr[3*pixel_step] * vp8_filter[5]) + + (VP8_FILTER_WEIGHT >> 1); /* Rounding */ + + /* Normalize back to 0-255 */ + Temp = Temp >> VP8_FILTER_SHIFT; + + if (Temp < 0) + Temp = 0; + else if (Temp > 255) + Temp = 255; + + output_ptr[j] = (unsigned char)Temp; + src_ptr++; + } + + /* Start next row */ + src_ptr += src_pixels_per_line - output_width; + output_ptr += output_pitch; + } +} + + +static void filter_block2d +( + unsigned char *src_ptr, + unsigned char *output_ptr, + unsigned int src_pixels_per_line, + int output_pitch, + const short *HFilter, + const short *VFilter +) +{ + int FData[9*4]; /* Temp data buffer used in filtering */ + + /* First filter 1-D horizontally... */ + filter_block2d_first_pass(src_ptr - (2 * src_pixels_per_line), FData, src_pixels_per_line, 1, 9, 4, HFilter); + + /* then filter verticaly... */ + filter_block2d_second_pass(FData + 8, output_ptr, output_pitch, 4, 4, 4, 4, VFilter); +} + + +void vp8_sixtap_predict4x4_c +( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch +) +{ + const short *HFilter; + const short *VFilter; + + HFilter = vp8_sub_pel_filters[xoffset]; /* 6 tap */ + VFilter = vp8_sub_pel_filters[yoffset]; /* 6 tap */ + + filter_block2d(src_ptr, dst_ptr, src_pixels_per_line, dst_pitch, HFilter, VFilter); +} +void vp8_sixtap_predict8x8_c +( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch +) +{ + const short *HFilter; + const short *VFilter; + int FData[13*16]; /* Temp data buffer used in filtering */ + + HFilter = vp8_sub_pel_filters[xoffset]; /* 6 tap */ + VFilter = vp8_sub_pel_filters[yoffset]; /* 6 tap */ + + /* First filter 1-D horizontally... */ + filter_block2d_first_pass(src_ptr - (2 * src_pixels_per_line), FData, src_pixels_per_line, 1, 13, 8, HFilter); + + + /* then filter verticaly... */ + filter_block2d_second_pass(FData + 16, dst_ptr, dst_pitch, 8, 8, 8, 8, VFilter); + +} + +void vp8_sixtap_predict8x4_c +( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch +) +{ + const short *HFilter; + const short *VFilter; + int FData[13*16]; /* Temp data buffer used in filtering */ + + HFilter = vp8_sub_pel_filters[xoffset]; /* 6 tap */ + VFilter = vp8_sub_pel_filters[yoffset]; /* 6 tap */ + + /* First filter 1-D horizontally... */ + filter_block2d_first_pass(src_ptr - (2 * src_pixels_per_line), FData, src_pixels_per_line, 1, 9, 8, HFilter); + + + /* then filter verticaly... */ + filter_block2d_second_pass(FData + 16, dst_ptr, dst_pitch, 8, 8, 4, 8, VFilter); + +} + +void vp8_sixtap_predict16x16_c +( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch +) +{ + const short *HFilter; + const short *VFilter; + int FData[21*24]; /* Temp data buffer used in filtering */ + + + HFilter = vp8_sub_pel_filters[xoffset]; /* 6 tap */ + VFilter = vp8_sub_pel_filters[yoffset]; /* 6 tap */ + + /* First filter 1-D horizontally... */ + filter_block2d_first_pass(src_ptr - (2 * src_pixels_per_line), FData, src_pixels_per_line, 1, 21, 16, HFilter); + + /* then filter verticaly... */ + filter_block2d_second_pass(FData + 32, dst_ptr, dst_pitch, 16, 16, 16, 16, VFilter); + +} + + +/**************************************************************************** + * + * ROUTINE : filter_block2d_bil_first_pass + * + * INPUTS : UINT8 *src_ptr : Pointer to source block. + * UINT32 src_stride : Stride of source block. + * UINT32 height : Block height. + * UINT32 width : Block width. + * INT32 *vp8_filter : Array of 2 bi-linear filter taps. + * + * OUTPUTS : INT32 *dst_ptr : Pointer to filtered block. + * + * RETURNS : void + * + * FUNCTION : Applies a 1-D 2-tap bi-linear filter to the source block + * in the horizontal direction to produce the filtered output + * block. Used to implement first-pass of 2-D separable filter. + * + * SPECIAL NOTES : Produces INT32 output to retain precision for next pass. + * Two filter taps should sum to VP8_FILTER_WEIGHT. + * + ****************************************************************************/ +static void filter_block2d_bil_first_pass +( + unsigned char *src_ptr, + unsigned short *dst_ptr, + unsigned int src_stride, + unsigned int height, + unsigned int width, + const short *vp8_filter +) +{ + unsigned int i, j; + + for (i = 0; i < height; i++) + { + for (j = 0; j < width; j++) + { + /* Apply bilinear filter */ + dst_ptr[j] = (((int)src_ptr[0] * vp8_filter[0]) + + ((int)src_ptr[1] * vp8_filter[1]) + + (VP8_FILTER_WEIGHT / 2)) >> VP8_FILTER_SHIFT; + src_ptr++; + } + + /* Next row... */ + src_ptr += src_stride - width; + dst_ptr += width; + } +} + +/**************************************************************************** + * + * ROUTINE : filter_block2d_bil_second_pass + * + * INPUTS : INT32 *src_ptr : Pointer to source block. + * UINT32 dst_pitch : Destination block pitch. + * UINT32 height : Block height. + * UINT32 width : Block width. + * INT32 *vp8_filter : Array of 2 bi-linear filter taps. + * + * OUTPUTS : UINT16 *dst_ptr : Pointer to filtered block. + * + * RETURNS : void + * + * FUNCTION : Applies a 1-D 2-tap bi-linear filter to the source block + * in the vertical direction to produce the filtered output + * block. Used to implement second-pass of 2-D separable filter. + * + * SPECIAL NOTES : Requires 32-bit input as produced by filter_block2d_bil_first_pass. + * Two filter taps should sum to VP8_FILTER_WEIGHT. + * + ****************************************************************************/ +static void filter_block2d_bil_second_pass +( + unsigned short *src_ptr, + unsigned char *dst_ptr, + int dst_pitch, + unsigned int height, + unsigned int width, + const short *vp8_filter +) +{ + unsigned int i, j; + int Temp; + + for (i = 0; i < height; i++) + { + for (j = 0; j < width; j++) + { + /* Apply filter */ + Temp = ((int)src_ptr[0] * vp8_filter[0]) + + ((int)src_ptr[width] * vp8_filter[1]) + + (VP8_FILTER_WEIGHT / 2); + dst_ptr[j] = (unsigned int)(Temp >> VP8_FILTER_SHIFT); + src_ptr++; + } + + /* Next row... */ + dst_ptr += dst_pitch; + } +} + + +/**************************************************************************** + * + * ROUTINE : filter_block2d_bil + * + * INPUTS : UINT8 *src_ptr : Pointer to source block. + * UINT32 src_pitch : Stride of source block. + * UINT32 dst_pitch : Stride of destination block. + * INT32 *HFilter : Array of 2 horizontal filter taps. + * INT32 *VFilter : Array of 2 vertical filter taps. + * INT32 Width : Block width + * INT32 Height : Block height + * + * OUTPUTS : UINT16 *dst_ptr : Pointer to filtered block. + * + * RETURNS : void + * + * FUNCTION : 2-D filters an input block by applying a 2-tap + * bi-linear filter horizontally followed by a 2-tap + * bi-linear filter vertically on the result. + * + * SPECIAL NOTES : The largest block size can be handled here is 16x16 + * + ****************************************************************************/ +static void filter_block2d_bil +( + unsigned char *src_ptr, + unsigned char *dst_ptr, + unsigned int src_pitch, + unsigned int dst_pitch, + const short *HFilter, + const short *VFilter, + int Width, + int Height +) +{ + + unsigned short FData[17*16]; /* Temp data buffer used in filtering */ + + /* First filter 1-D horizontally... */ + filter_block2d_bil_first_pass(src_ptr, FData, src_pitch, Height + 1, Width, HFilter); + + /* then 1-D vertically... */ + filter_block2d_bil_second_pass(FData, dst_ptr, dst_pitch, Height, Width, VFilter); +} + + +void vp8_bilinear_predict4x4_c +( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch +) +{ + const short *HFilter; + const short *VFilter; + + HFilter = vp8_bilinear_filters[xoffset]; + VFilter = vp8_bilinear_filters[yoffset]; +#if 0 + { + int i; + unsigned char temp1[16]; + unsigned char temp2[16]; + + bilinear_predict4x4_mmx(src_ptr, src_pixels_per_line, xoffset, yoffset, temp1, 4); + filter_block2d_bil(src_ptr, temp2, src_pixels_per_line, 4, HFilter, VFilter, 4, 4); + + for (i = 0; i < 16; i++) + { + if (temp1[i] != temp2[i]) + { + bilinear_predict4x4_mmx(src_ptr, src_pixels_per_line, xoffset, yoffset, temp1, 4); + filter_block2d_bil(src_ptr, temp2, src_pixels_per_line, 4, HFilter, VFilter, 4, 4); + } + } + } +#endif + filter_block2d_bil(src_ptr, dst_ptr, src_pixels_per_line, dst_pitch, HFilter, VFilter, 4, 4); + +} + +void vp8_bilinear_predict8x8_c +( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch +) +{ + const short *HFilter; + const short *VFilter; + + HFilter = vp8_bilinear_filters[xoffset]; + VFilter = vp8_bilinear_filters[yoffset]; + + filter_block2d_bil(src_ptr, dst_ptr, src_pixels_per_line, dst_pitch, HFilter, VFilter, 8, 8); + +} + +void vp8_bilinear_predict8x4_c +( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch +) +{ + const short *HFilter; + const short *VFilter; + + HFilter = vp8_bilinear_filters[xoffset]; + VFilter = vp8_bilinear_filters[yoffset]; + + filter_block2d_bil(src_ptr, dst_ptr, src_pixels_per_line, dst_pitch, HFilter, VFilter, 8, 4); + +} + +void vp8_bilinear_predict16x16_c +( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch +) +{ + const short *HFilter; + const short *VFilter; + + HFilter = vp8_bilinear_filters[xoffset]; + VFilter = vp8_bilinear_filters[yoffset]; + + filter_block2d_bil(src_ptr, dst_ptr, src_pixels_per_line, dst_pitch, HFilter, VFilter, 16, 16); +} diff --git a/thirdparty/libvpx/vp8/common/filter.h b/thirdparty/libvpx/vp8/common/filter.h new file mode 100644 index 0000000000..cfba775fce --- /dev/null +++ b/thirdparty/libvpx/vp8/common/filter.h @@ -0,0 +1,32 @@ +/* + * Copyright (c) 2011 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_FILTER_H_ +#define VP8_COMMON_FILTER_H_ + +#include "vpx_ports/mem.h" + +#ifdef __cplusplus +extern "C" { +#endif + +#define BLOCK_HEIGHT_WIDTH 4 +#define VP8_FILTER_WEIGHT 128 +#define VP8_FILTER_SHIFT 7 + +extern DECLARE_ALIGNED(16, const short, vp8_bilinear_filters[8][2]); +extern DECLARE_ALIGNED(16, const short, vp8_sub_pel_filters[8][6]); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_FILTER_H_ diff --git a/thirdparty/libvpx/vp8/common/findnearmv.c b/thirdparty/libvpx/vp8/common/findnearmv.c new file mode 100644 index 0000000000..e8ee40f56c --- /dev/null +++ b/thirdparty/libvpx/vp8/common/findnearmv.c @@ -0,0 +1,193 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "findnearmv.h" + +const unsigned char vp8_mbsplit_offset[4][16] = { + { 0, 8, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}, + { 0, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}, + { 0, 2, 8, 10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}, + { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15} +}; + +/* Predict motion vectors using those from already-decoded nearby blocks. + Note that we only consider one 4x4 subblock from each candidate 16x16 + macroblock. */ +void vp8_find_near_mvs +( + MACROBLOCKD *xd, + const MODE_INFO *here, + int_mv *nearest, + int_mv *nearby, + int_mv *best_mv, + int cnt[4], + int refframe, + int *ref_frame_sign_bias +) +{ + const MODE_INFO *above = here - xd->mode_info_stride; + const MODE_INFO *left = here - 1; + const MODE_INFO *aboveleft = above - 1; + int_mv near_mvs[4]; + int_mv *mv = near_mvs; + int *cntx = cnt; + enum {CNT_INTRA, CNT_NEAREST, CNT_NEAR, CNT_SPLITMV}; + + /* Zero accumulators */ + mv[0].as_int = mv[1].as_int = mv[2].as_int = 0; + cnt[0] = cnt[1] = cnt[2] = cnt[3] = 0; + + /* Process above */ + if (above->mbmi.ref_frame != INTRA_FRAME) + { + if (above->mbmi.mv.as_int) + { + (++mv)->as_int = above->mbmi.mv.as_int; + mv_bias(ref_frame_sign_bias[above->mbmi.ref_frame], refframe, mv, ref_frame_sign_bias); + ++cntx; + } + + *cntx += 2; + } + + /* Process left */ + if (left->mbmi.ref_frame != INTRA_FRAME) + { + if (left->mbmi.mv.as_int) + { + int_mv this_mv; + + this_mv.as_int = left->mbmi.mv.as_int; + mv_bias(ref_frame_sign_bias[left->mbmi.ref_frame], refframe, &this_mv, ref_frame_sign_bias); + + if (this_mv.as_int != mv->as_int) + { + (++mv)->as_int = this_mv.as_int; + ++cntx; + } + + *cntx += 2; + } + else + cnt[CNT_INTRA] += 2; + } + + /* Process above left */ + if (aboveleft->mbmi.ref_frame != INTRA_FRAME) + { + if (aboveleft->mbmi.mv.as_int) + { + int_mv this_mv; + + this_mv.as_int = aboveleft->mbmi.mv.as_int; + mv_bias(ref_frame_sign_bias[aboveleft->mbmi.ref_frame], refframe, &this_mv, ref_frame_sign_bias); + + if (this_mv.as_int != mv->as_int) + { + (++mv)->as_int = this_mv.as_int; + ++cntx; + } + + *cntx += 1; + } + else + cnt[CNT_INTRA] += 1; + } + + /* If we have three distinct MV's ... */ + if (cnt[CNT_SPLITMV]) + { + /* See if above-left MV can be merged with NEAREST */ + if (mv->as_int == near_mvs[CNT_NEAREST].as_int) + cnt[CNT_NEAREST] += 1; + } + + cnt[CNT_SPLITMV] = ((above->mbmi.mode == SPLITMV) + + (left->mbmi.mode == SPLITMV)) * 2 + + (aboveleft->mbmi.mode == SPLITMV); + + /* Swap near and nearest if necessary */ + if (cnt[CNT_NEAR] > cnt[CNT_NEAREST]) + { + int tmp; + tmp = cnt[CNT_NEAREST]; + cnt[CNT_NEAREST] = cnt[CNT_NEAR]; + cnt[CNT_NEAR] = tmp; + tmp = near_mvs[CNT_NEAREST].as_int; + near_mvs[CNT_NEAREST].as_int = near_mvs[CNT_NEAR].as_int; + near_mvs[CNT_NEAR].as_int = tmp; + } + + /* Use near_mvs[0] to store the "best" MV */ + if (cnt[CNT_NEAREST] >= cnt[CNT_INTRA]) + near_mvs[CNT_INTRA] = near_mvs[CNT_NEAREST]; + + /* Set up return values */ + best_mv->as_int = near_mvs[0].as_int; + nearest->as_int = near_mvs[CNT_NEAREST].as_int; + nearby->as_int = near_mvs[CNT_NEAR].as_int; +} + + +static void invert_and_clamp_mvs(int_mv *inv, int_mv *src, MACROBLOCKD *xd) +{ + inv->as_mv.row = src->as_mv.row * -1; + inv->as_mv.col = src->as_mv.col * -1; + vp8_clamp_mv2(inv, xd); + vp8_clamp_mv2(src, xd); +} + + +int vp8_find_near_mvs_bias +( + MACROBLOCKD *xd, + const MODE_INFO *here, + int_mv mode_mv_sb[2][MB_MODE_COUNT], + int_mv best_mv_sb[2], + int cnt[4], + int refframe, + int *ref_frame_sign_bias +) +{ + int sign_bias = ref_frame_sign_bias[refframe]; + + vp8_find_near_mvs(xd, + here, + &mode_mv_sb[sign_bias][NEARESTMV], + &mode_mv_sb[sign_bias][NEARMV], + &best_mv_sb[sign_bias], + cnt, + refframe, + ref_frame_sign_bias); + + invert_and_clamp_mvs(&mode_mv_sb[!sign_bias][NEARESTMV], + &mode_mv_sb[sign_bias][NEARESTMV], xd); + invert_and_clamp_mvs(&mode_mv_sb[!sign_bias][NEARMV], + &mode_mv_sb[sign_bias][NEARMV], xd); + invert_and_clamp_mvs(&best_mv_sb[!sign_bias], + &best_mv_sb[sign_bias], xd); + + return sign_bias; +} + + +vp8_prob *vp8_mv_ref_probs( + vp8_prob p[VP8_MVREFS-1], const int near_mv_ref_ct[4] +) +{ + p[0] = vp8_mode_contexts [near_mv_ref_ct[0]] [0]; + p[1] = vp8_mode_contexts [near_mv_ref_ct[1]] [1]; + p[2] = vp8_mode_contexts [near_mv_ref_ct[2]] [2]; + p[3] = vp8_mode_contexts [near_mv_ref_ct[3]] [3]; + /*p[3] = vp8_mode_contexts [near_mv_ref_ct[1] + near_mv_ref_ct[2] + near_mv_ref_ct[3]] [3];*/ + return p; +} + diff --git a/thirdparty/libvpx/vp8/common/findnearmv.h b/thirdparty/libvpx/vp8/common/findnearmv.h new file mode 100644 index 0000000000..472a7b5d8d --- /dev/null +++ b/thirdparty/libvpx/vp8/common/findnearmv.h @@ -0,0 +1,195 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_FINDNEARMV_H_ +#define VP8_COMMON_FINDNEARMV_H_ + +#include "./vpx_config.h" +#include "mv.h" +#include "blockd.h" +#include "modecont.h" +#include "treecoder.h" + +#ifdef __cplusplus +extern "C" { +#endif + + +static INLINE void mv_bias(int refmb_ref_frame_sign_bias, int refframe, + int_mv *mvp, const int *ref_frame_sign_bias) +{ + if (refmb_ref_frame_sign_bias != ref_frame_sign_bias[refframe]) + { + mvp->as_mv.row *= -1; + mvp->as_mv.col *= -1; + } +} + +#define LEFT_TOP_MARGIN (16 << 3) +#define RIGHT_BOTTOM_MARGIN (16 << 3) +static INLINE void vp8_clamp_mv2(int_mv *mv, const MACROBLOCKD *xd) +{ + if (mv->as_mv.col < (xd->mb_to_left_edge - LEFT_TOP_MARGIN)) + mv->as_mv.col = xd->mb_to_left_edge - LEFT_TOP_MARGIN; + else if (mv->as_mv.col > xd->mb_to_right_edge + RIGHT_BOTTOM_MARGIN) + mv->as_mv.col = xd->mb_to_right_edge + RIGHT_BOTTOM_MARGIN; + + if (mv->as_mv.row < (xd->mb_to_top_edge - LEFT_TOP_MARGIN)) + mv->as_mv.row = xd->mb_to_top_edge - LEFT_TOP_MARGIN; + else if (mv->as_mv.row > xd->mb_to_bottom_edge + RIGHT_BOTTOM_MARGIN) + mv->as_mv.row = xd->mb_to_bottom_edge + RIGHT_BOTTOM_MARGIN; +} + +static INLINE void vp8_clamp_mv(int_mv *mv, int mb_to_left_edge, + int mb_to_right_edge, int mb_to_top_edge, + int mb_to_bottom_edge) +{ + mv->as_mv.col = (mv->as_mv.col < mb_to_left_edge) ? + mb_to_left_edge : mv->as_mv.col; + mv->as_mv.col = (mv->as_mv.col > mb_to_right_edge) ? + mb_to_right_edge : mv->as_mv.col; + mv->as_mv.row = (mv->as_mv.row < mb_to_top_edge) ? + mb_to_top_edge : mv->as_mv.row; + mv->as_mv.row = (mv->as_mv.row > mb_to_bottom_edge) ? + mb_to_bottom_edge : mv->as_mv.row; +} +static INLINE unsigned int vp8_check_mv_bounds(int_mv *mv, int mb_to_left_edge, + int mb_to_right_edge, + int mb_to_top_edge, + int mb_to_bottom_edge) +{ + unsigned int need_to_clamp; + need_to_clamp = (mv->as_mv.col < mb_to_left_edge); + need_to_clamp |= (mv->as_mv.col > mb_to_right_edge); + need_to_clamp |= (mv->as_mv.row < mb_to_top_edge); + need_to_clamp |= (mv->as_mv.row > mb_to_bottom_edge); + return need_to_clamp; +} + +void vp8_find_near_mvs +( + MACROBLOCKD *xd, + const MODE_INFO *here, + int_mv *nearest, int_mv *nearby, int_mv *best, + int near_mv_ref_cts[4], + int refframe, + int *ref_frame_sign_bias +); + + +int vp8_find_near_mvs_bias +( + MACROBLOCKD *xd, + const MODE_INFO *here, + int_mv mode_mv_sb[2][MB_MODE_COUNT], + int_mv best_mv_sb[2], + int cnt[4], + int refframe, + int *ref_frame_sign_bias +); + + +vp8_prob *vp8_mv_ref_probs( + vp8_prob p[VP8_MVREFS-1], const int near_mv_ref_ct[4] +); + +extern const unsigned char vp8_mbsplit_offset[4][16]; + + +static INLINE uint32_t left_block_mv(const MODE_INFO *cur_mb, int b) +{ + if (!(b & 3)) + { + /* On L edge, get from MB to left of us */ + --cur_mb; + + if(cur_mb->mbmi.mode != SPLITMV) + return cur_mb->mbmi.mv.as_int; + b += 4; + } + + return (cur_mb->bmi + b - 1)->mv.as_int; +} + +static INLINE uint32_t above_block_mv(const MODE_INFO *cur_mb, int b, + int mi_stride) +{ + if (!(b >> 2)) + { + /* On top edge, get from MB above us */ + cur_mb -= mi_stride; + + if(cur_mb->mbmi.mode != SPLITMV) + return cur_mb->mbmi.mv.as_int; + b += 16; + } + + return (cur_mb->bmi + (b - 4))->mv.as_int; +} +static INLINE B_PREDICTION_MODE left_block_mode(const MODE_INFO *cur_mb, int b) +{ + if (!(b & 3)) + { + /* On L edge, get from MB to left of us */ + --cur_mb; + switch (cur_mb->mbmi.mode) + { + case B_PRED: + return (cur_mb->bmi + b + 3)->as_mode; + case DC_PRED: + return B_DC_PRED; + case V_PRED: + return B_VE_PRED; + case H_PRED: + return B_HE_PRED; + case TM_PRED: + return B_TM_PRED; + default: + return B_DC_PRED; + } + } + + return (cur_mb->bmi + b - 1)->as_mode; +} + +static INLINE B_PREDICTION_MODE above_block_mode(const MODE_INFO *cur_mb, int b, + int mi_stride) +{ + if (!(b >> 2)) + { + /* On top edge, get from MB above us */ + cur_mb -= mi_stride; + + switch (cur_mb->mbmi.mode) + { + case B_PRED: + return (cur_mb->bmi + b + 12)->as_mode; + case DC_PRED: + return B_DC_PRED; + case V_PRED: + return B_VE_PRED; + case H_PRED: + return B_HE_PRED; + case TM_PRED: + return B_TM_PRED; + default: + return B_DC_PRED; + } + } + + return (cur_mb->bmi + b - 4)->as_mode; +} + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_FINDNEARMV_H_ diff --git a/thirdparty/libvpx/vp8/common/generic/systemdependent.c b/thirdparty/libvpx/vp8/common/generic/systemdependent.c new file mode 100644 index 0000000000..6d5f302d7a --- /dev/null +++ b/thirdparty/libvpx/vp8/common/generic/systemdependent.c @@ -0,0 +1,106 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "vpx_config.h" +#include "vp8_rtcd.h" +#if ARCH_ARM +#include "vpx_ports/arm.h" +#elif ARCH_X86 || ARCH_X86_64 +#include "vpx_ports/x86.h" +#endif +#include "vp8/common/onyxc_int.h" +#include "vp8/common/systemdependent.h" + +#if CONFIG_MULTITHREAD +#if HAVE_UNISTD_H && !defined(__OS2__) +#include <unistd.h> +#elif defined(_WIN32) +#include <windows.h> +typedef void (WINAPI *PGNSI)(LPSYSTEM_INFO); +#elif defined(__OS2__) +#define INCL_DOS +#define INCL_DOSSPINLOCK +#include <os2.h> +#endif +#endif + +#if CONFIG_MULTITHREAD +static int get_cpu_count() +{ + int core_count = 16; + +#if HAVE_UNISTD_H && !defined(__OS2__) +#if defined(_SC_NPROCESSORS_ONLN) + core_count = sysconf(_SC_NPROCESSORS_ONLN); +#elif defined(_SC_NPROC_ONLN) + core_count = sysconf(_SC_NPROC_ONLN); +#endif +#elif defined(_WIN32) + { +#if _WIN32_WINNT >= 0x0501 + SYSTEM_INFO sysinfo; + GetNativeSystemInfo(&sysinfo); +#else + PGNSI pGNSI; + SYSTEM_INFO sysinfo; + + /* Call GetNativeSystemInfo if supported or + * GetSystemInfo otherwise. */ + + pGNSI = (PGNSI) GetProcAddress( + GetModuleHandle(TEXT("kernel32.dll")), "GetNativeSystemInfo"); + if (pGNSI != NULL) + pGNSI(&sysinfo); + else + GetSystemInfo(&sysinfo); +#endif + + core_count = sysinfo.dwNumberOfProcessors; + } +#elif defined(__OS2__) + { + ULONG proc_id; + ULONG status; + + core_count = 0; + for (proc_id = 1; ; proc_id++) + { + if (DosGetProcessorStatus(proc_id, &status)) + break; + + if (status == PROC_ONLINE) + core_count++; + } + } +#else + /* other platforms */ +#endif + + return core_count > 0 ? core_count : 1; +} +#endif + +void vp8_clear_system_state_c() {}; + +void vp8_machine_specific_config(VP8_COMMON *ctx) +{ +#if CONFIG_MULTITHREAD + ctx->processor_core_count = get_cpu_count(); +#else + (void)ctx; +#endif /* CONFIG_MULTITHREAD */ + +#if ARCH_ARM + ctx->cpu_caps = arm_cpu_caps(); +#elif ARCH_X86 || ARCH_X86_64 + ctx->cpu_caps = x86_simd_caps(); +#endif +} diff --git a/thirdparty/libvpx/vp8/common/header.h b/thirdparty/libvpx/vp8/common/header.h new file mode 100644 index 0000000000..e27bca16bd --- /dev/null +++ b/thirdparty/libvpx/vp8/common/header.h @@ -0,0 +1,51 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_HEADER_H_ +#define VP8_COMMON_HEADER_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +/* 24 bits total */ +typedef struct +{ + unsigned int type: 1; + unsigned int version: 3; + unsigned int show_frame: 1; + + /* Allow 2^20 bytes = 8 megabits for first partition */ + + unsigned int first_partition_length_in_bytes: 19; + +#ifdef PACKET_TESTING + unsigned int frame_number; + unsigned int update_gold: 1; + unsigned int uses_gold: 1; + unsigned int update_last: 1; + unsigned int uses_last: 1; +#endif + +} VP8_HEADER; + +#ifdef PACKET_TESTING +#define VP8_HEADER_SIZE 8 +#else +#define VP8_HEADER_SIZE 3 +#endif + + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_HEADER_H_ diff --git a/thirdparty/libvpx/vp8/common/idct_blk.c b/thirdparty/libvpx/vp8/common/idct_blk.c new file mode 100644 index 0000000000..8aa7d9bf0f --- /dev/null +++ b/thirdparty/libvpx/vp8/common/idct_blk.c @@ -0,0 +1,90 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "vpx_config.h" +#include "vp8_rtcd.h" +#include "vpx_mem/vpx_mem.h" + +void vp8_dequant_idct_add_c(short *input, short *dq, + unsigned char *dest, int stride); +void vp8_dc_only_idct_add_c(short input_dc, unsigned char * pred, + int pred_stride, unsigned char *dst_ptr, + int dst_stride); + +void vp8_dequant_idct_add_y_block_c + (short *q, short *dq, + unsigned char *dst, int stride, char *eobs) +{ + int i, j; + + for (i = 0; i < 4; i++) + { + for (j = 0; j < 4; j++) + { + if (*eobs++ > 1) + vp8_dequant_idct_add_c (q, dq, dst, stride); + else + { + vp8_dc_only_idct_add_c (q[0]*dq[0], dst, stride, dst, stride); + memset(q, 0, 2 * sizeof(q[0])); + } + + q += 16; + dst += 4; + } + + dst += 4*stride - 16; + } +} + +void vp8_dequant_idct_add_uv_block_c + (short *q, short *dq, + unsigned char *dstu, unsigned char *dstv, int stride, char *eobs) +{ + int i, j; + + for (i = 0; i < 2; i++) + { + for (j = 0; j < 2; j++) + { + if (*eobs++ > 1) + vp8_dequant_idct_add_c (q, dq, dstu, stride); + else + { + vp8_dc_only_idct_add_c (q[0]*dq[0], dstu, stride, dstu, stride); + memset(q, 0, 2 * sizeof(q[0])); + } + + q += 16; + dstu += 4; + } + + dstu += 4*stride - 8; + } + + for (i = 0; i < 2; i++) + { + for (j = 0; j < 2; j++) + { + if (*eobs++ > 1) + vp8_dequant_idct_add_c (q, dq, dstv, stride); + else + { + vp8_dc_only_idct_add_c (q[0]*dq[0], dstv, stride, dstv, stride); + memset(q, 0, 2 * sizeof(q[0])); + } + + q += 16; + dstv += 4; + } + + dstv += 4*stride - 8; + } +} diff --git a/thirdparty/libvpx/vp8/common/idctllm.c b/thirdparty/libvpx/vp8/common/idctllm.c new file mode 100644 index 0000000000..f5403c5aaf --- /dev/null +++ b/thirdparty/libvpx/vp8/common/idctllm.c @@ -0,0 +1,205 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "./vp8_rtcd.h" + +/**************************************************************************** + * Notes: + * + * This implementation makes use of 16 bit fixed point verio of two multiply + * constants: + * 1. sqrt(2) * cos (pi/8) + * 2. sqrt(2) * sin (pi/8) + * Becuase the first constant is bigger than 1, to maintain the same 16 bit + * fixed point precision as the second one, we use a trick of + * x * a = x + x*(a-1) + * so + * x * sqrt(2) * cos (pi/8) = x + x * (sqrt(2) *cos(pi/8)-1). + **************************************************************************/ +static const int cospi8sqrt2minus1 = 20091; +static const int sinpi8sqrt2 = 35468; + +void vp8_short_idct4x4llm_c(short *input, unsigned char *pred_ptr, + int pred_stride, unsigned char *dst_ptr, + int dst_stride) +{ + int i; + int r, c; + int a1, b1, c1, d1; + short output[16]; + short *ip = input; + short *op = output; + int temp1, temp2; + int shortpitch = 4; + + for (i = 0; i < 4; i++) + { + a1 = ip[0] + ip[8]; + b1 = ip[0] - ip[8]; + + temp1 = (ip[4] * sinpi8sqrt2) >> 16; + temp2 = ip[12] + ((ip[12] * cospi8sqrt2minus1) >> 16); + c1 = temp1 - temp2; + + temp1 = ip[4] + ((ip[4] * cospi8sqrt2minus1) >> 16); + temp2 = (ip[12] * sinpi8sqrt2) >> 16; + d1 = temp1 + temp2; + + op[shortpitch*0] = a1 + d1; + op[shortpitch*3] = a1 - d1; + + op[shortpitch*1] = b1 + c1; + op[shortpitch*2] = b1 - c1; + + ip++; + op++; + } + + ip = output; + op = output; + + for (i = 0; i < 4; i++) + { + a1 = ip[0] + ip[2]; + b1 = ip[0] - ip[2]; + + temp1 = (ip[1] * sinpi8sqrt2) >> 16; + temp2 = ip[3] + ((ip[3] * cospi8sqrt2minus1) >> 16); + c1 = temp1 - temp2; + + temp1 = ip[1] + ((ip[1] * cospi8sqrt2minus1) >> 16); + temp2 = (ip[3] * sinpi8sqrt2) >> 16; + d1 = temp1 + temp2; + + + op[0] = (a1 + d1 + 4) >> 3; + op[3] = (a1 - d1 + 4) >> 3; + + op[1] = (b1 + c1 + 4) >> 3; + op[2] = (b1 - c1 + 4) >> 3; + + ip += shortpitch; + op += shortpitch; + } + + ip = output; + for (r = 0; r < 4; r++) + { + for (c = 0; c < 4; c++) + { + int a = ip[c] + pred_ptr[c] ; + + if (a < 0) + a = 0; + + if (a > 255) + a = 255; + + dst_ptr[c] = (unsigned char) a ; + } + ip += 4; + dst_ptr += dst_stride; + pred_ptr += pred_stride; + } +} + +void vp8_dc_only_idct_add_c(short input_dc, unsigned char *pred_ptr, + int pred_stride, unsigned char *dst_ptr, + int dst_stride) +{ + int a1 = ((input_dc + 4) >> 3); + int r, c; + + for (r = 0; r < 4; r++) + { + for (c = 0; c < 4; c++) + { + int a = a1 + pred_ptr[c] ; + + if (a < 0) + a = 0; + + if (a > 255) + a = 255; + + dst_ptr[c] = (unsigned char) a ; + } + + dst_ptr += dst_stride; + pred_ptr += pred_stride; + } + +} + +void vp8_short_inv_walsh4x4_c(short *input, short *mb_dqcoeff) +{ + short output[16]; + int i; + int a1, b1, c1, d1; + int a2, b2, c2, d2; + short *ip = input; + short *op = output; + + for (i = 0; i < 4; i++) + { + a1 = ip[0] + ip[12]; + b1 = ip[4] + ip[8]; + c1 = ip[4] - ip[8]; + d1 = ip[0] - ip[12]; + + op[0] = a1 + b1; + op[4] = c1 + d1; + op[8] = a1 - b1; + op[12] = d1 - c1; + ip++; + op++; + } + + ip = output; + op = output; + + for (i = 0; i < 4; i++) + { + a1 = ip[0] + ip[3]; + b1 = ip[1] + ip[2]; + c1 = ip[1] - ip[2]; + d1 = ip[0] - ip[3]; + + a2 = a1 + b1; + b2 = c1 + d1; + c2 = a1 - b1; + d2 = d1 - c1; + + op[0] = (a2 + 3) >> 3; + op[1] = (b2 + 3) >> 3; + op[2] = (c2 + 3) >> 3; + op[3] = (d2 + 3) >> 3; + + ip += 4; + op += 4; + } + + for(i = 0; i < 16; i++) + { + mb_dqcoeff[i * 16] = output[i]; + } +} + +void vp8_short_inv_walsh4x4_1_c(short *input, short *mb_dqcoeff) +{ + int i; + int a1; + + a1 = ((input[0] + 3) >> 3); + for(i = 0; i < 16; i++) + { + mb_dqcoeff[i * 16] = a1; + } +} diff --git a/thirdparty/libvpx/vp8/common/invtrans.h b/thirdparty/libvpx/vp8/common/invtrans.h new file mode 100644 index 0000000000..9cfea8d513 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/invtrans.h @@ -0,0 +1,70 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_INVTRANS_H_ +#define VP8_COMMON_INVTRANS_H_ + +#include "./vpx_config.h" +#include "vp8_rtcd.h" +#include "blockd.h" +#include "onyxc_int.h" + +#if CONFIG_MULTITHREAD +#include "vpx_mem/vpx_mem.h" +#endif + +#ifdef __cplusplus +extern "C" { +#endif + +static void eob_adjust(char *eobs, short *diff) +{ + /* eob adjust.... the idct can only skip if both the dc and eob are zero */ + int js; + for(js = 0; js < 16; js++) + { + if((eobs[js] == 0) && (diff[0] != 0)) + eobs[js]++; + diff+=16; + } +} + +static INLINE void vp8_inverse_transform_mby(MACROBLOCKD *xd) +{ + short *DQC = xd->dequant_y1; + + if (xd->mode_info_context->mbmi.mode != SPLITMV) + { + /* do 2nd order transform on the dc block */ + if (xd->eobs[24] > 1) + { + vp8_short_inv_walsh4x4 + (&xd->block[24].dqcoeff[0], xd->qcoeff); + } + else + { + vp8_short_inv_walsh4x4_1 + (&xd->block[24].dqcoeff[0], xd->qcoeff); + } + eob_adjust(xd->eobs, xd->qcoeff); + + DQC = xd->dequant_y1_dc; + } + vp8_dequant_idct_add_y_block + (xd->qcoeff, DQC, + xd->dst.y_buffer, + xd->dst.y_stride, xd->eobs); +} +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_INVTRANS_H_ diff --git a/thirdparty/libvpx/vp8/common/loopfilter.h b/thirdparty/libvpx/vp8/common/loopfilter.h new file mode 100644 index 0000000000..20a6bd375b --- /dev/null +++ b/thirdparty/libvpx/vp8/common/loopfilter.h @@ -0,0 +1,113 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_LOOPFILTER_H_ +#define VP8_COMMON_LOOPFILTER_H_ + +#include "vpx_ports/mem.h" +#include "vpx_config.h" +#include "vp8_rtcd.h" + +#ifdef __cplusplus +extern "C" { +#endif + +#define MAX_LOOP_FILTER 63 +/* fraction of total macroblock rows to be used in fast filter level picking */ +/* has to be > 2 */ +#define PARTIAL_FRAME_FRACTION 8 + +typedef enum +{ + NORMAL_LOOPFILTER = 0, + SIMPLE_LOOPFILTER = 1 +} LOOPFILTERTYPE; + +#if ARCH_ARM +#define SIMD_WIDTH 1 +#else +#define SIMD_WIDTH 16 +#endif + +/* Need to align this structure so when it is declared and + * passed it can be loaded into vector registers. + */ +typedef struct +{ + DECLARE_ALIGNED(SIMD_WIDTH, unsigned char, mblim[MAX_LOOP_FILTER + 1][SIMD_WIDTH]); + DECLARE_ALIGNED(SIMD_WIDTH, unsigned char, blim[MAX_LOOP_FILTER + 1][SIMD_WIDTH]); + DECLARE_ALIGNED(SIMD_WIDTH, unsigned char, lim[MAX_LOOP_FILTER + 1][SIMD_WIDTH]); + DECLARE_ALIGNED(SIMD_WIDTH, unsigned char, hev_thr[4][SIMD_WIDTH]); + unsigned char lvl[4][4][4]; + unsigned char hev_thr_lut[2][MAX_LOOP_FILTER + 1]; + unsigned char mode_lf_lut[10]; +} loop_filter_info_n; + +typedef struct loop_filter_info +{ + const unsigned char * mblim; + const unsigned char * blim; + const unsigned char * lim; + const unsigned char * hev_thr; +} loop_filter_info; + + +typedef void loop_filter_uvfunction +( + unsigned char *u, /* source pointer */ + int p, /* pitch */ + const unsigned char *blimit, + const unsigned char *limit, + const unsigned char *thresh, + unsigned char *v +); + +/* assorted loopfilter functions which get used elsewhere */ +struct VP8Common; +struct macroblockd; +struct modeinfo; + +void vp8_loop_filter_init(struct VP8Common *cm); + +void vp8_loop_filter_frame_init(struct VP8Common *cm, + struct macroblockd *mbd, + int default_filt_lvl); + +void vp8_loop_filter_frame(struct VP8Common *cm, struct macroblockd *mbd, + int frame_type); + +void vp8_loop_filter_partial_frame(struct VP8Common *cm, + struct macroblockd *mbd, + int default_filt_lvl); + +void vp8_loop_filter_frame_yonly(struct VP8Common *cm, + struct macroblockd *mbd, + int default_filt_lvl); + +void vp8_loop_filter_update_sharpness(loop_filter_info_n *lfi, + int sharpness_lvl); + +void vp8_loop_filter_row_normal(struct VP8Common *cm, + struct modeinfo *mode_info_context, + int mb_row, int post_ystride, int post_uvstride, + unsigned char *y_ptr, unsigned char *u_ptr, + unsigned char *v_ptr); + +void vp8_loop_filter_row_simple(struct VP8Common *cm, + struct modeinfo *mode_info_context, + int mb_row, int post_ystride, int post_uvstride, + unsigned char *y_ptr, unsigned char *u_ptr, + unsigned char *v_ptr); +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_LOOPFILTER_H_ diff --git a/thirdparty/libvpx/vp8/common/loopfilter_filters.c b/thirdparty/libvpx/vp8/common/loopfilter_filters.c new file mode 100644 index 0000000000..1d51696ff7 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/loopfilter_filters.c @@ -0,0 +1,430 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include <stdlib.h> +#include "loopfilter.h" +#include "onyxc_int.h" + +typedef unsigned char uc; + +static signed char vp8_signed_char_clamp(int t) +{ + t = (t < -128 ? -128 : t); + t = (t > 127 ? 127 : t); + return (signed char) t; +} + + +/* should we apply any filter at all ( 11111111 yes, 00000000 no) */ +static signed char vp8_filter_mask(uc limit, uc blimit, + uc p3, uc p2, uc p1, uc p0, + uc q0, uc q1, uc q2, uc q3) +{ + signed char mask = 0; + mask |= (abs(p3 - p2) > limit); + mask |= (abs(p2 - p1) > limit); + mask |= (abs(p1 - p0) > limit); + mask |= (abs(q1 - q0) > limit); + mask |= (abs(q2 - q1) > limit); + mask |= (abs(q3 - q2) > limit); + mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2 > blimit); + return mask - 1; +} + +/* is there high variance internal edge ( 11111111 yes, 00000000 no) */ +static signed char vp8_hevmask(uc thresh, uc p1, uc p0, uc q0, uc q1) +{ + signed char hev = 0; + hev |= (abs(p1 - p0) > thresh) * -1; + hev |= (abs(q1 - q0) > thresh) * -1; + return hev; +} + +static void vp8_filter(signed char mask, uc hev, uc *op1, + uc *op0, uc *oq0, uc *oq1) + +{ + signed char ps0, qs0; + signed char ps1, qs1; + signed char filter_value, Filter1, Filter2; + signed char u; + + ps1 = (signed char) * op1 ^ 0x80; + ps0 = (signed char) * op0 ^ 0x80; + qs0 = (signed char) * oq0 ^ 0x80; + qs1 = (signed char) * oq1 ^ 0x80; + + /* add outer taps if we have high edge variance */ + filter_value = vp8_signed_char_clamp(ps1 - qs1); + filter_value &= hev; + + /* inner taps */ + filter_value = vp8_signed_char_clamp(filter_value + 3 * (qs0 - ps0)); + filter_value &= mask; + + /* save bottom 3 bits so that we round one side +4 and the other +3 + * if it equals 4 we'll set to adjust by -1 to account for the fact + * we'd round 3 the other way + */ + Filter1 = vp8_signed_char_clamp(filter_value + 4); + Filter2 = vp8_signed_char_clamp(filter_value + 3); + Filter1 >>= 3; + Filter2 >>= 3; + u = vp8_signed_char_clamp(qs0 - Filter1); + *oq0 = u ^ 0x80; + u = vp8_signed_char_clamp(ps0 + Filter2); + *op0 = u ^ 0x80; + filter_value = Filter1; + + /* outer tap adjustments */ + filter_value += 1; + filter_value >>= 1; + filter_value &= ~hev; + + u = vp8_signed_char_clamp(qs1 - filter_value); + *oq1 = u ^ 0x80; + u = vp8_signed_char_clamp(ps1 + filter_value); + *op1 = u ^ 0x80; + +} +void vp8_loop_filter_horizontal_edge_c +( + unsigned char *s, + int p, /* pitch */ + const unsigned char *blimit, + const unsigned char *limit, + const unsigned char *thresh, + int count +) +{ + int hev = 0; /* high edge variance */ + signed char mask = 0; + int i = 0; + + /* loop filter designed to work using chars so that we can make maximum use + * of 8 bit simd instructions. + */ + do + { + mask = vp8_filter_mask(limit[0], blimit[0], + s[-4*p], s[-3*p], s[-2*p], s[-1*p], + s[0*p], s[1*p], s[2*p], s[3*p]); + + hev = vp8_hevmask(thresh[0], s[-2*p], s[-1*p], s[0*p], s[1*p]); + + vp8_filter(mask, hev, s - 2 * p, s - 1 * p, s, s + 1 * p); + + ++s; + } + while (++i < count * 8); +} + +void vp8_loop_filter_vertical_edge_c +( + unsigned char *s, + int p, + const unsigned char *blimit, + const unsigned char *limit, + const unsigned char *thresh, + int count +) +{ + int hev = 0; /* high edge variance */ + signed char mask = 0; + int i = 0; + + /* loop filter designed to work using chars so that we can make maximum use + * of 8 bit simd instructions. + */ + do + { + mask = vp8_filter_mask(limit[0], blimit[0], + s[-4], s[-3], s[-2], s[-1], s[0], s[1], s[2], s[3]); + + hev = vp8_hevmask(thresh[0], s[-2], s[-1], s[0], s[1]); + + vp8_filter(mask, hev, s - 2, s - 1, s, s + 1); + + s += p; + } + while (++i < count * 8); +} + +static void vp8_mbfilter(signed char mask, uc hev, + uc *op2, uc *op1, uc *op0, uc *oq0, uc *oq1, uc *oq2) +{ + signed char s, u; + signed char filter_value, Filter1, Filter2; + signed char ps2 = (signed char) * op2 ^ 0x80; + signed char ps1 = (signed char) * op1 ^ 0x80; + signed char ps0 = (signed char) * op0 ^ 0x80; + signed char qs0 = (signed char) * oq0 ^ 0x80; + signed char qs1 = (signed char) * oq1 ^ 0x80; + signed char qs2 = (signed char) * oq2 ^ 0x80; + + /* add outer taps if we have high edge variance */ + filter_value = vp8_signed_char_clamp(ps1 - qs1); + filter_value = vp8_signed_char_clamp(filter_value + 3 * (qs0 - ps0)); + filter_value &= mask; + + Filter2 = filter_value; + Filter2 &= hev; + + /* save bottom 3 bits so that we round one side +4 and the other +3 */ + Filter1 = vp8_signed_char_clamp(Filter2 + 4); + Filter2 = vp8_signed_char_clamp(Filter2 + 3); + Filter1 >>= 3; + Filter2 >>= 3; + qs0 = vp8_signed_char_clamp(qs0 - Filter1); + ps0 = vp8_signed_char_clamp(ps0 + Filter2); + + + /* only apply wider filter if not high edge variance */ + filter_value &= ~hev; + Filter2 = filter_value; + + /* roughly 3/7th difference across boundary */ + u = vp8_signed_char_clamp((63 + Filter2 * 27) >> 7); + s = vp8_signed_char_clamp(qs0 - u); + *oq0 = s ^ 0x80; + s = vp8_signed_char_clamp(ps0 + u); + *op0 = s ^ 0x80; + + /* roughly 2/7th difference across boundary */ + u = vp8_signed_char_clamp((63 + Filter2 * 18) >> 7); + s = vp8_signed_char_clamp(qs1 - u); + *oq1 = s ^ 0x80; + s = vp8_signed_char_clamp(ps1 + u); + *op1 = s ^ 0x80; + + /* roughly 1/7th difference across boundary */ + u = vp8_signed_char_clamp((63 + Filter2 * 9) >> 7); + s = vp8_signed_char_clamp(qs2 - u); + *oq2 = s ^ 0x80; + s = vp8_signed_char_clamp(ps2 + u); + *op2 = s ^ 0x80; +} + +void vp8_mbloop_filter_horizontal_edge_c +( + unsigned char *s, + int p, + const unsigned char *blimit, + const unsigned char *limit, + const unsigned char *thresh, + int count +) +{ + signed char hev = 0; /* high edge variance */ + signed char mask = 0; + int i = 0; + + /* loop filter designed to work using chars so that we can make maximum use + * of 8 bit simd instructions. + */ + do + { + + mask = vp8_filter_mask(limit[0], blimit[0], + s[-4*p], s[-3*p], s[-2*p], s[-1*p], + s[0*p], s[1*p], s[2*p], s[3*p]); + + hev = vp8_hevmask(thresh[0], s[-2*p], s[-1*p], s[0*p], s[1*p]); + + vp8_mbfilter(mask, hev, s - 3 * p, s - 2 * p, s - 1 * p, s, s + 1 * p, s + 2 * p); + + ++s; + } + while (++i < count * 8); + +} + + +void vp8_mbloop_filter_vertical_edge_c +( + unsigned char *s, + int p, + const unsigned char *blimit, + const unsigned char *limit, + const unsigned char *thresh, + int count +) +{ + signed char hev = 0; /* high edge variance */ + signed char mask = 0; + int i = 0; + + do + { + + mask = vp8_filter_mask(limit[0], blimit[0], + s[-4], s[-3], s[-2], s[-1], s[0], s[1], s[2], s[3]); + + hev = vp8_hevmask(thresh[0], s[-2], s[-1], s[0], s[1]); + + vp8_mbfilter(mask, hev, s - 3, s - 2, s - 1, s, s + 1, s + 2); + + s += p; + } + while (++i < count * 8); + +} + +/* should we apply any filter at all ( 11111111 yes, 00000000 no) */ +static signed char vp8_simple_filter_mask(uc blimit, uc p1, uc p0, uc q0, uc q1) +{ +/* Why does this cause problems for win32? + * error C2143: syntax error : missing ';' before 'type' + * (void) limit; + */ + signed char mask = (abs(p0 - q0) * 2 + abs(p1 - q1) / 2 <= blimit) * -1; + return mask; +} + +static void vp8_simple_filter(signed char mask, uc *op1, uc *op0, uc *oq0, uc *oq1) +{ + signed char filter_value, Filter1, Filter2; + signed char p1 = (signed char) * op1 ^ 0x80; + signed char p0 = (signed char) * op0 ^ 0x80; + signed char q0 = (signed char) * oq0 ^ 0x80; + signed char q1 = (signed char) * oq1 ^ 0x80; + signed char u; + + filter_value = vp8_signed_char_clamp(p1 - q1); + filter_value = vp8_signed_char_clamp(filter_value + 3 * (q0 - p0)); + filter_value &= mask; + + /* save bottom 3 bits so that we round one side +4 and the other +3 */ + Filter1 = vp8_signed_char_clamp(filter_value + 4); + Filter1 >>= 3; + u = vp8_signed_char_clamp(q0 - Filter1); + *oq0 = u ^ 0x80; + + Filter2 = vp8_signed_char_clamp(filter_value + 3); + Filter2 >>= 3; + u = vp8_signed_char_clamp(p0 + Filter2); + *op0 = u ^ 0x80; +} + +void vp8_loop_filter_simple_horizontal_edge_c +( + unsigned char *s, + int p, + const unsigned char *blimit +) +{ + signed char mask = 0; + int i = 0; + + do + { + mask = vp8_simple_filter_mask(blimit[0], s[-2*p], s[-1*p], s[0*p], s[1*p]); + vp8_simple_filter(mask, s - 2 * p, s - 1 * p, s, s + 1 * p); + ++s; + } + while (++i < 16); +} + +void vp8_loop_filter_simple_vertical_edge_c +( + unsigned char *s, + int p, + const unsigned char *blimit +) +{ + signed char mask = 0; + int i = 0; + + do + { + mask = vp8_simple_filter_mask(blimit[0], s[-2], s[-1], s[0], s[1]); + vp8_simple_filter(mask, s - 2, s - 1, s, s + 1); + s += p; + } + while (++i < 16); + +} + +/* Horizontal MB filtering */ +void vp8_loop_filter_mbh_c(unsigned char *y_ptr, unsigned char *u_ptr, + unsigned char *v_ptr, int y_stride, int uv_stride, + loop_filter_info *lfi) +{ + vp8_mbloop_filter_horizontal_edge_c(y_ptr, y_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 2); + + if (u_ptr) + vp8_mbloop_filter_horizontal_edge_c(u_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 1); + + if (v_ptr) + vp8_mbloop_filter_horizontal_edge_c(v_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 1); +} + +/* Vertical MB Filtering */ +void vp8_loop_filter_mbv_c(unsigned char *y_ptr, unsigned char *u_ptr, + unsigned char *v_ptr, int y_stride, int uv_stride, + loop_filter_info *lfi) +{ + vp8_mbloop_filter_vertical_edge_c(y_ptr, y_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 2); + + if (u_ptr) + vp8_mbloop_filter_vertical_edge_c(u_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 1); + + if (v_ptr) + vp8_mbloop_filter_vertical_edge_c(v_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 1); +} + +/* Horizontal B Filtering */ +void vp8_loop_filter_bh_c(unsigned char *y_ptr, unsigned char *u_ptr, + unsigned char *v_ptr, int y_stride, int uv_stride, + loop_filter_info *lfi) +{ + vp8_loop_filter_horizontal_edge_c(y_ptr + 4 * y_stride, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2); + vp8_loop_filter_horizontal_edge_c(y_ptr + 8 * y_stride, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2); + vp8_loop_filter_horizontal_edge_c(y_ptr + 12 * y_stride, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2); + + if (u_ptr) + vp8_loop_filter_horizontal_edge_c(u_ptr + 4 * uv_stride, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, 1); + + if (v_ptr) + vp8_loop_filter_horizontal_edge_c(v_ptr + 4 * uv_stride, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, 1); +} + +void vp8_loop_filter_bhs_c(unsigned char *y_ptr, int y_stride, + const unsigned char *blimit) +{ + vp8_loop_filter_simple_horizontal_edge_c(y_ptr + 4 * y_stride, y_stride, blimit); + vp8_loop_filter_simple_horizontal_edge_c(y_ptr + 8 * y_stride, y_stride, blimit); + vp8_loop_filter_simple_horizontal_edge_c(y_ptr + 12 * y_stride, y_stride, blimit); +} + +/* Vertical B Filtering */ +void vp8_loop_filter_bv_c(unsigned char *y_ptr, unsigned char *u_ptr, + unsigned char *v_ptr, int y_stride, int uv_stride, + loop_filter_info *lfi) +{ + vp8_loop_filter_vertical_edge_c(y_ptr + 4, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2); + vp8_loop_filter_vertical_edge_c(y_ptr + 8, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2); + vp8_loop_filter_vertical_edge_c(y_ptr + 12, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2); + + if (u_ptr) + vp8_loop_filter_vertical_edge_c(u_ptr + 4, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, 1); + + if (v_ptr) + vp8_loop_filter_vertical_edge_c(v_ptr + 4, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, 1); +} + +void vp8_loop_filter_bvs_c(unsigned char *y_ptr, int y_stride, + const unsigned char *blimit) +{ + vp8_loop_filter_simple_vertical_edge_c(y_ptr + 4, y_stride, blimit); + vp8_loop_filter_simple_vertical_edge_c(y_ptr + 8, y_stride, blimit); + vp8_loop_filter_simple_vertical_edge_c(y_ptr + 12, y_stride, blimit); +} diff --git a/thirdparty/libvpx/vp8/common/mbpitch.c b/thirdparty/libvpx/vp8/common/mbpitch.c new file mode 100644 index 0000000000..32e1b66409 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/mbpitch.c @@ -0,0 +1,68 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "blockd.h" + +void vp8_setup_block_dptrs(MACROBLOCKD *x) +{ + int r, c; + + for (r = 0; r < 4; r++) + { + for (c = 0; c < 4; c++) + { + x->block[r*4+c].predictor = x->predictor + r * 4 * 16 + c * 4; + } + } + + for (r = 0; r < 2; r++) + { + for (c = 0; c < 2; c++) + { + x->block[16+r*2+c].predictor = x->predictor + 256 + r * 4 * 8 + c * 4; + + } + } + + for (r = 0; r < 2; r++) + { + for (c = 0; c < 2; c++) + { + x->block[20+r*2+c].predictor = x->predictor + 320 + r * 4 * 8 + c * 4; + + } + } + + for (r = 0; r < 25; r++) + { + x->block[r].qcoeff = x->qcoeff + r * 16; + x->block[r].dqcoeff = x->dqcoeff + r * 16; + x->block[r].eob = x->eobs + r; + } +} + +void vp8_build_block_doffsets(MACROBLOCKD *x) +{ + int block; + + for (block = 0; block < 16; block++) /* y blocks */ + { + x->block[block].offset = + (block >> 2) * 4 * x->dst.y_stride + (block & 3) * 4; + } + + for (block = 16; block < 20; block++) /* U and V blocks */ + { + x->block[block+4].offset = + x->block[block].offset = + ((block - 16) >> 1) * 4 * x->dst.uv_stride + (block & 1) * 4; + } +} diff --git a/thirdparty/libvpx/vp8/common/modecont.c b/thirdparty/libvpx/vp8/common/modecont.c new file mode 100644 index 0000000000..86a74bc0ff --- /dev/null +++ b/thirdparty/libvpx/vp8/common/modecont.c @@ -0,0 +1,40 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "entropy.h" + +const int vp8_mode_contexts[6][4] = +{ + { + /* 0 */ + 7, 1, 1, 143, + }, + { + /* 1 */ + 14, 18, 14, 107, + }, + { + /* 2 */ + 135, 64, 57, 68, + }, + { + /* 3 */ + 60, 56, 128, 65, + }, + { + /* 4 */ + 159, 134, 128, 34, + }, + { + /* 5 */ + 234, 188, 128, 28, + }, +}; diff --git a/thirdparty/libvpx/vp8/common/modecont.h b/thirdparty/libvpx/vp8/common/modecont.h new file mode 100644 index 0000000000..ff34c33c55 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/modecont.h @@ -0,0 +1,25 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_MODECONT_H_ +#define VP8_COMMON_MODECONT_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +extern const int vp8_mode_contexts[6][4]; + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_MODECONT_H_ diff --git a/thirdparty/libvpx/vp8/common/mv.h b/thirdparty/libvpx/vp8/common/mv.h new file mode 100644 index 0000000000..111ccd63c7 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/mv.h @@ -0,0 +1,36 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_MV_H_ +#define VP8_COMMON_MV_H_ +#include "vpx/vpx_integer.h" + +#ifdef __cplusplus +extern "C" { +#endif + +typedef struct +{ + short row; + short col; +} MV; + +typedef union int_mv +{ + uint32_t as_int; + MV as_mv; +} int_mv; /* facilitates faster equality tests and copies */ + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_MV_H_ diff --git a/thirdparty/libvpx/vp8/common/onyxc_int.h b/thirdparty/libvpx/vp8/common/onyxc_int.h new file mode 100644 index 0000000000..6d89865c60 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/onyxc_int.h @@ -0,0 +1,185 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_ONYXC_INT_H_ +#define VP8_COMMON_ONYXC_INT_H_ + +#include "vpx_config.h" +#include "vp8_rtcd.h" +#include "vpx/internal/vpx_codec_internal.h" +#include "loopfilter.h" +#include "entropymv.h" +#include "entropy.h" +#if CONFIG_POSTPROC +#include "postproc.h" +#endif + +/*#ifdef PACKET_TESTING*/ +#include "header.h" +/*#endif*/ + +#ifdef __cplusplus +extern "C" { +#endif + +#define MINQ 0 +#define MAXQ 127 +#define QINDEX_RANGE (MAXQ + 1) + +#define NUM_YV12_BUFFERS 4 + +#define MAX_PARTITIONS 9 + +typedef struct frame_contexts +{ + vp8_prob bmode_prob [VP8_BINTRAMODES-1]; + vp8_prob ymode_prob [VP8_YMODES-1]; /* interframe intra mode probs */ + vp8_prob uv_mode_prob [VP8_UV_MODES-1]; + vp8_prob sub_mv_ref_prob [VP8_SUBMVREFS-1]; + vp8_prob coef_probs [BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [ENTROPY_NODES]; + MV_CONTEXT mvc[2]; +} FRAME_CONTEXT; + +typedef enum +{ + ONE_PARTITION = 0, + TWO_PARTITION = 1, + FOUR_PARTITION = 2, + EIGHT_PARTITION = 3 +} TOKEN_PARTITION; + +typedef enum +{ + RECON_CLAMP_REQUIRED = 0, + RECON_CLAMP_NOTREQUIRED = 1 +} CLAMP_TYPE; + +typedef struct VP8Common + +{ + struct vpx_internal_error_info error; + + DECLARE_ALIGNED(16, short, Y1dequant[QINDEX_RANGE][2]); + DECLARE_ALIGNED(16, short, Y2dequant[QINDEX_RANGE][2]); + DECLARE_ALIGNED(16, short, UVdequant[QINDEX_RANGE][2]); + + int Width; + int Height; + int horiz_scale; + int vert_scale; + + CLAMP_TYPE clamp_type; + + YV12_BUFFER_CONFIG *frame_to_show; + + YV12_BUFFER_CONFIG yv12_fb[NUM_YV12_BUFFERS]; + int fb_idx_ref_cnt[NUM_YV12_BUFFERS]; + int new_fb_idx, lst_fb_idx, gld_fb_idx, alt_fb_idx; + + YV12_BUFFER_CONFIG temp_scale_frame; + +#if CONFIG_POSTPROC + YV12_BUFFER_CONFIG post_proc_buffer; + YV12_BUFFER_CONFIG post_proc_buffer_int; + int post_proc_buffer_int_used; + unsigned char *pp_limits_buffer; /* post-processing filter coefficients */ +#endif + + FRAME_TYPE last_frame_type; /* Save last frame's frame type for motion search. */ + FRAME_TYPE frame_type; + + int show_frame; + + int frame_flags; + int MBs; + int mb_rows; + int mb_cols; + int mode_info_stride; + + /* profile settings */ + int mb_no_coeff_skip; + int no_lpf; + int use_bilinear_mc_filter; + int full_pixel; + + int base_qindex; + + int y1dc_delta_q; + int y2dc_delta_q; + int y2ac_delta_q; + int uvdc_delta_q; + int uvac_delta_q; + + /* We allocate a MODE_INFO struct for each macroblock, together with + an extra row on top and column on the left to simplify prediction. */ + + MODE_INFO *mip; /* Base of allocated array */ + MODE_INFO *mi; /* Corresponds to upper left visible macroblock */ +#if CONFIG_ERROR_CONCEALMENT + MODE_INFO *prev_mip; /* MODE_INFO array 'mip' from last decoded frame */ + MODE_INFO *prev_mi; /* 'mi' from last frame (points into prev_mip) */ +#endif + MODE_INFO *show_frame_mi; /* MODE_INFO for the last decoded frame + to show */ + LOOPFILTERTYPE filter_type; + + loop_filter_info_n lf_info; + + int filter_level; + int last_sharpness_level; + int sharpness_level; + + int refresh_last_frame; /* Two state 0 = NO, 1 = YES */ + int refresh_golden_frame; /* Two state 0 = NO, 1 = YES */ + int refresh_alt_ref_frame; /* Two state 0 = NO, 1 = YES */ + + int copy_buffer_to_gf; /* 0 none, 1 Last to GF, 2 ARF to GF */ + int copy_buffer_to_arf; /* 0 none, 1 Last to ARF, 2 GF to ARF */ + + int refresh_entropy_probs; /* Two state 0 = NO, 1 = YES */ + + int ref_frame_sign_bias[MAX_REF_FRAMES]; /* Two state 0, 1 */ + + /* Y,U,V,Y2 */ + ENTROPY_CONTEXT_PLANES *above_context; /* row of context for each plane */ + ENTROPY_CONTEXT_PLANES left_context; /* (up to) 4 contexts "" */ + + FRAME_CONTEXT lfc; /* last frame entropy */ + FRAME_CONTEXT fc; /* this frame entropy */ + + unsigned int current_video_frame; + + int version; + + TOKEN_PARTITION multi_token_partition; + +#ifdef PACKET_TESTING + VP8_HEADER oh; +#endif +#if CONFIG_POSTPROC_VISUALIZER + double bitrate; + double framerate; +#endif + +#if CONFIG_MULTITHREAD + int processor_core_count; +#endif +#if CONFIG_POSTPROC + struct postproc_state postproc_state; +#endif + int cpu_caps; +} VP8_COMMON; + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_ONYXC_INT_H_ diff --git a/thirdparty/libvpx/vp8/common/onyxd.h b/thirdparty/libvpx/vp8/common/onyxd.h new file mode 100644 index 0000000000..e37b29f32c --- /dev/null +++ b/thirdparty/libvpx/vp8/common/onyxd.h @@ -0,0 +1,63 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_ONYXD_H_ +#define VP8_COMMON_ONYXD_H_ + + +/* Create/destroy static data structures. */ +#ifdef __cplusplus +extern "C" +{ +#endif +#include "vpx_scale/yv12config.h" +#include "ppflags.h" +#include "vpx_ports/mem.h" +#include "vpx/vpx_codec.h" +#include "vpx/vp8.h" + + struct VP8D_COMP; + + typedef struct + { + int Width; + int Height; + int Version; + int postprocess; + int max_threads; + int error_concealment; + } VP8D_CONFIG; + + typedef enum + { + VP8D_OK = 0 + } VP8D_SETTING; + + void vp8dx_initialize(void); + + void vp8dx_set_setting(struct VP8D_COMP* comp, VP8D_SETTING oxst, int x); + + int vp8dx_get_setting(struct VP8D_COMP* comp, VP8D_SETTING oxst); + + int vp8dx_receive_compressed_data(struct VP8D_COMP* comp, + size_t size, const uint8_t *dest, + int64_t time_stamp); + int vp8dx_get_raw_frame(struct VP8D_COMP* comp, YV12_BUFFER_CONFIG *sd, int64_t *time_stamp, int64_t *time_end_stamp, vp8_ppflags_t *flags); + + vpx_codec_err_t vp8dx_get_reference(struct VP8D_COMP* comp, enum vpx_ref_frame_type ref_frame_flag, YV12_BUFFER_CONFIG *sd); + vpx_codec_err_t vp8dx_set_reference(struct VP8D_COMP* comp, enum vpx_ref_frame_type ref_frame_flag, YV12_BUFFER_CONFIG *sd); + +#ifdef __cplusplus +} +#endif + + +#endif // VP8_COMMON_ONYXD_H_ diff --git a/thirdparty/libvpx/vp8/common/ppflags.h b/thirdparty/libvpx/vp8/common/ppflags.h new file mode 100644 index 0000000000..768224aad5 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/ppflags.h @@ -0,0 +1,49 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_PPFLAGS_H_ +#define VP8_COMMON_PPFLAGS_H_ + +#ifdef __cplusplus +extern "C" { +#endif +enum +{ + VP8D_NOFILTERING = 0, + VP8D_DEBLOCK = 1<<0, + VP8D_DEMACROBLOCK = 1<<1, + VP8D_ADDNOISE = 1<<2, + VP8D_DEBUG_TXT_FRAME_INFO = 1<<3, + VP8D_DEBUG_TXT_MBLK_MODES = 1<<4, + VP8D_DEBUG_TXT_DC_DIFF = 1<<5, + VP8D_DEBUG_TXT_RATE_INFO = 1<<6, + VP8D_DEBUG_DRAW_MV = 1<<7, + VP8D_DEBUG_CLR_BLK_MODES = 1<<8, + VP8D_DEBUG_CLR_FRM_REF_BLKS = 1<<9, + VP8D_MFQE = 1<<10 +}; + +typedef struct +{ + int post_proc_flag; + int deblocking_level; + int noise_level; + int display_ref_frame_flag; + int display_mb_modes_flag; + int display_b_modes_flag; + int display_mv_flag; +} vp8_ppflags_t; + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_PPFLAGS_H_ diff --git a/thirdparty/libvpx/vp8/common/quant_common.c b/thirdparty/libvpx/vp8/common/quant_common.c new file mode 100644 index 0000000000..05f9210702 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/quant_common.c @@ -0,0 +1,135 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "quant_common.h" + +static const int dc_qlookup[QINDEX_RANGE] = +{ + 4, 5, 6, 7, 8, 9, 10, 10, 11, 12, 13, 14, 15, 16, 17, 17, + 18, 19, 20, 20, 21, 21, 22, 22, 23, 23, 24, 25, 25, 26, 27, 28, + 29, 30, 31, 32, 33, 34, 35, 36, 37, 37, 38, 39, 40, 41, 42, 43, + 44, 45, 46, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, + 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, + 75, 76, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, + 91, 93, 95, 96, 98, 100, 101, 102, 104, 106, 108, 110, 112, 114, 116, 118, + 122, 124, 126, 128, 130, 132, 134, 136, 138, 140, 143, 145, 148, 151, 154, 157, +}; + +static const int ac_qlookup[QINDEX_RANGE] = +{ + 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, + 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, + 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, + 52, 53, 54, 55, 56, 57, 58, 60, 62, 64, 66, 68, 70, 72, 74, 76, + 78, 80, 82, 84, 86, 88, 90, 92, 94, 96, 98, 100, 102, 104, 106, 108, + 110, 112, 114, 116, 119, 122, 125, 128, 131, 134, 137, 140, 143, 146, 149, 152, + 155, 158, 161, 164, 167, 170, 173, 177, 181, 185, 189, 193, 197, 201, 205, 209, + 213, 217, 221, 225, 229, 234, 239, 245, 249, 254, 259, 264, 269, 274, 279, 284, +}; + + +int vp8_dc_quant(int QIndex, int Delta) +{ + int retval; + + QIndex = QIndex + Delta; + + if (QIndex > 127) + QIndex = 127; + else if (QIndex < 0) + QIndex = 0; + + retval = dc_qlookup[ QIndex ]; + return retval; +} + +int vp8_dc2quant(int QIndex, int Delta) +{ + int retval; + + QIndex = QIndex + Delta; + + if (QIndex > 127) + QIndex = 127; + else if (QIndex < 0) + QIndex = 0; + + retval = dc_qlookup[ QIndex ] * 2; + return retval; + +} +int vp8_dc_uv_quant(int QIndex, int Delta) +{ + int retval; + + QIndex = QIndex + Delta; + + if (QIndex > 127) + QIndex = 127; + else if (QIndex < 0) + QIndex = 0; + + retval = dc_qlookup[ QIndex ]; + + if (retval > 132) + retval = 132; + + return retval; +} + +int vp8_ac_yquant(int QIndex) +{ + int retval; + + if (QIndex > 127) + QIndex = 127; + else if (QIndex < 0) + QIndex = 0; + + retval = ac_qlookup[ QIndex ]; + return retval; +} + +int vp8_ac2quant(int QIndex, int Delta) +{ + int retval; + + QIndex = QIndex + Delta; + + if (QIndex > 127) + QIndex = 127; + else if (QIndex < 0) + QIndex = 0; + + /* For all x in [0..284], x*155/100 is bitwise equal to (x*101581) >> 16. + * The smallest precision for that is '(x*6349) >> 12' but 16 is a good + * word size. */ + retval = (ac_qlookup[ QIndex ] * 101581) >> 16; + + if (retval < 8) + retval = 8; + + return retval; +} +int vp8_ac_uv_quant(int QIndex, int Delta) +{ + int retval; + + QIndex = QIndex + Delta; + + if (QIndex > 127) + QIndex = 127; + else if (QIndex < 0) + QIndex = 0; + + retval = ac_qlookup[ QIndex ]; + return retval; +} diff --git a/thirdparty/libvpx/vp8/common/quant_common.h b/thirdparty/libvpx/vp8/common/quant_common.h new file mode 100644 index 0000000000..700b5e6d72 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/quant_common.h @@ -0,0 +1,34 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP8_COMMON_QUANT_COMMON_H_ +#define VP8_COMMON_QUANT_COMMON_H_ + + +#include "string.h" +#include "blockd.h" +#include "onyxc_int.h" + +#ifdef __cplusplus +extern "C" { +#endif + +extern int vp8_ac_yquant(int QIndex); +extern int vp8_dc_quant(int QIndex, int Delta); +extern int vp8_dc2quant(int QIndex, int Delta); +extern int vp8_ac2quant(int QIndex, int Delta); +extern int vp8_dc_uv_quant(int QIndex, int Delta); +extern int vp8_ac_uv_quant(int QIndex, int Delta); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_QUANT_COMMON_H_ diff --git a/thirdparty/libvpx/vp8/common/reconinter.c b/thirdparty/libvpx/vp8/common/reconinter.c new file mode 100644 index 0000000000..e302595587 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/reconinter.c @@ -0,0 +1,544 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include <limits.h> +#include <string.h> + +#include "vpx_config.h" +#include "vp8_rtcd.h" +#include "vpx/vpx_integer.h" +#include "blockd.h" +#include "reconinter.h" +#if CONFIG_RUNTIME_CPU_DETECT +#include "onyxc_int.h" +#endif + +void vp8_copy_mem16x16_c( + unsigned char *src, + int src_stride, + unsigned char *dst, + int dst_stride) +{ + + int r; + + for (r = 0; r < 16; r++) + { + memcpy(dst, src, 16); + + src += src_stride; + dst += dst_stride; + + } + +} + +void vp8_copy_mem8x8_c( + unsigned char *src, + int src_stride, + unsigned char *dst, + int dst_stride) +{ + int r; + + for (r = 0; r < 8; r++) + { + memcpy(dst, src, 8); + + src += src_stride; + dst += dst_stride; + + } + +} + +void vp8_copy_mem8x4_c( + unsigned char *src, + int src_stride, + unsigned char *dst, + int dst_stride) +{ + int r; + + for (r = 0; r < 4; r++) + { + memcpy(dst, src, 8); + + src += src_stride; + dst += dst_stride; + + } + +} + + +void vp8_build_inter_predictors_b(BLOCKD *d, int pitch, unsigned char *base_pre, int pre_stride, vp8_subpix_fn_t sppf) +{ + int r; + unsigned char *pred_ptr = d->predictor; + unsigned char *ptr; + ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride + (d->bmi.mv.as_mv.col >> 3); + + if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7) + { + sppf(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, pred_ptr, pitch); + } + else + { + for (r = 0; r < 4; r++) + { + pred_ptr[0] = ptr[0]; + pred_ptr[1] = ptr[1]; + pred_ptr[2] = ptr[2]; + pred_ptr[3] = ptr[3]; + pred_ptr += pitch; + ptr += pre_stride; + } + } +} + +static void build_inter_predictors4b(MACROBLOCKD *x, BLOCKD *d, unsigned char *dst, int dst_stride, unsigned char *base_pre, int pre_stride) +{ + unsigned char *ptr; + ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride + (d->bmi.mv.as_mv.col >> 3); + + if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7) + { + x->subpixel_predict8x8(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, dst, dst_stride); + } + else + { + vp8_copy_mem8x8(ptr, pre_stride, dst, dst_stride); + } +} + +static void build_inter_predictors2b(MACROBLOCKD *x, BLOCKD *d, unsigned char *dst, int dst_stride, unsigned char *base_pre, int pre_stride) +{ + unsigned char *ptr; + ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride + (d->bmi.mv.as_mv.col >> 3); + + if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7) + { + x->subpixel_predict8x4(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, dst, dst_stride); + } + else + { + vp8_copy_mem8x4(ptr, pre_stride, dst, dst_stride); + } +} + +static void build_inter_predictors_b(BLOCKD *d, unsigned char *dst, int dst_stride, unsigned char *base_pre, int pre_stride, vp8_subpix_fn_t sppf) +{ + int r; + unsigned char *ptr; + ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride + (d->bmi.mv.as_mv.col >> 3); + + if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7) + { + sppf(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, dst, dst_stride); + } + else + { + for (r = 0; r < 4; r++) + { + dst[0] = ptr[0]; + dst[1] = ptr[1]; + dst[2] = ptr[2]; + dst[3] = ptr[3]; + dst += dst_stride; + ptr += pre_stride; + } + } +} + + +/*encoder only*/ +void vp8_build_inter16x16_predictors_mbuv(MACROBLOCKD *x) +{ + unsigned char *uptr, *vptr; + unsigned char *upred_ptr = &x->predictor[256]; + unsigned char *vpred_ptr = &x->predictor[320]; + + int mv_row = x->mode_info_context->mbmi.mv.as_mv.row; + int mv_col = x->mode_info_context->mbmi.mv.as_mv.col; + int offset; + int pre_stride = x->pre.uv_stride; + + /* calc uv motion vectors */ + mv_row += 1 | (mv_row >> (sizeof(int) * CHAR_BIT - 1)); + mv_col += 1 | (mv_col >> (sizeof(int) * CHAR_BIT - 1)); + mv_row /= 2; + mv_col /= 2; + mv_row &= x->fullpixel_mask; + mv_col &= x->fullpixel_mask; + + offset = (mv_row >> 3) * pre_stride + (mv_col >> 3); + uptr = x->pre.u_buffer + offset; + vptr = x->pre.v_buffer + offset; + + if ((mv_row | mv_col) & 7) + { + x->subpixel_predict8x8(uptr, pre_stride, mv_col & 7, mv_row & 7, upred_ptr, 8); + x->subpixel_predict8x8(vptr, pre_stride, mv_col & 7, mv_row & 7, vpred_ptr, 8); + } + else + { + vp8_copy_mem8x8(uptr, pre_stride, upred_ptr, 8); + vp8_copy_mem8x8(vptr, pre_stride, vpred_ptr, 8); + } +} + +/*encoder only*/ +void vp8_build_inter4x4_predictors_mbuv(MACROBLOCKD *x) +{ + int i, j; + int pre_stride = x->pre.uv_stride; + unsigned char *base_pre; + + /* build uv mvs */ + for (i = 0; i < 2; i++) + { + for (j = 0; j < 2; j++) + { + int yoffset = i * 8 + j * 2; + int uoffset = 16 + i * 2 + j; + int voffset = 20 + i * 2 + j; + + int temp; + + temp = x->block[yoffset ].bmi.mv.as_mv.row + + x->block[yoffset+1].bmi.mv.as_mv.row + + x->block[yoffset+4].bmi.mv.as_mv.row + + x->block[yoffset+5].bmi.mv.as_mv.row; + + temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8); + + x->block[uoffset].bmi.mv.as_mv.row = (temp / 8) & x->fullpixel_mask; + + temp = x->block[yoffset ].bmi.mv.as_mv.col + + x->block[yoffset+1].bmi.mv.as_mv.col + + x->block[yoffset+4].bmi.mv.as_mv.col + + x->block[yoffset+5].bmi.mv.as_mv.col; + + temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8); + + x->block[uoffset].bmi.mv.as_mv.col = (temp / 8) & x->fullpixel_mask; + + x->block[voffset].bmi.mv.as_int = x->block[uoffset].bmi.mv.as_int; + } + } + + base_pre = x->pre.u_buffer; + for (i = 16; i < 20; i += 2) + { + BLOCKD *d0 = &x->block[i]; + BLOCKD *d1 = &x->block[i+1]; + + if (d0->bmi.mv.as_int == d1->bmi.mv.as_int) + build_inter_predictors2b(x, d0, d0->predictor, 8, base_pre, pre_stride); + else + { + vp8_build_inter_predictors_b(d0, 8, base_pre, pre_stride, x->subpixel_predict); + vp8_build_inter_predictors_b(d1, 8, base_pre, pre_stride, x->subpixel_predict); + } + } + + base_pre = x->pre.v_buffer; + for (i = 20; i < 24; i += 2) + { + BLOCKD *d0 = &x->block[i]; + BLOCKD *d1 = &x->block[i+1]; + + if (d0->bmi.mv.as_int == d1->bmi.mv.as_int) + build_inter_predictors2b(x, d0, d0->predictor, 8, base_pre, pre_stride); + else + { + vp8_build_inter_predictors_b(d0, 8, base_pre, pre_stride, x->subpixel_predict); + vp8_build_inter_predictors_b(d1, 8, base_pre, pre_stride, x->subpixel_predict); + } + } +} + + +/*encoder only*/ +void vp8_build_inter16x16_predictors_mby(MACROBLOCKD *x, + unsigned char *dst_y, + int dst_ystride) +{ + unsigned char *ptr_base; + unsigned char *ptr; + int mv_row = x->mode_info_context->mbmi.mv.as_mv.row; + int mv_col = x->mode_info_context->mbmi.mv.as_mv.col; + int pre_stride = x->pre.y_stride; + + ptr_base = x->pre.y_buffer; + ptr = ptr_base + (mv_row >> 3) * pre_stride + (mv_col >> 3); + + if ((mv_row | mv_col) & 7) + { + x->subpixel_predict16x16(ptr, pre_stride, mv_col & 7, mv_row & 7, + dst_y, dst_ystride); + } + else + { + vp8_copy_mem16x16(ptr, pre_stride, dst_y, + dst_ystride); + } +} + +static void clamp_mv_to_umv_border(MV *mv, const MACROBLOCKD *xd) +{ + /* If the MV points so far into the UMV border that no visible pixels + * are used for reconstruction, the subpel part of the MV can be + * discarded and the MV limited to 16 pixels with equivalent results. + * + * This limit kicks in at 19 pixels for the top and left edges, for + * the 16 pixels plus 3 taps right of the central pixel when subpel + * filtering. The bottom and right edges use 16 pixels plus 2 pixels + * left of the central pixel when filtering. + */ + if (mv->col < (xd->mb_to_left_edge - (19 << 3))) + mv->col = xd->mb_to_left_edge - (16 << 3); + else if (mv->col > xd->mb_to_right_edge + (18 << 3)) + mv->col = xd->mb_to_right_edge + (16 << 3); + + if (mv->row < (xd->mb_to_top_edge - (19 << 3))) + mv->row = xd->mb_to_top_edge - (16 << 3); + else if (mv->row > xd->mb_to_bottom_edge + (18 << 3)) + mv->row = xd->mb_to_bottom_edge + (16 << 3); +} + +/* A version of the above function for chroma block MVs.*/ +static void clamp_uvmv_to_umv_border(MV *mv, const MACROBLOCKD *xd) +{ + mv->col = (2*mv->col < (xd->mb_to_left_edge - (19 << 3))) ? + (xd->mb_to_left_edge - (16 << 3)) >> 1 : mv->col; + mv->col = (2*mv->col > xd->mb_to_right_edge + (18 << 3)) ? + (xd->mb_to_right_edge + (16 << 3)) >> 1 : mv->col; + + mv->row = (2*mv->row < (xd->mb_to_top_edge - (19 << 3))) ? + (xd->mb_to_top_edge - (16 << 3)) >> 1 : mv->row; + mv->row = (2*mv->row > xd->mb_to_bottom_edge + (18 << 3)) ? + (xd->mb_to_bottom_edge + (16 << 3)) >> 1 : mv->row; +} + +void vp8_build_inter16x16_predictors_mb(MACROBLOCKD *x, + unsigned char *dst_y, + unsigned char *dst_u, + unsigned char *dst_v, + int dst_ystride, + int dst_uvstride) +{ + int offset; + unsigned char *ptr; + unsigned char *uptr, *vptr; + + int_mv _16x16mv; + + unsigned char *ptr_base = x->pre.y_buffer; + int pre_stride = x->pre.y_stride; + + _16x16mv.as_int = x->mode_info_context->mbmi.mv.as_int; + + if (x->mode_info_context->mbmi.need_to_clamp_mvs) + { + clamp_mv_to_umv_border(&_16x16mv.as_mv, x); + } + + ptr = ptr_base + ( _16x16mv.as_mv.row >> 3) * pre_stride + (_16x16mv.as_mv.col >> 3); + + if ( _16x16mv.as_int & 0x00070007) + { + x->subpixel_predict16x16(ptr, pre_stride, _16x16mv.as_mv.col & 7, _16x16mv.as_mv.row & 7, dst_y, dst_ystride); + } + else + { + vp8_copy_mem16x16(ptr, pre_stride, dst_y, dst_ystride); + } + + /* calc uv motion vectors */ + _16x16mv.as_mv.row += 1 | (_16x16mv.as_mv.row >> (sizeof(int) * CHAR_BIT - 1)); + _16x16mv.as_mv.col += 1 | (_16x16mv.as_mv.col >> (sizeof(int) * CHAR_BIT - 1)); + _16x16mv.as_mv.row /= 2; + _16x16mv.as_mv.col /= 2; + _16x16mv.as_mv.row &= x->fullpixel_mask; + _16x16mv.as_mv.col &= x->fullpixel_mask; + + pre_stride >>= 1; + offset = ( _16x16mv.as_mv.row >> 3) * pre_stride + (_16x16mv.as_mv.col >> 3); + uptr = x->pre.u_buffer + offset; + vptr = x->pre.v_buffer + offset; + + if ( _16x16mv.as_int & 0x00070007) + { + x->subpixel_predict8x8(uptr, pre_stride, _16x16mv.as_mv.col & 7, _16x16mv.as_mv.row & 7, dst_u, dst_uvstride); + x->subpixel_predict8x8(vptr, pre_stride, _16x16mv.as_mv.col & 7, _16x16mv.as_mv.row & 7, dst_v, dst_uvstride); + } + else + { + vp8_copy_mem8x8(uptr, pre_stride, dst_u, dst_uvstride); + vp8_copy_mem8x8(vptr, pre_stride, dst_v, dst_uvstride); + } +} + +static void build_inter4x4_predictors_mb(MACROBLOCKD *x) +{ + int i; + unsigned char *base_dst = x->dst.y_buffer; + unsigned char *base_pre = x->pre.y_buffer; + + if (x->mode_info_context->mbmi.partitioning < 3) + { + BLOCKD *b; + int dst_stride = x->dst.y_stride; + + x->block[ 0].bmi = x->mode_info_context->bmi[ 0]; + x->block[ 2].bmi = x->mode_info_context->bmi[ 2]; + x->block[ 8].bmi = x->mode_info_context->bmi[ 8]; + x->block[10].bmi = x->mode_info_context->bmi[10]; + if (x->mode_info_context->mbmi.need_to_clamp_mvs) + { + clamp_mv_to_umv_border(&x->block[ 0].bmi.mv.as_mv, x); + clamp_mv_to_umv_border(&x->block[ 2].bmi.mv.as_mv, x); + clamp_mv_to_umv_border(&x->block[ 8].bmi.mv.as_mv, x); + clamp_mv_to_umv_border(&x->block[10].bmi.mv.as_mv, x); + } + + b = &x->block[ 0]; + build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre, dst_stride); + b = &x->block[ 2]; + build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre, dst_stride); + b = &x->block[ 8]; + build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre, dst_stride); + b = &x->block[10]; + build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre, dst_stride); + } + else + { + for (i = 0; i < 16; i += 2) + { + BLOCKD *d0 = &x->block[i]; + BLOCKD *d1 = &x->block[i+1]; + int dst_stride = x->dst.y_stride; + + x->block[i+0].bmi = x->mode_info_context->bmi[i+0]; + x->block[i+1].bmi = x->mode_info_context->bmi[i+1]; + if (x->mode_info_context->mbmi.need_to_clamp_mvs) + { + clamp_mv_to_umv_border(&x->block[i+0].bmi.mv.as_mv, x); + clamp_mv_to_umv_border(&x->block[i+1].bmi.mv.as_mv, x); + } + + if (d0->bmi.mv.as_int == d1->bmi.mv.as_int) + build_inter_predictors2b(x, d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride); + else + { + build_inter_predictors_b(d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict); + build_inter_predictors_b(d1, base_dst + d1->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict); + } + + } + + } + base_dst = x->dst.u_buffer; + base_pre = x->pre.u_buffer; + for (i = 16; i < 20; i += 2) + { + BLOCKD *d0 = &x->block[i]; + BLOCKD *d1 = &x->block[i+1]; + int dst_stride = x->dst.uv_stride; + + /* Note: uv mvs already clamped in build_4x4uvmvs() */ + + if (d0->bmi.mv.as_int == d1->bmi.mv.as_int) + build_inter_predictors2b(x, d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride); + else + { + build_inter_predictors_b(d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict); + build_inter_predictors_b(d1, base_dst + d1->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict); + } + } + + base_dst = x->dst.v_buffer; + base_pre = x->pre.v_buffer; + for (i = 20; i < 24; i += 2) + { + BLOCKD *d0 = &x->block[i]; + BLOCKD *d1 = &x->block[i+1]; + int dst_stride = x->dst.uv_stride; + + /* Note: uv mvs already clamped in build_4x4uvmvs() */ + + if (d0->bmi.mv.as_int == d1->bmi.mv.as_int) + build_inter_predictors2b(x, d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride); + else + { + build_inter_predictors_b(d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict); + build_inter_predictors_b(d1, base_dst + d1->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict); + } + } +} + +static +void build_4x4uvmvs(MACROBLOCKD *x) +{ + int i, j; + + for (i = 0; i < 2; i++) + { + for (j = 0; j < 2; j++) + { + int yoffset = i * 8 + j * 2; + int uoffset = 16 + i * 2 + j; + int voffset = 20 + i * 2 + j; + + int temp; + + temp = x->mode_info_context->bmi[yoffset + 0].mv.as_mv.row + + x->mode_info_context->bmi[yoffset + 1].mv.as_mv.row + + x->mode_info_context->bmi[yoffset + 4].mv.as_mv.row + + x->mode_info_context->bmi[yoffset + 5].mv.as_mv.row; + + temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8); + + x->block[uoffset].bmi.mv.as_mv.row = (temp / 8) & x->fullpixel_mask; + + temp = x->mode_info_context->bmi[yoffset + 0].mv.as_mv.col + + x->mode_info_context->bmi[yoffset + 1].mv.as_mv.col + + x->mode_info_context->bmi[yoffset + 4].mv.as_mv.col + + x->mode_info_context->bmi[yoffset + 5].mv.as_mv.col; + + temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8); + + x->block[uoffset].bmi.mv.as_mv.col = (temp / 8) & x->fullpixel_mask; + + if (x->mode_info_context->mbmi.need_to_clamp_mvs) + clamp_uvmv_to_umv_border(&x->block[uoffset].bmi.mv.as_mv, x); + + x->block[voffset].bmi.mv.as_int = x->block[uoffset].bmi.mv.as_int; + } + } +} + +void vp8_build_inter_predictors_mb(MACROBLOCKD *xd) +{ + if (xd->mode_info_context->mbmi.mode != SPLITMV) + { + vp8_build_inter16x16_predictors_mb(xd, xd->dst.y_buffer, + xd->dst.u_buffer, xd->dst.v_buffer, + xd->dst.y_stride, xd->dst.uv_stride); + } + else + { + build_4x4uvmvs(xd); + build_inter4x4_predictors_mb(xd); + } +} diff --git a/thirdparty/libvpx/vp8/common/reconinter.h b/thirdparty/libvpx/vp8/common/reconinter.h new file mode 100644 index 0000000000..ba979b9664 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/reconinter.h @@ -0,0 +1,43 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_RECONINTER_H_ +#define VP8_COMMON_RECONINTER_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +extern void vp8_build_inter_predictors_mb(MACROBLOCKD *x); +extern void vp8_build_inter16x16_predictors_mb(MACROBLOCKD *x, + unsigned char *dst_y, + unsigned char *dst_u, + unsigned char *dst_v, + int dst_ystride, + int dst_uvstride); + + +extern void vp8_build_inter16x16_predictors_mby(MACROBLOCKD *x, + unsigned char *dst_y, + int dst_ystride); +extern void vp8_build_inter_predictors_b(BLOCKD *d, int pitch, + unsigned char *base_pre, + int pre_stride, + vp8_subpix_fn_t sppf); + +extern void vp8_build_inter16x16_predictors_mbuv(MACROBLOCKD *x); +extern void vp8_build_inter4x4_predictors_mbuv(MACROBLOCKD *x); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_RECONINTER_H_ diff --git a/thirdparty/libvpx/vp8/common/reconintra.c b/thirdparty/libvpx/vp8/common/reconintra.c new file mode 100644 index 0000000000..356655dac7 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/reconintra.c @@ -0,0 +1,117 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "./vpx_config.h" +#include "./vpx_dsp_rtcd.h" +#include "./vp8_rtcd.h" +#include "vpx_mem/vpx_mem.h" +#include "vpx_ports/vpx_once.h" +#include "blockd.h" +#include "vp8/common/reconintra.h" +#include "vp8/common/reconintra4x4.h" + +enum { + SIZE_16, + SIZE_8, + NUM_SIZES, +}; + +typedef void (*intra_pred_fn)(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left); + +static intra_pred_fn pred[4][NUM_SIZES]; +static intra_pred_fn dc_pred[2][2][NUM_SIZES]; + +static void vp8_init_intra_predictors_internal(void) +{ +#define INIT_SIZE(sz) \ + pred[V_PRED][SIZE_##sz] = vpx_v_predictor_##sz##x##sz; \ + pred[H_PRED][SIZE_##sz] = vpx_h_predictor_##sz##x##sz; \ + pred[TM_PRED][SIZE_##sz] = vpx_tm_predictor_##sz##x##sz; \ + \ + dc_pred[0][0][SIZE_##sz] = vpx_dc_128_predictor_##sz##x##sz; \ + dc_pred[0][1][SIZE_##sz] = vpx_dc_top_predictor_##sz##x##sz; \ + dc_pred[1][0][SIZE_##sz] = vpx_dc_left_predictor_##sz##x##sz; \ + dc_pred[1][1][SIZE_##sz] = vpx_dc_predictor_##sz##x##sz + + INIT_SIZE(16); + INIT_SIZE(8); + vp8_init_intra4x4_predictors_internal(); +} + +void vp8_build_intra_predictors_mby_s(MACROBLOCKD *x, + unsigned char * yabove_row, + unsigned char * yleft, + int left_stride, + unsigned char * ypred_ptr, + int y_stride) +{ + MB_PREDICTION_MODE mode = x->mode_info_context->mbmi.mode; + DECLARE_ALIGNED(16, uint8_t, yleft_col[16]); + int i; + intra_pred_fn fn; + + for (i = 0; i < 16; i++) + { + yleft_col[i] = yleft[i* left_stride]; + } + + if (mode == DC_PRED) + { + fn = dc_pred[x->left_available][x->up_available][SIZE_16]; + } + else + { + fn = pred[mode][SIZE_16]; + } + + fn(ypred_ptr, y_stride, yabove_row, yleft_col); +} + +void vp8_build_intra_predictors_mbuv_s(MACROBLOCKD *x, + unsigned char * uabove_row, + unsigned char * vabove_row, + unsigned char * uleft, + unsigned char * vleft, + int left_stride, + unsigned char * upred_ptr, + unsigned char * vpred_ptr, + int pred_stride) +{ + MB_PREDICTION_MODE uvmode = x->mode_info_context->mbmi.uv_mode; + unsigned char uleft_col[8]; + unsigned char vleft_col[8]; + int i; + intra_pred_fn fn; + + for (i = 0; i < 8; i++) + { + uleft_col[i] = uleft[i * left_stride]; + vleft_col[i] = vleft[i * left_stride]; + } + + if (uvmode == DC_PRED) + { + fn = dc_pred[x->left_available][x->up_available][SIZE_8]; + } + else + { + fn = pred[uvmode][SIZE_8]; + } + + fn(upred_ptr, pred_stride, uabove_row, uleft_col); + fn(vpred_ptr, pred_stride, vabove_row, vleft_col); +} + +void vp8_init_intra_predictors(void) +{ + once(vp8_init_intra_predictors_internal); +} diff --git a/thirdparty/libvpx/vp8/common/reconintra.h b/thirdparty/libvpx/vp8/common/reconintra.h new file mode 100644 index 0000000000..b6225a6637 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/reconintra.h @@ -0,0 +1,44 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_RECONINTRA_H_ +#define VP8_COMMON_RECONINTRA_H_ + +#include "vp8/common/blockd.h" + +#ifdef __cplusplus +extern "C" { +#endif + +void vp8_build_intra_predictors_mby_s(MACROBLOCKD *x, + unsigned char *yabove_row, + unsigned char *yleft, + int left_stride, + unsigned char *ypred_ptr, + int y_stride); + +void vp8_build_intra_predictors_mbuv_s(MACROBLOCKD *x, + unsigned char * uabove_row, + unsigned char * vabove_row, + unsigned char * uleft, + unsigned char * vleft, + int left_stride, + unsigned char * upred_ptr, + unsigned char * vpred_ptr, + int pred_stride); + +void vp8_init_intra_predictors(void); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_RECONINTRA_H_ diff --git a/thirdparty/libvpx/vp8/common/reconintra4x4.c b/thirdparty/libvpx/vp8/common/reconintra4x4.c new file mode 100644 index 0000000000..35ad891eff --- /dev/null +++ b/thirdparty/libvpx/vp8/common/reconintra4x4.c @@ -0,0 +1,54 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <string.h> + +#include "vpx_config.h" +#include "./vpx_dsp_rtcd.h" +#include "vp8_rtcd.h" +#include "blockd.h" + +typedef void (*intra_pred_fn)(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left); + +static intra_pred_fn pred[10]; + +void vp8_init_intra4x4_predictors_internal(void) +{ + pred[B_DC_PRED] = vpx_dc_predictor_4x4; + pred[B_TM_PRED] = vpx_tm_predictor_4x4; + pred[B_VE_PRED] = vpx_ve_predictor_4x4; + pred[B_HE_PRED] = vpx_he_predictor_4x4; + pred[B_LD_PRED] = vpx_d45e_predictor_4x4; + pred[B_RD_PRED] = vpx_d135_predictor_4x4; + pred[B_VR_PRED] = vpx_d117_predictor_4x4; + pred[B_VL_PRED] = vpx_d63f_predictor_4x4; + pred[B_HD_PRED] = vpx_d153_predictor_4x4; + pred[B_HU_PRED] = vpx_d207_predictor_4x4; +} + +void vp8_intra4x4_predict(unsigned char *above, + unsigned char *yleft, int left_stride, + B_PREDICTION_MODE b_mode, + unsigned char *dst, int dst_stride, + unsigned char top_left) +{ + unsigned char Left[4]; + unsigned char Aboveb[12], *Above = Aboveb + 4; + + Left[0] = yleft[0]; + Left[1] = yleft[left_stride]; + Left[2] = yleft[2 * left_stride]; + Left[3] = yleft[3 * left_stride]; + memcpy(Above, above, 8); + Above[-1] = top_left; + + pred[b_mode](dst, dst_stride, Above, Left); +} diff --git a/thirdparty/libvpx/vp8/common/reconintra4x4.h b/thirdparty/libvpx/vp8/common/reconintra4x4.h new file mode 100644 index 0000000000..5dc5d13a5c --- /dev/null +++ b/thirdparty/libvpx/vp8/common/reconintra4x4.h @@ -0,0 +1,48 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_RECONINTRA4X4_H_ +#define VP8_COMMON_RECONINTRA4X4_H_ +#include "vp8/common/blockd.h" + +#ifdef __cplusplus +extern "C" { +#endif + +static INLINE void intra_prediction_down_copy(MACROBLOCKD *xd, + unsigned char *above_right_src) +{ + int dst_stride = xd->dst.y_stride; + unsigned char *above_right_dst = xd->dst.y_buffer - dst_stride + 16; + + unsigned int *src_ptr = (unsigned int *)above_right_src; + unsigned int *dst_ptr0 = (unsigned int *)(above_right_dst + 4 * dst_stride); + unsigned int *dst_ptr1 = (unsigned int *)(above_right_dst + 8 * dst_stride); + unsigned int *dst_ptr2 = (unsigned int *)(above_right_dst + 12 * dst_stride); + + *dst_ptr0 = *src_ptr; + *dst_ptr1 = *src_ptr; + *dst_ptr2 = *src_ptr; +} + +void vp8_intra4x4_predict(unsigned char *Above, + unsigned char *yleft, int left_stride, + B_PREDICTION_MODE b_mode, + unsigned char *dst, int dst_stride, + unsigned char top_left); + +void vp8_init_intra4x4_predictors_internal(void); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_RECONINTRA4X4_H_ diff --git a/thirdparty/libvpx/vp8/common/rtcd.c b/thirdparty/libvpx/vp8/common/rtcd.c new file mode 100644 index 0000000000..ab0e9b47fe --- /dev/null +++ b/thirdparty/libvpx/vp8/common/rtcd.c @@ -0,0 +1,19 @@ +/* + * Copyright (c) 2011 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ +#include "./vpx_config.h" +#define RTCD_C +#include "./vp8_rtcd.h" +#include "vpx_ports/vpx_once.h" + + +void vp8_rtcd() +{ + once(setup_rtcd_internal); +} diff --git a/thirdparty/libvpx/vp8/common/setupintrarecon.c b/thirdparty/libvpx/vp8/common/setupintrarecon.c new file mode 100644 index 0000000000..669564db42 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/setupintrarecon.c @@ -0,0 +1,39 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "setupintrarecon.h" +#include "vpx_mem/vpx_mem.h" + +void vp8_setup_intra_recon(YV12_BUFFER_CONFIG *ybf) +{ + int i; + + /* set up frame new frame for intra coded blocks */ + memset(ybf->y_buffer - 1 - ybf->y_stride, 127, ybf->y_width + 5); + for (i = 0; i < ybf->y_height; i++) + ybf->y_buffer[ybf->y_stride *i - 1] = (unsigned char) 129; + + memset(ybf->u_buffer - 1 - ybf->uv_stride, 127, ybf->uv_width + 5); + for (i = 0; i < ybf->uv_height; i++) + ybf->u_buffer[ybf->uv_stride *i - 1] = (unsigned char) 129; + + memset(ybf->v_buffer - 1 - ybf->uv_stride, 127, ybf->uv_width + 5); + for (i = 0; i < ybf->uv_height; i++) + ybf->v_buffer[ybf->uv_stride *i - 1] = (unsigned char) 129; + +} + +void vp8_setup_intra_recon_top_line(YV12_BUFFER_CONFIG *ybf) +{ + memset(ybf->y_buffer - 1 - ybf->y_stride, 127, ybf->y_width + 5); + memset(ybf->u_buffer - 1 - ybf->uv_stride, 127, ybf->uv_width + 5); + memset(ybf->v_buffer - 1 - ybf->uv_stride, 127, ybf->uv_width + 5); +} diff --git a/thirdparty/libvpx/vp8/common/setupintrarecon.h b/thirdparty/libvpx/vp8/common/setupintrarecon.h new file mode 100644 index 0000000000..1857c4e26a --- /dev/null +++ b/thirdparty/libvpx/vp8/common/setupintrarecon.h @@ -0,0 +1,45 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP8_COMMON_SETUPINTRARECON_H_ +#define VP8_COMMON_SETUPINTRARECON_H_ + +#include "./vpx_config.h" +#include "vpx_scale/yv12config.h" + +#ifdef __cplusplus +extern "C" { +#endif +extern void vp8_setup_intra_recon(YV12_BUFFER_CONFIG *ybf); +extern void vp8_setup_intra_recon_top_line(YV12_BUFFER_CONFIG *ybf); + +static INLINE void setup_intra_recon_left(unsigned char *y_buffer, + unsigned char *u_buffer, + unsigned char *v_buffer, + int y_stride, + int uv_stride) +{ + int i; + + for (i = 0; i < 16; i++) + y_buffer[y_stride *i] = (unsigned char) 129; + + for (i = 0; i < 8; i++) + u_buffer[uv_stride *i] = (unsigned char) 129; + + for (i = 0; i < 8; i++) + v_buffer[uv_stride *i] = (unsigned char) 129; +} + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_SETUPINTRARECON_H_ diff --git a/thirdparty/libvpx/vp8/common/swapyv12buffer.c b/thirdparty/libvpx/vp8/common/swapyv12buffer.c new file mode 100644 index 0000000000..73656b3d72 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/swapyv12buffer.c @@ -0,0 +1,34 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "swapyv12buffer.h" + +void vp8_swap_yv12_buffer(YV12_BUFFER_CONFIG *new_frame, YV12_BUFFER_CONFIG *last_frame) +{ + unsigned char *temp; + + temp = last_frame->buffer_alloc; + last_frame->buffer_alloc = new_frame->buffer_alloc; + new_frame->buffer_alloc = temp; + + temp = last_frame->y_buffer; + last_frame->y_buffer = new_frame->y_buffer; + new_frame->y_buffer = temp; + + temp = last_frame->u_buffer; + last_frame->u_buffer = new_frame->u_buffer; + new_frame->u_buffer = temp; + + temp = last_frame->v_buffer; + last_frame->v_buffer = new_frame->v_buffer; + new_frame->v_buffer = temp; + +} diff --git a/thirdparty/libvpx/vp8/common/swapyv12buffer.h b/thirdparty/libvpx/vp8/common/swapyv12buffer.h new file mode 100644 index 0000000000..1d66cd3d62 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/swapyv12buffer.h @@ -0,0 +1,27 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_SWAPYV12BUFFER_H_ +#define VP8_COMMON_SWAPYV12BUFFER_H_ + +#include "vpx_scale/yv12config.h" + +#ifdef __cplusplus +extern "C" { +#endif + +void vp8_swap_yv12_buffer(YV12_BUFFER_CONFIG *new_frame, YV12_BUFFER_CONFIG *last_frame); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_SWAPYV12BUFFER_H_ diff --git a/thirdparty/libvpx/vp8/common/systemdependent.h b/thirdparty/libvpx/vp8/common/systemdependent.h new file mode 100644 index 0000000000..3d44e37cf2 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/systemdependent.h @@ -0,0 +1,27 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP8_COMMON_SYSTEMDEPENDENT_H_ +#define VP8_COMMON_SYSTEMDEPENDENT_H_ + +#include "vpx_config.h" + +#ifdef __cplusplus +extern "C" { +#endif + +struct VP8Common; +void vp8_machine_specific_config(struct VP8Common *); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_SYSTEMDEPENDENT_H_ diff --git a/thirdparty/libvpx/vp8/common/threading.h b/thirdparty/libvpx/vp8/common/threading.h new file mode 100644 index 0000000000..183b49b8ff --- /dev/null +++ b/thirdparty/libvpx/vp8/common/threading.h @@ -0,0 +1,232 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_THREADING_H_ +#define VP8_COMMON_THREADING_H_ + +#include "./vpx_config.h" + +#ifdef __cplusplus +extern "C" { +#endif + +#if CONFIG_OS_SUPPORT && CONFIG_MULTITHREAD + +/* Thread management macros */ +#if defined(_WIN32) && !HAVE_PTHREAD_H +/* Win32 */ +#include <process.h> +#include <windows.h> +#define THREAD_FUNCTION unsigned int __stdcall +#define THREAD_FUNCTION_RETURN DWORD +#define THREAD_SPECIFIC_INDEX DWORD +#define pthread_t HANDLE +#define pthread_attr_t DWORD +#define pthread_detach(thread) if(thread!=NULL)CloseHandle(thread) +#define thread_sleep(nms) Sleep(nms) +#define pthread_cancel(thread) terminate_thread(thread,0) +#define ts_key_create(ts_key, destructor) {ts_key = TlsAlloc();}; +#define pthread_getspecific(ts_key) TlsGetValue(ts_key) +#define pthread_setspecific(ts_key, value) TlsSetValue(ts_key, (void *)value) +#define pthread_self() GetCurrentThreadId() + +#elif defined(__OS2__) +/* OS/2 */ +#define INCL_DOS +#include <os2.h> + +#include <stdlib.h> +#define THREAD_FUNCTION void * +#define THREAD_FUNCTION_RETURN void * +#define THREAD_SPECIFIC_INDEX PULONG +#define pthread_t TID +#define pthread_attr_t ULONG +#define pthread_detach(thread) 0 +#define thread_sleep(nms) DosSleep(nms) +#define pthread_cancel(thread) DosKillThread(thread) +#define ts_key_create(ts_key, destructor) \ + DosAllocThreadLocalMemory(1, &(ts_key)); +#define pthread_getspecific(ts_key) ((void *)(*(ts_key))) +#define pthread_setspecific(ts_key, value) (*(ts_key)=(ULONG)(value)) +#define pthread_self() _gettid() +#else +#ifdef __APPLE__ +#include <mach/mach_init.h> +#include <mach/semaphore.h> +#include <mach/task.h> +#include <time.h> +#include <unistd.h> + +#else +#include <semaphore.h> +#endif + +#include <pthread.h> +/* pthreads */ +/* Nearly everything is already defined */ +#define THREAD_FUNCTION void * +#define THREAD_FUNCTION_RETURN void * +#define THREAD_SPECIFIC_INDEX pthread_key_t +#define ts_key_create(ts_key, destructor) pthread_key_create (&(ts_key), destructor); +#endif + +/* Synchronization macros: Win32 and Pthreads */ +#if defined(_WIN32) && !HAVE_PTHREAD_H +#define sem_t HANDLE +#define pause(voidpara) __asm PAUSE +#define sem_init(sem, sem_attr1, sem_init_value) (int)((*sem = CreateSemaphore(NULL,0,32768,NULL))==NULL) +#define sem_wait(sem) (int)(WAIT_OBJECT_0 != WaitForSingleObject(*sem,INFINITE)) +#define sem_post(sem) ReleaseSemaphore(*sem,1,NULL) +#define sem_destroy(sem) if(*sem)((int)(CloseHandle(*sem))==TRUE) +#define thread_sleep(nms) Sleep(nms) + +#elif defined(__OS2__) +typedef struct +{ + HEV event; + HMTX wait_mutex; + HMTX count_mutex; + int count; +} sem_t; + +static inline int sem_init(sem_t *sem, int pshared, unsigned int value) +{ + DosCreateEventSem(NULL, &sem->event, pshared ? DC_SEM_SHARED : 0, + value > 0 ? TRUE : FALSE); + DosCreateMutexSem(NULL, &sem->wait_mutex, 0, FALSE); + DosCreateMutexSem(NULL, &sem->count_mutex, 0, FALSE); + + sem->count = value; + + return 0; +} + +static inline int sem_wait(sem_t * sem) +{ + DosRequestMutexSem(sem->wait_mutex, -1); + + DosWaitEventSem(sem->event, -1); + + DosRequestMutexSem(sem->count_mutex, -1); + + sem->count--; + if (sem->count == 0) + { + ULONG post_count; + + DosResetEventSem(sem->event, &post_count); + } + + DosReleaseMutexSem(sem->count_mutex); + + DosReleaseMutexSem(sem->wait_mutex); + + return 0; +} + +static inline int sem_post(sem_t * sem) +{ + DosRequestMutexSem(sem->count_mutex, -1); + + if (sem->count < 32768) + { + sem->count++; + DosPostEventSem(sem->event); + } + + DosReleaseMutexSem(sem->count_mutex); + + return 0; +} + +static inline int sem_destroy(sem_t * sem) +{ + DosCloseEventSem(sem->event); + DosCloseMutexSem(sem->wait_mutex); + DosCloseMutexSem(sem->count_mutex); + + return 0; +} + +#define thread_sleep(nms) DosSleep(nms) + +#else + +#ifdef __APPLE__ +#define sem_t semaphore_t +#define sem_init(X,Y,Z) semaphore_create(mach_task_self(), X, SYNC_POLICY_FIFO, Z) +#define sem_wait(sem) (semaphore_wait(*sem) ) +#define sem_post(sem) semaphore_signal(*sem) +#define sem_destroy(sem) semaphore_destroy(mach_task_self(),*sem) +#define thread_sleep(nms) /* { struct timespec ts;ts.tv_sec=0; ts.tv_nsec = 1000*nms;nanosleep(&ts, NULL);} */ +#else +#include <unistd.h> +#include <sched.h> +#define thread_sleep(nms) sched_yield();/* {struct timespec ts;ts.tv_sec=0; ts.tv_nsec = 1000*nms;nanosleep(&ts, NULL);} */ +#endif +/* Not Windows. Assume pthreads */ + +#endif + +#if ARCH_X86 || ARCH_X86_64 +#include "vpx_ports/x86.h" +#else +#define x86_pause_hint() +#endif + +#include "vpx_util/vpx_thread.h" + +static INLINE void mutex_lock(pthread_mutex_t *const mutex) { + const int kMaxTryLocks = 4000; + int locked = 0; + int i; + + for (i = 0; i < kMaxTryLocks; ++i) { + if (!pthread_mutex_trylock(mutex)) { + locked = 1; + break; + } + } + + if (!locked) + pthread_mutex_lock(mutex); +} + +static INLINE int protected_read(pthread_mutex_t *const mutex, const int *p) { + int ret; + mutex_lock(mutex); + ret = *p; + pthread_mutex_unlock(mutex); + return ret; +} + +static INLINE void sync_read(pthread_mutex_t *const mutex, int mb_col, + const int *last_row_current_mb_col, + const int nsync) { + while (mb_col > (protected_read(mutex, last_row_current_mb_col) - nsync)) { + x86_pause_hint(); + thread_sleep(0); + } +} + +static INLINE void protected_write(pthread_mutex_t *mutex, int *p, int v) { + mutex_lock(mutex); + *p = v; + pthread_mutex_unlock(mutex); +} + +#endif /* CONFIG_OS_SUPPORT && CONFIG_MULTITHREAD */ + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_THREADING_H_ diff --git a/thirdparty/libvpx/vp8/common/treecoder.c b/thirdparty/libvpx/vp8/common/treecoder.c new file mode 100644 index 0000000000..d80c64bdfa --- /dev/null +++ b/thirdparty/libvpx/vp8/common/treecoder.c @@ -0,0 +1,143 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#if CONFIG_DEBUG +#include <assert.h> +#endif +#include <stdio.h> + +#include "treecoder.h" + +static void tree2tok( + struct vp8_token_struct *const p, + vp8_tree t, + int i, + int v, + int L +) +{ + v += v; + ++L; + + do + { + const vp8_tree_index j = t[i++]; + + if (j <= 0) + { + p[-j].value = v; + p[-j].Len = L; + } + else + tree2tok(p, t, j, v, L); + } + while (++v & 1); +} + +void vp8_tokens_from_tree(struct vp8_token_struct *p, vp8_tree t) +{ + tree2tok(p, t, 0, 0, 0); +} + +void vp8_tokens_from_tree_offset(struct vp8_token_struct *p, vp8_tree t, + int offset) +{ + tree2tok(p - offset, t, 0, 0, 0); +} + +static void branch_counts( + int n, /* n = size of alphabet */ + vp8_token tok [ /* n */ ], + vp8_tree tree, + unsigned int branch_ct [ /* n-1 */ ] [2], + const unsigned int num_events[ /* n */ ] +) +{ + const int tree_len = n - 1; + int t = 0; + +#if CONFIG_DEBUG + assert(tree_len); +#endif + + do + { + branch_ct[t][0] = branch_ct[t][1] = 0; + } + while (++t < tree_len); + + t = 0; + + do + { + int L = tok[t].Len; + const int enc = tok[t].value; + const unsigned int ct = num_events[t]; + + vp8_tree_index i = 0; + + do + { + const int b = (enc >> --L) & 1; + const int j = i >> 1; +#if CONFIG_DEBUG + assert(j < tree_len && 0 <= L); +#endif + + branch_ct [j] [b] += ct; + i = tree[ i + b]; + } + while (i > 0); + +#if CONFIG_DEBUG + assert(!L); +#endif + } + while (++t < n); + +} + + +void vp8_tree_probs_from_distribution( + int n, /* n = size of alphabet */ + vp8_token tok [ /* n */ ], + vp8_tree tree, + vp8_prob probs [ /* n-1 */ ], + unsigned int branch_ct [ /* n-1 */ ] [2], + const unsigned int num_events[ /* n */ ], + unsigned int Pfac, + int rd +) +{ + const int tree_len = n - 1; + int t = 0; + + branch_counts(n, tok, tree, branch_ct, num_events); + + do + { + const unsigned int *const c = branch_ct[t]; + const unsigned int tot = c[0] + c[1]; + +#if CONFIG_DEBUG + assert(tot < (1 << 24)); /* no overflow below */ +#endif + + if (tot) + { + const unsigned int p = ((c[0] * Pfac) + (rd ? tot >> 1 : 0)) / tot; + probs[t] = p < 256 ? (p ? p : 1) : 255; /* agree w/old version for now */ + } + else + probs[t] = vp8_prob_half; + } + while (++t < tree_len); +} diff --git a/thirdparty/libvpx/vp8/common/treecoder.h b/thirdparty/libvpx/vp8/common/treecoder.h new file mode 100644 index 0000000000..d22b7c570c --- /dev/null +++ b/thirdparty/libvpx/vp8/common/treecoder.h @@ -0,0 +1,98 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_COMMON_TREECODER_H_ +#define VP8_COMMON_TREECODER_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +typedef unsigned char vp8bc_index_t; /* probability index */ + + +typedef unsigned char vp8_prob; + +#define vp8_prob_half ( (vp8_prob) 128) + +typedef signed char vp8_tree_index; +struct bool_coder_spec; + +typedef struct bool_coder_spec bool_coder_spec; +typedef struct bool_writer bool_writer; +typedef struct bool_reader bool_reader; + +typedef const bool_coder_spec c_bool_coder_spec; +typedef const bool_writer c_bool_writer; +typedef const bool_reader c_bool_reader; + + + +# define vp8_complement( x) (255 - x) + + +/* We build coding trees compactly in arrays. + Each node of the tree is a pair of vp8_tree_indices. + Array index often references a corresponding probability table. + Index <= 0 means done encoding/decoding and value = -Index, + Index > 0 means need another bit, specification at index. + Nonnegative indices are always even; processing begins at node 0. */ + +typedef const vp8_tree_index vp8_tree[], *vp8_tree_p; + + +typedef const struct vp8_token_struct +{ + int value; + int Len; +} vp8_token; + +/* Construct encoding array from tree. */ + +void vp8_tokens_from_tree(struct vp8_token_struct *, vp8_tree); +void vp8_tokens_from_tree_offset(struct vp8_token_struct *, vp8_tree, + int offset); + + +/* Convert array of token occurrence counts into a table of probabilities + for the associated binary encoding tree. Also writes count of branches + taken for each node on the tree; this facilitiates decisions as to + probability updates. */ + +void vp8_tree_probs_from_distribution( + int n, /* n = size of alphabet */ + vp8_token tok [ /* n */ ], + vp8_tree tree, + vp8_prob probs [ /* n-1 */ ], + unsigned int branch_ct [ /* n-1 */ ] [2], + const unsigned int num_events[ /* n */ ], + unsigned int Pfactor, + int Round +); + +/* Variant of above using coder spec rather than hardwired 8-bit probs. */ + +void vp8bc_tree_probs_from_distribution( + int n, /* n = size of alphabet */ + vp8_token tok [ /* n */ ], + vp8_tree tree, + vp8_prob probs [ /* n-1 */ ], + unsigned int branch_ct [ /* n-1 */ ] [2], + const unsigned int num_events[ /* n */ ], + c_bool_coder_spec *s +); + + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_TREECODER_H_ diff --git a/thirdparty/libvpx/vp8/common/vp8_entropymodedata.h b/thirdparty/libvpx/vp8/common/vp8_entropymodedata.h new file mode 100644 index 0000000000..c4aed49897 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/vp8_entropymodedata.h @@ -0,0 +1,254 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. +*/ + +#ifndef VP8_COMMON_VP8_ENTROPYMODEDATA_H_ +#define VP8_COMMON_VP8_ENTROPYMODEDATA_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +/*Generated file, included by entropymode.c*/ + + +const struct vp8_token_struct vp8_bmode_encodings[VP8_BINTRAMODES] = +{ + { 0, 1 }, + { 2, 2 }, + { 6, 3 }, + { 28, 5 }, + { 30, 5 }, + { 58, 6 }, + { 59, 6 }, + { 62, 6 }, + { 126, 7 }, + { 127, 7 } +}; + +const struct vp8_token_struct vp8_ymode_encodings[VP8_YMODES] = +{ + { 0, 1 }, + { 4, 3 }, + { 5, 3 }, + { 6, 3 }, + { 7, 3 } +}; + +const struct vp8_token_struct vp8_kf_ymode_encodings[VP8_YMODES] = +{ + { 4, 3 }, + { 5, 3 }, + { 6, 3 }, + { 7, 3 }, + { 0, 1 } +}; + +const struct vp8_token_struct vp8_uv_mode_encodings[VP8_UV_MODES] = +{ + { 0, 1 }, + { 2, 2 }, + { 6, 3 }, + { 7, 3 } +}; + +const struct vp8_token_struct vp8_mbsplit_encodings[VP8_NUMMBSPLITS] = +{ + { 6, 3 }, + { 7, 3 }, + { 2, 2 }, + { 0, 1 } +}; + +const struct vp8_token_struct vp8_mv_ref_encoding_array[VP8_MVREFS] = +{ + { 2, 2 }, + { 6, 3 }, + { 0, 1 }, + { 14, 4 }, + { 15, 4 } +}; + +const struct vp8_token_struct vp8_sub_mv_ref_encoding_array[VP8_SUBMVREFS] = +{ + { 0, 1 }, + { 2, 2 }, + { 6, 3 }, + { 7, 3 } +}; + +const struct vp8_token_struct vp8_small_mvencodings[8] = +{ + { 0, 3 }, + { 1, 3 }, + { 2, 3 }, + { 3, 3 }, + { 4, 3 }, + { 5, 3 }, + { 6, 3 }, + { 7, 3 } +}; + +const vp8_prob vp8_ymode_prob[VP8_YMODES-1] = +{ + 112, 86, 140, 37 +}; + +const vp8_prob vp8_kf_ymode_prob[VP8_YMODES-1] = +{ + 145, 156, 163, 128 +}; + +const vp8_prob vp8_uv_mode_prob[VP8_UV_MODES-1] = +{ + 162, 101, 204 +}; + +const vp8_prob vp8_kf_uv_mode_prob[VP8_UV_MODES-1] = +{ + 142, 114, 183 +}; + +const vp8_prob vp8_bmode_prob[VP8_BINTRAMODES-1] = +{ + 120, 90, 79, 133, 87, 85, 80, 111, 151 +}; + + + +const vp8_prob vp8_kf_bmode_prob +[VP8_BINTRAMODES] [VP8_BINTRAMODES] [VP8_BINTRAMODES-1] = +{ + { + { 231, 120, 48, 89, 115, 113, 120, 152, 112 }, + { 152, 179, 64, 126, 170, 118, 46, 70, 95 }, + { 175, 69, 143, 80, 85, 82, 72, 155, 103 }, + { 56, 58, 10, 171, 218, 189, 17, 13, 152 }, + { 144, 71, 10, 38, 171, 213, 144, 34, 26 }, + { 114, 26, 17, 163, 44, 195, 21, 10, 173 }, + { 121, 24, 80, 195, 26, 62, 44, 64, 85 }, + { 170, 46, 55, 19, 136, 160, 33, 206, 71 }, + { 63, 20, 8, 114, 114, 208, 12, 9, 226 }, + { 81, 40, 11, 96, 182, 84, 29, 16, 36 } + }, + { + { 134, 183, 89, 137, 98, 101, 106, 165, 148 }, + { 72, 187, 100, 130, 157, 111, 32, 75, 80 }, + { 66, 102, 167, 99, 74, 62, 40, 234, 128 }, + { 41, 53, 9, 178, 241, 141, 26, 8, 107 }, + { 104, 79, 12, 27, 217, 255, 87, 17, 7 }, + { 74, 43, 26, 146, 73, 166, 49, 23, 157 }, + { 65, 38, 105, 160, 51, 52, 31, 115, 128 }, + { 87, 68, 71, 44, 114, 51, 15, 186, 23 }, + { 47, 41, 14, 110, 182, 183, 21, 17, 194 }, + { 66, 45, 25, 102, 197, 189, 23, 18, 22 } + }, + { + { 88, 88, 147, 150, 42, 46, 45, 196, 205 }, + { 43, 97, 183, 117, 85, 38, 35, 179, 61 }, + { 39, 53, 200, 87, 26, 21, 43, 232, 171 }, + { 56, 34, 51, 104, 114, 102, 29, 93, 77 }, + { 107, 54, 32, 26, 51, 1, 81, 43, 31 }, + { 39, 28, 85, 171, 58, 165, 90, 98, 64 }, + { 34, 22, 116, 206, 23, 34, 43, 166, 73 }, + { 68, 25, 106, 22, 64, 171, 36, 225, 114 }, + { 34, 19, 21, 102, 132, 188, 16, 76, 124 }, + { 62, 18, 78, 95, 85, 57, 50, 48, 51 } + }, + { + { 193, 101, 35, 159, 215, 111, 89, 46, 111 }, + { 60, 148, 31, 172, 219, 228, 21, 18, 111 }, + { 112, 113, 77, 85, 179, 255, 38, 120, 114 }, + { 40, 42, 1, 196, 245, 209, 10, 25, 109 }, + { 100, 80, 8, 43, 154, 1, 51, 26, 71 }, + { 88, 43, 29, 140, 166, 213, 37, 43, 154 }, + { 61, 63, 30, 155, 67, 45, 68, 1, 209 }, + { 142, 78, 78, 16, 255, 128, 34, 197, 171 }, + { 41, 40, 5, 102, 211, 183, 4, 1, 221 }, + { 51, 50, 17, 168, 209, 192, 23, 25, 82 } + }, + { + { 125, 98, 42, 88, 104, 85, 117, 175, 82 }, + { 95, 84, 53, 89, 128, 100, 113, 101, 45 }, + { 75, 79, 123, 47, 51, 128, 81, 171, 1 }, + { 57, 17, 5, 71, 102, 57, 53, 41, 49 }, + { 115, 21, 2, 10, 102, 255, 166, 23, 6 }, + { 38, 33, 13, 121, 57, 73, 26, 1, 85 }, + { 41, 10, 67, 138, 77, 110, 90, 47, 114 }, + { 101, 29, 16, 10, 85, 128, 101, 196, 26 }, + { 57, 18, 10, 102, 102, 213, 34, 20, 43 }, + { 117, 20, 15, 36, 163, 128, 68, 1, 26 } + }, + { + { 138, 31, 36, 171, 27, 166, 38, 44, 229 }, + { 67, 87, 58, 169, 82, 115, 26, 59, 179 }, + { 63, 59, 90, 180, 59, 166, 93, 73, 154 }, + { 40, 40, 21, 116, 143, 209, 34, 39, 175 }, + { 57, 46, 22, 24, 128, 1, 54, 17, 37 }, + { 47, 15, 16, 183, 34, 223, 49, 45, 183 }, + { 46, 17, 33, 183, 6, 98, 15, 32, 183 }, + { 65, 32, 73, 115, 28, 128, 23, 128, 205 }, + { 40, 3, 9, 115, 51, 192, 18, 6, 223 }, + { 87, 37, 9, 115, 59, 77, 64, 21, 47 } + }, + { + { 104, 55, 44, 218, 9, 54, 53, 130, 226 }, + { 64, 90, 70, 205, 40, 41, 23, 26, 57 }, + { 54, 57, 112, 184, 5, 41, 38, 166, 213 }, + { 30, 34, 26, 133, 152, 116, 10, 32, 134 }, + { 75, 32, 12, 51, 192, 255, 160, 43, 51 }, + { 39, 19, 53, 221, 26, 114, 32, 73, 255 }, + { 31, 9, 65, 234, 2, 15, 1, 118, 73 }, + { 88, 31, 35, 67, 102, 85, 55, 186, 85 }, + { 56, 21, 23, 111, 59, 205, 45, 37, 192 }, + { 55, 38, 70, 124, 73, 102, 1, 34, 98 } + }, + { + { 102, 61, 71, 37, 34, 53, 31, 243, 192 }, + { 69, 60, 71, 38, 73, 119, 28, 222, 37 }, + { 68, 45, 128, 34, 1, 47, 11, 245, 171 }, + { 62, 17, 19, 70, 146, 85, 55, 62, 70 }, + { 75, 15, 9, 9, 64, 255, 184, 119, 16 }, + { 37, 43, 37, 154, 100, 163, 85, 160, 1 }, + { 63, 9, 92, 136, 28, 64, 32, 201, 85 }, + { 86, 6, 28, 5, 64, 255, 25, 248, 1 }, + { 56, 8, 17, 132, 137, 255, 55, 116, 128 }, + { 58, 15, 20, 82, 135, 57, 26, 121, 40 } + }, + { + { 164, 50, 31, 137, 154, 133, 25, 35, 218 }, + { 51, 103, 44, 131, 131, 123, 31, 6, 158 }, + { 86, 40, 64, 135, 148, 224, 45, 183, 128 }, + { 22, 26, 17, 131, 240, 154, 14, 1, 209 }, + { 83, 12, 13, 54, 192, 255, 68, 47, 28 }, + { 45, 16, 21, 91, 64, 222, 7, 1, 197 }, + { 56, 21, 39, 155, 60, 138, 23, 102, 213 }, + { 85, 26, 85, 85, 128, 128, 32, 146, 171 }, + { 18, 11, 7, 63, 144, 171, 4, 4, 246 }, + { 35, 27, 10, 146, 174, 171, 12, 26, 128 } + }, + { + { 190, 80, 35, 99, 180, 80, 126, 54, 45 }, + { 85, 126, 47, 87, 176, 51, 41, 20, 32 }, + { 101, 75, 128, 139, 118, 146, 116, 128, 85 }, + { 56, 41, 15, 176, 236, 85, 37, 9, 62 }, + { 146, 36, 19, 30, 171, 255, 97, 27, 20 }, + { 71, 30, 17, 119, 118, 255, 17, 18, 138 }, + { 101, 38, 60, 138, 55, 70, 43, 26, 142 }, + { 138, 45, 61, 62, 219, 1, 81, 188, 64 }, + { 32, 41, 20, 117, 151, 142, 20, 21, 163 }, + { 112, 19, 12, 61, 195, 128, 48, 4, 24 } + } +}; + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_VP8_ENTROPYMODEDATA_H_ diff --git a/thirdparty/libvpx/vp8/common/vp8_loopfilter.c b/thirdparty/libvpx/vp8/common/vp8_loopfilter.c new file mode 100644 index 0000000000..756ad488f9 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/vp8_loopfilter.c @@ -0,0 +1,661 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "vpx_config.h" +#include "vp8_rtcd.h" +#include "loopfilter.h" +#include "onyxc_int.h" +#include "vpx_mem/vpx_mem.h" + + +static void lf_init_lut(loop_filter_info_n *lfi) +{ + int filt_lvl; + + for (filt_lvl = 0; filt_lvl <= MAX_LOOP_FILTER; filt_lvl++) + { + if (filt_lvl >= 40) + { + lfi->hev_thr_lut[KEY_FRAME][filt_lvl] = 2; + lfi->hev_thr_lut[INTER_FRAME][filt_lvl] = 3; + } + else if (filt_lvl >= 20) + { + lfi->hev_thr_lut[KEY_FRAME][filt_lvl] = 1; + lfi->hev_thr_lut[INTER_FRAME][filt_lvl] = 2; + } + else if (filt_lvl >= 15) + { + lfi->hev_thr_lut[KEY_FRAME][filt_lvl] = 1; + lfi->hev_thr_lut[INTER_FRAME][filt_lvl] = 1; + } + else + { + lfi->hev_thr_lut[KEY_FRAME][filt_lvl] = 0; + lfi->hev_thr_lut[INTER_FRAME][filt_lvl] = 0; + } + } + + lfi->mode_lf_lut[DC_PRED] = 1; + lfi->mode_lf_lut[V_PRED] = 1; + lfi->mode_lf_lut[H_PRED] = 1; + lfi->mode_lf_lut[TM_PRED] = 1; + lfi->mode_lf_lut[B_PRED] = 0; + + lfi->mode_lf_lut[ZEROMV] = 1; + lfi->mode_lf_lut[NEARESTMV] = 2; + lfi->mode_lf_lut[NEARMV] = 2; + lfi->mode_lf_lut[NEWMV] = 2; + lfi->mode_lf_lut[SPLITMV] = 3; + +} + +void vp8_loop_filter_update_sharpness(loop_filter_info_n *lfi, + int sharpness_lvl) +{ + int i; + + /* For each possible value for the loop filter fill out limits */ + for (i = 0; i <= MAX_LOOP_FILTER; i++) + { + int filt_lvl = i; + int block_inside_limit = 0; + + /* Set loop filter paramaeters that control sharpness. */ + block_inside_limit = filt_lvl >> (sharpness_lvl > 0); + block_inside_limit = block_inside_limit >> (sharpness_lvl > 4); + + if (sharpness_lvl > 0) + { + if (block_inside_limit > (9 - sharpness_lvl)) + block_inside_limit = (9 - sharpness_lvl); + } + + if (block_inside_limit < 1) + block_inside_limit = 1; + + memset(lfi->lim[i], block_inside_limit, SIMD_WIDTH); + memset(lfi->blim[i], (2 * filt_lvl + block_inside_limit), SIMD_WIDTH); + memset(lfi->mblim[i], (2 * (filt_lvl + 2) + block_inside_limit), + SIMD_WIDTH); + } +} + +void vp8_loop_filter_init(VP8_COMMON *cm) +{ + loop_filter_info_n *lfi = &cm->lf_info; + int i; + + /* init limits for given sharpness*/ + vp8_loop_filter_update_sharpness(lfi, cm->sharpness_level); + cm->last_sharpness_level = cm->sharpness_level; + + /* init LUT for lvl and hev thr picking */ + lf_init_lut(lfi); + + /* init hev threshold const vectors */ + for(i = 0; i < 4 ; i++) + { + memset(lfi->hev_thr[i], i, SIMD_WIDTH); + } +} + +void vp8_loop_filter_frame_init(VP8_COMMON *cm, + MACROBLOCKD *mbd, + int default_filt_lvl) +{ + int seg, /* segment number */ + ref, /* index in ref_lf_deltas */ + mode; /* index in mode_lf_deltas */ + + loop_filter_info_n *lfi = &cm->lf_info; + + /* update limits if sharpness has changed */ + if(cm->last_sharpness_level != cm->sharpness_level) + { + vp8_loop_filter_update_sharpness(lfi, cm->sharpness_level); + cm->last_sharpness_level = cm->sharpness_level; + } + + for(seg = 0; seg < MAX_MB_SEGMENTS; seg++) + { + int lvl_seg = default_filt_lvl; + int lvl_ref, lvl_mode; + + /* Note the baseline filter values for each segment */ + if (mbd->segmentation_enabled) + { + /* Abs value */ + if (mbd->mb_segement_abs_delta == SEGMENT_ABSDATA) + { + lvl_seg = mbd->segment_feature_data[MB_LVL_ALT_LF][seg]; + } + else /* Delta Value */ + { + lvl_seg += mbd->segment_feature_data[MB_LVL_ALT_LF][seg]; + } + lvl_seg = (lvl_seg > 0) ? ((lvl_seg > 63) ? 63: lvl_seg) : 0; + } + + if (!mbd->mode_ref_lf_delta_enabled) + { + /* we could get rid of this if we assume that deltas are set to + * zero when not in use; encoder always uses deltas + */ + memset(lfi->lvl[seg][0], lvl_seg, 4 * 4 ); + continue; + } + + /* INTRA_FRAME */ + ref = INTRA_FRAME; + + /* Apply delta for reference frame */ + lvl_ref = lvl_seg + mbd->ref_lf_deltas[ref]; + + /* Apply delta for Intra modes */ + mode = 0; /* B_PRED */ + /* Only the split mode BPRED has a further special case */ + lvl_mode = lvl_ref + mbd->mode_lf_deltas[mode]; + /* clamp */ + lvl_mode = (lvl_mode > 0) ? (lvl_mode > 63 ? 63 : lvl_mode) : 0; + + lfi->lvl[seg][ref][mode] = lvl_mode; + + mode = 1; /* all the rest of Intra modes */ + /* clamp */ + lvl_mode = (lvl_ref > 0) ? (lvl_ref > 63 ? 63 : lvl_ref) : 0; + lfi->lvl[seg][ref][mode] = lvl_mode; + + /* LAST, GOLDEN, ALT */ + for(ref = 1; ref < MAX_REF_FRAMES; ref++) + { + /* Apply delta for reference frame */ + lvl_ref = lvl_seg + mbd->ref_lf_deltas[ref]; + + /* Apply delta for Inter modes */ + for (mode = 1; mode < 4; mode++) + { + lvl_mode = lvl_ref + mbd->mode_lf_deltas[mode]; + /* clamp */ + lvl_mode = (lvl_mode > 0) ? (lvl_mode > 63 ? 63 : lvl_mode) : 0; + + lfi->lvl[seg][ref][mode] = lvl_mode; + } + } + } +} + + +void vp8_loop_filter_row_normal(VP8_COMMON *cm, MODE_INFO *mode_info_context, + int mb_row, int post_ystride, int post_uvstride, + unsigned char *y_ptr, unsigned char *u_ptr, + unsigned char *v_ptr) +{ + int mb_col; + int filter_level; + loop_filter_info_n *lfi_n = &cm->lf_info; + loop_filter_info lfi; + FRAME_TYPE frame_type = cm->frame_type; + + for (mb_col = 0; mb_col < cm->mb_cols; mb_col++) + { + int skip_lf = (mode_info_context->mbmi.mode != B_PRED && + mode_info_context->mbmi.mode != SPLITMV && + mode_info_context->mbmi.mb_skip_coeff); + + const int mode_index = lfi_n->mode_lf_lut[mode_info_context->mbmi.mode]; + const int seg = mode_info_context->mbmi.segment_id; + const int ref_frame = mode_info_context->mbmi.ref_frame; + + filter_level = lfi_n->lvl[seg][ref_frame][mode_index]; + + if (filter_level) + { + const int hev_index = lfi_n->hev_thr_lut[frame_type][filter_level]; + lfi.mblim = lfi_n->mblim[filter_level]; + lfi.blim = lfi_n->blim[filter_level]; + lfi.lim = lfi_n->lim[filter_level]; + lfi.hev_thr = lfi_n->hev_thr[hev_index]; + + if (mb_col > 0) + vp8_loop_filter_mbv + (y_ptr, u_ptr, v_ptr, post_ystride, post_uvstride, &lfi); + + if (!skip_lf) + vp8_loop_filter_bv + (y_ptr, u_ptr, v_ptr, post_ystride, post_uvstride, &lfi); + + /* don't apply across umv border */ + if (mb_row > 0) + vp8_loop_filter_mbh + (y_ptr, u_ptr, v_ptr, post_ystride, post_uvstride, &lfi); + + if (!skip_lf) + vp8_loop_filter_bh + (y_ptr, u_ptr, v_ptr, post_ystride, post_uvstride, &lfi); + } + + y_ptr += 16; + u_ptr += 8; + v_ptr += 8; + + mode_info_context++; /* step to next MB */ + } + +} + +void vp8_loop_filter_row_simple(VP8_COMMON *cm, MODE_INFO *mode_info_context, + int mb_row, int post_ystride, int post_uvstride, + unsigned char *y_ptr, unsigned char *u_ptr, + unsigned char *v_ptr) +{ + int mb_col; + int filter_level; + loop_filter_info_n *lfi_n = &cm->lf_info; + (void)post_uvstride; + + for (mb_col = 0; mb_col < cm->mb_cols; mb_col++) + { + int skip_lf = (mode_info_context->mbmi.mode != B_PRED && + mode_info_context->mbmi.mode != SPLITMV && + mode_info_context->mbmi.mb_skip_coeff); + + const int mode_index = lfi_n->mode_lf_lut[mode_info_context->mbmi.mode]; + const int seg = mode_info_context->mbmi.segment_id; + const int ref_frame = mode_info_context->mbmi.ref_frame; + + filter_level = lfi_n->lvl[seg][ref_frame][mode_index]; + + if (filter_level) + { + if (mb_col > 0) + vp8_loop_filter_simple_mbv + (y_ptr, post_ystride, lfi_n->mblim[filter_level]); + + if (!skip_lf) + vp8_loop_filter_simple_bv + (y_ptr, post_ystride, lfi_n->blim[filter_level]); + + /* don't apply across umv border */ + if (mb_row > 0) + vp8_loop_filter_simple_mbh + (y_ptr, post_ystride, lfi_n->mblim[filter_level]); + + if (!skip_lf) + vp8_loop_filter_simple_bh + (y_ptr, post_ystride, lfi_n->blim[filter_level]); + } + + y_ptr += 16; + u_ptr += 8; + v_ptr += 8; + + mode_info_context++; /* step to next MB */ + } + +} +void vp8_loop_filter_frame(VP8_COMMON *cm, + MACROBLOCKD *mbd, + int frame_type) +{ + YV12_BUFFER_CONFIG *post = cm->frame_to_show; + loop_filter_info_n *lfi_n = &cm->lf_info; + loop_filter_info lfi; + + int mb_row; + int mb_col; + int mb_rows = cm->mb_rows; + int mb_cols = cm->mb_cols; + + int filter_level; + + unsigned char *y_ptr, *u_ptr, *v_ptr; + + /* Point at base of Mb MODE_INFO list */ + const MODE_INFO *mode_info_context = cm->mi; + int post_y_stride = post->y_stride; + int post_uv_stride = post->uv_stride; + + /* Initialize the loop filter for this frame. */ + vp8_loop_filter_frame_init(cm, mbd, cm->filter_level); + + /* Set up the buffer pointers */ + y_ptr = post->y_buffer; + u_ptr = post->u_buffer; + v_ptr = post->v_buffer; + + /* vp8_filter each macro block */ + if (cm->filter_type == NORMAL_LOOPFILTER) + { + for (mb_row = 0; mb_row < mb_rows; mb_row++) + { + for (mb_col = 0; mb_col < mb_cols; mb_col++) + { + int skip_lf = (mode_info_context->mbmi.mode != B_PRED && + mode_info_context->mbmi.mode != SPLITMV && + mode_info_context->mbmi.mb_skip_coeff); + + const int mode_index = lfi_n->mode_lf_lut[mode_info_context->mbmi.mode]; + const int seg = mode_info_context->mbmi.segment_id; + const int ref_frame = mode_info_context->mbmi.ref_frame; + + filter_level = lfi_n->lvl[seg][ref_frame][mode_index]; + + if (filter_level) + { + const int hev_index = lfi_n->hev_thr_lut[frame_type][filter_level]; + lfi.mblim = lfi_n->mblim[filter_level]; + lfi.blim = lfi_n->blim[filter_level]; + lfi.lim = lfi_n->lim[filter_level]; + lfi.hev_thr = lfi_n->hev_thr[hev_index]; + + if (mb_col > 0) + vp8_loop_filter_mbv + (y_ptr, u_ptr, v_ptr, post_y_stride, post_uv_stride, &lfi); + + if (!skip_lf) + vp8_loop_filter_bv + (y_ptr, u_ptr, v_ptr, post_y_stride, post_uv_stride, &lfi); + + /* don't apply across umv border */ + if (mb_row > 0) + vp8_loop_filter_mbh + (y_ptr, u_ptr, v_ptr, post_y_stride, post_uv_stride, &lfi); + + if (!skip_lf) + vp8_loop_filter_bh + (y_ptr, u_ptr, v_ptr, post_y_stride, post_uv_stride, &lfi); + } + + y_ptr += 16; + u_ptr += 8; + v_ptr += 8; + + mode_info_context++; /* step to next MB */ + } + y_ptr += post_y_stride * 16 - post->y_width; + u_ptr += post_uv_stride * 8 - post->uv_width; + v_ptr += post_uv_stride * 8 - post->uv_width; + + mode_info_context++; /* Skip border mb */ + + } + } + else /* SIMPLE_LOOPFILTER */ + { + for (mb_row = 0; mb_row < mb_rows; mb_row++) + { + for (mb_col = 0; mb_col < mb_cols; mb_col++) + { + int skip_lf = (mode_info_context->mbmi.mode != B_PRED && + mode_info_context->mbmi.mode != SPLITMV && + mode_info_context->mbmi.mb_skip_coeff); + + const int mode_index = lfi_n->mode_lf_lut[mode_info_context->mbmi.mode]; + const int seg = mode_info_context->mbmi.segment_id; + const int ref_frame = mode_info_context->mbmi.ref_frame; + + filter_level = lfi_n->lvl[seg][ref_frame][mode_index]; + if (filter_level) + { + const unsigned char * mblim = lfi_n->mblim[filter_level]; + const unsigned char * blim = lfi_n->blim[filter_level]; + + if (mb_col > 0) + vp8_loop_filter_simple_mbv + (y_ptr, post_y_stride, mblim); + + if (!skip_lf) + vp8_loop_filter_simple_bv + (y_ptr, post_y_stride, blim); + + /* don't apply across umv border */ + if (mb_row > 0) + vp8_loop_filter_simple_mbh + (y_ptr, post_y_stride, mblim); + + if (!skip_lf) + vp8_loop_filter_simple_bh + (y_ptr, post_y_stride, blim); + } + + y_ptr += 16; + u_ptr += 8; + v_ptr += 8; + + mode_info_context++; /* step to next MB */ + } + y_ptr += post_y_stride * 16 - post->y_width; + u_ptr += post_uv_stride * 8 - post->uv_width; + v_ptr += post_uv_stride * 8 - post->uv_width; + + mode_info_context++; /* Skip border mb */ + + } + } +} + +void vp8_loop_filter_frame_yonly +( + VP8_COMMON *cm, + MACROBLOCKD *mbd, + int default_filt_lvl +) +{ + YV12_BUFFER_CONFIG *post = cm->frame_to_show; + + unsigned char *y_ptr; + int mb_row; + int mb_col; + + loop_filter_info_n *lfi_n = &cm->lf_info; + loop_filter_info lfi; + + int filter_level; + FRAME_TYPE frame_type = cm->frame_type; + + /* Point at base of Mb MODE_INFO list */ + const MODE_INFO *mode_info_context = cm->mi; + +#if 0 + if(default_filt_lvl == 0) /* no filter applied */ + return; +#endif + + /* Initialize the loop filter for this frame. */ + vp8_loop_filter_frame_init( cm, mbd, default_filt_lvl); + + /* Set up the buffer pointers */ + y_ptr = post->y_buffer; + + /* vp8_filter each macro block */ + for (mb_row = 0; mb_row < cm->mb_rows; mb_row++) + { + for (mb_col = 0; mb_col < cm->mb_cols; mb_col++) + { + int skip_lf = (mode_info_context->mbmi.mode != B_PRED && + mode_info_context->mbmi.mode != SPLITMV && + mode_info_context->mbmi.mb_skip_coeff); + + const int mode_index = lfi_n->mode_lf_lut[mode_info_context->mbmi.mode]; + const int seg = mode_info_context->mbmi.segment_id; + const int ref_frame = mode_info_context->mbmi.ref_frame; + + filter_level = lfi_n->lvl[seg][ref_frame][mode_index]; + + if (filter_level) + { + if (cm->filter_type == NORMAL_LOOPFILTER) + { + const int hev_index = lfi_n->hev_thr_lut[frame_type][filter_level]; + lfi.mblim = lfi_n->mblim[filter_level]; + lfi.blim = lfi_n->blim[filter_level]; + lfi.lim = lfi_n->lim[filter_level]; + lfi.hev_thr = lfi_n->hev_thr[hev_index]; + + if (mb_col > 0) + vp8_loop_filter_mbv + (y_ptr, 0, 0, post->y_stride, 0, &lfi); + + if (!skip_lf) + vp8_loop_filter_bv + (y_ptr, 0, 0, post->y_stride, 0, &lfi); + + /* don't apply across umv border */ + if (mb_row > 0) + vp8_loop_filter_mbh + (y_ptr, 0, 0, post->y_stride, 0, &lfi); + + if (!skip_lf) + vp8_loop_filter_bh + (y_ptr, 0, 0, post->y_stride, 0, &lfi); + } + else + { + if (mb_col > 0) + vp8_loop_filter_simple_mbv + (y_ptr, post->y_stride, lfi_n->mblim[filter_level]); + + if (!skip_lf) + vp8_loop_filter_simple_bv + (y_ptr, post->y_stride, lfi_n->blim[filter_level]); + + /* don't apply across umv border */ + if (mb_row > 0) + vp8_loop_filter_simple_mbh + (y_ptr, post->y_stride, lfi_n->mblim[filter_level]); + + if (!skip_lf) + vp8_loop_filter_simple_bh + (y_ptr, post->y_stride, lfi_n->blim[filter_level]); + } + } + + y_ptr += 16; + mode_info_context ++; /* step to next MB */ + + } + + y_ptr += post->y_stride * 16 - post->y_width; + mode_info_context ++; /* Skip border mb */ + } + +} + +void vp8_loop_filter_partial_frame +( + VP8_COMMON *cm, + MACROBLOCKD *mbd, + int default_filt_lvl +) +{ + YV12_BUFFER_CONFIG *post = cm->frame_to_show; + + unsigned char *y_ptr; + int mb_row; + int mb_col; + int mb_cols = post->y_width >> 4; + int mb_rows = post->y_height >> 4; + + int linestocopy; + + loop_filter_info_n *lfi_n = &cm->lf_info; + loop_filter_info lfi; + + int filter_level; + FRAME_TYPE frame_type = cm->frame_type; + + const MODE_INFO *mode_info_context; + +#if 0 + if(default_filt_lvl == 0) /* no filter applied */ + return; +#endif + + /* Initialize the loop filter for this frame. */ + vp8_loop_filter_frame_init( cm, mbd, default_filt_lvl); + + /* number of MB rows to use in partial filtering */ + linestocopy = mb_rows / PARTIAL_FRAME_FRACTION; + linestocopy = linestocopy ? linestocopy << 4 : 16; /* 16 lines per MB */ + + /* Set up the buffer pointers; partial image starts at ~middle of frame */ + y_ptr = post->y_buffer + ((post->y_height >> 5) * 16) * post->y_stride; + mode_info_context = cm->mi + (post->y_height >> 5) * (mb_cols + 1); + + /* vp8_filter each macro block */ + for (mb_row = 0; mb_row<(linestocopy >> 4); mb_row++) + { + for (mb_col = 0; mb_col < mb_cols; mb_col++) + { + int skip_lf = (mode_info_context->mbmi.mode != B_PRED && + mode_info_context->mbmi.mode != SPLITMV && + mode_info_context->mbmi.mb_skip_coeff); + + const int mode_index = + lfi_n->mode_lf_lut[mode_info_context->mbmi.mode]; + const int seg = mode_info_context->mbmi.segment_id; + const int ref_frame = mode_info_context->mbmi.ref_frame; + + filter_level = lfi_n->lvl[seg][ref_frame][mode_index]; + + if (filter_level) + { + if (cm->filter_type == NORMAL_LOOPFILTER) + { + const int hev_index = lfi_n->hev_thr_lut[frame_type][filter_level]; + lfi.mblim = lfi_n->mblim[filter_level]; + lfi.blim = lfi_n->blim[filter_level]; + lfi.lim = lfi_n->lim[filter_level]; + lfi.hev_thr = lfi_n->hev_thr[hev_index]; + + if (mb_col > 0) + vp8_loop_filter_mbv + (y_ptr, 0, 0, post->y_stride, 0, &lfi); + + if (!skip_lf) + vp8_loop_filter_bv + (y_ptr, 0, 0, post->y_stride, 0, &lfi); + + vp8_loop_filter_mbh + (y_ptr, 0, 0, post->y_stride, 0, &lfi); + + if (!skip_lf) + vp8_loop_filter_bh + (y_ptr, 0, 0, post->y_stride, 0, &lfi); + } + else + { + if (mb_col > 0) + vp8_loop_filter_simple_mbv + (y_ptr, post->y_stride, lfi_n->mblim[filter_level]); + + if (!skip_lf) + vp8_loop_filter_simple_bv + (y_ptr, post->y_stride, lfi_n->blim[filter_level]); + + vp8_loop_filter_simple_mbh + (y_ptr, post->y_stride, lfi_n->mblim[filter_level]); + + if (!skip_lf) + vp8_loop_filter_simple_bh + (y_ptr, post->y_stride, lfi_n->blim[filter_level]); + } + } + + y_ptr += 16; + mode_info_context += 1; /* step to next MB */ + } + + y_ptr += post->y_stride * 16 - post->y_width; + mode_info_context += 1; /* Skip border mb */ + } +} diff --git a/thirdparty/libvpx/vp8/common/x86/copy_sse2.asm b/thirdparty/libvpx/vp8/common/x86/copy_sse2.asm new file mode 100644 index 0000000000..86fae26956 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/x86/copy_sse2.asm @@ -0,0 +1,93 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + + +%include "vpx_ports/x86_abi_support.asm" + + +;void vp8_copy32xn_sse2( +; unsigned char *src_ptr, +; int src_stride, +; unsigned char *dst_ptr, +; int dst_stride, +; int height); +global sym(vp8_copy32xn_sse2) PRIVATE +sym(vp8_copy32xn_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 5 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + mov rsi, arg(0) ;src_ptr + mov rdi, arg(2) ;dst_ptr + + movsxd rax, dword ptr arg(1) ;src_stride + movsxd rdx, dword ptr arg(3) ;dst_stride + movsxd rcx, dword ptr arg(4) ;height + +.block_copy_sse2_loopx4: + movdqu xmm0, XMMWORD PTR [rsi] + movdqu xmm1, XMMWORD PTR [rsi + 16] + movdqu xmm2, XMMWORD PTR [rsi + rax] + movdqu xmm3, XMMWORD PTR [rsi + rax + 16] + + lea rsi, [rsi+rax*2] + + movdqu xmm4, XMMWORD PTR [rsi] + movdqu xmm5, XMMWORD PTR [rsi + 16] + movdqu xmm6, XMMWORD PTR [rsi + rax] + movdqu xmm7, XMMWORD PTR [rsi + rax + 16] + + lea rsi, [rsi+rax*2] + + movdqa XMMWORD PTR [rdi], xmm0 + movdqa XMMWORD PTR [rdi + 16], xmm1 + movdqa XMMWORD PTR [rdi + rdx], xmm2 + movdqa XMMWORD PTR [rdi + rdx + 16], xmm3 + + lea rdi, [rdi+rdx*2] + + movdqa XMMWORD PTR [rdi], xmm4 + movdqa XMMWORD PTR [rdi + 16], xmm5 + movdqa XMMWORD PTR [rdi + rdx], xmm6 + movdqa XMMWORD PTR [rdi + rdx + 16], xmm7 + + lea rdi, [rdi+rdx*2] + + sub rcx, 4 + cmp rcx, 4 + jge .block_copy_sse2_loopx4 + + cmp rcx, 0 + je .copy_is_done + +.block_copy_sse2_loop: + movdqu xmm0, XMMWORD PTR [rsi] + movdqu xmm1, XMMWORD PTR [rsi + 16] + lea rsi, [rsi+rax] + + movdqa XMMWORD PTR [rdi], xmm0 + movdqa XMMWORD PTR [rdi + 16], xmm1 + lea rdi, [rdi+rdx] + + sub rcx, 1 + jne .block_copy_sse2_loop + +.copy_is_done: + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret diff --git a/thirdparty/libvpx/vp8/common/x86/copy_sse3.asm b/thirdparty/libvpx/vp8/common/x86/copy_sse3.asm new file mode 100644 index 0000000000..d789a40ccf --- /dev/null +++ b/thirdparty/libvpx/vp8/common/x86/copy_sse3.asm @@ -0,0 +1,146 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + +%include "vpx_ports/x86_abi_support.asm" + +%macro STACK_FRAME_CREATE_X3 0 +%if ABI_IS_32BIT + %define src_ptr rsi + %define src_stride rax + %define ref_ptr rdi + %define ref_stride rdx + %define end_ptr rcx + %define ret_var rbx + %define result_ptr arg(4) + %define max_sad arg(4) + %define height dword ptr arg(4) + push rbp + mov rbp, rsp + push rsi + push rdi + push rbx + + mov rsi, arg(0) ; src_ptr + mov rdi, arg(2) ; ref_ptr + + movsxd rax, dword ptr arg(1) ; src_stride + movsxd rdx, dword ptr arg(3) ; ref_stride +%else + %if LIBVPX_YASM_WIN64 + SAVE_XMM 7, u + %define src_ptr rcx + %define src_stride rdx + %define ref_ptr r8 + %define ref_stride r9 + %define end_ptr r10 + %define ret_var r11 + %define result_ptr [rsp+xmm_stack_space+8+4*8] + %define max_sad [rsp+xmm_stack_space+8+4*8] + %define height dword ptr [rsp+xmm_stack_space+8+4*8] + %else + %define src_ptr rdi + %define src_stride rsi + %define ref_ptr rdx + %define ref_stride rcx + %define end_ptr r9 + %define ret_var r10 + %define result_ptr r8 + %define max_sad r8 + %define height r8 + %endif +%endif + +%endmacro + +%macro STACK_FRAME_DESTROY_X3 0 + %define src_ptr + %define src_stride + %define ref_ptr + %define ref_stride + %define end_ptr + %define ret_var + %define result_ptr + %define max_sad + %define height + +%if ABI_IS_32BIT + pop rbx + pop rdi + pop rsi + pop rbp +%else + %if LIBVPX_YASM_WIN64 + RESTORE_XMM + %endif +%endif + ret +%endmacro + + +;void vp8_copy32xn_sse3( +; unsigned char *src_ptr, +; int src_stride, +; unsigned char *dst_ptr, +; int dst_stride, +; int height); +global sym(vp8_copy32xn_sse3) PRIVATE +sym(vp8_copy32xn_sse3): + + STACK_FRAME_CREATE_X3 + +.block_copy_sse3_loopx4: + lea end_ptr, [src_ptr+src_stride*2] + + movdqu xmm0, XMMWORD PTR [src_ptr] + movdqu xmm1, XMMWORD PTR [src_ptr + 16] + movdqu xmm2, XMMWORD PTR [src_ptr + src_stride] + movdqu xmm3, XMMWORD PTR [src_ptr + src_stride + 16] + movdqu xmm4, XMMWORD PTR [end_ptr] + movdqu xmm5, XMMWORD PTR [end_ptr + 16] + movdqu xmm6, XMMWORD PTR [end_ptr + src_stride] + movdqu xmm7, XMMWORD PTR [end_ptr + src_stride + 16] + + lea src_ptr, [src_ptr+src_stride*4] + + lea end_ptr, [ref_ptr+ref_stride*2] + + movdqa XMMWORD PTR [ref_ptr], xmm0 + movdqa XMMWORD PTR [ref_ptr + 16], xmm1 + movdqa XMMWORD PTR [ref_ptr + ref_stride], xmm2 + movdqa XMMWORD PTR [ref_ptr + ref_stride + 16], xmm3 + movdqa XMMWORD PTR [end_ptr], xmm4 + movdqa XMMWORD PTR [end_ptr + 16], xmm5 + movdqa XMMWORD PTR [end_ptr + ref_stride], xmm6 + movdqa XMMWORD PTR [end_ptr + ref_stride + 16], xmm7 + + lea ref_ptr, [ref_ptr+ref_stride*4] + + sub height, 4 + cmp height, 4 + jge .block_copy_sse3_loopx4 + + ;Check to see if there is more rows need to be copied. + cmp height, 0 + je .copy_is_done + +.block_copy_sse3_loop: + movdqu xmm0, XMMWORD PTR [src_ptr] + movdqu xmm1, XMMWORD PTR [src_ptr + 16] + lea src_ptr, [src_ptr+src_stride] + + movdqa XMMWORD PTR [ref_ptr], xmm0 + movdqa XMMWORD PTR [ref_ptr + 16], xmm1 + lea ref_ptr, [ref_ptr+ref_stride] + + sub height, 1 + jne .block_copy_sse3_loop + +.copy_is_done: + STACK_FRAME_DESTROY_X3 diff --git a/thirdparty/libvpx/vp8/common/x86/dequantize_mmx.asm b/thirdparty/libvpx/vp8/common/x86/dequantize_mmx.asm new file mode 100644 index 0000000000..4e551f00aa --- /dev/null +++ b/thirdparty/libvpx/vp8/common/x86/dequantize_mmx.asm @@ -0,0 +1,258 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + + +%include "vpx_ports/x86_abi_support.asm" + + +;void vp8_dequantize_b_impl_mmx(short *sq, short *dq, short *q) +global sym(vp8_dequantize_b_impl_mmx) PRIVATE +sym(vp8_dequantize_b_impl_mmx): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 3 + push rsi + push rdi + ; end prolog + + mov rsi, arg(0) ;sq + mov rdi, arg(1) ;dq + mov rax, arg(2) ;q + + movq mm1, [rsi] + pmullw mm1, [rax+0] ; mm4 *= kernel 0 modifiers. + movq [rdi], mm1 + + movq mm1, [rsi+8] + pmullw mm1, [rax+8] ; mm4 *= kernel 0 modifiers. + movq [rdi+8], mm1 + + movq mm1, [rsi+16] + pmullw mm1, [rax+16] ; mm4 *= kernel 0 modifiers. + movq [rdi+16], mm1 + + movq mm1, [rsi+24] + pmullw mm1, [rax+24] ; mm4 *= kernel 0 modifiers. + movq [rdi+24], mm1 + + ; begin epilog + pop rdi + pop rsi + UNSHADOW_ARGS + pop rbp + ret + + +;void dequant_idct_add_mmx( +;short *input, 0 +;short *dq, 1 +;unsigned char *dest, 2 +;int stride) 3 +global sym(vp8_dequant_idct_add_mmx) PRIVATE +sym(vp8_dequant_idct_add_mmx): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 4 + GET_GOT rbx + push rdi + ; end prolog + + mov rax, arg(0) ;input + mov rdx, arg(1) ;dq + + + movq mm0, [rax ] + pmullw mm0, [rdx] + + movq mm1, [rax +8] + pmullw mm1, [rdx +8] + + movq mm2, [rax+16] + pmullw mm2, [rdx+16] + + movq mm3, [rax+24] + pmullw mm3, [rdx+24] + + mov rdx, arg(2) ;dest + + pxor mm7, mm7 + + + movq [rax], mm7 + movq [rax+8], mm7 + + movq [rax+16],mm7 + movq [rax+24],mm7 + + + movsxd rdi, dword ptr arg(3) ;stride + + psubw mm0, mm2 ; b1= 0-2 + paddw mm2, mm2 ; + + movq mm5, mm1 + paddw mm2, mm0 ; a1 =0+2 + + pmulhw mm5, [GLOBAL(x_s1sqr2)]; + paddw mm5, mm1 ; ip1 * sin(pi/8) * sqrt(2) + + movq mm7, mm3 ; + pmulhw mm7, [GLOBAL(x_c1sqr2less1)]; + + paddw mm7, mm3 ; ip3 * cos(pi/8) * sqrt(2) + psubw mm7, mm5 ; c1 + + movq mm5, mm1 + movq mm4, mm3 + + pmulhw mm5, [GLOBAL(x_c1sqr2less1)] + paddw mm5, mm1 + + pmulhw mm3, [GLOBAL(x_s1sqr2)] + paddw mm3, mm4 + + paddw mm3, mm5 ; d1 + movq mm6, mm2 ; a1 + + movq mm4, mm0 ; b1 + paddw mm2, mm3 ;0 + + paddw mm4, mm7 ;1 + psubw mm0, mm7 ;2 + + psubw mm6, mm3 ;3 + + movq mm1, mm2 ; 03 02 01 00 + movq mm3, mm4 ; 23 22 21 20 + + punpcklwd mm1, mm0 ; 11 01 10 00 + punpckhwd mm2, mm0 ; 13 03 12 02 + + punpcklwd mm3, mm6 ; 31 21 30 20 + punpckhwd mm4, mm6 ; 33 23 32 22 + + movq mm0, mm1 ; 11 01 10 00 + movq mm5, mm2 ; 13 03 12 02 + + punpckldq mm0, mm3 ; 30 20 10 00 + punpckhdq mm1, mm3 ; 31 21 11 01 + + punpckldq mm2, mm4 ; 32 22 12 02 + punpckhdq mm5, mm4 ; 33 23 13 03 + + movq mm3, mm5 ; 33 23 13 03 + + psubw mm0, mm2 ; b1= 0-2 + paddw mm2, mm2 ; + + movq mm5, mm1 + paddw mm2, mm0 ; a1 =0+2 + + pmulhw mm5, [GLOBAL(x_s1sqr2)]; + paddw mm5, mm1 ; ip1 * sin(pi/8) * sqrt(2) + + movq mm7, mm3 ; + pmulhw mm7, [GLOBAL(x_c1sqr2less1)]; + + paddw mm7, mm3 ; ip3 * cos(pi/8) * sqrt(2) + psubw mm7, mm5 ; c1 + + movq mm5, mm1 + movq mm4, mm3 + + pmulhw mm5, [GLOBAL(x_c1sqr2less1)] + paddw mm5, mm1 + + pmulhw mm3, [GLOBAL(x_s1sqr2)] + paddw mm3, mm4 + + paddw mm3, mm5 ; d1 + paddw mm0, [GLOBAL(fours)] + + paddw mm2, [GLOBAL(fours)] + movq mm6, mm2 ; a1 + + movq mm4, mm0 ; b1 + paddw mm2, mm3 ;0 + + paddw mm4, mm7 ;1 + psubw mm0, mm7 ;2 + + psubw mm6, mm3 ;3 + psraw mm2, 3 + + psraw mm0, 3 + psraw mm4, 3 + + psraw mm6, 3 + + movq mm1, mm2 ; 03 02 01 00 + movq mm3, mm4 ; 23 22 21 20 + + punpcklwd mm1, mm0 ; 11 01 10 00 + punpckhwd mm2, mm0 ; 13 03 12 02 + + punpcklwd mm3, mm6 ; 31 21 30 20 + punpckhwd mm4, mm6 ; 33 23 32 22 + + movq mm0, mm1 ; 11 01 10 00 + movq mm5, mm2 ; 13 03 12 02 + + punpckldq mm0, mm3 ; 30 20 10 00 + punpckhdq mm1, mm3 ; 31 21 11 01 + + punpckldq mm2, mm4 ; 32 22 12 02 + punpckhdq mm5, mm4 ; 33 23 13 03 + + pxor mm7, mm7 + + movd mm4, [rdx] + punpcklbw mm4, mm7 + paddsw mm0, mm4 + packuswb mm0, mm7 + movd [rdx], mm0 + + movd mm4, [rdx+rdi] + punpcklbw mm4, mm7 + paddsw mm1, mm4 + packuswb mm1, mm7 + movd [rdx+rdi], mm1 + + movd mm4, [rdx+2*rdi] + punpcklbw mm4, mm7 + paddsw mm2, mm4 + packuswb mm2, mm7 + movd [rdx+rdi*2], mm2 + + add rdx, rdi + + movd mm4, [rdx+2*rdi] + punpcklbw mm4, mm7 + paddsw mm5, mm4 + packuswb mm5, mm7 + movd [rdx+rdi*2], mm5 + + ; begin epilog + pop rdi + RESTORE_GOT + UNSHADOW_ARGS + pop rbp + ret + +SECTION_RODATA +align 16 +x_s1sqr2: + times 4 dw 0x8A8C +align 16 +x_c1sqr2less1: + times 4 dw 0x4E7B +align 16 +fours: + times 4 dw 0x0004 diff --git a/thirdparty/libvpx/vp8/common/x86/filter_x86.c b/thirdparty/libvpx/vp8/common/x86/filter_x86.c new file mode 100644 index 0000000000..7f496ed7db --- /dev/null +++ b/thirdparty/libvpx/vp8/common/x86/filter_x86.c @@ -0,0 +1,35 @@ +/* + * Copyright (c) 2011 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "vp8/common/x86/filter_x86.h" + +DECLARE_ALIGNED(16, const short, vp8_bilinear_filters_x86_4[8][8]) = +{ + { 128, 128, 128, 128, 0, 0, 0, 0 }, + { 112, 112, 112, 112, 16, 16, 16, 16 }, + { 96, 96, 96, 96, 32, 32, 32, 32 }, + { 80, 80, 80, 80, 48, 48, 48, 48 }, + { 64, 64, 64, 64, 64, 64, 64, 64 }, + { 48, 48, 48, 48, 80, 80, 80, 80 }, + { 32, 32, 32, 32, 96, 96, 96, 96 }, + { 16, 16, 16, 16, 112, 112, 112, 112 } +}; + +DECLARE_ALIGNED(16, const short, vp8_bilinear_filters_x86_8[8][16]) = +{ + { 128, 128, 128, 128, 128, 128, 128, 128, 0, 0, 0, 0, 0, 0, 0, 0 }, + { 112, 112, 112, 112, 112, 112, 112, 112, 16, 16, 16, 16, 16, 16, 16, 16 }, + { 96, 96, 96, 96, 96, 96, 96, 96, 32, 32, 32, 32, 32, 32, 32, 32 }, + { 80, 80, 80, 80, 80, 80, 80, 80, 48, 48, 48, 48, 48, 48, 48, 48 }, + { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 }, + { 48, 48, 48, 48, 48, 48, 48, 48, 80, 80, 80, 80, 80, 80, 80, 80 }, + { 32, 32, 32, 32, 32, 32, 32, 32, 96, 96, 96, 96, 96, 96, 96, 96 }, + { 16, 16, 16, 16, 16, 16, 16, 16, 112, 112, 112, 112, 112, 112, 112, 112 } +}; diff --git a/thirdparty/libvpx/vp8/common/x86/filter_x86.h b/thirdparty/libvpx/vp8/common/x86/filter_x86.h new file mode 100644 index 0000000000..d282841bee --- /dev/null +++ b/thirdparty/libvpx/vp8/common/x86/filter_x86.h @@ -0,0 +1,33 @@ +/* + * Copyright (c) 2011 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP8_COMMON_X86_FILTER_X86_H_ +#define VP8_COMMON_X86_FILTER_X86_H_ + +#include "vpx_ports/mem.h" + +#ifdef __cplusplus +extern "C" { +#endif + +/* x86 assembly specific copy of vp8/common/filter.c:vp8_bilinear_filters with + * duplicated values */ + +/* duplicated 4x */ +extern DECLARE_ALIGNED(16, const short, vp8_bilinear_filters_x86_4[8][8]); + +/* duplicated 8x */ +extern DECLARE_ALIGNED(16, const short, vp8_bilinear_filters_x86_8[8][16]); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_X86_FILTER_X86_H_ diff --git a/thirdparty/libvpx/vp8/common/x86/idct_blk_mmx.c b/thirdparty/libvpx/vp8/common/x86/idct_blk_mmx.c new file mode 100644 index 0000000000..f2532b34da --- /dev/null +++ b/thirdparty/libvpx/vp8/common/x86/idct_blk_mmx.c @@ -0,0 +1,128 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "vpx_config.h" +#include "vp8_rtcd.h" +#include "vp8/common/blockd.h" +#include "vpx_mem/vpx_mem.h" + +extern void vp8_dequantize_b_impl_mmx(short *sq, short *dq, short *q); + +void vp8_dequantize_b_mmx(BLOCKD *d, short *DQC) +{ + short *sq = (short *) d->qcoeff; + short *dq = (short *) d->dqcoeff; + + vp8_dequantize_b_impl_mmx(sq, dq, DQC); +} + +void vp8_dequant_idct_add_y_block_mmx + (short *q, short *dq, + unsigned char *dst, int stride, char *eobs) +{ + int i; + + for (i = 0; i < 4; i++) + { + if (eobs[0] > 1) + vp8_dequant_idct_add_mmx (q, dq, dst, stride); + else if (eobs[0] == 1) + { + vp8_dc_only_idct_add_mmx (q[0]*dq[0], dst, stride, dst, stride); + memset(q, 0, 2 * sizeof(q[0])); + } + + if (eobs[1] > 1) + vp8_dequant_idct_add_mmx (q+16, dq, dst+4, stride); + else if (eobs[1] == 1) + { + vp8_dc_only_idct_add_mmx (q[16]*dq[0], dst+4, stride, + dst+4, stride); + memset(q + 16, 0, 2 * sizeof(q[0])); + } + + if (eobs[2] > 1) + vp8_dequant_idct_add_mmx (q+32, dq, dst+8, stride); + else if (eobs[2] == 1) + { + vp8_dc_only_idct_add_mmx (q[32]*dq[0], dst+8, stride, + dst+8, stride); + memset(q + 32, 0, 2 * sizeof(q[0])); + } + + if (eobs[3] > 1) + vp8_dequant_idct_add_mmx (q+48, dq, dst+12, stride); + else if (eobs[3] == 1) + { + vp8_dc_only_idct_add_mmx (q[48]*dq[0], dst+12, stride, + dst+12, stride); + memset(q + 48, 0, 2 * sizeof(q[0])); + } + + q += 64; + dst += 4*stride; + eobs += 4; + } +} + +void vp8_dequant_idct_add_uv_block_mmx + (short *q, short *dq, + unsigned char *dstu, unsigned char *dstv, int stride, char *eobs) +{ + int i; + + for (i = 0; i < 2; i++) + { + if (eobs[0] > 1) + vp8_dequant_idct_add_mmx (q, dq, dstu, stride); + else if (eobs[0] == 1) + { + vp8_dc_only_idct_add_mmx (q[0]*dq[0], dstu, stride, dstu, stride); + memset(q, 0, 2 * sizeof(q[0])); + } + + if (eobs[1] > 1) + vp8_dequant_idct_add_mmx (q+16, dq, dstu+4, stride); + else if (eobs[1] == 1) + { + vp8_dc_only_idct_add_mmx (q[16]*dq[0], dstu+4, stride, + dstu+4, stride); + memset(q + 16, 0, 2 * sizeof(q[0])); + } + + q += 32; + dstu += 4*stride; + eobs += 2; + } + + for (i = 0; i < 2; i++) + { + if (eobs[0] > 1) + vp8_dequant_idct_add_mmx (q, dq, dstv, stride); + else if (eobs[0] == 1) + { + vp8_dc_only_idct_add_mmx (q[0]*dq[0], dstv, stride, dstv, stride); + memset(q, 0, 2 * sizeof(q[0])); + } + + if (eobs[1] > 1) + vp8_dequant_idct_add_mmx (q+16, dq, dstv+4, stride); + else if (eobs[1] == 1) + { + vp8_dc_only_idct_add_mmx (q[16]*dq[0], dstv+4, stride, + dstv+4, stride); + memset(q + 16, 0, 2 * sizeof(q[0])); + } + + q += 32; + dstv += 4*stride; + eobs += 2; + } +} diff --git a/thirdparty/libvpx/vp8/common/x86/idct_blk_sse2.c b/thirdparty/libvpx/vp8/common/x86/idct_blk_sse2.c new file mode 100644 index 0000000000..ae96ec858c --- /dev/null +++ b/thirdparty/libvpx/vp8/common/x86/idct_blk_sse2.c @@ -0,0 +1,89 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "vpx_config.h" +#include "vp8_rtcd.h" + +void vp8_idct_dequant_0_2x_sse2 + (short *q, short *dq , + unsigned char *dst, int dst_stride); +void vp8_idct_dequant_full_2x_sse2 + (short *q, short *dq , + unsigned char *dst, int dst_stride); + +void vp8_dequant_idct_add_y_block_sse2 + (short *q, short *dq, + unsigned char *dst, int stride, char *eobs) +{ + int i; + + for (i = 0; i < 4; i++) + { + if (((short *)(eobs))[0]) + { + if (((short *)(eobs))[0] & 0xfefe) + vp8_idct_dequant_full_2x_sse2 (q, dq, dst, stride); + else + vp8_idct_dequant_0_2x_sse2 (q, dq, dst, stride); + } + if (((short *)(eobs))[1]) + { + if (((short *)(eobs))[1] & 0xfefe) + vp8_idct_dequant_full_2x_sse2 (q+32, dq, dst+8, stride); + else + vp8_idct_dequant_0_2x_sse2 (q+32, dq, dst+8, stride); + } + q += 64; + dst += stride*4; + eobs += 4; + } +} + +void vp8_dequant_idct_add_uv_block_sse2 + (short *q, short *dq, + unsigned char *dstu, unsigned char *dstv, int stride, char *eobs) +{ + if (((short *)(eobs))[0]) + { + if (((short *)(eobs))[0] & 0xfefe) + vp8_idct_dequant_full_2x_sse2 (q, dq, dstu, stride); + else + vp8_idct_dequant_0_2x_sse2 (q, dq, dstu, stride); + } + q += 32; + dstu += stride*4; + + if (((short *)(eobs))[1]) + { + if (((short *)(eobs))[1] & 0xfefe) + vp8_idct_dequant_full_2x_sse2 (q, dq, dstu, stride); + else + vp8_idct_dequant_0_2x_sse2 (q, dq, dstu, stride); + } + q += 32; + + if (((short *)(eobs))[2]) + { + if (((short *)(eobs))[2] & 0xfefe) + vp8_idct_dequant_full_2x_sse2 (q, dq, dstv, stride); + else + vp8_idct_dequant_0_2x_sse2 (q, dq, dstv, stride); + } + q += 32; + dstv += stride*4; + + if (((short *)(eobs))[3]) + { + if (((short *)(eobs))[3] & 0xfefe) + vp8_idct_dequant_full_2x_sse2 (q, dq, dstv, stride); + else + vp8_idct_dequant_0_2x_sse2 (q, dq, dstv, stride); + } +} diff --git a/thirdparty/libvpx/vp8/common/x86/idctllm_mmx.asm b/thirdparty/libvpx/vp8/common/x86/idctllm_mmx.asm new file mode 100644 index 0000000000..96fa2c60d0 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/x86/idctllm_mmx.asm @@ -0,0 +1,295 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + + +%include "vpx_ports/x86_abi_support.asm" + +; /**************************************************************************** +; * Notes: +; * +; * This implementation makes use of 16 bit fixed point version of two multiply +; * constants: +; * 1. sqrt(2) * cos (pi/8) +; * 2. sqrt(2) * sin (pi/8) +; * Because the first constant is bigger than 1, to maintain the same 16 bit +; * fixed point precision as the second one, we use a trick of +; * x * a = x + x*(a-1) +; * so +; * x * sqrt(2) * cos (pi/8) = x + x * (sqrt(2) *cos(pi/8)-1). +; * +; * For the second constant, because of the 16bit version is 35468, which +; * is bigger than 32768, in signed 16 bit multiply, it becomes a negative +; * number. +; * (x * (unsigned)35468 >> 16) = x * (signed)35468 >> 16 + x +; * +; **************************************************************************/ + + +;void vp8_short_idct4x4llm_mmx(short *input, unsigned char *pred, +;int pitch, unsigned char *dest,int stride) +global sym(vp8_short_idct4x4llm_mmx) PRIVATE +sym(vp8_short_idct4x4llm_mmx): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 5 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + mov rax, arg(0) ;input + mov rsi, arg(1) ;pred + + movq mm0, [rax ] + movq mm1, [rax+ 8] + movq mm2, [rax+16] + movq mm3, [rax+24] + +%if 0 + pxor mm7, mm7 + movq [rax], mm7 + movq [rax+8], mm7 + movq [rax+16],mm7 + movq [rax+24],mm7 +%endif + movsxd rax, dword ptr arg(2) ;pitch + mov rdx, arg(3) ;dest + movsxd rdi, dword ptr arg(4) ;stride + + + psubw mm0, mm2 ; b1= 0-2 + paddw mm2, mm2 ; + + movq mm5, mm1 + paddw mm2, mm0 ; a1 =0+2 + + pmulhw mm5, [GLOBAL(x_s1sqr2)]; + paddw mm5, mm1 ; ip1 * sin(pi/8) * sqrt(2) + + movq mm7, mm3 ; + pmulhw mm7, [GLOBAL(x_c1sqr2less1)]; + + paddw mm7, mm3 ; ip3 * cos(pi/8) * sqrt(2) + psubw mm7, mm5 ; c1 + + movq mm5, mm1 + movq mm4, mm3 + + pmulhw mm5, [GLOBAL(x_c1sqr2less1)] + paddw mm5, mm1 + + pmulhw mm3, [GLOBAL(x_s1sqr2)] + paddw mm3, mm4 + + paddw mm3, mm5 ; d1 + movq mm6, mm2 ; a1 + + movq mm4, mm0 ; b1 + paddw mm2, mm3 ;0 + + paddw mm4, mm7 ;1 + psubw mm0, mm7 ;2 + + psubw mm6, mm3 ;3 + + movq mm1, mm2 ; 03 02 01 00 + movq mm3, mm4 ; 23 22 21 20 + + punpcklwd mm1, mm0 ; 11 01 10 00 + punpckhwd mm2, mm0 ; 13 03 12 02 + + punpcklwd mm3, mm6 ; 31 21 30 20 + punpckhwd mm4, mm6 ; 33 23 32 22 + + movq mm0, mm1 ; 11 01 10 00 + movq mm5, mm2 ; 13 03 12 02 + + punpckldq mm0, mm3 ; 30 20 10 00 + punpckhdq mm1, mm3 ; 31 21 11 01 + + punpckldq mm2, mm4 ; 32 22 12 02 + punpckhdq mm5, mm4 ; 33 23 13 03 + + movq mm3, mm5 ; 33 23 13 03 + + psubw mm0, mm2 ; b1= 0-2 + paddw mm2, mm2 ; + + movq mm5, mm1 + paddw mm2, mm0 ; a1 =0+2 + + pmulhw mm5, [GLOBAL(x_s1sqr2)]; + paddw mm5, mm1 ; ip1 * sin(pi/8) * sqrt(2) + + movq mm7, mm3 ; + pmulhw mm7, [GLOBAL(x_c1sqr2less1)]; + + paddw mm7, mm3 ; ip3 * cos(pi/8) * sqrt(2) + psubw mm7, mm5 ; c1 + + movq mm5, mm1 + movq mm4, mm3 + + pmulhw mm5, [GLOBAL(x_c1sqr2less1)] + paddw mm5, mm1 + + pmulhw mm3, [GLOBAL(x_s1sqr2)] + paddw mm3, mm4 + + paddw mm3, mm5 ; d1 + paddw mm0, [GLOBAL(fours)] + + paddw mm2, [GLOBAL(fours)] + movq mm6, mm2 ; a1 + + movq mm4, mm0 ; b1 + paddw mm2, mm3 ;0 + + paddw mm4, mm7 ;1 + psubw mm0, mm7 ;2 + + psubw mm6, mm3 ;3 + psraw mm2, 3 + + psraw mm0, 3 + psraw mm4, 3 + + psraw mm6, 3 + + movq mm1, mm2 ; 03 02 01 00 + movq mm3, mm4 ; 23 22 21 20 + + punpcklwd mm1, mm0 ; 11 01 10 00 + punpckhwd mm2, mm0 ; 13 03 12 02 + + punpcklwd mm3, mm6 ; 31 21 30 20 + punpckhwd mm4, mm6 ; 33 23 32 22 + + movq mm0, mm1 ; 11 01 10 00 + movq mm5, mm2 ; 13 03 12 02 + + punpckldq mm0, mm3 ; 30 20 10 00 + punpckhdq mm1, mm3 ; 31 21 11 01 + + punpckldq mm2, mm4 ; 32 22 12 02 + punpckhdq mm5, mm4 ; 33 23 13 03 + + pxor mm7, mm7 + + movd mm4, [rsi] + punpcklbw mm4, mm7 + paddsw mm0, mm4 + packuswb mm0, mm7 + movd [rdx], mm0 + + movd mm4, [rsi+rax] + punpcklbw mm4, mm7 + paddsw mm1, mm4 + packuswb mm1, mm7 + movd [rdx+rdi], mm1 + + movd mm4, [rsi+2*rax] + punpcklbw mm4, mm7 + paddsw mm2, mm4 + packuswb mm2, mm7 + movd [rdx+rdi*2], mm2 + + add rdx, rdi + add rsi, rax + + movd mm4, [rsi+2*rax] + punpcklbw mm4, mm7 + paddsw mm5, mm4 + packuswb mm5, mm7 + movd [rdx+rdi*2], mm5 + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + UNSHADOW_ARGS + pop rbp + ret + +;void vp8_dc_only_idct_add_mmx( +;short input_dc, +;unsigned char *pred_ptr, +;int pred_stride, +;unsigned char *dst_ptr, +;int stride) +global sym(vp8_dc_only_idct_add_mmx) PRIVATE +sym(vp8_dc_only_idct_add_mmx): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 5 + GET_GOT rbx + ; end prolog + + movd mm5, arg(0) ;input_dc + mov rax, arg(1) ;pred_ptr + movsxd rdx, dword ptr arg(2) ;pred_stride + + pxor mm0, mm0 + + paddw mm5, [GLOBAL(fours)] + lea rcx, [rdx + rdx*2] + + psraw mm5, 3 + + punpcklwd mm5, mm5 + + punpckldq mm5, mm5 + + movd mm1, [rax] + movd mm2, [rax+rdx] + movd mm3, [rax+2*rdx] + movd mm4, [rax+rcx] + + mov rax, arg(3) ;d -- destination + movsxd rdx, dword ptr arg(4) ;dst_stride + + punpcklbw mm1, mm0 + paddsw mm1, mm5 + packuswb mm1, mm0 ; pack and unpack to saturate + lea rcx, [rdx + rdx*2] + + punpcklbw mm2, mm0 + paddsw mm2, mm5 + packuswb mm2, mm0 ; pack and unpack to saturate + + punpcklbw mm3, mm0 + paddsw mm3, mm5 + packuswb mm3, mm0 ; pack and unpack to saturate + + punpcklbw mm4, mm0 + paddsw mm4, mm5 + packuswb mm4, mm0 ; pack and unpack to saturate + + movd [rax], mm1 + movd [rax+rdx], mm2 + movd [rax+2*rdx], mm3 + movd [rax+rcx], mm4 + + ; begin epilog + RESTORE_GOT + UNSHADOW_ARGS + pop rbp + ret + +SECTION_RODATA +align 16 +x_s1sqr2: + times 4 dw 0x8A8C +align 16 +x_c1sqr2less1: + times 4 dw 0x4E7B +align 16 +fours: + times 4 dw 0x0004 diff --git a/thirdparty/libvpx/vp8/common/x86/idctllm_sse2.asm b/thirdparty/libvpx/vp8/common/x86/idctllm_sse2.asm new file mode 100644 index 0000000000..bf8e2c4021 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/x86/idctllm_sse2.asm @@ -0,0 +1,708 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + + +%include "vpx_ports/x86_abi_support.asm" + +;void vp8_idct_dequant_0_2x_sse2 +; ( +; short *qcoeff - 0 +; short *dequant - 1 +; unsigned char *dst - 2 +; int dst_stride - 3 +; ) + +global sym(vp8_idct_dequant_0_2x_sse2) PRIVATE +sym(vp8_idct_dequant_0_2x_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 4 + GET_GOT rbx + ; end prolog + + mov rdx, arg(1) ; dequant + mov rax, arg(0) ; qcoeff + + movd xmm4, [rax] + movd xmm5, [rdx] + + pinsrw xmm4, [rax+32], 4 + pinsrw xmm5, [rdx], 4 + + pmullw xmm4, xmm5 + + ; Zero out xmm5, for use unpacking + pxor xmm5, xmm5 + + ; clear coeffs + movd [rax], xmm5 + movd [rax+32], xmm5 +;pshufb + mov rax, arg(2) ; dst + movsxd rdx, dword ptr arg(3) ; dst_stride + + pshuflw xmm4, xmm4, 00000000b + pshufhw xmm4, xmm4, 00000000b + + lea rcx, [rdx + rdx*2] + paddw xmm4, [GLOBAL(fours)] + + psraw xmm4, 3 + + movq xmm0, [rax] + movq xmm1, [rax+rdx] + movq xmm2, [rax+2*rdx] + movq xmm3, [rax+rcx] + + punpcklbw xmm0, xmm5 + punpcklbw xmm1, xmm5 + punpcklbw xmm2, xmm5 + punpcklbw xmm3, xmm5 + + + ; Add to predict buffer + paddw xmm0, xmm4 + paddw xmm1, xmm4 + paddw xmm2, xmm4 + paddw xmm3, xmm4 + + ; pack up before storing + packuswb xmm0, xmm5 + packuswb xmm1, xmm5 + packuswb xmm2, xmm5 + packuswb xmm3, xmm5 + + ; store blocks back out + movq [rax], xmm0 + movq [rax + rdx], xmm1 + + lea rax, [rax + 2*rdx] + + movq [rax], xmm2 + movq [rax + rdx], xmm3 + + ; begin epilog + RESTORE_GOT + UNSHADOW_ARGS + pop rbp + ret + +;void vp8_idct_dequant_full_2x_sse2 +; ( +; short *qcoeff - 0 +; short *dequant - 1 +; unsigned char *dst - 2 +; int dst_stride - 3 +; ) +global sym(vp8_idct_dequant_full_2x_sse2) PRIVATE +sym(vp8_idct_dequant_full_2x_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 4 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + ; special case when 2 blocks have 0 or 1 coeffs + ; dc is set as first coeff, so no need to load qcoeff + mov rax, arg(0) ; qcoeff + mov rdx, arg(1) ; dequant + mov rdi, arg(2) ; dst + + + ; Zero out xmm7, for use unpacking + pxor xmm7, xmm7 + + + ; note the transpose of xmm1 and xmm2, necessary for shuffle + ; to spit out sensicle data + movdqa xmm0, [rax] + movdqa xmm2, [rax+16] + movdqa xmm1, [rax+32] + movdqa xmm3, [rax+48] + + ; Clear out coeffs + movdqa [rax], xmm7 + movdqa [rax+16], xmm7 + movdqa [rax+32], xmm7 + movdqa [rax+48], xmm7 + + ; dequantize qcoeff buffer + pmullw xmm0, [rdx] + pmullw xmm2, [rdx+16] + pmullw xmm1, [rdx] + pmullw xmm3, [rdx+16] + movsxd rdx, dword ptr arg(3) ; dst_stride + + ; repack so block 0 row x and block 1 row x are together + movdqa xmm4, xmm0 + punpckldq xmm0, xmm1 + punpckhdq xmm4, xmm1 + + pshufd xmm0, xmm0, 11011000b + pshufd xmm1, xmm4, 11011000b + + movdqa xmm4, xmm2 + punpckldq xmm2, xmm3 + punpckhdq xmm4, xmm3 + + pshufd xmm2, xmm2, 11011000b + pshufd xmm3, xmm4, 11011000b + + ; first pass + psubw xmm0, xmm2 ; b1 = 0-2 + paddw xmm2, xmm2 ; + + movdqa xmm5, xmm1 + paddw xmm2, xmm0 ; a1 = 0+2 + + pmulhw xmm5, [GLOBAL(x_s1sqr2)] + lea rcx, [rdx + rdx*2] ;dst_stride * 3 + paddw xmm5, xmm1 ; ip1 * sin(pi/8) * sqrt(2) + + movdqa xmm7, xmm3 + pmulhw xmm7, [GLOBAL(x_c1sqr2less1)] + + paddw xmm7, xmm3 ; ip3 * cos(pi/8) * sqrt(2) + psubw xmm7, xmm5 ; c1 + + movdqa xmm5, xmm1 + movdqa xmm4, xmm3 + + pmulhw xmm5, [GLOBAL(x_c1sqr2less1)] + paddw xmm5, xmm1 + + pmulhw xmm3, [GLOBAL(x_s1sqr2)] + paddw xmm3, xmm4 + + paddw xmm3, xmm5 ; d1 + movdqa xmm6, xmm2 ; a1 + + movdqa xmm4, xmm0 ; b1 + paddw xmm2, xmm3 ;0 + + paddw xmm4, xmm7 ;1 + psubw xmm0, xmm7 ;2 + + psubw xmm6, xmm3 ;3 + + ; transpose for the second pass + movdqa xmm7, xmm2 ; 103 102 101 100 003 002 001 000 + punpcklwd xmm2, xmm0 ; 007 003 006 002 005 001 004 000 + punpckhwd xmm7, xmm0 ; 107 103 106 102 105 101 104 100 + + movdqa xmm5, xmm4 ; 111 110 109 108 011 010 009 008 + punpcklwd xmm4, xmm6 ; 015 011 014 010 013 009 012 008 + punpckhwd xmm5, xmm6 ; 115 111 114 110 113 109 112 108 + + + movdqa xmm1, xmm2 ; 007 003 006 002 005 001 004 000 + punpckldq xmm2, xmm4 ; 013 009 005 001 012 008 004 000 + punpckhdq xmm1, xmm4 ; 015 011 007 003 014 010 006 002 + + movdqa xmm6, xmm7 ; 107 103 106 102 105 101 104 100 + punpckldq xmm7, xmm5 ; 113 109 105 101 112 108 104 100 + punpckhdq xmm6, xmm5 ; 115 111 107 103 114 110 106 102 + + + movdqa xmm5, xmm2 ; 013 009 005 001 012 008 004 000 + punpckldq xmm2, xmm7 ; 112 108 012 008 104 100 004 000 + punpckhdq xmm5, xmm7 ; 113 109 013 009 105 101 005 001 + + movdqa xmm7, xmm1 ; 015 011 007 003 014 010 006 002 + punpckldq xmm1, xmm6 ; 114 110 014 010 106 102 006 002 + punpckhdq xmm7, xmm6 ; 115 111 015 011 107 103 007 003 + + pshufd xmm0, xmm2, 11011000b + pshufd xmm2, xmm1, 11011000b + + pshufd xmm1, xmm5, 11011000b + pshufd xmm3, xmm7, 11011000b + + ; second pass + psubw xmm0, xmm2 ; b1 = 0-2 + paddw xmm2, xmm2 + + movdqa xmm5, xmm1 + paddw xmm2, xmm0 ; a1 = 0+2 + + pmulhw xmm5, [GLOBAL(x_s1sqr2)] + paddw xmm5, xmm1 ; ip1 * sin(pi/8) * sqrt(2) + + movdqa xmm7, xmm3 + pmulhw xmm7, [GLOBAL(x_c1sqr2less1)] + + paddw xmm7, xmm3 ; ip3 * cos(pi/8) * sqrt(2) + psubw xmm7, xmm5 ; c1 + + movdqa xmm5, xmm1 + movdqa xmm4, xmm3 + + pmulhw xmm5, [GLOBAL(x_c1sqr2less1)] + paddw xmm5, xmm1 + + pmulhw xmm3, [GLOBAL(x_s1sqr2)] + paddw xmm3, xmm4 + + paddw xmm3, xmm5 ; d1 + paddw xmm0, [GLOBAL(fours)] + + paddw xmm2, [GLOBAL(fours)] + movdqa xmm6, xmm2 ; a1 + + movdqa xmm4, xmm0 ; b1 + paddw xmm2, xmm3 ;0 + + paddw xmm4, xmm7 ;1 + psubw xmm0, xmm7 ;2 + + psubw xmm6, xmm3 ;3 + psraw xmm2, 3 + + psraw xmm0, 3 + psraw xmm4, 3 + + psraw xmm6, 3 + + ; transpose to save + movdqa xmm7, xmm2 ; 103 102 101 100 003 002 001 000 + punpcklwd xmm2, xmm0 ; 007 003 006 002 005 001 004 000 + punpckhwd xmm7, xmm0 ; 107 103 106 102 105 101 104 100 + + movdqa xmm5, xmm4 ; 111 110 109 108 011 010 009 008 + punpcklwd xmm4, xmm6 ; 015 011 014 010 013 009 012 008 + punpckhwd xmm5, xmm6 ; 115 111 114 110 113 109 112 108 + + + movdqa xmm1, xmm2 ; 007 003 006 002 005 001 004 000 + punpckldq xmm2, xmm4 ; 013 009 005 001 012 008 004 000 + punpckhdq xmm1, xmm4 ; 015 011 007 003 014 010 006 002 + + movdqa xmm6, xmm7 ; 107 103 106 102 105 101 104 100 + punpckldq xmm7, xmm5 ; 113 109 105 101 112 108 104 100 + punpckhdq xmm6, xmm5 ; 115 111 107 103 114 110 106 102 + + + movdqa xmm5, xmm2 ; 013 009 005 001 012 008 004 000 + punpckldq xmm2, xmm7 ; 112 108 012 008 104 100 004 000 + punpckhdq xmm5, xmm7 ; 113 109 013 009 105 101 005 001 + + movdqa xmm7, xmm1 ; 015 011 007 003 014 010 006 002 + punpckldq xmm1, xmm6 ; 114 110 014 010 106 102 006 002 + punpckhdq xmm7, xmm6 ; 115 111 015 011 107 103 007 003 + + pshufd xmm0, xmm2, 11011000b + pshufd xmm2, xmm1, 11011000b + + pshufd xmm1, xmm5, 11011000b + pshufd xmm3, xmm7, 11011000b + + pxor xmm7, xmm7 + + ; Load up predict blocks + movq xmm4, [rdi] + movq xmm5, [rdi+rdx] + + punpcklbw xmm4, xmm7 + punpcklbw xmm5, xmm7 + + paddw xmm0, xmm4 + paddw xmm1, xmm5 + + movq xmm4, [rdi+2*rdx] + movq xmm5, [rdi+rcx] + + punpcklbw xmm4, xmm7 + punpcklbw xmm5, xmm7 + + paddw xmm2, xmm4 + paddw xmm3, xmm5 + +.finish: + + ; pack up before storing + packuswb xmm0, xmm7 + packuswb xmm1, xmm7 + packuswb xmm2, xmm7 + packuswb xmm3, xmm7 + + ; store blocks back out + movq [rdi], xmm0 + movq [rdi + rdx], xmm1 + movq [rdi + rdx*2], xmm2 + movq [rdi + rcx], xmm3 + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +;void vp8_idct_dequant_dc_0_2x_sse2 +; ( +; short *qcoeff - 0 +; short *dequant - 1 +; unsigned char *dst - 2 +; int dst_stride - 3 +; short *dc - 4 +; ) +global sym(vp8_idct_dequant_dc_0_2x_sse2) PRIVATE +sym(vp8_idct_dequant_dc_0_2x_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 5 + GET_GOT rbx + push rdi + ; end prolog + + ; special case when 2 blocks have 0 or 1 coeffs + ; dc is set as first coeff, so no need to load qcoeff + mov rax, arg(0) ; qcoeff + + mov rdi, arg(2) ; dst + mov rdx, arg(4) ; dc + + ; Zero out xmm5, for use unpacking + pxor xmm5, xmm5 + + ; load up 2 dc words here == 2*16 = doubleword + movd xmm4, [rdx] + + movsxd rdx, dword ptr arg(3) ; dst_stride + lea rcx, [rdx + rdx*2] + ; Load up predict blocks + movq xmm0, [rdi] + movq xmm1, [rdi+rdx*1] + movq xmm2, [rdi+rdx*2] + movq xmm3, [rdi+rcx] + + ; Duplicate and expand dc across + punpcklwd xmm4, xmm4 + punpckldq xmm4, xmm4 + + ; Rounding to dequant and downshift + paddw xmm4, [GLOBAL(fours)] + psraw xmm4, 3 + + ; Predict buffer needs to be expanded from bytes to words + punpcklbw xmm0, xmm5 + punpcklbw xmm1, xmm5 + punpcklbw xmm2, xmm5 + punpcklbw xmm3, xmm5 + + ; Add to predict buffer + paddw xmm0, xmm4 + paddw xmm1, xmm4 + paddw xmm2, xmm4 + paddw xmm3, xmm4 + + ; pack up before storing + packuswb xmm0, xmm5 + packuswb xmm1, xmm5 + packuswb xmm2, xmm5 + packuswb xmm3, xmm5 + + ; store blocks back out + movq [rdi], xmm0 + movq [rdi + rdx], xmm1 + movq [rdi + rdx*2], xmm2 + movq [rdi + rcx], xmm3 + + ; begin epilog + pop rdi + RESTORE_GOT + UNSHADOW_ARGS + pop rbp + ret +;void vp8_idct_dequant_dc_full_2x_sse2 +; ( +; short *qcoeff - 0 +; short *dequant - 1 +; unsigned char *dst - 2 +; int dst_stride - 3 +; short *dc - 4 +; ) +global sym(vp8_idct_dequant_dc_full_2x_sse2) PRIVATE +sym(vp8_idct_dequant_dc_full_2x_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 5 + SAVE_XMM 7 + GET_GOT rbx + push rdi + ; end prolog + + ; special case when 2 blocks have 0 or 1 coeffs + ; dc is set as first coeff, so no need to load qcoeff + mov rax, arg(0) ; qcoeff + mov rdx, arg(1) ; dequant + + mov rdi, arg(2) ; dst + + ; Zero out xmm7, for use unpacking + pxor xmm7, xmm7 + + + ; note the transpose of xmm1 and xmm2, necessary for shuffle + ; to spit out sensicle data + movdqa xmm0, [rax] + movdqa xmm2, [rax+16] + movdqa xmm1, [rax+32] + movdqa xmm3, [rax+48] + + ; Clear out coeffs + movdqa [rax], xmm7 + movdqa [rax+16], xmm7 + movdqa [rax+32], xmm7 + movdqa [rax+48], xmm7 + + ; dequantize qcoeff buffer + pmullw xmm0, [rdx] + pmullw xmm2, [rdx+16] + pmullw xmm1, [rdx] + pmullw xmm3, [rdx+16] + + ; DC component + mov rdx, arg(4) + + ; repack so block 0 row x and block 1 row x are together + movdqa xmm4, xmm0 + punpckldq xmm0, xmm1 + punpckhdq xmm4, xmm1 + + pshufd xmm0, xmm0, 11011000b + pshufd xmm1, xmm4, 11011000b + + movdqa xmm4, xmm2 + punpckldq xmm2, xmm3 + punpckhdq xmm4, xmm3 + + pshufd xmm2, xmm2, 11011000b + pshufd xmm3, xmm4, 11011000b + + ; insert DC component + pinsrw xmm0, [rdx], 0 + pinsrw xmm0, [rdx+2], 4 + + ; first pass + psubw xmm0, xmm2 ; b1 = 0-2 + paddw xmm2, xmm2 ; + + movdqa xmm5, xmm1 + paddw xmm2, xmm0 ; a1 = 0+2 + + pmulhw xmm5, [GLOBAL(x_s1sqr2)] + paddw xmm5, xmm1 ; ip1 * sin(pi/8) * sqrt(2) + + movdqa xmm7, xmm3 + pmulhw xmm7, [GLOBAL(x_c1sqr2less1)] + + paddw xmm7, xmm3 ; ip3 * cos(pi/8) * sqrt(2) + psubw xmm7, xmm5 ; c1 + + movdqa xmm5, xmm1 + movdqa xmm4, xmm3 + + pmulhw xmm5, [GLOBAL(x_c1sqr2less1)] + paddw xmm5, xmm1 + + pmulhw xmm3, [GLOBAL(x_s1sqr2)] + paddw xmm3, xmm4 + + paddw xmm3, xmm5 ; d1 + movdqa xmm6, xmm2 ; a1 + + movdqa xmm4, xmm0 ; b1 + paddw xmm2, xmm3 ;0 + + paddw xmm4, xmm7 ;1 + psubw xmm0, xmm7 ;2 + + psubw xmm6, xmm3 ;3 + + ; transpose for the second pass + movdqa xmm7, xmm2 ; 103 102 101 100 003 002 001 000 + punpcklwd xmm2, xmm0 ; 007 003 006 002 005 001 004 000 + punpckhwd xmm7, xmm0 ; 107 103 106 102 105 101 104 100 + + movdqa xmm5, xmm4 ; 111 110 109 108 011 010 009 008 + punpcklwd xmm4, xmm6 ; 015 011 014 010 013 009 012 008 + punpckhwd xmm5, xmm6 ; 115 111 114 110 113 109 112 108 + + + movdqa xmm1, xmm2 ; 007 003 006 002 005 001 004 000 + punpckldq xmm2, xmm4 ; 013 009 005 001 012 008 004 000 + punpckhdq xmm1, xmm4 ; 015 011 007 003 014 010 006 002 + + movdqa xmm6, xmm7 ; 107 103 106 102 105 101 104 100 + punpckldq xmm7, xmm5 ; 113 109 105 101 112 108 104 100 + punpckhdq xmm6, xmm5 ; 115 111 107 103 114 110 106 102 + + + movdqa xmm5, xmm2 ; 013 009 005 001 012 008 004 000 + punpckldq xmm2, xmm7 ; 112 108 012 008 104 100 004 000 + punpckhdq xmm5, xmm7 ; 113 109 013 009 105 101 005 001 + + movdqa xmm7, xmm1 ; 015 011 007 003 014 010 006 002 + punpckldq xmm1, xmm6 ; 114 110 014 010 106 102 006 002 + punpckhdq xmm7, xmm6 ; 115 111 015 011 107 103 007 003 + + pshufd xmm0, xmm2, 11011000b + pshufd xmm2, xmm1, 11011000b + + pshufd xmm1, xmm5, 11011000b + pshufd xmm3, xmm7, 11011000b + + ; second pass + psubw xmm0, xmm2 ; b1 = 0-2 + paddw xmm2, xmm2 + + movdqa xmm5, xmm1 + paddw xmm2, xmm0 ; a1 = 0+2 + + pmulhw xmm5, [GLOBAL(x_s1sqr2)] + paddw xmm5, xmm1 ; ip1 * sin(pi/8) * sqrt(2) + + movdqa xmm7, xmm3 + pmulhw xmm7, [GLOBAL(x_c1sqr2less1)] + + paddw xmm7, xmm3 ; ip3 * cos(pi/8) * sqrt(2) + psubw xmm7, xmm5 ; c1 + + movdqa xmm5, xmm1 + movdqa xmm4, xmm3 + + pmulhw xmm5, [GLOBAL(x_c1sqr2less1)] + paddw xmm5, xmm1 + + pmulhw xmm3, [GLOBAL(x_s1sqr2)] + paddw xmm3, xmm4 + + paddw xmm3, xmm5 ; d1 + paddw xmm0, [GLOBAL(fours)] + + paddw xmm2, [GLOBAL(fours)] + movdqa xmm6, xmm2 ; a1 + + movdqa xmm4, xmm0 ; b1 + paddw xmm2, xmm3 ;0 + + paddw xmm4, xmm7 ;1 + psubw xmm0, xmm7 ;2 + + psubw xmm6, xmm3 ;3 + psraw xmm2, 3 + + psraw xmm0, 3 + psraw xmm4, 3 + + psraw xmm6, 3 + + ; transpose to save + movdqa xmm7, xmm2 ; 103 102 101 100 003 002 001 000 + punpcklwd xmm2, xmm0 ; 007 003 006 002 005 001 004 000 + punpckhwd xmm7, xmm0 ; 107 103 106 102 105 101 104 100 + + movdqa xmm5, xmm4 ; 111 110 109 108 011 010 009 008 + punpcklwd xmm4, xmm6 ; 015 011 014 010 013 009 012 008 + punpckhwd xmm5, xmm6 ; 115 111 114 110 113 109 112 108 + + + movdqa xmm1, xmm2 ; 007 003 006 002 005 001 004 000 + punpckldq xmm2, xmm4 ; 013 009 005 001 012 008 004 000 + punpckhdq xmm1, xmm4 ; 015 011 007 003 014 010 006 002 + + movdqa xmm6, xmm7 ; 107 103 106 102 105 101 104 100 + punpckldq xmm7, xmm5 ; 113 109 105 101 112 108 104 100 + punpckhdq xmm6, xmm5 ; 115 111 107 103 114 110 106 102 + + + movdqa xmm5, xmm2 ; 013 009 005 001 012 008 004 000 + punpckldq xmm2, xmm7 ; 112 108 012 008 104 100 004 000 + punpckhdq xmm5, xmm7 ; 113 109 013 009 105 101 005 001 + + movdqa xmm7, xmm1 ; 015 011 007 003 014 010 006 002 + punpckldq xmm1, xmm6 ; 114 110 014 010 106 102 006 002 + punpckhdq xmm7, xmm6 ; 115 111 015 011 107 103 007 003 + + pshufd xmm0, xmm2, 11011000b + pshufd xmm2, xmm1, 11011000b + + pshufd xmm1, xmm5, 11011000b + pshufd xmm3, xmm7, 11011000b + + pxor xmm7, xmm7 + + ; Load up predict blocks + movsxd rdx, dword ptr arg(3) ; dst_stride + movq xmm4, [rdi] + movq xmm5, [rdi+rdx] + lea rcx, [rdx + rdx*2] + + punpcklbw xmm4, xmm7 + punpcklbw xmm5, xmm7 + + paddw xmm0, xmm4 + paddw xmm1, xmm5 + + movq xmm4, [rdi+rdx*2] + movq xmm5, [rdi+rcx] + + punpcklbw xmm4, xmm7 + punpcklbw xmm5, xmm7 + + paddw xmm2, xmm4 + paddw xmm3, xmm5 + +.finish: + + ; pack up before storing + packuswb xmm0, xmm7 + packuswb xmm1, xmm7 + packuswb xmm2, xmm7 + packuswb xmm3, xmm7 + + ; Load destination stride before writing out, + ; doesn't need to persist + movsxd rdx, dword ptr arg(3) ; dst_stride + + ; store blocks back out + movq [rdi], xmm0 + movq [rdi + rdx], xmm1 + + lea rdi, [rdi + 2*rdx] + + movq [rdi], xmm2 + movq [rdi + rdx], xmm3 + + + ; begin epilog + pop rdi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +SECTION_RODATA +align 16 +fours: + times 8 dw 0x0004 +align 16 +x_s1sqr2: + times 8 dw 0x8A8C +align 16 +x_c1sqr2less1: + times 8 dw 0x4E7B diff --git a/thirdparty/libvpx/vp8/common/x86/iwalsh_mmx.asm b/thirdparty/libvpx/vp8/common/x86/iwalsh_mmx.asm new file mode 100644 index 0000000000..158c3b7458 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/x86/iwalsh_mmx.asm @@ -0,0 +1,140 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + + +%include "vpx_ports/x86_abi_support.asm" + +;void vp8_short_inv_walsh4x4_mmx(short *input, short *output) +global sym(vp8_short_inv_walsh4x4_mmx) PRIVATE +sym(vp8_short_inv_walsh4x4_mmx): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 2 + ; end prolog + + mov rdx, arg(0) + mov rax, 30003h + + movq mm0, [rdx + 0] ;ip[0] + movq mm1, [rdx + 8] ;ip[4] + movq mm7, rax + + movq mm2, [rdx + 16] ;ip[8] + movq mm3, [rdx + 24] ;ip[12] + punpcklwd mm7, mm7 ;0003000300030003h + mov rdx, arg(1) + + movq mm4, mm0 + movq mm5, mm1 + + paddw mm4, mm3 ;ip[0] + ip[12] aka al + paddw mm5, mm2 ;ip[4] + ip[8] aka bl + + movq mm6, mm4 ;temp al + paddw mm4, mm5 ;al + bl + psubw mm6, mm5 ;al - bl + + psubw mm0, mm3 ;ip[0] - ip[12] aka d1 + psubw mm1, mm2 ;ip[4] - ip[8] aka c1 + + movq mm5, mm0 ;temp dl + paddw mm0, mm1 ;dl + cl + psubw mm5, mm1 ;dl - cl + + ; 03 02 01 00 + ; 13 12 11 10 + ; 23 22 21 20 + ; 33 32 31 30 + + movq mm3, mm4 ; 03 02 01 00 + punpcklwd mm4, mm0 ; 11 01 10 00 + punpckhwd mm3, mm0 ; 13 03 12 02 + + movq mm1, mm6 ; 23 22 21 20 + punpcklwd mm6, mm5 ; 31 21 30 20 + punpckhwd mm1, mm5 ; 33 23 32 22 + + movq mm0, mm4 ; 11 01 10 00 + movq mm2, mm3 ; 13 03 12 02 + + punpckldq mm0, mm6 ; 30 20 10 00 aka ip[0] + punpckhdq mm4, mm6 ; 31 21 11 01 aka ip[4] + + punpckldq mm2, mm1 ; 32 22 12 02 aka ip[8] + punpckhdq mm3, mm1 ; 33 23 13 03 aka ip[12] +;~~~~~~~~~~~~~~~~~~~~~ + movq mm1, mm0 + movq mm5, mm4 + paddw mm1, mm3 ;ip[0] + ip[12] aka al + paddw mm5, mm2 ;ip[4] + ip[8] aka bl + + movq mm6, mm1 ;temp al + paddw mm1, mm5 ;al + bl + psubw mm6, mm5 ;al - bl + paddw mm1, mm7 + paddw mm6, mm7 + psraw mm1, 3 + psraw mm6, 3 + + psubw mm0, mm3 ;ip[0] - ip[12] aka d1 + psubw mm4, mm2 ;ip[4] - ip[8] aka c1 + + movq mm5, mm0 ;temp dl + paddw mm0, mm4 ;dl + cl + psubw mm5, mm4 ;dl - cl + paddw mm0, mm7 + paddw mm5, mm7 + psraw mm0, 3 + psraw mm5, 3 +;~~~~~~~~~~~~~~~~~~~~~ + + movd eax, mm1 + movd ecx, mm0 + psrlq mm0, 32 + psrlq mm1, 32 + mov word ptr[rdx+32*0], ax + mov word ptr[rdx+32*1], cx + shr eax, 16 + shr ecx, 16 + mov word ptr[rdx+32*4], ax + mov word ptr[rdx+32*5], cx + movd eax, mm1 + movd ecx, mm0 + mov word ptr[rdx+32*8], ax + mov word ptr[rdx+32*9], cx + shr eax, 16 + shr ecx, 16 + mov word ptr[rdx+32*12], ax + mov word ptr[rdx+32*13], cx + + movd eax, mm6 + movd ecx, mm5 + psrlq mm5, 32 + psrlq mm6, 32 + mov word ptr[rdx+32*2], ax + mov word ptr[rdx+32*3], cx + shr eax, 16 + shr ecx, 16 + mov word ptr[rdx+32*6], ax + mov word ptr[rdx+32*7], cx + movd eax, mm6 + movd ecx, mm5 + mov word ptr[rdx+32*10], ax + mov word ptr[rdx+32*11], cx + shr eax, 16 + shr ecx, 16 + mov word ptr[rdx+32*14], ax + mov word ptr[rdx+32*15], cx + + ; begin epilog + UNSHADOW_ARGS + pop rbp + ret + diff --git a/thirdparty/libvpx/vp8/common/x86/iwalsh_sse2.asm b/thirdparty/libvpx/vp8/common/x86/iwalsh_sse2.asm new file mode 100644 index 0000000000..06e86a80b6 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/x86/iwalsh_sse2.asm @@ -0,0 +1,121 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + + +%include "vpx_ports/x86_abi_support.asm" + +;void vp8_short_inv_walsh4x4_sse2(short *input, short *output) +global sym(vp8_short_inv_walsh4x4_sse2) PRIVATE +sym(vp8_short_inv_walsh4x4_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 2 + ; end prolog + + mov rcx, arg(0) + mov rdx, arg(1) + mov rax, 30003h + + movdqa xmm0, [rcx + 0] ;ip[4] ip[0] + movdqa xmm1, [rcx + 16] ;ip[12] ip[8] + + + pshufd xmm2, xmm1, 4eh ;ip[8] ip[12] + movdqa xmm3, xmm0 ;ip[4] ip[0] + + paddw xmm0, xmm2 ;ip[4]+ip[8] ip[0]+ip[12] aka b1 a1 + psubw xmm3, xmm2 ;ip[4]-ip[8] ip[0]-ip[12] aka c1 d1 + + movdqa xmm4, xmm0 + punpcklqdq xmm0, xmm3 ;d1 a1 + punpckhqdq xmm4, xmm3 ;c1 b1 + + movdqa xmm1, xmm4 ;c1 b1 + paddw xmm4, xmm0 ;dl+cl a1+b1 aka op[4] op[0] + psubw xmm0, xmm1 ;d1-c1 a1-b1 aka op[12] op[8] + + ;~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + ; 13 12 11 10 03 02 01 00 + ; + ; 33 32 31 30 23 22 21 20 + ; + movdqa xmm3, xmm4 ; 13 12 11 10 03 02 01 00 + punpcklwd xmm4, xmm0 ; 23 03 22 02 21 01 20 00 + punpckhwd xmm3, xmm0 ; 33 13 32 12 31 11 30 10 + movdqa xmm1, xmm4 ; 23 03 22 02 21 01 20 00 + punpcklwd xmm4, xmm3 ; 31 21 11 01 30 20 10 00 + punpckhwd xmm1, xmm3 ; 33 23 13 03 32 22 12 02 + ;~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + movd xmm0, eax + pshufd xmm2, xmm1, 4eh ;ip[8] ip[12] + movdqa xmm3, xmm4 ;ip[4] ip[0] + + pshufd xmm0, xmm0, 0 ;03 03 03 03 03 03 03 03 + + paddw xmm4, xmm2 ;ip[4]+ip[8] ip[0]+ip[12] aka b1 a1 + psubw xmm3, xmm2 ;ip[4]-ip[8] ip[0]-ip[12] aka c1 d1 + + movdqa xmm5, xmm4 + punpcklqdq xmm4, xmm3 ;d1 a1 + punpckhqdq xmm5, xmm3 ;c1 b1 + + movdqa xmm1, xmm5 ;c1 b1 + paddw xmm5, xmm4 ;dl+cl a1+b1 aka op[4] op[0] + psubw xmm4, xmm1 ;d1-c1 a1-b1 aka op[12] op[8] + + paddw xmm5, xmm0 + paddw xmm4, xmm0 + psraw xmm5, 3 + psraw xmm4, 3 + + movd eax, xmm5 + movd ecx, xmm4 + psrldq xmm5, 4 + psrldq xmm4, 4 + mov word ptr[rdx+32*0], ax + mov word ptr[rdx+32*2], cx + shr eax, 16 + shr ecx, 16 + mov word ptr[rdx+32*4], ax + mov word ptr[rdx+32*6], cx + movd eax, xmm5 + movd ecx, xmm4 + psrldq xmm5, 4 + psrldq xmm4, 4 + mov word ptr[rdx+32*8], ax + mov word ptr[rdx+32*10], cx + shr eax, 16 + shr ecx, 16 + mov word ptr[rdx+32*12], ax + mov word ptr[rdx+32*14], cx + + movd eax, xmm5 + movd ecx, xmm4 + psrldq xmm5, 4 + psrldq xmm4, 4 + mov word ptr[rdx+32*1], ax + mov word ptr[rdx+32*3], cx + shr eax, 16 + shr ecx, 16 + mov word ptr[rdx+32*5], ax + mov word ptr[rdx+32*7], cx + movd eax, xmm5 + movd ecx, xmm4 + mov word ptr[rdx+32*9], ax + mov word ptr[rdx+32*11], cx + shr eax, 16 + shr ecx, 16 + mov word ptr[rdx+32*13], ax + mov word ptr[rdx+32*15], cx + + ; begin epilog + UNSHADOW_ARGS + pop rbp + ret diff --git a/thirdparty/libvpx/vp8/common/x86/loopfilter_block_sse2_x86_64.asm b/thirdparty/libvpx/vp8/common/x86/loopfilter_block_sse2_x86_64.asm new file mode 100644 index 0000000000..6d5aaa19db --- /dev/null +++ b/thirdparty/libvpx/vp8/common/x86/loopfilter_block_sse2_x86_64.asm @@ -0,0 +1,815 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + + +%include "vpx_ports/x86_abi_support.asm" + +%macro LF_ABS 2 + ; %1 value not preserved + ; %2 value preserved + ; output in %1 + movdqa scratch1, %2 ; v2 + + psubusb scratch1, %1 ; v2 - v1 + psubusb %1, %2 ; v1 - v2 + por %1, scratch1 ; abs(v2 - v1) +%endmacro + +%macro LF_FILTER_HEV_MASK 8-9 + + LF_ABS %1, %2 ; abs(p3 - p2) + LF_ABS %2, %3 ; abs(p2 - p1) + pmaxub %1, %2 ; accumulate mask +%if %0 == 8 + movdqa scratch2, %3 ; save p1 + LF_ABS scratch2, %4 ; abs(p1 - p0) +%endif + LF_ABS %4, %5 ; abs(p0 - q0) + LF_ABS %5, %6 ; abs(q0 - q1) +%if %0 == 8 + pmaxub %5, scratch2 ; accumulate hev +%else + pmaxub %5, %9 +%endif + pmaxub %1, %5 ; accumulate mask + + LF_ABS %3, %6 ; abs(p1 - q1) + LF_ABS %6, %7 ; abs(q1 - q2) + pmaxub %1, %6 ; accumulate mask + LF_ABS %7, %8 ; abs(q2 - q3) + pmaxub %1, %7 ; accumulate mask + + paddusb %4, %4 ; 2 * abs(p0 - q0) + pand %3, [GLOBAL(tfe)] + psrlw %3, 1 ; abs(p1 - q1) / 2 + paddusb %4, %3 ; abs(p0 - q0) * 2 + abs(p1 - q1) / 2 + + psubusb %1, [limit] + psubusb %4, [blimit] + por %1, %4 + pcmpeqb %1, zero ; mask + + psubusb %5, [thresh] + pcmpeqb %5, zero ; ~hev +%endmacro + +%macro LF_FILTER 6 + ; %1-%4: p1-q1 + ; %5: mask + ; %6: hev + + movdqa scratch2, %6 ; save hev + + pxor %1, [GLOBAL(t80)] ; ps1 + pxor %4, [GLOBAL(t80)] ; qs1 + movdqa scratch1, %1 + psubsb scratch1, %4 ; signed_char_clamp(ps1 - qs1) + pandn scratch2, scratch1 ; vp8_filter &= hev + + pxor %2, [GLOBAL(t80)] ; ps0 + pxor %3, [GLOBAL(t80)] ; qs0 + movdqa scratch1, %3 + psubsb scratch1, %2 ; qs0 - ps0 + paddsb scratch2, scratch1 ; vp8_filter += (qs0 - ps0) + paddsb scratch2, scratch1 ; vp8_filter += (qs0 - ps0) + paddsb scratch2, scratch1 ; vp8_filter += (qs0 - ps0) + pand %5, scratch2 ; &= mask + + movdqa scratch2, %5 + paddsb %5, [GLOBAL(t4)] ; Filter1 + paddsb scratch2, [GLOBAL(t3)] ; Filter2 + + ; Filter1 >> 3 + movdqa scratch1, zero + pcmpgtb scratch1, %5 + psrlw %5, 3 + pand scratch1, [GLOBAL(te0)] + pand %5, [GLOBAL(t1f)] + por %5, scratch1 + + psubsb %3, %5 ; qs0 - Filter1 + pxor %3, [GLOBAL(t80)] + + ; Filter2 >> 3 + movdqa scratch1, zero + pcmpgtb scratch1, scratch2 + psrlw scratch2, 3 + pand scratch1, [GLOBAL(te0)] + pand scratch2, [GLOBAL(t1f)] + por scratch2, scratch1 + + paddsb %2, scratch2 ; ps0 + Filter2 + pxor %2, [GLOBAL(t80)] + + ; outer tap adjustments + paddsb %5, [GLOBAL(t1)] + movdqa scratch1, zero + pcmpgtb scratch1, %5 + psrlw %5, 1 + pand scratch1, [GLOBAL(t80)] + pand %5, [GLOBAL(t7f)] + por %5, scratch1 + pand %5, %6 ; vp8_filter &= ~hev + + psubsb %4, %5 ; qs1 - vp8_filter + pxor %4, [GLOBAL(t80)] + + paddsb %1, %5 ; ps1 + vp8_filter + pxor %1, [GLOBAL(t80)] +%endmacro + +;void vp8_loop_filter_bh_y_sse2 +;( +; unsigned char *src_ptr, +; int src_pixel_step, +; const char *blimit, +; const char *limit, +; const char *thresh +;) +global sym(vp8_loop_filter_bh_y_sse2) PRIVATE +sym(vp8_loop_filter_bh_y_sse2): + +%if LIBVPX_YASM_WIN64 + %define src rcx ; src_ptr + %define stride rdx ; src_pixel_step + %define blimit r8 + %define limit r9 + %define thresh r10 + + %define spp rax + %define stride3 r11 + %define stride5 r12 + %define stride7 r13 + + push rbp + mov rbp, rsp + SAVE_XMM 11 + push r12 + push r13 + mov thresh, arg(4) +%else + %define src rdi ; src_ptr + %define stride rsi ; src_pixel_step + %define blimit rdx + %define limit rcx + %define thresh r8 + + %define spp rax + %define stride3 r9 + %define stride5 r10 + %define stride7 r11 +%endif + + %define scratch1 xmm5 + %define scratch2 xmm6 + %define zero xmm7 + + %define i0 [src] + %define i1 [spp] + %define i2 [src + 2 * stride] + %define i3 [spp + 2 * stride] + %define i4 [src + 4 * stride] + %define i5 [spp + 4 * stride] + %define i6 [src + 2 * stride3] + %define i7 [spp + 2 * stride3] + %define i8 [src + 8 * stride] + %define i9 [spp + 8 * stride] + %define i10 [src + 2 * stride5] + %define i11 [spp + 2 * stride5] + %define i12 [src + 4 * stride3] + %define i13 [spp + 4 * stride3] + %define i14 [src + 2 * stride7] + %define i15 [spp + 2 * stride7] + + ; prep work + lea spp, [src + stride] + lea stride3, [stride + 2 * stride] + lea stride5, [stride3 + 2 * stride] + lea stride7, [stride3 + 4 * stride] + pxor zero, zero + + ; load the first set into registers + movdqa xmm0, i0 + movdqa xmm1, i1 + movdqa xmm2, i2 + movdqa xmm3, i3 + movdqa xmm4, i4 + movdqa xmm8, i5 + movdqa xmm9, i6 ; q2, will contain abs(p1-p0) + movdqa xmm10, i7 +LF_FILTER_HEV_MASK xmm0, xmm1, xmm2, xmm3, xmm4, xmm8, xmm9, xmm10 + + movdqa xmm1, i2 + movdqa xmm2, i3 + movdqa xmm3, i4 + movdqa xmm8, i5 +LF_FILTER xmm1, xmm2, xmm3, xmm8, xmm0, xmm4 + movdqa i2, xmm1 + movdqa i3, xmm2 + +; second set + movdqa i4, xmm3 + movdqa i5, xmm8 + + movdqa xmm0, i6 + movdqa xmm1, i7 + movdqa xmm2, i8 + movdqa xmm4, i9 + movdqa xmm10, i10 ; q2, will contain abs(p1-p0) + movdqa xmm11, i11 +LF_FILTER_HEV_MASK xmm3, xmm8, xmm0, xmm1, xmm2, xmm4, xmm10, xmm11, xmm9 + + movdqa xmm0, i6 + movdqa xmm1, i7 + movdqa xmm4, i8 + movdqa xmm8, i9 +LF_FILTER xmm0, xmm1, xmm4, xmm8, xmm3, xmm2 + movdqa i6, xmm0 + movdqa i7, xmm1 + +; last set + movdqa i8, xmm4 + movdqa i9, xmm8 + + movdqa xmm0, i10 + movdqa xmm1, i11 + movdqa xmm2, i12 + movdqa xmm3, i13 + movdqa xmm9, i14 ; q2, will contain abs(p1-p0) + movdqa xmm11, i15 +LF_FILTER_HEV_MASK xmm4, xmm8, xmm0, xmm1, xmm2, xmm3, xmm9, xmm11, xmm10 + + movdqa xmm0, i10 + movdqa xmm1, i11 + movdqa xmm3, i12 + movdqa xmm8, i13 +LF_FILTER xmm0, xmm1, xmm3, xmm8, xmm4, xmm2 + movdqa i10, xmm0 + movdqa i11, xmm1 + movdqa i12, xmm3 + movdqa i13, xmm8 + +%if LIBVPX_YASM_WIN64 + pop r13 + pop r12 + RESTORE_XMM + pop rbp +%endif + + ret + + +;void vp8_loop_filter_bv_y_sse2 +;( +; unsigned char *src_ptr, +; int src_pixel_step, +; const char *blimit, +; const char *limit, +; const char *thresh +;) + +global sym(vp8_loop_filter_bv_y_sse2) PRIVATE +sym(vp8_loop_filter_bv_y_sse2): + +%if LIBVPX_YASM_WIN64 + %define src rcx ; src_ptr + %define stride rdx ; src_pixel_step + %define blimit r8 + %define limit r9 + %define thresh r10 + + %define spp rax + %define stride3 r11 + %define stride5 r12 + %define stride7 r13 + + push rbp + mov rbp, rsp + SAVE_XMM 15 + push r12 + push r13 + mov thresh, arg(4) +%else + %define src rdi + %define stride rsi + %define blimit rdx + %define limit rcx + %define thresh r8 + + %define spp rax + %define stride3 r9 + %define stride5 r10 + %define stride7 r11 +%endif + + %define scratch1 xmm5 + %define scratch2 xmm6 + %define zero xmm7 + + %define s0 [src] + %define s1 [spp] + %define s2 [src + 2 * stride] + %define s3 [spp + 2 * stride] + %define s4 [src + 4 * stride] + %define s5 [spp + 4 * stride] + %define s6 [src + 2 * stride3] + %define s7 [spp + 2 * stride3] + %define s8 [src + 8 * stride] + %define s9 [spp + 8 * stride] + %define s10 [src + 2 * stride5] + %define s11 [spp + 2 * stride5] + %define s12 [src + 4 * stride3] + %define s13 [spp + 4 * stride3] + %define s14 [src + 2 * stride7] + %define s15 [spp + 2 * stride7] + + %define i0 [rsp] + %define i1 [rsp + 16] + %define i2 [rsp + 32] + %define i3 [rsp + 48] + %define i4 [rsp + 64] + %define i5 [rsp + 80] + %define i6 [rsp + 96] + %define i7 [rsp + 112] + %define i8 [rsp + 128] + %define i9 [rsp + 144] + %define i10 [rsp + 160] + %define i11 [rsp + 176] + %define i12 [rsp + 192] + %define i13 [rsp + 208] + %define i14 [rsp + 224] + %define i15 [rsp + 240] + + ALIGN_STACK 16, rax + + ; reserve stack space + %define temp_storage 0 ; size is 256 (16*16) + %define stack_size 256 + sub rsp, stack_size + + ; prep work + lea spp, [src + stride] + lea stride3, [stride + 2 * stride] + lea stride5, [stride3 + 2 * stride] + lea stride7, [stride3 + 4 * stride] + + ; 8-f + movdqa xmm0, s8 + movdqa xmm1, xmm0 + punpcklbw xmm0, s9 ; 80 90 + punpckhbw xmm1, s9 ; 88 98 + + movdqa xmm2, s10 + movdqa xmm3, xmm2 + punpcklbw xmm2, s11 ; a0 b0 + punpckhbw xmm3, s11 ; a8 b8 + + movdqa xmm4, xmm0 + punpcklwd xmm0, xmm2 ; 80 90 a0 b0 + punpckhwd xmm4, xmm2 ; 84 94 a4 b4 + + movdqa xmm2, xmm1 + punpcklwd xmm1, xmm3 ; 88 98 a8 b8 + punpckhwd xmm2, xmm3 ; 8c 9c ac bc + + ; using xmm[0124] + ; work on next 4 rows + + movdqa xmm3, s12 + movdqa xmm5, xmm3 + punpcklbw xmm3, s13 ; c0 d0 + punpckhbw xmm5, s13 ; c8 d8 + + movdqa xmm6, s14 + movdqa xmm7, xmm6 + punpcklbw xmm6, s15 ; e0 f0 + punpckhbw xmm7, s15 ; e8 f8 + + movdqa xmm8, xmm3 + punpcklwd xmm3, xmm6 ; c0 d0 e0 f0 + punpckhwd xmm8, xmm6 ; c4 d4 e4 f4 + + movdqa xmm6, xmm5 + punpcklwd xmm5, xmm7 ; c8 d8 e8 f8 + punpckhwd xmm6, xmm7 ; cc dc ec fc + + ; pull the third and fourth sets together + + movdqa xmm7, xmm0 + punpckldq xmm0, xmm3 ; 80 90 a0 b0 c0 d0 e0 f0 + punpckhdq xmm7, xmm3 ; 82 92 a2 b2 c2 d2 e2 f2 + + movdqa xmm3, xmm4 + punpckldq xmm4, xmm8 ; 84 94 a4 b4 c4 d4 e4 f4 + punpckhdq xmm3, xmm8 ; 86 96 a6 b6 c6 d6 e6 f6 + + movdqa xmm8, xmm1 + punpckldq xmm1, xmm5 ; 88 88 a8 b8 c8 d8 e8 f8 + punpckhdq xmm8, xmm5 ; 8a 9a aa ba ca da ea fa + + movdqa xmm5, xmm2 + punpckldq xmm2, xmm6 ; 8c 9c ac bc cc dc ec fc + punpckhdq xmm5, xmm6 ; 8e 9e ae be ce de ee fe + + ; save the calculations. we only have 15 registers ... + movdqa i0, xmm0 + movdqa i1, xmm7 + movdqa i2, xmm4 + movdqa i3, xmm3 + movdqa i4, xmm1 + movdqa i5, xmm8 + movdqa i6, xmm2 + movdqa i7, xmm5 + + ; 0-7 + movdqa xmm0, s0 + movdqa xmm1, xmm0 + punpcklbw xmm0, s1 ; 00 10 + punpckhbw xmm1, s1 ; 08 18 + + movdqa xmm2, s2 + movdqa xmm3, xmm2 + punpcklbw xmm2, s3 ; 20 30 + punpckhbw xmm3, s3 ; 28 38 + + movdqa xmm4, xmm0 + punpcklwd xmm0, xmm2 ; 00 10 20 30 + punpckhwd xmm4, xmm2 ; 04 14 24 34 + + movdqa xmm2, xmm1 + punpcklwd xmm1, xmm3 ; 08 18 28 38 + punpckhwd xmm2, xmm3 ; 0c 1c 2c 3c + + ; using xmm[0124] + ; work on next 4 rows + + movdqa xmm3, s4 + movdqa xmm5, xmm3 + punpcklbw xmm3, s5 ; 40 50 + punpckhbw xmm5, s5 ; 48 58 + + movdqa xmm6, s6 + movdqa xmm7, xmm6 + punpcklbw xmm6, s7 ; 60 70 + punpckhbw xmm7, s7 ; 68 78 + + movdqa xmm8, xmm3 + punpcklwd xmm3, xmm6 ; 40 50 60 70 + punpckhwd xmm8, xmm6 ; 44 54 64 74 + + movdqa xmm6, xmm5 + punpcklwd xmm5, xmm7 ; 48 58 68 78 + punpckhwd xmm6, xmm7 ; 4c 5c 6c 7c + + ; pull the first two sets together + + movdqa xmm7, xmm0 + punpckldq xmm0, xmm3 ; 00 10 20 30 40 50 60 70 + punpckhdq xmm7, xmm3 ; 02 12 22 32 42 52 62 72 + + movdqa xmm3, xmm4 + punpckldq xmm4, xmm8 ; 04 14 24 34 44 54 64 74 + punpckhdq xmm3, xmm8 ; 06 16 26 36 46 56 66 76 + + movdqa xmm8, xmm1 + punpckldq xmm1, xmm5 ; 08 18 28 38 48 58 68 78 + punpckhdq xmm8, xmm5 ; 0a 1a 2a 3a 4a 5a 6a 7a + + movdqa xmm5, xmm2 + punpckldq xmm2, xmm6 ; 0c 1c 2c 3c 4c 5c 6c 7c + punpckhdq xmm5, xmm6 ; 0e 1e 2e 3e 4e 5e 6e 7e + ; final combination + + movdqa xmm6, xmm0 + punpcklqdq xmm0, i0 + punpckhqdq xmm6, i0 + + movdqa xmm9, xmm7 + punpcklqdq xmm7, i1 + punpckhqdq xmm9, i1 + + movdqa xmm10, xmm4 + punpcklqdq xmm4, i2 + punpckhqdq xmm10, i2 + + movdqa xmm11, xmm3 + punpcklqdq xmm3, i3 + punpckhqdq xmm11, i3 + + movdqa xmm12, xmm1 + punpcklqdq xmm1, i4 + punpckhqdq xmm12, i4 + + movdqa xmm13, xmm8 + punpcklqdq xmm8, i5 + punpckhqdq xmm13, i5 + + movdqa xmm14, xmm2 + punpcklqdq xmm2, i6 + punpckhqdq xmm14, i6 + + movdqa xmm15, xmm5 + punpcklqdq xmm5, i7 + punpckhqdq xmm15, i7 + + movdqa i0, xmm0 + movdqa i1, xmm6 + movdqa i2, xmm7 + movdqa i3, xmm9 + movdqa i4, xmm4 + movdqa i5, xmm10 + movdqa i6, xmm3 + movdqa i7, xmm11 + movdqa i8, xmm1 + movdqa i9, xmm12 + movdqa i10, xmm8 + movdqa i11, xmm13 + movdqa i12, xmm2 + movdqa i13, xmm14 + movdqa i14, xmm5 + movdqa i15, xmm15 + +; TRANSPOSED DATA AVAILABLE ON THE STACK + + movdqa xmm12, xmm6 + movdqa xmm13, xmm7 + + pxor zero, zero + +LF_FILTER_HEV_MASK xmm0, xmm12, xmm13, xmm9, xmm4, xmm10, xmm3, xmm11 + + movdqa xmm1, i2 + movdqa xmm2, i3 + movdqa xmm8, i4 + movdqa xmm9, i5 +LF_FILTER xmm1, xmm2, xmm8, xmm9, xmm0, xmm4 + movdqa i2, xmm1 + movdqa i3, xmm2 + +; second set + movdqa i4, xmm8 + movdqa i5, xmm9 + + movdqa xmm0, i6 + movdqa xmm1, i7 + movdqa xmm2, i8 + movdqa xmm4, i9 + movdqa xmm10, i10 ; q2, will contain abs(p1-p0) + movdqa xmm11, i11 +LF_FILTER_HEV_MASK xmm8, xmm9, xmm0, xmm1, xmm2, xmm4, xmm10, xmm11, xmm3 + + movdqa xmm0, i6 + movdqa xmm1, i7 + movdqa xmm3, i8 + movdqa xmm4, i9 +LF_FILTER xmm0, xmm1, xmm3, xmm4, xmm8, xmm2 + movdqa i6, xmm0 + movdqa i7, xmm1 + +; last set + movdqa i8, xmm3 + movdqa i9, xmm4 + + movdqa xmm0, i10 + movdqa xmm1, i11 + movdqa xmm2, i12 + movdqa xmm8, i13 + movdqa xmm9, i14 ; q2, will contain abs(p1-p0) + movdqa xmm11, i15 +LF_FILTER_HEV_MASK xmm3, xmm4, xmm0, xmm1, xmm2, xmm8, xmm9, xmm11, xmm10 + + movdqa xmm0, i10 + movdqa xmm1, i11 + movdqa xmm4, i12 + movdqa xmm8, i13 +LF_FILTER xmm0, xmm1, xmm4, xmm8, xmm3, xmm2 + movdqa i10, xmm0 + movdqa i11, xmm1 + movdqa i12, xmm4 + movdqa i13, xmm8 + + +; RESHUFFLE AND WRITE OUT + ; 8-f + movdqa xmm0, i8 + movdqa xmm1, xmm0 + punpcklbw xmm0, i9 ; 80 90 + punpckhbw xmm1, i9 ; 88 98 + + movdqa xmm2, i10 + movdqa xmm3, xmm2 + punpcklbw xmm2, i11 ; a0 b0 + punpckhbw xmm3, i11 ; a8 b8 + + movdqa xmm4, xmm0 + punpcklwd xmm0, xmm2 ; 80 90 a0 b0 + punpckhwd xmm4, xmm2 ; 84 94 a4 b4 + + movdqa xmm2, xmm1 + punpcklwd xmm1, xmm3 ; 88 98 a8 b8 + punpckhwd xmm2, xmm3 ; 8c 9c ac bc + + ; using xmm[0124] + ; work on next 4 rows + + movdqa xmm3, i12 + movdqa xmm5, xmm3 + punpcklbw xmm3, i13 ; c0 d0 + punpckhbw xmm5, i13 ; c8 d8 + + movdqa xmm6, i14 + movdqa xmm7, xmm6 + punpcklbw xmm6, i15 ; e0 f0 + punpckhbw xmm7, i15 ; e8 f8 + + movdqa xmm8, xmm3 + punpcklwd xmm3, xmm6 ; c0 d0 e0 f0 + punpckhwd xmm8, xmm6 ; c4 d4 e4 f4 + + movdqa xmm6, xmm5 + punpcklwd xmm5, xmm7 ; c8 d8 e8 f8 + punpckhwd xmm6, xmm7 ; cc dc ec fc + + ; pull the third and fourth sets together + + movdqa xmm7, xmm0 + punpckldq xmm0, xmm3 ; 80 90 a0 b0 c0 d0 e0 f0 + punpckhdq xmm7, xmm3 ; 82 92 a2 b2 c2 d2 e2 f2 + + movdqa xmm3, xmm4 + punpckldq xmm4, xmm8 ; 84 94 a4 b4 c4 d4 e4 f4 + punpckhdq xmm3, xmm8 ; 86 96 a6 b6 c6 d6 e6 f6 + + movdqa xmm8, xmm1 + punpckldq xmm1, xmm5 ; 88 88 a8 b8 c8 d8 e8 f8 + punpckhdq xmm8, xmm5 ; 8a 9a aa ba ca da ea fa + + movdqa xmm5, xmm2 + punpckldq xmm2, xmm6 ; 8c 9c ac bc cc dc ec fc + punpckhdq xmm5, xmm6 ; 8e 9e ae be ce de ee fe + + ; save the calculations. we only have 15 registers ... + movdqa i8, xmm0 + movdqa i9, xmm7 + movdqa i10, xmm4 + movdqa i11, xmm3 + movdqa i12, xmm1 + movdqa i13, xmm8 + movdqa i14, xmm2 + movdqa i15, xmm5 + + ; 0-7 + movdqa xmm0, i0 + movdqa xmm1, xmm0 + punpcklbw xmm0, i1 ; 00 10 + punpckhbw xmm1, i1 ; 08 18 + + movdqa xmm2, i2 + movdqa xmm3, xmm2 + punpcklbw xmm2, i3 ; 20 30 + punpckhbw xmm3, i3 ; 28 38 + + movdqa xmm4, xmm0 + punpcklwd xmm0, xmm2 ; 00 10 20 30 + punpckhwd xmm4, xmm2 ; 04 14 24 34 + + movdqa xmm2, xmm1 + punpcklwd xmm1, xmm3 ; 08 18 28 38 + punpckhwd xmm2, xmm3 ; 0c 1c 2c 3c + + ; using xmm[0124] + ; work on next 4 rows + + movdqa xmm3, i4 + movdqa xmm5, xmm3 + punpcklbw xmm3, i5 ; 40 50 + punpckhbw xmm5, i5 ; 48 58 + + movdqa xmm6, i6 + movdqa xmm7, xmm6 + punpcklbw xmm6, i7 ; 60 70 + punpckhbw xmm7, i7 ; 68 78 + + movdqa xmm8, xmm3 + punpcklwd xmm3, xmm6 ; 40 50 60 70 + punpckhwd xmm8, xmm6 ; 44 54 64 74 + + movdqa xmm6, xmm5 + punpcklwd xmm5, xmm7 ; 48 58 68 78 + punpckhwd xmm6, xmm7 ; 4c 5c 6c 7c + + ; pull the first two sets together + + movdqa xmm7, xmm0 + punpckldq xmm0, xmm3 ; 00 10 20 30 40 50 60 70 + punpckhdq xmm7, xmm3 ; 02 12 22 32 42 52 62 72 + + movdqa xmm3, xmm4 + punpckldq xmm4, xmm8 ; 04 14 24 34 44 54 64 74 + punpckhdq xmm3, xmm8 ; 06 16 26 36 46 56 66 76 + + movdqa xmm8, xmm1 + punpckldq xmm1, xmm5 ; 08 18 28 38 48 58 68 78 + punpckhdq xmm8, xmm5 ; 0a 1a 2a 3a 4a 5a 6a 7a + + movdqa xmm5, xmm2 + punpckldq xmm2, xmm6 ; 0c 1c 2c 3c 4c 5c 6c 7c + punpckhdq xmm5, xmm6 ; 0e 1e 2e 3e 4e 5e 6e 7e + ; final combination + + movdqa xmm6, xmm0 + punpcklqdq xmm0, i8 + punpckhqdq xmm6, i8 + + movdqa xmm9, xmm7 + punpcklqdq xmm7, i9 + punpckhqdq xmm9, i9 + + movdqa xmm10, xmm4 + punpcklqdq xmm4, i10 + punpckhqdq xmm10, i10 + + movdqa xmm11, xmm3 + punpcklqdq xmm3, i11 + punpckhqdq xmm11, i11 + + movdqa xmm12, xmm1 + punpcklqdq xmm1, i12 + punpckhqdq xmm12, i12 + + movdqa xmm13, xmm8 + punpcklqdq xmm8, i13 + punpckhqdq xmm13, i13 + + movdqa xmm14, xmm2 + punpcklqdq xmm2, i14 + punpckhqdq xmm14, i14 + + movdqa xmm15, xmm5 + punpcklqdq xmm5, i15 + punpckhqdq xmm15, i15 + + movdqa s0, xmm0 + movdqa s1, xmm6 + movdqa s2, xmm7 + movdqa s3, xmm9 + movdqa s4, xmm4 + movdqa s5, xmm10 + movdqa s6, xmm3 + movdqa s7, xmm11 + movdqa s8, xmm1 + movdqa s9, xmm12 + movdqa s10, xmm8 + movdqa s11, xmm13 + movdqa s12, xmm2 + movdqa s13, xmm14 + movdqa s14, xmm5 + movdqa s15, xmm15 + + ; free stack space + add rsp, stack_size + + ; un-ALIGN_STACK + pop rsp + +%if LIBVPX_YASM_WIN64 + pop r13 + pop r12 + RESTORE_XMM + pop rbp +%endif + + ret + +SECTION_RODATA +align 16 +te0: + times 16 db 0xe0 +align 16 +t7f: + times 16 db 0x7f +align 16 +tfe: + times 16 db 0xfe +align 16 +t1f: + times 16 db 0x1f +align 16 +t80: + times 16 db 0x80 +align 16 +t1: + times 16 db 0x01 +align 16 +t3: + times 16 db 0x03 +align 16 +t4: + times 16 db 0x04 diff --git a/thirdparty/libvpx/vp8/common/x86/loopfilter_sse2.asm b/thirdparty/libvpx/vp8/common/x86/loopfilter_sse2.asm new file mode 100644 index 0000000000..1913abc69b --- /dev/null +++ b/thirdparty/libvpx/vp8/common/x86/loopfilter_sse2.asm @@ -0,0 +1,1640 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + + +%include "vpx_ports/x86_abi_support.asm" +%define _t0 0 +%define _t1 _t0 + 16 +%define _p3 _t1 + 16 +%define _p2 _p3 + 16 +%define _p1 _p2 + 16 +%define _p0 _p1 + 16 +%define _q0 _p0 + 16 +%define _q1 _q0 + 16 +%define _q2 _q1 + 16 +%define _q3 _q2 + 16 +%define lf_var_size 160 + +; Use of pmaxub instead of psubusb to compute filter mask was seen +; in ffvp8 + +%macro LFH_FILTER_AND_HEV_MASK 1 +%if %1 + movdqa xmm2, [rdi+2*rax] ; q3 + movdqa xmm1, [rsi+2*rax] ; q2 + movdqa xmm4, [rsi+rax] ; q1 + movdqa xmm5, [rsi] ; q0 + neg rax ; negate pitch to deal with above border +%else + movlps xmm2, [rsi + rcx*2] ; q3 + movlps xmm1, [rsi + rcx] ; q2 + movlps xmm4, [rsi] ; q1 + movlps xmm5, [rsi + rax] ; q0 + + movhps xmm2, [rdi + rcx*2] + movhps xmm1, [rdi + rcx] + movhps xmm4, [rdi] + movhps xmm5, [rdi + rax] + + lea rsi, [rsi + rax*4] + lea rdi, [rdi + rax*4] + + movdqa [rsp+_q2], xmm1 ; store q2 + movdqa [rsp+_q1], xmm4 ; store q1 +%endif + movdqa xmm7, [rdx] ;limit + + movdqa xmm6, xmm1 ; q2 + movdqa xmm3, xmm4 ; q1 + + psubusb xmm1, xmm2 ; q2-=q3 + psubusb xmm2, xmm6 ; q3-=q2 + + psubusb xmm4, xmm6 ; q1-=q2 + psubusb xmm6, xmm3 ; q2-=q1 + + por xmm4, xmm6 ; abs(q2-q1) + por xmm1, xmm2 ; abs(q3-q2) + + movdqa xmm0, xmm5 ; q0 + pmaxub xmm1, xmm4 + + psubusb xmm5, xmm3 ; q0-=q1 + psubusb xmm3, xmm0 ; q1-=q0 + + por xmm5, xmm3 ; abs(q0-q1) + movdqa [rsp+_t0], xmm5 ; save to t0 + + pmaxub xmm1, xmm5 + +%if %1 + movdqa xmm2, [rsi+4*rax] ; p3 + movdqa xmm4, [rdi+4*rax] ; p2 + movdqa xmm6, [rsi+2*rax] ; p1 +%else + movlps xmm2, [rsi + rax] ; p3 + movlps xmm4, [rsi] ; p2 + movlps xmm6, [rsi + rcx] ; p1 + + movhps xmm2, [rdi + rax] + movhps xmm4, [rdi] + movhps xmm6, [rdi + rcx] + + movdqa [rsp+_p2], xmm4 ; store p2 + movdqa [rsp+_p1], xmm6 ; store p1 +%endif + + movdqa xmm5, xmm4 ; p2 + movdqa xmm3, xmm6 ; p1 + + psubusb xmm4, xmm2 ; p2-=p3 + psubusb xmm2, xmm5 ; p3-=p2 + + psubusb xmm3, xmm5 ; p1-=p2 + pmaxub xmm1, xmm4 ; abs(p3 - p2) + + psubusb xmm5, xmm6 ; p2-=p1 + pmaxub xmm1, xmm2 ; abs(p3 - p2) + + pmaxub xmm1, xmm5 ; abs(p2 - p1) + movdqa xmm2, xmm6 ; p1 + + pmaxub xmm1, xmm3 ; abs(p2 - p1) +%if %1 + movdqa xmm4, [rsi+rax] ; p0 + movdqa xmm3, [rdi] ; q1 +%else + movlps xmm4, [rsi + rcx*2] ; p0 + movhps xmm4, [rdi + rcx*2] + movdqa xmm3, [rsp+_q1] ; q1 +%endif + + movdqa xmm5, xmm4 ; p0 + psubusb xmm4, xmm6 ; p0-=p1 + + psubusb xmm6, xmm5 ; p1-=p0 + + por xmm6, xmm4 ; abs(p1 - p0) + mov rdx, arg(2) ; get blimit + + movdqa [rsp+_t1], xmm6 ; save to t1 + + movdqa xmm4, xmm3 ; q1 + pmaxub xmm1, xmm6 + + psubusb xmm3, xmm2 ; q1-=p1 + psubusb xmm2, xmm4 ; p1-=q1 + + psubusb xmm1, xmm7 + por xmm2, xmm3 ; abs(p1-q1) + + movdqa xmm7, [rdx] ; blimit + mov rdx, arg(4) ; hev get thresh + + movdqa xmm3, xmm0 ; q0 + pand xmm2, [GLOBAL(tfe)] ; set lsb of each byte to zero + + movdqa xmm6, xmm5 ; p0 + psrlw xmm2, 1 ; abs(p1-q1)/2 + + psubusb xmm5, xmm3 ; p0-=q0 + psubusb xmm3, xmm6 ; q0-=p0 + por xmm5, xmm3 ; abs(p0 - q0) + + paddusb xmm5, xmm5 ; abs(p0-q0)*2 + + movdqa xmm4, [rsp+_t0] ; hev get abs (q1 - q0) + movdqa xmm3, [rsp+_t1] ; get abs (p1 - p0) + + paddusb xmm5, xmm2 ; abs (p0 - q0) *2 + abs(p1-q1)/2 + + movdqa xmm2, [rdx] ; hev + + psubusb xmm5, xmm7 ; abs (p0 - q0) *2 + abs(p1-q1)/2 > blimit + psubusb xmm4, xmm2 ; hev + + psubusb xmm3, xmm2 ; hev + por xmm1, xmm5 + + pxor xmm7, xmm7 + paddb xmm4, xmm3 ; hev abs(q1 - q0) > thresh || abs(p1 - p0) > thresh + + pcmpeqb xmm4, xmm5 ; hev + pcmpeqb xmm3, xmm3 ; hev + + pcmpeqb xmm1, xmm7 ; mask xmm1 + pxor xmm4, xmm3 ; hev +%endmacro + +%macro B_FILTER 1 + movdqa xmm3, [GLOBAL(t80)] +%if %1 == 0 + movdqa xmm2, [rsp+_p1] ; p1 + movdqa xmm7, [rsp+_q1] ; q1 +%elif %1 == 1 + movdqa xmm2, [rsi+2*rax] ; p1 + movdqa xmm7, [rdi] ; q1 +%elif %1 == 2 + movdqa xmm2, [rsp+_p1] ; p1 + movdqa xmm6, [rsp+_p0] ; p0 + movdqa xmm0, [rsp+_q0] ; q0 + movdqa xmm7, [rsp+_q1] ; q1 +%endif + + pxor xmm2, xmm3 ; p1 offset to convert to signed values + pxor xmm7, xmm3 ; q1 offset to convert to signed values + + psubsb xmm2, xmm7 ; p1 - q1 + pxor xmm6, xmm3 ; offset to convert to signed values + + pand xmm2, xmm4 ; high var mask (hvm)(p1 - q1) + pxor xmm0, xmm3 ; offset to convert to signed values + + movdqa xmm3, xmm0 ; q0 + psubsb xmm0, xmm6 ; q0 - p0 + paddsb xmm2, xmm0 ; 1 * (q0 - p0) + hvm(p1 - q1) + paddsb xmm2, xmm0 ; 2 * (q0 - p0) + hvm(p1 - q1) + paddsb xmm2, xmm0 ; 3 * (q0 - p0) + hvm(p1 - q1) + pand xmm1, xmm2 ; mask filter values we don't care about + + movdqa xmm2, xmm1 + paddsb xmm1, [GLOBAL(t4)] ; 3* (q0 - p0) + hvm(p1 - q1) + 4 + paddsb xmm2, [GLOBAL(t3)] ; 3* (q0 - p0) + hvm(p1 - q1) + 3 + + punpckhbw xmm5, xmm2 ; axbxcxdx + punpcklbw xmm2, xmm2 ; exfxgxhx + + punpcklbw xmm0, xmm1 ; exfxgxhx + psraw xmm5, 11 ; sign extended shift right by 3 + + punpckhbw xmm1, xmm1 ; axbxcxdx + psraw xmm2, 11 ; sign extended shift right by 3 + + packsswb xmm2, xmm5 ; (3* (q0 - p0) + hvm(p1 - q1) + 3) >> 3; + psraw xmm0, 11 ; sign extended shift right by 3 + + psraw xmm1, 11 ; sign extended shift right by 3 + movdqa xmm5, xmm0 ; save results + + packsswb xmm0, xmm1 ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>3 + + paddsb xmm6, xmm2 ; p0+= p0 add + + movdqa xmm2, [GLOBAL(ones)] + paddsw xmm5, xmm2 + paddsw xmm1, xmm2 + psraw xmm5, 1 ; partial shifted one more time for 2nd tap + psraw xmm1, 1 ; partial shifted one more time for 2nd tap + packsswb xmm5, xmm1 ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>4 + movdqa xmm2, [GLOBAL(t80)] + +%if %1 == 0 + movdqa xmm1, [rsp+_p1] ; p1 + lea rsi, [rsi + rcx*2] + lea rdi, [rdi + rcx*2] +%elif %1 == 1 + movdqa xmm1, [rsi+2*rax] ; p1 +%elif %1 == 2 + movdqa xmm1, [rsp+_p1] ; p1 +%endif + + pandn xmm4, xmm5 ; high edge variance additive + pxor xmm6, xmm2 ; unoffset + + pxor xmm1, xmm2 ; reoffset + psubsb xmm3, xmm0 ; q0-= q0 add + + paddsb xmm1, xmm4 ; p1+= p1 add + pxor xmm3, xmm2 ; unoffset + + pxor xmm1, xmm2 ; unoffset + psubsb xmm7, xmm4 ; q1-= q1 add + + pxor xmm7, xmm2 ; unoffset +%if %1 == 0 + movq [rsi], xmm6 ; p0 + movhps [rdi], xmm6 + movq [rsi + rax], xmm1 ; p1 + movhps [rdi + rax], xmm1 + movq [rsi + rcx], xmm3 ; q0 + movhps [rdi + rcx], xmm3 + movq [rsi + rcx*2], xmm7 ; q1 + movhps [rdi + rcx*2], xmm7 +%elif %1 == 1 + movdqa [rsi+rax], xmm6 ; write back + movdqa [rsi+2*rax], xmm1 ; write back + movdqa [rsi], xmm3 ; write back + movdqa [rdi], xmm7 ; write back +%endif + +%endmacro + +%if ABI_IS_32BIT + +;void vp8_loop_filter_horizontal_edge_sse2 +;( +; unsigned char *src_ptr, +; int src_pixel_step, +; const char *blimit, +; const char *limit, +; const char *thresh, +;) +global sym(vp8_loop_filter_horizontal_edge_sse2) PRIVATE +sym(vp8_loop_filter_horizontal_edge_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 5 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, lf_var_size + + mov rsi, arg(0) ;src_ptr + movsxd rax, dword ptr arg(1) ;src_pixel_step + + mov rdx, arg(3) ;limit + + lea rdi, [rsi+rax] ; rdi points to row +1 for indirect addressing + + ; calculate breakout conditions and high edge variance + LFH_FILTER_AND_HEV_MASK 1 + ; filter and write back the result + B_FILTER 1 + + add rsp, lf_var_size + pop rsp + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +%endif + +;void vp8_loop_filter_horizontal_edge_uv_sse2 +;( +; unsigned char *src_ptr, +; int src_pixel_step, +; const char *blimit, +; const char *limit, +; const char *thresh, +; int count +;) +global sym(vp8_loop_filter_horizontal_edge_uv_sse2) PRIVATE +sym(vp8_loop_filter_horizontal_edge_uv_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, lf_var_size + + mov rsi, arg(0) ; u + mov rdi, arg(5) ; v + movsxd rax, dword ptr arg(1) ; src_pixel_step + mov rcx, rax + neg rax ; negate pitch to deal with above border + + mov rdx, arg(3) ;limit + + lea rsi, [rsi + rcx] + lea rdi, [rdi + rcx] + + ; calculate breakout conditions and high edge variance + LFH_FILTER_AND_HEV_MASK 0 + ; filter and write back the result + B_FILTER 0 + + add rsp, lf_var_size + pop rsp + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + + +%macro MB_FILTER_AND_WRITEBACK 1 + movdqa xmm3, [GLOBAL(t80)] +%if %1 == 0 + movdqa xmm2, [rsp+_p1] ; p1 + movdqa xmm7, [rsp+_q1] ; q1 +%elif %1 == 1 + movdqa xmm2, [rsi+2*rax] ; p1 + movdqa xmm7, [rdi] ; q1 + + mov rcx, rax + neg rcx +%elif %1 == 2 + movdqa xmm2, [rsp+_p1] ; p1 + movdqa xmm6, [rsp+_p0] ; p0 + movdqa xmm0, [rsp+_q0] ; q0 + movdqa xmm7, [rsp+_q1] ; q1 +%endif + + pxor xmm2, xmm3 ; p1 offset to convert to signed values + pxor xmm7, xmm3 ; q1 offset to convert to signed values + pxor xmm6, xmm3 ; offset to convert to signed values + pxor xmm0, xmm3 ; offset to convert to signed values + + psubsb xmm2, xmm7 ; p1 - q1 + + movdqa xmm3, xmm0 ; q0 + psubsb xmm0, xmm6 ; q0 - p0 + paddsb xmm2, xmm0 ; 1 * (q0 - p0) + (p1 - q1) + paddsb xmm2, xmm0 ; 2 * (q0 - p0) + paddsb xmm2, xmm0 ; 3 * (q0 - p0) + (p1 - q1) + pand xmm1, xmm2 ; mask filter values we don't care about + + movdqa xmm2, xmm1 ; vp8_filter + + pand xmm2, xmm4 ; Filter2 = vp8_filter & hev + pxor xmm0, xmm0 + + pandn xmm4, xmm1 ; vp8_filter&=~hev + pxor xmm1, xmm1 + + punpcklbw xmm0, xmm4 ; Filter 2 (hi) + punpckhbw xmm1, xmm4 ; Filter 2 (lo) + + movdqa xmm5, xmm2 + + movdqa xmm4, [GLOBAL(s9)] + paddsb xmm5, [GLOBAL(t3)] ; vp8_signed_char_clamp(Filter2 + 3) + paddsb xmm2, [GLOBAL(t4)] ; vp8_signed_char_clamp(Filter2 + 4) + + pmulhw xmm1, xmm4 ; Filter 2 (lo) * 9 + pmulhw xmm0, xmm4 ; Filter 2 (hi) * 9 + + punpckhbw xmm7, xmm5 ; axbxcxdx + punpcklbw xmm5, xmm5 ; exfxgxhx + + psraw xmm7, 11 ; sign extended shift right by 3 + + psraw xmm5, 11 ; sign extended shift right by 3 + punpckhbw xmm4, xmm2 ; axbxcxdx + + punpcklbw xmm2, xmm2 ; exfxgxhx + psraw xmm4, 11 ; sign extended shift right by 3 + + packsswb xmm5, xmm7 ; Filter2 >>=3; + psraw xmm2, 11 ; sign extended shift right by 3 + + packsswb xmm2, xmm4 ; Filter1 >>=3; + + paddsb xmm6, xmm5 ; ps0 =ps0 + Fitler2 + + psubsb xmm3, xmm2 ; qs0 =qs0 - Filter1 + movdqa xmm7, xmm1 + + movdqa xmm4, [GLOBAL(s63)] + movdqa xmm5, xmm0 + movdqa xmm2, xmm5 + paddw xmm0, xmm4 ; Filter 2 (hi) * 9 + 63 + paddw xmm1, xmm4 ; Filter 2 (lo) * 9 + 63 + movdqa xmm4, xmm7 + + paddw xmm5, xmm5 ; Filter 2 (hi) * 18 + + paddw xmm7, xmm7 ; Filter 2 (lo) * 18 + paddw xmm5, xmm0 ; Filter 2 (hi) * 27 + 63 + + paddw xmm7, xmm1 ; Filter 2 (lo) * 27 + 63 + paddw xmm2, xmm0 ; Filter 2 (hi) * 18 + 63 + psraw xmm0, 7 ; (Filter 2 (hi) * 9 + 63) >> 7 + + paddw xmm4, xmm1 ; Filter 2 (lo) * 18 + 63 + psraw xmm1, 7 ; (Filter 2 (lo) * 9 + 63) >> 7 + psraw xmm2, 7 ; (Filter 2 (hi) * 18 + 63) >> 7 + + packsswb xmm0, xmm1 ; u1 = vp8_signed_char_clamp((63 + Filter2 * 9)>>7) + + psraw xmm4, 7 ; (Filter 2 (lo) * 18 + 63) >> 7 + psraw xmm5, 7 ; (Filter 2 (hi) * 27 + 63) >> 7 + psraw xmm7, 7 ; (Filter 2 (lo) * 27 + 63) >> 7 + + packsswb xmm5, xmm7 ; u3 = vp8_signed_char_clamp((63 + Filter2 * 27)>>7) + packsswb xmm2, xmm4 ; u2 = vp8_signed_char_clamp((63 + Filter2 * 18)>>7) + movdqa xmm7, [GLOBAL(t80)] + +%if %1 == 0 + movdqa xmm1, [rsp+_q1] ; q1 + movdqa xmm4, [rsp+_p1] ; p1 + lea rsi, [rsi+rcx*2] + lea rdi, [rdi+rcx*2] + +%elif %1 == 1 + movdqa xmm1, [rdi] ; q1 + movdqa xmm4, [rsi+rax*2] ; p1 +%elif %1 == 2 + movdqa xmm4, [rsp+_p1] ; p1 + movdqa xmm1, [rsp+_q1] ; q1 +%endif + + pxor xmm1, xmm7 + pxor xmm4, xmm7 + + psubsb xmm3, xmm5 ; sq = vp8_signed_char_clamp(qs0 - u3) + paddsb xmm6, xmm5 ; sp = vp8_signed_char_clamp(ps0 - u3) + psubsb xmm1, xmm2 ; sq = vp8_signed_char_clamp(qs1 - u2) + paddsb xmm4, xmm2 ; sp = vp8_signed_char_clamp(ps1 - u2) + +%if %1 == 1 + movdqa xmm2, [rdi+rax*4] ; p2 + movdqa xmm5, [rdi+rcx] ; q2 +%else + movdqa xmm2, [rsp+_p2] ; p2 + movdqa xmm5, [rsp+_q2] ; q2 +%endif + + pxor xmm1, xmm7 ; *oq1 = sq^0x80; + pxor xmm4, xmm7 ; *op1 = sp^0x80; + pxor xmm2, xmm7 + pxor xmm5, xmm7 + paddsb xmm2, xmm0 ; sp = vp8_signed_char_clamp(ps2 - u) + psubsb xmm5, xmm0 ; sq = vp8_signed_char_clamp(qs2 - u) + pxor xmm2, xmm7 ; *op2 = sp^0x80; + pxor xmm5, xmm7 ; *oq2 = sq^0x80; + pxor xmm3, xmm7 ; *oq0 = sq^0x80 + pxor xmm6, xmm7 ; *oq0 = sp^0x80 +%if %1 == 0 + movq [rsi], xmm6 ; p0 + movhps [rdi], xmm6 + movq [rsi + rcx], xmm3 ; q0 + movhps [rdi + rcx], xmm3 + lea rdx, [rcx + rcx*2] + movq [rsi+rcx*2], xmm1 ; q1 + movhps [rdi+rcx*2], xmm1 + + movq [rsi + rax], xmm4 ; p1 + movhps [rdi + rax], xmm4 + + movq [rsi+rax*2], xmm2 ; p2 + movhps [rdi+rax*2], xmm2 + + movq [rsi+rdx], xmm5 ; q2 + movhps [rdi+rdx], xmm5 +%elif %1 == 1 + movdqa [rdi+rcx], xmm5 ; q2 + movdqa [rdi], xmm1 ; q1 + movdqa [rsi], xmm3 ; q0 + movdqa [rsi+rax ], xmm6 ; p0 + movdqa [rsi+rax*2], xmm4 ; p1 + movdqa [rdi+rax*4], xmm2 ; p2 +%elif %1 == 2 + movdqa [rsp+_p1], xmm4 ; p1 + movdqa [rsp+_p0], xmm6 ; p0 + movdqa [rsp+_q0], xmm3 ; q0 + movdqa [rsp+_q1], xmm1 ; q1 +%endif + +%endmacro + + +;void vp8_mbloop_filter_horizontal_edge_sse2 +;( +; unsigned char *src_ptr, +; int src_pixel_step, +; const char *blimit, +; const char *limit, +; const char *thresh, +;) +global sym(vp8_mbloop_filter_horizontal_edge_sse2) PRIVATE +sym(vp8_mbloop_filter_horizontal_edge_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 5 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, lf_var_size + + mov rsi, arg(0) ;src_ptr + movsxd rax, dword ptr arg(1) ;src_pixel_step + mov rdx, arg(3) ;limit + + lea rdi, [rsi+rax] ; rdi points to row +1 for indirect addressing + + ; calculate breakout conditions and high edge variance + LFH_FILTER_AND_HEV_MASK 1 + ; filter and write back the results + MB_FILTER_AND_WRITEBACK 1 + + add rsp, lf_var_size + pop rsp + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + + +;void vp8_mbloop_filter_horizontal_edge_uv_sse2 +;( +; unsigned char *u, +; int src_pixel_step, +; const char *blimit, +; const char *limit, +; const char *thresh, +; unsigned char *v +;) +global sym(vp8_mbloop_filter_horizontal_edge_uv_sse2) PRIVATE +sym(vp8_mbloop_filter_horizontal_edge_uv_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, lf_var_size + + mov rsi, arg(0) ; u + mov rdi, arg(5) ; v + movsxd rax, dword ptr arg(1) ; src_pixel_step + mov rcx, rax + neg rax ; negate pitch to deal with above border + mov rdx, arg(3) ;limit + + lea rsi, [rsi + rcx] + lea rdi, [rdi + rcx] + + ; calculate breakout conditions and high edge variance + LFH_FILTER_AND_HEV_MASK 0 + ; filter and write back the results + MB_FILTER_AND_WRITEBACK 0 + + add rsp, lf_var_size + pop rsp + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + + +%macro TRANSPOSE_16X8 2 + movq xmm4, [rsi] ; xx xx xx xx xx xx xx xx 07 06 05 04 03 02 01 00 + movq xmm1, [rdi] ; xx xx xx xx xx xx xx xx 17 16 15 14 13 12 11 10 + movq xmm0, [rsi+2*rax] ; xx xx xx xx xx xx xx xx 27 26 25 24 23 22 21 20 + movq xmm7, [rdi+2*rax] ; xx xx xx xx xx xx xx xx 37 36 35 34 33 32 31 30 + movq xmm5, [rsi+4*rax] ; xx xx xx xx xx xx xx xx 47 46 45 44 43 42 41 40 + movq xmm2, [rdi+4*rax] ; xx xx xx xx xx xx xx xx 57 56 55 54 53 52 51 50 + + punpcklbw xmm4, xmm1 ; 17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00 + + movq xmm1, [rdi+2*rcx] ; xx xx xx xx xx xx xx xx 77 76 75 74 73 72 71 70 + + movdqa xmm3, xmm4 ; 17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00 + punpcklbw xmm0, xmm7 ; 37 27 36 36 35 25 34 24 33 23 32 22 31 21 30 20 + + movq xmm7, [rsi+2*rcx] ; xx xx xx xx xx xx xx xx 67 66 65 64 63 62 61 60 + + punpcklbw xmm5, xmm2 ; 57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40 +%if %1 + lea rsi, [rsi+rax*8] + lea rdi, [rdi+rax*8] +%else + mov rsi, arg(5) ; v_ptr +%endif + + movdqa xmm6, xmm5 ; 57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40 + punpcklbw xmm7, xmm1 ; 77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60 + punpcklwd xmm5, xmm7 ; 73 63 53 43 72 62 52 42 71 61 51 41 70 60 50 40 + punpckhwd xmm6, xmm7 ; 77 67 57 47 76 66 56 46 75 65 55 45 74 64 54 44 + punpcklwd xmm3, xmm0 ; 33 23 13 03 32 22 12 02 31 21 11 01 30 20 10 00 + +%if %1 == 0 + lea rdi, [rsi + rax - 4] ; rdi points to row +1 for indirect addressing + lea rsi, [rsi - 4] +%endif + + movdqa xmm2, xmm3 ; 33 23 13 03 32 22 12 02 31 21 11 01 30 20 10 00 + punpckhwd xmm4, xmm0 ; 37 27 17 07 36 26 16 06 35 25 15 05 34 24 14 04 + + movdqa xmm7, xmm4 ; 37 27 17 07 36 26 16 06 35 25 15 05 34 24 14 04 + punpckhdq xmm3, xmm5 ; 73 63 53 43 33 23 13 03 72 62 52 42 32 22 12 02 + + punpckhdq xmm7, xmm6 ; 77 67 57 47 37 27 17 07 76 66 56 46 36 26 16 06 + + punpckldq xmm4, xmm6 ; 75 65 55 45 35 25 15 05 74 64 54 44 34 24 14 04 + + punpckldq xmm2, xmm5 ; 71 61 51 41 31 21 11 01 70 60 50 40 30 20 10 00 + + movdqa [rsp+_t0], xmm2 ; save to free XMM2 + + movq xmm2, [rsi] ; xx xx xx xx xx xx xx xx 87 86 85 84 83 82 81 80 + movq xmm6, [rdi] ; xx xx xx xx xx xx xx xx 97 96 95 94 93 92 91 90 + movq xmm0, [rsi+2*rax] ; xx xx xx xx xx xx xx xx a7 a6 a5 a4 a3 a2 a1 a0 + movq xmm5, [rdi+2*rax] ; xx xx xx xx xx xx xx xx b7 b6 b5 b4 b3 b2 b1 b0 + movq xmm1, [rsi+4*rax] ; xx xx xx xx xx xx xx xx c7 c6 c5 c4 c3 c2 c1 c0 + + punpcklbw xmm2, xmm6 ; 97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80 + + movq xmm6, [rdi+4*rax] ; xx xx xx xx xx xx xx xx d7 d6 d5 d4 d3 d2 d1 d0 + + punpcklbw xmm0, xmm5 ; b7 a7 b6 a6 b5 a5 b4 a4 b3 a3 b2 a2 b1 a1 b0 a0 + + movq xmm5, [rsi+2*rcx] ; xx xx xx xx xx xx xx xx e7 e6 e5 e4 e3 e2 e1 e0 + + punpcklbw xmm1, xmm6 ; d7 c7 d6 c6 d5 c5 d4 c4 d3 c3 d2 c2 d1 e1 d0 c0 + + movq xmm6, [rdi+2*rcx] ; xx xx xx xx xx xx xx xx f7 f6 f5 f4 f3 f2 f1 f0 + + punpcklbw xmm5, xmm6 ; f7 e7 f6 e6 f5 e5 f4 e4 f3 e3 f2 e2 f1 e1 f0 e0 + + movdqa xmm6, xmm1 ; + punpckhwd xmm6, xmm5 ; f7 e7 d7 c7 f6 e6 d6 c6 f5 e5 d5 c5 f4 e4 d4 c4 + + punpcklwd xmm1, xmm5 ; f3 e3 d3 c3 f2 e2 d2 c2 f1 e1 d1 c1 f0 e0 d0 c0 + movdqa xmm5, xmm2 ; 97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80 + + punpcklwd xmm5, xmm0 ; b3 a3 93 83 b2 a2 92 82 b1 a1 91 81 b0 a0 90 80 + + punpckhwd xmm2, xmm0 ; b7 a7 97 87 b6 a6 96 86 b5 a5 95 85 b4 a4 94 84 + + movdqa xmm0, xmm5 + punpckldq xmm0, xmm1 ; f1 e1 d1 c1 b1 a1 91 81 f0 e0 d0 c0 b0 a0 90 80 + + punpckhdq xmm5, xmm1 ; f3 e3 d3 c3 b3 a3 93 83 f2 e2 d2 c2 b2 a2 92 82 + movdqa xmm1, xmm2 ; b7 a7 97 87 b6 a6 96 86 b5 a5 95 85 b4 a4 94 84 + + punpckldq xmm1, xmm6 ; f5 e5 d5 c5 b5 a5 95 85 f4 e4 d4 c4 b4 a4 94 84 + + punpckhdq xmm2, xmm6 ; f7 e7 d7 c7 b7 a7 97 87 f6 e6 d6 c6 b6 a6 96 86 + movdqa xmm6, xmm7 ; 77 67 57 47 37 27 17 07 76 66 56 46 36 26 16 06 + + punpcklqdq xmm6, xmm2 ; f6 e6 d6 c6 b6 a6 96 86 76 66 56 46 36 26 16 06 + + punpckhqdq xmm7, xmm2 ; f7 e7 d7 c7 b7 a7 97 87 77 67 57 47 37 27 17 07 + +%if %2 == 0 + movdqa [rsp+_q3], xmm7 ; save 7 + movdqa [rsp+_q2], xmm6 ; save 6 +%endif + movdqa xmm2, xmm3 ; 73 63 53 43 33 23 13 03 72 62 52 42 32 22 12 02 + punpckhqdq xmm3, xmm5 ; f3 e3 d3 c3 b3 a3 93 83 73 63 53 43 33 23 13 03 + punpcklqdq xmm2, xmm5 ; f2 e2 d2 c2 b2 a2 92 82 72 62 52 42 32 22 12 02 + movdqa [rsp+_p1], xmm2 ; save 2 + + movdqa xmm5, xmm4 ; 75 65 55 45 35 25 15 05 74 64 54 44 34 24 14 04 + punpcklqdq xmm4, xmm1 ; f4 e4 d4 c4 b4 a4 94 84 74 64 54 44 34 24 14 04 + movdqa [rsp+_p0], xmm3 ; save 3 + + punpckhqdq xmm5, xmm1 ; f5 e5 d5 c5 b5 a5 95 85 75 65 55 45 35 25 15 05 + + movdqa [rsp+_q0], xmm4 ; save 4 + movdqa [rsp+_q1], xmm5 ; save 5 + movdqa xmm1, [rsp+_t0] + + movdqa xmm2, xmm1 ; + punpckhqdq xmm1, xmm0 ; f1 e1 d1 c1 b1 a1 91 81 71 61 51 41 31 21 11 01 + punpcklqdq xmm2, xmm0 ; f0 e0 d0 c0 b0 a0 90 80 70 60 50 40 30 20 10 00 + +%if %2 == 0 + movdqa [rsp+_p2], xmm1 + movdqa [rsp+_p3], xmm2 +%endif + +%endmacro + +%macro LFV_FILTER_MASK_HEV_MASK 0 + movdqa xmm0, xmm6 ; q2 + psubusb xmm0, xmm7 ; q2-q3 + + psubusb xmm7, xmm6 ; q3-q2 + movdqa xmm4, xmm5 ; q1 + + por xmm7, xmm0 ; abs (q3-q2) + psubusb xmm4, xmm6 ; q1-q2 + + movdqa xmm0, xmm1 + psubusb xmm6, xmm5 ; q2-q1 + + por xmm6, xmm4 ; abs (q2-q1) + psubusb xmm0, xmm2 ; p2 - p3; + + psubusb xmm2, xmm1 ; p3 - p2; + por xmm0, xmm2 ; abs(p2-p3) + + movdqa xmm5, [rsp+_p1] ; p1 + pmaxub xmm0, xmm7 + + movdqa xmm2, xmm5 ; p1 + psubusb xmm5, xmm1 ; p1-p2 + psubusb xmm1, xmm2 ; p2-p1 + + movdqa xmm7, xmm3 ; p0 + psubusb xmm7, xmm2 ; p0-p1 + + por xmm1, xmm5 ; abs(p2-p1) + pmaxub xmm0, xmm6 + + pmaxub xmm0, xmm1 + movdqa xmm1, xmm2 ; p1 + + psubusb xmm2, xmm3 ; p1-p0 + + por xmm2, xmm7 ; abs(p1-p0) + + pmaxub xmm0, xmm2 + + movdqa xmm5, [rsp+_q0] ; q0 + movdqa xmm7, [rsp+_q1] ; q1 + + mov rdx, arg(3) ; limit + + movdqa xmm6, xmm5 ; q0 + movdqa xmm4, xmm7 ; q1 + + psubusb xmm5, xmm7 ; q0-q1 + psubusb xmm7, xmm6 ; q1-q0 + + por xmm7, xmm5 ; abs(q1-q0) + + pmaxub xmm0, xmm7 + + psubusb xmm0, [rdx] ; limit + + mov rdx, arg(2) ; blimit + movdqa xmm5, xmm4 ; q1 + + psubusb xmm5, xmm1 ; q1-=p1 + psubusb xmm1, xmm4 ; p1-=q1 + + por xmm5, xmm1 ; abs(p1-q1) + movdqa xmm1, xmm3 ; p0 + + pand xmm5, [GLOBAL(tfe)] ; set lsb of each byte to zero + psubusb xmm1, xmm6 ; p0-q0 + + movdqa xmm4, [rdx] ; blimit + mov rdx, arg(4) ; get thresh + + psrlw xmm5, 1 ; abs(p1-q1)/2 + psubusb xmm6, xmm3 ; q0-p0 + + por xmm1, xmm6 ; abs(q0-p0) + paddusb xmm1, xmm1 ; abs(q0-p0)*2 + movdqa xmm3, [rdx] + + paddusb xmm1, xmm5 ; abs (p0 - q0) *2 + abs(p1-q1)/2 + psubusb xmm2, xmm3 ; abs(q1 - q0) > thresh + + psubusb xmm7, xmm3 ; abs(p1 - p0)> thresh + + psubusb xmm1, xmm4 ; abs (p0 - q0) *2 + abs(p1-q1)/2 > blimit + por xmm2, xmm7 ; abs(q1 - q0) > thresh || abs(p1 - p0) > thresh + + por xmm1, xmm0 ; mask + pcmpeqb xmm2, xmm0 + + pxor xmm0, xmm0 + pcmpeqb xmm4, xmm4 + + pcmpeqb xmm1, xmm0 + pxor xmm4, xmm2 +%endmacro + +%macro BV_TRANSPOSE 0 + ; xmm1 = f2 e2 d2 c2 b2 a2 92 82 72 62 52 42 32 22 12 02 + ; xmm6 = f3 e3 d3 c3 b3 a3 93 83 73 63 53 43 33 23 13 03 + ; xmm3 = f4 e4 d4 c4 b4 a4 94 84 74 64 54 44 34 24 14 04 + ; xmm7 = f5 e5 d5 c5 b5 a5 95 85 75 65 55 45 35 25 15 05 + movdqa xmm2, xmm1 ; f2 e2 d2 c2 b2 a2 92 82 72 62 52 42 32 22 12 02 + punpcklbw xmm2, xmm6 ; 73 72 63 62 53 52 43 42 33 32 23 22 13 12 03 02 + + movdqa xmm4, xmm3 ; f4 e4 d4 c4 b4 a4 94 84 74 64 54 44 34 24 14 04 + punpckhbw xmm1, xmm6 ; f3 f2 e3 e2 d3 d2 c3 c2 b3 b2 a3 a2 93 92 83 82 + + punpcklbw xmm4, xmm7 ; 75 74 65 64 55 54 45 44 35 34 25 24 15 14 05 04 + + punpckhbw xmm3, xmm7 ; f5 f4 e5 e4 d5 d4 c5 c4 b5 b4 a5 a4 95 94 85 84 + + movdqa xmm6, xmm2 ; 73 72 63 62 53 52 43 42 33 32 23 22 13 12 03 02 + punpcklwd xmm2, xmm4 ; 35 34 33 32 25 24 23 22 15 14 13 12 05 04 03 02 + + punpckhwd xmm6, xmm4 ; 75 74 73 72 65 64 63 62 55 54 53 52 45 44 43 42 + movdqa xmm5, xmm1 ; f3 f2 e3 e2 d3 d2 c3 c2 b3 b2 a3 a2 93 92 83 82 + + punpcklwd xmm1, xmm3 ; b5 b4 b3 b2 a5 a4 a3 a2 95 94 93 92 85 84 83 82 + + punpckhwd xmm5, xmm3 ; f5 f4 f3 f2 e5 e4 e3 e2 d5 d4 d3 d2 c5 c4 c3 c2 + ; xmm2 = 35 34 33 32 25 24 23 22 15 14 13 12 05 04 03 02 + ; xmm6 = 75 74 73 72 65 64 63 62 55 54 53 52 45 44 43 42 + ; xmm1 = b5 b4 b3 b2 a5 a4 a3 a2 95 94 93 92 85 84 83 82 + ; xmm5 = f5 f4 f3 f2 e5 e4 e3 e2 d5 d4 d3 d2 c5 c4 c3 c2 +%endmacro + +%macro BV_WRITEBACK 2 + movd [rsi+2], %1 + movd [rsi+4*rax+2], %2 + psrldq %1, 4 + psrldq %2, 4 + movd [rdi+2], %1 + movd [rdi+4*rax+2], %2 + psrldq %1, 4 + psrldq %2, 4 + movd [rsi+2*rax+2], %1 + movd [rsi+2*rcx+2], %2 + psrldq %1, 4 + psrldq %2, 4 + movd [rdi+2*rax+2], %1 + movd [rdi+2*rcx+2], %2 +%endmacro + +%if ABI_IS_32BIT + +;void vp8_loop_filter_vertical_edge_sse2 +;( +; unsigned char *src_ptr, +; int src_pixel_step, +; const char *blimit, +; const char *limit, +; const char *thresh, +;) +global sym(vp8_loop_filter_vertical_edge_sse2) PRIVATE +sym(vp8_loop_filter_vertical_edge_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 5 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, lf_var_size + + mov rsi, arg(0) ; src_ptr + movsxd rax, dword ptr arg(1) ; src_pixel_step + + lea rsi, [rsi - 4] + lea rdi, [rsi + rax] ; rdi points to row +1 for indirect addressing + lea rcx, [rax*2+rax] + + ;transpose 16x8 to 8x16, and store the 8-line result on stack. + TRANSPOSE_16X8 1, 1 + + ; calculate filter mask and high edge variance + LFV_FILTER_MASK_HEV_MASK + + ; start work on filters + B_FILTER 2 + + ; transpose and write back - only work on q1, q0, p0, p1 + BV_TRANSPOSE + ; store 16-line result + + lea rdx, [rax] + neg rdx + + BV_WRITEBACK xmm1, xmm5 + + lea rsi, [rsi+rdx*8] + lea rdi, [rdi+rdx*8] + BV_WRITEBACK xmm2, xmm6 + + add rsp, lf_var_size + pop rsp + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +%endif + +;void vp8_loop_filter_vertical_edge_uv_sse2 +;( +; unsigned char *u, +; int src_pixel_step, +; const char *blimit, +; const char *limit, +; const char *thresh, +; unsigned char *v +;) +global sym(vp8_loop_filter_vertical_edge_uv_sse2) PRIVATE +sym(vp8_loop_filter_vertical_edge_uv_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, lf_var_size + + mov rsi, arg(0) ; u_ptr + movsxd rax, dword ptr arg(1) ; src_pixel_step + + lea rsi, [rsi - 4] + lea rdi, [rsi + rax] ; rdi points to row +1 for indirect addressing + lea rcx, [rax+2*rax] + + ;transpose 16x8 to 8x16, and store the 8-line result on stack. + TRANSPOSE_16X8 0, 1 + + ; calculate filter mask and high edge variance + LFV_FILTER_MASK_HEV_MASK + + ; start work on filters + B_FILTER 2 + + ; transpose and write back - only work on q1, q0, p0, p1 + BV_TRANSPOSE + + lea rdi, [rsi + rax] ; rdi points to row +1 for indirect addressing + + ; store 16-line result + BV_WRITEBACK xmm1, xmm5 + + mov rsi, arg(0) ; u_ptr + lea rsi, [rsi - 4] + lea rdi, [rsi + rax] ; rdi points to row +1 for indirect addressing + BV_WRITEBACK xmm2, xmm6 + + add rsp, lf_var_size + pop rsp + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +%macro MBV_TRANSPOSE 0 + movdqa xmm0, [rsp+_p3] ; f0 e0 d0 c0 b0 a0 90 80 70 60 50 40 30 20 10 00 + movdqa xmm1, xmm0 ; f0 e0 d0 c0 b0 a0 90 80 70 60 50 40 30 20 10 00 + + punpcklbw xmm0, xmm2 ; 71 70 61 60 51 50 41 40 31 30 21 20 11 10 01 00 + punpckhbw xmm1, xmm2 ; f1 f0 e1 e0 d1 d0 c1 c0 b1 b0 a1 a0 91 90 81 80 + + movdqa xmm7, [rsp+_p1] ; f2 e2 d2 c2 b2 a2 92 82 72 62 52 42 32 22 12 02 + movdqa xmm6, xmm7 ; f2 e2 d2 c2 b2 a2 92 82 72 62 52 42 32 22 12 02 + + punpcklbw xmm7, [rsp+_p0] ; 73 72 63 62 53 52 43 42 33 32 23 22 13 12 03 02 + punpckhbw xmm6, [rsp+_p0] ; f3 f2 e3 e2 d3 d2 c3 c2 b3 b2 a3 a2 93 92 83 82 + + movdqa xmm3, xmm0 ; 71 70 61 60 51 50 41 40 31 30 21 20 11 10 01 00 + punpcklwd xmm0, xmm7 ; 33 32 31 30 23 22 21 20 13 12 11 10 03 02 01 00 + + punpckhwd xmm3, xmm7 ; 73 72 71 70 63 62 61 60 53 52 51 50 43 42 41 40 + movdqa xmm4, xmm1 ; f1 f0 e1 e0 d1 d0 c1 c0 b1 b0 a1 a0 91 90 81 80 + + punpcklwd xmm1, xmm6 ; b3 b2 b1 b0 a3 a2 a1 a0 93 92 91 90 83 82 81 80 + punpckhwd xmm4, xmm6 ; f3 f2 f1 f0 e3 e2 e1 e0 d3 d2 d1 d0 c3 c2 c1 c0 + + movdqa xmm7, [rsp+_q0] ; f4 e4 d4 c4 b4 a4 94 84 74 64 54 44 34 24 14 04 + punpcklbw xmm7, [rsp+_q1] ; 75 74 65 64 55 54 45 44 35 34 25 24 15 14 05 04 + + movdqa xmm6, xmm5 ; f6 e6 d6 c6 b6 a6 96 86 76 66 56 46 36 26 16 06 + punpcklbw xmm6, [rsp+_q3] ; 77 76 67 66 57 56 47 46 37 36 27 26 17 16 07 06 + + movdqa xmm2, xmm7 ; 75 74 65 64 55 54 45 44 35 34 25 24 15 14 05 04 + punpcklwd xmm7, xmm6 ; 37 36 35 34 27 26 25 24 17 16 15 14 07 06 05 04 + + punpckhwd xmm2, xmm6 ; 77 76 75 74 67 66 65 64 57 56 55 54 47 46 45 44 + movdqa xmm6, xmm0 ; 33 32 31 30 23 22 21 20 13 12 11 10 03 02 01 00 + + punpckldq xmm0, xmm7 ; 17 16 15 14 13 12 11 10 07 06 05 04 03 02 01 00 + punpckhdq xmm6, xmm7 ; 37 36 35 34 33 32 31 30 27 26 25 24 23 22 21 20 +%endmacro + +%macro MBV_WRITEBACK_1 0 + movq [rsi], xmm0 + movhps [rdi], xmm0 + + movq [rsi+2*rax], xmm6 + movhps [rdi+2*rax], xmm6 + + movdqa xmm0, xmm3 ; 73 72 71 70 63 62 61 60 53 52 51 50 43 42 41 40 + punpckldq xmm0, xmm2 ; 57 56 55 54 53 52 51 50 47 46 45 44 43 42 41 40 + punpckhdq xmm3, xmm2 ; 77 76 75 74 73 72 71 70 67 66 65 64 63 62 61 60 + + movq [rsi+4*rax], xmm0 + movhps [rdi+4*rax], xmm0 + + movq [rsi+2*rcx], xmm3 + movhps [rdi+2*rcx], xmm3 + + movdqa xmm7, [rsp+_q0] ; f4 e4 d4 c4 b4 a4 94 84 74 64 54 44 34 24 14 04 + punpckhbw xmm7, [rsp+_q1] ; f5 f4 e5 e4 d5 d4 c5 c4 b5 b4 a5 a4 95 94 85 84 + punpckhbw xmm5, [rsp+_q3] ; f7 f6 e7 e6 d7 d6 c7 c6 b7 b6 a7 a6 97 96 87 86 + + movdqa xmm0, xmm7 + punpcklwd xmm0, xmm5 ; b7 b6 b4 b4 a7 a6 a5 a4 97 96 95 94 87 86 85 84 + punpckhwd xmm7, xmm5 ; f7 f6 f5 f4 e7 e6 e5 e4 d7 d6 d5 d4 c7 c6 c5 c4 + + movdqa xmm5, xmm1 ; b3 b2 b1 b0 a3 a2 a1 a0 93 92 91 90 83 82 81 80 + punpckldq xmm1, xmm0 ; 97 96 95 94 93 92 91 90 87 86 85 83 84 82 81 80 + punpckhdq xmm5, xmm0 ; b7 b6 b5 b4 b3 b2 b1 b0 a7 a6 a5 a4 a3 a2 a1 a0 +%endmacro + +%macro MBV_WRITEBACK_2 0 + movq [rsi], xmm1 + movhps [rdi], xmm1 + + movq [rsi+2*rax], xmm5 + movhps [rdi+2*rax], xmm5 + + movdqa xmm1, xmm4 ; f3 f2 f1 f0 e3 e2 e1 e0 d3 d2 d1 d0 c3 c2 c1 c0 + punpckldq xmm1, xmm7 ; d7 d6 d5 d4 d3 d2 d1 d0 c7 c6 c5 c4 c3 c2 c1 c0 + punpckhdq xmm4, xmm7 ; f7 f6 f4 f4 f3 f2 f1 f0 e7 e6 e5 e4 e3 e2 e1 e0 + + movq [rsi+4*rax], xmm1 + movhps [rdi+4*rax], xmm1 + + movq [rsi+2*rcx], xmm4 + movhps [rdi+2*rcx], xmm4 +%endmacro + + +;void vp8_mbloop_filter_vertical_edge_sse2 +;( +; unsigned char *src_ptr, +; int src_pixel_step, +; const char *blimit, +; const char *limit, +; const char *thresh, +;) +global sym(vp8_mbloop_filter_vertical_edge_sse2) PRIVATE +sym(vp8_mbloop_filter_vertical_edge_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 5 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, lf_var_size + + mov rsi, arg(0) ; src_ptr + movsxd rax, dword ptr arg(1) ; src_pixel_step + + lea rsi, [rsi - 4] + lea rdi, [rsi + rax] ; rdi points to row +1 for indirect addressing + lea rcx, [rax*2+rax] + + ; Transpose + TRANSPOSE_16X8 1, 0 + + ; calculate filter mask and high edge variance + LFV_FILTER_MASK_HEV_MASK + + neg rax + ; start work on filters + MB_FILTER_AND_WRITEBACK 2 + + lea rsi, [rsi+rax*8] + lea rdi, [rdi+rax*8] + + ; transpose and write back + MBV_TRANSPOSE + + neg rax + + MBV_WRITEBACK_1 + + + lea rsi, [rsi+rax*8] + lea rdi, [rdi+rax*8] + MBV_WRITEBACK_2 + + add rsp, lf_var_size + pop rsp + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + + +;void vp8_mbloop_filter_vertical_edge_uv_sse2 +;( +; unsigned char *u, +; int src_pixel_step, +; const char *blimit, +; const char *limit, +; const char *thresh, +; unsigned char *v +;) +global sym(vp8_mbloop_filter_vertical_edge_uv_sse2) PRIVATE +sym(vp8_mbloop_filter_vertical_edge_uv_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, lf_var_size + + mov rsi, arg(0) ; u_ptr + movsxd rax, dword ptr arg(1) ; src_pixel_step + + lea rsi, [rsi - 4] + lea rdi, [rsi + rax] ; rdi points to row +1 for indirect addressing + lea rcx, [rax+2*rax] + + ; Transpose + TRANSPOSE_16X8 0, 0 + + ; calculate filter mask and high edge variance + LFV_FILTER_MASK_HEV_MASK + + ; start work on filters + MB_FILTER_AND_WRITEBACK 2 + + ; transpose and write back + MBV_TRANSPOSE + + mov rsi, arg(0) ;u_ptr + lea rsi, [rsi - 4] + lea rdi, [rsi + rax] + MBV_WRITEBACK_1 + mov rsi, arg(5) ;v_ptr + lea rsi, [rsi - 4] + lea rdi, [rsi + rax] + MBV_WRITEBACK_2 + + add rsp, lf_var_size + pop rsp + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + + +;void vp8_loop_filter_simple_horizontal_edge_sse2 +;( +; unsigned char *src_ptr, +; int src_pixel_step, +; const char *blimit, +;) +global sym(vp8_loop_filter_simple_horizontal_edge_sse2) PRIVATE +sym(vp8_loop_filter_simple_horizontal_edge_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 3 + SAVE_XMM 7 + GET_GOT rbx + ; end prolog + + mov rcx, arg(0) ;src_ptr + movsxd rax, dword ptr arg(1) ;src_pixel_step ; destination pitch? + movdqa xmm6, [GLOBAL(tfe)] + lea rdx, [rcx + rax] + neg rax + + ; calculate mask + movdqa xmm0, [rdx] ; q1 + mov rdx, arg(2) ;blimit + movdqa xmm1, [rcx+2*rax] ; p1 + + movdqa xmm2, xmm1 + movdqa xmm3, xmm0 + + psubusb xmm0, xmm1 ; q1-=p1 + psubusb xmm1, xmm3 ; p1-=q1 + por xmm1, xmm0 ; abs(p1-q1) + pand xmm1, xmm6 ; set lsb of each byte to zero + psrlw xmm1, 1 ; abs(p1-q1)/2 + + movdqa xmm7, XMMWORD PTR [rdx] + + movdqa xmm5, [rcx+rax] ; p0 + movdqa xmm4, [rcx] ; q0 + movdqa xmm0, xmm4 ; q0 + movdqa xmm6, xmm5 ; p0 + psubusb xmm5, xmm4 ; p0-=q0 + psubusb xmm4, xmm6 ; q0-=p0 + por xmm5, xmm4 ; abs(p0 - q0) + + movdqa xmm4, [GLOBAL(t80)] + + paddusb xmm5, xmm5 ; abs(p0-q0)*2 + paddusb xmm5, xmm1 ; abs (p0 - q0) *2 + abs(p1-q1)/2 + psubusb xmm5, xmm7 ; abs(p0 - q0) *2 + abs(p1-q1)/2 > blimit + pxor xmm7, xmm7 + pcmpeqb xmm5, xmm7 + + + ; start work on filters + pxor xmm2, xmm4 ; p1 offset to convert to signed values + pxor xmm3, xmm4 ; q1 offset to convert to signed values + psubsb xmm2, xmm3 ; p1 - q1 + + pxor xmm6, xmm4 ; offset to convert to signed values + pxor xmm0, xmm4 ; offset to convert to signed values + movdqa xmm3, xmm0 ; q0 + psubsb xmm0, xmm6 ; q0 - p0 + paddsb xmm2, xmm0 ; p1 - q1 + 1 * (q0 - p0) + paddsb xmm2, xmm0 ; p1 - q1 + 2 * (q0 - p0) + paddsb xmm2, xmm0 ; p1 - q1 + 3 * (q0 - p0) + pand xmm5, xmm2 ; mask filter values we don't care about + + movdqa xmm0, xmm5 + paddsb xmm5, [GLOBAL(t3)] ; 3* (q0 - p0) + (p1 - q1) + 4 + paddsb xmm0, [GLOBAL(t4)] ; +3 instead of +4 + + movdqa xmm1, [GLOBAL(te0)] + movdqa xmm2, [GLOBAL(t1f)] + +; pxor xmm7, xmm7 + pcmpgtb xmm7, xmm0 ;save sign + pand xmm7, xmm1 ;preserve the upper 3 bits + psrlw xmm0, 3 + pand xmm0, xmm2 ;clear out upper 3 bits + por xmm0, xmm7 ;add sign + psubsb xmm3, xmm0 ; q0-= q0sz add + + pxor xmm7, xmm7 + pcmpgtb xmm7, xmm5 ;save sign + pand xmm7, xmm1 ;preserve the upper 3 bits + psrlw xmm5, 3 + pand xmm5, xmm2 ;clear out upper 3 bits + por xmm5, xmm7 ;add sign + paddsb xmm6, xmm5 ; p0+= p0 add + + pxor xmm3, xmm4 ; unoffset + movdqa [rcx], xmm3 ; write back + + pxor xmm6, xmm4 ; unoffset + movdqa [rcx+rax], xmm6 ; write back + + ; begin epilog + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + + +;void vp8_loop_filter_simple_vertical_edge_sse2 +;( +; unsigned char *src_ptr, +; int src_pixel_step, +; const char *blimit, +;) +global sym(vp8_loop_filter_simple_vertical_edge_sse2) PRIVATE +sym(vp8_loop_filter_simple_vertical_edge_sse2): + push rbp ; save old base pointer value. + mov rbp, rsp ; set new base pointer value. + SHADOW_ARGS_TO_STACK 3 + SAVE_XMM 7 + GET_GOT rbx ; save callee-saved reg + push rsi + push rdi + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, 32 ; reserve 32 bytes + %define t0 [rsp + 0] ;__declspec(align(16)) char t0[16]; + %define t1 [rsp + 16] ;__declspec(align(16)) char t1[16]; + + mov rsi, arg(0) ;src_ptr + movsxd rax, dword ptr arg(1) ;src_pixel_step ; destination pitch? + + lea rsi, [rsi - 2 ] + lea rdi, [rsi + rax] + lea rdx, [rsi + rax*4] + lea rcx, [rdx + rax] + + movd xmm0, [rsi] ; (high 96 bits unused) 03 02 01 00 + movd xmm1, [rdx] ; (high 96 bits unused) 43 42 41 40 + movd xmm2, [rdi] ; 13 12 11 10 + movd xmm3, [rcx] ; 53 52 51 50 + punpckldq xmm0, xmm1 ; (high 64 bits unused) 43 42 41 40 03 02 01 00 + punpckldq xmm2, xmm3 ; 53 52 51 50 13 12 11 10 + + movd xmm4, [rsi + rax*2] ; 23 22 21 20 + movd xmm5, [rdx + rax*2] ; 63 62 61 60 + movd xmm6, [rdi + rax*2] ; 33 32 31 30 + movd xmm7, [rcx + rax*2] ; 73 72 71 70 + punpckldq xmm4, xmm5 ; 63 62 61 60 23 22 21 20 + punpckldq xmm6, xmm7 ; 73 72 71 70 33 32 31 30 + + punpcklbw xmm0, xmm2 ; 53 43 52 42 51 41 50 40 13 03 12 02 11 01 10 00 + punpcklbw xmm4, xmm6 ; 73 63 72 62 71 61 70 60 33 23 32 22 31 21 30 20 + + movdqa xmm1, xmm0 + punpcklwd xmm0, xmm4 ; 33 23 13 03 32 22 12 02 31 21 11 01 30 20 10 00 + punpckhwd xmm1, xmm4 ; 73 63 53 43 72 62 52 42 71 61 51 41 70 60 50 40 + + movdqa xmm2, xmm0 + punpckldq xmm0, xmm1 ; 71 61 51 41 31 21 11 01 70 60 50 40 30 20 10 00 + punpckhdq xmm2, xmm1 ; 73 63 53 43 33 23 13 03 72 62 52 42 32 22 12 02 + + lea rsi, [rsi + rax*8] + lea rdi, [rsi + rax] + lea rdx, [rsi + rax*4] + lea rcx, [rdx + rax] + + movd xmm4, [rsi] ; 83 82 81 80 + movd xmm1, [rdx] ; c3 c2 c1 c0 + movd xmm6, [rdi] ; 93 92 91 90 + movd xmm3, [rcx] ; d3 d2 d1 d0 + punpckldq xmm4, xmm1 ; c3 c2 c1 c0 83 82 81 80 + punpckldq xmm6, xmm3 ; d3 d2 d1 d0 93 92 91 90 + + movd xmm1, [rsi + rax*2] ; a3 a2 a1 a0 + movd xmm5, [rdx + rax*2] ; e3 e2 e1 e0 + movd xmm3, [rdi + rax*2] ; b3 b2 b1 b0 + movd xmm7, [rcx + rax*2] ; f3 f2 f1 f0 + punpckldq xmm1, xmm5 ; e3 e2 e1 e0 a3 a2 a1 a0 + punpckldq xmm3, xmm7 ; f3 f2 f1 f0 b3 b2 b1 b0 + + punpcklbw xmm4, xmm6 ; d3 c3 d2 c2 d1 c1 d0 c0 93 83 92 82 91 81 90 80 + punpcklbw xmm1, xmm3 ; f3 e3 f2 e2 f1 e1 f0 e0 b3 a3 b2 a2 b1 a1 b0 a0 + + movdqa xmm7, xmm4 + punpcklwd xmm4, xmm1 ; b3 a3 93 83 b2 a2 92 82 b1 a1 91 81 b0 a0 90 80 + punpckhwd xmm7, xmm1 ; f3 e3 d3 c3 f2 e2 d2 c2 f1 e1 d1 c1 f0 e0 d0 c0 + + movdqa xmm6, xmm4 + punpckldq xmm4, xmm7 ; f1 e1 d1 c1 b1 a1 91 81 f0 e0 d0 c0 b0 a0 90 80 + punpckhdq xmm6, xmm7 ; f3 e3 d3 c3 b3 a3 93 83 f2 e2 d2 c2 b2 a2 92 82 + + movdqa xmm1, xmm0 + movdqa xmm3, xmm2 + + punpcklqdq xmm0, xmm4 ; p1 f0 e0 d0 c0 b0 a0 90 80 70 60 50 40 30 20 10 00 + punpckhqdq xmm1, xmm4 ; p0 f1 e1 d1 c1 b1 a1 91 81 71 61 51 41 31 21 11 01 + punpcklqdq xmm2, xmm6 ; q0 f2 e2 d2 c2 b2 a2 92 82 72 62 52 42 32 22 12 02 + punpckhqdq xmm3, xmm6 ; q1 f3 e3 d3 c3 b3 a3 93 83 73 63 53 43 33 23 13 03 + + mov rdx, arg(2) ;blimit + + ; calculate mask + movdqa xmm6, xmm0 ; p1 + movdqa xmm7, xmm3 ; q1 + psubusb xmm7, xmm0 ; q1-=p1 + psubusb xmm6, xmm3 ; p1-=q1 + por xmm6, xmm7 ; abs(p1-q1) + pand xmm6, [GLOBAL(tfe)] ; set lsb of each byte to zero + psrlw xmm6, 1 ; abs(p1-q1)/2 + + movdqa xmm7, [rdx] + + movdqa xmm5, xmm1 ; p0 + movdqa xmm4, xmm2 ; q0 + psubusb xmm5, xmm2 ; p0-=q0 + psubusb xmm4, xmm1 ; q0-=p0 + por xmm5, xmm4 ; abs(p0 - q0) + paddusb xmm5, xmm5 ; abs(p0-q0)*2 + paddusb xmm5, xmm6 ; abs (p0 - q0) *2 + abs(p1-q1)/2 + + movdqa xmm4, [GLOBAL(t80)] + + psubusb xmm5, xmm7 ; abs(p0 - q0) *2 + abs(p1-q1)/2 > blimit + pxor xmm7, xmm7 + pcmpeqb xmm5, xmm7 ; mm5 = mask + + ; start work on filters + movdqa t0, xmm0 + movdqa t1, xmm3 + + pxor xmm0, xmm4 ; p1 offset to convert to signed values + pxor xmm3, xmm4 ; q1 offset to convert to signed values + psubsb xmm0, xmm3 ; p1 - q1 + + pxor xmm1, xmm4 ; offset to convert to signed values + pxor xmm2, xmm4 ; offset to convert to signed values + + movdqa xmm3, xmm2 ; offseted ; q0 + psubsb xmm2, xmm1 ; q0 - p0 + paddsb xmm0, xmm2 ; p1 - q1 + 1 * (q0 - p0) + paddsb xmm0, xmm2 ; p1 - q1 + 2 * (q0 - p0) + paddsb xmm0, xmm2 ; p1 - q1 + 3 * (q0 - p0) + pand xmm5, xmm0 ; mask filter values we don't care about + + movdqa xmm0, xmm5 + paddsb xmm5, [GLOBAL(t3)] ; 3* (q0 - p0) + (p1 - q1) + 4 + paddsb xmm0, [GLOBAL(t4)] ; +3 instead of +4 + + movdqa xmm6, [GLOBAL(te0)] + movdqa xmm2, [GLOBAL(t1f)] + +; pxor xmm7, xmm7 + pcmpgtb xmm7, xmm0 ;save sign + pand xmm7, xmm6 ;preserve the upper 3 bits + psrlw xmm0, 3 + pand xmm0, xmm2 ;clear out upper 3 bits + por xmm0, xmm7 ;add sign + psubsb xmm3, xmm0 ; q0-= q0sz add + + pxor xmm7, xmm7 + pcmpgtb xmm7, xmm5 ;save sign + pand xmm7, xmm6 ;preserve the upper 3 bits + psrlw xmm5, 3 + pand xmm5, xmm2 ;clear out upper 3 bits + por xmm5, xmm7 ;add sign + paddsb xmm1, xmm5 ; p0+= p0 add + + pxor xmm3, xmm4 ; unoffset q0 + pxor xmm1, xmm4 ; unoffset p0 + + movdqa xmm0, t0 ; p1 + movdqa xmm4, t1 ; q1 + + ; write out order: xmm0 xmm2 xmm1 xmm3 + lea rdx, [rsi + rax*4] + + ; transpose back to write out + ; p1 f0 e0 d0 c0 b0 a0 90 80 70 60 50 40 30 20 10 00 + ; p0 f1 e1 d1 c1 b1 a1 91 81 71 61 51 41 31 21 11 01 + ; q0 f2 e2 d2 c2 b2 a2 92 82 72 62 52 42 32 22 12 02 + ; q1 f3 e3 d3 c3 b3 a3 93 83 73 63 53 43 33 23 13 03 + movdqa xmm6, xmm0 + punpcklbw xmm0, xmm1 ; 71 70 61 60 51 50 41 40 31 30 21 20 11 10 01 00 + punpckhbw xmm6, xmm1 ; f1 f0 e1 e0 d1 d0 c1 c0 b1 b0 a1 a0 91 90 81 80 + + movdqa xmm5, xmm3 + punpcklbw xmm3, xmm4 ; 73 72 63 62 53 52 43 42 33 32 23 22 13 12 03 02 + punpckhbw xmm5, xmm4 ; f3 f2 e3 e2 d3 d2 c3 c2 b3 b2 a3 a2 93 92 83 82 + + movdqa xmm2, xmm0 + punpcklwd xmm0, xmm3 ; 33 32 31 30 23 22 21 20 13 12 11 10 03 02 01 00 + punpckhwd xmm2, xmm3 ; 73 72 71 70 63 62 61 60 53 52 51 50 43 42 41 40 + + movdqa xmm3, xmm6 + punpcklwd xmm6, xmm5 ; b3 b2 b1 b0 a3 a2 a1 a0 93 92 91 90 83 82 81 80 + punpckhwd xmm3, xmm5 ; f3 f2 f1 f0 e3 e2 e1 e0 d3 d2 d1 d0 c3 c2 c1 c0 + + movd [rsi], xmm6 ; write the second 8-line result + movd [rdx], xmm3 + psrldq xmm6, 4 + psrldq xmm3, 4 + movd [rdi], xmm6 + movd [rcx], xmm3 + psrldq xmm6, 4 + psrldq xmm3, 4 + movd [rsi + rax*2], xmm6 + movd [rdx + rax*2], xmm3 + psrldq xmm6, 4 + psrldq xmm3, 4 + movd [rdi + rax*2], xmm6 + movd [rcx + rax*2], xmm3 + + neg rax + lea rsi, [rsi + rax*8] + neg rax + lea rdi, [rsi + rax] + lea rdx, [rsi + rax*4] + lea rcx, [rdx + rax] + + movd [rsi], xmm0 ; write the first 8-line result + movd [rdx], xmm2 + psrldq xmm0, 4 + psrldq xmm2, 4 + movd [rdi], xmm0 + movd [rcx], xmm2 + psrldq xmm0, 4 + psrldq xmm2, 4 + movd [rsi + rax*2], xmm0 + movd [rdx + rax*2], xmm2 + psrldq xmm0, 4 + psrldq xmm2, 4 + movd [rdi + rax*2], xmm0 + movd [rcx + rax*2], xmm2 + + add rsp, 32 + pop rsp + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +SECTION_RODATA +align 16 +tfe: + times 16 db 0xfe +align 16 +t80: + times 16 db 0x80 +align 16 +t1s: + times 16 db 0x01 +align 16 +t3: + times 16 db 0x03 +align 16 +t4: + times 16 db 0x04 +align 16 +ones: + times 8 dw 0x0001 +align 16 +s9: + times 8 dw 0x0900 +align 16 +s63: + times 8 dw 0x003f +align 16 +te0: + times 16 db 0xe0 +align 16 +t1f: + times 16 db 0x1f diff --git a/thirdparty/libvpx/vp8/common/x86/loopfilter_x86.c b/thirdparty/libvpx/vp8/common/x86/loopfilter_x86.c new file mode 100644 index 0000000000..6586004600 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/x86/loopfilter_x86.c @@ -0,0 +1,198 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "vpx_config.h" +#include "vp8/common/loopfilter.h" + +#define prototype_loopfilter(sym) \ + void sym(unsigned char *src, int pitch, const unsigned char *blimit,\ + const unsigned char *limit, const unsigned char *thresh, int count) + +#define prototype_loopfilter_nc(sym) \ + void sym(unsigned char *src, int pitch, const unsigned char *blimit,\ + const unsigned char *limit, const unsigned char *thresh) + +#define prototype_simple_loopfilter(sym) \ + void sym(unsigned char *y, int ystride, const unsigned char *blimit) + +prototype_loopfilter(vp8_mbloop_filter_vertical_edge_mmx); +prototype_loopfilter(vp8_mbloop_filter_horizontal_edge_mmx); +prototype_loopfilter(vp8_loop_filter_vertical_edge_mmx); +prototype_loopfilter(vp8_loop_filter_horizontal_edge_mmx); +prototype_simple_loopfilter(vp8_loop_filter_simple_horizontal_edge_mmx); +prototype_simple_loopfilter(vp8_loop_filter_simple_vertical_edge_mmx); + +#if HAVE_SSE2 && ARCH_X86_64 +prototype_loopfilter(vp8_loop_filter_bv_y_sse2); +prototype_loopfilter(vp8_loop_filter_bh_y_sse2); +#else +prototype_loopfilter_nc(vp8_loop_filter_vertical_edge_sse2); +prototype_loopfilter_nc(vp8_loop_filter_horizontal_edge_sse2); +#endif +prototype_loopfilter_nc(vp8_mbloop_filter_vertical_edge_sse2); +prototype_loopfilter_nc(vp8_mbloop_filter_horizontal_edge_sse2); + +extern loop_filter_uvfunction vp8_loop_filter_horizontal_edge_uv_sse2; +extern loop_filter_uvfunction vp8_loop_filter_vertical_edge_uv_sse2; +extern loop_filter_uvfunction vp8_mbloop_filter_horizontal_edge_uv_sse2; +extern loop_filter_uvfunction vp8_mbloop_filter_vertical_edge_uv_sse2; + +#if HAVE_MMX +/* Horizontal MB filtering */ +void vp8_loop_filter_mbh_mmx(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr, + int y_stride, int uv_stride, loop_filter_info *lfi) +{ + vp8_mbloop_filter_horizontal_edge_mmx(y_ptr, y_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 2); + + if (u_ptr) + vp8_mbloop_filter_horizontal_edge_mmx(u_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 1); + + if (v_ptr) + vp8_mbloop_filter_horizontal_edge_mmx(v_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 1); +} + + +/* Vertical MB Filtering */ +void vp8_loop_filter_mbv_mmx(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr, + int y_stride, int uv_stride, loop_filter_info *lfi) +{ + vp8_mbloop_filter_vertical_edge_mmx(y_ptr, y_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 2); + + if (u_ptr) + vp8_mbloop_filter_vertical_edge_mmx(u_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 1); + + if (v_ptr) + vp8_mbloop_filter_vertical_edge_mmx(v_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 1); +} + + +/* Horizontal B Filtering */ +void vp8_loop_filter_bh_mmx(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr, + int y_stride, int uv_stride, loop_filter_info *lfi) +{ + vp8_loop_filter_horizontal_edge_mmx(y_ptr + 4 * y_stride, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2); + vp8_loop_filter_horizontal_edge_mmx(y_ptr + 8 * y_stride, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2); + vp8_loop_filter_horizontal_edge_mmx(y_ptr + 12 * y_stride, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2); + + if (u_ptr) + vp8_loop_filter_horizontal_edge_mmx(u_ptr + 4 * uv_stride, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, 1); + + if (v_ptr) + vp8_loop_filter_horizontal_edge_mmx(v_ptr + 4 * uv_stride, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, 1); +} + + +void vp8_loop_filter_bhs_mmx(unsigned char *y_ptr, int y_stride, const unsigned char *blimit) +{ + vp8_loop_filter_simple_horizontal_edge_mmx(y_ptr + 4 * y_stride, y_stride, blimit); + vp8_loop_filter_simple_horizontal_edge_mmx(y_ptr + 8 * y_stride, y_stride, blimit); + vp8_loop_filter_simple_horizontal_edge_mmx(y_ptr + 12 * y_stride, y_stride, blimit); +} + + +/* Vertical B Filtering */ +void vp8_loop_filter_bv_mmx(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr, + int y_stride, int uv_stride, loop_filter_info *lfi) +{ + vp8_loop_filter_vertical_edge_mmx(y_ptr + 4, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2); + vp8_loop_filter_vertical_edge_mmx(y_ptr + 8, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2); + vp8_loop_filter_vertical_edge_mmx(y_ptr + 12, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2); + + if (u_ptr) + vp8_loop_filter_vertical_edge_mmx(u_ptr + 4, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, 1); + + if (v_ptr) + vp8_loop_filter_vertical_edge_mmx(v_ptr + 4, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, 1); +} + + +void vp8_loop_filter_bvs_mmx(unsigned char *y_ptr, int y_stride, const unsigned char *blimit) +{ + vp8_loop_filter_simple_vertical_edge_mmx(y_ptr + 4, y_stride, blimit); + vp8_loop_filter_simple_vertical_edge_mmx(y_ptr + 8, y_stride, blimit); + vp8_loop_filter_simple_vertical_edge_mmx(y_ptr + 12, y_stride, blimit); +} +#endif + + +/* Horizontal MB filtering */ +#if HAVE_SSE2 +void vp8_loop_filter_mbh_sse2(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr, + int y_stride, int uv_stride, loop_filter_info *lfi) +{ + vp8_mbloop_filter_horizontal_edge_sse2(y_ptr, y_stride, lfi->mblim, lfi->lim, lfi->hev_thr); + + if (u_ptr) + vp8_mbloop_filter_horizontal_edge_uv_sse2(u_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, v_ptr); +} + + +/* Vertical MB Filtering */ +void vp8_loop_filter_mbv_sse2(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr, + int y_stride, int uv_stride, loop_filter_info *lfi) +{ + vp8_mbloop_filter_vertical_edge_sse2(y_ptr, y_stride, lfi->mblim, lfi->lim, lfi->hev_thr); + + if (u_ptr) + vp8_mbloop_filter_vertical_edge_uv_sse2(u_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, v_ptr); +} + + +/* Horizontal B Filtering */ +void vp8_loop_filter_bh_sse2(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr, + int y_stride, int uv_stride, loop_filter_info *lfi) +{ +#if ARCH_X86_64 + vp8_loop_filter_bh_y_sse2(y_ptr, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2); +#else + vp8_loop_filter_horizontal_edge_sse2(y_ptr + 4 * y_stride, y_stride, lfi->blim, lfi->lim, lfi->hev_thr); + vp8_loop_filter_horizontal_edge_sse2(y_ptr + 8 * y_stride, y_stride, lfi->blim, lfi->lim, lfi->hev_thr); + vp8_loop_filter_horizontal_edge_sse2(y_ptr + 12 * y_stride, y_stride, lfi->blim, lfi->lim, lfi->hev_thr); +#endif + + if (u_ptr) + vp8_loop_filter_horizontal_edge_uv_sse2(u_ptr + 4 * uv_stride, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, v_ptr + 4 * uv_stride); +} + + +void vp8_loop_filter_bhs_sse2(unsigned char *y_ptr, int y_stride, const unsigned char *blimit) +{ + vp8_loop_filter_simple_horizontal_edge_sse2(y_ptr + 4 * y_stride, y_stride, blimit); + vp8_loop_filter_simple_horizontal_edge_sse2(y_ptr + 8 * y_stride, y_stride, blimit); + vp8_loop_filter_simple_horizontal_edge_sse2(y_ptr + 12 * y_stride, y_stride, blimit); +} + + +/* Vertical B Filtering */ +void vp8_loop_filter_bv_sse2(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr, + int y_stride, int uv_stride, loop_filter_info *lfi) +{ +#if ARCH_X86_64 + vp8_loop_filter_bv_y_sse2(y_ptr, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2); +#else + vp8_loop_filter_vertical_edge_sse2(y_ptr + 4, y_stride, lfi->blim, lfi->lim, lfi->hev_thr); + vp8_loop_filter_vertical_edge_sse2(y_ptr + 8, y_stride, lfi->blim, lfi->lim, lfi->hev_thr); + vp8_loop_filter_vertical_edge_sse2(y_ptr + 12, y_stride, lfi->blim, lfi->lim, lfi->hev_thr); +#endif + + if (u_ptr) + vp8_loop_filter_vertical_edge_uv_sse2(u_ptr + 4, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, v_ptr + 4); +} + + +void vp8_loop_filter_bvs_sse2(unsigned char *y_ptr, int y_stride, const unsigned char *blimit) +{ + vp8_loop_filter_simple_vertical_edge_sse2(y_ptr + 4, y_stride, blimit); + vp8_loop_filter_simple_vertical_edge_sse2(y_ptr + 8, y_stride, blimit); + vp8_loop_filter_simple_vertical_edge_sse2(y_ptr + 12, y_stride, blimit); +} + +#endif diff --git a/thirdparty/libvpx/vp8/common/x86/recon_mmx.asm b/thirdparty/libvpx/vp8/common/x86/recon_mmx.asm new file mode 100644 index 0000000000..15e98713c7 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/x86/recon_mmx.asm @@ -0,0 +1,274 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + + +%include "vpx_ports/x86_abi_support.asm" + + +;void copy_mem8x8_mmx( +; unsigned char *src, +; int src_stride, +; unsigned char *dst, +; int dst_stride +; ) +global sym(vp8_copy_mem8x8_mmx) PRIVATE +sym(vp8_copy_mem8x8_mmx): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 4 + push rsi + push rdi + ; end prolog + + mov rsi, arg(0) ;src; + movq mm0, [rsi] + + movsxd rax, dword ptr arg(1) ;src_stride; + mov rdi, arg(2) ;dst; + + movq mm1, [rsi+rax] + movq mm2, [rsi+rax*2] + + movsxd rcx, dword ptr arg(3) ;dst_stride + lea rsi, [rsi+rax*2] + + movq [rdi], mm0 + add rsi, rax + + movq [rdi+rcx], mm1 + movq [rdi+rcx*2], mm2 + + + lea rdi, [rdi+rcx*2] + movq mm3, [rsi] + + add rdi, rcx + movq mm4, [rsi+rax] + + movq mm5, [rsi+rax*2] + movq [rdi], mm3 + + lea rsi, [rsi+rax*2] + movq [rdi+rcx], mm4 + + movq [rdi+rcx*2], mm5 + lea rdi, [rdi+rcx*2] + + movq mm0, [rsi+rax] + movq mm1, [rsi+rax*2] + + movq [rdi+rcx], mm0 + movq [rdi+rcx*2],mm1 + + ; begin epilog + pop rdi + pop rsi + UNSHADOW_ARGS + pop rbp + ret + + +;void copy_mem8x4_mmx( +; unsigned char *src, +; int src_stride, +; unsigned char *dst, +; int dst_stride +; ) +global sym(vp8_copy_mem8x4_mmx) PRIVATE +sym(vp8_copy_mem8x4_mmx): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 4 + push rsi + push rdi + ; end prolog + + mov rsi, arg(0) ;src; + movq mm0, [rsi] + + movsxd rax, dword ptr arg(1) ;src_stride; + mov rdi, arg(2) ;dst; + + movq mm1, [rsi+rax] + movq mm2, [rsi+rax*2] + + movsxd rcx, dword ptr arg(3) ;dst_stride + lea rsi, [rsi+rax*2] + + movq [rdi], mm0 + movq [rdi+rcx], mm1 + + movq [rdi+rcx*2], mm2 + lea rdi, [rdi+rcx*2] + + movq mm3, [rsi+rax] + movq [rdi+rcx], mm3 + + ; begin epilog + pop rdi + pop rsi + UNSHADOW_ARGS + pop rbp + ret + + +;void copy_mem16x16_mmx( +; unsigned char *src, +; int src_stride, +; unsigned char *dst, +; int dst_stride +; ) +global sym(vp8_copy_mem16x16_mmx) PRIVATE +sym(vp8_copy_mem16x16_mmx): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 4 + push rsi + push rdi + ; end prolog + + mov rsi, arg(0) ;src; + movsxd rax, dword ptr arg(1) ;src_stride; + + mov rdi, arg(2) ;dst; + movsxd rcx, dword ptr arg(3) ;dst_stride + + movq mm0, [rsi] + movq mm3, [rsi+8]; + + movq mm1, [rsi+rax] + movq mm4, [rsi+rax+8] + + movq mm2, [rsi+rax*2] + movq mm5, [rsi+rax*2+8] + + lea rsi, [rsi+rax*2] + add rsi, rax + + movq [rdi], mm0 + movq [rdi+8], mm3 + + movq [rdi+rcx], mm1 + movq [rdi+rcx+8], mm4 + + movq [rdi+rcx*2], mm2 + movq [rdi+rcx*2+8], mm5 + + lea rdi, [rdi+rcx*2] + add rdi, rcx + + movq mm0, [rsi] + movq mm3, [rsi+8]; + + movq mm1, [rsi+rax] + movq mm4, [rsi+rax+8] + + movq mm2, [rsi+rax*2] + movq mm5, [rsi+rax*2+8] + + lea rsi, [rsi+rax*2] + add rsi, rax + + movq [rdi], mm0 + movq [rdi+8], mm3 + + movq [rdi+rcx], mm1 + movq [rdi+rcx+8], mm4 + + movq [rdi+rcx*2], mm2 + movq [rdi+rcx*2+8], mm5 + + lea rdi, [rdi+rcx*2] + add rdi, rcx + + movq mm0, [rsi] + movq mm3, [rsi+8]; + + movq mm1, [rsi+rax] + movq mm4, [rsi+rax+8] + + movq mm2, [rsi+rax*2] + movq mm5, [rsi+rax*2+8] + + lea rsi, [rsi+rax*2] + add rsi, rax + + movq [rdi], mm0 + movq [rdi+8], mm3 + + movq [rdi+rcx], mm1 + movq [rdi+rcx+8], mm4 + + movq [rdi+rcx*2], mm2 + movq [rdi+rcx*2+8], mm5 + + lea rdi, [rdi+rcx*2] + add rdi, rcx + + movq mm0, [rsi] + movq mm3, [rsi+8]; + + movq mm1, [rsi+rax] + movq mm4, [rsi+rax+8] + + movq mm2, [rsi+rax*2] + movq mm5, [rsi+rax*2+8] + + lea rsi, [rsi+rax*2] + add rsi, rax + + movq [rdi], mm0 + movq [rdi+8], mm3 + + movq [rdi+rcx], mm1 + movq [rdi+rcx+8], mm4 + + movq [rdi+rcx*2], mm2 + movq [rdi+rcx*2+8], mm5 + + lea rdi, [rdi+rcx*2] + add rdi, rcx + + movq mm0, [rsi] + movq mm3, [rsi+8]; + + movq mm1, [rsi+rax] + movq mm4, [rsi+rax+8] + + movq mm2, [rsi+rax*2] + movq mm5, [rsi+rax*2+8] + + lea rsi, [rsi+rax*2] + add rsi, rax + + movq [rdi], mm0 + movq [rdi+8], mm3 + + movq [rdi+rcx], mm1 + movq [rdi+rcx+8], mm4 + + movq [rdi+rcx*2], mm2 + movq [rdi+rcx*2+8], mm5 + + lea rdi, [rdi+rcx*2] + add rdi, rcx + + movq mm0, [rsi] + movq mm3, [rsi+8]; + + movq [rdi], mm0 + movq [rdi+8], mm3 + + ; begin epilog + pop rdi + pop rsi + UNSHADOW_ARGS + pop rbp + ret diff --git a/thirdparty/libvpx/vp8/common/x86/recon_sse2.asm b/thirdparty/libvpx/vp8/common/x86/recon_sse2.asm new file mode 100644 index 0000000000..cb89537f76 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/x86/recon_sse2.asm @@ -0,0 +1,116 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + + +%include "vpx_ports/x86_abi_support.asm" + +;void copy_mem16x16_sse2( +; unsigned char *src, +; int src_stride, +; unsigned char *dst, +; int dst_stride +; ) +global sym(vp8_copy_mem16x16_sse2) PRIVATE +sym(vp8_copy_mem16x16_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 4 + push rsi + push rdi + ; end prolog + + mov rsi, arg(0) ;src; + movdqu xmm0, [rsi] + + movsxd rax, dword ptr arg(1) ;src_stride; + mov rdi, arg(2) ;dst; + + movdqu xmm1, [rsi+rax] + movdqu xmm2, [rsi+rax*2] + + movsxd rcx, dword ptr arg(3) ;dst_stride + lea rsi, [rsi+rax*2] + + movdqa [rdi], xmm0 + add rsi, rax + + movdqa [rdi+rcx], xmm1 + movdqa [rdi+rcx*2],xmm2 + + lea rdi, [rdi+rcx*2] + movdqu xmm3, [rsi] + + add rdi, rcx + movdqu xmm4, [rsi+rax] + + movdqu xmm5, [rsi+rax*2] + lea rsi, [rsi+rax*2] + + movdqa [rdi], xmm3 + add rsi, rax + + movdqa [rdi+rcx], xmm4 + movdqa [rdi+rcx*2],xmm5 + + lea rdi, [rdi+rcx*2] + movdqu xmm0, [rsi] + + add rdi, rcx + movdqu xmm1, [rsi+rax] + + movdqu xmm2, [rsi+rax*2] + lea rsi, [rsi+rax*2] + + movdqa [rdi], xmm0 + add rsi, rax + + movdqa [rdi+rcx], xmm1 + + movdqa [rdi+rcx*2], xmm2 + movdqu xmm3, [rsi] + + movdqu xmm4, [rsi+rax] + lea rdi, [rdi+rcx*2] + + add rdi, rcx + movdqu xmm5, [rsi+rax*2] + + lea rsi, [rsi+rax*2] + movdqa [rdi], xmm3 + + add rsi, rax + movdqa [rdi+rcx], xmm4 + + movdqa [rdi+rcx*2],xmm5 + movdqu xmm0, [rsi] + + lea rdi, [rdi+rcx*2] + movdqu xmm1, [rsi+rax] + + add rdi, rcx + movdqu xmm2, [rsi+rax*2] + + lea rsi, [rsi+rax*2] + movdqa [rdi], xmm0 + + movdqa [rdi+rcx], xmm1 + movdqa [rdi+rcx*2],xmm2 + + movdqu xmm3, [rsi+rax] + lea rdi, [rdi+rcx*2] + + movdqa [rdi+rcx], xmm3 + + ; begin epilog + pop rdi + pop rsi + UNSHADOW_ARGS + pop rbp + ret diff --git a/thirdparty/libvpx/vp8/common/x86/subpixel_mmx.asm b/thirdparty/libvpx/vp8/common/x86/subpixel_mmx.asm new file mode 100644 index 0000000000..47dd452297 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/x86/subpixel_mmx.asm @@ -0,0 +1,702 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + + +%include "vpx_ports/x86_abi_support.asm" +extern sym(vp8_bilinear_filters_x86_8) + + +%define BLOCK_HEIGHT_WIDTH 4 +%define vp8_filter_weight 128 +%define VP8_FILTER_SHIFT 7 + + +;void vp8_filter_block1d_h6_mmx +;( +; unsigned char *src_ptr, +; unsigned short *output_ptr, +; unsigned int src_pixels_per_line, +; unsigned int pixel_step, +; unsigned int output_height, +; unsigned int output_width, +; short * vp8_filter +;) +global sym(vp8_filter_block1d_h6_mmx) PRIVATE +sym(vp8_filter_block1d_h6_mmx): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + mov rdx, arg(6) ;vp8_filter + + movq mm1, [rdx + 16] ; do both the negative taps first!!! + movq mm2, [rdx + 32] ; + movq mm6, [rdx + 48] ; + movq mm7, [rdx + 64] ; + + mov rdi, arg(1) ;output_ptr + mov rsi, arg(0) ;src_ptr + movsxd rcx, dword ptr arg(4) ;output_height + movsxd rax, dword ptr arg(5) ;output_width ; destination pitch? + pxor mm0, mm0 ; mm0 = 00000000 + +.nextrow: + movq mm3, [rsi-2] ; mm3 = p-2..p5 + movq mm4, mm3 ; mm4 = p-2..p5 + psrlq mm3, 8 ; mm3 = p-1..p5 + punpcklbw mm3, mm0 ; mm3 = p-1..p2 + pmullw mm3, mm1 ; mm3 *= kernel 1 modifiers. + + movq mm5, mm4 ; mm5 = p-2..p5 + punpckhbw mm4, mm0 ; mm5 = p2..p5 + pmullw mm4, mm7 ; mm5 *= kernel 4 modifiers + paddsw mm3, mm4 ; mm3 += mm5 + + movq mm4, mm5 ; mm4 = p-2..p5; + psrlq mm5, 16 ; mm5 = p0..p5; + punpcklbw mm5, mm0 ; mm5 = p0..p3 + pmullw mm5, mm2 ; mm5 *= kernel 2 modifiers + paddsw mm3, mm5 ; mm3 += mm5 + + movq mm5, mm4 ; mm5 = p-2..p5 + psrlq mm4, 24 ; mm4 = p1..p5 + punpcklbw mm4, mm0 ; mm4 = p1..p4 + pmullw mm4, mm6 ; mm5 *= kernel 3 modifiers + paddsw mm3, mm4 ; mm3 += mm5 + + ; do outer positive taps + movd mm4, [rsi+3] + punpcklbw mm4, mm0 ; mm5 = p3..p6 + pmullw mm4, [rdx+80] ; mm5 *= kernel 0 modifiers + paddsw mm3, mm4 ; mm3 += mm5 + + punpcklbw mm5, mm0 ; mm5 = p-2..p1 + pmullw mm5, [rdx] ; mm5 *= kernel 5 modifiers + paddsw mm3, mm5 ; mm3 += mm5 + + paddsw mm3, [GLOBAL(rd)] ; mm3 += round value + psraw mm3, VP8_FILTER_SHIFT ; mm3 /= 128 + packuswb mm3, mm0 ; pack and unpack to saturate + punpcklbw mm3, mm0 ; + + movq [rdi], mm3 ; store the results in the destination + +%if ABI_IS_32BIT + add rsi, dword ptr arg(2) ;src_pixels_per_line ; next line + add rdi, rax; +%else + movsxd r8, dword ptr arg(2) ;src_pixels_per_line + add rdi, rax; + + add rsi, r8 ; next line +%endif + + dec rcx ; decrement count + jnz .nextrow ; next row + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + UNSHADOW_ARGS + pop rbp + ret + + +;void vp8_filter_block1dc_v6_mmx +;( +; short *src_ptr, +; unsigned char *output_ptr, +; int output_pitch, +; unsigned int pixels_per_line, +; unsigned int pixel_step, +; unsigned int output_height, +; unsigned int output_width, +; short * vp8_filter +;) +global sym(vp8_filter_block1dc_v6_mmx) PRIVATE +sym(vp8_filter_block1dc_v6_mmx): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 8 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + movq mm5, [GLOBAL(rd)] + push rbx + mov rbx, arg(7) ;vp8_filter + movq mm1, [rbx + 16] ; do both the negative taps first!!! + movq mm2, [rbx + 32] ; + movq mm6, [rbx + 48] ; + movq mm7, [rbx + 64] ; + + movsxd rdx, dword ptr arg(3) ;pixels_per_line + mov rdi, arg(1) ;output_ptr + mov rsi, arg(0) ;src_ptr + sub rsi, rdx + sub rsi, rdx + movsxd rcx, DWORD PTR arg(5) ;output_height + movsxd rax, DWORD PTR arg(2) ;output_pitch ; destination pitch? + pxor mm0, mm0 ; mm0 = 00000000 + + +.nextrow_cv: + movq mm3, [rsi+rdx] ; mm3 = p0..p8 = row -1 + pmullw mm3, mm1 ; mm3 *= kernel 1 modifiers. + + + movq mm4, [rsi + 4*rdx] ; mm4 = p0..p3 = row 2 + pmullw mm4, mm7 ; mm4 *= kernel 4 modifiers. + paddsw mm3, mm4 ; mm3 += mm4 + + movq mm4, [rsi + 2*rdx] ; mm4 = p0..p3 = row 0 + pmullw mm4, mm2 ; mm4 *= kernel 2 modifiers. + paddsw mm3, mm4 ; mm3 += mm4 + + movq mm4, [rsi] ; mm4 = p0..p3 = row -2 + pmullw mm4, [rbx] ; mm4 *= kernel 0 modifiers. + paddsw mm3, mm4 ; mm3 += mm4 + + + add rsi, rdx ; move source forward 1 line to avoid 3 * pitch + movq mm4, [rsi + 2*rdx] ; mm4 = p0..p3 = row 1 + pmullw mm4, mm6 ; mm4 *= kernel 3 modifiers. + paddsw mm3, mm4 ; mm3 += mm4 + + movq mm4, [rsi + 4*rdx] ; mm4 = p0..p3 = row 3 + pmullw mm4, [rbx +80] ; mm4 *= kernel 3 modifiers. + paddsw mm3, mm4 ; mm3 += mm4 + + + paddsw mm3, mm5 ; mm3 += round value + psraw mm3, VP8_FILTER_SHIFT ; mm3 /= 128 + packuswb mm3, mm0 ; pack and saturate + + movd [rdi],mm3 ; store the results in the destination + ; the subsequent iterations repeat 3 out of 4 of these reads. Since the + ; recon block should be in cache this shouldn't cost much. Its obviously + ; avoidable!!!. + lea rdi, [rdi+rax] ; + dec rcx ; decrement count + jnz .nextrow_cv ; next row + + pop rbx + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + UNSHADOW_ARGS + pop rbp + ret + + +;void bilinear_predict8x8_mmx +;( +; unsigned char *src_ptr, +; int src_pixels_per_line, +; int xoffset, +; int yoffset, +; unsigned char *dst_ptr, +; int dst_pitch +;) +global sym(vp8_bilinear_predict8x8_mmx) PRIVATE +sym(vp8_bilinear_predict8x8_mmx): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + ;const short *HFilter = vp8_bilinear_filters_x86_8[xoffset]; + ;const short *VFilter = vp8_bilinear_filters_x86_8[yoffset]; + + movsxd rax, dword ptr arg(2) ;xoffset + mov rdi, arg(4) ;dst_ptr ; + + shl rax, 5 ; offset * 32 + lea rcx, [GLOBAL(sym(vp8_bilinear_filters_x86_8))] + + add rax, rcx ; HFilter + mov rsi, arg(0) ;src_ptr ; + + movsxd rdx, dword ptr arg(5) ;dst_pitch + movq mm1, [rax] ; + + movq mm2, [rax+16] ; + movsxd rax, dword ptr arg(3) ;yoffset + + pxor mm0, mm0 ; + + shl rax, 5 ; offset*32 + add rax, rcx ; VFilter + + lea rcx, [rdi+rdx*8] ; + movsxd rdx, dword ptr arg(1) ;src_pixels_per_line ; + + + + ; get the first horizontal line done ; + movq mm3, [rsi] ; xx 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 + movq mm4, mm3 ; make a copy of current line + + punpcklbw mm3, mm0 ; xx 00 01 02 03 04 05 06 + punpckhbw mm4, mm0 ; + + pmullw mm3, mm1 ; + pmullw mm4, mm1 ; + + movq mm5, [rsi+1] ; + movq mm6, mm5 ; + + punpcklbw mm5, mm0 ; + punpckhbw mm6, mm0 ; + + pmullw mm5, mm2 ; + pmullw mm6, mm2 ; + + paddw mm3, mm5 ; + paddw mm4, mm6 ; + + paddw mm3, [GLOBAL(rd)] ; xmm3 += round value + psraw mm3, VP8_FILTER_SHIFT ; xmm3 /= 128 + + paddw mm4, [GLOBAL(rd)] ; + psraw mm4, VP8_FILTER_SHIFT ; + + movq mm7, mm3 ; + packuswb mm7, mm4 ; + + add rsi, rdx ; next line +.next_row_8x8: + movq mm3, [rsi] ; xx 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 + movq mm4, mm3 ; make a copy of current line + + punpcklbw mm3, mm0 ; xx 00 01 02 03 04 05 06 + punpckhbw mm4, mm0 ; + + pmullw mm3, mm1 ; + pmullw mm4, mm1 ; + + movq mm5, [rsi+1] ; + movq mm6, mm5 ; + + punpcklbw mm5, mm0 ; + punpckhbw mm6, mm0 ; + + pmullw mm5, mm2 ; + pmullw mm6, mm2 ; + + paddw mm3, mm5 ; + paddw mm4, mm6 ; + + movq mm5, mm7 ; + movq mm6, mm7 ; + + punpcklbw mm5, mm0 ; + punpckhbw mm6, mm0 + + pmullw mm5, [rax] ; + pmullw mm6, [rax] ; + + paddw mm3, [GLOBAL(rd)] ; xmm3 += round value + psraw mm3, VP8_FILTER_SHIFT ; xmm3 /= 128 + + paddw mm4, [GLOBAL(rd)] ; + psraw mm4, VP8_FILTER_SHIFT ; + + movq mm7, mm3 ; + packuswb mm7, mm4 ; + + + pmullw mm3, [rax+16] ; + pmullw mm4, [rax+16] ; + + paddw mm3, mm5 ; + paddw mm4, mm6 ; + + + paddw mm3, [GLOBAL(rd)] ; xmm3 += round value + psraw mm3, VP8_FILTER_SHIFT ; xmm3 /= 128 + + paddw mm4, [GLOBAL(rd)] ; + psraw mm4, VP8_FILTER_SHIFT ; + + packuswb mm3, mm4 + + movq [rdi], mm3 ; store the results in the destination + +%if ABI_IS_32BIT + add rsi, rdx ; next line + add rdi, dword ptr arg(5) ;dst_pitch ; +%else + movsxd r8, dword ptr arg(5) ;dst_pitch + add rsi, rdx ; next line + add rdi, r8 ;dst_pitch +%endif + cmp rdi, rcx ; + jne .next_row_8x8 + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + UNSHADOW_ARGS + pop rbp + ret + + +;void bilinear_predict8x4_mmx +;( +; unsigned char *src_ptr, +; int src_pixels_per_line, +; int xoffset, +; int yoffset, +; unsigned char *dst_ptr, +; int dst_pitch +;) +global sym(vp8_bilinear_predict8x4_mmx) PRIVATE +sym(vp8_bilinear_predict8x4_mmx): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + ;const short *HFilter = vp8_bilinear_filters_x86_8[xoffset]; + ;const short *VFilter = vp8_bilinear_filters_x86_8[yoffset]; + + movsxd rax, dword ptr arg(2) ;xoffset + mov rdi, arg(4) ;dst_ptr ; + + lea rcx, [GLOBAL(sym(vp8_bilinear_filters_x86_8))] + shl rax, 5 + + mov rsi, arg(0) ;src_ptr ; + add rax, rcx + + movsxd rdx, dword ptr arg(5) ;dst_pitch + movq mm1, [rax] ; + + movq mm2, [rax+16] ; + movsxd rax, dword ptr arg(3) ;yoffset + + pxor mm0, mm0 ; + shl rax, 5 + + add rax, rcx + lea rcx, [rdi+rdx*4] ; + + movsxd rdx, dword ptr arg(1) ;src_pixels_per_line ; + + ; get the first horizontal line done ; + movq mm3, [rsi] ; xx 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 + movq mm4, mm3 ; make a copy of current line + + punpcklbw mm3, mm0 ; xx 00 01 02 03 04 05 06 + punpckhbw mm4, mm0 ; + + pmullw mm3, mm1 ; + pmullw mm4, mm1 ; + + movq mm5, [rsi+1] ; + movq mm6, mm5 ; + + punpcklbw mm5, mm0 ; + punpckhbw mm6, mm0 ; + + pmullw mm5, mm2 ; + pmullw mm6, mm2 ; + + paddw mm3, mm5 ; + paddw mm4, mm6 ; + + paddw mm3, [GLOBAL(rd)] ; xmm3 += round value + psraw mm3, VP8_FILTER_SHIFT ; xmm3 /= 128 + + paddw mm4, [GLOBAL(rd)] ; + psraw mm4, VP8_FILTER_SHIFT ; + + movq mm7, mm3 ; + packuswb mm7, mm4 ; + + add rsi, rdx ; next line +.next_row_8x4: + movq mm3, [rsi] ; xx 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 + movq mm4, mm3 ; make a copy of current line + + punpcklbw mm3, mm0 ; xx 00 01 02 03 04 05 06 + punpckhbw mm4, mm0 ; + + pmullw mm3, mm1 ; + pmullw mm4, mm1 ; + + movq mm5, [rsi+1] ; + movq mm6, mm5 ; + + punpcklbw mm5, mm0 ; + punpckhbw mm6, mm0 ; + + pmullw mm5, mm2 ; + pmullw mm6, mm2 ; + + paddw mm3, mm5 ; + paddw mm4, mm6 ; + + movq mm5, mm7 ; + movq mm6, mm7 ; + + punpcklbw mm5, mm0 ; + punpckhbw mm6, mm0 + + pmullw mm5, [rax] ; + pmullw mm6, [rax] ; + + paddw mm3, [GLOBAL(rd)] ; xmm3 += round value + psraw mm3, VP8_FILTER_SHIFT ; xmm3 /= 128 + + paddw mm4, [GLOBAL(rd)] ; + psraw mm4, VP8_FILTER_SHIFT ; + + movq mm7, mm3 ; + packuswb mm7, mm4 ; + + + pmullw mm3, [rax+16] ; + pmullw mm4, [rax+16] ; + + paddw mm3, mm5 ; + paddw mm4, mm6 ; + + + paddw mm3, [GLOBAL(rd)] ; xmm3 += round value + psraw mm3, VP8_FILTER_SHIFT ; xmm3 /= 128 + + paddw mm4, [GLOBAL(rd)] ; + psraw mm4, VP8_FILTER_SHIFT ; + + packuswb mm3, mm4 + + movq [rdi], mm3 ; store the results in the destination + +%if ABI_IS_32BIT + add rsi, rdx ; next line + add rdi, dword ptr arg(5) ;dst_pitch ; +%else + movsxd r8, dword ptr arg(5) ;dst_pitch + add rsi, rdx ; next line + add rdi, r8 +%endif + cmp rdi, rcx ; + jne .next_row_8x4 + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + UNSHADOW_ARGS + pop rbp + ret + + +;void bilinear_predict4x4_mmx +;( +; unsigned char *src_ptr, +; int src_pixels_per_line, +; int xoffset, +; int yoffset, +; unsigned char *dst_ptr, +; int dst_pitch +;) +global sym(vp8_bilinear_predict4x4_mmx) PRIVATE +sym(vp8_bilinear_predict4x4_mmx): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + ;const short *HFilter = vp8_bilinear_filters_x86_8[xoffset]; + ;const short *VFilter = vp8_bilinear_filters_x86_8[yoffset]; + + movsxd rax, dword ptr arg(2) ;xoffset + mov rdi, arg(4) ;dst_ptr ; + + lea rcx, [GLOBAL(sym(vp8_bilinear_filters_x86_8))] + shl rax, 5 + + add rax, rcx ; HFilter + mov rsi, arg(0) ;src_ptr ; + + movsxd rdx, dword ptr arg(5) ;ldst_pitch + movq mm1, [rax] ; + + movq mm2, [rax+16] ; + movsxd rax, dword ptr arg(3) ;yoffset + + pxor mm0, mm0 ; + shl rax, 5 + + add rax, rcx + lea rcx, [rdi+rdx*4] ; + + movsxd rdx, dword ptr arg(1) ;src_pixels_per_line ; + + ; get the first horizontal line done ; + movd mm3, [rsi] ; xx 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 + punpcklbw mm3, mm0 ; xx 00 01 02 03 04 05 06 + + pmullw mm3, mm1 ; + movd mm5, [rsi+1] ; + + punpcklbw mm5, mm0 ; + pmullw mm5, mm2 ; + + paddw mm3, mm5 ; + paddw mm3, [GLOBAL(rd)] ; xmm3 += round value + + psraw mm3, VP8_FILTER_SHIFT ; xmm3 /= 128 + + movq mm7, mm3 ; + packuswb mm7, mm0 ; + + add rsi, rdx ; next line +.next_row_4x4: + movd mm3, [rsi] ; xx 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 + punpcklbw mm3, mm0 ; xx 00 01 02 03 04 05 06 + + pmullw mm3, mm1 ; + movd mm5, [rsi+1] ; + + punpcklbw mm5, mm0 ; + pmullw mm5, mm2 ; + + paddw mm3, mm5 ; + + movq mm5, mm7 ; + punpcklbw mm5, mm0 ; + + pmullw mm5, [rax] ; + paddw mm3, [GLOBAL(rd)] ; xmm3 += round value + + psraw mm3, VP8_FILTER_SHIFT ; xmm3 /= 128 + movq mm7, mm3 ; + + packuswb mm7, mm0 ; + + pmullw mm3, [rax+16] ; + paddw mm3, mm5 ; + + + paddw mm3, [GLOBAL(rd)] ; xmm3 += round value + psraw mm3, VP8_FILTER_SHIFT ; xmm3 /= 128 + + packuswb mm3, mm0 + movd [rdi], mm3 ; store the results in the destination + +%if ABI_IS_32BIT + add rsi, rdx ; next line + add rdi, dword ptr arg(5) ;dst_pitch ; +%else + movsxd r8, dword ptr arg(5) ;dst_pitch ; + add rsi, rdx ; next line + add rdi, r8 +%endif + + cmp rdi, rcx ; + jne .next_row_4x4 + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + UNSHADOW_ARGS + pop rbp + ret + + + +SECTION_RODATA +align 16 +rd: + times 4 dw 0x40 + +align 16 +global HIDDEN_DATA(sym(vp8_six_tap_mmx)) +sym(vp8_six_tap_mmx): + times 8 dw 0 + times 8 dw 0 + times 8 dw 128 + times 8 dw 0 + times 8 dw 0 + times 8 dw 0 + + times 8 dw 0 + times 8 dw -6 + times 8 dw 123 + times 8 dw 12 + times 8 dw -1 + times 8 dw 0 + + times 8 dw 2 + times 8 dw -11 + times 8 dw 108 + times 8 dw 36 + times 8 dw -8 + times 8 dw 1 + + times 8 dw 0 + times 8 dw -9 + times 8 dw 93 + times 8 dw 50 + times 8 dw -6 + times 8 dw 0 + + times 8 dw 3 + times 8 dw -16 + times 8 dw 77 + times 8 dw 77 + times 8 dw -16 + times 8 dw 3 + + times 8 dw 0 + times 8 dw -6 + times 8 dw 50 + times 8 dw 93 + times 8 dw -9 + times 8 dw 0 + + times 8 dw 1 + times 8 dw -8 + times 8 dw 36 + times 8 dw 108 + times 8 dw -11 + times 8 dw 2 + + times 8 dw 0 + times 8 dw -1 + times 8 dw 12 + times 8 dw 123 + times 8 dw -6 + times 8 dw 0 + + diff --git a/thirdparty/libvpx/vp8/common/x86/subpixel_sse2.asm b/thirdparty/libvpx/vp8/common/x86/subpixel_sse2.asm new file mode 100644 index 0000000000..69f8d103c1 --- /dev/null +++ b/thirdparty/libvpx/vp8/common/x86/subpixel_sse2.asm @@ -0,0 +1,1372 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + + +%include "vpx_ports/x86_abi_support.asm" +extern sym(vp8_bilinear_filters_x86_8) + +%define BLOCK_HEIGHT_WIDTH 4 +%define VP8_FILTER_WEIGHT 128 +%define VP8_FILTER_SHIFT 7 + + +;/************************************************************************************ +; Notes: filter_block1d_h6 applies a 6 tap filter horizontally to the input pixels. The +; input pixel array has output_height rows. This routine assumes that output_height is an +; even number. This function handles 8 pixels in horizontal direction, calculating ONE +; rows each iteration to take advantage of the 128 bits operations. +;*************************************************************************************/ +;void vp8_filter_block1d8_h6_sse2 +;( +; unsigned char *src_ptr, +; unsigned short *output_ptr, +; unsigned int src_pixels_per_line, +; unsigned int pixel_step, +; unsigned int output_height, +; unsigned int output_width, +; short *vp8_filter +;) +global sym(vp8_filter_block1d8_h6_sse2) PRIVATE +sym(vp8_filter_block1d8_h6_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 7 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + mov rdx, arg(6) ;vp8_filter + mov rsi, arg(0) ;src_ptr + + mov rdi, arg(1) ;output_ptr + + movsxd rcx, dword ptr arg(4) ;output_height + movsxd rax, dword ptr arg(2) ;src_pixels_per_line ; Pitch for Source +%if ABI_IS_32BIT=0 + movsxd r8, dword ptr arg(5) ;output_width +%endif + pxor xmm0, xmm0 ; clear xmm0 for unpack + +.filter_block1d8_h6_rowloop: + movq xmm3, MMWORD PTR [rsi - 2] + movq xmm1, MMWORD PTR [rsi + 6] + + prefetcht2 [rsi+rax-2] + + pslldq xmm1, 8 + por xmm1, xmm3 + + movdqa xmm4, xmm1 + movdqa xmm5, xmm1 + + movdqa xmm6, xmm1 + movdqa xmm7, xmm1 + + punpcklbw xmm3, xmm0 ; xx05 xx04 xx03 xx02 xx01 xx01 xx-1 xx-2 + psrldq xmm4, 1 ; xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 00 -1 + + pmullw xmm3, XMMWORD PTR [rdx] ; x[-2] * H[-2]; Tap 1 + punpcklbw xmm4, xmm0 ; xx06 xx05 xx04 xx03 xx02 xx01 xx00 xx-1 + + psrldq xmm5, 2 ; xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 00 + pmullw xmm4, XMMWORD PTR [rdx+16] ; x[-1] * H[-1]; Tap 2 + + + punpcklbw xmm5, xmm0 ; xx07 xx06 xx05 xx04 xx03 xx02 xx01 xx00 + psrldq xmm6, 3 ; xx xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 + + pmullw xmm5, [rdx+32] ; x[ 0] * H[ 0]; Tap 3 + + punpcklbw xmm6, xmm0 ; xx08 xx07 xx06 xx05 xx04 xx03 xx02 xx01 + psrldq xmm7, 4 ; xx xx xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 + + pmullw xmm6, [rdx+48] ; x[ 1] * h[ 1] ; Tap 4 + + punpcklbw xmm7, xmm0 ; xx09 xx08 xx07 xx06 xx05 xx04 xx03 xx02 + psrldq xmm1, 5 ; xx xx xx xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 + + + pmullw xmm7, [rdx+64] ; x[ 2] * h[ 2] ; Tap 5 + + punpcklbw xmm1, xmm0 ; xx0a xx09 xx08 xx07 xx06 xx05 xx04 xx03 + pmullw xmm1, [rdx+80] ; x[ 3] * h[ 3] ; Tap 6 + + + paddsw xmm4, xmm7 + paddsw xmm4, xmm5 + + paddsw xmm4, xmm3 + paddsw xmm4, xmm6 + + paddsw xmm4, xmm1 + paddsw xmm4, [GLOBAL(rd)] + + psraw xmm4, 7 + + packuswb xmm4, xmm0 + punpcklbw xmm4, xmm0 + + movdqa XMMWORD Ptr [rdi], xmm4 + lea rsi, [rsi + rax] + +%if ABI_IS_32BIT + add rdi, DWORD Ptr arg(5) ;[output_width] +%else + add rdi, r8 +%endif + dec rcx + + jnz .filter_block1d8_h6_rowloop ; next row + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + + +;void vp8_filter_block1d16_h6_sse2 +;( +; unsigned char *src_ptr, +; unsigned short *output_ptr, +; unsigned int src_pixels_per_line, +; unsigned int pixel_step, +; unsigned int output_height, +; unsigned int output_width, +; short *vp8_filter +;) +;/************************************************************************************ +; Notes: filter_block1d_h6 applies a 6 tap filter horizontally to the input pixels. The +; input pixel array has output_height rows. This routine assumes that output_height is an +; even number. This function handles 8 pixels in horizontal direction, calculating ONE +; rows each iteration to take advantage of the 128 bits operations. +;*************************************************************************************/ +global sym(vp8_filter_block1d16_h6_sse2) PRIVATE +sym(vp8_filter_block1d16_h6_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 7 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + mov rdx, arg(6) ;vp8_filter + mov rsi, arg(0) ;src_ptr + + mov rdi, arg(1) ;output_ptr + + movsxd rcx, dword ptr arg(4) ;output_height + movsxd rax, dword ptr arg(2) ;src_pixels_per_line ; Pitch for Source +%if ABI_IS_32BIT=0 + movsxd r8, dword ptr arg(5) ;output_width +%endif + + pxor xmm0, xmm0 ; clear xmm0 for unpack + +.filter_block1d16_h6_sse2_rowloop: + movq xmm3, MMWORD PTR [rsi - 2] + movq xmm1, MMWORD PTR [rsi + 6] + + movq xmm2, MMWORD PTR [rsi +14] + pslldq xmm2, 8 + + por xmm2, xmm1 + prefetcht2 [rsi+rax-2] + + pslldq xmm1, 8 + por xmm1, xmm3 + + movdqa xmm4, xmm1 + movdqa xmm5, xmm1 + + movdqa xmm6, xmm1 + movdqa xmm7, xmm1 + + punpcklbw xmm3, xmm0 ; xx05 xx04 xx03 xx02 xx01 xx01 xx-1 xx-2 + psrldq xmm4, 1 ; xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 00 -1 + + pmullw xmm3, XMMWORD PTR [rdx] ; x[-2] * H[-2]; Tap 1 + punpcklbw xmm4, xmm0 ; xx06 xx05 xx04 xx03 xx02 xx01 xx00 xx-1 + + psrldq xmm5, 2 ; xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 00 + pmullw xmm4, XMMWORD PTR [rdx+16] ; x[-1] * H[-1]; Tap 2 + + + punpcklbw xmm5, xmm0 ; xx07 xx06 xx05 xx04 xx03 xx02 xx01 xx00 + psrldq xmm6, 3 ; xx xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 + + pmullw xmm5, [rdx+32] ; x[ 0] * H[ 0]; Tap 3 + + punpcklbw xmm6, xmm0 ; xx08 xx07 xx06 xx05 xx04 xx03 xx02 xx01 + psrldq xmm7, 4 ; xx xx xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 + + pmullw xmm6, [rdx+48] ; x[ 1] * h[ 1] ; Tap 4 + + punpcklbw xmm7, xmm0 ; xx09 xx08 xx07 xx06 xx05 xx04 xx03 xx02 + psrldq xmm1, 5 ; xx xx xx xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 + + + pmullw xmm7, [rdx+64] ; x[ 2] * h[ 2] ; Tap 5 + + punpcklbw xmm1, xmm0 ; xx0a xx09 xx08 xx07 xx06 xx05 xx04 xx03 + pmullw xmm1, [rdx+80] ; x[ 3] * h[ 3] ; Tap 6 + + paddsw xmm4, xmm7 + paddsw xmm4, xmm5 + + paddsw xmm4, xmm3 + paddsw xmm4, xmm6 + + paddsw xmm4, xmm1 + paddsw xmm4, [GLOBAL(rd)] + + psraw xmm4, 7 + + packuswb xmm4, xmm0 + punpcklbw xmm4, xmm0 + + movdqa XMMWORD Ptr [rdi], xmm4 + + movdqa xmm3, xmm2 + movdqa xmm4, xmm2 + + movdqa xmm5, xmm2 + movdqa xmm6, xmm2 + + movdqa xmm7, xmm2 + + punpcklbw xmm3, xmm0 ; xx05 xx04 xx03 xx02 xx01 xx01 xx-1 xx-2 + psrldq xmm4, 1 ; xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 00 -1 + + pmullw xmm3, XMMWORD PTR [rdx] ; x[-2] * H[-2]; Tap 1 + punpcklbw xmm4, xmm0 ; xx06 xx05 xx04 xx03 xx02 xx01 xx00 xx-1 + + psrldq xmm5, 2 ; xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 00 + pmullw xmm4, XMMWORD PTR [rdx+16] ; x[-1] * H[-1]; Tap 2 + + + punpcklbw xmm5, xmm0 ; xx07 xx06 xx05 xx04 xx03 xx02 xx01 xx00 + psrldq xmm6, 3 ; xx xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 + + pmullw xmm5, [rdx+32] ; x[ 0] * H[ 0]; Tap 3 + + punpcklbw xmm6, xmm0 ; xx08 xx07 xx06 xx05 xx04 xx03 xx02 xx01 + psrldq xmm7, 4 ; xx xx xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 + + pmullw xmm6, [rdx+48] ; x[ 1] * h[ 1] ; Tap 4 + + punpcklbw xmm7, xmm0 ; xx09 xx08 xx07 xx06 xx05 xx04 xx03 xx02 + psrldq xmm2, 5 ; xx xx xx xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 + + pmullw xmm7, [rdx+64] ; x[ 2] * h[ 2] ; Tap 5 + + punpcklbw xmm2, xmm0 ; xx0a xx09 xx08 xx07 xx06 xx05 xx04 xx03 + pmullw xmm2, [rdx+80] ; x[ 3] * h[ 3] ; Tap 6 + + + paddsw xmm4, xmm7 + paddsw xmm4, xmm5 + + paddsw xmm4, xmm3 + paddsw xmm4, xmm6 + + paddsw xmm4, xmm2 + paddsw xmm4, [GLOBAL(rd)] + + psraw xmm4, 7 + + packuswb xmm4, xmm0 + punpcklbw xmm4, xmm0 + + movdqa XMMWORD Ptr [rdi+16], xmm4 + + lea rsi, [rsi + rax] +%if ABI_IS_32BIT + add rdi, DWORD Ptr arg(5) ;[output_width] +%else + add rdi, r8 +%endif + + dec rcx + jnz .filter_block1d16_h6_sse2_rowloop ; next row + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + + +;void vp8_filter_block1d8_v6_sse2 +;( +; short *src_ptr, +; unsigned char *output_ptr, +; int dst_ptich, +; unsigned int pixels_per_line, +; unsigned int pixel_step, +; unsigned int output_height, +; unsigned int output_width, +; short * vp8_filter +;) +;/************************************************************************************ +; Notes: filter_block1d8_v6 applies a 6 tap filter vertically to the input pixels. The +; input pixel array has output_height rows. +;*************************************************************************************/ +global sym(vp8_filter_block1d8_v6_sse2) PRIVATE +sym(vp8_filter_block1d8_v6_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 8 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + mov rax, arg(7) ;vp8_filter + movsxd rdx, dword ptr arg(3) ;pixels_per_line + + mov rdi, arg(1) ;output_ptr + mov rsi, arg(0) ;src_ptr + + sub rsi, rdx + sub rsi, rdx + + movsxd rcx, DWORD PTR arg(5) ;[output_height] + pxor xmm0, xmm0 ; clear xmm0 + + movdqa xmm7, XMMWORD PTR [GLOBAL(rd)] +%if ABI_IS_32BIT=0 + movsxd r8, dword ptr arg(2) ; dst_ptich +%endif + +.vp8_filter_block1d8_v6_sse2_loop: + movdqa xmm1, XMMWORD PTR [rsi] + pmullw xmm1, [rax] + + movdqa xmm2, XMMWORD PTR [rsi + rdx] + pmullw xmm2, [rax + 16] + + movdqa xmm3, XMMWORD PTR [rsi + rdx * 2] + pmullw xmm3, [rax + 32] + + movdqa xmm5, XMMWORD PTR [rsi + rdx * 4] + pmullw xmm5, [rax + 64] + + add rsi, rdx + movdqa xmm4, XMMWORD PTR [rsi + rdx * 2] + + pmullw xmm4, [rax + 48] + movdqa xmm6, XMMWORD PTR [rsi + rdx * 4] + + pmullw xmm6, [rax + 80] + + paddsw xmm2, xmm5 + paddsw xmm2, xmm3 + + paddsw xmm2, xmm1 + paddsw xmm2, xmm4 + + paddsw xmm2, xmm6 + paddsw xmm2, xmm7 + + psraw xmm2, 7 + packuswb xmm2, xmm0 ; pack and saturate + + movq QWORD PTR [rdi], xmm2 ; store the results in the destination +%if ABI_IS_32BIT + add rdi, DWORD PTR arg(2) ;[dst_ptich] +%else + add rdi, r8 +%endif + dec rcx ; decrement count + jnz .vp8_filter_block1d8_v6_sse2_loop ; next row + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + + +;void vp8_filter_block1d16_v6_sse2 +;( +; unsigned short *src_ptr, +; unsigned char *output_ptr, +; int dst_ptich, +; unsigned int pixels_per_line, +; unsigned int pixel_step, +; unsigned int output_height, +; unsigned int output_width, +; const short *vp8_filter +;) +;/************************************************************************************ +; Notes: filter_block1d16_v6 applies a 6 tap filter vertically to the input pixels. The +; input pixel array has output_height rows. +;*************************************************************************************/ +global sym(vp8_filter_block1d16_v6_sse2) PRIVATE +sym(vp8_filter_block1d16_v6_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 8 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + mov rax, arg(7) ;vp8_filter + movsxd rdx, dword ptr arg(3) ;pixels_per_line + + mov rdi, arg(1) ;output_ptr + mov rsi, arg(0) ;src_ptr + + sub rsi, rdx + sub rsi, rdx + + movsxd rcx, DWORD PTR arg(5) ;[output_height] +%if ABI_IS_32BIT=0 + movsxd r8, dword ptr arg(2) ; dst_ptich +%endif + +.vp8_filter_block1d16_v6_sse2_loop: +; The order for adding 6-tap is 2 5 3 1 4 6. Read in data in that order. + movdqa xmm1, XMMWORD PTR [rsi + rdx] ; line 2 + movdqa xmm2, XMMWORD PTR [rsi + rdx + 16] + pmullw xmm1, [rax + 16] + pmullw xmm2, [rax + 16] + + movdqa xmm3, XMMWORD PTR [rsi + rdx * 4] ; line 5 + movdqa xmm4, XMMWORD PTR [rsi + rdx * 4 + 16] + pmullw xmm3, [rax + 64] + pmullw xmm4, [rax + 64] + + movdqa xmm5, XMMWORD PTR [rsi + rdx * 2] ; line 3 + movdqa xmm6, XMMWORD PTR [rsi + rdx * 2 + 16] + pmullw xmm5, [rax + 32] + pmullw xmm6, [rax + 32] + + movdqa xmm7, XMMWORD PTR [rsi] ; line 1 + movdqa xmm0, XMMWORD PTR [rsi + 16] + pmullw xmm7, [rax] + pmullw xmm0, [rax] + + paddsw xmm1, xmm3 + paddsw xmm2, xmm4 + paddsw xmm1, xmm5 + paddsw xmm2, xmm6 + paddsw xmm1, xmm7 + paddsw xmm2, xmm0 + + add rsi, rdx + + movdqa xmm3, XMMWORD PTR [rsi + rdx * 2] ; line 4 + movdqa xmm4, XMMWORD PTR [rsi + rdx * 2 + 16] + pmullw xmm3, [rax + 48] + pmullw xmm4, [rax + 48] + + movdqa xmm5, XMMWORD PTR [rsi + rdx * 4] ; line 6 + movdqa xmm6, XMMWORD PTR [rsi + rdx * 4 + 16] + pmullw xmm5, [rax + 80] + pmullw xmm6, [rax + 80] + + movdqa xmm7, XMMWORD PTR [GLOBAL(rd)] + pxor xmm0, xmm0 ; clear xmm0 + + paddsw xmm1, xmm3 + paddsw xmm2, xmm4 + paddsw xmm1, xmm5 + paddsw xmm2, xmm6 + + paddsw xmm1, xmm7 + paddsw xmm2, xmm7 + + psraw xmm1, 7 + psraw xmm2, 7 + + packuswb xmm1, xmm2 ; pack and saturate + movdqa XMMWORD PTR [rdi], xmm1 ; store the results in the destination +%if ABI_IS_32BIT + add rdi, DWORD PTR arg(2) ;[dst_ptich] +%else + add rdi, r8 +%endif + dec rcx ; decrement count + jnz .vp8_filter_block1d16_v6_sse2_loop ; next row + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + + +;void vp8_filter_block1d8_h6_only_sse2 +;( +; unsigned char *src_ptr, +; unsigned int src_pixels_per_line, +; unsigned char *output_ptr, +; int dst_ptich, +; unsigned int output_height, +; const short *vp8_filter +;) +; First-pass filter only when yoffset==0 +global sym(vp8_filter_block1d8_h6_only_sse2) PRIVATE +sym(vp8_filter_block1d8_h6_only_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + mov rdx, arg(5) ;vp8_filter + mov rsi, arg(0) ;src_ptr + + mov rdi, arg(2) ;output_ptr + + movsxd rcx, dword ptr arg(4) ;output_height + movsxd rax, dword ptr arg(1) ;src_pixels_per_line ; Pitch for Source +%if ABI_IS_32BIT=0 + movsxd r8, dword ptr arg(3) ;dst_ptich +%endif + pxor xmm0, xmm0 ; clear xmm0 for unpack + +.filter_block1d8_h6_only_rowloop: + movq xmm3, MMWORD PTR [rsi - 2] + movq xmm1, MMWORD PTR [rsi + 6] + + prefetcht2 [rsi+rax-2] + + pslldq xmm1, 8 + por xmm1, xmm3 + + movdqa xmm4, xmm1 + movdqa xmm5, xmm1 + + movdqa xmm6, xmm1 + movdqa xmm7, xmm1 + + punpcklbw xmm3, xmm0 ; xx05 xx04 xx03 xx02 xx01 xx01 xx-1 xx-2 + psrldq xmm4, 1 ; xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 00 -1 + + pmullw xmm3, XMMWORD PTR [rdx] ; x[-2] * H[-2]; Tap 1 + punpcklbw xmm4, xmm0 ; xx06 xx05 xx04 xx03 xx02 xx01 xx00 xx-1 + + psrldq xmm5, 2 ; xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 00 + pmullw xmm4, XMMWORD PTR [rdx+16] ; x[-1] * H[-1]; Tap 2 + + + punpcklbw xmm5, xmm0 ; xx07 xx06 xx05 xx04 xx03 xx02 xx01 xx00 + psrldq xmm6, 3 ; xx xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 + + pmullw xmm5, [rdx+32] ; x[ 0] * H[ 0]; Tap 3 + + punpcklbw xmm6, xmm0 ; xx08 xx07 xx06 xx05 xx04 xx03 xx02 xx01 + psrldq xmm7, 4 ; xx xx xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 + + pmullw xmm6, [rdx+48] ; x[ 1] * h[ 1] ; Tap 4 + + punpcklbw xmm7, xmm0 ; xx09 xx08 xx07 xx06 xx05 xx04 xx03 xx02 + psrldq xmm1, 5 ; xx xx xx xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 + + + pmullw xmm7, [rdx+64] ; x[ 2] * h[ 2] ; Tap 5 + + punpcklbw xmm1, xmm0 ; xx0a xx09 xx08 xx07 xx06 xx05 xx04 xx03 + pmullw xmm1, [rdx+80] ; x[ 3] * h[ 3] ; Tap 6 + + + paddsw xmm4, xmm7 + paddsw xmm4, xmm5 + + paddsw xmm4, xmm3 + paddsw xmm4, xmm6 + + paddsw xmm4, xmm1 + paddsw xmm4, [GLOBAL(rd)] + + psraw xmm4, 7 + + packuswb xmm4, xmm0 + + movq QWORD PTR [rdi], xmm4 ; store the results in the destination + lea rsi, [rsi + rax] + +%if ABI_IS_32BIT + add rdi, DWORD Ptr arg(3) ;dst_ptich +%else + add rdi, r8 +%endif + dec rcx + + jnz .filter_block1d8_h6_only_rowloop ; next row + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + + +;void vp8_filter_block1d16_h6_only_sse2 +;( +; unsigned char *src_ptr, +; unsigned int src_pixels_per_line, +; unsigned char *output_ptr, +; int dst_ptich, +; unsigned int output_height, +; const short *vp8_filter +;) +; First-pass filter only when yoffset==0 +global sym(vp8_filter_block1d16_h6_only_sse2) PRIVATE +sym(vp8_filter_block1d16_h6_only_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + mov rdx, arg(5) ;vp8_filter + mov rsi, arg(0) ;src_ptr + + mov rdi, arg(2) ;output_ptr + + movsxd rcx, dword ptr arg(4) ;output_height + movsxd rax, dword ptr arg(1) ;src_pixels_per_line ; Pitch for Source +%if ABI_IS_32BIT=0 + movsxd r8, dword ptr arg(3) ;dst_ptich +%endif + + pxor xmm0, xmm0 ; clear xmm0 for unpack + +.filter_block1d16_h6_only_sse2_rowloop: + movq xmm3, MMWORD PTR [rsi - 2] + movq xmm1, MMWORD PTR [rsi + 6] + + movq xmm2, MMWORD PTR [rsi +14] + pslldq xmm2, 8 + + por xmm2, xmm1 + prefetcht2 [rsi+rax-2] + + pslldq xmm1, 8 + por xmm1, xmm3 + + movdqa xmm4, xmm1 + movdqa xmm5, xmm1 + + movdqa xmm6, xmm1 + movdqa xmm7, xmm1 + + punpcklbw xmm3, xmm0 ; xx05 xx04 xx03 xx02 xx01 xx01 xx-1 xx-2 + psrldq xmm4, 1 ; xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 00 -1 + + pmullw xmm3, XMMWORD PTR [rdx] ; x[-2] * H[-2]; Tap 1 + punpcklbw xmm4, xmm0 ; xx06 xx05 xx04 xx03 xx02 xx01 xx00 xx-1 + + psrldq xmm5, 2 ; xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 00 + pmullw xmm4, XMMWORD PTR [rdx+16] ; x[-1] * H[-1]; Tap 2 + + punpcklbw xmm5, xmm0 ; xx07 xx06 xx05 xx04 xx03 xx02 xx01 xx00 + psrldq xmm6, 3 ; xx xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 + + pmullw xmm5, [rdx+32] ; x[ 0] * H[ 0]; Tap 3 + + punpcklbw xmm6, xmm0 ; xx08 xx07 xx06 xx05 xx04 xx03 xx02 xx01 + psrldq xmm7, 4 ; xx xx xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 + + pmullw xmm6, [rdx+48] ; x[ 1] * h[ 1] ; Tap 4 + + punpcklbw xmm7, xmm0 ; xx09 xx08 xx07 xx06 xx05 xx04 xx03 xx02 + psrldq xmm1, 5 ; xx xx xx xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 + + pmullw xmm7, [rdx+64] ; x[ 2] * h[ 2] ; Tap 5 + + punpcklbw xmm1, xmm0 ; xx0a xx09 xx08 xx07 xx06 xx05 xx04 xx03 + pmullw xmm1, [rdx+80] ; x[ 3] * h[ 3] ; Tap 6 + + paddsw xmm4, xmm7 + paddsw xmm4, xmm5 + + paddsw xmm4, xmm3 + paddsw xmm4, xmm6 + + paddsw xmm4, xmm1 + paddsw xmm4, [GLOBAL(rd)] + + psraw xmm4, 7 + + packuswb xmm4, xmm0 ; lower 8 bytes + + movq QWORD Ptr [rdi], xmm4 ; store the results in the destination + + movdqa xmm3, xmm2 + movdqa xmm4, xmm2 + + movdqa xmm5, xmm2 + movdqa xmm6, xmm2 + + movdqa xmm7, xmm2 + + punpcklbw xmm3, xmm0 ; xx05 xx04 xx03 xx02 xx01 xx01 xx-1 xx-2 + psrldq xmm4, 1 ; xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 00 -1 + + pmullw xmm3, XMMWORD PTR [rdx] ; x[-2] * H[-2]; Tap 1 + punpcklbw xmm4, xmm0 ; xx06 xx05 xx04 xx03 xx02 xx01 xx00 xx-1 + + psrldq xmm5, 2 ; xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 00 + pmullw xmm4, XMMWORD PTR [rdx+16] ; x[-1] * H[-1]; Tap 2 + + punpcklbw xmm5, xmm0 ; xx07 xx06 xx05 xx04 xx03 xx02 xx01 xx00 + psrldq xmm6, 3 ; xx xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 + + pmullw xmm5, [rdx+32] ; x[ 0] * H[ 0]; Tap 3 + + punpcklbw xmm6, xmm0 ; xx08 xx07 xx06 xx05 xx04 xx03 xx02 xx01 + psrldq xmm7, 4 ; xx xx xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 02 + + pmullw xmm6, [rdx+48] ; x[ 1] * h[ 1] ; Tap 4 + + punpcklbw xmm7, xmm0 ; xx09 xx08 xx07 xx06 xx05 xx04 xx03 xx02 + psrldq xmm2, 5 ; xx xx xx xx xx 0d 0c 0b 0a 09 08 07 06 05 04 03 + + pmullw xmm7, [rdx+64] ; x[ 2] * h[ 2] ; Tap 5 + + punpcklbw xmm2, xmm0 ; xx0a xx09 xx08 xx07 xx06 xx05 xx04 xx03 + pmullw xmm2, [rdx+80] ; x[ 3] * h[ 3] ; Tap 6 + + paddsw xmm4, xmm7 + paddsw xmm4, xmm5 + + paddsw xmm4, xmm3 + paddsw xmm4, xmm6 + + paddsw xmm4, xmm2 + paddsw xmm4, [GLOBAL(rd)] + + psraw xmm4, 7 + + packuswb xmm4, xmm0 ; higher 8 bytes + + movq QWORD Ptr [rdi+8], xmm4 ; store the results in the destination + + lea rsi, [rsi + rax] +%if ABI_IS_32BIT + add rdi, DWORD Ptr arg(3) ;dst_ptich +%else + add rdi, r8 +%endif + + dec rcx + jnz .filter_block1d16_h6_only_sse2_rowloop ; next row + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + + +;void vp8_filter_block1d8_v6_only_sse2 +;( +; unsigned char *src_ptr, +; unsigned int src_pixels_per_line, +; unsigned char *output_ptr, +; int dst_ptich, +; unsigned int output_height, +; const short *vp8_filter +;) +; Second-pass filter only when xoffset==0 +global sym(vp8_filter_block1d8_v6_only_sse2) PRIVATE +sym(vp8_filter_block1d8_v6_only_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + mov rsi, arg(0) ;src_ptr + mov rdi, arg(2) ;output_ptr + + movsxd rcx, dword ptr arg(4) ;output_height + movsxd rdx, dword ptr arg(1) ;src_pixels_per_line + + mov rax, arg(5) ;vp8_filter + + pxor xmm0, xmm0 ; clear xmm0 + + movdqa xmm7, XMMWORD PTR [GLOBAL(rd)] +%if ABI_IS_32BIT=0 + movsxd r8, dword ptr arg(3) ; dst_ptich +%endif + +.vp8_filter_block1d8_v6_only_sse2_loop: + movq xmm1, MMWORD PTR [rsi] + movq xmm2, MMWORD PTR [rsi + rdx] + movq xmm3, MMWORD PTR [rsi + rdx * 2] + movq xmm5, MMWORD PTR [rsi + rdx * 4] + add rsi, rdx + movq xmm4, MMWORD PTR [rsi + rdx * 2] + movq xmm6, MMWORD PTR [rsi + rdx * 4] + + punpcklbw xmm1, xmm0 + pmullw xmm1, [rax] + + punpcklbw xmm2, xmm0 + pmullw xmm2, [rax + 16] + + punpcklbw xmm3, xmm0 + pmullw xmm3, [rax + 32] + + punpcklbw xmm5, xmm0 + pmullw xmm5, [rax + 64] + + punpcklbw xmm4, xmm0 + pmullw xmm4, [rax + 48] + + punpcklbw xmm6, xmm0 + pmullw xmm6, [rax + 80] + + paddsw xmm2, xmm5 + paddsw xmm2, xmm3 + + paddsw xmm2, xmm1 + paddsw xmm2, xmm4 + + paddsw xmm2, xmm6 + paddsw xmm2, xmm7 + + psraw xmm2, 7 + packuswb xmm2, xmm0 ; pack and saturate + + movq QWORD PTR [rdi], xmm2 ; store the results in the destination +%if ABI_IS_32BIT + add rdi, DWORD PTR arg(3) ;[dst_ptich] +%else + add rdi, r8 +%endif + dec rcx ; decrement count + jnz .vp8_filter_block1d8_v6_only_sse2_loop ; next row + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + + +;void vp8_unpack_block1d16_h6_sse2 +;( +; unsigned char *src_ptr, +; unsigned short *output_ptr, +; unsigned int src_pixels_per_line, +; unsigned int output_height, +; unsigned int output_width +;) +global sym(vp8_unpack_block1d16_h6_sse2) PRIVATE +sym(vp8_unpack_block1d16_h6_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 5 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + mov rsi, arg(0) ;src_ptr + mov rdi, arg(1) ;output_ptr + + movsxd rcx, dword ptr arg(3) ;output_height + movsxd rax, dword ptr arg(2) ;src_pixels_per_line ; Pitch for Source + + pxor xmm0, xmm0 ; clear xmm0 for unpack +%if ABI_IS_32BIT=0 + movsxd r8, dword ptr arg(4) ;output_width ; Pitch for Source +%endif + +.unpack_block1d16_h6_sse2_rowloop: + movq xmm1, MMWORD PTR [rsi] ; 0d 0c 0b 0a 09 08 07 06 05 04 03 02 01 00 -1 -2 + movq xmm3, MMWORD PTR [rsi+8] ; make copy of xmm1 + + punpcklbw xmm3, xmm0 ; xx05 xx04 xx03 xx02 xx01 xx01 xx-1 xx-2 + punpcklbw xmm1, xmm0 + + movdqa XMMWORD Ptr [rdi], xmm1 + movdqa XMMWORD Ptr [rdi + 16], xmm3 + + lea rsi, [rsi + rax] +%if ABI_IS_32BIT + add rdi, DWORD Ptr arg(4) ;[output_width] +%else + add rdi, r8 +%endif + dec rcx + jnz .unpack_block1d16_h6_sse2_rowloop ; next row + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + UNSHADOW_ARGS + pop rbp + ret + + +;void vp8_bilinear_predict16x16_sse2 +;( +; unsigned char *src_ptr, +; int src_pixels_per_line, +; int xoffset, +; int yoffset, +; unsigned char *dst_ptr, +; int dst_pitch +;) +extern sym(vp8_bilinear_filters_x86_8) +global sym(vp8_bilinear_predict16x16_sse2) PRIVATE +sym(vp8_bilinear_predict16x16_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + ;const short *HFilter = vp8_bilinear_filters_x86_8[xoffset] + ;const short *VFilter = vp8_bilinear_filters_x86_8[yoffset] + + lea rcx, [GLOBAL(sym(vp8_bilinear_filters_x86_8))] + movsxd rax, dword ptr arg(2) ;xoffset + + cmp rax, 0 ;skip first_pass filter if xoffset=0 + je .b16x16_sp_only + + shl rax, 5 + add rax, rcx ;HFilter + + mov rdi, arg(4) ;dst_ptr + mov rsi, arg(0) ;src_ptr + movsxd rdx, dword ptr arg(5) ;dst_pitch + + movdqa xmm1, [rax] + movdqa xmm2, [rax+16] + + movsxd rax, dword ptr arg(3) ;yoffset + + cmp rax, 0 ;skip second_pass filter if yoffset=0 + je .b16x16_fp_only + + shl rax, 5 + add rax, rcx ;VFilter + + lea rcx, [rdi+rdx*8] + lea rcx, [rcx+rdx*8] + movsxd rdx, dword ptr arg(1) ;src_pixels_per_line + + pxor xmm0, xmm0 + +%if ABI_IS_32BIT=0 + movsxd r8, dword ptr arg(5) ;dst_pitch +%endif + ; get the first horizontal line done + movdqu xmm3, [rsi] ; xx 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 + movdqa xmm4, xmm3 ; make a copy of current line + + punpcklbw xmm3, xmm0 ; xx 00 01 02 03 04 05 06 + punpckhbw xmm4, xmm0 + + pmullw xmm3, xmm1 + pmullw xmm4, xmm1 + + movdqu xmm5, [rsi+1] + movdqa xmm6, xmm5 + + punpcklbw xmm5, xmm0 + punpckhbw xmm6, xmm0 + + pmullw xmm5, xmm2 + pmullw xmm6, xmm2 + + paddw xmm3, xmm5 + paddw xmm4, xmm6 + + paddw xmm3, [GLOBAL(rd)] ; xmm3 += round value + psraw xmm3, VP8_FILTER_SHIFT ; xmm3 /= 128 + + paddw xmm4, [GLOBAL(rd)] + psraw xmm4, VP8_FILTER_SHIFT + + movdqa xmm7, xmm3 + packuswb xmm7, xmm4 + + add rsi, rdx ; next line +.next_row: + movdqu xmm3, [rsi] ; xx 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 + movdqa xmm4, xmm3 ; make a copy of current line + + punpcklbw xmm3, xmm0 ; xx 00 01 02 03 04 05 06 + punpckhbw xmm4, xmm0 + + pmullw xmm3, xmm1 + pmullw xmm4, xmm1 + + movdqu xmm5, [rsi+1] + movdqa xmm6, xmm5 + + punpcklbw xmm5, xmm0 + punpckhbw xmm6, xmm0 + + pmullw xmm5, xmm2 + pmullw xmm6, xmm2 + + paddw xmm3, xmm5 + paddw xmm4, xmm6 + + movdqa xmm5, xmm7 + movdqa xmm6, xmm7 + + punpcklbw xmm5, xmm0 + punpckhbw xmm6, xmm0 + + pmullw xmm5, [rax] + pmullw xmm6, [rax] + + paddw xmm3, [GLOBAL(rd)] ; xmm3 += round value + psraw xmm3, VP8_FILTER_SHIFT ; xmm3 /= 128 + + paddw xmm4, [GLOBAL(rd)] + psraw xmm4, VP8_FILTER_SHIFT + + movdqa xmm7, xmm3 + packuswb xmm7, xmm4 + + pmullw xmm3, [rax+16] + pmullw xmm4, [rax+16] + + paddw xmm3, xmm5 + paddw xmm4, xmm6 + + paddw xmm3, [GLOBAL(rd)] ; xmm3 += round value + psraw xmm3, VP8_FILTER_SHIFT ; xmm3 /= 128 + + paddw xmm4, [GLOBAL(rd)] + psraw xmm4, VP8_FILTER_SHIFT + + packuswb xmm3, xmm4 + movdqa [rdi], xmm3 ; store the results in the destination + + add rsi, rdx ; next line +%if ABI_IS_32BIT + add rdi, DWORD PTR arg(5) ;dst_pitch +%else + add rdi, r8 +%endif + + cmp rdi, rcx + jne .next_row + + jmp .done + +.b16x16_sp_only: + movsxd rax, dword ptr arg(3) ;yoffset + shl rax, 5 + add rax, rcx ;VFilter + + mov rdi, arg(4) ;dst_ptr + mov rsi, arg(0) ;src_ptr + movsxd rdx, dword ptr arg(5) ;dst_pitch + + movdqa xmm1, [rax] + movdqa xmm2, [rax+16] + + lea rcx, [rdi+rdx*8] + lea rcx, [rcx+rdx*8] + movsxd rax, dword ptr arg(1) ;src_pixels_per_line + + pxor xmm0, xmm0 + + ; get the first horizontal line done + movdqu xmm7, [rsi] ; xx 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 + + add rsi, rax ; next line +.next_row_spo: + movdqu xmm3, [rsi] ; xx 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 + + movdqa xmm5, xmm7 + movdqa xmm6, xmm7 + + movdqa xmm4, xmm3 ; make a copy of current line + movdqa xmm7, xmm3 + + punpcklbw xmm5, xmm0 + punpckhbw xmm6, xmm0 + punpcklbw xmm3, xmm0 ; xx 00 01 02 03 04 05 06 + punpckhbw xmm4, xmm0 + + pmullw xmm5, xmm1 + pmullw xmm6, xmm1 + pmullw xmm3, xmm2 + pmullw xmm4, xmm2 + + paddw xmm3, xmm5 + paddw xmm4, xmm6 + + paddw xmm3, [GLOBAL(rd)] ; xmm3 += round value + psraw xmm3, VP8_FILTER_SHIFT ; xmm3 /= 128 + + paddw xmm4, [GLOBAL(rd)] + psraw xmm4, VP8_FILTER_SHIFT + + packuswb xmm3, xmm4 + movdqa [rdi], xmm3 ; store the results in the destination + + add rsi, rax ; next line + add rdi, rdx ;dst_pitch + cmp rdi, rcx + jne .next_row_spo + + jmp .done + +.b16x16_fp_only: + lea rcx, [rdi+rdx*8] + lea rcx, [rcx+rdx*8] + movsxd rax, dword ptr arg(1) ;src_pixels_per_line + pxor xmm0, xmm0 + +.next_row_fpo: + movdqu xmm3, [rsi] ; xx 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 + movdqa xmm4, xmm3 ; make a copy of current line + + punpcklbw xmm3, xmm0 ; xx 00 01 02 03 04 05 06 + punpckhbw xmm4, xmm0 + + pmullw xmm3, xmm1 + pmullw xmm4, xmm1 + + movdqu xmm5, [rsi+1] + movdqa xmm6, xmm5 + + punpcklbw xmm5, xmm0 + punpckhbw xmm6, xmm0 + + pmullw xmm5, xmm2 + pmullw xmm6, xmm2 + + paddw xmm3, xmm5 + paddw xmm4, xmm6 + + paddw xmm3, [GLOBAL(rd)] ; xmm3 += round value + psraw xmm3, VP8_FILTER_SHIFT ; xmm3 /= 128 + + paddw xmm4, [GLOBAL(rd)] + psraw xmm4, VP8_FILTER_SHIFT + + packuswb xmm3, xmm4 + movdqa [rdi], xmm3 ; store the results in the destination + + add rsi, rax ; next line + add rdi, rdx ; dst_pitch + cmp rdi, rcx + jne .next_row_fpo + +.done: + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + + +;void vp8_bilinear_predict8x8_sse2 +;( +; unsigned char *src_ptr, +; int src_pixels_per_line, +; int xoffset, +; int yoffset, +; unsigned char *dst_ptr, +; int dst_pitch +;) +global sym(vp8_bilinear_predict8x8_sse2) PRIVATE +sym(vp8_bilinear_predict8x8_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, 144 ; reserve 144 bytes + + ;const short *HFilter = vp8_bilinear_filters_x86_8[xoffset] + ;const short *VFilter = vp8_bilinear_filters_x86_8[yoffset] + lea rcx, [GLOBAL(sym(vp8_bilinear_filters_x86_8))] + + mov rsi, arg(0) ;src_ptr + movsxd rdx, dword ptr arg(1) ;src_pixels_per_line + + ;Read 9-line unaligned data in and put them on stack. This gives a big + ;performance boost. + movdqu xmm0, [rsi] + lea rax, [rdx + rdx*2] + movdqu xmm1, [rsi+rdx] + movdqu xmm2, [rsi+rdx*2] + add rsi, rax + movdqu xmm3, [rsi] + movdqu xmm4, [rsi+rdx] + movdqu xmm5, [rsi+rdx*2] + add rsi, rax + movdqu xmm6, [rsi] + movdqu xmm7, [rsi+rdx] + + movdqa XMMWORD PTR [rsp], xmm0 + + movdqu xmm0, [rsi+rdx*2] + + movdqa XMMWORD PTR [rsp+16], xmm1 + movdqa XMMWORD PTR [rsp+32], xmm2 + movdqa XMMWORD PTR [rsp+48], xmm3 + movdqa XMMWORD PTR [rsp+64], xmm4 + movdqa XMMWORD PTR [rsp+80], xmm5 + movdqa XMMWORD PTR [rsp+96], xmm6 + movdqa XMMWORD PTR [rsp+112], xmm7 + movdqa XMMWORD PTR [rsp+128], xmm0 + + movsxd rax, dword ptr arg(2) ;xoffset + shl rax, 5 + add rax, rcx ;HFilter + + mov rdi, arg(4) ;dst_ptr + movsxd rdx, dword ptr arg(5) ;dst_pitch + + movdqa xmm1, [rax] + movdqa xmm2, [rax+16] + + movsxd rax, dword ptr arg(3) ;yoffset + shl rax, 5 + add rax, rcx ;VFilter + + lea rcx, [rdi+rdx*8] + + movdqa xmm5, [rax] + movdqa xmm6, [rax+16] + + pxor xmm0, xmm0 + + ; get the first horizontal line done + movdqa xmm3, XMMWORD PTR [rsp] + movdqa xmm4, xmm3 ; make a copy of current line + psrldq xmm4, 1 + + punpcklbw xmm3, xmm0 ; 00 01 02 03 04 05 06 07 + punpcklbw xmm4, xmm0 ; 01 02 03 04 05 06 07 08 + + pmullw xmm3, xmm1 + pmullw xmm4, xmm2 + + paddw xmm3, xmm4 + + paddw xmm3, [GLOBAL(rd)] ; xmm3 += round value + psraw xmm3, VP8_FILTER_SHIFT ; xmm3 /= 128 + + movdqa xmm7, xmm3 + add rsp, 16 ; next line +.next_row8x8: + movdqa xmm3, XMMWORD PTR [rsp] ; 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 + movdqa xmm4, xmm3 ; make a copy of current line + psrldq xmm4, 1 + + punpcklbw xmm3, xmm0 ; 00 01 02 03 04 05 06 07 + punpcklbw xmm4, xmm0 ; 01 02 03 04 05 06 07 08 + + pmullw xmm3, xmm1 + pmullw xmm4, xmm2 + + paddw xmm3, xmm4 + pmullw xmm7, xmm5 + + paddw xmm3, [GLOBAL(rd)] ; xmm3 += round value + psraw xmm3, VP8_FILTER_SHIFT ; xmm3 /= 128 + + movdqa xmm4, xmm3 + + pmullw xmm3, xmm6 + paddw xmm3, xmm7 + + movdqa xmm7, xmm4 + + paddw xmm3, [GLOBAL(rd)] ; xmm3 += round value + psraw xmm3, VP8_FILTER_SHIFT ; xmm3 /= 128 + + packuswb xmm3, xmm0 + movq [rdi], xmm3 ; store the results in the destination + + add rsp, 16 ; next line + add rdi, rdx + + cmp rdi, rcx + jne .next_row8x8 + + ;add rsp, 144 + pop rsp + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + + +SECTION_RODATA +align 16 +rd: + times 8 dw 0x40 diff --git a/thirdparty/libvpx/vp8/common/x86/subpixel_ssse3.asm b/thirdparty/libvpx/vp8/common/x86/subpixel_ssse3.asm new file mode 100644 index 0000000000..c06f24556e --- /dev/null +++ b/thirdparty/libvpx/vp8/common/x86/subpixel_ssse3.asm @@ -0,0 +1,1508 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + + +%include "vpx_ports/x86_abi_support.asm" + +%define BLOCK_HEIGHT_WIDTH 4 +%define VP8_FILTER_WEIGHT 128 +%define VP8_FILTER_SHIFT 7 + + +;/************************************************************************************ +; Notes: filter_block1d_h6 applies a 6 tap filter horizontally to the input pixels. The +; input pixel array has output_height rows. This routine assumes that output_height is an +; even number. This function handles 8 pixels in horizontal direction, calculating ONE +; rows each iteration to take advantage of the 128 bits operations. +; +; This is an implementation of some of the SSE optimizations first seen in ffvp8 +; +;*************************************************************************************/ +;void vp8_filter_block1d8_h6_ssse3 +;( +; unsigned char *src_ptr, +; unsigned int src_pixels_per_line, +; unsigned char *output_ptr, +; unsigned int output_pitch, +; unsigned int output_height, +; unsigned int vp8_filter_index +;) +global sym(vp8_filter_block1d8_h6_ssse3) PRIVATE +sym(vp8_filter_block1d8_h6_ssse3): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + movsxd rdx, DWORD PTR arg(5) ;table index + xor rsi, rsi + shl rdx, 4 + + movdqa xmm7, [GLOBAL(rd)] + + lea rax, [GLOBAL(k0_k5)] + add rax, rdx + mov rdi, arg(2) ;output_ptr + + cmp esi, DWORD PTR [rax] + je vp8_filter_block1d8_h4_ssse3 + + movdqa xmm4, XMMWORD PTR [rax] ;k0_k5 + movdqa xmm5, XMMWORD PTR [rax+256] ;k2_k4 + movdqa xmm6, XMMWORD PTR [rax+128] ;k1_k3 + + mov rsi, arg(0) ;src_ptr + movsxd rax, dword ptr arg(1) ;src_pixels_per_line + movsxd rcx, dword ptr arg(4) ;output_height + + movsxd rdx, dword ptr arg(3) ;output_pitch + + sub rdi, rdx +;xmm3 free +.filter_block1d8_h6_rowloop_ssse3: + movq xmm0, MMWORD PTR [rsi - 2] ; -2 -1 0 1 2 3 4 5 + + movq xmm2, MMWORD PTR [rsi + 3] ; 3 4 5 6 7 8 9 10 + + punpcklbw xmm0, xmm2 ; -2 3 -1 4 0 5 1 6 2 7 3 8 4 9 5 10 + + movdqa xmm1, xmm0 + pmaddubsw xmm0, xmm4 + + movdqa xmm2, xmm1 + pshufb xmm1, [GLOBAL(shuf2bfrom1)] + + pshufb xmm2, [GLOBAL(shuf3bfrom1)] + pmaddubsw xmm1, xmm5 + + lea rdi, [rdi + rdx] + pmaddubsw xmm2, xmm6 + + lea rsi, [rsi + rax] + dec rcx + + paddsw xmm0, xmm1 + paddsw xmm2, xmm7 + + paddsw xmm0, xmm2 + + psraw xmm0, 7 + + packuswb xmm0, xmm0 + + movq MMWORD Ptr [rdi], xmm0 + jnz .filter_block1d8_h6_rowloop_ssse3 + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +vp8_filter_block1d8_h4_ssse3: + movdqa xmm5, XMMWORD PTR [rax+256] ;k2_k4 + movdqa xmm6, XMMWORD PTR [rax+128] ;k1_k3 + + movdqa xmm3, XMMWORD PTR [GLOBAL(shuf2bfrom1)] + movdqa xmm4, XMMWORD PTR [GLOBAL(shuf3bfrom1)] + + mov rsi, arg(0) ;src_ptr + + movsxd rax, dword ptr arg(1) ;src_pixels_per_line + movsxd rcx, dword ptr arg(4) ;output_height + + movsxd rdx, dword ptr arg(3) ;output_pitch + + sub rdi, rdx + +.filter_block1d8_h4_rowloop_ssse3: + movq xmm0, MMWORD PTR [rsi - 2] ; -2 -1 0 1 2 3 4 5 + + movq xmm1, MMWORD PTR [rsi + 3] ; 3 4 5 6 7 8 9 10 + + punpcklbw xmm0, xmm1 ; -2 3 -1 4 0 5 1 6 2 7 3 8 4 9 5 10 + + movdqa xmm2, xmm0 + pshufb xmm0, xmm3 + + pshufb xmm2, xmm4 + pmaddubsw xmm0, xmm5 + + lea rdi, [rdi + rdx] + pmaddubsw xmm2, xmm6 + + lea rsi, [rsi + rax] + dec rcx + + paddsw xmm0, xmm7 + + paddsw xmm0, xmm2 + + psraw xmm0, 7 + + packuswb xmm0, xmm0 + + movq MMWORD Ptr [rdi], xmm0 + + jnz .filter_block1d8_h4_rowloop_ssse3 + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret +;void vp8_filter_block1d16_h6_ssse3 +;( +; unsigned char *src_ptr, +; unsigned int src_pixels_per_line, +; unsigned char *output_ptr, +; unsigned int output_pitch, +; unsigned int output_height, +; unsigned int vp8_filter_index +;) +global sym(vp8_filter_block1d16_h6_ssse3) PRIVATE +sym(vp8_filter_block1d16_h6_ssse3): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + movsxd rdx, DWORD PTR arg(5) ;table index + xor rsi, rsi + shl rdx, 4 ; + + lea rax, [GLOBAL(k0_k5)] + add rax, rdx + + mov rdi, arg(2) ;output_ptr + + mov rsi, arg(0) ;src_ptr + + movdqa xmm4, XMMWORD PTR [rax] ;k0_k5 + movdqa xmm5, XMMWORD PTR [rax+256] ;k2_k4 + movdqa xmm6, XMMWORD PTR [rax+128] ;k1_k3 + + movsxd rax, dword ptr arg(1) ;src_pixels_per_line + movsxd rcx, dword ptr arg(4) ;output_height + movsxd rdx, dword ptr arg(3) ;output_pitch + +.filter_block1d16_h6_rowloop_ssse3: + movq xmm0, MMWORD PTR [rsi - 2] ; -2 -1 0 1 2 3 4 5 + + movq xmm3, MMWORD PTR [rsi + 3] ; 3 4 5 6 7 8 9 10 + + punpcklbw xmm0, xmm3 ; -2 3 -1 4 0 5 1 6 2 7 3 8 4 9 5 10 + + movdqa xmm1, xmm0 + pmaddubsw xmm0, xmm4 + + movdqa xmm2, xmm1 + pshufb xmm1, [GLOBAL(shuf2bfrom1)] + + pshufb xmm2, [GLOBAL(shuf3bfrom1)] + movq xmm3, MMWORD PTR [rsi + 6] + + pmaddubsw xmm1, xmm5 + movq xmm7, MMWORD PTR [rsi + 11] + + pmaddubsw xmm2, xmm6 + punpcklbw xmm3, xmm7 + + paddsw xmm0, xmm1 + movdqa xmm1, xmm3 + + pmaddubsw xmm3, xmm4 + paddsw xmm0, xmm2 + + movdqa xmm2, xmm1 + paddsw xmm0, [GLOBAL(rd)] + + pshufb xmm1, [GLOBAL(shuf2bfrom1)] + pshufb xmm2, [GLOBAL(shuf3bfrom1)] + + psraw xmm0, 7 + pmaddubsw xmm1, xmm5 + + pmaddubsw xmm2, xmm6 + packuswb xmm0, xmm0 + + lea rsi, [rsi + rax] + paddsw xmm3, xmm1 + + paddsw xmm3, xmm2 + + paddsw xmm3, [GLOBAL(rd)] + + psraw xmm3, 7 + + packuswb xmm3, xmm3 + + punpcklqdq xmm0, xmm3 + + movdqa XMMWORD Ptr [rdi], xmm0 + + lea rdi, [rdi + rdx] + dec rcx + jnz .filter_block1d16_h6_rowloop_ssse3 + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +;void vp8_filter_block1d4_h6_ssse3 +;( +; unsigned char *src_ptr, +; unsigned int src_pixels_per_line, +; unsigned char *output_ptr, +; unsigned int output_pitch, +; unsigned int output_height, +; unsigned int vp8_filter_index +;) +global sym(vp8_filter_block1d4_h6_ssse3) PRIVATE +sym(vp8_filter_block1d4_h6_ssse3): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + movsxd rdx, DWORD PTR arg(5) ;table index + xor rsi, rsi + shl rdx, 4 ; + + lea rax, [GLOBAL(k0_k5)] + add rax, rdx + movdqa xmm7, [GLOBAL(rd)] + + cmp esi, DWORD PTR [rax] + je .vp8_filter_block1d4_h4_ssse3 + + movdqa xmm4, XMMWORD PTR [rax] ;k0_k5 + movdqa xmm5, XMMWORD PTR [rax+256] ;k2_k4 + movdqa xmm6, XMMWORD PTR [rax+128] ;k1_k3 + + mov rsi, arg(0) ;src_ptr + mov rdi, arg(2) ;output_ptr + movsxd rax, dword ptr arg(1) ;src_pixels_per_line + movsxd rcx, dword ptr arg(4) ;output_height + + movsxd rdx, dword ptr arg(3) ;output_pitch + +;xmm3 free +.filter_block1d4_h6_rowloop_ssse3: + movdqu xmm0, XMMWORD PTR [rsi - 2] + + movdqa xmm1, xmm0 + pshufb xmm0, [GLOBAL(shuf1b)] + + movdqa xmm2, xmm1 + pshufb xmm1, [GLOBAL(shuf2b)] + pmaddubsw xmm0, xmm4 + pshufb xmm2, [GLOBAL(shuf3b)] + pmaddubsw xmm1, xmm5 + +;-- + pmaddubsw xmm2, xmm6 + + lea rsi, [rsi + rax] +;-- + paddsw xmm0, xmm1 + paddsw xmm0, xmm7 + pxor xmm1, xmm1 + paddsw xmm0, xmm2 + psraw xmm0, 7 + packuswb xmm0, xmm0 + + movd DWORD PTR [rdi], xmm0 + + add rdi, rdx + dec rcx + jnz .filter_block1d4_h6_rowloop_ssse3 + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +.vp8_filter_block1d4_h4_ssse3: + movdqa xmm5, XMMWORD PTR [rax+256] ;k2_k4 + movdqa xmm6, XMMWORD PTR [rax+128] ;k1_k3 + movdqa xmm0, XMMWORD PTR [GLOBAL(shuf2b)] + movdqa xmm3, XMMWORD PTR [GLOBAL(shuf3b)] + + mov rsi, arg(0) ;src_ptr + mov rdi, arg(2) ;output_ptr + movsxd rax, dword ptr arg(1) ;src_pixels_per_line + movsxd rcx, dword ptr arg(4) ;output_height + + movsxd rdx, dword ptr arg(3) ;output_pitch + +.filter_block1d4_h4_rowloop_ssse3: + movdqu xmm1, XMMWORD PTR [rsi - 2] + + movdqa xmm2, xmm1 + pshufb xmm1, xmm0 ;;[GLOBAL(shuf2b)] + pshufb xmm2, xmm3 ;;[GLOBAL(shuf3b)] + pmaddubsw xmm1, xmm5 + +;-- + pmaddubsw xmm2, xmm6 + + lea rsi, [rsi + rax] +;-- + paddsw xmm1, xmm7 + paddsw xmm1, xmm2 + psraw xmm1, 7 + packuswb xmm1, xmm1 + + movd DWORD PTR [rdi], xmm1 + + add rdi, rdx + dec rcx + jnz .filter_block1d4_h4_rowloop_ssse3 + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + + + +;void vp8_filter_block1d16_v6_ssse3 +;( +; unsigned char *src_ptr, +; unsigned int src_pitch, +; unsigned char *output_ptr, +; unsigned int out_pitch, +; unsigned int output_height, +; unsigned int vp8_filter_index +;) +global sym(vp8_filter_block1d16_v6_ssse3) PRIVATE +sym(vp8_filter_block1d16_v6_ssse3): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + movsxd rdx, DWORD PTR arg(5) ;table index + xor rsi, rsi + shl rdx, 4 ; + + lea rax, [GLOBAL(k0_k5)] + add rax, rdx + + cmp esi, DWORD PTR [rax] + je .vp8_filter_block1d16_v4_ssse3 + + movdqa xmm5, XMMWORD PTR [rax] ;k0_k5 + movdqa xmm6, XMMWORD PTR [rax+256] ;k2_k4 + movdqa xmm7, XMMWORD PTR [rax+128] ;k1_k3 + + mov rsi, arg(0) ;src_ptr + movsxd rdx, DWORD PTR arg(1) ;pixels_per_line + mov rdi, arg(2) ;output_ptr + +%if ABI_IS_32BIT=0 + movsxd r8, DWORD PTR arg(3) ;out_pitch +%endif + mov rax, rsi + movsxd rcx, DWORD PTR arg(4) ;output_height + add rax, rdx + + +.vp8_filter_block1d16_v6_ssse3_loop: + movq xmm1, MMWORD PTR [rsi] ;A + movq xmm2, MMWORD PTR [rsi + rdx] ;B + movq xmm3, MMWORD PTR [rsi + rdx * 2] ;C + movq xmm4, MMWORD PTR [rax + rdx * 2] ;D + movq xmm0, MMWORD PTR [rsi + rdx * 4] ;E + + punpcklbw xmm2, xmm4 ;B D + punpcklbw xmm3, xmm0 ;C E + + movq xmm0, MMWORD PTR [rax + rdx * 4] ;F + + pmaddubsw xmm3, xmm6 + punpcklbw xmm1, xmm0 ;A F + pmaddubsw xmm2, xmm7 + pmaddubsw xmm1, xmm5 + + paddsw xmm2, xmm3 + paddsw xmm2, xmm1 + paddsw xmm2, [GLOBAL(rd)] + psraw xmm2, 7 + packuswb xmm2, xmm2 + + movq MMWORD PTR [rdi], xmm2 ;store the results + + movq xmm1, MMWORD PTR [rsi + 8] ;A + movq xmm2, MMWORD PTR [rsi + rdx + 8] ;B + movq xmm3, MMWORD PTR [rsi + rdx * 2 + 8] ;C + movq xmm4, MMWORD PTR [rax + rdx * 2 + 8] ;D + movq xmm0, MMWORD PTR [rsi + rdx * 4 + 8] ;E + + punpcklbw xmm2, xmm4 ;B D + punpcklbw xmm3, xmm0 ;C E + + movq xmm0, MMWORD PTR [rax + rdx * 4 + 8] ;F + pmaddubsw xmm3, xmm6 + punpcklbw xmm1, xmm0 ;A F + pmaddubsw xmm2, xmm7 + pmaddubsw xmm1, xmm5 + + add rsi, rdx + add rax, rdx +;-- +;-- + paddsw xmm2, xmm3 + paddsw xmm2, xmm1 + paddsw xmm2, [GLOBAL(rd)] + psraw xmm2, 7 + packuswb xmm2, xmm2 + + movq MMWORD PTR [rdi+8], xmm2 + +%if ABI_IS_32BIT + add rdi, DWORD PTR arg(3) ;out_pitch +%else + add rdi, r8 +%endif + dec rcx + jnz .vp8_filter_block1d16_v6_ssse3_loop + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +.vp8_filter_block1d16_v4_ssse3: + movdqa xmm6, XMMWORD PTR [rax+256] ;k2_k4 + movdqa xmm7, XMMWORD PTR [rax+128] ;k1_k3 + + mov rsi, arg(0) ;src_ptr + movsxd rdx, DWORD PTR arg(1) ;pixels_per_line + mov rdi, arg(2) ;output_ptr + +%if ABI_IS_32BIT=0 + movsxd r8, DWORD PTR arg(3) ;out_pitch +%endif + mov rax, rsi + movsxd rcx, DWORD PTR arg(4) ;output_height + add rax, rdx + +.vp8_filter_block1d16_v4_ssse3_loop: + movq xmm2, MMWORD PTR [rsi + rdx] ;B + movq xmm3, MMWORD PTR [rsi + rdx * 2] ;C + movq xmm4, MMWORD PTR [rax + rdx * 2] ;D + movq xmm0, MMWORD PTR [rsi + rdx * 4] ;E + + punpcklbw xmm2, xmm4 ;B D + punpcklbw xmm3, xmm0 ;C E + + pmaddubsw xmm3, xmm6 + pmaddubsw xmm2, xmm7 + movq xmm5, MMWORD PTR [rsi + rdx + 8] ;B + movq xmm1, MMWORD PTR [rsi + rdx * 2 + 8] ;C + movq xmm4, MMWORD PTR [rax + rdx * 2 + 8] ;D + movq xmm0, MMWORD PTR [rsi + rdx * 4 + 8] ;E + + paddsw xmm2, [GLOBAL(rd)] + paddsw xmm2, xmm3 + psraw xmm2, 7 + packuswb xmm2, xmm2 + + punpcklbw xmm5, xmm4 ;B D + punpcklbw xmm1, xmm0 ;C E + + pmaddubsw xmm1, xmm6 + pmaddubsw xmm5, xmm7 + + movdqa xmm4, [GLOBAL(rd)] + add rsi, rdx + add rax, rdx +;-- +;-- + paddsw xmm5, xmm1 + paddsw xmm5, xmm4 + psraw xmm5, 7 + packuswb xmm5, xmm5 + + punpcklqdq xmm2, xmm5 + + movdqa XMMWORD PTR [rdi], xmm2 + +%if ABI_IS_32BIT + add rdi, DWORD PTR arg(3) ;out_pitch +%else + add rdi, r8 +%endif + dec rcx + jnz .vp8_filter_block1d16_v4_ssse3_loop + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +;void vp8_filter_block1d8_v6_ssse3 +;( +; unsigned char *src_ptr, +; unsigned int src_pitch, +; unsigned char *output_ptr, +; unsigned int out_pitch, +; unsigned int output_height, +; unsigned int vp8_filter_index +;) +global sym(vp8_filter_block1d8_v6_ssse3) PRIVATE +sym(vp8_filter_block1d8_v6_ssse3): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + movsxd rdx, DWORD PTR arg(5) ;table index + xor rsi, rsi + shl rdx, 4 ; + + lea rax, [GLOBAL(k0_k5)] + add rax, rdx + + movsxd rdx, DWORD PTR arg(1) ;pixels_per_line + mov rdi, arg(2) ;output_ptr +%if ABI_IS_32BIT=0 + movsxd r8, DWORD PTR arg(3) ; out_pitch +%endif + movsxd rcx, DWORD PTR arg(4) ;[output_height] + + cmp esi, DWORD PTR [rax] + je .vp8_filter_block1d8_v4_ssse3 + + movdqa xmm5, XMMWORD PTR [rax] ;k0_k5 + movdqa xmm6, XMMWORD PTR [rax+256] ;k2_k4 + movdqa xmm7, XMMWORD PTR [rax+128] ;k1_k3 + + mov rsi, arg(0) ;src_ptr + + mov rax, rsi + add rax, rdx + +.vp8_filter_block1d8_v6_ssse3_loop: + movq xmm1, MMWORD PTR [rsi] ;A + movq xmm2, MMWORD PTR [rsi + rdx] ;B + movq xmm3, MMWORD PTR [rsi + rdx * 2] ;C + movq xmm4, MMWORD PTR [rax + rdx * 2] ;D + movq xmm0, MMWORD PTR [rsi + rdx * 4] ;E + + punpcklbw xmm2, xmm4 ;B D + punpcklbw xmm3, xmm0 ;C E + + movq xmm0, MMWORD PTR [rax + rdx * 4] ;F + movdqa xmm4, [GLOBAL(rd)] + + pmaddubsw xmm3, xmm6 + punpcklbw xmm1, xmm0 ;A F + pmaddubsw xmm2, xmm7 + pmaddubsw xmm1, xmm5 + add rsi, rdx + add rax, rdx +;-- +;-- + paddsw xmm2, xmm3 + paddsw xmm2, xmm1 + paddsw xmm2, xmm4 + psraw xmm2, 7 + packuswb xmm2, xmm2 + + movq MMWORD PTR [rdi], xmm2 + +%if ABI_IS_32BIT + add rdi, DWORD PTR arg(3) ;[out_pitch] +%else + add rdi, r8 +%endif + dec rcx + jnz .vp8_filter_block1d8_v6_ssse3_loop + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +.vp8_filter_block1d8_v4_ssse3: + movdqa xmm6, XMMWORD PTR [rax+256] ;k2_k4 + movdqa xmm7, XMMWORD PTR [rax+128] ;k1_k3 + movdqa xmm5, [GLOBAL(rd)] + + mov rsi, arg(0) ;src_ptr + + mov rax, rsi + add rax, rdx + +.vp8_filter_block1d8_v4_ssse3_loop: + movq xmm2, MMWORD PTR [rsi + rdx] ;B + movq xmm3, MMWORD PTR [rsi + rdx * 2] ;C + movq xmm4, MMWORD PTR [rax + rdx * 2] ;D + movq xmm0, MMWORD PTR [rsi + rdx * 4] ;E + + punpcklbw xmm2, xmm4 ;B D + punpcklbw xmm3, xmm0 ;C E + + pmaddubsw xmm3, xmm6 + pmaddubsw xmm2, xmm7 + add rsi, rdx + add rax, rdx +;-- +;-- + paddsw xmm2, xmm3 + paddsw xmm2, xmm5 + psraw xmm2, 7 + packuswb xmm2, xmm2 + + movq MMWORD PTR [rdi], xmm2 + +%if ABI_IS_32BIT + add rdi, DWORD PTR arg(3) ;[out_pitch] +%else + add rdi, r8 +%endif + dec rcx + jnz .vp8_filter_block1d8_v4_ssse3_loop + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret +;void vp8_filter_block1d4_v6_ssse3 +;( +; unsigned char *src_ptr, +; unsigned int src_pitch, +; unsigned char *output_ptr, +; unsigned int out_pitch, +; unsigned int output_height, +; unsigned int vp8_filter_index +;) +global sym(vp8_filter_block1d4_v6_ssse3) PRIVATE +sym(vp8_filter_block1d4_v6_ssse3): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + movsxd rdx, DWORD PTR arg(5) ;table index + xor rsi, rsi + shl rdx, 4 ; + + lea rax, [GLOBAL(k0_k5)] + add rax, rdx + + movsxd rdx, DWORD PTR arg(1) ;pixels_per_line + mov rdi, arg(2) ;output_ptr +%if ABI_IS_32BIT=0 + movsxd r8, DWORD PTR arg(3) ; out_pitch +%endif + movsxd rcx, DWORD PTR arg(4) ;[output_height] + + cmp esi, DWORD PTR [rax] + je .vp8_filter_block1d4_v4_ssse3 + + movq mm5, MMWORD PTR [rax] ;k0_k5 + movq mm6, MMWORD PTR [rax+256] ;k2_k4 + movq mm7, MMWORD PTR [rax+128] ;k1_k3 + + mov rsi, arg(0) ;src_ptr + + mov rax, rsi + add rax, rdx + +.vp8_filter_block1d4_v6_ssse3_loop: + movd mm1, DWORD PTR [rsi] ;A + movd mm2, DWORD PTR [rsi + rdx] ;B + movd mm3, DWORD PTR [rsi + rdx * 2] ;C + movd mm4, DWORD PTR [rax + rdx * 2] ;D + movd mm0, DWORD PTR [rsi + rdx * 4] ;E + + punpcklbw mm2, mm4 ;B D + punpcklbw mm3, mm0 ;C E + + movd mm0, DWORD PTR [rax + rdx * 4] ;F + + movq mm4, [GLOBAL(rd)] + + pmaddubsw mm3, mm6 + punpcklbw mm1, mm0 ;A F + pmaddubsw mm2, mm7 + pmaddubsw mm1, mm5 + add rsi, rdx + add rax, rdx +;-- +;-- + paddsw mm2, mm3 + paddsw mm2, mm1 + paddsw mm2, mm4 + psraw mm2, 7 + packuswb mm2, mm2 + + movd DWORD PTR [rdi], mm2 + +%if ABI_IS_32BIT + add rdi, DWORD PTR arg(3) ;[out_pitch] +%else + add rdi, r8 +%endif + dec rcx + jnz .vp8_filter_block1d4_v6_ssse3_loop + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + UNSHADOW_ARGS + pop rbp + ret + +.vp8_filter_block1d4_v4_ssse3: + movq mm6, MMWORD PTR [rax+256] ;k2_k4 + movq mm7, MMWORD PTR [rax+128] ;k1_k3 + movq mm5, MMWORD PTR [GLOBAL(rd)] + + mov rsi, arg(0) ;src_ptr + + mov rax, rsi + add rax, rdx + +.vp8_filter_block1d4_v4_ssse3_loop: + movd mm2, DWORD PTR [rsi + rdx] ;B + movd mm3, DWORD PTR [rsi + rdx * 2] ;C + movd mm4, DWORD PTR [rax + rdx * 2] ;D + movd mm0, DWORD PTR [rsi + rdx * 4] ;E + + punpcklbw mm2, mm4 ;B D + punpcklbw mm3, mm0 ;C E + + pmaddubsw mm3, mm6 + pmaddubsw mm2, mm7 + add rsi, rdx + add rax, rdx +;-- +;-- + paddsw mm2, mm3 + paddsw mm2, mm5 + psraw mm2, 7 + packuswb mm2, mm2 + + movd DWORD PTR [rdi], mm2 + +%if ABI_IS_32BIT + add rdi, DWORD PTR arg(3) ;[out_pitch] +%else + add rdi, r8 +%endif + dec rcx + jnz .vp8_filter_block1d4_v4_ssse3_loop + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + UNSHADOW_ARGS + pop rbp + ret + +;void vp8_bilinear_predict16x16_ssse3 +;( +; unsigned char *src_ptr, +; int src_pixels_per_line, +; int xoffset, +; int yoffset, +; unsigned char *dst_ptr, +; int dst_pitch +;) +global sym(vp8_bilinear_predict16x16_ssse3) PRIVATE +sym(vp8_bilinear_predict16x16_ssse3): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + lea rcx, [GLOBAL(vp8_bilinear_filters_ssse3)] + movsxd rax, dword ptr arg(2) ; xoffset + + cmp rax, 0 ; skip first_pass filter if xoffset=0 + je .b16x16_sp_only + + shl rax, 4 + lea rax, [rax + rcx] ; HFilter + + mov rdi, arg(4) ; dst_ptr + mov rsi, arg(0) ; src_ptr + movsxd rdx, dword ptr arg(5) ; dst_pitch + + movdqa xmm1, [rax] + + movsxd rax, dword ptr arg(3) ; yoffset + + cmp rax, 0 ; skip second_pass filter if yoffset=0 + je .b16x16_fp_only + + shl rax, 4 + lea rax, [rax + rcx] ; VFilter + + lea rcx, [rdi+rdx*8] + lea rcx, [rcx+rdx*8] + movsxd rdx, dword ptr arg(1) ; src_pixels_per_line + + movdqa xmm2, [rax] + +%if ABI_IS_32BIT=0 + movsxd r8, dword ptr arg(5) ; dst_pitch +%endif + movq xmm3, [rsi] ; 00 01 02 03 04 05 06 07 + movq xmm5, [rsi+1] ; 01 02 03 04 05 06 07 08 + + punpcklbw xmm3, xmm5 ; 00 01 01 02 02 03 03 04 04 05 05 06 06 07 07 08 + movq xmm4, [rsi+8] ; 08 09 10 11 12 13 14 15 + + movq xmm5, [rsi+9] ; 09 10 11 12 13 14 15 16 + + lea rsi, [rsi + rdx] ; next line + + pmaddubsw xmm3, xmm1 ; 00 02 04 06 08 10 12 14 + + punpcklbw xmm4, xmm5 ; 08 09 09 10 10 11 11 12 12 13 13 14 14 15 15 16 + pmaddubsw xmm4, xmm1 ; 01 03 05 07 09 11 13 15 + + paddw xmm3, [GLOBAL(rd)] ; xmm3 += round value + psraw xmm3, VP8_FILTER_SHIFT ; xmm3 /= 128 + + paddw xmm4, [GLOBAL(rd)] ; xmm4 += round value + psraw xmm4, VP8_FILTER_SHIFT ; xmm4 /= 128 + + movdqa xmm7, xmm3 + packuswb xmm7, xmm4 ; 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 + +.next_row: + movq xmm6, [rsi] ; 00 01 02 03 04 05 06 07 + movq xmm5, [rsi+1] ; 01 02 03 04 05 06 07 08 + + punpcklbw xmm6, xmm5 + movq xmm4, [rsi+8] ; 08 09 10 11 12 13 14 15 + + movq xmm5, [rsi+9] ; 09 10 11 12 13 14 15 16 + lea rsi, [rsi + rdx] ; next line + + pmaddubsw xmm6, xmm1 + + punpcklbw xmm4, xmm5 + pmaddubsw xmm4, xmm1 + + paddw xmm6, [GLOBAL(rd)] ; xmm6 += round value + psraw xmm6, VP8_FILTER_SHIFT ; xmm6 /= 128 + + paddw xmm4, [GLOBAL(rd)] ; xmm4 += round value + psraw xmm4, VP8_FILTER_SHIFT ; xmm4 /= 128 + + packuswb xmm6, xmm4 + movdqa xmm5, xmm7 + + punpcklbw xmm5, xmm6 + pmaddubsw xmm5, xmm2 + + punpckhbw xmm7, xmm6 + pmaddubsw xmm7, xmm2 + + paddw xmm5, [GLOBAL(rd)] ; xmm5 += round value + psraw xmm5, VP8_FILTER_SHIFT ; xmm5 /= 128 + + paddw xmm7, [GLOBAL(rd)] ; xmm7 += round value + psraw xmm7, VP8_FILTER_SHIFT ; xmm7 /= 128 + + packuswb xmm5, xmm7 + movdqa xmm7, xmm6 + + movdqa [rdi], xmm5 ; store the results in the destination +%if ABI_IS_32BIT + add rdi, DWORD PTR arg(5) ; dst_pitch +%else + add rdi, r8 +%endif + + cmp rdi, rcx + jne .next_row + + jmp .done + +.b16x16_sp_only: + movsxd rax, dword ptr arg(3) ; yoffset + shl rax, 4 + lea rax, [rax + rcx] ; VFilter + + mov rdi, arg(4) ; dst_ptr + mov rsi, arg(0) ; src_ptr + movsxd rdx, dword ptr arg(5) ; dst_pitch + + movdqa xmm1, [rax] ; VFilter + + lea rcx, [rdi+rdx*8] + lea rcx, [rcx+rdx*8] + movsxd rax, dword ptr arg(1) ; src_pixels_per_line + + ; get the first horizontal line done + movq xmm4, [rsi] ; load row 0 + movq xmm2, [rsi + 8] ; load row 0 + + lea rsi, [rsi + rax] ; next line +.next_row_sp: + movq xmm3, [rsi] ; load row + 1 + movq xmm5, [rsi + 8] ; load row + 1 + + punpcklbw xmm4, xmm3 + punpcklbw xmm2, xmm5 + + pmaddubsw xmm4, xmm1 + movq xmm7, [rsi + rax] ; load row + 2 + + pmaddubsw xmm2, xmm1 + movq xmm6, [rsi + rax + 8] ; load row + 2 + + punpcklbw xmm3, xmm7 + punpcklbw xmm5, xmm6 + + pmaddubsw xmm3, xmm1 + paddw xmm4, [GLOBAL(rd)] + + pmaddubsw xmm5, xmm1 + paddw xmm2, [GLOBAL(rd)] + + psraw xmm4, VP8_FILTER_SHIFT + psraw xmm2, VP8_FILTER_SHIFT + + packuswb xmm4, xmm2 + paddw xmm3, [GLOBAL(rd)] + + movdqa [rdi], xmm4 ; store row 0 + paddw xmm5, [GLOBAL(rd)] + + psraw xmm3, VP8_FILTER_SHIFT + psraw xmm5, VP8_FILTER_SHIFT + + packuswb xmm3, xmm5 + movdqa xmm4, xmm7 + + movdqa [rdi + rdx],xmm3 ; store row 1 + lea rsi, [rsi + 2*rax] + + movdqa xmm2, xmm6 + lea rdi, [rdi + 2*rdx] + + cmp rdi, rcx + jne .next_row_sp + + jmp .done + +.b16x16_fp_only: + lea rcx, [rdi+rdx*8] + lea rcx, [rcx+rdx*8] + movsxd rax, dword ptr arg(1) ; src_pixels_per_line + +.next_row_fp: + movq xmm2, [rsi] ; 00 01 02 03 04 05 06 07 + movq xmm4, [rsi+1] ; 01 02 03 04 05 06 07 08 + + punpcklbw xmm2, xmm4 + movq xmm3, [rsi+8] ; 08 09 10 11 12 13 14 15 + + pmaddubsw xmm2, xmm1 + movq xmm4, [rsi+9] ; 09 10 11 12 13 14 15 16 + + lea rsi, [rsi + rax] ; next line + punpcklbw xmm3, xmm4 + + pmaddubsw xmm3, xmm1 + movq xmm5, [rsi] + + paddw xmm2, [GLOBAL(rd)] + movq xmm7, [rsi+1] + + movq xmm6, [rsi+8] + psraw xmm2, VP8_FILTER_SHIFT + + punpcklbw xmm5, xmm7 + movq xmm7, [rsi+9] + + paddw xmm3, [GLOBAL(rd)] + pmaddubsw xmm5, xmm1 + + psraw xmm3, VP8_FILTER_SHIFT + punpcklbw xmm6, xmm7 + + packuswb xmm2, xmm3 + pmaddubsw xmm6, xmm1 + + movdqa [rdi], xmm2 ; store the results in the destination + paddw xmm5, [GLOBAL(rd)] + + lea rdi, [rdi + rdx] ; dst_pitch + psraw xmm5, VP8_FILTER_SHIFT + + paddw xmm6, [GLOBAL(rd)] + psraw xmm6, VP8_FILTER_SHIFT + + packuswb xmm5, xmm6 + lea rsi, [rsi + rax] ; next line + + movdqa [rdi], xmm5 ; store the results in the destination + lea rdi, [rdi + rdx] ; dst_pitch + + cmp rdi, rcx + + jne .next_row_fp + +.done: + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +;void vp8_bilinear_predict8x8_ssse3 +;( +; unsigned char *src_ptr, +; int src_pixels_per_line, +; int xoffset, +; int yoffset, +; unsigned char *dst_ptr, +; int dst_pitch +;) +global sym(vp8_bilinear_predict8x8_ssse3) PRIVATE +sym(vp8_bilinear_predict8x8_ssse3): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, 144 ; reserve 144 bytes + + lea rcx, [GLOBAL(vp8_bilinear_filters_ssse3)] + + mov rsi, arg(0) ;src_ptr + movsxd rdx, dword ptr arg(1) ;src_pixels_per_line + + ;Read 9-line unaligned data in and put them on stack. This gives a big + ;performance boost. + movdqu xmm0, [rsi] + lea rax, [rdx + rdx*2] + movdqu xmm1, [rsi+rdx] + movdqu xmm2, [rsi+rdx*2] + add rsi, rax + movdqu xmm3, [rsi] + movdqu xmm4, [rsi+rdx] + movdqu xmm5, [rsi+rdx*2] + add rsi, rax + movdqu xmm6, [rsi] + movdqu xmm7, [rsi+rdx] + + movdqa XMMWORD PTR [rsp], xmm0 + + movdqu xmm0, [rsi+rdx*2] + + movdqa XMMWORD PTR [rsp+16], xmm1 + movdqa XMMWORD PTR [rsp+32], xmm2 + movdqa XMMWORD PTR [rsp+48], xmm3 + movdqa XMMWORD PTR [rsp+64], xmm4 + movdqa XMMWORD PTR [rsp+80], xmm5 + movdqa XMMWORD PTR [rsp+96], xmm6 + movdqa XMMWORD PTR [rsp+112], xmm7 + movdqa XMMWORD PTR [rsp+128], xmm0 + + movsxd rax, dword ptr arg(2) ; xoffset + cmp rax, 0 ; skip first_pass filter if xoffset=0 + je .b8x8_sp_only + + shl rax, 4 + add rax, rcx ; HFilter + + mov rdi, arg(4) ; dst_ptr + movsxd rdx, dword ptr arg(5) ; dst_pitch + + movdqa xmm0, [rax] + + movsxd rax, dword ptr arg(3) ; yoffset + cmp rax, 0 ; skip second_pass filter if yoffset=0 + je .b8x8_fp_only + + shl rax, 4 + lea rax, [rax + rcx] ; VFilter + + lea rcx, [rdi+rdx*8] + + movdqa xmm1, [rax] + + ; get the first horizontal line done + movdqa xmm3, [rsp] ; 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 + movdqa xmm5, xmm3 ; 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 xx + + psrldq xmm5, 1 + lea rsp, [rsp + 16] ; next line + + punpcklbw xmm3, xmm5 ; 00 01 01 02 02 03 03 04 04 05 05 06 06 07 07 08 + pmaddubsw xmm3, xmm0 ; 00 02 04 06 08 10 12 14 + + paddw xmm3, [GLOBAL(rd)] ; xmm3 += round value + psraw xmm3, VP8_FILTER_SHIFT ; xmm3 /= 128 + + movdqa xmm7, xmm3 + packuswb xmm7, xmm7 ; 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 + +.next_row: + movdqa xmm6, [rsp] ; 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 + lea rsp, [rsp + 16] ; next line + + movdqa xmm5, xmm6 + + psrldq xmm5, 1 + + punpcklbw xmm6, xmm5 + pmaddubsw xmm6, xmm0 + + paddw xmm6, [GLOBAL(rd)] ; xmm6 += round value + psraw xmm6, VP8_FILTER_SHIFT ; xmm6 /= 128 + + packuswb xmm6, xmm6 + + punpcklbw xmm7, xmm6 + pmaddubsw xmm7, xmm1 + + paddw xmm7, [GLOBAL(rd)] ; xmm7 += round value + psraw xmm7, VP8_FILTER_SHIFT ; xmm7 /= 128 + + packuswb xmm7, xmm7 + + movq [rdi], xmm7 ; store the results in the destination + lea rdi, [rdi + rdx] + + movdqa xmm7, xmm6 + + cmp rdi, rcx + jne .next_row + + jmp .done8x8 + +.b8x8_sp_only: + movsxd rax, dword ptr arg(3) ; yoffset + shl rax, 4 + lea rax, [rax + rcx] ; VFilter + + mov rdi, arg(4) ;dst_ptr + movsxd rdx, dword ptr arg(5) ; dst_pitch + + movdqa xmm0, [rax] ; VFilter + + movq xmm1, XMMWORD PTR [rsp] + movq xmm2, XMMWORD PTR [rsp+16] + + movq xmm3, XMMWORD PTR [rsp+32] + punpcklbw xmm1, xmm2 + + movq xmm4, XMMWORD PTR [rsp+48] + punpcklbw xmm2, xmm3 + + movq xmm5, XMMWORD PTR [rsp+64] + punpcklbw xmm3, xmm4 + + movq xmm6, XMMWORD PTR [rsp+80] + punpcklbw xmm4, xmm5 + + movq xmm7, XMMWORD PTR [rsp+96] + punpcklbw xmm5, xmm6 + + pmaddubsw xmm1, xmm0 + pmaddubsw xmm2, xmm0 + + pmaddubsw xmm3, xmm0 + pmaddubsw xmm4, xmm0 + + pmaddubsw xmm5, xmm0 + punpcklbw xmm6, xmm7 + + pmaddubsw xmm6, xmm0 + paddw xmm1, [GLOBAL(rd)] + + paddw xmm2, [GLOBAL(rd)] + psraw xmm1, VP8_FILTER_SHIFT + + paddw xmm3, [GLOBAL(rd)] + psraw xmm2, VP8_FILTER_SHIFT + + paddw xmm4, [GLOBAL(rd)] + psraw xmm3, VP8_FILTER_SHIFT + + paddw xmm5, [GLOBAL(rd)] + psraw xmm4, VP8_FILTER_SHIFT + + paddw xmm6, [GLOBAL(rd)] + psraw xmm5, VP8_FILTER_SHIFT + + psraw xmm6, VP8_FILTER_SHIFT + packuswb xmm1, xmm1 + + packuswb xmm2, xmm2 + movq [rdi], xmm1 + + packuswb xmm3, xmm3 + movq [rdi+rdx], xmm2 + + packuswb xmm4, xmm4 + movq xmm1, XMMWORD PTR [rsp+112] + + lea rdi, [rdi + 2*rdx] + movq xmm2, XMMWORD PTR [rsp+128] + + packuswb xmm5, xmm5 + movq [rdi], xmm3 + + packuswb xmm6, xmm6 + movq [rdi+rdx], xmm4 + + lea rdi, [rdi + 2*rdx] + punpcklbw xmm7, xmm1 + + movq [rdi], xmm5 + pmaddubsw xmm7, xmm0 + + movq [rdi+rdx], xmm6 + punpcklbw xmm1, xmm2 + + pmaddubsw xmm1, xmm0 + paddw xmm7, [GLOBAL(rd)] + + psraw xmm7, VP8_FILTER_SHIFT + paddw xmm1, [GLOBAL(rd)] + + psraw xmm1, VP8_FILTER_SHIFT + packuswb xmm7, xmm7 + + packuswb xmm1, xmm1 + lea rdi, [rdi + 2*rdx] + + movq [rdi], xmm7 + + movq [rdi+rdx], xmm1 + lea rsp, [rsp + 144] + + jmp .done8x8 + +.b8x8_fp_only: + lea rcx, [rdi+rdx*8] + +.next_row_fp: + movdqa xmm1, XMMWORD PTR [rsp] + movdqa xmm3, XMMWORD PTR [rsp+16] + + movdqa xmm2, xmm1 + movdqa xmm5, XMMWORD PTR [rsp+32] + + psrldq xmm2, 1 + movdqa xmm7, XMMWORD PTR [rsp+48] + + movdqa xmm4, xmm3 + psrldq xmm4, 1 + + movdqa xmm6, xmm5 + psrldq xmm6, 1 + + punpcklbw xmm1, xmm2 + pmaddubsw xmm1, xmm0 + + punpcklbw xmm3, xmm4 + pmaddubsw xmm3, xmm0 + + punpcklbw xmm5, xmm6 + pmaddubsw xmm5, xmm0 + + movdqa xmm2, xmm7 + psrldq xmm2, 1 + + punpcklbw xmm7, xmm2 + pmaddubsw xmm7, xmm0 + + paddw xmm1, [GLOBAL(rd)] + psraw xmm1, VP8_FILTER_SHIFT + + paddw xmm3, [GLOBAL(rd)] + psraw xmm3, VP8_FILTER_SHIFT + + paddw xmm5, [GLOBAL(rd)] + psraw xmm5, VP8_FILTER_SHIFT + + paddw xmm7, [GLOBAL(rd)] + psraw xmm7, VP8_FILTER_SHIFT + + packuswb xmm1, xmm1 + packuswb xmm3, xmm3 + + packuswb xmm5, xmm5 + movq [rdi], xmm1 + + packuswb xmm7, xmm7 + movq [rdi+rdx], xmm3 + + lea rdi, [rdi + 2*rdx] + movq [rdi], xmm5 + + lea rsp, [rsp + 4*16] + movq [rdi+rdx], xmm7 + + lea rdi, [rdi + 2*rdx] + cmp rdi, rcx + + jne .next_row_fp + + lea rsp, [rsp + 16] + +.done8x8: + ;add rsp, 144 + pop rsp + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +SECTION_RODATA +align 16 +shuf1b: + db 0, 5, 1, 6, 2, 7, 3, 8, 4, 9, 5, 10, 6, 11, 7, 12 +shuf2b: + db 2, 4, 3, 5, 4, 6, 5, 7, 6, 8, 7, 9, 8, 10, 9, 11 +shuf3b: + db 1, 3, 2, 4, 3, 5, 4, 6, 5, 7, 6, 8, 7, 9, 8, 10 + +align 16 +shuf2bfrom1: + db 4, 8, 6, 1, 8, 3, 1, 5, 3, 7, 5, 9, 7,11, 9,13 +align 16 +shuf3bfrom1: + db 2, 6, 4, 8, 6, 1, 8, 3, 1, 5, 3, 7, 5, 9, 7,11 + +align 16 +rd: + times 8 dw 0x40 + +align 16 +k0_k5: + times 8 db 0, 0 ;placeholder + times 8 db 0, 0 + times 8 db 2, 1 + times 8 db 0, 0 + times 8 db 3, 3 + times 8 db 0, 0 + times 8 db 1, 2 + times 8 db 0, 0 +k1_k3: + times 8 db 0, 0 ;placeholder + times 8 db -6, 12 + times 8 db -11, 36 + times 8 db -9, 50 + times 8 db -16, 77 + times 8 db -6, 93 + times 8 db -8, 108 + times 8 db -1, 123 +k2_k4: + times 8 db 128, 0 ;placeholder + times 8 db 123, -1 + times 8 db 108, -8 + times 8 db 93, -6 + times 8 db 77, -16 + times 8 db 50, -9 + times 8 db 36, -11 + times 8 db 12, -6 +align 16 +vp8_bilinear_filters_ssse3: + times 8 db 128, 0 + times 8 db 112, 16 + times 8 db 96, 32 + times 8 db 80, 48 + times 8 db 64, 64 + times 8 db 48, 80 + times 8 db 32, 96 + times 8 db 16, 112 + diff --git a/thirdparty/libvpx/vp8/common/x86/vp8_asm_stubs.c b/thirdparty/libvpx/vp8/common/x86/vp8_asm_stubs.c new file mode 100644 index 0000000000..fb0b57eb1c --- /dev/null +++ b/thirdparty/libvpx/vp8/common/x86/vp8_asm_stubs.c @@ -0,0 +1,625 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "vpx_config.h" +#include "vp8_rtcd.h" +#include "vpx_ports/mem.h" +#include "filter_x86.h" + +extern const short vp8_six_tap_mmx[8][6*8]; + +extern void vp8_filter_block1d_h6_mmx +( + unsigned char *src_ptr, + unsigned short *output_ptr, + unsigned int src_pixels_per_line, + unsigned int pixel_step, + unsigned int output_height, + unsigned int output_width, + const short *vp8_filter +); +extern void vp8_filter_block1dc_v6_mmx +( + unsigned short *src_ptr, + unsigned char *output_ptr, + int output_pitch, + unsigned int pixels_per_line, + unsigned int pixel_step, + unsigned int output_height, + unsigned int output_width, + const short *vp8_filter +); +extern void vp8_filter_block1d8_h6_sse2 +( + unsigned char *src_ptr, + unsigned short *output_ptr, + unsigned int src_pixels_per_line, + unsigned int pixel_step, + unsigned int output_height, + unsigned int output_width, + const short *vp8_filter +); +extern void vp8_filter_block1d16_h6_sse2 +( + unsigned char *src_ptr, + unsigned short *output_ptr, + unsigned int src_pixels_per_line, + unsigned int pixel_step, + unsigned int output_height, + unsigned int output_width, + const short *vp8_filter +); +extern void vp8_filter_block1d8_v6_sse2 +( + unsigned short *src_ptr, + unsigned char *output_ptr, + int dst_ptich, + unsigned int pixels_per_line, + unsigned int pixel_step, + unsigned int output_height, + unsigned int output_width, + const short *vp8_filter +); +extern void vp8_filter_block1d16_v6_sse2 +( + unsigned short *src_ptr, + unsigned char *output_ptr, + int dst_ptich, + unsigned int pixels_per_line, + unsigned int pixel_step, + unsigned int output_height, + unsigned int output_width, + const short *vp8_filter +); +extern void vp8_unpack_block1d16_h6_sse2 +( + unsigned char *src_ptr, + unsigned short *output_ptr, + unsigned int src_pixels_per_line, + unsigned int output_height, + unsigned int output_width +); +extern void vp8_filter_block1d8_h6_only_sse2 +( + unsigned char *src_ptr, + unsigned int src_pixels_per_line, + unsigned char *output_ptr, + int dst_ptich, + unsigned int output_height, + const short *vp8_filter +); +extern void vp8_filter_block1d16_h6_only_sse2 +( + unsigned char *src_ptr, + unsigned int src_pixels_per_line, + unsigned char *output_ptr, + int dst_ptich, + unsigned int output_height, + const short *vp8_filter +); +extern void vp8_filter_block1d8_v6_only_sse2 +( + unsigned char *src_ptr, + unsigned int src_pixels_per_line, + unsigned char *output_ptr, + int dst_ptich, + unsigned int output_height, + const short *vp8_filter +); + + +#if HAVE_MMX +void vp8_sixtap_predict4x4_mmx +( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch +) +{ + DECLARE_ALIGNED(16, unsigned short, FData2[16*16]); /* Temp data bufffer used in filtering */ + const short *HFilter, *VFilter; + HFilter = vp8_six_tap_mmx[xoffset]; + vp8_filter_block1d_h6_mmx(src_ptr - (2 * src_pixels_per_line), FData2, src_pixels_per_line, 1, 9, 8, HFilter); + VFilter = vp8_six_tap_mmx[yoffset]; + vp8_filter_block1dc_v6_mmx(FData2 + 8, dst_ptr, dst_pitch, 8, 4 , 4, 4, VFilter); + +} + + +void vp8_sixtap_predict16x16_mmx +( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch +) +{ + + DECLARE_ALIGNED(16, unsigned short, FData2[24*24]); /* Temp data bufffer used in filtering */ + + const short *HFilter, *VFilter; + + + HFilter = vp8_six_tap_mmx[xoffset]; + + vp8_filter_block1d_h6_mmx(src_ptr - (2 * src_pixels_per_line), FData2, src_pixels_per_line, 1, 21, 32, HFilter); + vp8_filter_block1d_h6_mmx(src_ptr - (2 * src_pixels_per_line) + 4, FData2 + 4, src_pixels_per_line, 1, 21, 32, HFilter); + vp8_filter_block1d_h6_mmx(src_ptr - (2 * src_pixels_per_line) + 8, FData2 + 8, src_pixels_per_line, 1, 21, 32, HFilter); + vp8_filter_block1d_h6_mmx(src_ptr - (2 * src_pixels_per_line) + 12, FData2 + 12, src_pixels_per_line, 1, 21, 32, HFilter); + + VFilter = vp8_six_tap_mmx[yoffset]; + vp8_filter_block1dc_v6_mmx(FData2 + 32, dst_ptr, dst_pitch, 32, 16 , 16, 16, VFilter); + vp8_filter_block1dc_v6_mmx(FData2 + 36, dst_ptr + 4, dst_pitch, 32, 16 , 16, 16, VFilter); + vp8_filter_block1dc_v6_mmx(FData2 + 40, dst_ptr + 8, dst_pitch, 32, 16 , 16, 16, VFilter); + vp8_filter_block1dc_v6_mmx(FData2 + 44, dst_ptr + 12, dst_pitch, 32, 16 , 16, 16, VFilter); + +} + + +void vp8_sixtap_predict8x8_mmx +( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch +) +{ + + DECLARE_ALIGNED(16, unsigned short, FData2[256]); /* Temp data bufffer used in filtering */ + + const short *HFilter, *VFilter; + + HFilter = vp8_six_tap_mmx[xoffset]; + vp8_filter_block1d_h6_mmx(src_ptr - (2 * src_pixels_per_line), FData2, src_pixels_per_line, 1, 13, 16, HFilter); + vp8_filter_block1d_h6_mmx(src_ptr - (2 * src_pixels_per_line) + 4, FData2 + 4, src_pixels_per_line, 1, 13, 16, HFilter); + + VFilter = vp8_six_tap_mmx[yoffset]; + vp8_filter_block1dc_v6_mmx(FData2 + 16, dst_ptr, dst_pitch, 16, 8 , 8, 8, VFilter); + vp8_filter_block1dc_v6_mmx(FData2 + 20, dst_ptr + 4, dst_pitch, 16, 8 , 8, 8, VFilter); + +} + + +void vp8_sixtap_predict8x4_mmx +( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch +) +{ + + DECLARE_ALIGNED(16, unsigned short, FData2[256]); /* Temp data bufffer used in filtering */ + + const short *HFilter, *VFilter; + + HFilter = vp8_six_tap_mmx[xoffset]; + vp8_filter_block1d_h6_mmx(src_ptr - (2 * src_pixels_per_line), FData2, src_pixels_per_line, 1, 9, 16, HFilter); + vp8_filter_block1d_h6_mmx(src_ptr - (2 * src_pixels_per_line) + 4, FData2 + 4, src_pixels_per_line, 1, 9, 16, HFilter); + + VFilter = vp8_six_tap_mmx[yoffset]; + vp8_filter_block1dc_v6_mmx(FData2 + 16, dst_ptr, dst_pitch, 16, 8 , 4, 8, VFilter); + vp8_filter_block1dc_v6_mmx(FData2 + 20, dst_ptr + 4, dst_pitch, 16, 8 , 4, 8, VFilter); + +} + + + +void vp8_bilinear_predict16x16_mmx +( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch +) +{ + vp8_bilinear_predict8x8_mmx(src_ptr, src_pixels_per_line, xoffset, yoffset, dst_ptr, dst_pitch); + vp8_bilinear_predict8x8_mmx(src_ptr + 8, src_pixels_per_line, xoffset, yoffset, dst_ptr + 8, dst_pitch); + vp8_bilinear_predict8x8_mmx(src_ptr + 8 * src_pixels_per_line, src_pixels_per_line, xoffset, yoffset, dst_ptr + dst_pitch * 8, dst_pitch); + vp8_bilinear_predict8x8_mmx(src_ptr + 8 * src_pixels_per_line + 8, src_pixels_per_line, xoffset, yoffset, dst_ptr + dst_pitch * 8 + 8, dst_pitch); +} +#endif + + +#if HAVE_SSE2 +void vp8_sixtap_predict16x16_sse2 +( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch + +) +{ + DECLARE_ALIGNED(16, unsigned short, FData2[24*24]); /* Temp data bufffer used in filtering */ + + const short *HFilter, *VFilter; + + if (xoffset) + { + if (yoffset) + { + HFilter = vp8_six_tap_mmx[xoffset]; + vp8_filter_block1d16_h6_sse2(src_ptr - (2 * src_pixels_per_line), FData2, src_pixels_per_line, 1, 21, 32, HFilter); + VFilter = vp8_six_tap_mmx[yoffset]; + vp8_filter_block1d16_v6_sse2(FData2 + 32, dst_ptr, dst_pitch, 32, 16 , 16, dst_pitch, VFilter); + } + else + { + /* First-pass only */ + HFilter = vp8_six_tap_mmx[xoffset]; + vp8_filter_block1d16_h6_only_sse2(src_ptr, src_pixels_per_line, dst_ptr, dst_pitch, 16, HFilter); + } + } + else + { + /* Second-pass only */ + VFilter = vp8_six_tap_mmx[yoffset]; + vp8_unpack_block1d16_h6_sse2(src_ptr - (2 * src_pixels_per_line), FData2, src_pixels_per_line, 21, 32); + vp8_filter_block1d16_v6_sse2(FData2 + 32, dst_ptr, dst_pitch, 32, 16 , 16, dst_pitch, VFilter); + } +} + + +void vp8_sixtap_predict8x8_sse2 +( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch +) +{ + DECLARE_ALIGNED(16, unsigned short, FData2[256]); /* Temp data bufffer used in filtering */ + const short *HFilter, *VFilter; + + if (xoffset) + { + if (yoffset) + { + HFilter = vp8_six_tap_mmx[xoffset]; + vp8_filter_block1d8_h6_sse2(src_ptr - (2 * src_pixels_per_line), FData2, src_pixels_per_line, 1, 13, 16, HFilter); + VFilter = vp8_six_tap_mmx[yoffset]; + vp8_filter_block1d8_v6_sse2(FData2 + 16, dst_ptr, dst_pitch, 16, 8 , 8, dst_pitch, VFilter); + } + else + { + /* First-pass only */ + HFilter = vp8_six_tap_mmx[xoffset]; + vp8_filter_block1d8_h6_only_sse2(src_ptr, src_pixels_per_line, dst_ptr, dst_pitch, 8, HFilter); + } + } + else + { + /* Second-pass only */ + VFilter = vp8_six_tap_mmx[yoffset]; + vp8_filter_block1d8_v6_only_sse2(src_ptr - (2 * src_pixels_per_line), src_pixels_per_line, dst_ptr, dst_pitch, 8, VFilter); + } +} + + +void vp8_sixtap_predict8x4_sse2 +( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch +) +{ + DECLARE_ALIGNED(16, unsigned short, FData2[256]); /* Temp data bufffer used in filtering */ + const short *HFilter, *VFilter; + + if (xoffset) + { + if (yoffset) + { + HFilter = vp8_six_tap_mmx[xoffset]; + vp8_filter_block1d8_h6_sse2(src_ptr - (2 * src_pixels_per_line), FData2, src_pixels_per_line, 1, 9, 16, HFilter); + VFilter = vp8_six_tap_mmx[yoffset]; + vp8_filter_block1d8_v6_sse2(FData2 + 16, dst_ptr, dst_pitch, 16, 8 , 4, dst_pitch, VFilter); + } + else + { + /* First-pass only */ + HFilter = vp8_six_tap_mmx[xoffset]; + vp8_filter_block1d8_h6_only_sse2(src_ptr, src_pixels_per_line, dst_ptr, dst_pitch, 4, HFilter); + } + } + else + { + /* Second-pass only */ + VFilter = vp8_six_tap_mmx[yoffset]; + vp8_filter_block1d8_v6_only_sse2(src_ptr - (2 * src_pixels_per_line), src_pixels_per_line, dst_ptr, dst_pitch, 4, VFilter); + } +} + +#endif + +#if HAVE_SSSE3 + +extern void vp8_filter_block1d8_h6_ssse3 +( + unsigned char *src_ptr, + unsigned int src_pixels_per_line, + unsigned char *output_ptr, + unsigned int output_pitch, + unsigned int output_height, + unsigned int vp8_filter_index +); + +extern void vp8_filter_block1d16_h6_ssse3 +( + unsigned char *src_ptr, + unsigned int src_pixels_per_line, + unsigned char *output_ptr, + unsigned int output_pitch, + unsigned int output_height, + unsigned int vp8_filter_index +); + +extern void vp8_filter_block1d16_v6_ssse3 +( + unsigned char *src_ptr, + unsigned int src_pitch, + unsigned char *output_ptr, + unsigned int out_pitch, + unsigned int output_height, + unsigned int vp8_filter_index +); + +extern void vp8_filter_block1d8_v6_ssse3 +( + unsigned char *src_ptr, + unsigned int src_pitch, + unsigned char *output_ptr, + unsigned int out_pitch, + unsigned int output_height, + unsigned int vp8_filter_index +); + +extern void vp8_filter_block1d4_h6_ssse3 +( + unsigned char *src_ptr, + unsigned int src_pixels_per_line, + unsigned char *output_ptr, + unsigned int output_pitch, + unsigned int output_height, + unsigned int vp8_filter_index +); + +extern void vp8_filter_block1d4_v6_ssse3 +( + unsigned char *src_ptr, + unsigned int src_pitch, + unsigned char *output_ptr, + unsigned int out_pitch, + unsigned int output_height, + unsigned int vp8_filter_index +); + +void vp8_sixtap_predict16x16_ssse3 +( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch + +) +{ + DECLARE_ALIGNED(16, unsigned char, FData2[24*24]); + + if (xoffset) + { + if (yoffset) + { + vp8_filter_block1d16_h6_ssse3(src_ptr - (2 * src_pixels_per_line), + src_pixels_per_line, FData2, + 16, 21, xoffset); + vp8_filter_block1d16_v6_ssse3(FData2 , 16, dst_ptr, dst_pitch, + 16, yoffset); + } + else + { + /* First-pass only */ + vp8_filter_block1d16_h6_ssse3(src_ptr, src_pixels_per_line, + dst_ptr, dst_pitch, 16, xoffset); + } + } + else + { + if (yoffset) + { + /* Second-pass only */ + vp8_filter_block1d16_v6_ssse3(src_ptr - (2 * src_pixels_per_line), + src_pixels_per_line, + dst_ptr, dst_pitch, 16, yoffset); + } + else + { + /* ssse3 second-pass only function couldn't handle (xoffset==0 && + * yoffset==0) case correctly. Add copy function here to guarantee + * six-tap function handles all possible offsets. */ + vp8_copy_mem16x16(src_ptr, src_pixels_per_line, dst_ptr, dst_pitch); + } + } +} + +void vp8_sixtap_predict8x8_ssse3 +( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch +) +{ + DECLARE_ALIGNED(16, unsigned char, FData2[256]); + + if (xoffset) + { + if (yoffset) + { + vp8_filter_block1d8_h6_ssse3(src_ptr - (2 * src_pixels_per_line), + src_pixels_per_line, FData2, + 8, 13, xoffset); + vp8_filter_block1d8_v6_ssse3(FData2, 8, dst_ptr, dst_pitch, + 8, yoffset); + } + else + { + vp8_filter_block1d8_h6_ssse3(src_ptr, src_pixels_per_line, + dst_ptr, dst_pitch, 8, xoffset); + } + } + else + { + if (yoffset) + { + /* Second-pass only */ + vp8_filter_block1d8_v6_ssse3(src_ptr - (2 * src_pixels_per_line), + src_pixels_per_line, + dst_ptr, dst_pitch, 8, yoffset); + } + else + { + /* ssse3 second-pass only function couldn't handle (xoffset==0 && + * yoffset==0) case correctly. Add copy function here to guarantee + * six-tap function handles all possible offsets. */ + vp8_copy_mem8x8(src_ptr, src_pixels_per_line, dst_ptr, dst_pitch); + } + } +} + + +void vp8_sixtap_predict8x4_ssse3 +( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch +) +{ + DECLARE_ALIGNED(16, unsigned char, FData2[256]); + + if (xoffset) + { + if (yoffset) + { + vp8_filter_block1d8_h6_ssse3(src_ptr - (2 * src_pixels_per_line), + src_pixels_per_line, FData2, + 8, 9, xoffset); + vp8_filter_block1d8_v6_ssse3(FData2, 8, dst_ptr, dst_pitch, + 4, yoffset); + } + else + { + /* First-pass only */ + vp8_filter_block1d8_h6_ssse3(src_ptr, src_pixels_per_line, + dst_ptr, dst_pitch, 4, xoffset); + } + } + else + { + if (yoffset) + { + /* Second-pass only */ + vp8_filter_block1d8_v6_ssse3(src_ptr - (2 * src_pixels_per_line), + src_pixels_per_line, + dst_ptr, dst_pitch, 4, yoffset); + } + else + { + /* ssse3 second-pass only function couldn't handle (xoffset==0 && + * yoffset==0) case correctly. Add copy function here to guarantee + * six-tap function handles all possible offsets. */ + vp8_copy_mem8x4(src_ptr, src_pixels_per_line, dst_ptr, dst_pitch); + } + } +} + +void vp8_sixtap_predict4x4_ssse3 +( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch +) +{ + DECLARE_ALIGNED(16, unsigned char, FData2[4*9]); + + if (xoffset) + { + if (yoffset) + { + vp8_filter_block1d4_h6_ssse3(src_ptr - (2 * src_pixels_per_line), + src_pixels_per_line, + FData2, 4, 9, xoffset); + vp8_filter_block1d4_v6_ssse3(FData2, 4, dst_ptr, dst_pitch, + 4, yoffset); + } + else + { + vp8_filter_block1d4_h6_ssse3(src_ptr, src_pixels_per_line, + dst_ptr, dst_pitch, 4, xoffset); + } + } + else + { + if (yoffset) + { + vp8_filter_block1d4_v6_ssse3(src_ptr - (2 * src_pixels_per_line), + src_pixels_per_line, + dst_ptr, dst_pitch, 4, yoffset); + } + else + { + /* ssse3 second-pass only function couldn't handle (xoffset==0 && + * yoffset==0) case correctly. Add copy function here to guarantee + * six-tap function handles all possible offsets. */ + int r; + + for (r = 0; r < 4; r++) + { + dst_ptr[0] = src_ptr[0]; + dst_ptr[1] = src_ptr[1]; + dst_ptr[2] = src_ptr[2]; + dst_ptr[3] = src_ptr[3]; + dst_ptr += dst_pitch; + src_ptr += src_pixels_per_line; + } + } + } +} + +#endif diff --git a/thirdparty/libvpx/vp8/common/x86/vp8_loopfilter_mmx.asm b/thirdparty/libvpx/vp8/common/x86/vp8_loopfilter_mmx.asm new file mode 100644 index 0000000000..88a07b9f3f --- /dev/null +++ b/thirdparty/libvpx/vp8/common/x86/vp8_loopfilter_mmx.asm @@ -0,0 +1,1753 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + + +%include "vpx_ports/x86_abi_support.asm" + + +;void vp8_loop_filter_horizontal_edge_mmx +;( +; unsigned char *src_ptr, +; int src_pixel_step, +; const char *blimit, +; const char *limit, +; const char *thresh, +; int count +;) +global sym(vp8_loop_filter_horizontal_edge_mmx) PRIVATE +sym(vp8_loop_filter_horizontal_edge_mmx): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, 32 ; reserve 32 bytes + %define t0 [rsp + 0] ;__declspec(align(16)) char t0[8]; + %define t1 [rsp + 16] ;__declspec(align(16)) char t1[8]; + + mov rsi, arg(0) ;src_ptr + movsxd rax, dword ptr arg(1) ;src_pixel_step ; destination pitch? + + movsxd rcx, dword ptr arg(5) ;count +.next8_h: + mov rdx, arg(3) ;limit + movq mm7, [rdx] + mov rdi, rsi ; rdi points to row +1 for indirect addressing + add rdi, rax + + ; calculate breakout conditions + movq mm2, [rdi+2*rax] ; q3 + movq mm1, [rsi+2*rax] ; q2 + movq mm6, mm1 ; q2 + psubusb mm1, mm2 ; q2-=q3 + psubusb mm2, mm6 ; q3-=q2 + por mm1, mm2 ; abs(q3-q2) + psubusb mm1, mm7 ; + + + movq mm4, [rsi+rax] ; q1 + movq mm3, mm4 ; q1 + psubusb mm4, mm6 ; q1-=q2 + psubusb mm6, mm3 ; q2-=q1 + por mm4, mm6 ; abs(q2-q1) + + psubusb mm4, mm7 + por mm1, mm4 + + movq mm4, [rsi] ; q0 + movq mm0, mm4 ; q0 + psubusb mm4, mm3 ; q0-=q1 + psubusb mm3, mm0 ; q1-=q0 + por mm4, mm3 ; abs(q0-q1) + movq t0, mm4 ; save to t0 + psubusb mm4, mm7 + por mm1, mm4 + + + neg rax ; negate pitch to deal with above border + + movq mm2, [rsi+4*rax] ; p3 + movq mm4, [rdi+4*rax] ; p2 + movq mm5, mm4 ; p2 + psubusb mm4, mm2 ; p2-=p3 + psubusb mm2, mm5 ; p3-=p2 + por mm4, mm2 ; abs(p3 - p2) + psubusb mm4, mm7 + por mm1, mm4 + + + movq mm4, [rsi+2*rax] ; p1 + movq mm3, mm4 ; p1 + psubusb mm4, mm5 ; p1-=p2 + psubusb mm5, mm3 ; p2-=p1 + por mm4, mm5 ; abs(p2 - p1) + psubusb mm4, mm7 + por mm1, mm4 + + movq mm2, mm3 ; p1 + + movq mm4, [rsi+rax] ; p0 + movq mm5, mm4 ; p0 + psubusb mm4, mm3 ; p0-=p1 + psubusb mm3, mm5 ; p1-=p0 + por mm4, mm3 ; abs(p1 - p0) + movq t1, mm4 ; save to t1 + psubusb mm4, mm7 + por mm1, mm4 + + movq mm3, [rdi] ; q1 + movq mm4, mm3 ; q1 + psubusb mm3, mm2 ; q1-=p1 + psubusb mm2, mm4 ; p1-=q1 + por mm2, mm3 ; abs(p1-q1) + pand mm2, [GLOBAL(tfe)] ; set lsb of each byte to zero + psrlw mm2, 1 ; abs(p1-q1)/2 + + movq mm6, mm5 ; p0 + movq mm3, [rsi] ; q0 + psubusb mm5, mm3 ; p0-=q0 + psubusb mm3, mm6 ; q0-=p0 + por mm5, mm3 ; abs(p0 - q0) + paddusb mm5, mm5 ; abs(p0-q0)*2 + paddusb mm5, mm2 ; abs (p0 - q0) *2 + abs(p1-q1)/2 + + mov rdx, arg(2) ;blimit ; get blimit + movq mm7, [rdx] ; blimit + + psubusb mm5, mm7 ; abs (p0 - q0) *2 + abs(p1-q1)/2 > blimit + por mm1, mm5 + pxor mm5, mm5 + pcmpeqb mm1, mm5 ; mask mm1 + + ; calculate high edge variance + mov rdx, arg(4) ;thresh ; get thresh + movq mm7, [rdx] ; + movq mm4, t0 ; get abs (q1 - q0) + psubusb mm4, mm7 + movq mm3, t1 ; get abs (p1 - p0) + psubusb mm3, mm7 + paddb mm4, mm3 ; abs(q1 - q0) > thresh || abs(p1 - p0) > thresh + + pcmpeqb mm4, mm5 + + pcmpeqb mm5, mm5 + pxor mm4, mm5 + + + ; start work on filters + movq mm2, [rsi+2*rax] ; p1 + movq mm7, [rdi] ; q1 + pxor mm2, [GLOBAL(t80)] ; p1 offset to convert to signed values + pxor mm7, [GLOBAL(t80)] ; q1 offset to convert to signed values + psubsb mm2, mm7 ; p1 - q1 + pand mm2, mm4 ; high var mask (hvm)(p1 - q1) + pxor mm6, [GLOBAL(t80)] ; offset to convert to signed values + pxor mm0, [GLOBAL(t80)] ; offset to convert to signed values + movq mm3, mm0 ; q0 + psubsb mm0, mm6 ; q0 - p0 + paddsb mm2, mm0 ; 1 * (q0 - p0) + hvm(p1 - q1) + paddsb mm2, mm0 ; 2 * (q0 - p0) + hvm(p1 - q1) + paddsb mm2, mm0 ; 3 * (q0 - p0) + hvm(p1 - q1) + pand mm1, mm2 ; mask filter values we don't care about + movq mm2, mm1 + paddsb mm1, [GLOBAL(t4)] ; 3* (q0 - p0) + hvm(p1 - q1) + 4 + paddsb mm2, [GLOBAL(t3)] ; 3* (q0 - p0) + hvm(p1 - q1) + 3 + + pxor mm0, mm0 ; + pxor mm5, mm5 + punpcklbw mm0, mm2 ; + punpckhbw mm5, mm2 ; + psraw mm0, 11 ; + psraw mm5, 11 + packsswb mm0, mm5 + movq mm2, mm0 ; (3* (q0 - p0) + hvm(p1 - q1) + 3) >> 3; + + pxor mm0, mm0 ; 0 + movq mm5, mm1 ; abcdefgh + punpcklbw mm0, mm1 ; e0f0g0h0 + psraw mm0, 11 ; sign extended shift right by 3 + pxor mm1, mm1 ; 0 + punpckhbw mm1, mm5 ; a0b0c0d0 + psraw mm1, 11 ; sign extended shift right by 3 + movq mm5, mm0 ; save results + + packsswb mm0, mm1 ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>3 + paddsw mm5, [GLOBAL(ones)] + paddsw mm1, [GLOBAL(ones)] + psraw mm5, 1 ; partial shifted one more time for 2nd tap + psraw mm1, 1 ; partial shifted one more time for 2nd tap + packsswb mm5, mm1 ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>4 + pandn mm4, mm5 ; high edge variance additive + + paddsb mm6, mm2 ; p0+= p0 add + pxor mm6, [GLOBAL(t80)] ; unoffset + movq [rsi+rax], mm6 ; write back + + movq mm6, [rsi+2*rax] ; p1 + pxor mm6, [GLOBAL(t80)] ; reoffset + paddsb mm6, mm4 ; p1+= p1 add + pxor mm6, [GLOBAL(t80)] ; unoffset + movq [rsi+2*rax], mm6 ; write back + + psubsb mm3, mm0 ; q0-= q0 add + pxor mm3, [GLOBAL(t80)] ; unoffset + movq [rsi], mm3 ; write back + + psubsb mm7, mm4 ; q1-= q1 add + pxor mm7, [GLOBAL(t80)] ; unoffset + movq [rdi], mm7 ; write back + + add rsi,8 + neg rax + dec rcx + jnz .next8_h + + add rsp, 32 + pop rsp + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + UNSHADOW_ARGS + pop rbp + ret + + +;void vp8_loop_filter_vertical_edge_mmx +;( +; unsigned char *src_ptr, +; int src_pixel_step, +; const char *blimit, +; const char *limit, +; const char *thresh, +; int count +;) +global sym(vp8_loop_filter_vertical_edge_mmx) PRIVATE +sym(vp8_loop_filter_vertical_edge_mmx): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, 64 ; reserve 64 bytes + %define t0 [rsp + 0] ;__declspec(align(16)) char t0[8]; + %define t1 [rsp + 16] ;__declspec(align(16)) char t1[8]; + %define srct [rsp + 32] ;__declspec(align(16)) char srct[32]; + + mov rsi, arg(0) ;src_ptr + movsxd rax, dword ptr arg(1) ;src_pixel_step ; destination pitch? + + lea rsi, [rsi + rax*4 - 4] + + movsxd rcx, dword ptr arg(5) ;count +.next8_v: + mov rdi, rsi ; rdi points to row +1 for indirect addressing + add rdi, rax + + + ;transpose + movq mm6, [rsi+2*rax] ; 67 66 65 64 63 62 61 60 + movq mm7, mm6 ; 77 76 75 74 73 72 71 70 + + punpckhbw mm7, [rdi+2*rax] ; 77 67 76 66 75 65 74 64 + punpcklbw mm6, [rdi+2*rax] ; 73 63 72 62 71 61 70 60 + + movq mm4, [rsi] ; 47 46 45 44 43 42 41 40 + movq mm5, mm4 ; 47 46 45 44 43 42 41 40 + + punpckhbw mm5, [rsi+rax] ; 57 47 56 46 55 45 54 44 + punpcklbw mm4, [rsi+rax] ; 53 43 52 42 51 41 50 40 + + movq mm3, mm5 ; 57 47 56 46 55 45 54 44 + punpckhwd mm5, mm7 ; 77 67 57 47 76 66 56 46 + + punpcklwd mm3, mm7 ; 75 65 55 45 74 64 54 44 + movq mm2, mm4 ; 53 43 52 42 51 41 50 40 + + punpckhwd mm4, mm6 ; 73 63 53 43 72 62 52 42 + punpcklwd mm2, mm6 ; 71 61 51 41 70 60 50 40 + + neg rax + movq mm6, [rsi+rax*2] ; 27 26 25 24 23 22 21 20 + + movq mm1, mm6 ; 27 26 25 24 23 22 21 20 + punpckhbw mm6, [rsi+rax] ; 37 27 36 36 35 25 34 24 + + punpcklbw mm1, [rsi+rax] ; 33 23 32 22 31 21 30 20 + movq mm7, [rsi+rax*4]; ; 07 06 05 04 03 02 01 00 + + punpckhbw mm7, [rdi+rax*4] ; 17 07 16 06 15 05 14 04 + movq mm0, mm7 ; 17 07 16 06 15 05 14 04 + + punpckhwd mm7, mm6 ; 37 27 17 07 36 26 16 06 + punpcklwd mm0, mm6 ; 35 25 15 05 34 24 14 04 + + movq mm6, mm7 ; 37 27 17 07 36 26 16 06 + punpckhdq mm7, mm5 ; 77 67 57 47 37 27 17 07 = q3 + + punpckldq mm6, mm5 ; 76 66 56 46 36 26 16 06 = q2 + + movq mm5, mm6 ; 76 66 56 46 36 26 16 06 + psubusb mm5, mm7 ; q2-q3 + + psubusb mm7, mm6 ; q3-q2 + por mm7, mm5; ; mm7=abs (q3-q2) + + movq mm5, mm0 ; 35 25 15 05 34 24 14 04 + punpckhdq mm5, mm3 ; 75 65 55 45 35 25 15 05 = q1 + + punpckldq mm0, mm3 ; 74 64 54 44 34 24 15 04 = q0 + movq mm3, mm5 ; 75 65 55 45 35 25 15 05 = q1 + + psubusb mm3, mm6 ; q1-q2 + psubusb mm6, mm5 ; q2-q1 + + por mm6, mm3 ; mm6=abs(q2-q1) + lea rdx, srct + + movq [rdx+24], mm5 ; save q1 + movq [rdx+16], mm0 ; save q0 + + movq mm3, [rsi+rax*4] ; 07 06 05 04 03 02 01 00 + punpcklbw mm3, [rdi+rax*4] ; 13 03 12 02 11 01 10 00 + + movq mm0, mm3 ; 13 03 12 02 11 01 10 00 + punpcklwd mm0, mm1 ; 31 21 11 01 30 20 10 00 + + punpckhwd mm3, mm1 ; 33 23 13 03 32 22 12 02 + movq mm1, mm0 ; 31 21 11 01 30 20 10 00 + + punpckldq mm0, mm2 ; 70 60 50 40 30 20 10 00 =p3 + punpckhdq mm1, mm2 ; 71 61 51 41 31 21 11 01 =p2 + + movq mm2, mm1 ; 71 61 51 41 31 21 11 01 =p2 + psubusb mm2, mm0 ; p2-p3 + + psubusb mm0, mm1 ; p3-p2 + por mm0, mm2 ; mm0=abs(p3-p2) + + movq mm2, mm3 ; 33 23 13 03 32 22 12 02 + punpckldq mm2, mm4 ; 72 62 52 42 32 22 12 02 = p1 + + punpckhdq mm3, mm4 ; 73 63 53 43 33 23 13 03 = p0 + movq [rdx+8], mm3 ; save p0 + + movq [rdx], mm2 ; save p1 + movq mm5, mm2 ; mm5 = p1 + + psubusb mm2, mm1 ; p1-p2 + psubusb mm1, mm5 ; p2-p1 + + por mm1, mm2 ; mm1=abs(p2-p1) + mov rdx, arg(3) ;limit + + movq mm4, [rdx] ; mm4 = limit + psubusb mm7, mm4 + + psubusb mm0, mm4 + psubusb mm1, mm4 + + psubusb mm6, mm4 + por mm7, mm6 + + por mm0, mm1 + por mm0, mm7 ; abs(q3-q2) > limit || abs(p3-p2) > limit ||abs(p2-p1) > limit || abs(q2-q1) > limit + + movq mm1, mm5 ; p1 + + movq mm7, mm3 ; mm3=mm7=p0 + psubusb mm7, mm5 ; p0 - p1 + + psubusb mm5, mm3 ; p1 - p0 + por mm5, mm7 ; abs(p1-p0) + + movq t0, mm5 ; save abs(p1-p0) + lea rdx, srct + + psubusb mm5, mm4 + por mm0, mm5 ; mm0=mask + + movq mm5, [rdx+16] ; mm5=q0 + movq mm7, [rdx+24] ; mm7=q1 + + movq mm6, mm5 ; mm6=q0 + movq mm2, mm7 ; q1 + psubusb mm5, mm7 ; q0-q1 + + psubusb mm7, mm6 ; q1-q0 + por mm7, mm5 ; abs(q1-q0) + + movq t1, mm7 ; save abs(q1-q0) + psubusb mm7, mm4 + + por mm0, mm7 ; mask + + movq mm5, mm2 ; q1 + psubusb mm5, mm1 ; q1-=p1 + psubusb mm1, mm2 ; p1-=q1 + por mm5, mm1 ; abs(p1-q1) + pand mm5, [GLOBAL(tfe)] ; set lsb of each byte to zero + psrlw mm5, 1 ; abs(p1-q1)/2 + + mov rdx, arg(2) ;blimit ; + + movq mm4, [rdx] ;blimit + movq mm1, mm3 ; mm1=mm3=p0 + + movq mm7, mm6 ; mm7=mm6=q0 + psubusb mm1, mm7 ; p0-q0 + + psubusb mm7, mm3 ; q0-p0 + por mm1, mm7 ; abs(q0-p0) + paddusb mm1, mm1 ; abs(q0-p0)*2 + paddusb mm1, mm5 ; abs (p0 - q0) *2 + abs(p1-q1)/2 + + psubusb mm1, mm4 ; abs (p0 - q0) *2 + abs(p1-q1)/2 > blimit + por mm1, mm0; ; mask + + pxor mm0, mm0 + pcmpeqb mm1, mm0 + + ; calculate high edge variance + mov rdx, arg(4) ;thresh ; get thresh + movq mm7, [rdx] + ; + movq mm4, t0 ; get abs (q1 - q0) + psubusb mm4, mm7 + + movq mm3, t1 ; get abs (p1 - p0) + psubusb mm3, mm7 + + por mm4, mm3 ; abs(q1 - q0) > thresh || abs(p1 - p0) > thresh + pcmpeqb mm4, mm0 + + pcmpeqb mm0, mm0 + pxor mm4, mm0 + + + + ; start work on filters + lea rdx, srct + + movq mm2, [rdx] ; p1 + movq mm7, [rdx+24] ; q1 + + movq mm6, [rdx+8] ; p0 + movq mm0, [rdx+16] ; q0 + + pxor mm2, [GLOBAL(t80)] ; p1 offset to convert to signed values + pxor mm7, [GLOBAL(t80)] ; q1 offset to convert to signed values + + psubsb mm2, mm7 ; p1 - q1 + pand mm2, mm4 ; high var mask (hvm)(p1 - q1) + + pxor mm6, [GLOBAL(t80)] ; offset to convert to signed values + pxor mm0, [GLOBAL(t80)] ; offset to convert to signed values + + movq mm3, mm0 ; q0 + psubsb mm0, mm6 ; q0 - p0 + + paddsb mm2, mm0 ; 1 * (q0 - p0) + hvm(p1 - q1) + paddsb mm2, mm0 ; 2 * (q0 - p0) + hvm(p1 - q1) + + paddsb mm2, mm0 ; 3 * (q0 - p0) + hvm(p1 - q1) + pand mm1, mm2 ; mask filter values we don't care about + + movq mm2, mm1 + paddsb mm1, [GLOBAL(t4)] ; 3* (q0 - p0) + hvm(p1 - q1) + 4 + + paddsb mm2, [GLOBAL(t3)] ; 3* (q0 - p0) + hvm(p1 - q1) + 3 + pxor mm0, mm0 ; + + pxor mm5, mm5 + punpcklbw mm0, mm2 ; + + punpckhbw mm5, mm2 ; + psraw mm0, 11 ; + + psraw mm5, 11 + packsswb mm0, mm5 + + movq mm2, mm0 ; (3* (q0 - p0) + hvm(p1 - q1) + 3) >> 3; + + pxor mm0, mm0 ; 0 + movq mm5, mm1 ; abcdefgh + + punpcklbw mm0, mm1 ; e0f0g0h0 + psraw mm0, 11 ; sign extended shift right by 3 + + pxor mm1, mm1 ; 0 + punpckhbw mm1, mm5 ; a0b0c0d0 + + psraw mm1, 11 ; sign extended shift right by 3 + movq mm5, mm0 ; save results + + packsswb mm0, mm1 ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>3 + paddsw mm5, [GLOBAL(ones)] + + paddsw mm1, [GLOBAL(ones)] + psraw mm5, 1 ; partial shifted one more time for 2nd tap + + psraw mm1, 1 ; partial shifted one more time for 2nd tap + packsswb mm5, mm1 ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>4 + + pandn mm4, mm5 ; high edge variance additive + + paddsb mm6, mm2 ; p0+= p0 add + pxor mm6, [GLOBAL(t80)] ; unoffset + + ; mm6=p0 ; + movq mm1, [rdx] ; p1 + pxor mm1, [GLOBAL(t80)] ; reoffset + + paddsb mm1, mm4 ; p1+= p1 add + pxor mm1, [GLOBAL(t80)] ; unoffset + ; mm6 = p0 mm1 = p1 + + psubsb mm3, mm0 ; q0-= q0 add + pxor mm3, [GLOBAL(t80)] ; unoffset + + ; mm3 = q0 + psubsb mm7, mm4 ; q1-= q1 add + pxor mm7, [GLOBAL(t80)] ; unoffset + ; mm7 = q1 + + ; transpose and write back + ; mm1 = 72 62 52 42 32 22 12 02 + ; mm6 = 73 63 53 43 33 23 13 03 + ; mm3 = 74 64 54 44 34 24 14 04 + ; mm7 = 75 65 55 45 35 25 15 05 + + movq mm2, mm1 ; 72 62 52 42 32 22 12 02 + punpcklbw mm2, mm6 ; 33 32 23 22 13 12 03 02 + + movq mm4, mm3 ; 74 64 54 44 34 24 14 04 + punpckhbw mm1, mm6 ; 73 72 63 62 53 52 43 42 + + punpcklbw mm4, mm7 ; 35 34 25 24 15 14 05 04 + punpckhbw mm3, mm7 ; 75 74 65 64 55 54 45 44 + + movq mm6, mm2 ; 33 32 23 22 13 12 03 02 + punpcklwd mm2, mm4 ; 15 14 13 12 05 04 03 02 + + punpckhwd mm6, mm4 ; 35 34 33 32 25 24 23 22 + movq mm5, mm1 ; 73 72 63 62 53 52 43 42 + + punpcklwd mm1, mm3 ; 55 54 53 52 45 44 43 42 + punpckhwd mm5, mm3 ; 75 74 73 72 65 64 63 62 + + + ; mm2 = 15 14 13 12 05 04 03 02 + ; mm6 = 35 34 33 32 25 24 23 22 + ; mm5 = 55 54 53 52 45 44 43 42 + ; mm1 = 75 74 73 72 65 64 63 62 + + + + movd [rsi+rax*4+2], mm2 + psrlq mm2, 32 + + movd [rdi+rax*4+2], mm2 + movd [rsi+rax*2+2], mm6 + + psrlq mm6, 32 + movd [rsi+rax+2],mm6 + + movd [rsi+2], mm1 + psrlq mm1, 32 + + movd [rdi+2], mm1 + neg rax + + movd [rdi+rax+2],mm5 + psrlq mm5, 32 + + movd [rdi+rax*2+2], mm5 + + lea rsi, [rsi+rax*8] + dec rcx + jnz .next8_v + + add rsp, 64 + pop rsp + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + UNSHADOW_ARGS + pop rbp + ret + + +;void vp8_mbloop_filter_horizontal_edge_mmx +;( +; unsigned char *src_ptr, +; int src_pixel_step, +; const char *blimit, +; const char *limit, +; const char *thresh, +; int count +;) +global sym(vp8_mbloop_filter_horizontal_edge_mmx) PRIVATE +sym(vp8_mbloop_filter_horizontal_edge_mmx): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, 32 ; reserve 32 bytes + %define t0 [rsp + 0] ;__declspec(align(16)) char t0[8]; + %define t1 [rsp + 16] ;__declspec(align(16)) char t1[8]; + + mov rsi, arg(0) ;src_ptr + movsxd rax, dword ptr arg(1) ;src_pixel_step ; destination pitch? + + movsxd rcx, dword ptr arg(5) ;count +.next8_mbh: + mov rdx, arg(3) ;limit + movq mm7, [rdx] + mov rdi, rsi ; rdi points to row +1 for indirect addressing + add rdi, rax + + ; calculate breakout conditions + movq mm2, [rdi+2*rax] ; q3 + + movq mm1, [rsi+2*rax] ; q2 + movq mm6, mm1 ; q2 + psubusb mm1, mm2 ; q2-=q3 + psubusb mm2, mm6 ; q3-=q2 + por mm1, mm2 ; abs(q3-q2) + psubusb mm1, mm7 + + + ; mm1 = abs(q3-q2), mm6 =q2, mm7 = limit + movq mm4, [rsi+rax] ; q1 + movq mm3, mm4 ; q1 + psubusb mm4, mm6 ; q1-=q2 + psubusb mm6, mm3 ; q2-=q1 + por mm4, mm6 ; abs(q2-q1) + psubusb mm4, mm7 + por mm1, mm4 + + + ; mm1 = mask, mm3=q1, mm7 = limit + + movq mm4, [rsi] ; q0 + movq mm0, mm4 ; q0 + psubusb mm4, mm3 ; q0-=q1 + psubusb mm3, mm0 ; q1-=q0 + por mm4, mm3 ; abs(q0-q1) + movq t0, mm4 ; save to t0 + psubusb mm4, mm7 + por mm1, mm4 + + + ; mm1 = mask, mm0=q0, mm7 = limit, t0 = abs(q0-q1) + + neg rax ; negate pitch to deal with above border + + movq mm2, [rsi+4*rax] ; p3 + movq mm4, [rdi+4*rax] ; p2 + movq mm5, mm4 ; p2 + psubusb mm4, mm2 ; p2-=p3 + psubusb mm2, mm5 ; p3-=p2 + por mm4, mm2 ; abs(p3 - p2) + psubusb mm4, mm7 + por mm1, mm4 + ; mm1 = mask, mm0=q0, mm7 = limit, t0 = abs(q0-q1) + + movq mm4, [rsi+2*rax] ; p1 + movq mm3, mm4 ; p1 + psubusb mm4, mm5 ; p1-=p2 + psubusb mm5, mm3 ; p2-=p1 + por mm4, mm5 ; abs(p2 - p1) + psubusb mm4, mm7 + por mm1, mm4 + + movq mm2, mm3 ; p1 + + + ; mm1 = mask, mm0=q0, mm7 = limit, t0 = abs(q0-q1) + + movq mm4, [rsi+rax] ; p0 + movq mm5, mm4 ; p0 + psubusb mm4, mm3 ; p0-=p1 + psubusb mm3, mm5 ; p1-=p0 + por mm4, mm3 ; abs(p1 - p0) + movq t1, mm4 ; save to t1 + psubusb mm4, mm7 + por mm1, mm4 + ; mm1 = mask, mm0=q0, mm7 = limit, t0 = abs(q0-q1) t1 = abs(p1-p0) + ; mm5 = p0 + movq mm3, [rdi] ; q1 + movq mm4, mm3 ; q1 + psubusb mm3, mm2 ; q1-=p1 + psubusb mm2, mm4 ; p1-=q1 + por mm2, mm3 ; abs(p1-q1) + pand mm2, [GLOBAL(tfe)] ; set lsb of each byte to zero + psrlw mm2, 1 ; abs(p1-q1)/2 + + movq mm6, mm5 ; p0 + movq mm3, mm0 ; q0 + psubusb mm5, mm3 ; p0-=q0 + psubusb mm3, mm6 ; q0-=p0 + por mm5, mm3 ; abs(p0 - q0) + paddusb mm5, mm5 ; abs(p0-q0)*2 + paddusb mm5, mm2 ; abs (p0 - q0) *2 + abs(p1-q1)/2 + + mov rdx, arg(2) ;blimit ; get blimit + movq mm7, [rdx] ; blimit + + psubusb mm5, mm7 ; abs (p0 - q0) *2 + abs(p1-q1)/2 > blimit + por mm1, mm5 + pxor mm5, mm5 + pcmpeqb mm1, mm5 ; mask mm1 + + ; mm1 = mask, mm0=q0, mm7 = blimit, t0 = abs(q0-q1) t1 = abs(p1-p0) + ; mm6 = p0, + + ; calculate high edge variance + mov rdx, arg(4) ;thresh ; get thresh + movq mm7, [rdx] ; + movq mm4, t0 ; get abs (q1 - q0) + psubusb mm4, mm7 + movq mm3, t1 ; get abs (p1 - p0) + psubusb mm3, mm7 + paddb mm4, mm3 ; abs(q1 - q0) > thresh || abs(p1 - p0) > thresh + + pcmpeqb mm4, mm5 + + pcmpeqb mm5, mm5 + pxor mm4, mm5 + + + + ; mm1 = mask, mm0=q0, mm7 = thresh, t0 = abs(q0-q1) t1 = abs(p1-p0) + ; mm6 = p0, mm4=hev + ; start work on filters + movq mm2, [rsi+2*rax] ; p1 + movq mm7, [rdi] ; q1 + pxor mm2, [GLOBAL(t80)] ; p1 offset to convert to signed values + pxor mm7, [GLOBAL(t80)] ; q1 offset to convert to signed values + psubsb mm2, mm7 ; p1 - q1 + + pxor mm6, [GLOBAL(t80)] ; offset to convert to signed values + pxor mm0, [GLOBAL(t80)] ; offset to convert to signed values + movq mm3, mm0 ; q0 + psubsb mm0, mm6 ; q0 - p0 + paddsb mm2, mm0 ; 1 * (q0 - p0) + (p1 - q1) + paddsb mm2, mm0 ; 2 * (q0 - p0) + paddsb mm2, mm0 ; 3 * (q0 - p0) + (p1 - q1) + pand mm1, mm2 ; mask filter values we don't care about + + + ; mm1 = vp8_filter, mm4=hev, mm6=ps0, mm3=qs0 + movq mm2, mm1 ; vp8_filter + pand mm2, mm4; ; Filter2 = vp8_filter & hev + + movq mm5, mm2 ; + paddsb mm5, [GLOBAL(t3)]; + + pxor mm0, mm0 ; 0 + pxor mm7, mm7 ; 0 + + punpcklbw mm0, mm5 ; e0f0g0h0 + psraw mm0, 11 ; sign extended shift right by 3 + punpckhbw mm7, mm5 ; a0b0c0d0 + psraw mm7, 11 ; sign extended shift right by 3 + packsswb mm0, mm7 ; Filter2 >>=3; + + movq mm5, mm0 ; Filter2 + + paddsb mm2, [GLOBAL(t4)] ; vp8_signed_char_clamp(Filter2 + 4) + pxor mm0, mm0 ; 0 + pxor mm7, mm7 ; 0 + + punpcklbw mm0, mm2 ; e0f0g0h0 + psraw mm0, 11 ; sign extended shift right by 3 + punpckhbw mm7, mm2 ; a0b0c0d0 + psraw mm7, 11 ; sign extended shift right by 3 + packsswb mm0, mm7 ; Filter2 >>=3; + + ; mm0= filter2 mm1 = vp8_filter, mm3 =qs0 mm5=s mm4 =hev mm6=ps0 + psubsb mm3, mm0 ; qs0 =qs0 - filter1 + paddsb mm6, mm5 ; ps0 =ps0 + Fitler2 + + ; mm1=vp8_filter, mm3=qs0, mm4 =hev mm6=ps0 + ; vp8_filter &= ~hev; + ; Filter2 = vp8_filter; + pandn mm4, mm1 ; vp8_filter&=~hev + + + ; mm3=qs0, mm4=filter2, mm6=ps0 + + ; u = vp8_signed_char_clamp((63 + Filter2 * 27)>>7); + ; s = vp8_signed_char_clamp(qs0 - u); + ; *oq0 = s^0x80; + ; s = vp8_signed_char_clamp(ps0 + u); + ; *op0 = s^0x80; + pxor mm0, mm0 + + pxor mm1, mm1 + pxor mm2, mm2 + punpcklbw mm1, mm4 + punpckhbw mm2, mm4 + pmulhw mm1, [GLOBAL(s27)] + pmulhw mm2, [GLOBAL(s27)] + paddw mm1, [GLOBAL(s63)] + paddw mm2, [GLOBAL(s63)] + psraw mm1, 7 + psraw mm2, 7 + packsswb mm1, mm2 + + psubsb mm3, mm1 + paddsb mm6, mm1 + + pxor mm3, [GLOBAL(t80)] + pxor mm6, [GLOBAL(t80)] + movq [rsi+rax], mm6 + movq [rsi], mm3 + + ; roughly 2/7th difference across boundary + ; u = vp8_signed_char_clamp((63 + Filter2 * 18)>>7); + ; s = vp8_signed_char_clamp(qs1 - u); + ; *oq1 = s^0x80; + ; s = vp8_signed_char_clamp(ps1 + u); + ; *op1 = s^0x80; + pxor mm1, mm1 + pxor mm2, mm2 + punpcklbw mm1, mm4 + punpckhbw mm2, mm4 + pmulhw mm1, [GLOBAL(s18)] + pmulhw mm2, [GLOBAL(s18)] + paddw mm1, [GLOBAL(s63)] + paddw mm2, [GLOBAL(s63)] + psraw mm1, 7 + psraw mm2, 7 + packsswb mm1, mm2 + + movq mm3, [rdi] + movq mm6, [rsi+rax*2] ; p1 + + pxor mm3, [GLOBAL(t80)] + pxor mm6, [GLOBAL(t80)] + + paddsb mm6, mm1 + psubsb mm3, mm1 + + pxor mm6, [GLOBAL(t80)] + pxor mm3, [GLOBAL(t80)] + movq [rdi], mm3 + movq [rsi+rax*2], mm6 + + ; roughly 1/7th difference across boundary + ; u = vp8_signed_char_clamp((63 + Filter2 * 9)>>7); + ; s = vp8_signed_char_clamp(qs2 - u); + ; *oq2 = s^0x80; + ; s = vp8_signed_char_clamp(ps2 + u); + ; *op2 = s^0x80; + pxor mm1, mm1 + pxor mm2, mm2 + punpcklbw mm1, mm4 + punpckhbw mm2, mm4 + pmulhw mm1, [GLOBAL(s9)] + pmulhw mm2, [GLOBAL(s9)] + paddw mm1, [GLOBAL(s63)] + paddw mm2, [GLOBAL(s63)] + psraw mm1, 7 + psraw mm2, 7 + packsswb mm1, mm2 + + + movq mm6, [rdi+rax*4] + neg rax + movq mm3, [rdi+rax ] + + pxor mm6, [GLOBAL(t80)] + pxor mm3, [GLOBAL(t80)] + + paddsb mm6, mm1 + psubsb mm3, mm1 + + pxor mm6, [GLOBAL(t80)] + pxor mm3, [GLOBAL(t80)] + movq [rdi+rax ], mm3 + neg rax + movq [rdi+rax*4], mm6 + +;EARLY_BREAK_OUT: + neg rax + add rsi,8 + dec rcx + jnz .next8_mbh + + add rsp, 32 + pop rsp + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + UNSHADOW_ARGS + pop rbp + ret + + +;void vp8_mbloop_filter_vertical_edge_mmx +;( +; unsigned char *src_ptr, +; int src_pixel_step, +; const char *blimit, +; const char *limit, +; const char *thresh, +; int count +;) +global sym(vp8_mbloop_filter_vertical_edge_mmx) PRIVATE +sym(vp8_mbloop_filter_vertical_edge_mmx): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, 96 ; reserve 96 bytes + %define t0 [rsp + 0] ;__declspec(align(16)) char t0[8]; + %define t1 [rsp + 16] ;__declspec(align(16)) char t1[8]; + %define srct [rsp + 32] ;__declspec(align(16)) char srct[64]; + + mov rsi, arg(0) ;src_ptr + movsxd rax, dword ptr arg(1) ;src_pixel_step ; destination pitch? + + lea rsi, [rsi + rax*4 - 4] + + movsxd rcx, dword ptr arg(5) ;count +.next8_mbv: + lea rdi, [rsi + rax] ; rdi points to row +1 for indirect addressing + + ;transpose + movq mm0, [rdi+2*rax] ; 77 76 75 74 73 72 71 70 + movq mm6, [rsi+2*rax] ; 67 66 65 64 63 62 61 60 + + movq mm7, mm6 ; 77 76 75 74 73 72 71 70 + punpckhbw mm7, mm0 ; 77 67 76 66 75 65 74 64 + + punpcklbw mm6, mm0 ; 73 63 72 62 71 61 70 60 + movq mm0, [rsi+rax] ; 57 56 55 54 53 52 51 50 + + movq mm4, [rsi] ; 47 46 45 44 43 42 41 40 + movq mm5, mm4 ; 47 46 45 44 43 42 41 40 + + punpckhbw mm5, mm0 ; 57 47 56 46 55 45 54 44 + punpcklbw mm4, mm0 ; 53 43 52 42 51 41 50 40 + + movq mm3, mm5 ; 57 47 56 46 55 45 54 44 + punpckhwd mm5, mm7 ; 77 67 57 47 76 66 56 46 + + punpcklwd mm3, mm7 ; 75 65 55 45 74 64 54 44 + movq mm2, mm4 ; 53 43 52 42 51 41 50 40 + + punpckhwd mm4, mm6 ; 73 63 53 43 72 62 52 42 + punpcklwd mm2, mm6 ; 71 61 51 41 70 60 50 40 + + neg rax + + movq mm7, [rsi+rax] ; 37 36 35 34 33 32 31 30 + movq mm6, [rsi+rax*2] ; 27 26 25 24 23 22 21 20 + + movq mm1, mm6 ; 27 26 25 24 23 22 21 20 + punpckhbw mm6, mm7 ; 37 27 36 36 35 25 34 24 + + punpcklbw mm1, mm7 ; 33 23 32 22 31 21 30 20 + + movq mm7, [rsi+rax*4]; ; 07 06 05 04 03 02 01 00 + punpckhbw mm7, [rdi+rax*4] ; 17 07 16 06 15 05 14 04 + + movq mm0, mm7 ; 17 07 16 06 15 05 14 04 + punpckhwd mm7, mm6 ; 37 27 17 07 36 26 16 06 + + punpcklwd mm0, mm6 ; 35 25 15 05 34 24 14 04 + movq mm6, mm7 ; 37 27 17 07 36 26 16 06 + + punpckhdq mm7, mm5 ; 77 67 57 47 37 27 17 07 = q3 + punpckldq mm6, mm5 ; 76 66 56 46 36 26 16 06 = q2 + + lea rdx, srct + movq mm5, mm6 ; 76 66 56 46 36 26 16 06 + + movq [rdx+56], mm7 + psubusb mm5, mm7 ; q2-q3 + + + movq [rdx+48], mm6 + psubusb mm7, mm6 ; q3-q2 + + por mm7, mm5; ; mm7=abs (q3-q2) + movq mm5, mm0 ; 35 25 15 05 34 24 14 04 + + punpckhdq mm5, mm3 ; 75 65 55 45 35 25 15 05 = q1 + punpckldq mm0, mm3 ; 74 64 54 44 34 24 15 04 = q0 + + movq mm3, mm5 ; 75 65 55 45 35 25 15 05 = q1 + psubusb mm3, mm6 ; q1-q2 + + psubusb mm6, mm5 ; q2-q1 + por mm6, mm3 ; mm6=abs(q2-q1) + + movq [rdx+40], mm5 ; save q1 + movq [rdx+32], mm0 ; save q0 + + movq mm3, [rsi+rax*4] ; 07 06 05 04 03 02 01 00 + punpcklbw mm3, [rdi+rax*4] ; 13 03 12 02 11 01 10 00 + + movq mm0, mm3 ; 13 03 12 02 11 01 10 00 + punpcklwd mm0, mm1 ; 31 21 11 01 30 20 10 00 + + punpckhwd mm3, mm1 ; 33 23 13 03 32 22 12 02 + movq mm1, mm0 ; 31 21 11 01 30 20 10 00 + + punpckldq mm0, mm2 ; 70 60 50 40 30 20 10 00 =p3 + punpckhdq mm1, mm2 ; 71 61 51 41 31 21 11 01 =p2 + + movq [rdx], mm0 ; save p3 + movq [rdx+8], mm1 ; save p2 + + movq mm2, mm1 ; 71 61 51 41 31 21 11 01 =p2 + psubusb mm2, mm0 ; p2-p3 + + psubusb mm0, mm1 ; p3-p2 + por mm0, mm2 ; mm0=abs(p3-p2) + + movq mm2, mm3 ; 33 23 13 03 32 22 12 02 + punpckldq mm2, mm4 ; 72 62 52 42 32 22 12 02 = p1 + + punpckhdq mm3, mm4 ; 73 63 53 43 33 23 13 03 = p0 + movq [rdx+24], mm3 ; save p0 + + movq [rdx+16], mm2 ; save p1 + movq mm5, mm2 ; mm5 = p1 + + psubusb mm2, mm1 ; p1-p2 + psubusb mm1, mm5 ; p2-p1 + + por mm1, mm2 ; mm1=abs(p2-p1) + mov rdx, arg(3) ;limit + + movq mm4, [rdx] ; mm4 = limit + psubusb mm7, mm4 ; abs(q3-q2) > limit + + psubusb mm0, mm4 ; abs(p3-p2) > limit + psubusb mm1, mm4 ; abs(p2-p1) > limit + + psubusb mm6, mm4 ; abs(q2-q1) > limit + por mm7, mm6 ; or + + por mm0, mm1 ; + por mm0, mm7 ; abs(q3-q2) > limit || abs(p3-p2) > limit ||abs(p2-p1) > limit || abs(q2-q1) > limit + + movq mm1, mm5 ; p1 + + movq mm7, mm3 ; mm3=mm7=p0 + psubusb mm7, mm5 ; p0 - p1 + + psubusb mm5, mm3 ; p1 - p0 + por mm5, mm7 ; abs(p1-p0) + + movq t0, mm5 ; save abs(p1-p0) + lea rdx, srct + + psubusb mm5, mm4 ; mm5 = abs(p1-p0) > limit + por mm0, mm5 ; mm0=mask + + movq mm5, [rdx+32] ; mm5=q0 + movq mm7, [rdx+40] ; mm7=q1 + + movq mm6, mm5 ; mm6=q0 + movq mm2, mm7 ; q1 + psubusb mm5, mm7 ; q0-q1 + + psubusb mm7, mm6 ; q1-q0 + por mm7, mm5 ; abs(q1-q0) + + movq t1, mm7 ; save abs(q1-q0) + psubusb mm7, mm4 ; mm7=abs(q1-q0)> limit + + por mm0, mm7 ; mask + + movq mm5, mm2 ; q1 + psubusb mm5, mm1 ; q1-=p1 + psubusb mm1, mm2 ; p1-=q1 + por mm5, mm1 ; abs(p1-q1) + pand mm5, [GLOBAL(tfe)] ; set lsb of each byte to zero + psrlw mm5, 1 ; abs(p1-q1)/2 + + mov rdx, arg(2) ;blimit ; + + movq mm4, [rdx] ;blimit + movq mm1, mm3 ; mm1=mm3=p0 + + movq mm7, mm6 ; mm7=mm6=q0 + psubusb mm1, mm7 ; p0-q0 + + psubusb mm7, mm3 ; q0-p0 + por mm1, mm7 ; abs(q0-p0) + paddusb mm1, mm1 ; abs(q0-p0)*2 + paddusb mm1, mm5 ; abs (p0 - q0) *2 + abs(p1-q1)/2 + + psubusb mm1, mm4 ; abs (p0 - q0) *2 + abs(p1-q1)/2 > blimit + por mm1, mm0; ; mask + + pxor mm0, mm0 + pcmpeqb mm1, mm0 + + ; calculate high edge variance + mov rdx, arg(4) ;thresh ; get thresh + movq mm7, [rdx] + ; + movq mm4, t0 ; get abs (q1 - q0) + psubusb mm4, mm7 ; abs(q1 - q0) > thresh + + movq mm3, t1 ; get abs (p1 - p0) + psubusb mm3, mm7 ; abs(p1 - p0)> thresh + + por mm4, mm3 ; abs(q1 - q0) > thresh || abs(p1 - p0) > thresh + pcmpeqb mm4, mm0 + + pcmpeqb mm0, mm0 + pxor mm4, mm0 + + + + + ; start work on filters + lea rdx, srct + + ; start work on filters + movq mm2, [rdx+16] ; p1 + movq mm7, [rdx+40] ; q1 + pxor mm2, [GLOBAL(t80)] ; p1 offset to convert to signed values + pxor mm7, [GLOBAL(t80)] ; q1 offset to convert to signed values + psubsb mm2, mm7 ; p1 - q1 + + movq mm6, [rdx+24] ; p0 + movq mm0, [rdx+32] ; q0 + pxor mm6, [GLOBAL(t80)] ; offset to convert to signed values + pxor mm0, [GLOBAL(t80)] ; offset to convert to signed values + + movq mm3, mm0 ; q0 + psubsb mm0, mm6 ; q0 - p0 + paddsb mm2, mm0 ; 1 * (q0 - p0) + (p1 - q1) + paddsb mm2, mm0 ; 2 * (q0 - p0) + paddsb mm2, mm0 ; 3 * (q0 - p0) + (p1 - q1) + pand mm1, mm2 ; mask filter values we don't care about + + ; mm1 = vp8_filter, mm4=hev, mm6=ps0, mm3=qs0 + movq mm2, mm1 ; vp8_filter + pand mm2, mm4; ; Filter2 = vp8_filter & hev + + movq mm5, mm2 ; + paddsb mm5, [GLOBAL(t3)]; + + pxor mm0, mm0 ; 0 + pxor mm7, mm7 ; 0 + + punpcklbw mm0, mm5 ; e0f0g0h0 + psraw mm0, 11 ; sign extended shift right by 3 + punpckhbw mm7, mm5 ; a0b0c0d0 + psraw mm7, 11 ; sign extended shift right by 3 + packsswb mm0, mm7 ; Filter2 >>=3; + + movq mm5, mm0 ; Filter2 + + paddsb mm2, [GLOBAL(t4)] ; vp8_signed_char_clamp(Filter2 + 4) + pxor mm0, mm0 ; 0 + pxor mm7, mm7 ; 0 + + punpcklbw mm0, mm2 ; e0f0g0h0 + psraw mm0, 11 ; sign extended shift right by 3 + punpckhbw mm7, mm2 ; a0b0c0d0 + psraw mm7, 11 ; sign extended shift right by 3 + packsswb mm0, mm7 ; Filter2 >>=3; + + ; mm0= filter2 mm1 = vp8_filter, mm3 =qs0 mm5=s mm4 =hev mm6=ps0 + psubsb mm3, mm0 ; qs0 =qs0 - filter1 + paddsb mm6, mm5 ; ps0 =ps0 + Fitler2 + + ; mm1=vp8_filter, mm3=qs0, mm4 =hev mm6=ps0 + ; vp8_filter &= ~hev; + ; Filter2 = vp8_filter; + pandn mm4, mm1 ; vp8_filter&=~hev + + + ; mm3=qs0, mm4=filter2, mm6=ps0 + + ; u = vp8_signed_char_clamp((63 + Filter2 * 27)>>7); + ; s = vp8_signed_char_clamp(qs0 - u); + ; *oq0 = s^0x80; + ; s = vp8_signed_char_clamp(ps0 + u); + ; *op0 = s^0x80; + pxor mm0, mm0 + + pxor mm1, mm1 + pxor mm2, mm2 + punpcklbw mm1, mm4 + punpckhbw mm2, mm4 + pmulhw mm1, [GLOBAL(s27)] + pmulhw mm2, [GLOBAL(s27)] + paddw mm1, [GLOBAL(s63)] + paddw mm2, [GLOBAL(s63)] + psraw mm1, 7 + psraw mm2, 7 + packsswb mm1, mm2 + + psubsb mm3, mm1 + paddsb mm6, mm1 + + pxor mm3, [GLOBAL(t80)] + pxor mm6, [GLOBAL(t80)] + movq [rdx+24], mm6 + movq [rdx+32], mm3 + + ; roughly 2/7th difference across boundary + ; u = vp8_signed_char_clamp((63 + Filter2 * 18)>>7); + ; s = vp8_signed_char_clamp(qs1 - u); + ; *oq1 = s^0x80; + ; s = vp8_signed_char_clamp(ps1 + u); + ; *op1 = s^0x80; + pxor mm1, mm1 + pxor mm2, mm2 + punpcklbw mm1, mm4 + punpckhbw mm2, mm4 + pmulhw mm1, [GLOBAL(s18)] + pmulhw mm2, [GLOBAL(s18)] + paddw mm1, [GLOBAL(s63)] + paddw mm2, [GLOBAL(s63)] + psraw mm1, 7 + psraw mm2, 7 + packsswb mm1, mm2 + + movq mm3, [rdx + 40] + movq mm6, [rdx + 16] ; p1 + pxor mm3, [GLOBAL(t80)] + pxor mm6, [GLOBAL(t80)] + + paddsb mm6, mm1 + psubsb mm3, mm1 + + pxor mm6, [GLOBAL(t80)] + pxor mm3, [GLOBAL(t80)] + movq [rdx + 40], mm3 + movq [rdx + 16], mm6 + + ; roughly 1/7th difference across boundary + ; u = vp8_signed_char_clamp((63 + Filter2 * 9)>>7); + ; s = vp8_signed_char_clamp(qs2 - u); + ; *oq2 = s^0x80; + ; s = vp8_signed_char_clamp(ps2 + u); + ; *op2 = s^0x80; + pxor mm1, mm1 + pxor mm2, mm2 + punpcklbw mm1, mm4 + punpckhbw mm2, mm4 + pmulhw mm1, [GLOBAL(s9)] + pmulhw mm2, [GLOBAL(s9)] + paddw mm1, [GLOBAL(s63)] + paddw mm2, [GLOBAL(s63)] + psraw mm1, 7 + psraw mm2, 7 + packsswb mm1, mm2 + + movq mm6, [rdx+ 8] + movq mm3, [rdx+48] + + pxor mm6, [GLOBAL(t80)] + pxor mm3, [GLOBAL(t80)] + + paddsb mm6, mm1 + psubsb mm3, mm1 + + pxor mm6, [GLOBAL(t80)] ; mm6 = 71 61 51 41 31 21 11 01 + pxor mm3, [GLOBAL(t80)] ; mm3 = 76 66 56 46 36 26 15 06 + + ; transpose and write back + movq mm0, [rdx] ; mm0 = 70 60 50 40 30 20 10 00 + movq mm1, mm0 ; mm0 = 70 60 50 40 30 20 10 00 + + punpcklbw mm0, mm6 ; mm0 = 31 30 21 20 11 10 01 00 + punpckhbw mm1, mm6 ; mm3 = 71 70 61 60 51 50 41 40 + + movq mm2, [rdx+16] ; mm2 = 72 62 52 42 32 22 12 02 + movq mm6, mm2 ; mm3 = 72 62 52 42 32 22 12 02 + + punpcklbw mm2, [rdx+24] ; mm2 = 33 32 23 22 13 12 03 02 + punpckhbw mm6, [rdx+24] ; mm3 = 73 72 63 62 53 52 43 42 + + movq mm5, mm0 ; mm5 = 31 30 21 20 11 10 01 00 + punpcklwd mm0, mm2 ; mm0 = 13 12 11 10 03 02 01 00 + + punpckhwd mm5, mm2 ; mm5 = 33 32 31 30 23 22 21 20 + movq mm4, mm1 ; mm4 = 71 70 61 60 51 50 41 40 + + punpcklwd mm1, mm6 ; mm1 = 53 52 51 50 43 42 41 40 + punpckhwd mm4, mm6 ; mm4 = 73 72 71 70 63 62 61 60 + + movq mm2, [rdx+32] ; mm2 = 74 64 54 44 34 24 14 04 + punpcklbw mm2, [rdx+40] ; mm2 = 35 34 25 24 15 14 05 04 + + movq mm6, mm3 ; mm6 = 76 66 56 46 36 26 15 06 + punpcklbw mm6, [rdx+56] ; mm6 = 37 36 27 26 17 16 07 06 + + movq mm7, mm2 ; mm7 = 35 34 25 24 15 14 05 04 + punpcklwd mm2, mm6 ; mm2 = 17 16 15 14 07 06 05 04 + + punpckhwd mm7, mm6 ; mm7 = 37 36 35 34 27 26 25 24 + movq mm6, mm0 ; mm6 = 13 12 11 10 03 02 01 00 + + punpckldq mm0, mm2 ; mm0 = 07 06 05 04 03 02 01 00 + punpckhdq mm6, mm2 ; mm6 = 17 16 15 14 13 12 11 10 + + movq [rsi+rax*4], mm0 ; write out + movq [rdi+rax*4], mm6 ; write out + + movq mm0, mm5 ; mm0 = 33 32 31 30 23 22 21 20 + punpckldq mm0, mm7 ; mm0 = 27 26 25 24 23 22 20 20 + + punpckhdq mm5, mm7 ; mm5 = 37 36 35 34 33 32 31 30 + movq [rsi+rax*2], mm0 ; write out + + movq [rdi+rax*2], mm5 ; write out + movq mm2, [rdx+32] ; mm2 = 74 64 54 44 34 24 14 04 + + punpckhbw mm2, [rdx+40] ; mm2 = 75 74 65 64 54 54 45 44 + punpckhbw mm3, [rdx+56] ; mm3 = 77 76 67 66 57 56 47 46 + + movq mm5, mm2 ; mm5 = 75 74 65 64 54 54 45 44 + punpcklwd mm2, mm3 ; mm2 = 57 56 55 54 47 46 45 44 + + punpckhwd mm5, mm3 ; mm5 = 77 76 75 74 67 66 65 64 + movq mm0, mm1 ; mm0= 53 52 51 50 43 42 41 40 + + movq mm3, mm4 ; mm4 = 73 72 71 70 63 62 61 60 + punpckldq mm0, mm2 ; mm0 = 47 46 45 44 43 42 41 40 + + punpckhdq mm1, mm2 ; mm1 = 57 56 55 54 53 52 51 50 + movq [rsi], mm0 ; write out + + movq [rdi], mm1 ; write out + neg rax + + punpckldq mm3, mm5 ; mm3 = 67 66 65 64 63 62 61 60 + punpckhdq mm4, mm5 ; mm4 = 77 76 75 74 73 72 71 60 + + movq [rsi+rax*2], mm3 + movq [rdi+rax*2], mm4 + + lea rsi, [rsi+rax*8] + dec rcx + + jnz .next8_mbv + + add rsp, 96 + pop rsp + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + UNSHADOW_ARGS + pop rbp + ret + + +;void vp8_loop_filter_simple_horizontal_edge_mmx +;( +; unsigned char *src_ptr, +; int src_pixel_step, +; const char *blimit +;) +global sym(vp8_loop_filter_simple_horizontal_edge_mmx) PRIVATE +sym(vp8_loop_filter_simple_horizontal_edge_mmx): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 3 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + mov rsi, arg(0) ;src_ptr + movsxd rax, dword ptr arg(1) ;src_pixel_step ; destination pitch? + + mov rcx, 2 ; count +.nexts8_h: + mov rdx, arg(2) ;blimit ; get blimit + movq mm3, [rdx] ; + + mov rdi, rsi ; rdi points to row +1 for indirect addressing + add rdi, rax + neg rax + + ; calculate mask + movq mm1, [rsi+2*rax] ; p1 + movq mm0, [rdi] ; q1 + movq mm2, mm1 + movq mm7, mm0 + movq mm4, mm0 + psubusb mm0, mm1 ; q1-=p1 + psubusb mm1, mm4 ; p1-=q1 + por mm1, mm0 ; abs(p1-q1) + pand mm1, [GLOBAL(tfe)] ; set lsb of each byte to zero + psrlw mm1, 1 ; abs(p1-q1)/2 + + movq mm5, [rsi+rax] ; p0 + movq mm4, [rsi] ; q0 + movq mm0, mm4 ; q0 + movq mm6, mm5 ; p0 + psubusb mm5, mm4 ; p0-=q0 + psubusb mm4, mm6 ; q0-=p0 + por mm5, mm4 ; abs(p0 - q0) + paddusb mm5, mm5 ; abs(p0-q0)*2 + paddusb mm5, mm1 ; abs (p0 - q0) *2 + abs(p1-q1)/2 + + psubusb mm5, mm3 ; abs(p0 - q0) *2 + abs(p1-q1)/2 > blimit + pxor mm3, mm3 + pcmpeqb mm5, mm3 + + ; start work on filters + pxor mm2, [GLOBAL(t80)] ; p1 offset to convert to signed values + pxor mm7, [GLOBAL(t80)] ; q1 offset to convert to signed values + psubsb mm2, mm7 ; p1 - q1 + + pxor mm6, [GLOBAL(t80)] ; offset to convert to signed values + pxor mm0, [GLOBAL(t80)] ; offset to convert to signed values + movq mm3, mm0 ; q0 + psubsb mm0, mm6 ; q0 - p0 + paddsb mm2, mm0 ; p1 - q1 + 1 * (q0 - p0) + paddsb mm2, mm0 ; p1 - q1 + 2 * (q0 - p0) + paddsb mm2, mm0 ; p1 - q1 + 3 * (q0 - p0) + pand mm5, mm2 ; mask filter values we don't care about + + ; do + 4 side + paddsb mm5, [GLOBAL(t4)] ; 3* (q0 - p0) + (p1 - q1) + 4 + + movq mm0, mm5 ; get a copy of filters + psllw mm0, 8 ; shift left 8 + psraw mm0, 3 ; arithmetic shift right 11 + psrlw mm0, 8 + movq mm1, mm5 ; get a copy of filters + psraw mm1, 11 ; arithmetic shift right 11 + psllw mm1, 8 ; shift left 8 to put it back + + por mm0, mm1 ; put the two together to get result + + psubsb mm3, mm0 ; q0-= q0 add + pxor mm3, [GLOBAL(t80)] ; unoffset + movq [rsi], mm3 ; write back + + + ; now do +3 side + psubsb mm5, [GLOBAL(t1s)] ; +3 instead of +4 + + movq mm0, mm5 ; get a copy of filters + psllw mm0, 8 ; shift left 8 + psraw mm0, 3 ; arithmetic shift right 11 + psrlw mm0, 8 + psraw mm5, 11 ; arithmetic shift right 11 + psllw mm5, 8 ; shift left 8 to put it back + por mm0, mm5 ; put the two together to get result + + + paddsb mm6, mm0 ; p0+= p0 add + pxor mm6, [GLOBAL(t80)] ; unoffset + movq [rsi+rax], mm6 ; write back + + add rsi,8 + neg rax + dec rcx + jnz .nexts8_h + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + UNSHADOW_ARGS + pop rbp + ret + + +;void vp8_loop_filter_simple_vertical_edge_mmx +;( +; unsigned char *src_ptr, +; int src_pixel_step, +; const char *blimit +;) +global sym(vp8_loop_filter_simple_vertical_edge_mmx) PRIVATE +sym(vp8_loop_filter_simple_vertical_edge_mmx): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 3 + GET_GOT rbx + push rsi + push rdi + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, 32 ; reserve 32 bytes + %define t0 [rsp + 0] ;__declspec(align(16)) char t0[8]; + %define t1 [rsp + 16] ;__declspec(align(16)) char t1[8]; + + mov rsi, arg(0) ;src_ptr + movsxd rax, dword ptr arg(1) ;src_pixel_step ; destination pitch? + + lea rsi, [rsi + rax*4- 2]; ; + mov rcx, 2 ; count +.nexts8_v: + + lea rdi, [rsi + rax]; + movd mm0, [rdi + rax * 2] ; xx xx xx xx 73 72 71 70 + + movd mm6, [rsi + rax * 2] ; xx xx xx xx 63 62 61 60 + punpcklbw mm6, mm0 ; 73 63 72 62 71 61 70 60 + + movd mm0, [rsi + rax] ; xx xx xx xx 53 52 51 50 + movd mm4, [rsi] ; xx xx xx xx 43 42 41 40 + + punpcklbw mm4, mm0 ; 53 43 52 42 51 41 50 40 + movq mm5, mm4 ; 53 43 52 42 51 41 50 40 + + punpcklwd mm4, mm6 ; 71 61 51 41 70 60 50 40 + punpckhwd mm5, mm6 ; 73 63 53 43 72 62 52 42 + + neg rax + + movd mm7, [rsi + rax] ; xx xx xx xx 33 32 31 30 + movd mm6, [rsi + rax * 2] ; xx xx xx xx 23 22 21 20 + + punpcklbw mm6, mm7 ; 33 23 32 22 31 21 30 20 + movd mm1, [rdi + rax * 4] ; xx xx xx xx 13 12 11 10 + + movd mm0, [rsi + rax * 4] ; xx xx xx xx 03 02 01 00 + punpcklbw mm0, mm1 ; 13 03 12 02 11 01 10 00 + + movq mm2, mm0 ; 13 03 12 02 11 01 10 00 + punpcklwd mm0, mm6 ; 31 21 11 01 30 20 10 00 + + punpckhwd mm2, mm6 ; 33 23 13 03 32 22 12 02 + movq mm1, mm0 ; 13 03 12 02 11 01 10 00 + + punpckldq mm0, mm4 ; 70 60 50 40 30 20 10 00 = p1 + movq mm3, mm2 ; 33 23 13 03 32 22 12 02 + + punpckhdq mm1, mm4 ; 71 61 51 41 31 21 11 01 = p0 + punpckldq mm2, mm5 ; 72 62 52 42 32 22 12 02 = q0 + + punpckhdq mm3, mm5 ; 73 63 53 43 33 23 13 03 = q1 + + + ; calculate mask + movq mm6, mm0 ; p1 + movq mm7, mm3 ; q1 + psubusb mm7, mm6 ; q1-=p1 + psubusb mm6, mm3 ; p1-=q1 + por mm6, mm7 ; abs(p1-q1) + pand mm6, [GLOBAL(tfe)] ; set lsb of each byte to zero + psrlw mm6, 1 ; abs(p1-q1)/2 + + movq mm5, mm1 ; p0 + movq mm4, mm2 ; q0 + + psubusb mm5, mm2 ; p0-=q0 + psubusb mm4, mm1 ; q0-=p0 + + por mm5, mm4 ; abs(p0 - q0) + paddusb mm5, mm5 ; abs(p0-q0)*2 + paddusb mm5, mm6 ; abs (p0 - q0) *2 + abs(p1-q1)/2 + + mov rdx, arg(2) ;blimit ; get blimit + movq mm7, [rdx] + + psubusb mm5, mm7 ; abs(p0 - q0) *2 + abs(p1-q1)/2 > blimit + pxor mm7, mm7 + pcmpeqb mm5, mm7 ; mm5 = mask + + ; start work on filters + movq t0, mm0 + movq t1, mm3 + + pxor mm0, [GLOBAL(t80)] ; p1 offset to convert to signed values + pxor mm3, [GLOBAL(t80)] ; q1 offset to convert to signed values + + psubsb mm0, mm3 ; p1 - q1 + movq mm6, mm1 ; p0 + + movq mm7, mm2 ; q0 + pxor mm6, [GLOBAL(t80)] ; offset to convert to signed values + + pxor mm7, [GLOBAL(t80)] ; offset to convert to signed values + movq mm3, mm7 ; offseted ; q0 + + psubsb mm7, mm6 ; q0 - p0 + paddsb mm0, mm7 ; p1 - q1 + 1 * (q0 - p0) + + paddsb mm0, mm7 ; p1 - q1 + 2 * (q0 - p0) + paddsb mm0, mm7 ; p1 - q1 + 3 * (q0 - p0) + + pand mm5, mm0 ; mask filter values we don't care about + + paddsb mm5, [GLOBAL(t4)] ; 3* (q0 - p0) + (p1 - q1) + 4 + + movq mm0, mm5 ; get a copy of filters + psllw mm0, 8 ; shift left 8 + psraw mm0, 3 ; arithmetic shift right 11 + psrlw mm0, 8 + + movq mm7, mm5 ; get a copy of filters + psraw mm7, 11 ; arithmetic shift right 11 + psllw mm7, 8 ; shift left 8 to put it back + + por mm0, mm7 ; put the two together to get result + + psubsb mm3, mm0 ; q0-= q0sz add + pxor mm3, [GLOBAL(t80)] ; unoffset + + ; now do +3 side + psubsb mm5, [GLOBAL(t1s)] ; +3 instead of +4 + + movq mm0, mm5 ; get a copy of filters + psllw mm0, 8 ; shift left 8 + psraw mm0, 3 ; arithmetic shift right 11 + psrlw mm0, 8 + + psraw mm5, 11 ; arithmetic shift right 11 + psllw mm5, 8 ; shift left 8 to put it back + por mm0, mm5 ; put the two together to get result + + paddsb mm6, mm0 ; p0+= p0 add + pxor mm6, [GLOBAL(t80)] ; unoffset + + + movq mm0, t0 + movq mm4, t1 + + ; mm0 = 70 60 50 40 30 20 10 00 + ; mm6 = 71 61 51 41 31 21 11 01 + ; mm3 = 72 62 52 42 32 22 12 02 + ; mm4 = 73 63 53 43 33 23 13 03 + ; transpose back to write out + + movq mm1, mm0 ; + punpcklbw mm0, mm6 ; 31 30 21 20 11 10 01 00 + + punpckhbw mm1, mm6 ; 71 70 61 60 51 50 41 40 + movq mm2, mm3 ; + + punpcklbw mm2, mm4 ; 33 32 23 22 13 12 03 02 + movq mm5, mm1 ; 71 70 61 60 51 50 41 40 + + punpckhbw mm3, mm4 ; 73 72 63 62 53 52 43 42 + movq mm6, mm0 ; 31 30 21 20 11 10 01 00 + + punpcklwd mm0, mm2 ; 13 12 11 10 03 02 01 00 + punpckhwd mm6, mm2 ; 33 32 31 30 23 22 21 20 + + movd [rsi+rax*4], mm0 ; write 03 02 01 00 + punpcklwd mm1, mm3 ; 53 52 51 50 43 42 41 40 + + psrlq mm0, 32 ; xx xx xx xx 13 12 11 10 + punpckhwd mm5, mm3 ; 73 72 71 70 63 62 61 60 + + movd [rdi+rax*4], mm0 ; write 13 12 11 10 + movd [rsi+rax*2], mm6 ; write 23 22 21 20 + + psrlq mm6, 32 ; 33 32 31 30 + movd [rsi], mm1 ; write 43 42 41 40 + + movd [rsi + rax], mm6 ; write 33 32 31 30 + neg rax + + movd [rsi + rax*2], mm5 ; write 63 62 61 60 + psrlq mm1, 32 ; 53 52 51 50 + + movd [rdi], mm1 ; write out 53 52 51 50 + psrlq mm5, 32 ; 73 72 71 70 + + movd [rdi + rax*2], mm5 ; write 73 72 71 70 + + lea rsi, [rsi+rax*8] ; next 8 + + dec rcx + jnz .nexts8_v + + add rsp, 32 + pop rsp + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + UNSHADOW_ARGS + pop rbp + ret + + + +;void fast_loop_filter_vertical_edges_mmx(unsigned char *y_ptr, +; int y_stride, +; loop_filter_info *lfi) +;{ +; +; +; vp8_loop_filter_simple_vertical_edge_mmx(y_ptr+4, y_stride, lfi->flim,lfi->lim,lfi->thr,2); +; vp8_loop_filter_simple_vertical_edge_mmx(y_ptr+8, y_stride, lfi->flim,lfi->lim,lfi->thr,2); +; vp8_loop_filter_simple_vertical_edge_mmx(y_ptr+12, y_stride, lfi->flim,lfi->lim,lfi->thr,2); +;} + +SECTION_RODATA +align 16 +tfe: + times 8 db 0xfe +align 16 +t80: + times 8 db 0x80 +align 16 +t1s: + times 8 db 0x01 +align 16 +t3: + times 8 db 0x03 +align 16 +t4: + times 8 db 0x04 +align 16 +ones: + times 4 dw 0x0001 +align 16 +s27: + times 4 dw 0x1b00 +align 16 +s18: + times 4 dw 0x1200 +align 16 +s9: + times 4 dw 0x0900 +align 16 +s63: + times 4 dw 0x003f diff --git a/thirdparty/libvpx/vp8/decoder/dboolhuff.c b/thirdparty/libvpx/vp8/decoder/dboolhuff.c new file mode 100644 index 0000000000..5cdd2a2491 --- /dev/null +++ b/thirdparty/libvpx/vp8/decoder/dboolhuff.c @@ -0,0 +1,77 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "dboolhuff.h" +#include "vp8/common/common.h" +#include "vpx_dsp/vpx_dsp_common.h" + +int vp8dx_start_decode(BOOL_DECODER *br, + const unsigned char *source, + unsigned int source_sz, + vpx_decrypt_cb decrypt_cb, + void *decrypt_state) +{ + br->user_buffer_end = source+source_sz; + br->user_buffer = source; + br->value = 0; + br->count = -8; + br->range = 255; + br->decrypt_cb = decrypt_cb; + br->decrypt_state = decrypt_state; + + if (source_sz && !source) + return 1; + + /* Populate the buffer */ + vp8dx_bool_decoder_fill(br); + + return 0; +} + +void vp8dx_bool_decoder_fill(BOOL_DECODER *br) +{ + const unsigned char *bufptr = br->user_buffer; + VP8_BD_VALUE value = br->value; + int count = br->count; + int shift = VP8_BD_VALUE_SIZE - CHAR_BIT - (count + CHAR_BIT); + size_t bytes_left = br->user_buffer_end - bufptr; + size_t bits_left = bytes_left * CHAR_BIT; + int x = shift + CHAR_BIT - (int)bits_left; + int loop_end = 0; + unsigned char decrypted[sizeof(VP8_BD_VALUE) + 1]; + + if (br->decrypt_cb) { + size_t n = VPXMIN(sizeof(decrypted), bytes_left); + br->decrypt_cb(br->decrypt_state, bufptr, decrypted, (int)n); + bufptr = decrypted; + } + + if(x >= 0) + { + count += VP8_LOTS_OF_BITS; + loop_end = x; + } + + if (x < 0 || bits_left) + { + while(shift >= loop_end) + { + count += CHAR_BIT; + value |= (VP8_BD_VALUE)*bufptr << shift; + ++bufptr; + ++br->user_buffer; + shift -= CHAR_BIT; + } + } + + br->value = value; + br->count = count; +} diff --git a/thirdparty/libvpx/vp8/decoder/dboolhuff.h b/thirdparty/libvpx/vp8/decoder/dboolhuff.h new file mode 100644 index 0000000000..1b1bbf868e --- /dev/null +++ b/thirdparty/libvpx/vp8/decoder/dboolhuff.h @@ -0,0 +1,141 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_DECODER_DBOOLHUFF_H_ +#define VP8_DECODER_DBOOLHUFF_H_ + +#include <stddef.h> +#include <limits.h> + +#include "./vpx_config.h" +#include "vpx_ports/mem.h" +#include "vpx/vp8dx.h" +#include "vpx/vpx_integer.h" + +#ifdef __cplusplus +extern "C" { +#endif + +typedef size_t VP8_BD_VALUE; + +#define VP8_BD_VALUE_SIZE ((int)sizeof(VP8_BD_VALUE)*CHAR_BIT) + +/*This is meant to be a large, positive constant that can still be efficiently + loaded as an immediate (on platforms like ARM, for example). + Even relatively modest values like 100 would work fine.*/ +#define VP8_LOTS_OF_BITS (0x40000000) + +typedef struct +{ + const unsigned char *user_buffer_end; + const unsigned char *user_buffer; + VP8_BD_VALUE value; + int count; + unsigned int range; + vpx_decrypt_cb decrypt_cb; + void *decrypt_state; +} BOOL_DECODER; + +DECLARE_ALIGNED(16, extern const unsigned char, vp8_norm[256]); + +int vp8dx_start_decode(BOOL_DECODER *br, + const unsigned char *source, + unsigned int source_sz, + vpx_decrypt_cb decrypt_cb, + void *decrypt_state); + +void vp8dx_bool_decoder_fill(BOOL_DECODER *br); + + +static int vp8dx_decode_bool(BOOL_DECODER *br, int probability) { + unsigned int bit = 0; + VP8_BD_VALUE value; + unsigned int split; + VP8_BD_VALUE bigsplit; + int count; + unsigned int range; + + split = 1 + (((br->range - 1) * probability) >> 8); + + if(br->count < 0) + vp8dx_bool_decoder_fill(br); + + value = br->value; + count = br->count; + + bigsplit = (VP8_BD_VALUE)split << (VP8_BD_VALUE_SIZE - 8); + + range = split; + + if (value >= bigsplit) + { + range = br->range - split; + value = value - bigsplit; + bit = 1; + } + + { + register int shift = vp8_norm[range]; + range <<= shift; + value <<= shift; + count -= shift; + } + br->value = value; + br->count = count; + br->range = range; + + return bit; +} + +static INLINE int vp8_decode_value(BOOL_DECODER *br, int bits) +{ + int z = 0; + int bit; + + for (bit = bits - 1; bit >= 0; bit--) + { + z |= (vp8dx_decode_bool(br, 0x80) << bit); + } + + return z; +} + +static INLINE int vp8dx_bool_error(BOOL_DECODER *br) +{ + /* Check if we have reached the end of the buffer. + * + * Variable 'count' stores the number of bits in the 'value' buffer, minus + * 8. The top byte is part of the algorithm, and the remainder is buffered + * to be shifted into it. So if count == 8, the top 16 bits of 'value' are + * occupied, 8 for the algorithm and 8 in the buffer. + * + * When reading a byte from the user's buffer, count is filled with 8 and + * one byte is filled into the value buffer. When we reach the end of the + * data, count is additionally filled with VP8_LOTS_OF_BITS. So when + * count == VP8_LOTS_OF_BITS - 1, the user's data has been exhausted. + */ + if ((br->count > VP8_BD_VALUE_SIZE) && (br->count < VP8_LOTS_OF_BITS)) + { + /* We have tried to decode bits after the end of + * stream was encountered. + */ + return 1; + } + + /* No error. */ + return 0; +} + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_DECODER_DBOOLHUFF_H_ diff --git a/thirdparty/libvpx/vp8/decoder/decodeframe.c b/thirdparty/libvpx/vp8/decoder/decodeframe.c new file mode 100644 index 0000000000..51acdbb9c8 --- /dev/null +++ b/thirdparty/libvpx/vp8/decoder/decodeframe.c @@ -0,0 +1,1397 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "vpx_config.h" +#include "vp8_rtcd.h" +#include "./vpx_scale_rtcd.h" +#include "onyxd_int.h" +#include "vp8/common/header.h" +#include "vp8/common/reconintra4x4.h" +#include "vp8/common/reconinter.h" +#include "detokenize.h" +#include "vp8/common/common.h" +#include "vp8/common/invtrans.h" +#include "vp8/common/alloccommon.h" +#include "vp8/common/entropymode.h" +#include "vp8/common/quant_common.h" +#include "vpx_scale/vpx_scale.h" +#include "vp8/common/reconintra.h" +#include "vp8/common/setupintrarecon.h" + +#include "decodemv.h" +#include "vp8/common/extend.h" +#if CONFIG_ERROR_CONCEALMENT +#include "error_concealment.h" +#endif +#include "vpx_mem/vpx_mem.h" +#include "vp8/common/threading.h" +#include "decoderthreading.h" +#include "dboolhuff.h" +#include "vpx_dsp/vpx_dsp_common.h" + +#include <assert.h> +#include <stdio.h> + +void vp8cx_init_de_quantizer(VP8D_COMP *pbi) +{ + int Q; + VP8_COMMON *const pc = & pbi->common; + + for (Q = 0; Q < QINDEX_RANGE; Q++) + { + pc->Y1dequant[Q][0] = (short)vp8_dc_quant(Q, pc->y1dc_delta_q); + pc->Y2dequant[Q][0] = (short)vp8_dc2quant(Q, pc->y2dc_delta_q); + pc->UVdequant[Q][0] = (short)vp8_dc_uv_quant(Q, pc->uvdc_delta_q); + + pc->Y1dequant[Q][1] = (short)vp8_ac_yquant(Q); + pc->Y2dequant[Q][1] = (short)vp8_ac2quant(Q, pc->y2ac_delta_q); + pc->UVdequant[Q][1] = (short)vp8_ac_uv_quant(Q, pc->uvac_delta_q); + } +} + +void vp8_mb_init_dequantizer(VP8D_COMP *pbi, MACROBLOCKD *xd) +{ + int i; + int QIndex; + MB_MODE_INFO *mbmi = &xd->mode_info_context->mbmi; + VP8_COMMON *const pc = & pbi->common; + + /* Decide whether to use the default or alternate baseline Q value. */ + if (xd->segmentation_enabled) + { + /* Abs Value */ + if (xd->mb_segement_abs_delta == SEGMENT_ABSDATA) + QIndex = xd->segment_feature_data[MB_LVL_ALT_Q][mbmi->segment_id]; + + /* Delta Value */ + else + QIndex = pc->base_qindex + xd->segment_feature_data[MB_LVL_ALT_Q][mbmi->segment_id]; + + QIndex = (QIndex >= 0) ? ((QIndex <= MAXQ) ? QIndex : MAXQ) : 0; /* Clamp to valid range */ + } + else + QIndex = pc->base_qindex; + + /* Set up the macroblock dequant constants */ + xd->dequant_y1_dc[0] = 1; + xd->dequant_y1[0] = pc->Y1dequant[QIndex][0]; + xd->dequant_y2[0] = pc->Y2dequant[QIndex][0]; + xd->dequant_uv[0] = pc->UVdequant[QIndex][0]; + + for (i = 1; i < 16; i++) + { + xd->dequant_y1_dc[i] = + xd->dequant_y1[i] = pc->Y1dequant[QIndex][1]; + xd->dequant_y2[i] = pc->Y2dequant[QIndex][1]; + xd->dequant_uv[i] = pc->UVdequant[QIndex][1]; + } +} + +static void decode_macroblock(VP8D_COMP *pbi, MACROBLOCKD *xd, + unsigned int mb_idx) +{ + MB_PREDICTION_MODE mode; + int i; +#if CONFIG_ERROR_CONCEALMENT + int corruption_detected = 0; +#else + (void)mb_idx; +#endif + + if (xd->mode_info_context->mbmi.mb_skip_coeff) + { + vp8_reset_mb_tokens_context(xd); + } + else if (!vp8dx_bool_error(xd->current_bc)) + { + int eobtotal; + eobtotal = vp8_decode_mb_tokens(pbi, xd); + + /* Special case: Force the loopfilter to skip when eobtotal is zero */ + xd->mode_info_context->mbmi.mb_skip_coeff = (eobtotal==0); + } + + mode = xd->mode_info_context->mbmi.mode; + + if (xd->segmentation_enabled) + vp8_mb_init_dequantizer(pbi, xd); + + +#if CONFIG_ERROR_CONCEALMENT + + if(pbi->ec_active) + { + int throw_residual; + /* When we have independent partitions we can apply residual even + * though other partitions within the frame are corrupt. + */ + throw_residual = (!pbi->independent_partitions && + pbi->frame_corrupt_residual); + throw_residual = (throw_residual || vp8dx_bool_error(xd->current_bc)); + + if ((mb_idx >= pbi->mvs_corrupt_from_mb || throw_residual)) + { + /* MB with corrupt residuals or corrupt mode/motion vectors. + * Better to use the predictor as reconstruction. + */ + pbi->frame_corrupt_residual = 1; + memset(xd->qcoeff, 0, sizeof(xd->qcoeff)); + + corruption_detected = 1; + + /* force idct to be skipped for B_PRED and use the + * prediction only for reconstruction + * */ + memset(xd->eobs, 0, 25); + } + } +#endif + + /* do prediction */ + if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) + { + vp8_build_intra_predictors_mbuv_s(xd, + xd->recon_above[1], + xd->recon_above[2], + xd->recon_left[1], + xd->recon_left[2], + xd->recon_left_stride[1], + xd->dst.u_buffer, xd->dst.v_buffer, + xd->dst.uv_stride); + + if (mode != B_PRED) + { + vp8_build_intra_predictors_mby_s(xd, + xd->recon_above[0], + xd->recon_left[0], + xd->recon_left_stride[0], + xd->dst.y_buffer, + xd->dst.y_stride); + } + else + { + short *DQC = xd->dequant_y1; + int dst_stride = xd->dst.y_stride; + + /* clear out residual eob info */ + if(xd->mode_info_context->mbmi.mb_skip_coeff) + memset(xd->eobs, 0, 25); + + intra_prediction_down_copy(xd, xd->recon_above[0] + 16); + + for (i = 0; i < 16; i++) + { + BLOCKD *b = &xd->block[i]; + unsigned char *dst = xd->dst.y_buffer + b->offset; + B_PREDICTION_MODE b_mode = + xd->mode_info_context->bmi[i].as_mode; + unsigned char *Above = dst - dst_stride; + unsigned char *yleft = dst - 1; + int left_stride = dst_stride; + unsigned char top_left = Above[-1]; + + vp8_intra4x4_predict(Above, yleft, left_stride, b_mode, + dst, dst_stride, top_left); + + if (xd->eobs[i]) + { + if (xd->eobs[i] > 1) + { + vp8_dequant_idct_add(b->qcoeff, DQC, dst, dst_stride); + } + else + { + vp8_dc_only_idct_add + (b->qcoeff[0] * DQC[0], + dst, dst_stride, + dst, dst_stride); + memset(b->qcoeff, 0, 2 * sizeof(b->qcoeff[0])); + } + } + } + } + } + else + { + vp8_build_inter_predictors_mb(xd); + } + + +#if CONFIG_ERROR_CONCEALMENT + if (corruption_detected) + { + return; + } +#endif + + if(!xd->mode_info_context->mbmi.mb_skip_coeff) + { + /* dequantization and idct */ + if (mode != B_PRED) + { + short *DQC = xd->dequant_y1; + + if (mode != SPLITMV) + { + BLOCKD *b = &xd->block[24]; + + /* do 2nd order transform on the dc block */ + if (xd->eobs[24] > 1) + { + vp8_dequantize_b(b, xd->dequant_y2); + + vp8_short_inv_walsh4x4(&b->dqcoeff[0], + xd->qcoeff); + memset(b->qcoeff, 0, 16 * sizeof(b->qcoeff[0])); + } + else + { + b->dqcoeff[0] = b->qcoeff[0] * xd->dequant_y2[0]; + vp8_short_inv_walsh4x4_1(&b->dqcoeff[0], + xd->qcoeff); + memset(b->qcoeff, 0, 2 * sizeof(b->qcoeff[0])); + } + + /* override the dc dequant constant in order to preserve the + * dc components + */ + DQC = xd->dequant_y1_dc; + } + + vp8_dequant_idct_add_y_block + (xd->qcoeff, DQC, + xd->dst.y_buffer, + xd->dst.y_stride, xd->eobs); + } + + vp8_dequant_idct_add_uv_block + (xd->qcoeff+16*16, xd->dequant_uv, + xd->dst.u_buffer, xd->dst.v_buffer, + xd->dst.uv_stride, xd->eobs+16); + } +} + +static int get_delta_q(vp8_reader *bc, int prev, int *q_update) +{ + int ret_val = 0; + + if (vp8_read_bit(bc)) + { + ret_val = vp8_read_literal(bc, 4); + + if (vp8_read_bit(bc)) + ret_val = -ret_val; + } + + /* Trigger a quantizer update if the delta-q value has changed */ + if (ret_val != prev) + *q_update = 1; + + return ret_val; +} + +#ifdef PACKET_TESTING +#include <stdio.h> +FILE *vpxlog = 0; +#endif + +static void yv12_extend_frame_top_c(YV12_BUFFER_CONFIG *ybf) +{ + int i; + unsigned char *src_ptr1; + unsigned char *dest_ptr1; + + unsigned int Border; + int plane_stride; + + /***********/ + /* Y Plane */ + /***********/ + Border = ybf->border; + plane_stride = ybf->y_stride; + src_ptr1 = ybf->y_buffer - Border; + dest_ptr1 = src_ptr1 - (Border * plane_stride); + + for (i = 0; i < (int)Border; i++) + { + memcpy(dest_ptr1, src_ptr1, plane_stride); + dest_ptr1 += plane_stride; + } + + + /***********/ + /* U Plane */ + /***********/ + plane_stride = ybf->uv_stride; + Border /= 2; + src_ptr1 = ybf->u_buffer - Border; + dest_ptr1 = src_ptr1 - (Border * plane_stride); + + for (i = 0; i < (int)(Border); i++) + { + memcpy(dest_ptr1, src_ptr1, plane_stride); + dest_ptr1 += plane_stride; + } + + /***********/ + /* V Plane */ + /***********/ + + src_ptr1 = ybf->v_buffer - Border; + dest_ptr1 = src_ptr1 - (Border * plane_stride); + + for (i = 0; i < (int)(Border); i++) + { + memcpy(dest_ptr1, src_ptr1, plane_stride); + dest_ptr1 += plane_stride; + } +} + +static void yv12_extend_frame_bottom_c(YV12_BUFFER_CONFIG *ybf) +{ + int i; + unsigned char *src_ptr1, *src_ptr2; + unsigned char *dest_ptr2; + + unsigned int Border; + int plane_stride; + int plane_height; + + /***********/ + /* Y Plane */ + /***********/ + Border = ybf->border; + plane_stride = ybf->y_stride; + plane_height = ybf->y_height; + + src_ptr1 = ybf->y_buffer - Border; + src_ptr2 = src_ptr1 + (plane_height * plane_stride) - plane_stride; + dest_ptr2 = src_ptr2 + plane_stride; + + for (i = 0; i < (int)Border; i++) + { + memcpy(dest_ptr2, src_ptr2, plane_stride); + dest_ptr2 += plane_stride; + } + + + /***********/ + /* U Plane */ + /***********/ + plane_stride = ybf->uv_stride; + plane_height = ybf->uv_height; + Border /= 2; + + src_ptr1 = ybf->u_buffer - Border; + src_ptr2 = src_ptr1 + (plane_height * plane_stride) - plane_stride; + dest_ptr2 = src_ptr2 + plane_stride; + + for (i = 0; i < (int)(Border); i++) + { + memcpy(dest_ptr2, src_ptr2, plane_stride); + dest_ptr2 += plane_stride; + } + + /***********/ + /* V Plane */ + /***********/ + + src_ptr1 = ybf->v_buffer - Border; + src_ptr2 = src_ptr1 + (plane_height * plane_stride) - plane_stride; + dest_ptr2 = src_ptr2 + plane_stride; + + for (i = 0; i < (int)(Border); i++) + { + memcpy(dest_ptr2, src_ptr2, plane_stride); + dest_ptr2 += plane_stride; + } +} + +static void yv12_extend_frame_left_right_c(YV12_BUFFER_CONFIG *ybf, + unsigned char *y_src, + unsigned char *u_src, + unsigned char *v_src) +{ + int i; + unsigned char *src_ptr1, *src_ptr2; + unsigned char *dest_ptr1, *dest_ptr2; + + unsigned int Border; + int plane_stride; + int plane_height; + int plane_width; + + /***********/ + /* Y Plane */ + /***********/ + Border = ybf->border; + plane_stride = ybf->y_stride; + plane_height = 16; + plane_width = ybf->y_width; + + /* copy the left and right most columns out */ + src_ptr1 = y_src; + src_ptr2 = src_ptr1 + plane_width - 1; + dest_ptr1 = src_ptr1 - Border; + dest_ptr2 = src_ptr2 + 1; + + for (i = 0; i < plane_height; i++) + { + memset(dest_ptr1, src_ptr1[0], Border); + memset(dest_ptr2, src_ptr2[0], Border); + src_ptr1 += plane_stride; + src_ptr2 += plane_stride; + dest_ptr1 += plane_stride; + dest_ptr2 += plane_stride; + } + + /***********/ + /* U Plane */ + /***********/ + plane_stride = ybf->uv_stride; + plane_height = 8; + plane_width = ybf->uv_width; + Border /= 2; + + /* copy the left and right most columns out */ + src_ptr1 = u_src; + src_ptr2 = src_ptr1 + plane_width - 1; + dest_ptr1 = src_ptr1 - Border; + dest_ptr2 = src_ptr2 + 1; + + for (i = 0; i < plane_height; i++) + { + memset(dest_ptr1, src_ptr1[0], Border); + memset(dest_ptr2, src_ptr2[0], Border); + src_ptr1 += plane_stride; + src_ptr2 += plane_stride; + dest_ptr1 += plane_stride; + dest_ptr2 += plane_stride; + } + + /***********/ + /* V Plane */ + /***********/ + + /* copy the left and right most columns out */ + src_ptr1 = v_src; + src_ptr2 = src_ptr1 + plane_width - 1; + dest_ptr1 = src_ptr1 - Border; + dest_ptr2 = src_ptr2 + 1; + + for (i = 0; i < plane_height; i++) + { + memset(dest_ptr1, src_ptr1[0], Border); + memset(dest_ptr2, src_ptr2[0], Border); + src_ptr1 += plane_stride; + src_ptr2 += plane_stride; + dest_ptr1 += plane_stride; + dest_ptr2 += plane_stride; + } +} + +static void decode_mb_rows(VP8D_COMP *pbi) +{ + VP8_COMMON *const pc = & pbi->common; + MACROBLOCKD *const xd = & pbi->mb; + + MODE_INFO *lf_mic = xd->mode_info_context; + + int ibc = 0; + int num_part = 1 << pc->multi_token_partition; + + int recon_yoffset, recon_uvoffset; + int mb_row, mb_col; + int mb_idx = 0; + + YV12_BUFFER_CONFIG *yv12_fb_new = pbi->dec_fb_ref[INTRA_FRAME]; + + int recon_y_stride = yv12_fb_new->y_stride; + int recon_uv_stride = yv12_fb_new->uv_stride; + + unsigned char *ref_buffer[MAX_REF_FRAMES][3]; + unsigned char *dst_buffer[3]; + unsigned char *lf_dst[3]; + unsigned char *eb_dst[3]; + int i; + int ref_fb_corrupted[MAX_REF_FRAMES]; + + ref_fb_corrupted[INTRA_FRAME] = 0; + + for(i = 1; i < MAX_REF_FRAMES; i++) + { + YV12_BUFFER_CONFIG *this_fb = pbi->dec_fb_ref[i]; + + ref_buffer[i][0] = this_fb->y_buffer; + ref_buffer[i][1] = this_fb->u_buffer; + ref_buffer[i][2] = this_fb->v_buffer; + + ref_fb_corrupted[i] = this_fb->corrupted; + } + + /* Set up the buffer pointers */ + eb_dst[0] = lf_dst[0] = dst_buffer[0] = yv12_fb_new->y_buffer; + eb_dst[1] = lf_dst[1] = dst_buffer[1] = yv12_fb_new->u_buffer; + eb_dst[2] = lf_dst[2] = dst_buffer[2] = yv12_fb_new->v_buffer; + + xd->up_available = 0; + + /* Initialize the loop filter for this frame. */ + if(pc->filter_level) + vp8_loop_filter_frame_init(pc, xd, pc->filter_level); + + vp8_setup_intra_recon_top_line(yv12_fb_new); + + /* Decode the individual macro block */ + for (mb_row = 0; mb_row < pc->mb_rows; mb_row++) + { + if (num_part > 1) + { + xd->current_bc = & pbi->mbc[ibc]; + ibc++; + + if (ibc == num_part) + ibc = 0; + } + + recon_yoffset = mb_row * recon_y_stride * 16; + recon_uvoffset = mb_row * recon_uv_stride * 8; + + /* reset contexts */ + xd->above_context = pc->above_context; + memset(xd->left_context, 0, sizeof(ENTROPY_CONTEXT_PLANES)); + + xd->left_available = 0; + + xd->mb_to_top_edge = -((mb_row * 16) << 3); + xd->mb_to_bottom_edge = ((pc->mb_rows - 1 - mb_row) * 16) << 3; + + xd->recon_above[0] = dst_buffer[0] + recon_yoffset; + xd->recon_above[1] = dst_buffer[1] + recon_uvoffset; + xd->recon_above[2] = dst_buffer[2] + recon_uvoffset; + + xd->recon_left[0] = xd->recon_above[0] - 1; + xd->recon_left[1] = xd->recon_above[1] - 1; + xd->recon_left[2] = xd->recon_above[2] - 1; + + xd->recon_above[0] -= xd->dst.y_stride; + xd->recon_above[1] -= xd->dst.uv_stride; + xd->recon_above[2] -= xd->dst.uv_stride; + + /* TODO: move to outside row loop */ + xd->recon_left_stride[0] = xd->dst.y_stride; + xd->recon_left_stride[1] = xd->dst.uv_stride; + + setup_intra_recon_left(xd->recon_left[0], xd->recon_left[1], + xd->recon_left[2], xd->dst.y_stride, + xd->dst.uv_stride); + + for (mb_col = 0; mb_col < pc->mb_cols; mb_col++) + { + /* Distance of Mb to the various image edges. + * These are specified to 8th pel as they are always compared to values + * that are in 1/8th pel units + */ + xd->mb_to_left_edge = -((mb_col * 16) << 3); + xd->mb_to_right_edge = ((pc->mb_cols - 1 - mb_col) * 16) << 3; + +#if CONFIG_ERROR_CONCEALMENT + { + int corrupt_residual = (!pbi->independent_partitions && + pbi->frame_corrupt_residual) || + vp8dx_bool_error(xd->current_bc); + if (pbi->ec_active && + xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME && + corrupt_residual) + { + /* We have an intra block with corrupt coefficients, better to + * conceal with an inter block. Interpolate MVs from neighboring + * MBs. + * + * Note that for the first mb with corrupt residual in a frame, + * we might not discover that before decoding the residual. That + * happens after this check, and therefore no inter concealment + * will be done. + */ + vp8_interpolate_motion(xd, + mb_row, mb_col, + pc->mb_rows, pc->mb_cols); + } + } +#endif + + xd->dst.y_buffer = dst_buffer[0] + recon_yoffset; + xd->dst.u_buffer = dst_buffer[1] + recon_uvoffset; + xd->dst.v_buffer = dst_buffer[2] + recon_uvoffset; + + if (xd->mode_info_context->mbmi.ref_frame >= LAST_FRAME) { + const MV_REFERENCE_FRAME ref = xd->mode_info_context->mbmi.ref_frame; + xd->pre.y_buffer = ref_buffer[ref][0] + recon_yoffset; + xd->pre.u_buffer = ref_buffer[ref][1] + recon_uvoffset; + xd->pre.v_buffer = ref_buffer[ref][2] + recon_uvoffset; + } else { + // ref_frame is INTRA_FRAME, pre buffer should not be used. + xd->pre.y_buffer = 0; + xd->pre.u_buffer = 0; + xd->pre.v_buffer = 0; + } + + /* propagate errors from reference frames */ + xd->corrupted |= ref_fb_corrupted[xd->mode_info_context->mbmi.ref_frame]; + + decode_macroblock(pbi, xd, mb_idx); + + mb_idx++; + xd->left_available = 1; + + /* check if the boolean decoder has suffered an error */ + xd->corrupted |= vp8dx_bool_error(xd->current_bc); + + xd->recon_above[0] += 16; + xd->recon_above[1] += 8; + xd->recon_above[2] += 8; + xd->recon_left[0] += 16; + xd->recon_left[1] += 8; + xd->recon_left[2] += 8; + + recon_yoffset += 16; + recon_uvoffset += 8; + + ++xd->mode_info_context; /* next mb */ + + xd->above_context++; + } + + /* adjust to the next row of mbs */ + vp8_extend_mb_row(yv12_fb_new, xd->dst.y_buffer + 16, + xd->dst.u_buffer + 8, xd->dst.v_buffer + 8); + + ++xd->mode_info_context; /* skip prediction column */ + xd->up_available = 1; + + if(pc->filter_level) + { + if(mb_row > 0) + { + if (pc->filter_type == NORMAL_LOOPFILTER) + vp8_loop_filter_row_normal(pc, lf_mic, mb_row-1, + recon_y_stride, recon_uv_stride, + lf_dst[0], lf_dst[1], lf_dst[2]); + else + vp8_loop_filter_row_simple(pc, lf_mic, mb_row-1, + recon_y_stride, recon_uv_stride, + lf_dst[0], lf_dst[1], lf_dst[2]); + if(mb_row > 1) + { + yv12_extend_frame_left_right_c(yv12_fb_new, + eb_dst[0], + eb_dst[1], + eb_dst[2]); + + eb_dst[0] += recon_y_stride * 16; + eb_dst[1] += recon_uv_stride * 8; + eb_dst[2] += recon_uv_stride * 8; + } + + lf_dst[0] += recon_y_stride * 16; + lf_dst[1] += recon_uv_stride * 8; + lf_dst[2] += recon_uv_stride * 8; + lf_mic += pc->mb_cols; + lf_mic++; /* Skip border mb */ + } + } + else + { + if(mb_row > 0) + { + /**/ + yv12_extend_frame_left_right_c(yv12_fb_new, + eb_dst[0], + eb_dst[1], + eb_dst[2]); + eb_dst[0] += recon_y_stride * 16; + eb_dst[1] += recon_uv_stride * 8; + eb_dst[2] += recon_uv_stride * 8; + } + } + } + + if(pc->filter_level) + { + if (pc->filter_type == NORMAL_LOOPFILTER) + vp8_loop_filter_row_normal(pc, lf_mic, mb_row-1, recon_y_stride, + recon_uv_stride, lf_dst[0], lf_dst[1], + lf_dst[2]); + else + vp8_loop_filter_row_simple(pc, lf_mic, mb_row-1, recon_y_stride, + recon_uv_stride, lf_dst[0], lf_dst[1], + lf_dst[2]); + + yv12_extend_frame_left_right_c(yv12_fb_new, + eb_dst[0], + eb_dst[1], + eb_dst[2]); + eb_dst[0] += recon_y_stride * 16; + eb_dst[1] += recon_uv_stride * 8; + eb_dst[2] += recon_uv_stride * 8; + } + yv12_extend_frame_left_right_c(yv12_fb_new, + eb_dst[0], + eb_dst[1], + eb_dst[2]); + yv12_extend_frame_top_c(yv12_fb_new); + yv12_extend_frame_bottom_c(yv12_fb_new); + +} + +static unsigned int read_partition_size(VP8D_COMP *pbi, + const unsigned char *cx_size) +{ + unsigned char temp[3]; + if (pbi->decrypt_cb) + { + pbi->decrypt_cb(pbi->decrypt_state, cx_size, temp, 3); + cx_size = temp; + } + return cx_size[0] + (cx_size[1] << 8) + (cx_size[2] << 16); +} + +static int read_is_valid(const unsigned char *start, + size_t len, + const unsigned char *end) +{ + return (start + len > start && start + len <= end); +} + +static unsigned int read_available_partition_size( + VP8D_COMP *pbi, + const unsigned char *token_part_sizes, + const unsigned char *fragment_start, + const unsigned char *first_fragment_end, + const unsigned char *fragment_end, + int i, + int num_part) +{ + VP8_COMMON* pc = &pbi->common; + const unsigned char *partition_size_ptr = token_part_sizes + i * 3; + unsigned int partition_size = 0; + ptrdiff_t bytes_left = fragment_end - fragment_start; + /* Calculate the length of this partition. The last partition + * size is implicit. If the partition size can't be read, then + * either use the remaining data in the buffer (for EC mode) + * or throw an error. + */ + if (i < num_part - 1) + { + if (read_is_valid(partition_size_ptr, 3, first_fragment_end)) + partition_size = read_partition_size(pbi, partition_size_ptr); + else if (pbi->ec_active) + partition_size = (unsigned int)bytes_left; + else + vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME, + "Truncated partition size data"); + } + else + partition_size = (unsigned int)bytes_left; + + /* Validate the calculated partition length. If the buffer + * described by the partition can't be fully read, then restrict + * it to the portion that can be (for EC mode) or throw an error. + */ + if (!read_is_valid(fragment_start, partition_size, fragment_end)) + { + if (pbi->ec_active) + partition_size = (unsigned int)bytes_left; + else + vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME, + "Truncated packet or corrupt partition " + "%d length", i + 1); + } + return partition_size; +} + + +static void setup_token_decoder(VP8D_COMP *pbi, + const unsigned char* token_part_sizes) +{ + vp8_reader *bool_decoder = &pbi->mbc[0]; + unsigned int partition_idx; + unsigned int fragment_idx; + unsigned int num_token_partitions; + const unsigned char *first_fragment_end = pbi->fragments.ptrs[0] + + pbi->fragments.sizes[0]; + + TOKEN_PARTITION multi_token_partition = + (TOKEN_PARTITION)vp8_read_literal(&pbi->mbc[8], 2); + if (!vp8dx_bool_error(&pbi->mbc[8])) + pbi->common.multi_token_partition = multi_token_partition; + num_token_partitions = 1 << pbi->common.multi_token_partition; + + /* Check for partitions within the fragments and unpack the fragments + * so that each fragment pointer points to its corresponding partition. */ + for (fragment_idx = 0; fragment_idx < pbi->fragments.count; ++fragment_idx) + { + unsigned int fragment_size = pbi->fragments.sizes[fragment_idx]; + const unsigned char *fragment_end = pbi->fragments.ptrs[fragment_idx] + + fragment_size; + /* Special case for handling the first partition since we have already + * read its size. */ + if (fragment_idx == 0) + { + /* Size of first partition + token partition sizes element */ + ptrdiff_t ext_first_part_size = token_part_sizes - + pbi->fragments.ptrs[0] + 3 * (num_token_partitions - 1); + fragment_size -= (unsigned int)ext_first_part_size; + if (fragment_size > 0) + { + pbi->fragments.sizes[0] = (unsigned int)ext_first_part_size; + /* The fragment contains an additional partition. Move to + * next. */ + fragment_idx++; + pbi->fragments.ptrs[fragment_idx] = pbi->fragments.ptrs[0] + + pbi->fragments.sizes[0]; + } + } + /* Split the chunk into partitions read from the bitstream */ + while (fragment_size > 0) + { + ptrdiff_t partition_size = read_available_partition_size( + pbi, + token_part_sizes, + pbi->fragments.ptrs[fragment_idx], + first_fragment_end, + fragment_end, + fragment_idx - 1, + num_token_partitions); + pbi->fragments.sizes[fragment_idx] = (unsigned int)partition_size; + fragment_size -= (unsigned int)partition_size; + assert(fragment_idx <= num_token_partitions); + if (fragment_size > 0) + { + /* The fragment contains an additional partition. + * Move to next. */ + fragment_idx++; + pbi->fragments.ptrs[fragment_idx] = + pbi->fragments.ptrs[fragment_idx - 1] + partition_size; + } + } + } + + pbi->fragments.count = num_token_partitions + 1; + + for (partition_idx = 1; partition_idx < pbi->fragments.count; ++partition_idx) + { + if (vp8dx_start_decode(bool_decoder, + pbi->fragments.ptrs[partition_idx], + pbi->fragments.sizes[partition_idx], + pbi->decrypt_cb, pbi->decrypt_state)) + vpx_internal_error(&pbi->common.error, VPX_CODEC_MEM_ERROR, + "Failed to allocate bool decoder %d", + partition_idx); + + bool_decoder++; + } + +#if CONFIG_MULTITHREAD + /* Clamp number of decoder threads */ + if (pbi->decoding_thread_count > num_token_partitions - 1) + pbi->decoding_thread_count = num_token_partitions - 1; +#endif +} + + +static void init_frame(VP8D_COMP *pbi) +{ + VP8_COMMON *const pc = & pbi->common; + MACROBLOCKD *const xd = & pbi->mb; + + if (pc->frame_type == KEY_FRAME) + { + /* Various keyframe initializations */ + memcpy(pc->fc.mvc, vp8_default_mv_context, sizeof(vp8_default_mv_context)); + + vp8_init_mbmode_probs(pc); + + vp8_default_coef_probs(pc); + + /* reset the segment feature data to 0 with delta coding (Default state). */ + memset(xd->segment_feature_data, 0, sizeof(xd->segment_feature_data)); + xd->mb_segement_abs_delta = SEGMENT_DELTADATA; + + /* reset the mode ref deltasa for loop filter */ + memset(xd->ref_lf_deltas, 0, sizeof(xd->ref_lf_deltas)); + memset(xd->mode_lf_deltas, 0, sizeof(xd->mode_lf_deltas)); + + /* All buffers are implicitly updated on key frames. */ + pc->refresh_golden_frame = 1; + pc->refresh_alt_ref_frame = 1; + pc->copy_buffer_to_gf = 0; + pc->copy_buffer_to_arf = 0; + + /* Note that Golden and Altref modes cannot be used on a key frame so + * ref_frame_sign_bias[] is undefined and meaningless + */ + pc->ref_frame_sign_bias[GOLDEN_FRAME] = 0; + pc->ref_frame_sign_bias[ALTREF_FRAME] = 0; + } + else + { + /* To enable choice of different interploation filters */ + if (!pc->use_bilinear_mc_filter) + { + xd->subpixel_predict = vp8_sixtap_predict4x4; + xd->subpixel_predict8x4 = vp8_sixtap_predict8x4; + xd->subpixel_predict8x8 = vp8_sixtap_predict8x8; + xd->subpixel_predict16x16 = vp8_sixtap_predict16x16; + } + else + { + xd->subpixel_predict = vp8_bilinear_predict4x4; + xd->subpixel_predict8x4 = vp8_bilinear_predict8x4; + xd->subpixel_predict8x8 = vp8_bilinear_predict8x8; + xd->subpixel_predict16x16 = vp8_bilinear_predict16x16; + } + + if (pbi->decoded_key_frame && pbi->ec_enabled && !pbi->ec_active) + pbi->ec_active = 1; + } + + xd->left_context = &pc->left_context; + xd->mode_info_context = pc->mi; + xd->frame_type = pc->frame_type; + xd->mode_info_context->mbmi.mode = DC_PRED; + xd->mode_info_stride = pc->mode_info_stride; + xd->corrupted = 0; /* init without corruption */ + + xd->fullpixel_mask = 0xffffffff; + if(pc->full_pixel) + xd->fullpixel_mask = 0xfffffff8; + +} + +int vp8_decode_frame(VP8D_COMP *pbi) +{ + vp8_reader *const bc = &pbi->mbc[8]; + VP8_COMMON *const pc = &pbi->common; + MACROBLOCKD *const xd = &pbi->mb; + const unsigned char *data = pbi->fragments.ptrs[0]; + const unsigned int data_sz = pbi->fragments.sizes[0]; + const unsigned char *data_end = data + data_sz; + ptrdiff_t first_partition_length_in_bytes; + + int i, j, k, l; + const int *const mb_feature_data_bits = vp8_mb_feature_data_bits; + int corrupt_tokens = 0; + int prev_independent_partitions = pbi->independent_partitions; + + YV12_BUFFER_CONFIG *yv12_fb_new = pbi->dec_fb_ref[INTRA_FRAME]; + + /* start with no corruption of current frame */ + xd->corrupted = 0; + yv12_fb_new->corrupted = 0; + + if (data_end - data < 3) + { + if (!pbi->ec_active) + { + vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME, + "Truncated packet"); + } + + /* Declare the missing frame as an inter frame since it will + be handled as an inter frame when we have estimated its + motion vectors. */ + pc->frame_type = INTER_FRAME; + pc->version = 0; + pc->show_frame = 1; + first_partition_length_in_bytes = 0; + } + else + { + unsigned char clear_buffer[10]; + const unsigned char *clear = data; + if (pbi->decrypt_cb) + { + int n = (int)VPXMIN(sizeof(clear_buffer), data_sz); + pbi->decrypt_cb(pbi->decrypt_state, data, clear_buffer, n); + clear = clear_buffer; + } + + pc->frame_type = (FRAME_TYPE)(clear[0] & 1); + pc->version = (clear[0] >> 1) & 7; + pc->show_frame = (clear[0] >> 4) & 1; + first_partition_length_in_bytes = + (clear[0] | (clear[1] << 8) | (clear[2] << 16)) >> 5; + + if (!pbi->ec_active && + (data + first_partition_length_in_bytes > data_end + || data + first_partition_length_in_bytes < data)) + vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME, + "Truncated packet or corrupt partition 0 length"); + + data += 3; + clear += 3; + + vp8_setup_version(pc); + + + if (pc->frame_type == KEY_FRAME) + { + /* vet via sync code */ + /* When error concealment is enabled we should only check the sync + * code if we have enough bits available + */ + if (!pbi->ec_active || data + 3 < data_end) + { + if (clear[0] != 0x9d || clear[1] != 0x01 || clear[2] != 0x2a) + vpx_internal_error(&pc->error, VPX_CODEC_UNSUP_BITSTREAM, + "Invalid frame sync code"); + } + + /* If error concealment is enabled we should only parse the new size + * if we have enough data. Otherwise we will end up with the wrong + * size. + */ + if (!pbi->ec_active || data + 6 < data_end) + { + pc->Width = (clear[3] | (clear[4] << 8)) & 0x3fff; + pc->horiz_scale = clear[4] >> 6; + pc->Height = (clear[5] | (clear[6] << 8)) & 0x3fff; + pc->vert_scale = clear[6] >> 6; + } + data += 7; + } + else + { + memcpy(&xd->pre, yv12_fb_new, sizeof(YV12_BUFFER_CONFIG)); + memcpy(&xd->dst, yv12_fb_new, sizeof(YV12_BUFFER_CONFIG)); + } + } + if ((!pbi->decoded_key_frame && pc->frame_type != KEY_FRAME)) + { + return -1; + } + + init_frame(pbi); + + if (vp8dx_start_decode(bc, data, (unsigned int)(data_end - data), + pbi->decrypt_cb, pbi->decrypt_state)) + vpx_internal_error(&pc->error, VPX_CODEC_MEM_ERROR, + "Failed to allocate bool decoder 0"); + if (pc->frame_type == KEY_FRAME) { + (void)vp8_read_bit(bc); // colorspace + pc->clamp_type = (CLAMP_TYPE)vp8_read_bit(bc); + } + + /* Is segmentation enabled */ + xd->segmentation_enabled = (unsigned char)vp8_read_bit(bc); + + if (xd->segmentation_enabled) + { + /* Signal whether or not the segmentation map is being explicitly updated this frame. */ + xd->update_mb_segmentation_map = (unsigned char)vp8_read_bit(bc); + xd->update_mb_segmentation_data = (unsigned char)vp8_read_bit(bc); + + if (xd->update_mb_segmentation_data) + { + xd->mb_segement_abs_delta = (unsigned char)vp8_read_bit(bc); + + memset(xd->segment_feature_data, 0, sizeof(xd->segment_feature_data)); + + /* For each segmentation feature (Quant and loop filter level) */ + for (i = 0; i < MB_LVL_MAX; i++) + { + for (j = 0; j < MAX_MB_SEGMENTS; j++) + { + /* Frame level data */ + if (vp8_read_bit(bc)) + { + xd->segment_feature_data[i][j] = (signed char)vp8_read_literal(bc, mb_feature_data_bits[i]); + + if (vp8_read_bit(bc)) + xd->segment_feature_data[i][j] = -xd->segment_feature_data[i][j]; + } + else + xd->segment_feature_data[i][j] = 0; + } + } + } + + if (xd->update_mb_segmentation_map) + { + /* Which macro block level features are enabled */ + memset(xd->mb_segment_tree_probs, 255, sizeof(xd->mb_segment_tree_probs)); + + /* Read the probs used to decode the segment id for each macro block. */ + for (i = 0; i < MB_FEATURE_TREE_PROBS; i++) + { + /* If not explicitly set value is defaulted to 255 by memset above */ + if (vp8_read_bit(bc)) + xd->mb_segment_tree_probs[i] = (vp8_prob)vp8_read_literal(bc, 8); + } + } + } + else + { + /* No segmentation updates on this frame */ + xd->update_mb_segmentation_map = 0; + xd->update_mb_segmentation_data = 0; + } + + /* Read the loop filter level and type */ + pc->filter_type = (LOOPFILTERTYPE) vp8_read_bit(bc); + pc->filter_level = vp8_read_literal(bc, 6); + pc->sharpness_level = vp8_read_literal(bc, 3); + + /* Read in loop filter deltas applied at the MB level based on mode or ref frame. */ + xd->mode_ref_lf_delta_update = 0; + xd->mode_ref_lf_delta_enabled = (unsigned char)vp8_read_bit(bc); + + if (xd->mode_ref_lf_delta_enabled) + { + /* Do the deltas need to be updated */ + xd->mode_ref_lf_delta_update = (unsigned char)vp8_read_bit(bc); + + if (xd->mode_ref_lf_delta_update) + { + /* Send update */ + for (i = 0; i < MAX_REF_LF_DELTAS; i++) + { + if (vp8_read_bit(bc)) + { + /*sign = vp8_read_bit( bc );*/ + xd->ref_lf_deltas[i] = (signed char)vp8_read_literal(bc, 6); + + if (vp8_read_bit(bc)) /* Apply sign */ + xd->ref_lf_deltas[i] = xd->ref_lf_deltas[i] * -1; + } + } + + /* Send update */ + for (i = 0; i < MAX_MODE_LF_DELTAS; i++) + { + if (vp8_read_bit(bc)) + { + /*sign = vp8_read_bit( bc );*/ + xd->mode_lf_deltas[i] = (signed char)vp8_read_literal(bc, 6); + + if (vp8_read_bit(bc)) /* Apply sign */ + xd->mode_lf_deltas[i] = xd->mode_lf_deltas[i] * -1; + } + } + } + } + + setup_token_decoder(pbi, data + first_partition_length_in_bytes); + + xd->current_bc = &pbi->mbc[0]; + + /* Read the default quantizers. */ + { + int Q, q_update; + + Q = vp8_read_literal(bc, 7); /* AC 1st order Q = default */ + pc->base_qindex = Q; + q_update = 0; + pc->y1dc_delta_q = get_delta_q(bc, pc->y1dc_delta_q, &q_update); + pc->y2dc_delta_q = get_delta_q(bc, pc->y2dc_delta_q, &q_update); + pc->y2ac_delta_q = get_delta_q(bc, pc->y2ac_delta_q, &q_update); + pc->uvdc_delta_q = get_delta_q(bc, pc->uvdc_delta_q, &q_update); + pc->uvac_delta_q = get_delta_q(bc, pc->uvac_delta_q, &q_update); + + if (q_update) + vp8cx_init_de_quantizer(pbi); + + /* MB level dequantizer setup */ + vp8_mb_init_dequantizer(pbi, &pbi->mb); + } + + /* Determine if the golden frame or ARF buffer should be updated and how. + * For all non key frames the GF and ARF refresh flags and sign bias + * flags must be set explicitly. + */ + if (pc->frame_type != KEY_FRAME) + { + /* Should the GF or ARF be updated from the current frame */ + pc->refresh_golden_frame = vp8_read_bit(bc); +#if CONFIG_ERROR_CONCEALMENT + /* Assume we shouldn't refresh golden if the bit is missing */ + xd->corrupted |= vp8dx_bool_error(bc); + if (pbi->ec_active && xd->corrupted) + pc->refresh_golden_frame = 0; +#endif + + pc->refresh_alt_ref_frame = vp8_read_bit(bc); +#if CONFIG_ERROR_CONCEALMENT + /* Assume we shouldn't refresh altref if the bit is missing */ + xd->corrupted |= vp8dx_bool_error(bc); + if (pbi->ec_active && xd->corrupted) + pc->refresh_alt_ref_frame = 0; +#endif + + /* Buffer to buffer copy flags. */ + pc->copy_buffer_to_gf = 0; + + if (!pc->refresh_golden_frame) + pc->copy_buffer_to_gf = vp8_read_literal(bc, 2); + +#if CONFIG_ERROR_CONCEALMENT + /* Assume we shouldn't copy to the golden if the bit is missing */ + xd->corrupted |= vp8dx_bool_error(bc); + if (pbi->ec_active && xd->corrupted) + pc->copy_buffer_to_gf = 0; +#endif + + pc->copy_buffer_to_arf = 0; + + if (!pc->refresh_alt_ref_frame) + pc->copy_buffer_to_arf = vp8_read_literal(bc, 2); + +#if CONFIG_ERROR_CONCEALMENT + /* Assume we shouldn't copy to the alt-ref if the bit is missing */ + xd->corrupted |= vp8dx_bool_error(bc); + if (pbi->ec_active && xd->corrupted) + pc->copy_buffer_to_arf = 0; +#endif + + + pc->ref_frame_sign_bias[GOLDEN_FRAME] = vp8_read_bit(bc); + pc->ref_frame_sign_bias[ALTREF_FRAME] = vp8_read_bit(bc); + } + + pc->refresh_entropy_probs = vp8_read_bit(bc); +#if CONFIG_ERROR_CONCEALMENT + /* Assume we shouldn't refresh the probabilities if the bit is + * missing */ + xd->corrupted |= vp8dx_bool_error(bc); + if (pbi->ec_active && xd->corrupted) + pc->refresh_entropy_probs = 0; +#endif + if (pc->refresh_entropy_probs == 0) + { + memcpy(&pc->lfc, &pc->fc, sizeof(pc->fc)); + } + + pc->refresh_last_frame = pc->frame_type == KEY_FRAME || vp8_read_bit(bc); + +#if CONFIG_ERROR_CONCEALMENT + /* Assume we should refresh the last frame if the bit is missing */ + xd->corrupted |= vp8dx_bool_error(bc); + if (pbi->ec_active && xd->corrupted) + pc->refresh_last_frame = 1; +#endif + + if (0) + { + FILE *z = fopen("decodestats.stt", "a"); + fprintf(z, "%6d F:%d,G:%d,A:%d,L:%d,Q:%d\n", + pc->current_video_frame, + pc->frame_type, + pc->refresh_golden_frame, + pc->refresh_alt_ref_frame, + pc->refresh_last_frame, + pc->base_qindex); + fclose(z); + } + + { + pbi->independent_partitions = 1; + + /* read coef probability tree */ + for (i = 0; i < BLOCK_TYPES; i++) + for (j = 0; j < COEF_BANDS; j++) + for (k = 0; k < PREV_COEF_CONTEXTS; k++) + for (l = 0; l < ENTROPY_NODES; l++) + { + + vp8_prob *const p = pc->fc.coef_probs [i][j][k] + l; + + if (vp8_read(bc, vp8_coef_update_probs [i][j][k][l])) + { + *p = (vp8_prob)vp8_read_literal(bc, 8); + + } + if (k > 0 && *p != pc->fc.coef_probs[i][j][k-1][l]) + pbi->independent_partitions = 0; + + } + } + + /* clear out the coeff buffer */ + memset(xd->qcoeff, 0, sizeof(xd->qcoeff)); + + vp8_decode_mode_mvs(pbi); + +#if CONFIG_ERROR_CONCEALMENT + if (pbi->ec_active && + pbi->mvs_corrupt_from_mb < (unsigned int)pc->mb_cols * pc->mb_rows) + { + /* Motion vectors are missing in this frame. We will try to estimate + * them and then continue decoding the frame as usual */ + vp8_estimate_missing_mvs(pbi); + } +#endif + + memset(pc->above_context, 0, sizeof(ENTROPY_CONTEXT_PLANES) * pc->mb_cols); + pbi->frame_corrupt_residual = 0; + +#if CONFIG_MULTITHREAD + if (pbi->b_multithreaded_rd && pc->multi_token_partition != ONE_PARTITION) + { + unsigned int thread; + vp8mt_decode_mb_rows(pbi, xd); + vp8_yv12_extend_frame_borders(yv12_fb_new); + for (thread = 0; thread < pbi->decoding_thread_count; ++thread) + corrupt_tokens |= pbi->mb_row_di[thread].mbd.corrupted; + } + else +#endif + { + decode_mb_rows(pbi); + corrupt_tokens |= xd->corrupted; + } + + /* Collect information about decoder corruption. */ + /* 1. Check first boolean decoder for errors. */ + yv12_fb_new->corrupted = vp8dx_bool_error(bc); + /* 2. Check the macroblock information */ + yv12_fb_new->corrupted |= corrupt_tokens; + + if (!pbi->decoded_key_frame) + { + if (pc->frame_type == KEY_FRAME && + !yv12_fb_new->corrupted) + pbi->decoded_key_frame = 1; + else + vpx_internal_error(&pbi->common.error, VPX_CODEC_CORRUPT_FRAME, + "A stream must start with a complete key frame"); + } + + /* vpx_log("Decoder: Frame Decoded, Size Roughly:%d bytes \n",bc->pos+pbi->bc2.pos); */ + + if (pc->refresh_entropy_probs == 0) + { + memcpy(&pc->fc, &pc->lfc, sizeof(pc->fc)); + pbi->independent_partitions = prev_independent_partitions; + } + +#ifdef PACKET_TESTING + { + FILE *f = fopen("decompressor.VP8", "ab"); + unsigned int size = pbi->bc2.pos + pbi->bc.pos + 8; + fwrite((void *) &size, 4, 1, f); + fwrite((void *) pbi->Source, size, 1, f); + fclose(f); + } +#endif + + return 0; +} diff --git a/thirdparty/libvpx/vp8/decoder/decodemv.c b/thirdparty/libvpx/vp8/decoder/decodemv.c new file mode 100644 index 0000000000..1d155e7e16 --- /dev/null +++ b/thirdparty/libvpx/vp8/decoder/decodemv.c @@ -0,0 +1,670 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "treereader.h" +#include "vp8/common/entropymv.h" +#include "vp8/common/entropymode.h" +#include "onyxd_int.h" +#include "vp8/common/findnearmv.h" + +#if CONFIG_DEBUG +#include <assert.h> +#endif +static B_PREDICTION_MODE read_bmode(vp8_reader *bc, const vp8_prob *p) +{ + const int i = vp8_treed_read(bc, vp8_bmode_tree, p); + + return (B_PREDICTION_MODE)i; +} + +static MB_PREDICTION_MODE read_ymode(vp8_reader *bc, const vp8_prob *p) +{ + const int i = vp8_treed_read(bc, vp8_ymode_tree, p); + + return (MB_PREDICTION_MODE)i; +} + +static MB_PREDICTION_MODE read_kf_ymode(vp8_reader *bc, const vp8_prob *p) +{ + const int i = vp8_treed_read(bc, vp8_kf_ymode_tree, p); + + return (MB_PREDICTION_MODE)i; +} + +static MB_PREDICTION_MODE read_uv_mode(vp8_reader *bc, const vp8_prob *p) +{ + const int i = vp8_treed_read(bc, vp8_uv_mode_tree, p); + + return (MB_PREDICTION_MODE)i; +} + +static void read_kf_modes(VP8D_COMP *pbi, MODE_INFO *mi) +{ + vp8_reader *const bc = & pbi->mbc[8]; + const int mis = pbi->common.mode_info_stride; + + mi->mbmi.ref_frame = INTRA_FRAME; + mi->mbmi.mode = read_kf_ymode(bc, vp8_kf_ymode_prob); + + if (mi->mbmi.mode == B_PRED) + { + int i = 0; + mi->mbmi.is_4x4 = 1; + + do + { + const B_PREDICTION_MODE A = above_block_mode(mi, i, mis); + const B_PREDICTION_MODE L = left_block_mode(mi, i); + + mi->bmi[i].as_mode = + read_bmode(bc, vp8_kf_bmode_prob [A] [L]); + } + while (++i < 16); + } + + mi->mbmi.uv_mode = read_uv_mode(bc, vp8_kf_uv_mode_prob); +} + +static int read_mvcomponent(vp8_reader *r, const MV_CONTEXT *mvc) +{ + const vp8_prob *const p = (const vp8_prob *) mvc; + int x = 0; + + if (vp8_read(r, p [mvpis_short])) /* Large */ + { + int i = 0; + + do + { + x += vp8_read(r, p [MVPbits + i]) << i; + } + while (++i < 3); + + i = mvlong_width - 1; /* Skip bit 3, which is sometimes implicit */ + + do + { + x += vp8_read(r, p [MVPbits + i]) << i; + } + while (--i > 3); + + if (!(x & 0xFFF0) || vp8_read(r, p [MVPbits + 3])) + x += 8; + } + else /* small */ + x = vp8_treed_read(r, vp8_small_mvtree, p + MVPshort); + + if (x && vp8_read(r, p [MVPsign])) + x = -x; + + return x; +} + +static void read_mv(vp8_reader *r, MV *mv, const MV_CONTEXT *mvc) +{ + mv->row = (short)(read_mvcomponent(r, mvc) * 2); + mv->col = (short)(read_mvcomponent(r, ++mvc) * 2); +} + + +static void read_mvcontexts(vp8_reader *bc, MV_CONTEXT *mvc) +{ + int i = 0; + + do + { + const vp8_prob *up = vp8_mv_update_probs[i].prob; + vp8_prob *p = (vp8_prob *)(mvc + i); + vp8_prob *const pstop = p + MVPcount; + + do + { + if (vp8_read(bc, *up++)) + { + const vp8_prob x = (vp8_prob)vp8_read_literal(bc, 7); + + *p = x ? x << 1 : 1; + } + } + while (++p < pstop); + } + while (++i < 2); +} + +static const unsigned char mbsplit_fill_count[4] = {8, 8, 4, 1}; +static const unsigned char mbsplit_fill_offset[4][16] = { + { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15}, + { 0, 1, 4, 5, 8, 9, 12, 13, 2, 3, 6, 7, 10, 11, 14, 15}, + { 0, 1, 4, 5, 2, 3, 6, 7, 8, 9, 12, 13, 10, 11, 14, 15}, + { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15} +}; + + +static void mb_mode_mv_init(VP8D_COMP *pbi) +{ + vp8_reader *const bc = & pbi->mbc[8]; + MV_CONTEXT *const mvc = pbi->common.fc.mvc; + +#if CONFIG_ERROR_CONCEALMENT + /* Default is that no macroblock is corrupt, therefore we initialize + * mvs_corrupt_from_mb to something very big, which we can be sure is + * outside the frame. */ + pbi->mvs_corrupt_from_mb = UINT_MAX; +#endif + /* Read the mb_no_coeff_skip flag */ + pbi->common.mb_no_coeff_skip = (int)vp8_read_bit(bc); + + pbi->prob_skip_false = 0; + if (pbi->common.mb_no_coeff_skip) + pbi->prob_skip_false = (vp8_prob)vp8_read_literal(bc, 8); + + if(pbi->common.frame_type != KEY_FRAME) + { + pbi->prob_intra = (vp8_prob)vp8_read_literal(bc, 8); + pbi->prob_last = (vp8_prob)vp8_read_literal(bc, 8); + pbi->prob_gf = (vp8_prob)vp8_read_literal(bc, 8); + + if (vp8_read_bit(bc)) + { + int i = 0; + + do + { + pbi->common.fc.ymode_prob[i] = + (vp8_prob) vp8_read_literal(bc, 8); + } + while (++i < 4); + } + + if (vp8_read_bit(bc)) + { + int i = 0; + + do + { + pbi->common.fc.uv_mode_prob[i] = + (vp8_prob) vp8_read_literal(bc, 8); + } + while (++i < 3); + } + + read_mvcontexts(bc, mvc); + } +} + +const vp8_prob vp8_sub_mv_ref_prob3 [8][VP8_SUBMVREFS-1] = +{ + { 147, 136, 18 }, /* SUBMVREF_NORMAL */ + { 223, 1 , 34 }, /* SUBMVREF_LEFT_ABOVE_SAME */ + { 106, 145, 1 }, /* SUBMVREF_LEFT_ZED */ + { 208, 1 , 1 }, /* SUBMVREF_LEFT_ABOVE_ZED */ + { 179, 121, 1 }, /* SUBMVREF_ABOVE_ZED */ + { 223, 1 , 34 }, /* SUBMVREF_LEFT_ABOVE_SAME */ + { 179, 121, 1 }, /* SUBMVREF_ABOVE_ZED */ + { 208, 1 , 1 } /* SUBMVREF_LEFT_ABOVE_ZED */ +}; + +static +const vp8_prob * get_sub_mv_ref_prob(const int left, const int above) +{ + int lez = (left == 0); + int aez = (above == 0); + int lea = (left == above); + const vp8_prob * prob; + + prob = vp8_sub_mv_ref_prob3[(aez << 2) | + (lez << 1) | + (lea)]; + + return prob; +} + +static void decode_split_mv(vp8_reader *const bc, MODE_INFO *mi, + const MODE_INFO *left_mb, const MODE_INFO *above_mb, + MB_MODE_INFO *mbmi, int_mv best_mv, + MV_CONTEXT *const mvc, int mb_to_left_edge, + int mb_to_right_edge, int mb_to_top_edge, + int mb_to_bottom_edge) +{ + int s; /* split configuration (16x8, 8x16, 8x8, 4x4) */ + int num_p; /* number of partitions in the split configuration + (see vp8_mbsplit_count) */ + int j = 0; + + s = 3; + num_p = 16; + if( vp8_read(bc, 110) ) + { + s = 2; + num_p = 4; + if( vp8_read(bc, 111) ) + { + s = vp8_read(bc, 150); + num_p = 2; + } + } + + do /* for each subset j */ + { + int_mv leftmv, abovemv; + int_mv blockmv; + int k; /* first block in subset j */ + + const vp8_prob *prob; + k = vp8_mbsplit_offset[s][j]; + + if (!(k & 3)) + { + /* On L edge, get from MB to left of us */ + if(left_mb->mbmi.mode != SPLITMV) + leftmv.as_int = left_mb->mbmi.mv.as_int; + else + leftmv.as_int = (left_mb->bmi + k + 4 - 1)->mv.as_int; + } + else + leftmv.as_int = (mi->bmi + k - 1)->mv.as_int; + + if (!(k >> 2)) + { + /* On top edge, get from MB above us */ + if(above_mb->mbmi.mode != SPLITMV) + abovemv.as_int = above_mb->mbmi.mv.as_int; + else + abovemv.as_int = (above_mb->bmi + k + 16 - 4)->mv.as_int; + } + else + abovemv.as_int = (mi->bmi + k - 4)->mv.as_int; + + prob = get_sub_mv_ref_prob(leftmv.as_int, abovemv.as_int); + + if( vp8_read(bc, prob[0]) ) + { + if( vp8_read(bc, prob[1]) ) + { + blockmv.as_int = 0; + if( vp8_read(bc, prob[2]) ) + { + blockmv.as_mv.row = read_mvcomponent(bc, &mvc[0]) * 2; + blockmv.as_mv.row += best_mv.as_mv.row; + blockmv.as_mv.col = read_mvcomponent(bc, &mvc[1]) * 2; + blockmv.as_mv.col += best_mv.as_mv.col; + } + } + else + { + blockmv.as_int = abovemv.as_int; + } + } + else + { + blockmv.as_int = leftmv.as_int; + } + + mbmi->need_to_clamp_mvs |= vp8_check_mv_bounds(&blockmv, + mb_to_left_edge, + mb_to_right_edge, + mb_to_top_edge, + mb_to_bottom_edge); + + { + /* Fill (uniform) modes, mvs of jth subset. + Must do it here because ensuing subsets can + refer back to us via "left" or "above". */ + const unsigned char *fill_offset; + unsigned int fill_count = mbsplit_fill_count[s]; + + fill_offset = &mbsplit_fill_offset[s] + [(unsigned char)j * mbsplit_fill_count[s]]; + + do { + mi->bmi[ *fill_offset].mv.as_int = blockmv.as_int; + fill_offset++; + }while (--fill_count); + } + + } + while (++j < num_p); + + mbmi->partitioning = s; +} + +static void read_mb_modes_mv(VP8D_COMP *pbi, MODE_INFO *mi, MB_MODE_INFO *mbmi) +{ + vp8_reader *const bc = & pbi->mbc[8]; + mbmi->ref_frame = (MV_REFERENCE_FRAME) vp8_read(bc, pbi->prob_intra); + if (mbmi->ref_frame) /* inter MB */ + { + enum {CNT_INTRA, CNT_NEAREST, CNT_NEAR, CNT_SPLITMV}; + int cnt[4]; + int *cntx = cnt; + int_mv near_mvs[4]; + int_mv *nmv = near_mvs; + const int mis = pbi->mb.mode_info_stride; + const MODE_INFO *above = mi - mis; + const MODE_INFO *left = mi - 1; + const MODE_INFO *aboveleft = above - 1; + int *ref_frame_sign_bias = pbi->common.ref_frame_sign_bias; + + mbmi->need_to_clamp_mvs = 0; + + if (vp8_read(bc, pbi->prob_last)) + { + mbmi->ref_frame = + (MV_REFERENCE_FRAME)((int)(2 + vp8_read(bc, pbi->prob_gf))); + } + + /* Zero accumulators */ + nmv[0].as_int = nmv[1].as_int = nmv[2].as_int = 0; + cnt[0] = cnt[1] = cnt[2] = cnt[3] = 0; + + /* Process above */ + if (above->mbmi.ref_frame != INTRA_FRAME) + { + if (above->mbmi.mv.as_int) + { + (++nmv)->as_int = above->mbmi.mv.as_int; + mv_bias(ref_frame_sign_bias[above->mbmi.ref_frame], + mbmi->ref_frame, nmv, ref_frame_sign_bias); + ++cntx; + } + + *cntx += 2; + } + + /* Process left */ + if (left->mbmi.ref_frame != INTRA_FRAME) + { + if (left->mbmi.mv.as_int) + { + int_mv this_mv; + + this_mv.as_int = left->mbmi.mv.as_int; + mv_bias(ref_frame_sign_bias[left->mbmi.ref_frame], + mbmi->ref_frame, &this_mv, ref_frame_sign_bias); + + if (this_mv.as_int != nmv->as_int) + { + (++nmv)->as_int = this_mv.as_int; + ++cntx; + } + + *cntx += 2; + } + else + cnt[CNT_INTRA] += 2; + } + + /* Process above left */ + if (aboveleft->mbmi.ref_frame != INTRA_FRAME) + { + if (aboveleft->mbmi.mv.as_int) + { + int_mv this_mv; + + this_mv.as_int = aboveleft->mbmi.mv.as_int; + mv_bias(ref_frame_sign_bias[aboveleft->mbmi.ref_frame], + mbmi->ref_frame, &this_mv, ref_frame_sign_bias); + + if (this_mv.as_int != nmv->as_int) + { + (++nmv)->as_int = this_mv.as_int; + ++cntx; + } + + *cntx += 1; + } + else + cnt[CNT_INTRA] += 1; + } + + if( vp8_read(bc, vp8_mode_contexts [cnt[CNT_INTRA]] [0]) ) + { + + /* If we have three distinct MV's ... */ + /* See if above-left MV can be merged with NEAREST */ + cnt[CNT_NEAREST] += ( (cnt[CNT_SPLITMV] > 0) & + (nmv->as_int == near_mvs[CNT_NEAREST].as_int)); + + /* Swap near and nearest if necessary */ + if (cnt[CNT_NEAR] > cnt[CNT_NEAREST]) + { + int tmp; + tmp = cnt[CNT_NEAREST]; + cnt[CNT_NEAREST] = cnt[CNT_NEAR]; + cnt[CNT_NEAR] = tmp; + tmp = near_mvs[CNT_NEAREST].as_int; + near_mvs[CNT_NEAREST].as_int = near_mvs[CNT_NEAR].as_int; + near_mvs[CNT_NEAR].as_int = tmp; + } + + if( vp8_read(bc, vp8_mode_contexts [cnt[CNT_NEAREST]] [1]) ) + { + + if( vp8_read(bc, vp8_mode_contexts [cnt[CNT_NEAR]] [2]) ) + { + int mb_to_top_edge; + int mb_to_bottom_edge; + int mb_to_left_edge; + int mb_to_right_edge; + MV_CONTEXT *const mvc = pbi->common.fc.mvc; + int near_index; + + mb_to_top_edge = pbi->mb.mb_to_top_edge; + mb_to_bottom_edge = pbi->mb.mb_to_bottom_edge; + mb_to_top_edge -= LEFT_TOP_MARGIN; + mb_to_bottom_edge += RIGHT_BOTTOM_MARGIN; + mb_to_right_edge = pbi->mb.mb_to_right_edge; + mb_to_right_edge += RIGHT_BOTTOM_MARGIN; + mb_to_left_edge = pbi->mb.mb_to_left_edge; + mb_to_left_edge -= LEFT_TOP_MARGIN; + + /* Use near_mvs[0] to store the "best" MV */ + near_index = CNT_INTRA + + (cnt[CNT_NEAREST] >= cnt[CNT_INTRA]); + + vp8_clamp_mv2(&near_mvs[near_index], &pbi->mb); + + cnt[CNT_SPLITMV] = ((above->mbmi.mode == SPLITMV) + + (left->mbmi.mode == SPLITMV)) * 2 + + (aboveleft->mbmi.mode == SPLITMV); + + if( vp8_read(bc, vp8_mode_contexts [cnt[CNT_SPLITMV]] [3]) ) + { + decode_split_mv(bc, mi, left, above, + mbmi, + near_mvs[near_index], + mvc, mb_to_left_edge, + mb_to_right_edge, + mb_to_top_edge, + mb_to_bottom_edge); + mbmi->mv.as_int = mi->bmi[15].mv.as_int; + mbmi->mode = SPLITMV; + mbmi->is_4x4 = 1; + } + else + { + int_mv *const mbmi_mv = & mbmi->mv; + read_mv(bc, &mbmi_mv->as_mv, (const MV_CONTEXT *) mvc); + mbmi_mv->as_mv.row += near_mvs[near_index].as_mv.row; + mbmi_mv->as_mv.col += near_mvs[near_index].as_mv.col; + + /* Don't need to check this on NEARMV and NEARESTMV + * modes since those modes clamp the MV. The NEWMV mode + * does not, so signal to the prediction stage whether + * special handling may be required. + */ + mbmi->need_to_clamp_mvs = + vp8_check_mv_bounds(mbmi_mv, mb_to_left_edge, + mb_to_right_edge, + mb_to_top_edge, + mb_to_bottom_edge); + mbmi->mode = NEWMV; + } + } + else + { + mbmi->mode = NEARMV; + mbmi->mv.as_int = near_mvs[CNT_NEAR].as_int; + vp8_clamp_mv2(&mbmi->mv, &pbi->mb); + } + } + else + { + mbmi->mode = NEARESTMV; + mbmi->mv.as_int = near_mvs[CNT_NEAREST].as_int; + vp8_clamp_mv2(&mbmi->mv, &pbi->mb); + } + } + else + { + mbmi->mode = ZEROMV; + mbmi->mv.as_int = 0; + } + +#if CONFIG_ERROR_CONCEALMENT + if(pbi->ec_enabled && (mbmi->mode != SPLITMV)) + { + mi->bmi[ 0].mv.as_int = + mi->bmi[ 1].mv.as_int = + mi->bmi[ 2].mv.as_int = + mi->bmi[ 3].mv.as_int = + mi->bmi[ 4].mv.as_int = + mi->bmi[ 5].mv.as_int = + mi->bmi[ 6].mv.as_int = + mi->bmi[ 7].mv.as_int = + mi->bmi[ 8].mv.as_int = + mi->bmi[ 9].mv.as_int = + mi->bmi[10].mv.as_int = + mi->bmi[11].mv.as_int = + mi->bmi[12].mv.as_int = + mi->bmi[13].mv.as_int = + mi->bmi[14].mv.as_int = + mi->bmi[15].mv.as_int = mbmi->mv.as_int; + } +#endif + } + else + { + /* required for left and above block mv */ + mbmi->mv.as_int = 0; + + /* MB is intra coded */ + if ((mbmi->mode = read_ymode(bc, pbi->common.fc.ymode_prob)) == B_PRED) + { + int j = 0; + mbmi->is_4x4 = 1; + do + { + mi->bmi[j].as_mode = read_bmode(bc, pbi->common.fc.bmode_prob); + } + while (++j < 16); + } + + mbmi->uv_mode = read_uv_mode(bc, pbi->common.fc.uv_mode_prob); + } + +} + +static void read_mb_features(vp8_reader *r, MB_MODE_INFO *mi, MACROBLOCKD *x) +{ + /* Is segmentation enabled */ + if (x->segmentation_enabled && x->update_mb_segmentation_map) + { + /* If so then read the segment id. */ + if (vp8_read(r, x->mb_segment_tree_probs[0])) + mi->segment_id = + (unsigned char)(2 + vp8_read(r, x->mb_segment_tree_probs[2])); + else + mi->segment_id = + (unsigned char)(vp8_read(r, x->mb_segment_tree_probs[1])); + } +} + +static void decode_mb_mode_mvs(VP8D_COMP *pbi, MODE_INFO *mi, + MB_MODE_INFO *mbmi) +{ + (void)mbmi; + + /* Read the Macroblock segmentation map if it is being updated explicitly + * this frame (reset to 0 above by default) + * By default on a key frame reset all MBs to segment 0 + */ + if (pbi->mb.update_mb_segmentation_map) + read_mb_features(&pbi->mbc[8], &mi->mbmi, &pbi->mb); + else if(pbi->common.frame_type == KEY_FRAME) + mi->mbmi.segment_id = 0; + + /* Read the macroblock coeff skip flag if this feature is in use, + * else default to 0 */ + if (pbi->common.mb_no_coeff_skip) + mi->mbmi.mb_skip_coeff = vp8_read(&pbi->mbc[8], pbi->prob_skip_false); + else + mi->mbmi.mb_skip_coeff = 0; + + mi->mbmi.is_4x4 = 0; + if(pbi->common.frame_type == KEY_FRAME) + read_kf_modes(pbi, mi); + else + read_mb_modes_mv(pbi, mi, &mi->mbmi); + +} + +void vp8_decode_mode_mvs(VP8D_COMP *pbi) +{ + MODE_INFO *mi = pbi->common.mi; + int mb_row = -1; + int mb_to_right_edge_start; + + mb_mode_mv_init(pbi); + + pbi->mb.mb_to_top_edge = 0; + pbi->mb.mb_to_bottom_edge = ((pbi->common.mb_rows - 1) * 16) << 3; + mb_to_right_edge_start = ((pbi->common.mb_cols - 1) * 16) << 3; + + while (++mb_row < pbi->common.mb_rows) + { + int mb_col = -1; + + pbi->mb.mb_to_left_edge = 0; + pbi->mb.mb_to_right_edge = mb_to_right_edge_start; + + while (++mb_col < pbi->common.mb_cols) + { +#if CONFIG_ERROR_CONCEALMENT + int mb_num = mb_row * pbi->common.mb_cols + mb_col; +#endif + + decode_mb_mode_mvs(pbi, mi, &mi->mbmi); + +#if CONFIG_ERROR_CONCEALMENT + /* look for corruption. set mvs_corrupt_from_mb to the current + * mb_num if the frame is corrupt from this macroblock. */ + if (vp8dx_bool_error(&pbi->mbc[8]) && mb_num < + (int)pbi->mvs_corrupt_from_mb) + { + pbi->mvs_corrupt_from_mb = mb_num; + /* no need to continue since the partition is corrupt from + * here on. + */ + return; + } +#endif + + pbi->mb.mb_to_left_edge -= (16 << 3); + pbi->mb.mb_to_right_edge -= (16 << 3); + mi++; /* next macroblock */ + } + pbi->mb.mb_to_top_edge -= (16 << 3); + pbi->mb.mb_to_bottom_edge -= (16 << 3); + + mi++; /* skip left predictor each row */ + } +} diff --git a/thirdparty/libvpx/vp8/decoder/decodemv.h b/thirdparty/libvpx/vp8/decoder/decodemv.h new file mode 100644 index 0000000000..f33b07351d --- /dev/null +++ b/thirdparty/libvpx/vp8/decoder/decodemv.h @@ -0,0 +1,26 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP8_DECODER_DECODEMV_H_ +#define VP8_DECODER_DECODEMV_H_ + +#include "onyxd_int.h" + +#ifdef __cplusplus +extern "C" { +#endif + +void vp8_decode_mode_mvs(VP8D_COMP *); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_DECODER_DECODEMV_H_ diff --git a/thirdparty/libvpx/vp8/decoder/decoderthreading.h b/thirdparty/libvpx/vp8/decoder/decoderthreading.h new file mode 100644 index 0000000000..c563cf6e93 --- /dev/null +++ b/thirdparty/libvpx/vp8/decoder/decoderthreading.h @@ -0,0 +1,30 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP8_DECODER_DECODERTHREADING_H_ +#define VP8_DECODER_DECODERTHREADING_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +#if CONFIG_MULTITHREAD +void vp8mt_decode_mb_rows(VP8D_COMP *pbi, MACROBLOCKD *xd); +void vp8_decoder_remove_threads(VP8D_COMP *pbi); +void vp8_decoder_create_threads(VP8D_COMP *pbi); +void vp8mt_alloc_temp_buffers(VP8D_COMP *pbi, int width, int prev_mb_rows); +void vp8mt_de_alloc_temp_buffers(VP8D_COMP *pbi, int mb_rows); +#endif + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_DECODER_DECODERTHREADING_H_ diff --git a/thirdparty/libvpx/vp8/decoder/detokenize.c b/thirdparty/libvpx/vp8/decoder/detokenize.c new file mode 100644 index 0000000000..fcc7533c50 --- /dev/null +++ b/thirdparty/libvpx/vp8/decoder/detokenize.c @@ -0,0 +1,245 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "vp8/common/blockd.h" +#include "onyxd_int.h" +#include "vpx_mem/vpx_mem.h" +#include "vpx_ports/mem.h" +#include "detokenize.h" + +void vp8_reset_mb_tokens_context(MACROBLOCKD *x) +{ + ENTROPY_CONTEXT *a_ctx = ((ENTROPY_CONTEXT *)x->above_context); + ENTROPY_CONTEXT *l_ctx = ((ENTROPY_CONTEXT *)x->left_context); + + memset(a_ctx, 0, sizeof(ENTROPY_CONTEXT_PLANES)-1); + memset(l_ctx, 0, sizeof(ENTROPY_CONTEXT_PLANES)-1); + + /* Clear entropy contexts for Y2 blocks */ + if (!x->mode_info_context->mbmi.is_4x4) + { + a_ctx[8] = l_ctx[8] = 0; + } +} + +/* + ------------------------------------------------------------------------------ + Residual decoding (Paragraph 13.2 / 13.3) +*/ +static const uint8_t kBands[16 + 1] = { + 0, 1, 2, 3, 6, 4, 5, 6, 6, 6, 6, 6, 6, 6, 6, 7, + 0 /* extra entry as sentinel */ +}; + +static const uint8_t kCat3[] = { 173, 148, 140, 0 }; +static const uint8_t kCat4[] = { 176, 155, 140, 135, 0 }; +static const uint8_t kCat5[] = { 180, 157, 141, 134, 130, 0 }; +static const uint8_t kCat6[] = + { 254, 254, 243, 230, 196, 177, 153, 140, 133, 130, 129, 0 }; +static const uint8_t* const kCat3456[] = { kCat3, kCat4, kCat5, kCat6 }; +static const uint8_t kZigzag[16] = { + 0, 1, 4, 8, 5, 2, 3, 6, 9, 12, 13, 10, 7, 11, 14, 15 +}; + +#define VP8GetBit vp8dx_decode_bool +#define NUM_PROBAS 11 +#define NUM_CTX 3 + +/* for const-casting */ +typedef const uint8_t (*ProbaArray)[NUM_CTX][NUM_PROBAS]; + +static int GetSigned(BOOL_DECODER *br, int value_to_sign) +{ + int split = (br->range + 1) >> 1; + VP8_BD_VALUE bigsplit = (VP8_BD_VALUE)split << (VP8_BD_VALUE_SIZE - 8); + int v; + + if(br->count < 0) + vp8dx_bool_decoder_fill(br); + + if ( br->value < bigsplit ) + { + br->range = split; + v= value_to_sign; + } + else + { + br->range = br->range-split; + br->value = br->value-bigsplit; + v = -value_to_sign; + } + br->range +=br->range; + br->value +=br->value; + br->count--; + + return v; +} +/* + Returns the position of the last non-zero coeff plus one + (and 0 if there's no coeff at all) +*/ +static int GetCoeffs(BOOL_DECODER *br, ProbaArray prob, + int ctx, int n, int16_t* out) +{ + const uint8_t* p = prob[n][ctx]; + if (!VP8GetBit(br, p[0])) + { /* first EOB is more a 'CBP' bit. */ + return 0; + } + while (1) + { + ++n; + if (!VP8GetBit(br, p[1])) + { + p = prob[kBands[n]][0]; + } + else + { /* non zero coeff */ + int v, j; + if (!VP8GetBit(br, p[2])) + { + p = prob[kBands[n]][1]; + v = 1; + } + else + { + if (!VP8GetBit(br, p[3])) + { + if (!VP8GetBit(br, p[4])) + { + v = 2; + } + else + { + v = 3 + VP8GetBit(br, p[5]); + } + } + else + { + if (!VP8GetBit(br, p[6])) + { + if (!VP8GetBit(br, p[7])) + { + v = 5 + VP8GetBit(br, 159); + } else + { + v = 7 + 2 * VP8GetBit(br, 165); + v += VP8GetBit(br, 145); + } + } + else + { + const uint8_t* tab; + const int bit1 = VP8GetBit(br, p[8]); + const int bit0 = VP8GetBit(br, p[9 + bit1]); + const int cat = 2 * bit1 + bit0; + v = 0; + for (tab = kCat3456[cat]; *tab; ++tab) + { + v += v + VP8GetBit(br, *tab); + } + v += 3 + (8 << cat); + } + } + p = prob[kBands[n]][2]; + } + j = kZigzag[n - 1]; + + out[j] = GetSigned(br, v); + + if (n == 16 || !VP8GetBit(br, p[0])) + { /* EOB */ + return n; + } + } + if (n == 16) + { + return 16; + } + } +} + +int vp8_decode_mb_tokens(VP8D_COMP *dx, MACROBLOCKD *x) +{ + BOOL_DECODER *bc = x->current_bc; + const FRAME_CONTEXT * const fc = &dx->common.fc; + char *eobs = x->eobs; + + int i; + int nonzeros; + int eobtotal = 0; + + short *qcoeff_ptr; + ProbaArray coef_probs; + ENTROPY_CONTEXT *a_ctx = ((ENTROPY_CONTEXT *)x->above_context); + ENTROPY_CONTEXT *l_ctx = ((ENTROPY_CONTEXT *)x->left_context); + ENTROPY_CONTEXT *a; + ENTROPY_CONTEXT *l; + int skip_dc = 0; + + qcoeff_ptr = &x->qcoeff[0]; + + if (!x->mode_info_context->mbmi.is_4x4) + { + a = a_ctx + 8; + l = l_ctx + 8; + + coef_probs = fc->coef_probs [1]; + + nonzeros = GetCoeffs(bc, coef_probs, (*a + *l), 0, qcoeff_ptr + 24 * 16); + *a = *l = (nonzeros > 0); + + eobs[24] = nonzeros; + eobtotal += nonzeros - 16; + + coef_probs = fc->coef_probs [0]; + skip_dc = 1; + } + else + { + coef_probs = fc->coef_probs [3]; + skip_dc = 0; + } + + for (i = 0; i < 16; ++i) + { + a = a_ctx + (i&3); + l = l_ctx + ((i&0xc)>>2); + + nonzeros = GetCoeffs(bc, coef_probs, (*a + *l), skip_dc, qcoeff_ptr); + *a = *l = (nonzeros > 0); + + nonzeros += skip_dc; + eobs[i] = nonzeros; + eobtotal += nonzeros; + qcoeff_ptr += 16; + } + + coef_probs = fc->coef_probs [2]; + + a_ctx += 4; + l_ctx += 4; + for (i = 16; i < 24; ++i) + { + a = a_ctx + ((i > 19)<<1) + (i&1); + l = l_ctx + ((i > 19)<<1) + ((i&3)>1); + + nonzeros = GetCoeffs(bc, coef_probs, (*a + *l), 0, qcoeff_ptr); + *a = *l = (nonzeros > 0); + + eobs[i] = nonzeros; + eobtotal += nonzeros; + qcoeff_ptr += 16; + } + + return eobtotal; +} + diff --git a/thirdparty/libvpx/vp8/decoder/detokenize.h b/thirdparty/libvpx/vp8/decoder/detokenize.h new file mode 100644 index 0000000000..f0b125444f --- /dev/null +++ b/thirdparty/libvpx/vp8/decoder/detokenize.h @@ -0,0 +1,27 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP8_DECODER_DETOKENIZE_H_ +#define VP8_DECODER_DETOKENIZE_H_ + +#include "onyxd_int.h" + +#ifdef __cplusplus +extern "C" { +#endif + +void vp8_reset_mb_tokens_context(MACROBLOCKD *x); +int vp8_decode_mb_tokens(VP8D_COMP *, MACROBLOCKD *); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_DECODER_DETOKENIZE_H_ diff --git a/thirdparty/libvpx/vp8/decoder/onyxd_if.c b/thirdparty/libvpx/vp8/decoder/onyxd_if.c new file mode 100644 index 0000000000..3468268a2a --- /dev/null +++ b/thirdparty/libvpx/vp8/decoder/onyxd_if.c @@ -0,0 +1,521 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "vp8/common/onyxc_int.h" +#if CONFIG_POSTPROC +#include "vp8/common/postproc.h" +#endif +#include "vp8/common/onyxd.h" +#include "onyxd_int.h" +#include "vpx_mem/vpx_mem.h" +#include "vp8/common/alloccommon.h" +#include "vp8/common/loopfilter.h" +#include "vp8/common/swapyv12buffer.h" +#include "vp8/common/threading.h" +#include "decoderthreading.h" +#include <stdio.h> +#include <assert.h> + +#include "vp8/common/quant_common.h" +#include "vp8/common/reconintra.h" +#include "./vpx_dsp_rtcd.h" +#include "./vpx_scale_rtcd.h" +#include "vpx_scale/vpx_scale.h" +#include "vp8/common/systemdependent.h" +#include "vpx_ports/vpx_once.h" +#include "vpx_ports/vpx_timer.h" +#include "detokenize.h" +#if CONFIG_ERROR_CONCEALMENT +#include "error_concealment.h" +#endif +#if ARCH_ARM +#include "vpx_ports/arm.h" +#endif + +extern void vp8_init_loop_filter(VP8_COMMON *cm); +extern void vp8cx_init_de_quantizer(VP8D_COMP *pbi); +static int get_free_fb (VP8_COMMON *cm); +static void ref_cnt_fb (int *buf, int *idx, int new_idx); + +static void initialize_dec(void) { + static volatile int init_done = 0; + + if (!init_done) + { + vpx_dsp_rtcd(); + vp8_init_intra_predictors(); + init_done = 1; + } +} + +static void remove_decompressor(VP8D_COMP *pbi) +{ +#if CONFIG_ERROR_CONCEALMENT + vp8_de_alloc_overlap_lists(pbi); +#endif + vp8_remove_common(&pbi->common); + vpx_free(pbi); +} + +static struct VP8D_COMP * create_decompressor(VP8D_CONFIG *oxcf) +{ + VP8D_COMP *pbi = vpx_memalign(32, sizeof(VP8D_COMP)); + + if (!pbi) + return NULL; + + memset(pbi, 0, sizeof(VP8D_COMP)); + + if (setjmp(pbi->common.error.jmp)) + { + pbi->common.error.setjmp = 0; + remove_decompressor(pbi); + return 0; + } + + pbi->common.error.setjmp = 1; + + vp8_create_common(&pbi->common); + + pbi->common.current_video_frame = 0; + pbi->ready_for_new_data = 1; + + /* vp8cx_init_de_quantizer() is first called here. Add check in frame_init_dequantizer() to avoid + * unnecessary calling of vp8cx_init_de_quantizer() for every frame. + */ + vp8cx_init_de_quantizer(pbi); + + vp8_loop_filter_init(&pbi->common); + + pbi->common.error.setjmp = 0; + +#if CONFIG_ERROR_CONCEALMENT + pbi->ec_enabled = oxcf->error_concealment; + pbi->overlaps = NULL; +#else + (void)oxcf; + pbi->ec_enabled = 0; +#endif + /* Error concealment is activated after a key frame has been + * decoded without errors when error concealment is enabled. + */ + pbi->ec_active = 0; + + pbi->decoded_key_frame = 0; + + /* Independent partitions is activated when a frame updates the + * token probability table to have equal probabilities over the + * PREV_COEF context. + */ + pbi->independent_partitions = 0; + + vp8_setup_block_dptrs(&pbi->mb); + + once(initialize_dec); + + return pbi; +} + +vpx_codec_err_t vp8dx_get_reference(VP8D_COMP *pbi, enum vpx_ref_frame_type ref_frame_flag, YV12_BUFFER_CONFIG *sd) +{ + VP8_COMMON *cm = &pbi->common; + int ref_fb_idx; + + if (ref_frame_flag == VP8_LAST_FRAME) + ref_fb_idx = cm->lst_fb_idx; + else if (ref_frame_flag == VP8_GOLD_FRAME) + ref_fb_idx = cm->gld_fb_idx; + else if (ref_frame_flag == VP8_ALTR_FRAME) + ref_fb_idx = cm->alt_fb_idx; + else{ + vpx_internal_error(&pbi->common.error, VPX_CODEC_ERROR, + "Invalid reference frame"); + return pbi->common.error.error_code; + } + + if(cm->yv12_fb[ref_fb_idx].y_height != sd->y_height || + cm->yv12_fb[ref_fb_idx].y_width != sd->y_width || + cm->yv12_fb[ref_fb_idx].uv_height != sd->uv_height || + cm->yv12_fb[ref_fb_idx].uv_width != sd->uv_width){ + vpx_internal_error(&pbi->common.error, VPX_CODEC_ERROR, + "Incorrect buffer dimensions"); + } + else + vp8_yv12_copy_frame(&cm->yv12_fb[ref_fb_idx], sd); + + return pbi->common.error.error_code; +} + + +vpx_codec_err_t vp8dx_set_reference(VP8D_COMP *pbi, enum vpx_ref_frame_type ref_frame_flag, YV12_BUFFER_CONFIG *sd) +{ + VP8_COMMON *cm = &pbi->common; + int *ref_fb_ptr = NULL; + int free_fb; + + if (ref_frame_flag == VP8_LAST_FRAME) + ref_fb_ptr = &cm->lst_fb_idx; + else if (ref_frame_flag == VP8_GOLD_FRAME) + ref_fb_ptr = &cm->gld_fb_idx; + else if (ref_frame_flag == VP8_ALTR_FRAME) + ref_fb_ptr = &cm->alt_fb_idx; + else{ + vpx_internal_error(&pbi->common.error, VPX_CODEC_ERROR, + "Invalid reference frame"); + return pbi->common.error.error_code; + } + + if(cm->yv12_fb[*ref_fb_ptr].y_height != sd->y_height || + cm->yv12_fb[*ref_fb_ptr].y_width != sd->y_width || + cm->yv12_fb[*ref_fb_ptr].uv_height != sd->uv_height || + cm->yv12_fb[*ref_fb_ptr].uv_width != sd->uv_width){ + vpx_internal_error(&pbi->common.error, VPX_CODEC_ERROR, + "Incorrect buffer dimensions"); + } + else{ + /* Find an empty frame buffer. */ + free_fb = get_free_fb(cm); + /* Decrease fb_idx_ref_cnt since it will be increased again in + * ref_cnt_fb() below. */ + cm->fb_idx_ref_cnt[free_fb]--; + + /* Manage the reference counters and copy image. */ + ref_cnt_fb (cm->fb_idx_ref_cnt, ref_fb_ptr, free_fb); + vp8_yv12_copy_frame(sd, &cm->yv12_fb[*ref_fb_ptr]); + } + + return pbi->common.error.error_code; +} + +static int get_free_fb (VP8_COMMON *cm) +{ + int i; + for (i = 0; i < NUM_YV12_BUFFERS; i++) + if (cm->fb_idx_ref_cnt[i] == 0) + break; + + assert(i < NUM_YV12_BUFFERS); + cm->fb_idx_ref_cnt[i] = 1; + return i; +} + +static void ref_cnt_fb (int *buf, int *idx, int new_idx) +{ + if (buf[*idx] > 0) + buf[*idx]--; + + *idx = new_idx; + + buf[new_idx]++; +} + +/* If any buffer copy / swapping is signalled it should be done here. */ +static int swap_frame_buffers (VP8_COMMON *cm) +{ + int err = 0; + + /* The alternate reference frame or golden frame can be updated + * using the new, last, or golden/alt ref frame. If it + * is updated using the newly decoded frame it is a refresh. + * An update using the last or golden/alt ref frame is a copy. + */ + if (cm->copy_buffer_to_arf) + { + int new_fb = 0; + + if (cm->copy_buffer_to_arf == 1) + new_fb = cm->lst_fb_idx; + else if (cm->copy_buffer_to_arf == 2) + new_fb = cm->gld_fb_idx; + else + err = -1; + + ref_cnt_fb (cm->fb_idx_ref_cnt, &cm->alt_fb_idx, new_fb); + } + + if (cm->copy_buffer_to_gf) + { + int new_fb = 0; + + if (cm->copy_buffer_to_gf == 1) + new_fb = cm->lst_fb_idx; + else if (cm->copy_buffer_to_gf == 2) + new_fb = cm->alt_fb_idx; + else + err = -1; + + ref_cnt_fb (cm->fb_idx_ref_cnt, &cm->gld_fb_idx, new_fb); + } + + if (cm->refresh_golden_frame) + ref_cnt_fb (cm->fb_idx_ref_cnt, &cm->gld_fb_idx, cm->new_fb_idx); + + if (cm->refresh_alt_ref_frame) + ref_cnt_fb (cm->fb_idx_ref_cnt, &cm->alt_fb_idx, cm->new_fb_idx); + + if (cm->refresh_last_frame) + { + ref_cnt_fb (cm->fb_idx_ref_cnt, &cm->lst_fb_idx, cm->new_fb_idx); + + cm->frame_to_show = &cm->yv12_fb[cm->lst_fb_idx]; + } + else + cm->frame_to_show = &cm->yv12_fb[cm->new_fb_idx]; + + cm->fb_idx_ref_cnt[cm->new_fb_idx]--; + + return err; +} + +static int check_fragments_for_errors(VP8D_COMP *pbi) +{ + if (!pbi->ec_active && + pbi->fragments.count <= 1 && pbi->fragments.sizes[0] == 0) + { + VP8_COMMON *cm = &pbi->common; + + /* If error concealment is disabled we won't signal missing frames + * to the decoder. + */ + if (cm->fb_idx_ref_cnt[cm->lst_fb_idx] > 1) + { + /* The last reference shares buffer with another reference + * buffer. Move it to its own buffer before setting it as + * corrupt, otherwise we will make multiple buffers corrupt. + */ + const int prev_idx = cm->lst_fb_idx; + cm->fb_idx_ref_cnt[prev_idx]--; + cm->lst_fb_idx = get_free_fb(cm); + vp8_yv12_copy_frame(&cm->yv12_fb[prev_idx], + &cm->yv12_fb[cm->lst_fb_idx]); + } + /* This is used to signal that we are missing frames. + * We do not know if the missing frame(s) was supposed to update + * any of the reference buffers, but we act conservative and + * mark only the last buffer as corrupted. + */ + cm->yv12_fb[cm->lst_fb_idx].corrupted = 1; + + /* Signal that we have no frame to show. */ + cm->show_frame = 0; + + /* Nothing more to do. */ + return 0; + } + + return 1; +} + +int vp8dx_receive_compressed_data(VP8D_COMP *pbi, size_t size, + const uint8_t *source, + int64_t time_stamp) +{ + VP8_COMMON *cm = &pbi->common; + int retcode = -1; + (void)size; + (void)source; + + pbi->common.error.error_code = VPX_CODEC_OK; + + retcode = check_fragments_for_errors(pbi); + if(retcode <= 0) + return retcode; + + cm->new_fb_idx = get_free_fb (cm); + + /* setup reference frames for vp8_decode_frame */ + pbi->dec_fb_ref[INTRA_FRAME] = &cm->yv12_fb[cm->new_fb_idx]; + pbi->dec_fb_ref[LAST_FRAME] = &cm->yv12_fb[cm->lst_fb_idx]; + pbi->dec_fb_ref[GOLDEN_FRAME] = &cm->yv12_fb[cm->gld_fb_idx]; + pbi->dec_fb_ref[ALTREF_FRAME] = &cm->yv12_fb[cm->alt_fb_idx]; + + if (setjmp(pbi->common.error.jmp)) + { + /* We do not know if the missing frame(s) was supposed to update + * any of the reference buffers, but we act conservative and + * mark only the last buffer as corrupted. + */ + cm->yv12_fb[cm->lst_fb_idx].corrupted = 1; + + if (cm->fb_idx_ref_cnt[cm->new_fb_idx] > 0) + cm->fb_idx_ref_cnt[cm->new_fb_idx]--; + + goto decode_exit; + } + + pbi->common.error.setjmp = 1; + + retcode = vp8_decode_frame(pbi); + + if (retcode < 0) + { + if (cm->fb_idx_ref_cnt[cm->new_fb_idx] > 0) + cm->fb_idx_ref_cnt[cm->new_fb_idx]--; + + pbi->common.error.error_code = VPX_CODEC_ERROR; + goto decode_exit; + } + + if (swap_frame_buffers (cm)) + { + pbi->common.error.error_code = VPX_CODEC_ERROR; + goto decode_exit; + } + + vp8_clear_system_state(); + + if (cm->show_frame) + { + cm->current_video_frame++; + cm->show_frame_mi = cm->mi; + } + + #if CONFIG_ERROR_CONCEALMENT + /* swap the mode infos to storage for future error concealment */ + if (pbi->ec_enabled && pbi->common.prev_mi) + { + MODE_INFO* tmp = pbi->common.prev_mi; + int row, col; + pbi->common.prev_mi = pbi->common.mi; + pbi->common.mi = tmp; + + /* Propagate the segment_ids to the next frame */ + for (row = 0; row < pbi->common.mb_rows; ++row) + { + for (col = 0; col < pbi->common.mb_cols; ++col) + { + const int i = row*pbi->common.mode_info_stride + col; + pbi->common.mi[i].mbmi.segment_id = + pbi->common.prev_mi[i].mbmi.segment_id; + } + } + } +#endif + + pbi->ready_for_new_data = 0; + pbi->last_time_stamp = time_stamp; + +decode_exit: + pbi->common.error.setjmp = 0; + vp8_clear_system_state(); + return retcode; +} +int vp8dx_get_raw_frame(VP8D_COMP *pbi, YV12_BUFFER_CONFIG *sd, int64_t *time_stamp, int64_t *time_end_stamp, vp8_ppflags_t *flags) +{ + int ret = -1; + + if (pbi->ready_for_new_data == 1) + return ret; + + /* ie no raw frame to show!!! */ + if (pbi->common.show_frame == 0) + return ret; + + pbi->ready_for_new_data = 1; + *time_stamp = pbi->last_time_stamp; + *time_end_stamp = 0; + +#if CONFIG_POSTPROC + ret = vp8_post_proc_frame(&pbi->common, sd, flags); +#else + (void)flags; + + if (pbi->common.frame_to_show) + { + *sd = *pbi->common.frame_to_show; + sd->y_width = pbi->common.Width; + sd->y_height = pbi->common.Height; + sd->uv_height = pbi->common.Height / 2; + ret = 0; + } + else + { + ret = -1; + } + +#endif /*!CONFIG_POSTPROC*/ + vp8_clear_system_state(); + return ret; +} + + +/* This function as written isn't decoder specific, but the encoder has + * much faster ways of computing this, so it's ok for it to live in a + * decode specific file. + */ +int vp8dx_references_buffer( VP8_COMMON *oci, int ref_frame ) +{ + const MODE_INFO *mi = oci->mi; + int mb_row, mb_col; + + for (mb_row = 0; mb_row < oci->mb_rows; mb_row++) + { + for (mb_col = 0; mb_col < oci->mb_cols; mb_col++,mi++) + { + if( mi->mbmi.ref_frame == ref_frame) + return 1; + } + mi++; + } + return 0; + +} + +int vp8_create_decoder_instances(struct frame_buffers *fb, VP8D_CONFIG *oxcf) +{ + if(!fb->use_frame_threads) + { + /* decoder instance for single thread mode */ + fb->pbi[0] = create_decompressor(oxcf); + if(!fb->pbi[0]) + return VPX_CODEC_ERROR; + +#if CONFIG_MULTITHREAD + /* enable row-based threading only when use_frame_threads + * is disabled */ + fb->pbi[0]->max_threads = oxcf->max_threads; + vp8_decoder_create_threads(fb->pbi[0]); +#endif + } + else + { + /* TODO : create frame threads and decoder instances for each + * thread here */ + } + + return VPX_CODEC_OK; +} + +int vp8_remove_decoder_instances(struct frame_buffers *fb) +{ + if(!fb->use_frame_threads) + { + VP8D_COMP *pbi = fb->pbi[0]; + + if (!pbi) + return VPX_CODEC_ERROR; +#if CONFIG_MULTITHREAD + if (pbi->b_multithreaded_rd) + vp8mt_de_alloc_temp_buffers(pbi, pbi->common.mb_rows); + vp8_decoder_remove_threads(pbi); +#endif + + /* decoder instance for single thread mode */ + remove_decompressor(pbi); + } + else + { + /* TODO : remove frame threads and decoder instances for each + * thread here */ + } + + return VPX_CODEC_OK; +} diff --git a/thirdparty/libvpx/vp8/decoder/onyxd_int.h b/thirdparty/libvpx/vp8/decoder/onyxd_int.h new file mode 100644 index 0000000000..313fe01c07 --- /dev/null +++ b/thirdparty/libvpx/vp8/decoder/onyxd_int.h @@ -0,0 +1,161 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_DECODER_ONYXD_INT_H_ +#define VP8_DECODER_ONYXD_INT_H_ + +#include "vpx_config.h" +#include "vp8/common/onyxd.h" +#include "treereader.h" +#include "vp8/common/onyxc_int.h" +#include "vp8/common/threading.h" + +#if CONFIG_ERROR_CONCEALMENT +#include "ec_types.h" +#endif + +#ifdef __cplusplus +extern "C" { +#endif + +typedef struct +{ + int ithread; + void *ptr1; + void *ptr2; +} DECODETHREAD_DATA; + +typedef struct +{ + MACROBLOCKD mbd; +} MB_ROW_DEC; + + +typedef struct +{ + int enabled; + unsigned int count; + const unsigned char *ptrs[MAX_PARTITIONS]; + unsigned int sizes[MAX_PARTITIONS]; +} FRAGMENT_DATA; + +#define MAX_FB_MT_DEC 32 + +struct frame_buffers +{ + /* + * this struct will be populated with frame buffer management + * info in future commits. */ + + /* enable/disable frame-based threading */ + int use_frame_threads; + + /* decoder instances */ + struct VP8D_COMP *pbi[MAX_FB_MT_DEC]; + +}; + +typedef struct VP8D_COMP +{ + DECLARE_ALIGNED(16, MACROBLOCKD, mb); + + YV12_BUFFER_CONFIG *dec_fb_ref[NUM_YV12_BUFFERS]; + + DECLARE_ALIGNED(16, VP8_COMMON, common); + + /* the last partition will be used for the modes/mvs */ + vp8_reader mbc[MAX_PARTITIONS]; + + VP8D_CONFIG oxcf; + + FRAGMENT_DATA fragments; + +#if CONFIG_MULTITHREAD + /* variable for threading */ + + int b_multithreaded_rd; + int max_threads; + int current_mb_col_main; + unsigned int decoding_thread_count; + int allocated_decoding_thread_count; + + int mt_baseline_filter_level[MAX_MB_SEGMENTS]; + int sync_range; + int *mt_current_mb_col; /* Each row remembers its already decoded column. */ + pthread_mutex_t *pmutex; + pthread_mutex_t mt_mutex; /* mutex for b_multithreaded_rd */ + + unsigned char **mt_yabove_row; /* mb_rows x width */ + unsigned char **mt_uabove_row; + unsigned char **mt_vabove_row; + unsigned char **mt_yleft_col; /* mb_rows x 16 */ + unsigned char **mt_uleft_col; /* mb_rows x 8 */ + unsigned char **mt_vleft_col; /* mb_rows x 8 */ + + MB_ROW_DEC *mb_row_di; + DECODETHREAD_DATA *de_thread_data; + + pthread_t *h_decoding_thread; + sem_t *h_event_start_decoding; + sem_t h_event_end_decoding; + /* end of threading data */ +#endif + + int64_t last_time_stamp; + int ready_for_new_data; + + vp8_prob prob_intra; + vp8_prob prob_last; + vp8_prob prob_gf; + vp8_prob prob_skip_false; + +#if CONFIG_ERROR_CONCEALMENT + MB_OVERLAP *overlaps; + /* the mb num from which modes and mvs (first partition) are corrupt */ + unsigned int mvs_corrupt_from_mb; +#endif + int ec_enabled; + int ec_active; + int decoded_key_frame; + int independent_partitions; + int frame_corrupt_residual; + + vpx_decrypt_cb decrypt_cb; + void *decrypt_state; +} VP8D_COMP; + +int vp8_decode_frame(VP8D_COMP *cpi); + +int vp8_create_decoder_instances(struct frame_buffers *fb, VP8D_CONFIG *oxcf); +int vp8_remove_decoder_instances(struct frame_buffers *fb); + +#if CONFIG_DEBUG +#define CHECK_MEM_ERROR(lval,expr) do {\ + lval = (expr); \ + if(!lval) \ + vpx_internal_error(&pbi->common.error, VPX_CODEC_MEM_ERROR,\ + "Failed to allocate "#lval" at %s:%d", \ + __FILE__,__LINE__);\ + } while(0) +#else +#define CHECK_MEM_ERROR(lval,expr) do {\ + lval = (expr); \ + if(!lval) \ + vpx_internal_error(&pbi->common.error, VPX_CODEC_MEM_ERROR,\ + "Failed to allocate "#lval);\ + } while(0) +#endif + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_DECODER_ONYXD_INT_H_ diff --git a/thirdparty/libvpx/vp8/decoder/threading.c b/thirdparty/libvpx/vp8/decoder/threading.c new file mode 100644 index 0000000000..3c1b8387ec --- /dev/null +++ b/thirdparty/libvpx/vp8/decoder/threading.c @@ -0,0 +1,928 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "vpx_config.h" +#include "vp8_rtcd.h" +#if !defined(WIN32) && CONFIG_OS_SUPPORT == 1 +# include <unistd.h> +#endif +#include "onyxd_int.h" +#include "vpx_mem/vpx_mem.h" +#include "vp8/common/threading.h" + +#include "vp8/common/loopfilter.h" +#include "vp8/common/extend.h" +#include "vpx_ports/vpx_timer.h" +#include "detokenize.h" +#include "vp8/common/reconintra4x4.h" +#include "vp8/common/reconinter.h" +#include "vp8/common/reconintra.h" +#include "vp8/common/setupintrarecon.h" +#if CONFIG_ERROR_CONCEALMENT +#include "error_concealment.h" +#endif + +#define CALLOC_ARRAY(p, n) CHECK_MEM_ERROR((p), vpx_calloc(sizeof(*(p)), (n))) +#define CALLOC_ARRAY_ALIGNED(p, n, algn) do { \ + CHECK_MEM_ERROR((p), vpx_memalign((algn), sizeof(*(p)) * (n))); \ + memset((p), 0, (n) * sizeof(*(p))); \ +} while (0) + + +void vp8_mb_init_dequantizer(VP8D_COMP *pbi, MACROBLOCKD *xd); + +static void setup_decoding_thread_data(VP8D_COMP *pbi, MACROBLOCKD *xd, MB_ROW_DEC *mbrd, int count) +{ + VP8_COMMON *const pc = & pbi->common; + int i; + + for (i = 0; i < count; i++) + { + MACROBLOCKD *mbd = &mbrd[i].mbd; + mbd->subpixel_predict = xd->subpixel_predict; + mbd->subpixel_predict8x4 = xd->subpixel_predict8x4; + mbd->subpixel_predict8x8 = xd->subpixel_predict8x8; + mbd->subpixel_predict16x16 = xd->subpixel_predict16x16; + + mbd->frame_type = pc->frame_type; + mbd->pre = xd->pre; + mbd->dst = xd->dst; + + mbd->segmentation_enabled = xd->segmentation_enabled; + mbd->mb_segement_abs_delta = xd->mb_segement_abs_delta; + memcpy(mbd->segment_feature_data, xd->segment_feature_data, sizeof(xd->segment_feature_data)); + + /*signed char ref_lf_deltas[MAX_REF_LF_DELTAS];*/ + memcpy(mbd->ref_lf_deltas, xd->ref_lf_deltas, sizeof(xd->ref_lf_deltas)); + /*signed char mode_lf_deltas[MAX_MODE_LF_DELTAS];*/ + memcpy(mbd->mode_lf_deltas, xd->mode_lf_deltas, sizeof(xd->mode_lf_deltas)); + /*unsigned char mode_ref_lf_delta_enabled; + unsigned char mode_ref_lf_delta_update;*/ + mbd->mode_ref_lf_delta_enabled = xd->mode_ref_lf_delta_enabled; + mbd->mode_ref_lf_delta_update = xd->mode_ref_lf_delta_update; + + mbd->current_bc = &pbi->mbc[0]; + + memcpy(mbd->dequant_y1_dc, xd->dequant_y1_dc, sizeof(xd->dequant_y1_dc)); + memcpy(mbd->dequant_y1, xd->dequant_y1, sizeof(xd->dequant_y1)); + memcpy(mbd->dequant_y2, xd->dequant_y2, sizeof(xd->dequant_y2)); + memcpy(mbd->dequant_uv, xd->dequant_uv, sizeof(xd->dequant_uv)); + + mbd->fullpixel_mask = 0xffffffff; + + if (pc->full_pixel) + mbd->fullpixel_mask = 0xfffffff8; + + } + + for (i = 0; i < pc->mb_rows; i++) + pbi->mt_current_mb_col[i] = -1; +} + +static void mt_decode_macroblock(VP8D_COMP *pbi, MACROBLOCKD *xd, + unsigned int mb_idx) +{ + MB_PREDICTION_MODE mode; + int i; +#if CONFIG_ERROR_CONCEALMENT + int corruption_detected = 0; +#else + (void)mb_idx; +#endif + + if (xd->mode_info_context->mbmi.mb_skip_coeff) + { + vp8_reset_mb_tokens_context(xd); + } + else if (!vp8dx_bool_error(xd->current_bc)) + { + int eobtotal; + eobtotal = vp8_decode_mb_tokens(pbi, xd); + + /* Special case: Force the loopfilter to skip when eobtotal is zero */ + xd->mode_info_context->mbmi.mb_skip_coeff = (eobtotal==0); + } + + mode = xd->mode_info_context->mbmi.mode; + + if (xd->segmentation_enabled) + vp8_mb_init_dequantizer(pbi, xd); + + +#if CONFIG_ERROR_CONCEALMENT + + if(pbi->ec_active) + { + int throw_residual; + /* When we have independent partitions we can apply residual even + * though other partitions within the frame are corrupt. + */ + throw_residual = (!pbi->independent_partitions && + pbi->frame_corrupt_residual); + throw_residual = (throw_residual || vp8dx_bool_error(xd->current_bc)); + + if ((mb_idx >= pbi->mvs_corrupt_from_mb || throw_residual)) + { + /* MB with corrupt residuals or corrupt mode/motion vectors. + * Better to use the predictor as reconstruction. + */ + pbi->frame_corrupt_residual = 1; + memset(xd->qcoeff, 0, sizeof(xd->qcoeff)); + + corruption_detected = 1; + + /* force idct to be skipped for B_PRED and use the + * prediction only for reconstruction + * */ + memset(xd->eobs, 0, 25); + } + } +#endif + + /* do prediction */ + if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) + { + vp8_build_intra_predictors_mbuv_s(xd, + xd->recon_above[1], + xd->recon_above[2], + xd->recon_left[1], + xd->recon_left[2], + xd->recon_left_stride[1], + xd->dst.u_buffer, xd->dst.v_buffer, + xd->dst.uv_stride); + + if (mode != B_PRED) + { + vp8_build_intra_predictors_mby_s(xd, + xd->recon_above[0], + xd->recon_left[0], + xd->recon_left_stride[0], + xd->dst.y_buffer, + xd->dst.y_stride); + } + else + { + short *DQC = xd->dequant_y1; + int dst_stride = xd->dst.y_stride; + + /* clear out residual eob info */ + if(xd->mode_info_context->mbmi.mb_skip_coeff) + memset(xd->eobs, 0, 25); + + intra_prediction_down_copy(xd, xd->recon_above[0] + 16); + + for (i = 0; i < 16; i++) + { + BLOCKD *b = &xd->block[i]; + unsigned char *dst = xd->dst.y_buffer + b->offset; + B_PREDICTION_MODE b_mode = + xd->mode_info_context->bmi[i].as_mode; + unsigned char *Above; + unsigned char *yleft; + int left_stride; + unsigned char top_left; + + /*Caution: For some b_mode, it needs 8 pixels (4 above + 4 above-right).*/ + if (i < 4 && pbi->common.filter_level) + Above = xd->recon_above[0] + b->offset; + else + Above = dst - dst_stride; + + if (i%4==0 && pbi->common.filter_level) + { + yleft = xd->recon_left[0] + i; + left_stride = 1; + } + else + { + yleft = dst - 1; + left_stride = dst_stride; + } + + if ((i==4 || i==8 || i==12) && pbi->common.filter_level) + top_left = *(xd->recon_left[0] + i - 1); + else + top_left = Above[-1]; + + vp8_intra4x4_predict(Above, yleft, left_stride, + b_mode, dst, dst_stride, top_left); + + if (xd->eobs[i] ) + { + if (xd->eobs[i] > 1) + { + vp8_dequant_idct_add(b->qcoeff, DQC, dst, dst_stride); + } + else + { + vp8_dc_only_idct_add(b->qcoeff[0] * DQC[0], + dst, dst_stride, dst, dst_stride); + memset(b->qcoeff, 0, 2 * sizeof(b->qcoeff[0])); + } + } + } + } + } + else + { + vp8_build_inter_predictors_mb(xd); + } + + +#if CONFIG_ERROR_CONCEALMENT + if (corruption_detected) + { + return; + } +#endif + + if(!xd->mode_info_context->mbmi.mb_skip_coeff) + { + /* dequantization and idct */ + if (mode != B_PRED) + { + short *DQC = xd->dequant_y1; + + if (mode != SPLITMV) + { + BLOCKD *b = &xd->block[24]; + + /* do 2nd order transform on the dc block */ + if (xd->eobs[24] > 1) + { + vp8_dequantize_b(b, xd->dequant_y2); + + vp8_short_inv_walsh4x4(&b->dqcoeff[0], + xd->qcoeff); + memset(b->qcoeff, 0, 16 * sizeof(b->qcoeff[0])); + } + else + { + b->dqcoeff[0] = b->qcoeff[0] * xd->dequant_y2[0]; + vp8_short_inv_walsh4x4_1(&b->dqcoeff[0], + xd->qcoeff); + memset(b->qcoeff, 0, 2 * sizeof(b->qcoeff[0])); + } + + /* override the dc dequant constant in order to preserve the + * dc components + */ + DQC = xd->dequant_y1_dc; + } + + vp8_dequant_idct_add_y_block + (xd->qcoeff, DQC, + xd->dst.y_buffer, + xd->dst.y_stride, xd->eobs); + } + + vp8_dequant_idct_add_uv_block + (xd->qcoeff+16*16, xd->dequant_uv, + xd->dst.u_buffer, xd->dst.v_buffer, + xd->dst.uv_stride, xd->eobs+16); + } +} + +static void mt_decode_mb_rows(VP8D_COMP *pbi, MACROBLOCKD *xd, int start_mb_row) +{ + const int *last_row_current_mb_col; + int *current_mb_col; + int mb_row; + VP8_COMMON *pc = &pbi->common; + const int nsync = pbi->sync_range; + const int first_row_no_sync_above = pc->mb_cols + nsync; + int num_part = 1 << pbi->common.multi_token_partition; + int last_mb_row = start_mb_row; + + YV12_BUFFER_CONFIG *yv12_fb_new = pbi->dec_fb_ref[INTRA_FRAME]; + YV12_BUFFER_CONFIG *yv12_fb_lst = pbi->dec_fb_ref[LAST_FRAME]; + + int recon_y_stride = yv12_fb_new->y_stride; + int recon_uv_stride = yv12_fb_new->uv_stride; + + unsigned char *ref_buffer[MAX_REF_FRAMES][3]; + unsigned char *dst_buffer[3]; + int i; + int ref_fb_corrupted[MAX_REF_FRAMES]; + + ref_fb_corrupted[INTRA_FRAME] = 0; + + for(i = 1; i < MAX_REF_FRAMES; i++) + { + YV12_BUFFER_CONFIG *this_fb = pbi->dec_fb_ref[i]; + + ref_buffer[i][0] = this_fb->y_buffer; + ref_buffer[i][1] = this_fb->u_buffer; + ref_buffer[i][2] = this_fb->v_buffer; + + ref_fb_corrupted[i] = this_fb->corrupted; + } + + dst_buffer[0] = yv12_fb_new->y_buffer; + dst_buffer[1] = yv12_fb_new->u_buffer; + dst_buffer[2] = yv12_fb_new->v_buffer; + + xd->up_available = (start_mb_row != 0); + + xd->mode_info_context = pc->mi + pc->mode_info_stride * start_mb_row; + xd->mode_info_stride = pc->mode_info_stride; + + for (mb_row = start_mb_row; mb_row < pc->mb_rows; mb_row += (pbi->decoding_thread_count + 1)) + { + int recon_yoffset, recon_uvoffset; + int mb_col; + int filter_level; + loop_filter_info_n *lfi_n = &pc->lf_info; + + /* save last row processed by this thread */ + last_mb_row = mb_row; + /* select bool coder for current partition */ + xd->current_bc = &pbi->mbc[mb_row%num_part]; + + if (mb_row > 0) + last_row_current_mb_col = &pbi->mt_current_mb_col[mb_row -1]; + else + last_row_current_mb_col = &first_row_no_sync_above; + + current_mb_col = &pbi->mt_current_mb_col[mb_row]; + + recon_yoffset = mb_row * recon_y_stride * 16; + recon_uvoffset = mb_row * recon_uv_stride * 8; + + /* reset contexts */ + xd->above_context = pc->above_context; + memset(xd->left_context, 0, sizeof(ENTROPY_CONTEXT_PLANES)); + + xd->left_available = 0; + + xd->mb_to_top_edge = -((mb_row * 16)) << 3; + xd->mb_to_bottom_edge = ((pc->mb_rows - 1 - mb_row) * 16) << 3; + + if (pbi->common.filter_level) + { + xd->recon_above[0] = pbi->mt_yabove_row[mb_row] + 0*16 +32; + xd->recon_above[1] = pbi->mt_uabove_row[mb_row] + 0*8 +16; + xd->recon_above[2] = pbi->mt_vabove_row[mb_row] + 0*8 +16; + + xd->recon_left[0] = pbi->mt_yleft_col[mb_row]; + xd->recon_left[1] = pbi->mt_uleft_col[mb_row]; + xd->recon_left[2] = pbi->mt_vleft_col[mb_row]; + + /* TODO: move to outside row loop */ + xd->recon_left_stride[0] = 1; + xd->recon_left_stride[1] = 1; + } + else + { + xd->recon_above[0] = dst_buffer[0] + recon_yoffset; + xd->recon_above[1] = dst_buffer[1] + recon_uvoffset; + xd->recon_above[2] = dst_buffer[2] + recon_uvoffset; + + xd->recon_left[0] = xd->recon_above[0] - 1; + xd->recon_left[1] = xd->recon_above[1] - 1; + xd->recon_left[2] = xd->recon_above[2] - 1; + + xd->recon_above[0] -= xd->dst.y_stride; + xd->recon_above[1] -= xd->dst.uv_stride; + xd->recon_above[2] -= xd->dst.uv_stride; + + /* TODO: move to outside row loop */ + xd->recon_left_stride[0] = xd->dst.y_stride; + xd->recon_left_stride[1] = xd->dst.uv_stride; + + setup_intra_recon_left(xd->recon_left[0], xd->recon_left[1], + xd->recon_left[2], xd->dst.y_stride, + xd->dst.uv_stride); + } + + for (mb_col = 0; mb_col < pc->mb_cols; mb_col++) { + if (((mb_col - 1) % nsync) == 0) { + pthread_mutex_t *mutex = &pbi->pmutex[mb_row]; + protected_write(mutex, current_mb_col, mb_col - 1); + } + + if (mb_row && !(mb_col & (nsync - 1))) { + pthread_mutex_t *mutex = &pbi->pmutex[mb_row-1]; + sync_read(mutex, mb_col, last_row_current_mb_col, nsync); + } + + /* Distance of MB to the various image edges. + * These are specified to 8th pel as they are always + * compared to values that are in 1/8th pel units. + */ + xd->mb_to_left_edge = -((mb_col * 16) << 3); + xd->mb_to_right_edge = ((pc->mb_cols - 1 - mb_col) * 16) << 3; + + #if CONFIG_ERROR_CONCEALMENT + { + int corrupt_residual = + (!pbi->independent_partitions && + pbi->frame_corrupt_residual) || + vp8dx_bool_error(xd->current_bc); + if (pbi->ec_active && + (xd->mode_info_context->mbmi.ref_frame == + INTRA_FRAME) && + corrupt_residual) + { + /* We have an intra block with corrupt + * coefficients, better to conceal with an inter + * block. + * Interpolate MVs from neighboring MBs + * + * Note that for the first mb with corrupt + * residual in a frame, we might not discover + * that before decoding the residual. That + * happens after this check, and therefore no + * inter concealment will be done. + */ + vp8_interpolate_motion(xd, + mb_row, mb_col, + pc->mb_rows, pc->mb_cols); + } + } + #endif + + + xd->dst.y_buffer = dst_buffer[0] + recon_yoffset; + xd->dst.u_buffer = dst_buffer[1] + recon_uvoffset; + xd->dst.v_buffer = dst_buffer[2] + recon_uvoffset; + + xd->pre.y_buffer = ref_buffer[xd->mode_info_context->mbmi.ref_frame][0] + recon_yoffset; + xd->pre.u_buffer = ref_buffer[xd->mode_info_context->mbmi.ref_frame][1] + recon_uvoffset; + xd->pre.v_buffer = ref_buffer[xd->mode_info_context->mbmi.ref_frame][2] + recon_uvoffset; + + /* propagate errors from reference frames */ + xd->corrupted |= ref_fb_corrupted[xd->mode_info_context->mbmi.ref_frame]; + + mt_decode_macroblock(pbi, xd, 0); + + xd->left_available = 1; + + /* check if the boolean decoder has suffered an error */ + xd->corrupted |= vp8dx_bool_error(xd->current_bc); + + xd->recon_above[0] += 16; + xd->recon_above[1] += 8; + xd->recon_above[2] += 8; + + if (!pbi->common.filter_level) + { + xd->recon_left[0] += 16; + xd->recon_left[1] += 8; + xd->recon_left[2] += 8; + } + + if (pbi->common.filter_level) + { + int skip_lf = (xd->mode_info_context->mbmi.mode != B_PRED && + xd->mode_info_context->mbmi.mode != SPLITMV && + xd->mode_info_context->mbmi.mb_skip_coeff); + + const int mode_index = lfi_n->mode_lf_lut[xd->mode_info_context->mbmi.mode]; + const int seg = xd->mode_info_context->mbmi.segment_id; + const int ref_frame = xd->mode_info_context->mbmi.ref_frame; + + filter_level = lfi_n->lvl[seg][ref_frame][mode_index]; + + if( mb_row != pc->mb_rows-1 ) + { + /* Save decoded MB last row data for next-row decoding */ + memcpy((pbi->mt_yabove_row[mb_row + 1] + 32 + mb_col*16), (xd->dst.y_buffer + 15 * recon_y_stride), 16); + memcpy((pbi->mt_uabove_row[mb_row + 1] + 16 + mb_col*8), (xd->dst.u_buffer + 7 * recon_uv_stride), 8); + memcpy((pbi->mt_vabove_row[mb_row + 1] + 16 + mb_col*8), (xd->dst.v_buffer + 7 * recon_uv_stride), 8); + } + + /* save left_col for next MB decoding */ + if(mb_col != pc->mb_cols-1) + { + MODE_INFO *next = xd->mode_info_context +1; + + if (next->mbmi.ref_frame == INTRA_FRAME) + { + for (i = 0; i < 16; i++) + pbi->mt_yleft_col[mb_row][i] = xd->dst.y_buffer [i* recon_y_stride + 15]; + for (i = 0; i < 8; i++) + { + pbi->mt_uleft_col[mb_row][i] = xd->dst.u_buffer [i* recon_uv_stride + 7]; + pbi->mt_vleft_col[mb_row][i] = xd->dst.v_buffer [i* recon_uv_stride + 7]; + } + } + } + + /* loopfilter on this macroblock. */ + if (filter_level) + { + if(pc->filter_type == NORMAL_LOOPFILTER) + { + loop_filter_info lfi; + FRAME_TYPE frame_type = pc->frame_type; + const int hev_index = lfi_n->hev_thr_lut[frame_type][filter_level]; + lfi.mblim = lfi_n->mblim[filter_level]; + lfi.blim = lfi_n->blim[filter_level]; + lfi.lim = lfi_n->lim[filter_level]; + lfi.hev_thr = lfi_n->hev_thr[hev_index]; + + if (mb_col > 0) + vp8_loop_filter_mbv + (xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi); + + if (!skip_lf) + vp8_loop_filter_bv + (xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi); + + /* don't apply across umv border */ + if (mb_row > 0) + vp8_loop_filter_mbh + (xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi); + + if (!skip_lf) + vp8_loop_filter_bh + (xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi); + } + else + { + if (mb_col > 0) + vp8_loop_filter_simple_mbv + (xd->dst.y_buffer, recon_y_stride, lfi_n->mblim[filter_level]); + + if (!skip_lf) + vp8_loop_filter_simple_bv + (xd->dst.y_buffer, recon_y_stride, lfi_n->blim[filter_level]); + + /* don't apply across umv border */ + if (mb_row > 0) + vp8_loop_filter_simple_mbh + (xd->dst.y_buffer, recon_y_stride, lfi_n->mblim[filter_level]); + + if (!skip_lf) + vp8_loop_filter_simple_bh + (xd->dst.y_buffer, recon_y_stride, lfi_n->blim[filter_level]); + } + } + + } + + recon_yoffset += 16; + recon_uvoffset += 8; + + ++xd->mode_info_context; /* next mb */ + + xd->above_context++; + } + + /* adjust to the next row of mbs */ + if (pbi->common.filter_level) + { + if(mb_row != pc->mb_rows-1) + { + int lasty = yv12_fb_lst->y_width + VP8BORDERINPIXELS; + int lastuv = (yv12_fb_lst->y_width>>1) + (VP8BORDERINPIXELS>>1); + + for (i = 0; i < 4; i++) + { + pbi->mt_yabove_row[mb_row +1][lasty + i] = pbi->mt_yabove_row[mb_row +1][lasty -1]; + pbi->mt_uabove_row[mb_row +1][lastuv + i] = pbi->mt_uabove_row[mb_row +1][lastuv -1]; + pbi->mt_vabove_row[mb_row +1][lastuv + i] = pbi->mt_vabove_row[mb_row +1][lastuv -1]; + } + } + } + else + vp8_extend_mb_row(yv12_fb_new, xd->dst.y_buffer + 16, + xd->dst.u_buffer + 8, xd->dst.v_buffer + 8); + + /* last MB of row is ready just after extension is done */ + protected_write(&pbi->pmutex[mb_row], current_mb_col, mb_col + nsync); + + ++xd->mode_info_context; /* skip prediction column */ + xd->up_available = 1; + + /* since we have multithread */ + xd->mode_info_context += xd->mode_info_stride * pbi->decoding_thread_count; + } + + /* signal end of frame decoding if this thread processed the last mb_row */ + if (last_mb_row == (pc->mb_rows - 1)) + sem_post(&pbi->h_event_end_decoding); + +} + + +static THREAD_FUNCTION thread_decoding_proc(void *p_data) +{ + int ithread = ((DECODETHREAD_DATA *)p_data)->ithread; + VP8D_COMP *pbi = (VP8D_COMP *)(((DECODETHREAD_DATA *)p_data)->ptr1); + MB_ROW_DEC *mbrd = (MB_ROW_DEC *)(((DECODETHREAD_DATA *)p_data)->ptr2); + ENTROPY_CONTEXT_PLANES mb_row_left_context; + + while (1) + { + if (protected_read(&pbi->mt_mutex, &pbi->b_multithreaded_rd) == 0) + break; + + if (sem_wait(&pbi->h_event_start_decoding[ithread]) == 0) + { + if (protected_read(&pbi->mt_mutex, &pbi->b_multithreaded_rd) == 0) + break; + else + { + MACROBLOCKD *xd = &mbrd->mbd; + xd->left_context = &mb_row_left_context; + + mt_decode_mb_rows(pbi, xd, ithread+1); + } + } + } + + return 0 ; +} + + +void vp8_decoder_create_threads(VP8D_COMP *pbi) +{ + int core_count = 0; + unsigned int ithread; + + pbi->b_multithreaded_rd = 0; + pbi->allocated_decoding_thread_count = 0; + pthread_mutex_init(&pbi->mt_mutex, NULL); + + /* limit decoding threads to the max number of token partitions */ + core_count = (pbi->max_threads > 8) ? 8 : pbi->max_threads; + + /* limit decoding threads to the available cores */ + if (core_count > pbi->common.processor_core_count) + core_count = pbi->common.processor_core_count; + + if (core_count > 1) + { + pbi->b_multithreaded_rd = 1; + pbi->decoding_thread_count = core_count - 1; + + CALLOC_ARRAY(pbi->h_decoding_thread, pbi->decoding_thread_count); + CALLOC_ARRAY(pbi->h_event_start_decoding, pbi->decoding_thread_count); + CALLOC_ARRAY_ALIGNED(pbi->mb_row_di, pbi->decoding_thread_count, 32); + CALLOC_ARRAY(pbi->de_thread_data, pbi->decoding_thread_count); + + for (ithread = 0; ithread < pbi->decoding_thread_count; ithread++) + { + sem_init(&pbi->h_event_start_decoding[ithread], 0, 0); + + vp8_setup_block_dptrs(&pbi->mb_row_di[ithread].mbd); + + pbi->de_thread_data[ithread].ithread = ithread; + pbi->de_thread_data[ithread].ptr1 = (void *)pbi; + pbi->de_thread_data[ithread].ptr2 = (void *) &pbi->mb_row_di[ithread]; + + pthread_create(&pbi->h_decoding_thread[ithread], 0, thread_decoding_proc, (&pbi->de_thread_data[ithread])); + } + + sem_init(&pbi->h_event_end_decoding, 0, 0); + + pbi->allocated_decoding_thread_count = pbi->decoding_thread_count; + } +} + + +void vp8mt_de_alloc_temp_buffers(VP8D_COMP *pbi, int mb_rows) +{ + int i; + + if (protected_read(&pbi->mt_mutex, &pbi->b_multithreaded_rd)) + { + /* De-allocate mutex */ + if (pbi->pmutex != NULL) { + for (i = 0; i < mb_rows; i++) { + pthread_mutex_destroy(&pbi->pmutex[i]); + } + vpx_free(pbi->pmutex); + pbi->pmutex = NULL; + } + + vpx_free(pbi->mt_current_mb_col); + pbi->mt_current_mb_col = NULL ; + + /* Free above_row buffers. */ + if (pbi->mt_yabove_row) + { + for (i=0; i< mb_rows; i++) + { + vpx_free(pbi->mt_yabove_row[i]); + pbi->mt_yabove_row[i] = NULL ; + } + vpx_free(pbi->mt_yabove_row); + pbi->mt_yabove_row = NULL ; + } + + if (pbi->mt_uabove_row) + { + for (i=0; i< mb_rows; i++) + { + vpx_free(pbi->mt_uabove_row[i]); + pbi->mt_uabove_row[i] = NULL ; + } + vpx_free(pbi->mt_uabove_row); + pbi->mt_uabove_row = NULL ; + } + + if (pbi->mt_vabove_row) + { + for (i=0; i< mb_rows; i++) + { + vpx_free(pbi->mt_vabove_row[i]); + pbi->mt_vabove_row[i] = NULL ; + } + vpx_free(pbi->mt_vabove_row); + pbi->mt_vabove_row = NULL ; + } + + /* Free left_col buffers. */ + if (pbi->mt_yleft_col) + { + for (i=0; i< mb_rows; i++) + { + vpx_free(pbi->mt_yleft_col[i]); + pbi->mt_yleft_col[i] = NULL ; + } + vpx_free(pbi->mt_yleft_col); + pbi->mt_yleft_col = NULL ; + } + + if (pbi->mt_uleft_col) + { + for (i=0; i< mb_rows; i++) + { + vpx_free(pbi->mt_uleft_col[i]); + pbi->mt_uleft_col[i] = NULL ; + } + vpx_free(pbi->mt_uleft_col); + pbi->mt_uleft_col = NULL ; + } + + if (pbi->mt_vleft_col) + { + for (i=0; i< mb_rows; i++) + { + vpx_free(pbi->mt_vleft_col[i]); + pbi->mt_vleft_col[i] = NULL ; + } + vpx_free(pbi->mt_vleft_col); + pbi->mt_vleft_col = NULL ; + } + } +} + + +void vp8mt_alloc_temp_buffers(VP8D_COMP *pbi, int width, int prev_mb_rows) +{ + VP8_COMMON *const pc = & pbi->common; + int i; + int uv_width; + + if (protected_read(&pbi->mt_mutex, &pbi->b_multithreaded_rd)) + { + vp8mt_de_alloc_temp_buffers(pbi, prev_mb_rows); + + /* our internal buffers are always multiples of 16 */ + if ((width & 0xf) != 0) + width += 16 - (width & 0xf); + + if (width < 640) pbi->sync_range = 1; + else if (width <= 1280) pbi->sync_range = 8; + else if (width <= 2560) pbi->sync_range =16; + else pbi->sync_range = 32; + + uv_width = width >>1; + + /* Allocate mutex */ + CHECK_MEM_ERROR(pbi->pmutex, vpx_malloc(sizeof(*pbi->pmutex) * + pc->mb_rows)); + if (pbi->pmutex) { + for (i = 0; i < pc->mb_rows; i++) { + pthread_mutex_init(&pbi->pmutex[i], NULL); + } + } + + /* Allocate an int for each mb row. */ + CALLOC_ARRAY(pbi->mt_current_mb_col, pc->mb_rows); + + /* Allocate memory for above_row buffers. */ + CALLOC_ARRAY(pbi->mt_yabove_row, pc->mb_rows); + for (i = 0; i < pc->mb_rows; i++) + CHECK_MEM_ERROR(pbi->mt_yabove_row[i], vpx_memalign(16,sizeof(unsigned char) * (width + (VP8BORDERINPIXELS<<1)))); + + CALLOC_ARRAY(pbi->mt_uabove_row, pc->mb_rows); + for (i = 0; i < pc->mb_rows; i++) + CHECK_MEM_ERROR(pbi->mt_uabove_row[i], vpx_memalign(16,sizeof(unsigned char) * (uv_width + VP8BORDERINPIXELS))); + + CALLOC_ARRAY(pbi->mt_vabove_row, pc->mb_rows); + for (i = 0; i < pc->mb_rows; i++) + CHECK_MEM_ERROR(pbi->mt_vabove_row[i], vpx_memalign(16,sizeof(unsigned char) * (uv_width + VP8BORDERINPIXELS))); + + /* Allocate memory for left_col buffers. */ + CALLOC_ARRAY(pbi->mt_yleft_col, pc->mb_rows); + for (i = 0; i < pc->mb_rows; i++) + CHECK_MEM_ERROR(pbi->mt_yleft_col[i], vpx_calloc(sizeof(unsigned char) * 16, 1)); + + CALLOC_ARRAY(pbi->mt_uleft_col, pc->mb_rows); + for (i = 0; i < pc->mb_rows; i++) + CHECK_MEM_ERROR(pbi->mt_uleft_col[i], vpx_calloc(sizeof(unsigned char) * 8, 1)); + + CALLOC_ARRAY(pbi->mt_vleft_col, pc->mb_rows); + for (i = 0; i < pc->mb_rows; i++) + CHECK_MEM_ERROR(pbi->mt_vleft_col[i], vpx_calloc(sizeof(unsigned char) * 8, 1)); + } +} + + +void vp8_decoder_remove_threads(VP8D_COMP *pbi) +{ + /* shutdown MB Decoding thread; */ + if (protected_read(&pbi->mt_mutex, &pbi->b_multithreaded_rd)) + { + int i; + + protected_write(&pbi->mt_mutex, &pbi->b_multithreaded_rd, 0); + + /* allow all threads to exit */ + for (i = 0; i < pbi->allocated_decoding_thread_count; i++) + { + sem_post(&pbi->h_event_start_decoding[i]); + pthread_join(pbi->h_decoding_thread[i], NULL); + } + + for (i = 0; i < pbi->allocated_decoding_thread_count; i++) + { + sem_destroy(&pbi->h_event_start_decoding[i]); + } + + sem_destroy(&pbi->h_event_end_decoding); + + vpx_free(pbi->h_decoding_thread); + pbi->h_decoding_thread = NULL; + + vpx_free(pbi->h_event_start_decoding); + pbi->h_event_start_decoding = NULL; + + vpx_free(pbi->mb_row_di); + pbi->mb_row_di = NULL ; + + vpx_free(pbi->de_thread_data); + pbi->de_thread_data = NULL; + } + pthread_mutex_destroy(&pbi->mt_mutex); +} + +void vp8mt_decode_mb_rows( VP8D_COMP *pbi, MACROBLOCKD *xd) +{ + VP8_COMMON *pc = &pbi->common; + unsigned int i; + int j; + + int filter_level = pc->filter_level; + YV12_BUFFER_CONFIG *yv12_fb_new = pbi->dec_fb_ref[INTRA_FRAME]; + + if (filter_level) + { + /* Set above_row buffer to 127 for decoding first MB row */ + memset(pbi->mt_yabove_row[0] + VP8BORDERINPIXELS-1, 127, yv12_fb_new->y_width + 5); + memset(pbi->mt_uabove_row[0] + (VP8BORDERINPIXELS>>1)-1, 127, (yv12_fb_new->y_width>>1) +5); + memset(pbi->mt_vabove_row[0] + (VP8BORDERINPIXELS>>1)-1, 127, (yv12_fb_new->y_width>>1) +5); + + for (j=1; j<pc->mb_rows; j++) + { + memset(pbi->mt_yabove_row[j] + VP8BORDERINPIXELS-1, (unsigned char)129, 1); + memset(pbi->mt_uabove_row[j] + (VP8BORDERINPIXELS>>1)-1, (unsigned char)129, 1); + memset(pbi->mt_vabove_row[j] + (VP8BORDERINPIXELS>>1)-1, (unsigned char)129, 1); + } + + /* Set left_col to 129 initially */ + for (j=0; j<pc->mb_rows; j++) + { + memset(pbi->mt_yleft_col[j], (unsigned char)129, 16); + memset(pbi->mt_uleft_col[j], (unsigned char)129, 8); + memset(pbi->mt_vleft_col[j], (unsigned char)129, 8); + } + + /* Initialize the loop filter for this frame. */ + vp8_loop_filter_frame_init(pc, &pbi->mb, filter_level); + } + else + vp8_setup_intra_recon_top_line(yv12_fb_new); + + setup_decoding_thread_data(pbi, xd, pbi->mb_row_di, pbi->decoding_thread_count); + + for (i = 0; i < pbi->decoding_thread_count; i++) + sem_post(&pbi->h_event_start_decoding[i]); + + mt_decode_mb_rows(pbi, xd, 0); + + sem_wait(&pbi->h_event_end_decoding); /* add back for each frame */ +} diff --git a/thirdparty/libvpx/vp8/decoder/treereader.h b/thirdparty/libvpx/vp8/decoder/treereader.h new file mode 100644 index 0000000000..f7d23c3698 --- /dev/null +++ b/thirdparty/libvpx/vp8/decoder/treereader.h @@ -0,0 +1,49 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP8_DECODER_TREEREADER_H_ +#define VP8_DECODER_TREEREADER_H_ + +#include "./vpx_config.h" +#include "vp8/common/treecoder.h" +#include "dboolhuff.h" + +#ifdef __cplusplus +extern "C" { +#endif + +typedef BOOL_DECODER vp8_reader; + +#define vp8_read vp8dx_decode_bool +#define vp8_read_literal vp8_decode_value +#define vp8_read_bit(R) vp8_read(R, vp8_prob_half) + + +/* Intent of tree data structure is to make decoding trivial. */ + +static INLINE int vp8_treed_read( + vp8_reader *const r, /* !!! must return a 0 or 1 !!! */ + vp8_tree t, + const vp8_prob *const p +) +{ + register vp8_tree_index i = 0; + + while ((i = t[ i + vp8_read(r, p[i>>1])]) > 0) ; + + return -i; +} + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_DECODER_TREEREADER_H_ diff --git a/thirdparty/libvpx/vp8/vp8_dx_iface.c b/thirdparty/libvpx/vp8/vp8_dx_iface.c new file mode 100644 index 0000000000..fc9288d62b --- /dev/null +++ b/thirdparty/libvpx/vp8/vp8_dx_iface.c @@ -0,0 +1,828 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include <assert.h> +#include <stdlib.h> +#include <string.h> +#include "./vp8_rtcd.h" +#include "./vpx_dsp_rtcd.h" +#include "./vpx_scale_rtcd.h" +#include "vpx/vpx_decoder.h" +#include "vpx/vp8dx.h" +#include "vpx/internal/vpx_codec_internal.h" +#include "vpx_version.h" +#include "common/alloccommon.h" +#include "common/common.h" +#include "common/onyxd.h" +#include "decoder/onyxd_int.h" +#include "vpx_dsp/vpx_dsp_common.h" +#include "vpx_mem/vpx_mem.h" +#if CONFIG_ERROR_CONCEALMENT +#include "decoder/error_concealment.h" +#endif +#include "decoder/decoderthreading.h" + +#define VP8_CAP_POSTPROC (CONFIG_POSTPROC ? VPX_CODEC_CAP_POSTPROC : 0) +#define VP8_CAP_ERROR_CONCEALMENT (CONFIG_ERROR_CONCEALMENT ? \ + VPX_CODEC_CAP_ERROR_CONCEALMENT : 0) + +typedef vpx_codec_stream_info_t vp8_stream_info_t; + +/* Structures for handling memory allocations */ +typedef enum +{ + VP8_SEG_ALG_PRIV = 256, + VP8_SEG_MAX +} mem_seg_id_t; +#define NELEMENTS(x) ((int)(sizeof(x)/sizeof(x[0]))) + +struct vpx_codec_alg_priv +{ + vpx_codec_priv_t base; + vpx_codec_dec_cfg_t cfg; + vp8_stream_info_t si; + int decoder_init; + int postproc_cfg_set; + vp8_postproc_cfg_t postproc_cfg; +#if CONFIG_POSTPROC_VISUALIZER + unsigned int dbg_postproc_flag; + int dbg_color_ref_frame_flag; + int dbg_color_mb_modes_flag; + int dbg_color_b_modes_flag; + int dbg_display_mv_flag; +#endif + vpx_decrypt_cb decrypt_cb; + void *decrypt_state; + vpx_image_t img; + int img_setup; + struct frame_buffers yv12_frame_buffers; + void *user_priv; + FRAGMENT_DATA fragments; +}; + +static int vp8_init_ctx(vpx_codec_ctx_t *ctx) +{ + vpx_codec_alg_priv_t *priv = + (vpx_codec_alg_priv_t *)vpx_calloc(1, sizeof(*priv)); + if (!priv) return 1; + + ctx->priv = (vpx_codec_priv_t *)priv; + ctx->priv->init_flags = ctx->init_flags; + + priv->si.sz = sizeof(priv->si); + priv->decrypt_cb = NULL; + priv->decrypt_state = NULL; + + if (ctx->config.dec) + { + /* Update the reference to the config structure to an internal copy. */ + priv->cfg = *ctx->config.dec; + ctx->config.dec = &priv->cfg; + } + + return 0; +} + +static vpx_codec_err_t vp8_init(vpx_codec_ctx_t *ctx, + vpx_codec_priv_enc_mr_cfg_t *data) +{ + vpx_codec_err_t res = VPX_CODEC_OK; + vpx_codec_alg_priv_t *priv = NULL; + (void) data; + + vp8_rtcd(); + vpx_dsp_rtcd(); + vpx_scale_rtcd(); + + /* This function only allocates space for the vpx_codec_alg_priv_t + * structure. More memory may be required at the time the stream + * information becomes known. + */ + if (!ctx->priv) { + if (vp8_init_ctx(ctx)) return VPX_CODEC_MEM_ERROR; + priv = (vpx_codec_alg_priv_t *)ctx->priv; + + /* initialize number of fragments to zero */ + priv->fragments.count = 0; + /* is input fragments enabled? */ + priv->fragments.enabled = + (priv->base.init_flags & VPX_CODEC_USE_INPUT_FRAGMENTS); + + /*post processing level initialized to do nothing */ + } else { + priv = (vpx_codec_alg_priv_t *)ctx->priv; + } + + priv->yv12_frame_buffers.use_frame_threads = + (ctx->priv->init_flags & VPX_CODEC_USE_FRAME_THREADING); + + /* for now, disable frame threading */ + priv->yv12_frame_buffers.use_frame_threads = 0; + + if (priv->yv12_frame_buffers.use_frame_threads && + ((ctx->priv->init_flags & VPX_CODEC_USE_ERROR_CONCEALMENT) || + (ctx->priv->init_flags & VPX_CODEC_USE_INPUT_FRAGMENTS))) { + /* row-based threading, error concealment, and input fragments will + * not be supported when using frame-based threading */ + res = VPX_CODEC_INVALID_PARAM; + } + + return res; +} + +static vpx_codec_err_t vp8_destroy(vpx_codec_alg_priv_t *ctx) +{ + vp8_remove_decoder_instances(&ctx->yv12_frame_buffers); + + vpx_free(ctx); + + return VPX_CODEC_OK; +} + +static vpx_codec_err_t vp8_peek_si_internal(const uint8_t *data, + unsigned int data_sz, + vpx_codec_stream_info_t *si, + vpx_decrypt_cb decrypt_cb, + void *decrypt_state) +{ + vpx_codec_err_t res = VPX_CODEC_OK; + + assert(data != NULL); + + if(data + data_sz <= data) + { + res = VPX_CODEC_INVALID_PARAM; + } + else + { + /* Parse uncompresssed part of key frame header. + * 3 bytes:- including version, frame type and an offset + * 3 bytes:- sync code (0x9d, 0x01, 0x2a) + * 4 bytes:- including image width and height in the lowest 14 bits + * of each 2-byte value. + */ + uint8_t clear_buffer[10]; + const uint8_t *clear = data; + if (decrypt_cb) + { + int n = VPXMIN(sizeof(clear_buffer), data_sz); + decrypt_cb(decrypt_state, data, clear_buffer, n); + clear = clear_buffer; + } + si->is_kf = 0; + + if (data_sz >= 10 && !(clear[0] & 0x01)) /* I-Frame */ + { + si->is_kf = 1; + + /* vet via sync code */ + if (clear[3] != 0x9d || clear[4] != 0x01 || clear[5] != 0x2a) + return VPX_CODEC_UNSUP_BITSTREAM; + + si->w = (clear[6] | (clear[7] << 8)) & 0x3fff; + si->h = (clear[8] | (clear[9] << 8)) & 0x3fff; + + /*printf("w=%d, h=%d\n", si->w, si->h);*/ + if (!(si->h | si->w)) + res = VPX_CODEC_UNSUP_BITSTREAM; + } + else + { + res = VPX_CODEC_UNSUP_BITSTREAM; + } + } + + return res; +} + +static vpx_codec_err_t vp8_peek_si(const uint8_t *data, + unsigned int data_sz, + vpx_codec_stream_info_t *si) { + return vp8_peek_si_internal(data, data_sz, si, NULL, NULL); +} + +static vpx_codec_err_t vp8_get_si(vpx_codec_alg_priv_t *ctx, + vpx_codec_stream_info_t *si) +{ + + unsigned int sz; + + if (si->sz >= sizeof(vp8_stream_info_t)) + sz = sizeof(vp8_stream_info_t); + else + sz = sizeof(vpx_codec_stream_info_t); + + memcpy(si, &ctx->si, sz); + si->sz = sz; + + return VPX_CODEC_OK; +} + + +static vpx_codec_err_t +update_error_state(vpx_codec_alg_priv_t *ctx, + const struct vpx_internal_error_info *error) +{ + vpx_codec_err_t res; + + if ((res = error->error_code)) + ctx->base.err_detail = error->has_detail + ? error->detail + : NULL; + + return res; +} + +static void yuvconfig2image(vpx_image_t *img, + const YV12_BUFFER_CONFIG *yv12, + void *user_priv) +{ + /** vpx_img_wrap() doesn't allow specifying independent strides for + * the Y, U, and V planes, nor other alignment adjustments that + * might be representable by a YV12_BUFFER_CONFIG, so we just + * initialize all the fields.*/ + img->fmt = VPX_IMG_FMT_I420; + img->w = yv12->y_stride; + img->h = (yv12->y_height + 2 * VP8BORDERINPIXELS + 15) & ~15; + img->d_w = img->r_w = yv12->y_width; + img->d_h = img->r_h = yv12->y_height; + img->x_chroma_shift = 1; + img->y_chroma_shift = 1; + img->planes[VPX_PLANE_Y] = yv12->y_buffer; + img->planes[VPX_PLANE_U] = yv12->u_buffer; + img->planes[VPX_PLANE_V] = yv12->v_buffer; + img->planes[VPX_PLANE_ALPHA] = NULL; + img->stride[VPX_PLANE_Y] = yv12->y_stride; + img->stride[VPX_PLANE_U] = yv12->uv_stride; + img->stride[VPX_PLANE_V] = yv12->uv_stride; + img->stride[VPX_PLANE_ALPHA] = yv12->y_stride; + img->bit_depth = 8; + img->bps = 12; + img->user_priv = user_priv; + img->img_data = yv12->buffer_alloc; + img->img_data_owner = 0; + img->self_allocd = 0; +} + +static int +update_fragments(vpx_codec_alg_priv_t *ctx, + const uint8_t *data, + unsigned int data_sz, + vpx_codec_err_t *res) +{ + *res = VPX_CODEC_OK; + + if (ctx->fragments.count == 0) + { + /* New frame, reset fragment pointers and sizes */ + memset((void*)ctx->fragments.ptrs, 0, sizeof(ctx->fragments.ptrs)); + memset(ctx->fragments.sizes, 0, sizeof(ctx->fragments.sizes)); + } + if (ctx->fragments.enabled && !(data == NULL && data_sz == 0)) + { + /* Store a pointer to this fragment and return. We haven't + * received the complete frame yet, so we will wait with decoding. + */ + ctx->fragments.ptrs[ctx->fragments.count] = data; + ctx->fragments.sizes[ctx->fragments.count] = data_sz; + ctx->fragments.count++; + if (ctx->fragments.count > (1 << EIGHT_PARTITION) + 1) + { + ctx->fragments.count = 0; + *res = VPX_CODEC_INVALID_PARAM; + return -1; + } + return 0; + } + + if (!ctx->fragments.enabled && (data == NULL && data_sz == 0)) + { + return 0; + } + + if (!ctx->fragments.enabled) + { + ctx->fragments.ptrs[0] = data; + ctx->fragments.sizes[0] = data_sz; + ctx->fragments.count = 1; + } + + return 1; +} + +static vpx_codec_err_t vp8_decode(vpx_codec_alg_priv_t *ctx, + const uint8_t *data, + unsigned int data_sz, + void *user_priv, + long deadline) +{ + vpx_codec_err_t res = VPX_CODEC_OK; + unsigned int resolution_change = 0; + unsigned int w, h; + + if (!ctx->fragments.enabled && (data == NULL && data_sz == 0)) + { + return 0; + } + + /* Update the input fragment data */ + if(update_fragments(ctx, data, data_sz, &res) <= 0) + return res; + + /* Determine the stream parameters. Note that we rely on peek_si to + * validate that we have a buffer that does not wrap around the top + * of the heap. + */ + w = ctx->si.w; + h = ctx->si.h; + + res = vp8_peek_si_internal(ctx->fragments.ptrs[0], ctx->fragments.sizes[0], + &ctx->si, ctx->decrypt_cb, ctx->decrypt_state); + + if((res == VPX_CODEC_UNSUP_BITSTREAM) && !ctx->si.is_kf) + { + /* the peek function returns an error for non keyframes, however for + * this case, it is not an error */ + res = VPX_CODEC_OK; + } + + if(!ctx->decoder_init && !ctx->si.is_kf) + res = VPX_CODEC_UNSUP_BITSTREAM; + + if ((ctx->si.h != h) || (ctx->si.w != w)) + resolution_change = 1; + + /* Initialize the decoder instance on the first frame*/ + if (!res && !ctx->decoder_init) + { + VP8D_CONFIG oxcf; + + oxcf.Width = ctx->si.w; + oxcf.Height = ctx->si.h; + oxcf.Version = 9; + oxcf.postprocess = 0; + oxcf.max_threads = ctx->cfg.threads; + oxcf.error_concealment = + (ctx->base.init_flags & VPX_CODEC_USE_ERROR_CONCEALMENT); + + /* If postprocessing was enabled by the application and a + * configuration has not been provided, default it. + */ + if (!ctx->postproc_cfg_set + && (ctx->base.init_flags & VPX_CODEC_USE_POSTPROC)) { + ctx->postproc_cfg.post_proc_flag = + VP8_DEBLOCK | VP8_DEMACROBLOCK | VP8_MFQE; + ctx->postproc_cfg.deblocking_level = 4; + ctx->postproc_cfg.noise_level = 0; + } + + res = vp8_create_decoder_instances(&ctx->yv12_frame_buffers, &oxcf); + ctx->decoder_init = 1; + } + + /* Set these even if already initialized. The caller may have changed the + * decrypt config between frames. + */ + if (ctx->decoder_init) { + ctx->yv12_frame_buffers.pbi[0]->decrypt_cb = ctx->decrypt_cb; + ctx->yv12_frame_buffers.pbi[0]->decrypt_state = ctx->decrypt_state; + } + + if (!res) + { + VP8D_COMP *pbi = ctx->yv12_frame_buffers.pbi[0]; + if (resolution_change) + { + VP8_COMMON *const pc = & pbi->common; + MACROBLOCKD *const xd = & pbi->mb; +#if CONFIG_MULTITHREAD + int i; +#endif + pc->Width = ctx->si.w; + pc->Height = ctx->si.h; + { + int prev_mb_rows = pc->mb_rows; + + if (setjmp(pbi->common.error.jmp)) + { + pbi->common.error.setjmp = 0; + vp8_clear_system_state(); + /* same return value as used in vp8dx_receive_compressed_data */ + return -1; + } + + pbi->common.error.setjmp = 1; + + if (pc->Width <= 0) + { + pc->Width = w; + vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME, + "Invalid frame width"); + } + + if (pc->Height <= 0) + { + pc->Height = h; + vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME, + "Invalid frame height"); + } + + if (vp8_alloc_frame_buffers(pc, pc->Width, pc->Height)) + vpx_internal_error(&pc->error, VPX_CODEC_MEM_ERROR, + "Failed to allocate frame buffers"); + + xd->pre = pc->yv12_fb[pc->lst_fb_idx]; + xd->dst = pc->yv12_fb[pc->new_fb_idx]; + +#if CONFIG_MULTITHREAD + for (i = 0; i < pbi->allocated_decoding_thread_count; i++) + { + pbi->mb_row_di[i].mbd.dst = pc->yv12_fb[pc->new_fb_idx]; + vp8_build_block_doffsets(&pbi->mb_row_di[i].mbd); + } +#endif + vp8_build_block_doffsets(&pbi->mb); + + /* allocate memory for last frame MODE_INFO array */ +#if CONFIG_ERROR_CONCEALMENT + + if (pbi->ec_enabled) + { + /* old prev_mip was released by vp8_de_alloc_frame_buffers() + * called in vp8_alloc_frame_buffers() */ + pc->prev_mip = vpx_calloc( + (pc->mb_cols + 1) * (pc->mb_rows + 1), + sizeof(MODE_INFO)); + + if (!pc->prev_mip) + { + vp8_de_alloc_frame_buffers(pc); + vpx_internal_error(&pc->error, VPX_CODEC_MEM_ERROR, + "Failed to allocate" + "last frame MODE_INFO array"); + } + + pc->prev_mi = pc->prev_mip + pc->mode_info_stride + 1; + + if (vp8_alloc_overlap_lists(pbi)) + vpx_internal_error(&pc->error, VPX_CODEC_MEM_ERROR, + "Failed to allocate overlap lists " + "for error concealment"); + } + +#endif + +#if CONFIG_MULTITHREAD + if (pbi->b_multithreaded_rd) + vp8mt_alloc_temp_buffers(pbi, pc->Width, prev_mb_rows); +#else + (void)prev_mb_rows; +#endif + } + + pbi->common.error.setjmp = 0; + + /* required to get past the first get_free_fb() call */ + pbi->common.fb_idx_ref_cnt[0] = 0; + } + + /* update the pbi fragment data */ + pbi->fragments = ctx->fragments; + + ctx->user_priv = user_priv; + if (vp8dx_receive_compressed_data(pbi, data_sz, data, deadline)) + { + res = update_error_state(ctx, &pbi->common.error); + } + + /* get ready for the next series of fragments */ + ctx->fragments.count = 0; + } + + return res; +} + +static vpx_image_t *vp8_get_frame(vpx_codec_alg_priv_t *ctx, + vpx_codec_iter_t *iter) +{ + vpx_image_t *img = NULL; + + /* iter acts as a flip flop, so an image is only returned on the first + * call to get_frame. + */ + if (!(*iter) && ctx->yv12_frame_buffers.pbi[0]) + { + YV12_BUFFER_CONFIG sd; + int64_t time_stamp = 0, time_end_stamp = 0; + vp8_ppflags_t flags; + vp8_zero(flags); + + if (ctx->base.init_flags & VPX_CODEC_USE_POSTPROC) + { + flags.post_proc_flag= ctx->postproc_cfg.post_proc_flag +#if CONFIG_POSTPROC_VISUALIZER + + | ((ctx->dbg_color_ref_frame_flag != 0) ? VP8D_DEBUG_CLR_FRM_REF_BLKS : 0) + | ((ctx->dbg_color_mb_modes_flag != 0) ? VP8D_DEBUG_CLR_BLK_MODES : 0) + | ((ctx->dbg_color_b_modes_flag != 0) ? VP8D_DEBUG_CLR_BLK_MODES : 0) + | ((ctx->dbg_display_mv_flag != 0) ? VP8D_DEBUG_DRAW_MV : 0) +#endif + ; + flags.deblocking_level = ctx->postproc_cfg.deblocking_level; + flags.noise_level = ctx->postproc_cfg.noise_level; +#if CONFIG_POSTPROC_VISUALIZER + flags.display_ref_frame_flag= ctx->dbg_color_ref_frame_flag; + flags.display_mb_modes_flag = ctx->dbg_color_mb_modes_flag; + flags.display_b_modes_flag = ctx->dbg_color_b_modes_flag; + flags.display_mv_flag = ctx->dbg_display_mv_flag; +#endif + } + + if (0 == vp8dx_get_raw_frame(ctx->yv12_frame_buffers.pbi[0], &sd, + &time_stamp, &time_end_stamp, &flags)) + { + yuvconfig2image(&ctx->img, &sd, ctx->user_priv); + + img = &ctx->img; + *iter = img; + } + } + + return img; +} + +static vpx_codec_err_t image2yuvconfig(const vpx_image_t *img, + YV12_BUFFER_CONFIG *yv12) +{ + const int y_w = img->d_w; + const int y_h = img->d_h; + const int uv_w = (img->d_w + 1) / 2; + const int uv_h = (img->d_h + 1) / 2; + vpx_codec_err_t res = VPX_CODEC_OK; + yv12->y_buffer = img->planes[VPX_PLANE_Y]; + yv12->u_buffer = img->planes[VPX_PLANE_U]; + yv12->v_buffer = img->planes[VPX_PLANE_V]; + + yv12->y_crop_width = y_w; + yv12->y_crop_height = y_h; + yv12->y_width = y_w; + yv12->y_height = y_h; + yv12->uv_crop_width = uv_w; + yv12->uv_crop_height = uv_h; + yv12->uv_width = uv_w; + yv12->uv_height = uv_h; + + yv12->y_stride = img->stride[VPX_PLANE_Y]; + yv12->uv_stride = img->stride[VPX_PLANE_U]; + + yv12->border = (img->stride[VPX_PLANE_Y] - img->d_w) / 2; + return res; +} + + +static vpx_codec_err_t vp8_set_reference(vpx_codec_alg_priv_t *ctx, + va_list args) +{ + + vpx_ref_frame_t *data = va_arg(args, vpx_ref_frame_t *); + + if (data && !ctx->yv12_frame_buffers.use_frame_threads) + { + vpx_ref_frame_t *frame = (vpx_ref_frame_t *)data; + YV12_BUFFER_CONFIG sd; + + image2yuvconfig(&frame->img, &sd); + + return vp8dx_set_reference(ctx->yv12_frame_buffers.pbi[0], + frame->frame_type, &sd); + } + else + return VPX_CODEC_INVALID_PARAM; + +} + +static vpx_codec_err_t vp8_get_reference(vpx_codec_alg_priv_t *ctx, + va_list args) +{ + + vpx_ref_frame_t *data = va_arg(args, vpx_ref_frame_t *); + + if (data && !ctx->yv12_frame_buffers.use_frame_threads) + { + vpx_ref_frame_t *frame = (vpx_ref_frame_t *)data; + YV12_BUFFER_CONFIG sd; + + image2yuvconfig(&frame->img, &sd); + + return vp8dx_get_reference(ctx->yv12_frame_buffers.pbi[0], + frame->frame_type, &sd); + } + else + return VPX_CODEC_INVALID_PARAM; + +} + +static vpx_codec_err_t vp8_set_postproc(vpx_codec_alg_priv_t *ctx, + va_list args) +{ +#if CONFIG_POSTPROC + vp8_postproc_cfg_t *data = va_arg(args, vp8_postproc_cfg_t *); + + if (data) + { + ctx->postproc_cfg_set = 1; + ctx->postproc_cfg = *((vp8_postproc_cfg_t *)data); + return VPX_CODEC_OK; + } + else + return VPX_CODEC_INVALID_PARAM; + +#else + (void)ctx; + (void)args; + return VPX_CODEC_INCAPABLE; +#endif +} + + +static vpx_codec_err_t vp8_set_dbg_color_ref_frame(vpx_codec_alg_priv_t *ctx, + va_list args) { +#if CONFIG_POSTPROC_VISUALIZER && CONFIG_POSTPROC + ctx->dbg_color_ref_frame_flag = va_arg(args, int); + return VPX_CODEC_OK; +#else + (void)ctx; + (void)args; + return VPX_CODEC_INCAPABLE; +#endif +} + +static vpx_codec_err_t vp8_set_dbg_color_mb_modes(vpx_codec_alg_priv_t *ctx, + va_list args) { +#if CONFIG_POSTPROC_VISUALIZER && CONFIG_POSTPROC + ctx->dbg_color_mb_modes_flag = va_arg(args, int); + return VPX_CODEC_OK; +#else + (void)ctx; + (void)args; + return VPX_CODEC_INCAPABLE; +#endif +} + +static vpx_codec_err_t vp8_set_dbg_color_b_modes(vpx_codec_alg_priv_t *ctx, + va_list args) { +#if CONFIG_POSTPROC_VISUALIZER && CONFIG_POSTPROC + ctx->dbg_color_b_modes_flag = va_arg(args, int); + return VPX_CODEC_OK; +#else + (void)ctx; + (void)args; + return VPX_CODEC_INCAPABLE; +#endif +} + +static vpx_codec_err_t vp8_set_dbg_display_mv(vpx_codec_alg_priv_t *ctx, + va_list args) { +#if CONFIG_POSTPROC_VISUALIZER && CONFIG_POSTPROC + ctx->dbg_display_mv_flag = va_arg(args, int); + return VPX_CODEC_OK; +#else + (void)ctx; + (void)args; + return VPX_CODEC_INCAPABLE; +#endif +} + +static vpx_codec_err_t vp8_get_last_ref_updates(vpx_codec_alg_priv_t *ctx, + va_list args) +{ + int *update_info = va_arg(args, int *); + + if (update_info && !ctx->yv12_frame_buffers.use_frame_threads) + { + VP8D_COMP *pbi = (VP8D_COMP *)ctx->yv12_frame_buffers.pbi[0]; + + *update_info = pbi->common.refresh_alt_ref_frame * (int) VP8_ALTR_FRAME + + pbi->common.refresh_golden_frame * (int) VP8_GOLD_FRAME + + pbi->common.refresh_last_frame * (int) VP8_LAST_FRAME; + + return VPX_CODEC_OK; + } + else + return VPX_CODEC_INVALID_PARAM; +} + +extern int vp8dx_references_buffer( VP8_COMMON *oci, int ref_frame ); +static vpx_codec_err_t vp8_get_last_ref_frame(vpx_codec_alg_priv_t *ctx, + va_list args) +{ + int *ref_info = va_arg(args, int *); + + if (ref_info && !ctx->yv12_frame_buffers.use_frame_threads) + { + VP8D_COMP *pbi = (VP8D_COMP *)ctx->yv12_frame_buffers.pbi[0]; + VP8_COMMON *oci = &pbi->common; + *ref_info = + (vp8dx_references_buffer( oci, ALTREF_FRAME )?VP8_ALTR_FRAME:0) | + (vp8dx_references_buffer( oci, GOLDEN_FRAME )?VP8_GOLD_FRAME:0) | + (vp8dx_references_buffer( oci, LAST_FRAME )?VP8_LAST_FRAME:0); + + return VPX_CODEC_OK; + } + else + return VPX_CODEC_INVALID_PARAM; +} + +static vpx_codec_err_t vp8_get_frame_corrupted(vpx_codec_alg_priv_t *ctx, + va_list args) +{ + + int *corrupted = va_arg(args, int *); + VP8D_COMP *pbi = (VP8D_COMP *)ctx->yv12_frame_buffers.pbi[0]; + + if (corrupted && pbi) + { + const YV12_BUFFER_CONFIG *const frame = pbi->common.frame_to_show; + if (frame == NULL) return VPX_CODEC_ERROR; + *corrupted = frame->corrupted; + return VPX_CODEC_OK; + } + else + return VPX_CODEC_INVALID_PARAM; + +} + +static vpx_codec_err_t vp8_set_decryptor(vpx_codec_alg_priv_t *ctx, + va_list args) +{ + vpx_decrypt_init *init = va_arg(args, vpx_decrypt_init *); + + if (init) + { + ctx->decrypt_cb = init->decrypt_cb; + ctx->decrypt_state = init->decrypt_state; + } + else + { + ctx->decrypt_cb = NULL; + ctx->decrypt_state = NULL; + } + return VPX_CODEC_OK; +} + +vpx_codec_ctrl_fn_map_t vp8_ctf_maps[] = +{ + {VP8_SET_REFERENCE, vp8_set_reference}, + {VP8_COPY_REFERENCE, vp8_get_reference}, + {VP8_SET_POSTPROC, vp8_set_postproc}, + {VP8_SET_DBG_COLOR_REF_FRAME, vp8_set_dbg_color_ref_frame}, + {VP8_SET_DBG_COLOR_MB_MODES, vp8_set_dbg_color_mb_modes}, + {VP8_SET_DBG_COLOR_B_MODES, vp8_set_dbg_color_b_modes}, + {VP8_SET_DBG_DISPLAY_MV, vp8_set_dbg_display_mv}, + {VP8D_GET_LAST_REF_UPDATES, vp8_get_last_ref_updates}, + {VP8D_GET_FRAME_CORRUPTED, vp8_get_frame_corrupted}, + {VP8D_GET_LAST_REF_USED, vp8_get_last_ref_frame}, + {VPXD_SET_DECRYPTOR, vp8_set_decryptor}, + { -1, NULL}, +}; + + +#ifndef VERSION_STRING +#define VERSION_STRING +#endif +CODEC_INTERFACE(vpx_codec_vp8_dx) = +{ + "WebM Project VP8 Decoder" VERSION_STRING, + VPX_CODEC_INTERNAL_ABI_VERSION, + VPX_CODEC_CAP_DECODER | VP8_CAP_POSTPROC | VP8_CAP_ERROR_CONCEALMENT | + VPX_CODEC_CAP_INPUT_FRAGMENTS, + /* vpx_codec_caps_t caps; */ + vp8_init, /* vpx_codec_init_fn_t init; */ + vp8_destroy, /* vpx_codec_destroy_fn_t destroy; */ + vp8_ctf_maps, /* vpx_codec_ctrl_fn_map_t *ctrl_maps; */ + { + vp8_peek_si, /* vpx_codec_peek_si_fn_t peek_si; */ + vp8_get_si, /* vpx_codec_get_si_fn_t get_si; */ + vp8_decode, /* vpx_codec_decode_fn_t decode; */ + vp8_get_frame, /* vpx_codec_frame_get_fn_t frame_get; */ + NULL, + }, + { /* encoder functions */ + 0, + NULL, /* vpx_codec_enc_cfg_map_t */ + NULL, /* vpx_codec_encode_fn_t */ + NULL, /* vpx_codec_get_cx_data_fn_t */ + NULL, /* vpx_codec_enc_config_set_fn_t */ + NULL, /* vpx_codec_get_global_headers_fn_t */ + NULL, /* vpx_codec_get_preview_frame_fn_t */ + NULL /* vpx_codec_enc_mr_get_mem_loc_fn_t */ + } +}; diff --git a/thirdparty/libvpx/vp9/common/arm/neon/vp9_iht4x4_add_neon.c b/thirdparty/libvpx/vp9/common/arm/neon/vp9_iht4x4_add_neon.c new file mode 100644 index 0000000000..1761fada2f --- /dev/null +++ b/thirdparty/libvpx/vp9/common/arm/neon/vp9_iht4x4_add_neon.c @@ -0,0 +1,248 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> +#include <assert.h> + +#include "./vp9_rtcd.h" +#include "./vpx_config.h" +#include "vp9/common/vp9_common.h" + +static int16_t sinpi_1_9 = 0x14a3; +static int16_t sinpi_2_9 = 0x26c9; +static int16_t sinpi_3_9 = 0x3441; +static int16_t sinpi_4_9 = 0x3b6c; +static int16_t cospi_8_64 = 0x3b21; +static int16_t cospi_16_64 = 0x2d41; +static int16_t cospi_24_64 = 0x187e; + +static INLINE void TRANSPOSE4X4( + int16x8_t *q8s16, + int16x8_t *q9s16) { + int32x4_t q8s32, q9s32; + int16x4x2_t d0x2s16, d1x2s16; + int32x4x2_t q0x2s32; + + d0x2s16 = vtrn_s16(vget_low_s16(*q8s16), vget_high_s16(*q8s16)); + d1x2s16 = vtrn_s16(vget_low_s16(*q9s16), vget_high_s16(*q9s16)); + + q8s32 = vreinterpretq_s32_s16(vcombine_s16(d0x2s16.val[0], d0x2s16.val[1])); + q9s32 = vreinterpretq_s32_s16(vcombine_s16(d1x2s16.val[0], d1x2s16.val[1])); + q0x2s32 = vtrnq_s32(q8s32, q9s32); + + *q8s16 = vreinterpretq_s16_s32(q0x2s32.val[0]); + *q9s16 = vreinterpretq_s16_s32(q0x2s32.val[1]); + return; +} + +static INLINE void GENERATE_COSINE_CONSTANTS( + int16x4_t *d0s16, + int16x4_t *d1s16, + int16x4_t *d2s16) { + *d0s16 = vdup_n_s16(cospi_8_64); + *d1s16 = vdup_n_s16(cospi_16_64); + *d2s16 = vdup_n_s16(cospi_24_64); + return; +} + +static INLINE void GENERATE_SINE_CONSTANTS( + int16x4_t *d3s16, + int16x4_t *d4s16, + int16x4_t *d5s16, + int16x8_t *q3s16) { + *d3s16 = vdup_n_s16(sinpi_1_9); + *d4s16 = vdup_n_s16(sinpi_2_9); + *q3s16 = vdupq_n_s16(sinpi_3_9); + *d5s16 = vdup_n_s16(sinpi_4_9); + return; +} + +static INLINE void IDCT4x4_1D( + int16x4_t *d0s16, + int16x4_t *d1s16, + int16x4_t *d2s16, + int16x8_t *q8s16, + int16x8_t *q9s16) { + int16x4_t d16s16, d17s16, d18s16, d19s16, d23s16, d24s16; + int16x4_t d26s16, d27s16, d28s16, d29s16; + int32x4_t q10s32, q13s32, q14s32, q15s32; + int16x8_t q13s16, q14s16; + + d16s16 = vget_low_s16(*q8s16); + d17s16 = vget_high_s16(*q8s16); + d18s16 = vget_low_s16(*q9s16); + d19s16 = vget_high_s16(*q9s16); + + d23s16 = vadd_s16(d16s16, d18s16); + d24s16 = vsub_s16(d16s16, d18s16); + + q15s32 = vmull_s16(d17s16, *d2s16); + q10s32 = vmull_s16(d17s16, *d0s16); + q13s32 = vmull_s16(d23s16, *d1s16); + q14s32 = vmull_s16(d24s16, *d1s16); + q15s32 = vmlsl_s16(q15s32, d19s16, *d0s16); + q10s32 = vmlal_s16(q10s32, d19s16, *d2s16); + + d26s16 = vqrshrn_n_s32(q13s32, 14); + d27s16 = vqrshrn_n_s32(q14s32, 14); + d29s16 = vqrshrn_n_s32(q15s32, 14); + d28s16 = vqrshrn_n_s32(q10s32, 14); + + q13s16 = vcombine_s16(d26s16, d27s16); + q14s16 = vcombine_s16(d28s16, d29s16); + *q8s16 = vaddq_s16(q13s16, q14s16); + *q9s16 = vsubq_s16(q13s16, q14s16); + *q9s16 = vcombine_s16(vget_high_s16(*q9s16), + vget_low_s16(*q9s16)); // vswp + return; +} + +static INLINE void IADST4x4_1D( + int16x4_t *d3s16, + int16x4_t *d4s16, + int16x4_t *d5s16, + int16x8_t *q3s16, + int16x8_t *q8s16, + int16x8_t *q9s16) { + int16x4_t d6s16, d16s16, d17s16, d18s16, d19s16; + int32x4_t q8s32, q9s32, q10s32, q11s32, q12s32, q13s32, q14s32, q15s32; + + d6s16 = vget_low_s16(*q3s16); + + d16s16 = vget_low_s16(*q8s16); + d17s16 = vget_high_s16(*q8s16); + d18s16 = vget_low_s16(*q9s16); + d19s16 = vget_high_s16(*q9s16); + + q10s32 = vmull_s16(*d3s16, d16s16); + q11s32 = vmull_s16(*d4s16, d16s16); + q12s32 = vmull_s16(d6s16, d17s16); + q13s32 = vmull_s16(*d5s16, d18s16); + q14s32 = vmull_s16(*d3s16, d18s16); + q15s32 = vmovl_s16(d16s16); + q15s32 = vaddw_s16(q15s32, d19s16); + q8s32 = vmull_s16(*d4s16, d19s16); + q15s32 = vsubw_s16(q15s32, d18s16); + q9s32 = vmull_s16(*d5s16, d19s16); + + q10s32 = vaddq_s32(q10s32, q13s32); + q10s32 = vaddq_s32(q10s32, q8s32); + q11s32 = vsubq_s32(q11s32, q14s32); + q8s32 = vdupq_n_s32(sinpi_3_9); + q11s32 = vsubq_s32(q11s32, q9s32); + q15s32 = vmulq_s32(q15s32, q8s32); + + q13s32 = vaddq_s32(q10s32, q12s32); + q10s32 = vaddq_s32(q10s32, q11s32); + q14s32 = vaddq_s32(q11s32, q12s32); + q10s32 = vsubq_s32(q10s32, q12s32); + + d16s16 = vqrshrn_n_s32(q13s32, 14); + d17s16 = vqrshrn_n_s32(q14s32, 14); + d18s16 = vqrshrn_n_s32(q15s32, 14); + d19s16 = vqrshrn_n_s32(q10s32, 14); + + *q8s16 = vcombine_s16(d16s16, d17s16); + *q9s16 = vcombine_s16(d18s16, d19s16); + return; +} + +void vp9_iht4x4_16_add_neon(const tran_low_t *input, uint8_t *dest, + int dest_stride, int tx_type) { + uint8x8_t d26u8, d27u8; + int16x4_t d0s16, d1s16, d2s16, d3s16, d4s16, d5s16; + uint32x2_t d26u32, d27u32; + int16x8_t q3s16, q8s16, q9s16; + uint16x8_t q8u16, q9u16; + + d26u32 = d27u32 = vdup_n_u32(0); + + q8s16 = vld1q_s16(input); + q9s16 = vld1q_s16(input + 8); + + TRANSPOSE4X4(&q8s16, &q9s16); + + switch (tx_type) { + case 0: // idct_idct is not supported. Fall back to C + vp9_iht4x4_16_add_c(input, dest, dest_stride, tx_type); + return; + break; + case 1: // iadst_idct + // generate constants + GENERATE_COSINE_CONSTANTS(&d0s16, &d1s16, &d2s16); + GENERATE_SINE_CONSTANTS(&d3s16, &d4s16, &d5s16, &q3s16); + + // first transform rows + IDCT4x4_1D(&d0s16, &d1s16, &d2s16, &q8s16, &q9s16); + + // transpose the matrix + TRANSPOSE4X4(&q8s16, &q9s16); + + // then transform columns + IADST4x4_1D(&d3s16, &d4s16, &d5s16, &q3s16, &q8s16, &q9s16); + break; + case 2: // idct_iadst + // generate constantsyy + GENERATE_COSINE_CONSTANTS(&d0s16, &d1s16, &d2s16); + GENERATE_SINE_CONSTANTS(&d3s16, &d4s16, &d5s16, &q3s16); + + // first transform rows + IADST4x4_1D(&d3s16, &d4s16, &d5s16, &q3s16, &q8s16, &q9s16); + + // transpose the matrix + TRANSPOSE4X4(&q8s16, &q9s16); + + // then transform columns + IDCT4x4_1D(&d0s16, &d1s16, &d2s16, &q8s16, &q9s16); + break; + case 3: // iadst_iadst + // generate constants + GENERATE_SINE_CONSTANTS(&d3s16, &d4s16, &d5s16, &q3s16); + + // first transform rows + IADST4x4_1D(&d3s16, &d4s16, &d5s16, &q3s16, &q8s16, &q9s16); + + // transpose the matrix + TRANSPOSE4X4(&q8s16, &q9s16); + + // then transform columns + IADST4x4_1D(&d3s16, &d4s16, &d5s16, &q3s16, &q8s16, &q9s16); + break; + default: // iadst_idct + assert(0); + break; + } + + q8s16 = vrshrq_n_s16(q8s16, 4); + q9s16 = vrshrq_n_s16(q9s16, 4); + + d26u32 = vld1_lane_u32((const uint32_t *)dest, d26u32, 0); + dest += dest_stride; + d26u32 = vld1_lane_u32((const uint32_t *)dest, d26u32, 1); + dest += dest_stride; + d27u32 = vld1_lane_u32((const uint32_t *)dest, d27u32, 0); + dest += dest_stride; + d27u32 = vld1_lane_u32((const uint32_t *)dest, d27u32, 1); + + q8u16 = vaddw_u8(vreinterpretq_u16_s16(q8s16), vreinterpret_u8_u32(d26u32)); + q9u16 = vaddw_u8(vreinterpretq_u16_s16(q9s16), vreinterpret_u8_u32(d27u32)); + + d26u8 = vqmovun_s16(vreinterpretq_s16_u16(q8u16)); + d27u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16)); + + vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d27u8), 1); + dest -= dest_stride; + vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d27u8), 0); + dest -= dest_stride; + vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d26u8), 1); + dest -= dest_stride; + vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d26u8), 0); + return; +} diff --git a/thirdparty/libvpx/vp9/common/arm/neon/vp9_iht8x8_add_neon.c b/thirdparty/libvpx/vp9/common/arm/neon/vp9_iht8x8_add_neon.c new file mode 100644 index 0000000000..04b342c3d3 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/arm/neon/vp9_iht8x8_add_neon.c @@ -0,0 +1,624 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> +#include <assert.h> + +#include "./vp9_rtcd.h" +#include "./vpx_config.h" +#include "vp9/common/vp9_common.h" + +static int16_t cospi_2_64 = 16305; +static int16_t cospi_4_64 = 16069; +static int16_t cospi_6_64 = 15679; +static int16_t cospi_8_64 = 15137; +static int16_t cospi_10_64 = 14449; +static int16_t cospi_12_64 = 13623; +static int16_t cospi_14_64 = 12665; +static int16_t cospi_16_64 = 11585; +static int16_t cospi_18_64 = 10394; +static int16_t cospi_20_64 = 9102; +static int16_t cospi_22_64 = 7723; +static int16_t cospi_24_64 = 6270; +static int16_t cospi_26_64 = 4756; +static int16_t cospi_28_64 = 3196; +static int16_t cospi_30_64 = 1606; + +static INLINE void TRANSPOSE8X8( + int16x8_t *q8s16, + int16x8_t *q9s16, + int16x8_t *q10s16, + int16x8_t *q11s16, + int16x8_t *q12s16, + int16x8_t *q13s16, + int16x8_t *q14s16, + int16x8_t *q15s16) { + int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16; + int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16; + int32x4x2_t q0x2s32, q1x2s32, q2x2s32, q3x2s32; + int16x8x2_t q0x2s16, q1x2s16, q2x2s16, q3x2s16; + + d16s16 = vget_low_s16(*q8s16); + d17s16 = vget_high_s16(*q8s16); + d18s16 = vget_low_s16(*q9s16); + d19s16 = vget_high_s16(*q9s16); + d20s16 = vget_low_s16(*q10s16); + d21s16 = vget_high_s16(*q10s16); + d22s16 = vget_low_s16(*q11s16); + d23s16 = vget_high_s16(*q11s16); + d24s16 = vget_low_s16(*q12s16); + d25s16 = vget_high_s16(*q12s16); + d26s16 = vget_low_s16(*q13s16); + d27s16 = vget_high_s16(*q13s16); + d28s16 = vget_low_s16(*q14s16); + d29s16 = vget_high_s16(*q14s16); + d30s16 = vget_low_s16(*q15s16); + d31s16 = vget_high_s16(*q15s16); + + *q8s16 = vcombine_s16(d16s16, d24s16); // vswp d17, d24 + *q9s16 = vcombine_s16(d18s16, d26s16); // vswp d19, d26 + *q10s16 = vcombine_s16(d20s16, d28s16); // vswp d21, d28 + *q11s16 = vcombine_s16(d22s16, d30s16); // vswp d23, d30 + *q12s16 = vcombine_s16(d17s16, d25s16); + *q13s16 = vcombine_s16(d19s16, d27s16); + *q14s16 = vcombine_s16(d21s16, d29s16); + *q15s16 = vcombine_s16(d23s16, d31s16); + + q0x2s32 = vtrnq_s32(vreinterpretq_s32_s16(*q8s16), + vreinterpretq_s32_s16(*q10s16)); + q1x2s32 = vtrnq_s32(vreinterpretq_s32_s16(*q9s16), + vreinterpretq_s32_s16(*q11s16)); + q2x2s32 = vtrnq_s32(vreinterpretq_s32_s16(*q12s16), + vreinterpretq_s32_s16(*q14s16)); + q3x2s32 = vtrnq_s32(vreinterpretq_s32_s16(*q13s16), + vreinterpretq_s32_s16(*q15s16)); + + q0x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[0]), // q8 + vreinterpretq_s16_s32(q1x2s32.val[0])); // q9 + q1x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[1]), // q10 + vreinterpretq_s16_s32(q1x2s32.val[1])); // q11 + q2x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[0]), // q12 + vreinterpretq_s16_s32(q3x2s32.val[0])); // q13 + q3x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[1]), // q14 + vreinterpretq_s16_s32(q3x2s32.val[1])); // q15 + + *q8s16 = q0x2s16.val[0]; + *q9s16 = q0x2s16.val[1]; + *q10s16 = q1x2s16.val[0]; + *q11s16 = q1x2s16.val[1]; + *q12s16 = q2x2s16.val[0]; + *q13s16 = q2x2s16.val[1]; + *q14s16 = q3x2s16.val[0]; + *q15s16 = q3x2s16.val[1]; + return; +} + +static INLINE void IDCT8x8_1D( + int16x8_t *q8s16, + int16x8_t *q9s16, + int16x8_t *q10s16, + int16x8_t *q11s16, + int16x8_t *q12s16, + int16x8_t *q13s16, + int16x8_t *q14s16, + int16x8_t *q15s16) { + int16x4_t d0s16, d1s16, d2s16, d3s16; + int16x4_t d8s16, d9s16, d10s16, d11s16, d12s16, d13s16, d14s16, d15s16; + int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16; + int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16; + int16x8_t q0s16, q1s16, q2s16, q3s16, q4s16, q5s16, q6s16, q7s16; + int32x4_t q2s32, q3s32, q5s32, q6s32, q8s32, q9s32; + int32x4_t q10s32, q11s32, q12s32, q13s32, q15s32; + + d0s16 = vdup_n_s16(cospi_28_64); + d1s16 = vdup_n_s16(cospi_4_64); + d2s16 = vdup_n_s16(cospi_12_64); + d3s16 = vdup_n_s16(cospi_20_64); + + d16s16 = vget_low_s16(*q8s16); + d17s16 = vget_high_s16(*q8s16); + d18s16 = vget_low_s16(*q9s16); + d19s16 = vget_high_s16(*q9s16); + d20s16 = vget_low_s16(*q10s16); + d21s16 = vget_high_s16(*q10s16); + d22s16 = vget_low_s16(*q11s16); + d23s16 = vget_high_s16(*q11s16); + d24s16 = vget_low_s16(*q12s16); + d25s16 = vget_high_s16(*q12s16); + d26s16 = vget_low_s16(*q13s16); + d27s16 = vget_high_s16(*q13s16); + d28s16 = vget_low_s16(*q14s16); + d29s16 = vget_high_s16(*q14s16); + d30s16 = vget_low_s16(*q15s16); + d31s16 = vget_high_s16(*q15s16); + + q2s32 = vmull_s16(d18s16, d0s16); + q3s32 = vmull_s16(d19s16, d0s16); + q5s32 = vmull_s16(d26s16, d2s16); + q6s32 = vmull_s16(d27s16, d2s16); + + q2s32 = vmlsl_s16(q2s32, d30s16, d1s16); + q3s32 = vmlsl_s16(q3s32, d31s16, d1s16); + q5s32 = vmlsl_s16(q5s32, d22s16, d3s16); + q6s32 = vmlsl_s16(q6s32, d23s16, d3s16); + + d8s16 = vqrshrn_n_s32(q2s32, 14); + d9s16 = vqrshrn_n_s32(q3s32, 14); + d10s16 = vqrshrn_n_s32(q5s32, 14); + d11s16 = vqrshrn_n_s32(q6s32, 14); + q4s16 = vcombine_s16(d8s16, d9s16); + q5s16 = vcombine_s16(d10s16, d11s16); + + q2s32 = vmull_s16(d18s16, d1s16); + q3s32 = vmull_s16(d19s16, d1s16); + q9s32 = vmull_s16(d26s16, d3s16); + q13s32 = vmull_s16(d27s16, d3s16); + + q2s32 = vmlal_s16(q2s32, d30s16, d0s16); + q3s32 = vmlal_s16(q3s32, d31s16, d0s16); + q9s32 = vmlal_s16(q9s32, d22s16, d2s16); + q13s32 = vmlal_s16(q13s32, d23s16, d2s16); + + d14s16 = vqrshrn_n_s32(q2s32, 14); + d15s16 = vqrshrn_n_s32(q3s32, 14); + d12s16 = vqrshrn_n_s32(q9s32, 14); + d13s16 = vqrshrn_n_s32(q13s32, 14); + q6s16 = vcombine_s16(d12s16, d13s16); + q7s16 = vcombine_s16(d14s16, d15s16); + + d0s16 = vdup_n_s16(cospi_16_64); + + q2s32 = vmull_s16(d16s16, d0s16); + q3s32 = vmull_s16(d17s16, d0s16); + q13s32 = vmull_s16(d16s16, d0s16); + q15s32 = vmull_s16(d17s16, d0s16); + + q2s32 = vmlal_s16(q2s32, d24s16, d0s16); + q3s32 = vmlal_s16(q3s32, d25s16, d0s16); + q13s32 = vmlsl_s16(q13s32, d24s16, d0s16); + q15s32 = vmlsl_s16(q15s32, d25s16, d0s16); + + d0s16 = vdup_n_s16(cospi_24_64); + d1s16 = vdup_n_s16(cospi_8_64); + + d18s16 = vqrshrn_n_s32(q2s32, 14); + d19s16 = vqrshrn_n_s32(q3s32, 14); + d22s16 = vqrshrn_n_s32(q13s32, 14); + d23s16 = vqrshrn_n_s32(q15s32, 14); + *q9s16 = vcombine_s16(d18s16, d19s16); + *q11s16 = vcombine_s16(d22s16, d23s16); + + q2s32 = vmull_s16(d20s16, d0s16); + q3s32 = vmull_s16(d21s16, d0s16); + q8s32 = vmull_s16(d20s16, d1s16); + q12s32 = vmull_s16(d21s16, d1s16); + + q2s32 = vmlsl_s16(q2s32, d28s16, d1s16); + q3s32 = vmlsl_s16(q3s32, d29s16, d1s16); + q8s32 = vmlal_s16(q8s32, d28s16, d0s16); + q12s32 = vmlal_s16(q12s32, d29s16, d0s16); + + d26s16 = vqrshrn_n_s32(q2s32, 14); + d27s16 = vqrshrn_n_s32(q3s32, 14); + d30s16 = vqrshrn_n_s32(q8s32, 14); + d31s16 = vqrshrn_n_s32(q12s32, 14); + *q13s16 = vcombine_s16(d26s16, d27s16); + *q15s16 = vcombine_s16(d30s16, d31s16); + + q0s16 = vaddq_s16(*q9s16, *q15s16); + q1s16 = vaddq_s16(*q11s16, *q13s16); + q2s16 = vsubq_s16(*q11s16, *q13s16); + q3s16 = vsubq_s16(*q9s16, *q15s16); + + *q13s16 = vsubq_s16(q4s16, q5s16); + q4s16 = vaddq_s16(q4s16, q5s16); + *q14s16 = vsubq_s16(q7s16, q6s16); + q7s16 = vaddq_s16(q7s16, q6s16); + d26s16 = vget_low_s16(*q13s16); + d27s16 = vget_high_s16(*q13s16); + d28s16 = vget_low_s16(*q14s16); + d29s16 = vget_high_s16(*q14s16); + + d16s16 = vdup_n_s16(cospi_16_64); + + q9s32 = vmull_s16(d28s16, d16s16); + q10s32 = vmull_s16(d29s16, d16s16); + q11s32 = vmull_s16(d28s16, d16s16); + q12s32 = vmull_s16(d29s16, d16s16); + + q9s32 = vmlsl_s16(q9s32, d26s16, d16s16); + q10s32 = vmlsl_s16(q10s32, d27s16, d16s16); + q11s32 = vmlal_s16(q11s32, d26s16, d16s16); + q12s32 = vmlal_s16(q12s32, d27s16, d16s16); + + d10s16 = vqrshrn_n_s32(q9s32, 14); + d11s16 = vqrshrn_n_s32(q10s32, 14); + d12s16 = vqrshrn_n_s32(q11s32, 14); + d13s16 = vqrshrn_n_s32(q12s32, 14); + q5s16 = vcombine_s16(d10s16, d11s16); + q6s16 = vcombine_s16(d12s16, d13s16); + + *q8s16 = vaddq_s16(q0s16, q7s16); + *q9s16 = vaddq_s16(q1s16, q6s16); + *q10s16 = vaddq_s16(q2s16, q5s16); + *q11s16 = vaddq_s16(q3s16, q4s16); + *q12s16 = vsubq_s16(q3s16, q4s16); + *q13s16 = vsubq_s16(q2s16, q5s16); + *q14s16 = vsubq_s16(q1s16, q6s16); + *q15s16 = vsubq_s16(q0s16, q7s16); + return; +} + +static INLINE void IADST8X8_1D( + int16x8_t *q8s16, + int16x8_t *q9s16, + int16x8_t *q10s16, + int16x8_t *q11s16, + int16x8_t *q12s16, + int16x8_t *q13s16, + int16x8_t *q14s16, + int16x8_t *q15s16) { + int16x4_t d0s16, d1s16, d2s16, d3s16, d4s16, d5s16, d6s16, d7s16; + int16x4_t d8s16, d9s16, d10s16, d11s16, d12s16, d13s16, d14s16, d15s16; + int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16; + int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16; + int16x8_t q2s16, q4s16, q5s16, q6s16; + int32x4_t q0s32, q1s32, q2s32, q3s32, q4s32, q5s32, q6s32, q7s32, q8s32; + int32x4_t q9s32, q10s32, q11s32, q12s32, q13s32, q14s32, q15s32; + + d16s16 = vget_low_s16(*q8s16); + d17s16 = vget_high_s16(*q8s16); + d18s16 = vget_low_s16(*q9s16); + d19s16 = vget_high_s16(*q9s16); + d20s16 = vget_low_s16(*q10s16); + d21s16 = vget_high_s16(*q10s16); + d22s16 = vget_low_s16(*q11s16); + d23s16 = vget_high_s16(*q11s16); + d24s16 = vget_low_s16(*q12s16); + d25s16 = vget_high_s16(*q12s16); + d26s16 = vget_low_s16(*q13s16); + d27s16 = vget_high_s16(*q13s16); + d28s16 = vget_low_s16(*q14s16); + d29s16 = vget_high_s16(*q14s16); + d30s16 = vget_low_s16(*q15s16); + d31s16 = vget_high_s16(*q15s16); + + d14s16 = vdup_n_s16(cospi_2_64); + d15s16 = vdup_n_s16(cospi_30_64); + + q1s32 = vmull_s16(d30s16, d14s16); + q2s32 = vmull_s16(d31s16, d14s16); + q3s32 = vmull_s16(d30s16, d15s16); + q4s32 = vmull_s16(d31s16, d15s16); + + d30s16 = vdup_n_s16(cospi_18_64); + d31s16 = vdup_n_s16(cospi_14_64); + + q1s32 = vmlal_s16(q1s32, d16s16, d15s16); + q2s32 = vmlal_s16(q2s32, d17s16, d15s16); + q3s32 = vmlsl_s16(q3s32, d16s16, d14s16); + q4s32 = vmlsl_s16(q4s32, d17s16, d14s16); + + q5s32 = vmull_s16(d22s16, d30s16); + q6s32 = vmull_s16(d23s16, d30s16); + q7s32 = vmull_s16(d22s16, d31s16); + q8s32 = vmull_s16(d23s16, d31s16); + + q5s32 = vmlal_s16(q5s32, d24s16, d31s16); + q6s32 = vmlal_s16(q6s32, d25s16, d31s16); + q7s32 = vmlsl_s16(q7s32, d24s16, d30s16); + q8s32 = vmlsl_s16(q8s32, d25s16, d30s16); + + q11s32 = vaddq_s32(q1s32, q5s32); + q12s32 = vaddq_s32(q2s32, q6s32); + q1s32 = vsubq_s32(q1s32, q5s32); + q2s32 = vsubq_s32(q2s32, q6s32); + + d22s16 = vqrshrn_n_s32(q11s32, 14); + d23s16 = vqrshrn_n_s32(q12s32, 14); + *q11s16 = vcombine_s16(d22s16, d23s16); + + q12s32 = vaddq_s32(q3s32, q7s32); + q15s32 = vaddq_s32(q4s32, q8s32); + q3s32 = vsubq_s32(q3s32, q7s32); + q4s32 = vsubq_s32(q4s32, q8s32); + + d2s16 = vqrshrn_n_s32(q1s32, 14); + d3s16 = vqrshrn_n_s32(q2s32, 14); + d24s16 = vqrshrn_n_s32(q12s32, 14); + d25s16 = vqrshrn_n_s32(q15s32, 14); + d6s16 = vqrshrn_n_s32(q3s32, 14); + d7s16 = vqrshrn_n_s32(q4s32, 14); + *q12s16 = vcombine_s16(d24s16, d25s16); + + d0s16 = vdup_n_s16(cospi_10_64); + d1s16 = vdup_n_s16(cospi_22_64); + q4s32 = vmull_s16(d26s16, d0s16); + q5s32 = vmull_s16(d27s16, d0s16); + q2s32 = vmull_s16(d26s16, d1s16); + q6s32 = vmull_s16(d27s16, d1s16); + + d30s16 = vdup_n_s16(cospi_26_64); + d31s16 = vdup_n_s16(cospi_6_64); + + q4s32 = vmlal_s16(q4s32, d20s16, d1s16); + q5s32 = vmlal_s16(q5s32, d21s16, d1s16); + q2s32 = vmlsl_s16(q2s32, d20s16, d0s16); + q6s32 = vmlsl_s16(q6s32, d21s16, d0s16); + + q0s32 = vmull_s16(d18s16, d30s16); + q13s32 = vmull_s16(d19s16, d30s16); + + q0s32 = vmlal_s16(q0s32, d28s16, d31s16); + q13s32 = vmlal_s16(q13s32, d29s16, d31s16); + + q10s32 = vmull_s16(d18s16, d31s16); + q9s32 = vmull_s16(d19s16, d31s16); + + q10s32 = vmlsl_s16(q10s32, d28s16, d30s16); + q9s32 = vmlsl_s16(q9s32, d29s16, d30s16); + + q14s32 = vaddq_s32(q2s32, q10s32); + q15s32 = vaddq_s32(q6s32, q9s32); + q2s32 = vsubq_s32(q2s32, q10s32); + q6s32 = vsubq_s32(q6s32, q9s32); + + d28s16 = vqrshrn_n_s32(q14s32, 14); + d29s16 = vqrshrn_n_s32(q15s32, 14); + d4s16 = vqrshrn_n_s32(q2s32, 14); + d5s16 = vqrshrn_n_s32(q6s32, 14); + *q14s16 = vcombine_s16(d28s16, d29s16); + + q9s32 = vaddq_s32(q4s32, q0s32); + q10s32 = vaddq_s32(q5s32, q13s32); + q4s32 = vsubq_s32(q4s32, q0s32); + q5s32 = vsubq_s32(q5s32, q13s32); + + d30s16 = vdup_n_s16(cospi_8_64); + d31s16 = vdup_n_s16(cospi_24_64); + + d18s16 = vqrshrn_n_s32(q9s32, 14); + d19s16 = vqrshrn_n_s32(q10s32, 14); + d8s16 = vqrshrn_n_s32(q4s32, 14); + d9s16 = vqrshrn_n_s32(q5s32, 14); + *q9s16 = vcombine_s16(d18s16, d19s16); + + q5s32 = vmull_s16(d2s16, d30s16); + q6s32 = vmull_s16(d3s16, d30s16); + q7s32 = vmull_s16(d2s16, d31s16); + q0s32 = vmull_s16(d3s16, d31s16); + + q5s32 = vmlal_s16(q5s32, d6s16, d31s16); + q6s32 = vmlal_s16(q6s32, d7s16, d31s16); + q7s32 = vmlsl_s16(q7s32, d6s16, d30s16); + q0s32 = vmlsl_s16(q0s32, d7s16, d30s16); + + q1s32 = vmull_s16(d4s16, d30s16); + q3s32 = vmull_s16(d5s16, d30s16); + q10s32 = vmull_s16(d4s16, d31s16); + q2s32 = vmull_s16(d5s16, d31s16); + + q1s32 = vmlsl_s16(q1s32, d8s16, d31s16); + q3s32 = vmlsl_s16(q3s32, d9s16, d31s16); + q10s32 = vmlal_s16(q10s32, d8s16, d30s16); + q2s32 = vmlal_s16(q2s32, d9s16, d30s16); + + *q8s16 = vaddq_s16(*q11s16, *q9s16); + *q11s16 = vsubq_s16(*q11s16, *q9s16); + q4s16 = vaddq_s16(*q12s16, *q14s16); + *q12s16 = vsubq_s16(*q12s16, *q14s16); + + q14s32 = vaddq_s32(q5s32, q1s32); + q15s32 = vaddq_s32(q6s32, q3s32); + q5s32 = vsubq_s32(q5s32, q1s32); + q6s32 = vsubq_s32(q6s32, q3s32); + + d18s16 = vqrshrn_n_s32(q14s32, 14); + d19s16 = vqrshrn_n_s32(q15s32, 14); + d10s16 = vqrshrn_n_s32(q5s32, 14); + d11s16 = vqrshrn_n_s32(q6s32, 14); + *q9s16 = vcombine_s16(d18s16, d19s16); + + q1s32 = vaddq_s32(q7s32, q10s32); + q3s32 = vaddq_s32(q0s32, q2s32); + q7s32 = vsubq_s32(q7s32, q10s32); + q0s32 = vsubq_s32(q0s32, q2s32); + + d28s16 = vqrshrn_n_s32(q1s32, 14); + d29s16 = vqrshrn_n_s32(q3s32, 14); + d14s16 = vqrshrn_n_s32(q7s32, 14); + d15s16 = vqrshrn_n_s32(q0s32, 14); + *q14s16 = vcombine_s16(d28s16, d29s16); + + d30s16 = vdup_n_s16(cospi_16_64); + + d22s16 = vget_low_s16(*q11s16); + d23s16 = vget_high_s16(*q11s16); + q2s32 = vmull_s16(d22s16, d30s16); + q3s32 = vmull_s16(d23s16, d30s16); + q13s32 = vmull_s16(d22s16, d30s16); + q1s32 = vmull_s16(d23s16, d30s16); + + d24s16 = vget_low_s16(*q12s16); + d25s16 = vget_high_s16(*q12s16); + q2s32 = vmlal_s16(q2s32, d24s16, d30s16); + q3s32 = vmlal_s16(q3s32, d25s16, d30s16); + q13s32 = vmlsl_s16(q13s32, d24s16, d30s16); + q1s32 = vmlsl_s16(q1s32, d25s16, d30s16); + + d4s16 = vqrshrn_n_s32(q2s32, 14); + d5s16 = vqrshrn_n_s32(q3s32, 14); + d24s16 = vqrshrn_n_s32(q13s32, 14); + d25s16 = vqrshrn_n_s32(q1s32, 14); + q2s16 = vcombine_s16(d4s16, d5s16); + *q12s16 = vcombine_s16(d24s16, d25s16); + + q13s32 = vmull_s16(d10s16, d30s16); + q1s32 = vmull_s16(d11s16, d30s16); + q11s32 = vmull_s16(d10s16, d30s16); + q0s32 = vmull_s16(d11s16, d30s16); + + q13s32 = vmlal_s16(q13s32, d14s16, d30s16); + q1s32 = vmlal_s16(q1s32, d15s16, d30s16); + q11s32 = vmlsl_s16(q11s32, d14s16, d30s16); + q0s32 = vmlsl_s16(q0s32, d15s16, d30s16); + + d20s16 = vqrshrn_n_s32(q13s32, 14); + d21s16 = vqrshrn_n_s32(q1s32, 14); + d12s16 = vqrshrn_n_s32(q11s32, 14); + d13s16 = vqrshrn_n_s32(q0s32, 14); + *q10s16 = vcombine_s16(d20s16, d21s16); + q6s16 = vcombine_s16(d12s16, d13s16); + + q5s16 = vdupq_n_s16(0); + + *q9s16 = vsubq_s16(q5s16, *q9s16); + *q11s16 = vsubq_s16(q5s16, q2s16); + *q13s16 = vsubq_s16(q5s16, q6s16); + *q15s16 = vsubq_s16(q5s16, q4s16); + return; +} + +void vp9_iht8x8_64_add_neon(const tran_low_t *input, uint8_t *dest, + int dest_stride, int tx_type) { + int i; + uint8_t *d1, *d2; + uint8x8_t d0u8, d1u8, d2u8, d3u8; + uint64x1_t d0u64, d1u64, d2u64, d3u64; + int16x8_t q8s16, q9s16, q10s16, q11s16, q12s16, q13s16, q14s16, q15s16; + uint16x8_t q8u16, q9u16, q10u16, q11u16; + + q8s16 = vld1q_s16(input); + q9s16 = vld1q_s16(input + 8); + q10s16 = vld1q_s16(input + 8 * 2); + q11s16 = vld1q_s16(input + 8 * 3); + q12s16 = vld1q_s16(input + 8 * 4); + q13s16 = vld1q_s16(input + 8 * 5); + q14s16 = vld1q_s16(input + 8 * 6); + q15s16 = vld1q_s16(input + 8 * 7); + + TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, + &q12s16, &q13s16, &q14s16, &q15s16); + + switch (tx_type) { + case 0: // idct_idct is not supported. Fall back to C + vp9_iht8x8_64_add_c(input, dest, dest_stride, tx_type); + return; + break; + case 1: // iadst_idct + // generate IDCT constants + // GENERATE_IDCT_CONSTANTS + + // first transform rows + IDCT8x8_1D(&q8s16, &q9s16, &q10s16, &q11s16, + &q12s16, &q13s16, &q14s16, &q15s16); + + // transpose the matrix + TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, + &q12s16, &q13s16, &q14s16, &q15s16); + + // generate IADST constants + // GENERATE_IADST_CONSTANTS + + // then transform columns + IADST8X8_1D(&q8s16, &q9s16, &q10s16, &q11s16, + &q12s16, &q13s16, &q14s16, &q15s16); + break; + case 2: // idct_iadst + // generate IADST constants + // GENERATE_IADST_CONSTANTS + + // first transform rows + IADST8X8_1D(&q8s16, &q9s16, &q10s16, &q11s16, + &q12s16, &q13s16, &q14s16, &q15s16); + + // transpose the matrix + TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, + &q12s16, &q13s16, &q14s16, &q15s16); + + // generate IDCT constants + // GENERATE_IDCT_CONSTANTS + + // then transform columns + IDCT8x8_1D(&q8s16, &q9s16, &q10s16, &q11s16, + &q12s16, &q13s16, &q14s16, &q15s16); + break; + case 3: // iadst_iadst + // generate IADST constants + // GENERATE_IADST_CONSTANTS + + // first transform rows + IADST8X8_1D(&q8s16, &q9s16, &q10s16, &q11s16, + &q12s16, &q13s16, &q14s16, &q15s16); + + // transpose the matrix + TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, + &q12s16, &q13s16, &q14s16, &q15s16); + + // then transform columns + IADST8X8_1D(&q8s16, &q9s16, &q10s16, &q11s16, + &q12s16, &q13s16, &q14s16, &q15s16); + break; + default: // iadst_idct + assert(0); + break; + } + + q8s16 = vrshrq_n_s16(q8s16, 5); + q9s16 = vrshrq_n_s16(q9s16, 5); + q10s16 = vrshrq_n_s16(q10s16, 5); + q11s16 = vrshrq_n_s16(q11s16, 5); + q12s16 = vrshrq_n_s16(q12s16, 5); + q13s16 = vrshrq_n_s16(q13s16, 5); + q14s16 = vrshrq_n_s16(q14s16, 5); + q15s16 = vrshrq_n_s16(q15s16, 5); + + for (d1 = d2 = dest, i = 0; i < 2; i++) { + if (i != 0) { + q8s16 = q12s16; + q9s16 = q13s16; + q10s16 = q14s16; + q11s16 = q15s16; + } + + d0u64 = vld1_u64((uint64_t *)d1); + d1 += dest_stride; + d1u64 = vld1_u64((uint64_t *)d1); + d1 += dest_stride; + d2u64 = vld1_u64((uint64_t *)d1); + d1 += dest_stride; + d3u64 = vld1_u64((uint64_t *)d1); + d1 += dest_stride; + + q8u16 = vaddw_u8(vreinterpretq_u16_s16(q8s16), + vreinterpret_u8_u64(d0u64)); + q9u16 = vaddw_u8(vreinterpretq_u16_s16(q9s16), + vreinterpret_u8_u64(d1u64)); + q10u16 = vaddw_u8(vreinterpretq_u16_s16(q10s16), + vreinterpret_u8_u64(d2u64)); + q11u16 = vaddw_u8(vreinterpretq_u16_s16(q11s16), + vreinterpret_u8_u64(d3u64)); + + d0u8 = vqmovun_s16(vreinterpretq_s16_u16(q8u16)); + d1u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16)); + d2u8 = vqmovun_s16(vreinterpretq_s16_u16(q10u16)); + d3u8 = vqmovun_s16(vreinterpretq_s16_u16(q11u16)); + + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d0u8)); + d2 += dest_stride; + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d1u8)); + d2 += dest_stride; + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d2u8)); + d2 += dest_stride; + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d3u8)); + d2 += dest_stride; + } + return; +} diff --git a/thirdparty/libvpx/vp9/common/vp9_alloccommon.c b/thirdparty/libvpx/vp9/common/vp9_alloccommon.c new file mode 100644 index 0000000000..7dd1005d3f --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_alloccommon.c @@ -0,0 +1,201 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "./vpx_config.h" +#include "vpx_mem/vpx_mem.h" + +#include "vp9/common/vp9_alloccommon.h" +#include "vp9/common/vp9_blockd.h" +#include "vp9/common/vp9_entropymode.h" +#include "vp9/common/vp9_entropymv.h" +#include "vp9/common/vp9_onyxc_int.h" + +// TODO(hkuang): Don't need to lock the whole pool after implementing atomic +// frame reference count. +void lock_buffer_pool(BufferPool *const pool) { +#if CONFIG_MULTITHREAD + pthread_mutex_lock(&pool->pool_mutex); +#else + (void)pool; +#endif +} + +void unlock_buffer_pool(BufferPool *const pool) { +#if CONFIG_MULTITHREAD + pthread_mutex_unlock(&pool->pool_mutex); +#else + (void)pool; +#endif +} + +void vp9_set_mb_mi(VP9_COMMON *cm, int width, int height) { + const int aligned_width = ALIGN_POWER_OF_TWO(width, MI_SIZE_LOG2); + const int aligned_height = ALIGN_POWER_OF_TWO(height, MI_SIZE_LOG2); + + cm->mi_cols = aligned_width >> MI_SIZE_LOG2; + cm->mi_rows = aligned_height >> MI_SIZE_LOG2; + cm->mi_stride = calc_mi_size(cm->mi_cols); + + cm->mb_cols = (cm->mi_cols + 1) >> 1; + cm->mb_rows = (cm->mi_rows + 1) >> 1; + cm->MBs = cm->mb_rows * cm->mb_cols; +} + +static int alloc_seg_map(VP9_COMMON *cm, int seg_map_size) { + int i; + + for (i = 0; i < NUM_PING_PONG_BUFFERS; ++i) { + cm->seg_map_array[i] = (uint8_t *)vpx_calloc(seg_map_size, 1); + if (cm->seg_map_array[i] == NULL) + return 1; + } + cm->seg_map_alloc_size = seg_map_size; + + // Init the index. + cm->seg_map_idx = 0; + cm->prev_seg_map_idx = 1; + + cm->current_frame_seg_map = cm->seg_map_array[cm->seg_map_idx]; + if (!cm->frame_parallel_decode) + cm->last_frame_seg_map = cm->seg_map_array[cm->prev_seg_map_idx]; + + return 0; +} + +static void free_seg_map(VP9_COMMON *cm) { + int i; + + for (i = 0; i < NUM_PING_PONG_BUFFERS; ++i) { + vpx_free(cm->seg_map_array[i]); + cm->seg_map_array[i] = NULL; + } + + cm->current_frame_seg_map = NULL; + + if (!cm->frame_parallel_decode) { + cm->last_frame_seg_map = NULL; + } +} + +void vp9_free_ref_frame_buffers(BufferPool *pool) { + int i; + + for (i = 0; i < FRAME_BUFFERS; ++i) { + if (pool->frame_bufs[i].ref_count > 0 && + pool->frame_bufs[i].raw_frame_buffer.data != NULL) { + pool->release_fb_cb(pool->cb_priv, &pool->frame_bufs[i].raw_frame_buffer); + pool->frame_bufs[i].ref_count = 0; + } + vpx_free(pool->frame_bufs[i].mvs); + pool->frame_bufs[i].mvs = NULL; + vpx_free_frame_buffer(&pool->frame_bufs[i].buf); + } +} + +void vp9_free_postproc_buffers(VP9_COMMON *cm) { +#if CONFIG_VP9_POSTPROC + vpx_free_frame_buffer(&cm->post_proc_buffer); + vpx_free_frame_buffer(&cm->post_proc_buffer_int); +#else + (void)cm; +#endif +} + +void vp9_free_context_buffers(VP9_COMMON *cm) { + cm->free_mi(cm); + free_seg_map(cm); + vpx_free(cm->above_context); + cm->above_context = NULL; + vpx_free(cm->above_seg_context); + cm->above_seg_context = NULL; + vpx_free(cm->lf.lfm); + cm->lf.lfm = NULL; +} + + +int vp9_alloc_loop_filter(VP9_COMMON *cm) { + vpx_free(cm->lf.lfm); + // Each lfm holds bit masks for all the 8x8 blocks in a 64x64 region. The + // stride and rows are rounded up / truncated to a multiple of 8. + cm->lf.lfm_stride = (cm->mi_cols + (MI_BLOCK_SIZE - 1)) >> 3; + cm->lf.lfm = (LOOP_FILTER_MASK *)vpx_calloc( + ((cm->mi_rows + (MI_BLOCK_SIZE - 1)) >> 3) * cm->lf.lfm_stride, + sizeof(*cm->lf.lfm)); + if (!cm->lf.lfm) + return 1; + return 0; +} + +int vp9_alloc_context_buffers(VP9_COMMON *cm, int width, int height) { + int new_mi_size; + + vp9_set_mb_mi(cm, width, height); + new_mi_size = cm->mi_stride * calc_mi_size(cm->mi_rows); + if (cm->mi_alloc_size < new_mi_size) { + cm->free_mi(cm); + if (cm->alloc_mi(cm, new_mi_size)) + goto fail; + } + + if (cm->seg_map_alloc_size < cm->mi_rows * cm->mi_cols) { + // Create the segmentation map structure and set to 0. + free_seg_map(cm); + if (alloc_seg_map(cm, cm->mi_rows * cm->mi_cols)) + goto fail; + } + + if (cm->above_context_alloc_cols < cm->mi_cols) { + vpx_free(cm->above_context); + cm->above_context = (ENTROPY_CONTEXT *)vpx_calloc( + 2 * mi_cols_aligned_to_sb(cm->mi_cols) * MAX_MB_PLANE, + sizeof(*cm->above_context)); + if (!cm->above_context) goto fail; + + vpx_free(cm->above_seg_context); + cm->above_seg_context = (PARTITION_CONTEXT *)vpx_calloc( + mi_cols_aligned_to_sb(cm->mi_cols), sizeof(*cm->above_seg_context)); + if (!cm->above_seg_context) goto fail; + cm->above_context_alloc_cols = cm->mi_cols; + } + + if (vp9_alloc_loop_filter(cm)) + goto fail; + + return 0; + + fail: + vp9_free_context_buffers(cm); + return 1; +} + +void vp9_remove_common(VP9_COMMON *cm) { + vp9_free_context_buffers(cm); + + vpx_free(cm->fc); + cm->fc = NULL; + vpx_free(cm->frame_contexts); + cm->frame_contexts = NULL; +} + +void vp9_init_context_buffers(VP9_COMMON *cm) { + cm->setup_mi(cm); + if (cm->last_frame_seg_map && !cm->frame_parallel_decode) + memset(cm->last_frame_seg_map, 0, cm->mi_rows * cm->mi_cols); +} + +void vp9_swap_current_and_last_seg_map(VP9_COMMON *cm) { + // Swap indices. + const int tmp = cm->seg_map_idx; + cm->seg_map_idx = cm->prev_seg_map_idx; + cm->prev_seg_map_idx = tmp; + + cm->current_frame_seg_map = cm->seg_map_array[cm->seg_map_idx]; + cm->last_frame_seg_map = cm->seg_map_array[cm->prev_seg_map_idx]; +} diff --git a/thirdparty/libvpx/vp9/common/vp9_alloccommon.h b/thirdparty/libvpx/vp9/common/vp9_alloccommon.h new file mode 100644 index 0000000000..e53955b998 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_alloccommon.h @@ -0,0 +1,45 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP9_COMMON_VP9_ALLOCCOMMON_H_ +#define VP9_COMMON_VP9_ALLOCCOMMON_H_ + +#define INVALID_IDX -1 // Invalid buffer index. + +#ifdef __cplusplus +extern "C" { +#endif + +struct VP9Common; +struct BufferPool; + +void vp9_remove_common(struct VP9Common *cm); + +int vp9_alloc_loop_filter(struct VP9Common *cm); +int vp9_alloc_context_buffers(struct VP9Common *cm, int width, int height); +void vp9_init_context_buffers(struct VP9Common *cm); +void vp9_free_context_buffers(struct VP9Common *cm); + +void vp9_free_ref_frame_buffers(struct BufferPool *pool); +void vp9_free_postproc_buffers(struct VP9Common *cm); + +int vp9_alloc_state_buffers(struct VP9Common *cm, int width, int height); +void vp9_free_state_buffers(struct VP9Common *cm); + +void vp9_set_mb_mi(struct VP9Common *cm, int width, int height); + +void vp9_swap_current_and_last_seg_map(struct VP9Common *cm); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_ALLOCCOMMON_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_blockd.c b/thirdparty/libvpx/vp9/common/vp9_blockd.c new file mode 100644 index 0000000000..7bab27d4fd --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_blockd.c @@ -0,0 +1,135 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "vp9/common/vp9_blockd.h" + +PREDICTION_MODE vp9_left_block_mode(const MODE_INFO *cur_mi, + const MODE_INFO *left_mi, int b) { + if (b == 0 || b == 2) { + if (!left_mi || is_inter_block(left_mi)) + return DC_PRED; + + return get_y_mode(left_mi, b + 1); + } else { + assert(b == 1 || b == 3); + return cur_mi->bmi[b - 1].as_mode; + } +} + +PREDICTION_MODE vp9_above_block_mode(const MODE_INFO *cur_mi, + const MODE_INFO *above_mi, int b) { + if (b == 0 || b == 1) { + if (!above_mi || is_inter_block(above_mi)) + return DC_PRED; + + return get_y_mode(above_mi, b + 2); + } else { + assert(b == 2 || b == 3); + return cur_mi->bmi[b - 2].as_mode; + } +} + +void vp9_foreach_transformed_block_in_plane( + const MACROBLOCKD *const xd, BLOCK_SIZE bsize, int plane, + foreach_transformed_block_visitor visit, void *arg) { + const struct macroblockd_plane *const pd = &xd->plane[plane]; + const MODE_INFO* mi = xd->mi[0]; + // block and transform sizes, in number of 4x4 blocks log 2 ("*_b") + // 4x4=0, 8x8=2, 16x16=4, 32x32=6, 64x64=8 + // transform size varies per plane, look it up in a common way. + const TX_SIZE tx_size = plane ? get_uv_tx_size(mi, pd) + : mi->tx_size; + const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd); + const int num_4x4_w = num_4x4_blocks_wide_lookup[plane_bsize]; + const int num_4x4_h = num_4x4_blocks_high_lookup[plane_bsize]; + const int step = 1 << (tx_size << 1); + int i = 0, r, c; + + // If mb_to_right_edge is < 0 we are in a situation in which + // the current block size extends into the UMV and we won't + // visit the sub blocks that are wholly within the UMV. + const int max_blocks_wide = num_4x4_w + (xd->mb_to_right_edge >= 0 ? 0 : + xd->mb_to_right_edge >> (5 + pd->subsampling_x)); + const int max_blocks_high = num_4x4_h + (xd->mb_to_bottom_edge >= 0 ? 0 : + xd->mb_to_bottom_edge >> (5 + pd->subsampling_y)); + const int extra_step = ((num_4x4_w - max_blocks_wide) >> tx_size) * step; + + // Keep track of the row and column of the blocks we use so that we know + // if we are in the unrestricted motion border. + for (r = 0; r < max_blocks_high; r += (1 << tx_size)) { + // Skip visiting the sub blocks that are wholly within the UMV. + for (c = 0; c < max_blocks_wide; c += (1 << tx_size)) { + visit(plane, i, plane_bsize, tx_size, arg); + i += step; + } + i += extra_step; + } +} + +void vp9_foreach_transformed_block(const MACROBLOCKD* const xd, + BLOCK_SIZE bsize, + foreach_transformed_block_visitor visit, + void *arg) { + int plane; + + for (plane = 0; plane < MAX_MB_PLANE; ++plane) + vp9_foreach_transformed_block_in_plane(xd, bsize, plane, visit, arg); +} + +void vp9_set_contexts(const MACROBLOCKD *xd, struct macroblockd_plane *pd, + BLOCK_SIZE plane_bsize, TX_SIZE tx_size, int has_eob, + int aoff, int loff) { + ENTROPY_CONTEXT *const a = pd->above_context + aoff; + ENTROPY_CONTEXT *const l = pd->left_context + loff; + const int tx_size_in_blocks = 1 << tx_size; + + // above + if (has_eob && xd->mb_to_right_edge < 0) { + int i; + const int blocks_wide = num_4x4_blocks_wide_lookup[plane_bsize] + + (xd->mb_to_right_edge >> (5 + pd->subsampling_x)); + int above_contexts = tx_size_in_blocks; + if (above_contexts + aoff > blocks_wide) + above_contexts = blocks_wide - aoff; + + for (i = 0; i < above_contexts; ++i) + a[i] = has_eob; + for (i = above_contexts; i < tx_size_in_blocks; ++i) + a[i] = 0; + } else { + memset(a, has_eob, sizeof(ENTROPY_CONTEXT) * tx_size_in_blocks); + } + + // left + if (has_eob && xd->mb_to_bottom_edge < 0) { + int i; + const int blocks_high = num_4x4_blocks_high_lookup[plane_bsize] + + (xd->mb_to_bottom_edge >> (5 + pd->subsampling_y)); + int left_contexts = tx_size_in_blocks; + if (left_contexts + loff > blocks_high) + left_contexts = blocks_high - loff; + + for (i = 0; i < left_contexts; ++i) + l[i] = has_eob; + for (i = left_contexts; i < tx_size_in_blocks; ++i) + l[i] = 0; + } else { + memset(l, has_eob, sizeof(ENTROPY_CONTEXT) * tx_size_in_blocks); + } +} + +void vp9_setup_block_planes(MACROBLOCKD *xd, int ss_x, int ss_y) { + int i; + + for (i = 0; i < MAX_MB_PLANE; i++) { + xd->plane[i].subsampling_x = i ? ss_x : 0; + xd->plane[i].subsampling_y = i ? ss_y : 0; + } +} diff --git a/thirdparty/libvpx/vp9/common/vp9_blockd.h b/thirdparty/libvpx/vp9/common/vp9_blockd.h new file mode 100644 index 0000000000..3d26fb2b5d --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_blockd.h @@ -0,0 +1,305 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP9_COMMON_VP9_BLOCKD_H_ +#define VP9_COMMON_VP9_BLOCKD_H_ + +#include "./vpx_config.h" + +#include "vpx_dsp/vpx_dsp_common.h" +#include "vpx_ports/mem.h" +#include "vpx_scale/yv12config.h" + +#include "vp9/common/vp9_common_data.h" +#include "vp9/common/vp9_entropy.h" +#include "vp9/common/vp9_entropymode.h" +#include "vp9/common/vp9_mv.h" +#include "vp9/common/vp9_scale.h" +#include "vp9/common/vp9_seg_common.h" +#include "vp9/common/vp9_tile_common.h" + +#ifdef __cplusplus +extern "C" { +#endif + +#define MAX_MB_PLANE 3 + +typedef enum { + KEY_FRAME = 0, + INTER_FRAME = 1, + FRAME_TYPES, +} FRAME_TYPE; + +static INLINE int is_inter_mode(PREDICTION_MODE mode) { + return mode >= NEARESTMV && mode <= NEWMV; +} + +/* For keyframes, intra block modes are predicted by the (already decoded) + modes for the Y blocks to the left and above us; for interframes, there + is a single probability table. */ + +typedef struct { + PREDICTION_MODE as_mode; + int_mv as_mv[2]; // first, second inter predictor motion vectors +} b_mode_info; + +// Note that the rate-distortion optimization loop, bit-stream writer, and +// decoder implementation modules critically rely on the defined entry values +// specified herein. They should be refactored concurrently. + +#define NONE -1 +#define INTRA_FRAME 0 +#define LAST_FRAME 1 +#define GOLDEN_FRAME 2 +#define ALTREF_FRAME 3 +#define MAX_REF_FRAMES 4 +typedef int8_t MV_REFERENCE_FRAME; + +// This structure now relates to 8x8 block regions. +typedef struct MODE_INFO { + // Common for both INTER and INTRA blocks + BLOCK_SIZE sb_type; + PREDICTION_MODE mode; + TX_SIZE tx_size; + int8_t skip; + int8_t segment_id; + int8_t seg_id_predicted; // valid only when temporal_update is enabled + + // Only for INTRA blocks + PREDICTION_MODE uv_mode; + + // Only for INTER blocks + INTERP_FILTER interp_filter; + MV_REFERENCE_FRAME ref_frame[2]; + + // TODO(slavarnway): Delete and use bmi[3].as_mv[] instead. + int_mv mv[2]; + + b_mode_info bmi[4]; +} MODE_INFO; + +static INLINE PREDICTION_MODE get_y_mode(const MODE_INFO *mi, int block) { + return mi->sb_type < BLOCK_8X8 ? mi->bmi[block].as_mode + : mi->mode; +} + +static INLINE int is_inter_block(const MODE_INFO *mi) { + return mi->ref_frame[0] > INTRA_FRAME; +} + +static INLINE int has_second_ref(const MODE_INFO *mi) { + return mi->ref_frame[1] > INTRA_FRAME; +} + +PREDICTION_MODE vp9_left_block_mode(const MODE_INFO *cur_mi, + const MODE_INFO *left_mi, int b); + +PREDICTION_MODE vp9_above_block_mode(const MODE_INFO *cur_mi, + const MODE_INFO *above_mi, int b); + +enum mv_precision { + MV_PRECISION_Q3, + MV_PRECISION_Q4 +}; + +struct buf_2d { + uint8_t *buf; + int stride; +}; + +struct macroblockd_plane { + tran_low_t *dqcoeff; + int subsampling_x; + int subsampling_y; + struct buf_2d dst; + struct buf_2d pre[2]; + ENTROPY_CONTEXT *above_context; + ENTROPY_CONTEXT *left_context; + int16_t seg_dequant[MAX_SEGMENTS][2]; + + // number of 4x4s in current block + uint16_t n4_w, n4_h; + // log2 of n4_w, n4_h + uint8_t n4_wl, n4_hl; + + // encoder + const int16_t *dequant; +}; + +#define BLOCK_OFFSET(x, i) ((x) + (i) * 16) + +typedef struct RefBuffer { + // TODO(dkovalev): idx is not really required and should be removed, now it + // is used in vp9_onyxd_if.c + int idx; + YV12_BUFFER_CONFIG *buf; + struct scale_factors sf; +} RefBuffer; + +typedef struct macroblockd { + struct macroblockd_plane plane[MAX_MB_PLANE]; + uint8_t bmode_blocks_wl; + uint8_t bmode_blocks_hl; + + FRAME_COUNTS *counts; + TileInfo tile; + + int mi_stride; + + MODE_INFO **mi; + MODE_INFO *left_mi; + MODE_INFO *above_mi; + + unsigned int max_blocks_wide; + unsigned int max_blocks_high; + + const vpx_prob (*partition_probs)[PARTITION_TYPES - 1]; + + /* Distance of MB away from frame edges */ + int mb_to_left_edge; + int mb_to_right_edge; + int mb_to_top_edge; + int mb_to_bottom_edge; + + FRAME_CONTEXT *fc; + + /* pointers to reference frames */ + RefBuffer *block_refs[2]; + + /* pointer to current frame */ + const YV12_BUFFER_CONFIG *cur_buf; + + ENTROPY_CONTEXT *above_context[MAX_MB_PLANE]; + ENTROPY_CONTEXT left_context[MAX_MB_PLANE][16]; + + PARTITION_CONTEXT *above_seg_context; + PARTITION_CONTEXT left_seg_context[8]; + +#if CONFIG_VP9_HIGHBITDEPTH + /* Bit depth: 8, 10, 12 */ + int bd; +#endif + + int lossless; + int corrupted; + + struct vpx_internal_error_info *error_info; +} MACROBLOCKD; + +static INLINE PLANE_TYPE get_plane_type(int plane) { + return (PLANE_TYPE)(plane > 0); +} + +static INLINE BLOCK_SIZE get_subsize(BLOCK_SIZE bsize, + PARTITION_TYPE partition) { + return subsize_lookup[partition][bsize]; +} + +extern const TX_TYPE intra_mode_to_tx_type_lookup[INTRA_MODES]; + +static INLINE TX_TYPE get_tx_type(PLANE_TYPE plane_type, + const MACROBLOCKD *xd) { + const MODE_INFO *const mi = xd->mi[0]; + + if (plane_type != PLANE_TYPE_Y || xd->lossless || is_inter_block(mi)) + return DCT_DCT; + + return intra_mode_to_tx_type_lookup[mi->mode]; +} + +static INLINE TX_TYPE get_tx_type_4x4(PLANE_TYPE plane_type, + const MACROBLOCKD *xd, int ib) { + const MODE_INFO *const mi = xd->mi[0]; + + if (plane_type != PLANE_TYPE_Y || xd->lossless || is_inter_block(mi)) + return DCT_DCT; + + return intra_mode_to_tx_type_lookup[get_y_mode(mi, ib)]; +} + +void vp9_setup_block_planes(MACROBLOCKD *xd, int ss_x, int ss_y); + +static INLINE TX_SIZE get_uv_tx_size_impl(TX_SIZE y_tx_size, BLOCK_SIZE bsize, + int xss, int yss) { + if (bsize < BLOCK_8X8) { + return TX_4X4; + } else { + const BLOCK_SIZE plane_bsize = ss_size_lookup[bsize][xss][yss]; + return VPXMIN(y_tx_size, max_txsize_lookup[plane_bsize]); + } +} + +static INLINE TX_SIZE get_uv_tx_size(const MODE_INFO *mi, + const struct macroblockd_plane *pd) { + return get_uv_tx_size_impl(mi->tx_size, mi->sb_type, pd->subsampling_x, + pd->subsampling_y); +} + +static INLINE BLOCK_SIZE get_plane_block_size(BLOCK_SIZE bsize, + const struct macroblockd_plane *pd) { + return ss_size_lookup[bsize][pd->subsampling_x][pd->subsampling_y]; +} + +static INLINE void reset_skip_context(MACROBLOCKD *xd, BLOCK_SIZE bsize) { + int i; + for (i = 0; i < MAX_MB_PLANE; i++) { + struct macroblockd_plane *const pd = &xd->plane[i]; + const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd); + memset(pd->above_context, 0, + sizeof(ENTROPY_CONTEXT) * num_4x4_blocks_wide_lookup[plane_bsize]); + memset(pd->left_context, 0, + sizeof(ENTROPY_CONTEXT) * num_4x4_blocks_high_lookup[plane_bsize]); + } +} + +static INLINE const vpx_prob *get_y_mode_probs(const MODE_INFO *mi, + const MODE_INFO *above_mi, + const MODE_INFO *left_mi, + int block) { + const PREDICTION_MODE above = vp9_above_block_mode(mi, above_mi, block); + const PREDICTION_MODE left = vp9_left_block_mode(mi, left_mi, block); + return vp9_kf_y_mode_prob[above][left]; +} + +typedef void (*foreach_transformed_block_visitor)(int plane, int block, + BLOCK_SIZE plane_bsize, + TX_SIZE tx_size, + void *arg); + +void vp9_foreach_transformed_block_in_plane( + const MACROBLOCKD *const xd, BLOCK_SIZE bsize, int plane, + foreach_transformed_block_visitor visit, void *arg); + + +void vp9_foreach_transformed_block( + const MACROBLOCKD* const xd, BLOCK_SIZE bsize, + foreach_transformed_block_visitor visit, void *arg); + +static INLINE void txfrm_block_to_raster_xy(BLOCK_SIZE plane_bsize, + TX_SIZE tx_size, int block, + int *x, int *y) { + const int bwl = b_width_log2_lookup[plane_bsize]; + const int tx_cols_log2 = bwl - tx_size; + const int tx_cols = 1 << tx_cols_log2; + const int raster_mb = block >> (tx_size << 1); + *x = (raster_mb & (tx_cols - 1)) << tx_size; + *y = (raster_mb >> tx_cols_log2) << tx_size; +} + +void vp9_set_contexts(const MACROBLOCKD *xd, struct macroblockd_plane *pd, + BLOCK_SIZE plane_bsize, TX_SIZE tx_size, int has_eob, + int aoff, int loff); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_BLOCKD_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_common.h b/thirdparty/libvpx/vp9/common/vp9_common.h new file mode 100644 index 0000000000..908fa80a31 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_common.h @@ -0,0 +1,74 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_COMMON_VP9_COMMON_H_ +#define VP9_COMMON_VP9_COMMON_H_ + +/* Interface header for common constant data structures and lookup tables */ + +#include <assert.h> + +#include "./vpx_config.h" +#include "vpx_dsp/vpx_dsp_common.h" +#include "vpx_mem/vpx_mem.h" +#include "vpx/vpx_integer.h" +#include "vpx_ports/bitops.h" + +#ifdef __cplusplus +extern "C" { +#endif + +// Only need this for fixed-size arrays, for structs just assign. +#define vp9_copy(dest, src) { \ + assert(sizeof(dest) == sizeof(src)); \ + memcpy(dest, src, sizeof(src)); \ + } + +// Use this for variably-sized arrays. +#define vp9_copy_array(dest, src, n) { \ + assert(sizeof(*dest) == sizeof(*src)); \ + memcpy(dest, src, n * sizeof(*src)); \ + } + +#define vp9_zero(dest) memset(&(dest), 0, sizeof(dest)) +#define vp9_zero_array(dest, n) memset(dest, 0, n * sizeof(*dest)) + +static INLINE int get_unsigned_bits(unsigned int num_values) { + return num_values > 0 ? get_msb(num_values) + 1 : 0; +} + +#if CONFIG_DEBUG +#define CHECK_MEM_ERROR(cm, lval, expr) do { \ + lval = (expr); \ + if (!lval) \ + vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR, \ + "Failed to allocate "#lval" at %s:%d", \ + __FILE__, __LINE__); \ + } while (0) +#else +#define CHECK_MEM_ERROR(cm, lval, expr) do { \ + lval = (expr); \ + if (!lval) \ + vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR, \ + "Failed to allocate "#lval); \ + } while (0) +#endif + +#define VP9_SYNC_CODE_0 0x49 +#define VP9_SYNC_CODE_1 0x83 +#define VP9_SYNC_CODE_2 0x42 + +#define VP9_FRAME_MARKER 0x2 + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_COMMON_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_common_data.c b/thirdparty/libvpx/vp9/common/vp9_common_data.c new file mode 100644 index 0000000000..3409d04844 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_common_data.c @@ -0,0 +1,176 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "vp9/common/vp9_common_data.h" +#include "vpx_dsp/vpx_dsp_common.h" + +// Log 2 conversion lookup tables for block width and height +const uint8_t b_width_log2_lookup[BLOCK_SIZES] = + {0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3, 4, 4}; +const uint8_t b_height_log2_lookup[BLOCK_SIZES] = + {0, 1, 0, 1, 2, 1, 2, 3, 2, 3, 4, 3, 4}; +const uint8_t num_4x4_blocks_wide_lookup[BLOCK_SIZES] = + {1, 1, 2, 2, 2, 4, 4, 4, 8, 8, 8, 16, 16}; +const uint8_t num_4x4_blocks_high_lookup[BLOCK_SIZES] = + {1, 2, 1, 2, 4, 2, 4, 8, 4, 8, 16, 8, 16}; +// Log 2 conversion lookup tables for modeinfo width and height +const uint8_t mi_width_log2_lookup[BLOCK_SIZES] = + {0, 0, 0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3}; +const uint8_t num_8x8_blocks_wide_lookup[BLOCK_SIZES] = + {1, 1, 1, 1, 1, 2, 2, 2, 4, 4, 4, 8, 8}; +const uint8_t num_8x8_blocks_high_lookup[BLOCK_SIZES] = + {1, 1, 1, 1, 2, 1, 2, 4, 2, 4, 8, 4, 8}; + +// VPXMIN(3, VPXMIN(b_width_log2(bsize), b_height_log2(bsize))) +const uint8_t size_group_lookup[BLOCK_SIZES] = + {0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3, 3}; + +const uint8_t num_pels_log2_lookup[BLOCK_SIZES] = + {4, 5, 5, 6, 7, 7, 8, 9, 9, 10, 11, 11, 12}; + +const PARTITION_TYPE partition_lookup[][BLOCK_SIZES] = { + { // 4X4 + // 4X4, 4X8,8X4,8X8,8X16,16X8,16X16,16X32,32X16,32X32,32X64,64X32,64X64 + PARTITION_NONE, PARTITION_INVALID, PARTITION_INVALID, + PARTITION_INVALID, PARTITION_INVALID, PARTITION_INVALID, + PARTITION_INVALID, PARTITION_INVALID, PARTITION_INVALID, + PARTITION_INVALID, PARTITION_INVALID, PARTITION_INVALID, + PARTITION_INVALID + }, { // 8X8 + // 4X4, 4X8,8X4,8X8,8X16,16X8,16X16,16X32,32X16,32X32,32X64,64X32,64X64 + PARTITION_SPLIT, PARTITION_VERT, PARTITION_HORZ, PARTITION_NONE, + PARTITION_INVALID, PARTITION_INVALID, PARTITION_INVALID, + PARTITION_INVALID, PARTITION_INVALID, PARTITION_INVALID, + PARTITION_INVALID, PARTITION_INVALID, PARTITION_INVALID + }, { // 16X16 + // 4X4, 4X8,8X4,8X8,8X16,16X8,16X16,16X32,32X16,32X32,32X64,64X32,64X64 + PARTITION_SPLIT, PARTITION_SPLIT, PARTITION_SPLIT, PARTITION_SPLIT, + PARTITION_VERT, PARTITION_HORZ, PARTITION_NONE, PARTITION_INVALID, + PARTITION_INVALID, PARTITION_INVALID, PARTITION_INVALID, + PARTITION_INVALID, PARTITION_INVALID + }, { // 32X32 + // 4X4, 4X8,8X4,8X8,8X16,16X8,16X16,16X32,32X16,32X32,32X64,64X32,64X64 + PARTITION_SPLIT, PARTITION_SPLIT, PARTITION_SPLIT, PARTITION_SPLIT, + PARTITION_SPLIT, PARTITION_SPLIT, PARTITION_SPLIT, PARTITION_VERT, + PARTITION_HORZ, PARTITION_NONE, PARTITION_INVALID, + PARTITION_INVALID, PARTITION_INVALID + }, { // 64X64 + // 4X4, 4X8,8X4,8X8,8X16,16X8,16X16,16X32,32X16,32X32,32X64,64X32,64X64 + PARTITION_SPLIT, PARTITION_SPLIT, PARTITION_SPLIT, PARTITION_SPLIT, + PARTITION_SPLIT, PARTITION_SPLIT, PARTITION_SPLIT, PARTITION_SPLIT, + PARTITION_SPLIT, PARTITION_SPLIT, PARTITION_VERT, PARTITION_HORZ, + PARTITION_NONE + } +}; + +const BLOCK_SIZE subsize_lookup[PARTITION_TYPES][BLOCK_SIZES] = { + { // PARTITION_NONE + BLOCK_4X4, BLOCK_4X8, BLOCK_8X4, + BLOCK_8X8, BLOCK_8X16, BLOCK_16X8, + BLOCK_16X16, BLOCK_16X32, BLOCK_32X16, + BLOCK_32X32, BLOCK_32X64, BLOCK_64X32, + BLOCK_64X64, + }, { // PARTITION_HORZ + BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID, + BLOCK_8X4, BLOCK_INVALID, BLOCK_INVALID, + BLOCK_16X8, BLOCK_INVALID, BLOCK_INVALID, + BLOCK_32X16, BLOCK_INVALID, BLOCK_INVALID, + BLOCK_64X32, + }, { // PARTITION_VERT + BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID, + BLOCK_4X8, BLOCK_INVALID, BLOCK_INVALID, + BLOCK_8X16, BLOCK_INVALID, BLOCK_INVALID, + BLOCK_16X32, BLOCK_INVALID, BLOCK_INVALID, + BLOCK_32X64, + }, { // PARTITION_SPLIT + BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID, + BLOCK_4X4, BLOCK_INVALID, BLOCK_INVALID, + BLOCK_8X8, BLOCK_INVALID, BLOCK_INVALID, + BLOCK_16X16, BLOCK_INVALID, BLOCK_INVALID, + BLOCK_32X32, + } +}; + +const TX_SIZE max_txsize_lookup[BLOCK_SIZES] = { + TX_4X4, TX_4X4, TX_4X4, + TX_8X8, TX_8X8, TX_8X8, + TX_16X16, TX_16X16, TX_16X16, + TX_32X32, TX_32X32, TX_32X32, TX_32X32 +}; + +const BLOCK_SIZE txsize_to_bsize[TX_SIZES] = { + BLOCK_4X4, // TX_4X4 + BLOCK_8X8, // TX_8X8 + BLOCK_16X16, // TX_16X16 + BLOCK_32X32, // TX_32X32 +}; + +const TX_SIZE tx_mode_to_biggest_tx_size[TX_MODES] = { + TX_4X4, // ONLY_4X4 + TX_8X8, // ALLOW_8X8 + TX_16X16, // ALLOW_16X16 + TX_32X32, // ALLOW_32X32 + TX_32X32, // TX_MODE_SELECT +}; + +const BLOCK_SIZE ss_size_lookup[BLOCK_SIZES][2][2] = { +// ss_x == 0 ss_x == 0 ss_x == 1 ss_x == 1 +// ss_y == 0 ss_y == 1 ss_y == 0 ss_y == 1 + {{BLOCK_4X4, BLOCK_INVALID}, {BLOCK_INVALID, BLOCK_INVALID}}, + {{BLOCK_4X8, BLOCK_4X4}, {BLOCK_INVALID, BLOCK_INVALID}}, + {{BLOCK_8X4, BLOCK_INVALID}, {BLOCK_4X4, BLOCK_INVALID}}, + {{BLOCK_8X8, BLOCK_8X4}, {BLOCK_4X8, BLOCK_4X4}}, + {{BLOCK_8X16, BLOCK_8X8}, {BLOCK_INVALID, BLOCK_4X8}}, + {{BLOCK_16X8, BLOCK_INVALID}, {BLOCK_8X8, BLOCK_8X4}}, + {{BLOCK_16X16, BLOCK_16X8}, {BLOCK_8X16, BLOCK_8X8}}, + {{BLOCK_16X32, BLOCK_16X16}, {BLOCK_INVALID, BLOCK_8X16}}, + {{BLOCK_32X16, BLOCK_INVALID}, {BLOCK_16X16, BLOCK_16X8}}, + {{BLOCK_32X32, BLOCK_32X16}, {BLOCK_16X32, BLOCK_16X16}}, + {{BLOCK_32X64, BLOCK_32X32}, {BLOCK_INVALID, BLOCK_16X32}}, + {{BLOCK_64X32, BLOCK_INVALID}, {BLOCK_32X32, BLOCK_32X16}}, + {{BLOCK_64X64, BLOCK_64X32}, {BLOCK_32X64, BLOCK_32X32}}, +}; + +// Generates 4 bit field in which each bit set to 1 represents +// a blocksize partition 1111 means we split 64x64, 32x32, 16x16 +// and 8x8. 1000 means we just split the 64x64 to 32x32 +const struct { + PARTITION_CONTEXT above; + PARTITION_CONTEXT left; +} partition_context_lookup[BLOCK_SIZES]= { + {15, 15}, // 4X4 - {0b1111, 0b1111} + {15, 14}, // 4X8 - {0b1111, 0b1110} + {14, 15}, // 8X4 - {0b1110, 0b1111} + {14, 14}, // 8X8 - {0b1110, 0b1110} + {14, 12}, // 8X16 - {0b1110, 0b1100} + {12, 14}, // 16X8 - {0b1100, 0b1110} + {12, 12}, // 16X16 - {0b1100, 0b1100} + {12, 8 }, // 16X32 - {0b1100, 0b1000} + {8, 12}, // 32X16 - {0b1000, 0b1100} + {8, 8 }, // 32X32 - {0b1000, 0b1000} + {8, 0 }, // 32X64 - {0b1000, 0b0000} + {0, 8 }, // 64X32 - {0b0000, 0b1000} + {0, 0 }, // 64X64 - {0b0000, 0b0000} +}; + +#if CONFIG_BETTER_HW_COMPATIBILITY && CONFIG_VP9_HIGHBITDEPTH +const uint8_t need_top_left[INTRA_MODES] = { + 0, // DC_PRED + 0, // V_PRED + 0, // H_PRED + 0, // D45_PRED + 1, // D135_PRED + 1, // D117_PRED + 1, // D153_PRED + 0, // D207_PRED + 0, // D63_PRED + 1, // TM_PRED +}; +#endif // CONFIG_BETTER_HW_COMPATIBILITY && CONFIG_VP9_HIGHBITDEPTH diff --git a/thirdparty/libvpx/vp9/common/vp9_common_data.h b/thirdparty/libvpx/vp9/common/vp9_common_data.h new file mode 100644 index 0000000000..0ae24dad54 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_common_data.h @@ -0,0 +1,44 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_COMMON_VP9_COMMON_DATA_H_ +#define VP9_COMMON_VP9_COMMON_DATA_H_ + +#include "vp9/common/vp9_enums.h" +#include "vpx/vpx_integer.h" + +#ifdef __cplusplus +extern "C" { +#endif + +extern const uint8_t b_width_log2_lookup[BLOCK_SIZES]; +extern const uint8_t b_height_log2_lookup[BLOCK_SIZES]; +extern const uint8_t mi_width_log2_lookup[BLOCK_SIZES]; +extern const uint8_t num_8x8_blocks_wide_lookup[BLOCK_SIZES]; +extern const uint8_t num_8x8_blocks_high_lookup[BLOCK_SIZES]; +extern const uint8_t num_4x4_blocks_high_lookup[BLOCK_SIZES]; +extern const uint8_t num_4x4_blocks_wide_lookup[BLOCK_SIZES]; +extern const uint8_t size_group_lookup[BLOCK_SIZES]; +extern const uint8_t num_pels_log2_lookup[BLOCK_SIZES]; +extern const PARTITION_TYPE partition_lookup[][BLOCK_SIZES]; +extern const BLOCK_SIZE subsize_lookup[PARTITION_TYPES][BLOCK_SIZES]; +extern const TX_SIZE max_txsize_lookup[BLOCK_SIZES]; +extern const BLOCK_SIZE txsize_to_bsize[TX_SIZES]; +extern const TX_SIZE tx_mode_to_biggest_tx_size[TX_MODES]; +extern const BLOCK_SIZE ss_size_lookup[BLOCK_SIZES][2][2]; +#if CONFIG_BETTER_HW_COMPATIBILITY && CONFIG_VP9_HIGHBITDEPTH +extern const uint8_t need_top_left[INTRA_MODES]; +#endif // CONFIG_BETTER_HW_COMPATIBILITY && CONFIG_VP9_HIGHBITDEPTH + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_COMMON_DATA_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_debugmodes.c b/thirdparty/libvpx/vp9/common/vp9_debugmodes.c new file mode 100644 index 0000000000..d9c1fd9686 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_debugmodes.c @@ -0,0 +1,91 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <stdio.h> + +#include "vp9/common/vp9_blockd.h" +#include "vp9/common/vp9_onyxc_int.h" + +static void log_frame_info(VP9_COMMON *cm, const char *str, FILE *f) { + fprintf(f, "%s", str); + fprintf(f, "(Frame %d, Show:%d, Q:%d): \n", cm->current_video_frame, + cm->show_frame, cm->base_qindex); +} +/* This function dereferences a pointer to the mbmi structure + * and uses the passed in member offset to print out the value of an integer + * for each mbmi member value in the mi structure. + */ +static void print_mi_data(VP9_COMMON *cm, FILE *file, const char *descriptor, + size_t member_offset) { + int mi_row, mi_col; + MODE_INFO **mi = cm->mi_grid_visible; + int rows = cm->mi_rows; + int cols = cm->mi_cols; + char prefix = descriptor[0]; + + log_frame_info(cm, descriptor, file); + for (mi_row = 0; mi_row < rows; mi_row++) { + fprintf(file, "%c ", prefix); + for (mi_col = 0; mi_col < cols; mi_col++) { + fprintf(file, "%2d ", + *((int*) ((char *) (mi[0]) + + member_offset))); + mi++; + } + fprintf(file, "\n"); + mi += 8; + } + fprintf(file, "\n"); +} + +void vp9_print_modes_and_motion_vectors(VP9_COMMON *cm, const char *file) { + int mi_row; + int mi_col; + FILE *mvs = fopen(file, "a"); + MODE_INFO **mi = cm->mi_grid_visible; + int rows = cm->mi_rows; + int cols = cm->mi_cols; + + print_mi_data(cm, mvs, "Partitions:", offsetof(MODE_INFO, sb_type)); + print_mi_data(cm, mvs, "Modes:", offsetof(MODE_INFO, mode)); + print_mi_data(cm, mvs, "Ref frame:", offsetof(MODE_INFO, ref_frame[0])); + print_mi_data(cm, mvs, "Transform:", offsetof(MODE_INFO, tx_size)); + print_mi_data(cm, mvs, "UV Modes:", offsetof(MODE_INFO, uv_mode)); + + // output skip infomation. + log_frame_info(cm, "Skips:", mvs); + for (mi_row = 0; mi_row < rows; mi_row++) { + fprintf(mvs, "S "); + for (mi_col = 0; mi_col < cols; mi_col++) { + fprintf(mvs, "%2d ", mi[0]->skip); + mi++; + } + fprintf(mvs, "\n"); + mi += 8; + } + fprintf(mvs, "\n"); + + // output motion vectors. + log_frame_info(cm, "Vectors ", mvs); + mi = cm->mi_grid_visible; + for (mi_row = 0; mi_row < rows; mi_row++) { + fprintf(mvs, "V "); + for (mi_col = 0; mi_col < cols; mi_col++) { + fprintf(mvs, "%4d:%4d ", mi[0]->mv[0].as_mv.row, + mi[0]->mv[0].as_mv.col); + mi++; + } + fprintf(mvs, "\n"); + mi += 8; + } + fprintf(mvs, "\n"); + + fclose(mvs); +} diff --git a/thirdparty/libvpx/vp9/common/vp9_entropy.c b/thirdparty/libvpx/vp9/common/vp9_entropy.c new file mode 100644 index 0000000000..7b490af34f --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_entropy.c @@ -0,0 +1,802 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "vp9/common/vp9_entropy.h" +#include "vp9/common/vp9_blockd.h" +#include "vp9/common/vp9_onyxc_int.h" +#include "vp9/common/vp9_entropymode.h" +#include "vpx_mem/vpx_mem.h" +#include "vpx/vpx_integer.h" + +// Unconstrained Node Tree +const vpx_tree_index vp9_coef_con_tree[TREE_SIZE(ENTROPY_TOKENS)] = { + 2, 6, // 0 = LOW_VAL + -TWO_TOKEN, 4, // 1 = TWO + -THREE_TOKEN, -FOUR_TOKEN, // 2 = THREE + 8, 10, // 3 = HIGH_LOW + -CATEGORY1_TOKEN, -CATEGORY2_TOKEN, // 4 = CAT_ONE + 12, 14, // 5 = CAT_THREEFOUR + -CATEGORY3_TOKEN, -CATEGORY4_TOKEN, // 6 = CAT_THREE + -CATEGORY5_TOKEN, -CATEGORY6_TOKEN // 7 = CAT_FIVE +}; + +const vpx_prob vp9_cat1_prob[] = { 159 }; +const vpx_prob vp9_cat2_prob[] = { 165, 145 }; +const vpx_prob vp9_cat3_prob[] = { 173, 148, 140 }; +const vpx_prob vp9_cat4_prob[] = { 176, 155, 140, 135 }; +const vpx_prob vp9_cat5_prob[] = { 180, 157, 141, 134, 130 }; +const vpx_prob vp9_cat6_prob[] = { + 254, 254, 254, 252, 249, 243, 230, 196, 177, 153, 140, 133, 130, 129 +}; +#if CONFIG_VP9_HIGHBITDEPTH +const vpx_prob vp9_cat6_prob_high12[] = { + 255, 255, 255, 255, 254, 254, 254, 252, 249, + 243, 230, 196, 177, 153, 140, 133, 130, 129 +}; +#endif + +const uint8_t vp9_coefband_trans_8x8plus[1024] = { + 0, 1, 1, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, + 4, 4, 4, 4, 4, 5, + // beyond MAXBAND_INDEX+1 all values are filled as 5 + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, +}; + +const uint8_t vp9_coefband_trans_4x4[16] = { + 0, 1, 1, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 5, 5, 5, +}; + +const uint8_t vp9_pt_energy_class[ENTROPY_TOKENS] = { + 0, 1, 2, 3, 3, 4, 4, 5, 5, 5, 5, 5 +}; + +// Model obtained from a 2-sided zero-centerd distribuition derived +// from a Pareto distribution. The cdf of the distribution is: +// cdf(x) = 0.5 + 0.5 * sgn(x) * [1 - {alpha/(alpha + |x|)} ^ beta] +// +// For a given beta and a given probablity of the 1-node, the alpha +// is first solved, and then the {alpha, beta} pair is used to generate +// the probabilities for the rest of the nodes. + +// beta = 8 + +// Every odd line in this table can be generated from the even lines +// by averaging : +// vp9_pareto8_full[l][node] = (vp9_pareto8_full[l-1][node] + +// vp9_pareto8_full[l+1][node] ) >> 1; +const vpx_prob vp9_pareto8_full[COEFF_PROB_MODELS][MODEL_NODES] = { + { 3, 86, 128, 6, 86, 23, 88, 29}, + { 6, 86, 128, 11, 87, 42, 91, 52}, + { 9, 86, 129, 17, 88, 61, 94, 76}, + { 12, 86, 129, 22, 88, 77, 97, 93}, + { 15, 87, 129, 28, 89, 93, 100, 110}, + { 17, 87, 129, 33, 90, 105, 103, 123}, + { 20, 88, 130, 38, 91, 118, 106, 136}, + { 23, 88, 130, 43, 91, 128, 108, 146}, + { 26, 89, 131, 48, 92, 139, 111, 156}, + { 28, 89, 131, 53, 93, 147, 114, 163}, + { 31, 90, 131, 58, 94, 156, 117, 171}, + { 34, 90, 131, 62, 94, 163, 119, 177}, + { 37, 90, 132, 66, 95, 171, 122, 184}, + { 39, 90, 132, 70, 96, 177, 124, 189}, + { 42, 91, 132, 75, 97, 183, 127, 194}, + { 44, 91, 132, 79, 97, 188, 129, 198}, + { 47, 92, 133, 83, 98, 193, 132, 202}, + { 49, 92, 133, 86, 99, 197, 134, 205}, + { 52, 93, 133, 90, 100, 201, 137, 208}, + { 54, 93, 133, 94, 100, 204, 139, 211}, + { 57, 94, 134, 98, 101, 208, 142, 214}, + { 59, 94, 134, 101, 102, 211, 144, 216}, + { 62, 94, 135, 105, 103, 214, 146, 218}, + { 64, 94, 135, 108, 103, 216, 148, 220}, + { 66, 95, 135, 111, 104, 219, 151, 222}, + { 68, 95, 135, 114, 105, 221, 153, 223}, + { 71, 96, 136, 117, 106, 224, 155, 225}, + { 73, 96, 136, 120, 106, 225, 157, 226}, + { 76, 97, 136, 123, 107, 227, 159, 228}, + { 78, 97, 136, 126, 108, 229, 160, 229}, + { 80, 98, 137, 129, 109, 231, 162, 231}, + { 82, 98, 137, 131, 109, 232, 164, 232}, + { 84, 98, 138, 134, 110, 234, 166, 233}, + { 86, 98, 138, 137, 111, 235, 168, 234}, + { 89, 99, 138, 140, 112, 236, 170, 235}, + { 91, 99, 138, 142, 112, 237, 171, 235}, + { 93, 100, 139, 145, 113, 238, 173, 236}, + { 95, 100, 139, 147, 114, 239, 174, 237}, + { 97, 101, 140, 149, 115, 240, 176, 238}, + { 99, 101, 140, 151, 115, 241, 177, 238}, + {101, 102, 140, 154, 116, 242, 179, 239}, + {103, 102, 140, 156, 117, 242, 180, 239}, + {105, 103, 141, 158, 118, 243, 182, 240}, + {107, 103, 141, 160, 118, 243, 183, 240}, + {109, 104, 141, 162, 119, 244, 185, 241}, + {111, 104, 141, 164, 119, 244, 186, 241}, + {113, 104, 142, 166, 120, 245, 187, 242}, + {114, 104, 142, 168, 121, 245, 188, 242}, + {116, 105, 143, 170, 122, 246, 190, 243}, + {118, 105, 143, 171, 122, 246, 191, 243}, + {120, 106, 143, 173, 123, 247, 192, 244}, + {121, 106, 143, 175, 124, 247, 193, 244}, + {123, 107, 144, 177, 125, 248, 195, 244}, + {125, 107, 144, 178, 125, 248, 196, 244}, + {127, 108, 145, 180, 126, 249, 197, 245}, + {128, 108, 145, 181, 127, 249, 198, 245}, + {130, 109, 145, 183, 128, 249, 199, 245}, + {132, 109, 145, 184, 128, 249, 200, 245}, + {134, 110, 146, 186, 129, 250, 201, 246}, + {135, 110, 146, 187, 130, 250, 202, 246}, + {137, 111, 147, 189, 131, 251, 203, 246}, + {138, 111, 147, 190, 131, 251, 204, 246}, + {140, 112, 147, 192, 132, 251, 205, 247}, + {141, 112, 147, 193, 132, 251, 206, 247}, + {143, 113, 148, 194, 133, 251, 207, 247}, + {144, 113, 148, 195, 134, 251, 207, 247}, + {146, 114, 149, 197, 135, 252, 208, 248}, + {147, 114, 149, 198, 135, 252, 209, 248}, + {149, 115, 149, 199, 136, 252, 210, 248}, + {150, 115, 149, 200, 137, 252, 210, 248}, + {152, 115, 150, 201, 138, 252, 211, 248}, + {153, 115, 150, 202, 138, 252, 212, 248}, + {155, 116, 151, 204, 139, 253, 213, 249}, + {156, 116, 151, 205, 139, 253, 213, 249}, + {158, 117, 151, 206, 140, 253, 214, 249}, + {159, 117, 151, 207, 141, 253, 215, 249}, + {161, 118, 152, 208, 142, 253, 216, 249}, + {162, 118, 152, 209, 142, 253, 216, 249}, + {163, 119, 153, 210, 143, 253, 217, 249}, + {164, 119, 153, 211, 143, 253, 217, 249}, + {166, 120, 153, 212, 144, 254, 218, 250}, + {167, 120, 153, 212, 145, 254, 219, 250}, + {168, 121, 154, 213, 146, 254, 220, 250}, + {169, 121, 154, 214, 146, 254, 220, 250}, + {171, 122, 155, 215, 147, 254, 221, 250}, + {172, 122, 155, 216, 147, 254, 221, 250}, + {173, 123, 155, 217, 148, 254, 222, 250}, + {174, 123, 155, 217, 149, 254, 222, 250}, + {176, 124, 156, 218, 150, 254, 223, 250}, + {177, 124, 156, 219, 150, 254, 223, 250}, + {178, 125, 157, 220, 151, 254, 224, 251}, + {179, 125, 157, 220, 151, 254, 224, 251}, + {180, 126, 157, 221, 152, 254, 225, 251}, + {181, 126, 157, 221, 152, 254, 225, 251}, + {183, 127, 158, 222, 153, 254, 226, 251}, + {184, 127, 158, 223, 154, 254, 226, 251}, + {185, 128, 159, 224, 155, 255, 227, 251}, + {186, 128, 159, 224, 155, 255, 227, 251}, + {187, 129, 160, 225, 156, 255, 228, 251}, + {188, 130, 160, 225, 156, 255, 228, 251}, + {189, 131, 160, 226, 157, 255, 228, 251}, + {190, 131, 160, 226, 158, 255, 228, 251}, + {191, 132, 161, 227, 159, 255, 229, 251}, + {192, 132, 161, 227, 159, 255, 229, 251}, + {193, 133, 162, 228, 160, 255, 230, 252}, + {194, 133, 162, 229, 160, 255, 230, 252}, + {195, 134, 163, 230, 161, 255, 231, 252}, + {196, 134, 163, 230, 161, 255, 231, 252}, + {197, 135, 163, 231, 162, 255, 231, 252}, + {198, 135, 163, 231, 162, 255, 231, 252}, + {199, 136, 164, 232, 163, 255, 232, 252}, + {200, 136, 164, 232, 164, 255, 232, 252}, + {201, 137, 165, 233, 165, 255, 233, 252}, + {201, 137, 165, 233, 165, 255, 233, 252}, + {202, 138, 166, 233, 166, 255, 233, 252}, + {203, 138, 166, 233, 166, 255, 233, 252}, + {204, 139, 166, 234, 167, 255, 234, 252}, + {205, 139, 166, 234, 167, 255, 234, 252}, + {206, 140, 167, 235, 168, 255, 235, 252}, + {206, 140, 167, 235, 168, 255, 235, 252}, + {207, 141, 168, 236, 169, 255, 235, 252}, + {208, 141, 168, 236, 170, 255, 235, 252}, + {209, 142, 169, 237, 171, 255, 236, 252}, + {209, 143, 169, 237, 171, 255, 236, 252}, + {210, 144, 169, 237, 172, 255, 236, 252}, + {211, 144, 169, 237, 172, 255, 236, 252}, + {212, 145, 170, 238, 173, 255, 237, 252}, + {213, 145, 170, 238, 173, 255, 237, 252}, + {214, 146, 171, 239, 174, 255, 237, 253}, + {214, 146, 171, 239, 174, 255, 237, 253}, + {215, 147, 172, 240, 175, 255, 238, 253}, + {215, 147, 172, 240, 175, 255, 238, 253}, + {216, 148, 173, 240, 176, 255, 238, 253}, + {217, 148, 173, 240, 176, 255, 238, 253}, + {218, 149, 173, 241, 177, 255, 239, 253}, + {218, 149, 173, 241, 178, 255, 239, 253}, + {219, 150, 174, 241, 179, 255, 239, 253}, + {219, 151, 174, 241, 179, 255, 239, 253}, + {220, 152, 175, 242, 180, 255, 240, 253}, + {221, 152, 175, 242, 180, 255, 240, 253}, + {222, 153, 176, 242, 181, 255, 240, 253}, + {222, 153, 176, 242, 181, 255, 240, 253}, + {223, 154, 177, 243, 182, 255, 240, 253}, + {223, 154, 177, 243, 182, 255, 240, 253}, + {224, 155, 178, 244, 183, 255, 241, 253}, + {224, 155, 178, 244, 183, 255, 241, 253}, + {225, 156, 178, 244, 184, 255, 241, 253}, + {225, 157, 178, 244, 184, 255, 241, 253}, + {226, 158, 179, 244, 185, 255, 242, 253}, + {227, 158, 179, 244, 185, 255, 242, 253}, + {228, 159, 180, 245, 186, 255, 242, 253}, + {228, 159, 180, 245, 186, 255, 242, 253}, + {229, 160, 181, 245, 187, 255, 242, 253}, + {229, 160, 181, 245, 187, 255, 242, 253}, + {230, 161, 182, 246, 188, 255, 243, 253}, + {230, 162, 182, 246, 188, 255, 243, 253}, + {231, 163, 183, 246, 189, 255, 243, 253}, + {231, 163, 183, 246, 189, 255, 243, 253}, + {232, 164, 184, 247, 190, 255, 243, 253}, + {232, 164, 184, 247, 190, 255, 243, 253}, + {233, 165, 185, 247, 191, 255, 244, 253}, + {233, 165, 185, 247, 191, 255, 244, 253}, + {234, 166, 185, 247, 192, 255, 244, 253}, + {234, 167, 185, 247, 192, 255, 244, 253}, + {235, 168, 186, 248, 193, 255, 244, 253}, + {235, 168, 186, 248, 193, 255, 244, 253}, + {236, 169, 187, 248, 194, 255, 244, 253}, + {236, 169, 187, 248, 194, 255, 244, 253}, + {236, 170, 188, 248, 195, 255, 245, 253}, + {236, 170, 188, 248, 195, 255, 245, 253}, + {237, 171, 189, 249, 196, 255, 245, 254}, + {237, 172, 189, 249, 196, 255, 245, 254}, + {238, 173, 190, 249, 197, 255, 245, 254}, + {238, 173, 190, 249, 197, 255, 245, 254}, + {239, 174, 191, 249, 198, 255, 245, 254}, + {239, 174, 191, 249, 198, 255, 245, 254}, + {240, 175, 192, 249, 199, 255, 246, 254}, + {240, 176, 192, 249, 199, 255, 246, 254}, + {240, 177, 193, 250, 200, 255, 246, 254}, + {240, 177, 193, 250, 200, 255, 246, 254}, + {241, 178, 194, 250, 201, 255, 246, 254}, + {241, 178, 194, 250, 201, 255, 246, 254}, + {242, 179, 195, 250, 202, 255, 246, 254}, + {242, 180, 195, 250, 202, 255, 246, 254}, + {242, 181, 196, 250, 203, 255, 247, 254}, + {242, 181, 196, 250, 203, 255, 247, 254}, + {243, 182, 197, 251, 204, 255, 247, 254}, + {243, 183, 197, 251, 204, 255, 247, 254}, + {244, 184, 198, 251, 205, 255, 247, 254}, + {244, 184, 198, 251, 205, 255, 247, 254}, + {244, 185, 199, 251, 206, 255, 247, 254}, + {244, 185, 199, 251, 206, 255, 247, 254}, + {245, 186, 200, 251, 207, 255, 247, 254}, + {245, 187, 200, 251, 207, 255, 247, 254}, + {246, 188, 201, 252, 207, 255, 248, 254}, + {246, 188, 201, 252, 207, 255, 248, 254}, + {246, 189, 202, 252, 208, 255, 248, 254}, + {246, 190, 202, 252, 208, 255, 248, 254}, + {247, 191, 203, 252, 209, 255, 248, 254}, + {247, 191, 203, 252, 209, 255, 248, 254}, + {247, 192, 204, 252, 210, 255, 248, 254}, + {247, 193, 204, 252, 210, 255, 248, 254}, + {248, 194, 205, 252, 211, 255, 248, 254}, + {248, 194, 205, 252, 211, 255, 248, 254}, + {248, 195, 206, 252, 212, 255, 249, 254}, + {248, 196, 206, 252, 212, 255, 249, 254}, + {249, 197, 207, 253, 213, 255, 249, 254}, + {249, 197, 207, 253, 213, 255, 249, 254}, + {249, 198, 208, 253, 214, 255, 249, 254}, + {249, 199, 209, 253, 214, 255, 249, 254}, + {250, 200, 210, 253, 215, 255, 249, 254}, + {250, 200, 210, 253, 215, 255, 249, 254}, + {250, 201, 211, 253, 215, 255, 249, 254}, + {250, 202, 211, 253, 215, 255, 249, 254}, + {250, 203, 212, 253, 216, 255, 249, 254}, + {250, 203, 212, 253, 216, 255, 249, 254}, + {251, 204, 213, 253, 217, 255, 250, 254}, + {251, 205, 213, 253, 217, 255, 250, 254}, + {251, 206, 214, 254, 218, 255, 250, 254}, + {251, 206, 215, 254, 218, 255, 250, 254}, + {252, 207, 216, 254, 219, 255, 250, 254}, + {252, 208, 216, 254, 219, 255, 250, 254}, + {252, 209, 217, 254, 220, 255, 250, 254}, + {252, 210, 217, 254, 220, 255, 250, 254}, + {252, 211, 218, 254, 221, 255, 250, 254}, + {252, 212, 218, 254, 221, 255, 250, 254}, + {253, 213, 219, 254, 222, 255, 250, 254}, + {253, 213, 220, 254, 222, 255, 250, 254}, + {253, 214, 221, 254, 223, 255, 250, 254}, + {253, 215, 221, 254, 223, 255, 250, 254}, + {253, 216, 222, 254, 224, 255, 251, 254}, + {253, 217, 223, 254, 224, 255, 251, 254}, + {253, 218, 224, 254, 225, 255, 251, 254}, + {253, 219, 224, 254, 225, 255, 251, 254}, + {254, 220, 225, 254, 225, 255, 251, 254}, + {254, 221, 226, 254, 225, 255, 251, 254}, + {254, 222, 227, 255, 226, 255, 251, 254}, + {254, 223, 227, 255, 226, 255, 251, 254}, + {254, 224, 228, 255, 227, 255, 251, 254}, + {254, 225, 229, 255, 227, 255, 251, 254}, + {254, 226, 230, 255, 228, 255, 251, 254}, + {254, 227, 230, 255, 229, 255, 251, 254}, + {255, 228, 231, 255, 230, 255, 251, 254}, + {255, 229, 232, 255, 230, 255, 251, 254}, + {255, 230, 233, 255, 231, 255, 252, 254}, + {255, 231, 234, 255, 231, 255, 252, 254}, + {255, 232, 235, 255, 232, 255, 252, 254}, + {255, 233, 236, 255, 232, 255, 252, 254}, + {255, 235, 237, 255, 233, 255, 252, 254}, + {255, 236, 238, 255, 234, 255, 252, 254}, + {255, 238, 240, 255, 235, 255, 252, 255}, + {255, 239, 241, 255, 235, 255, 252, 254}, + {255, 241, 243, 255, 236, 255, 252, 254}, + {255, 243, 245, 255, 237, 255, 252, 254}, + {255, 246, 247, 255, 239, 255, 253, 255}, +}; + +static const vp9_coeff_probs_model default_coef_probs_4x4[PLANE_TYPES] = { + { // Y plane + { // Intra + { // Band 0 + { 195, 29, 183 }, { 84, 49, 136 }, { 8, 42, 71 } + }, { // Band 1 + { 31, 107, 169 }, { 35, 99, 159 }, { 17, 82, 140 }, + { 8, 66, 114 }, { 2, 44, 76 }, { 1, 19, 32 } + }, { // Band 2 + { 40, 132, 201 }, { 29, 114, 187 }, { 13, 91, 157 }, + { 7, 75, 127 }, { 3, 58, 95 }, { 1, 28, 47 } + }, { // Band 3 + { 69, 142, 221 }, { 42, 122, 201 }, { 15, 91, 159 }, + { 6, 67, 121 }, { 1, 42, 77 }, { 1, 17, 31 } + }, { // Band 4 + { 102, 148, 228 }, { 67, 117, 204 }, { 17, 82, 154 }, + { 6, 59, 114 }, { 2, 39, 75 }, { 1, 15, 29 } + }, { // Band 5 + { 156, 57, 233 }, { 119, 57, 212 }, { 58, 48, 163 }, + { 29, 40, 124 }, { 12, 30, 81 }, { 3, 12, 31 } + } + }, { // Inter + { // Band 0 + { 191, 107, 226 }, { 124, 117, 204 }, { 25, 99, 155 } + }, { // Band 1 + { 29, 148, 210 }, { 37, 126, 194 }, { 8, 93, 157 }, + { 2, 68, 118 }, { 1, 39, 69 }, { 1, 17, 33 } + }, { // Band 2 + { 41, 151, 213 }, { 27, 123, 193 }, { 3, 82, 144 }, + { 1, 58, 105 }, { 1, 32, 60 }, { 1, 13, 26 } + }, { // Band 3 + { 59, 159, 220 }, { 23, 126, 198 }, { 4, 88, 151 }, + { 1, 66, 114 }, { 1, 38, 71 }, { 1, 18, 34 } + }, { // Band 4 + { 114, 136, 232 }, { 51, 114, 207 }, { 11, 83, 155 }, + { 3, 56, 105 }, { 1, 33, 65 }, { 1, 17, 34 } + }, { // Band 5 + { 149, 65, 234 }, { 121, 57, 215 }, { 61, 49, 166 }, + { 28, 36, 114 }, { 12, 25, 76 }, { 3, 16, 42 } + } + } + }, { // UV plane + { // Intra + { // Band 0 + { 214, 49, 220 }, { 132, 63, 188 }, { 42, 65, 137 } + }, { // Band 1 + { 85, 137, 221 }, { 104, 131, 216 }, { 49, 111, 192 }, + { 21, 87, 155 }, { 2, 49, 87 }, { 1, 16, 28 } + }, { // Band 2 + { 89, 163, 230 }, { 90, 137, 220 }, { 29, 100, 183 }, + { 10, 70, 135 }, { 2, 42, 81 }, { 1, 17, 33 } + }, { // Band 3 + { 108, 167, 237 }, { 55, 133, 222 }, { 15, 97, 179 }, + { 4, 72, 135 }, { 1, 45, 85 }, { 1, 19, 38 } + }, { // Band 4 + { 124, 146, 240 }, { 66, 124, 224 }, { 17, 88, 175 }, + { 4, 58, 122 }, { 1, 36, 75 }, { 1, 18, 37 } + }, { // Band 5 + { 141, 79, 241 }, { 126, 70, 227 }, { 66, 58, 182 }, + { 30, 44, 136 }, { 12, 34, 96 }, { 2, 20, 47 } + } + }, { // Inter + { // Band 0 + { 229, 99, 249 }, { 143, 111, 235 }, { 46, 109, 192 } + }, { // Band 1 + { 82, 158, 236 }, { 94, 146, 224 }, { 25, 117, 191 }, + { 9, 87, 149 }, { 3, 56, 99 }, { 1, 33, 57 } + }, { // Band 2 + { 83, 167, 237 }, { 68, 145, 222 }, { 10, 103, 177 }, + { 2, 72, 131 }, { 1, 41, 79 }, { 1, 20, 39 } + }, { // Band 3 + { 99, 167, 239 }, { 47, 141, 224 }, { 10, 104, 178 }, + { 2, 73, 133 }, { 1, 44, 85 }, { 1, 22, 47 } + }, { // Band 4 + { 127, 145, 243 }, { 71, 129, 228 }, { 17, 93, 177 }, + { 3, 61, 124 }, { 1, 41, 84 }, { 1, 21, 52 } + }, { // Band 5 + { 157, 78, 244 }, { 140, 72, 231 }, { 69, 58, 184 }, + { 31, 44, 137 }, { 14, 38, 105 }, { 8, 23, 61 } + } + } + } +}; + +static const vp9_coeff_probs_model default_coef_probs_8x8[PLANE_TYPES] = { + { // Y plane + { // Intra + { // Band 0 + { 125, 34, 187 }, { 52, 41, 133 }, { 6, 31, 56 } + }, { // Band 1 + { 37, 109, 153 }, { 51, 102, 147 }, { 23, 87, 128 }, + { 8, 67, 101 }, { 1, 41, 63 }, { 1, 19, 29 } + }, { // Band 2 + { 31, 154, 185 }, { 17, 127, 175 }, { 6, 96, 145 }, + { 2, 73, 114 }, { 1, 51, 82 }, { 1, 28, 45 } + }, { // Band 3 + { 23, 163, 200 }, { 10, 131, 185 }, { 2, 93, 148 }, + { 1, 67, 111 }, { 1, 41, 69 }, { 1, 14, 24 } + }, { // Band 4 + { 29, 176, 217 }, { 12, 145, 201 }, { 3, 101, 156 }, + { 1, 69, 111 }, { 1, 39, 63 }, { 1, 14, 23 } + }, { // Band 5 + { 57, 192, 233 }, { 25, 154, 215 }, { 6, 109, 167 }, + { 3, 78, 118 }, { 1, 48, 69 }, { 1, 21, 29 } + } + }, { // Inter + { // Band 0 + { 202, 105, 245 }, { 108, 106, 216 }, { 18, 90, 144 } + }, { // Band 1 + { 33, 172, 219 }, { 64, 149, 206 }, { 14, 117, 177 }, + { 5, 90, 141 }, { 2, 61, 95 }, { 1, 37, 57 } + }, { // Band 2 + { 33, 179, 220 }, { 11, 140, 198 }, { 1, 89, 148 }, + { 1, 60, 104 }, { 1, 33, 57 }, { 1, 12, 21 } + }, { // Band 3 + { 30, 181, 221 }, { 8, 141, 198 }, { 1, 87, 145 }, + { 1, 58, 100 }, { 1, 31, 55 }, { 1, 12, 20 } + }, { // Band 4 + { 32, 186, 224 }, { 7, 142, 198 }, { 1, 86, 143 }, + { 1, 58, 100 }, { 1, 31, 55 }, { 1, 12, 22 } + }, { // Band 5 + { 57, 192, 227 }, { 20, 143, 204 }, { 3, 96, 154 }, + { 1, 68, 112 }, { 1, 42, 69 }, { 1, 19, 32 } + } + } + }, { // UV plane + { // Intra + { // Band 0 + { 212, 35, 215 }, { 113, 47, 169 }, { 29, 48, 105 } + }, { // Band 1 + { 74, 129, 203 }, { 106, 120, 203 }, { 49, 107, 178 }, + { 19, 84, 144 }, { 4, 50, 84 }, { 1, 15, 25 } + }, { // Band 2 + { 71, 172, 217 }, { 44, 141, 209 }, { 15, 102, 173 }, + { 6, 76, 133 }, { 2, 51, 89 }, { 1, 24, 42 } + }, { // Band 3 + { 64, 185, 231 }, { 31, 148, 216 }, { 8, 103, 175 }, + { 3, 74, 131 }, { 1, 46, 81 }, { 1, 18, 30 } + }, { // Band 4 + { 65, 196, 235 }, { 25, 157, 221 }, { 5, 105, 174 }, + { 1, 67, 120 }, { 1, 38, 69 }, { 1, 15, 30 } + }, { // Band 5 + { 65, 204, 238 }, { 30, 156, 224 }, { 7, 107, 177 }, + { 2, 70, 124 }, { 1, 42, 73 }, { 1, 18, 34 } + } + }, { // Inter + { // Band 0 + { 225, 86, 251 }, { 144, 104, 235 }, { 42, 99, 181 } + }, { // Band 1 + { 85, 175, 239 }, { 112, 165, 229 }, { 29, 136, 200 }, + { 12, 103, 162 }, { 6, 77, 123 }, { 2, 53, 84 } + }, { // Band 2 + { 75, 183, 239 }, { 30, 155, 221 }, { 3, 106, 171 }, + { 1, 74, 128 }, { 1, 44, 76 }, { 1, 17, 28 } + }, { // Band 3 + { 73, 185, 240 }, { 27, 159, 222 }, { 2, 107, 172 }, + { 1, 75, 127 }, { 1, 42, 73 }, { 1, 17, 29 } + }, { // Band 4 + { 62, 190, 238 }, { 21, 159, 222 }, { 2, 107, 172 }, + { 1, 72, 122 }, { 1, 40, 71 }, { 1, 18, 32 } + }, { // Band 5 + { 61, 199, 240 }, { 27, 161, 226 }, { 4, 113, 180 }, + { 1, 76, 129 }, { 1, 46, 80 }, { 1, 23, 41 } + } + } + } +}; + +static const vp9_coeff_probs_model default_coef_probs_16x16[PLANE_TYPES] = { + { // Y plane + { // Intra + { // Band 0 + { 7, 27, 153 }, { 5, 30, 95 }, { 1, 16, 30 } + }, { // Band 1 + { 50, 75, 127 }, { 57, 75, 124 }, { 27, 67, 108 }, + { 10, 54, 86 }, { 1, 33, 52 }, { 1, 12, 18 } + }, { // Band 2 + { 43, 125, 151 }, { 26, 108, 148 }, { 7, 83, 122 }, + { 2, 59, 89 }, { 1, 38, 60 }, { 1, 17, 27 } + }, { // Band 3 + { 23, 144, 163 }, { 13, 112, 154 }, { 2, 75, 117 }, + { 1, 50, 81 }, { 1, 31, 51 }, { 1, 14, 23 } + }, { // Band 4 + { 18, 162, 185 }, { 6, 123, 171 }, { 1, 78, 125 }, + { 1, 51, 86 }, { 1, 31, 54 }, { 1, 14, 23 } + }, { // Band 5 + { 15, 199, 227 }, { 3, 150, 204 }, { 1, 91, 146 }, + { 1, 55, 95 }, { 1, 30, 53 }, { 1, 11, 20 } + } + }, { // Inter + { // Band 0 + { 19, 55, 240 }, { 19, 59, 196 }, { 3, 52, 105 } + }, { // Band 1 + { 41, 166, 207 }, { 104, 153, 199 }, { 31, 123, 181 }, + { 14, 101, 152 }, { 5, 72, 106 }, { 1, 36, 52 } + }, { // Band 2 + { 35, 176, 211 }, { 12, 131, 190 }, { 2, 88, 144 }, + { 1, 60, 101 }, { 1, 36, 60 }, { 1, 16, 28 } + }, { // Band 3 + { 28, 183, 213 }, { 8, 134, 191 }, { 1, 86, 142 }, + { 1, 56, 96 }, { 1, 30, 53 }, { 1, 12, 20 } + }, { // Band 4 + { 20, 190, 215 }, { 4, 135, 192 }, { 1, 84, 139 }, + { 1, 53, 91 }, { 1, 28, 49 }, { 1, 11, 20 } + }, { // Band 5 + { 13, 196, 216 }, { 2, 137, 192 }, { 1, 86, 143 }, + { 1, 57, 99 }, { 1, 32, 56 }, { 1, 13, 24 } + } + } + }, { // UV plane + { // Intra + { // Band 0 + { 211, 29, 217 }, { 96, 47, 156 }, { 22, 43, 87 } + }, { // Band 1 + { 78, 120, 193 }, { 111, 116, 186 }, { 46, 102, 164 }, + { 15, 80, 128 }, { 2, 49, 76 }, { 1, 18, 28 } + }, { // Band 2 + { 71, 161, 203 }, { 42, 132, 192 }, { 10, 98, 150 }, + { 3, 69, 109 }, { 1, 44, 70 }, { 1, 18, 29 } + }, { // Band 3 + { 57, 186, 211 }, { 30, 140, 196 }, { 4, 93, 146 }, + { 1, 62, 102 }, { 1, 38, 65 }, { 1, 16, 27 } + }, { // Band 4 + { 47, 199, 217 }, { 14, 145, 196 }, { 1, 88, 142 }, + { 1, 57, 98 }, { 1, 36, 62 }, { 1, 15, 26 } + }, { // Band 5 + { 26, 219, 229 }, { 5, 155, 207 }, { 1, 94, 151 }, + { 1, 60, 104 }, { 1, 36, 62 }, { 1, 16, 28 } + } + }, { // Inter + { // Band 0 + { 233, 29, 248 }, { 146, 47, 220 }, { 43, 52, 140 } + }, { // Band 1 + { 100, 163, 232 }, { 179, 161, 222 }, { 63, 142, 204 }, + { 37, 113, 174 }, { 26, 89, 137 }, { 18, 68, 97 } + }, { // Band 2 + { 85, 181, 230 }, { 32, 146, 209 }, { 7, 100, 164 }, + { 3, 71, 121 }, { 1, 45, 77 }, { 1, 18, 30 } + }, { // Band 3 + { 65, 187, 230 }, { 20, 148, 207 }, { 2, 97, 159 }, + { 1, 68, 116 }, { 1, 40, 70 }, { 1, 14, 29 } + }, { // Band 4 + { 40, 194, 227 }, { 8, 147, 204 }, { 1, 94, 155 }, + { 1, 65, 112 }, { 1, 39, 66 }, { 1, 14, 26 } + }, { // Band 5 + { 16, 208, 228 }, { 3, 151, 207 }, { 1, 98, 160 }, + { 1, 67, 117 }, { 1, 41, 74 }, { 1, 17, 31 } + } + } + } +}; + +static const vp9_coeff_probs_model default_coef_probs_32x32[PLANE_TYPES] = { + { // Y plane + { // Intra + { // Band 0 + { 17, 38, 140 }, { 7, 34, 80 }, { 1, 17, 29 } + }, { // Band 1 + { 37, 75, 128 }, { 41, 76, 128 }, { 26, 66, 116 }, + { 12, 52, 94 }, { 2, 32, 55 }, { 1, 10, 16 } + }, { // Band 2 + { 50, 127, 154 }, { 37, 109, 152 }, { 16, 82, 121 }, + { 5, 59, 85 }, { 1, 35, 54 }, { 1, 13, 20 } + }, { // Band 3 + { 40, 142, 167 }, { 17, 110, 157 }, { 2, 71, 112 }, + { 1, 44, 72 }, { 1, 27, 45 }, { 1, 11, 17 } + }, { // Band 4 + { 30, 175, 188 }, { 9, 124, 169 }, { 1, 74, 116 }, + { 1, 48, 78 }, { 1, 30, 49 }, { 1, 11, 18 } + }, { // Band 5 + { 10, 222, 223 }, { 2, 150, 194 }, { 1, 83, 128 }, + { 1, 48, 79 }, { 1, 27, 45 }, { 1, 11, 17 } + } + }, { // Inter + { // Band 0 + { 36, 41, 235 }, { 29, 36, 193 }, { 10, 27, 111 } + }, { // Band 1 + { 85, 165, 222 }, { 177, 162, 215 }, { 110, 135, 195 }, + { 57, 113, 168 }, { 23, 83, 120 }, { 10, 49, 61 } + }, { // Band 2 + { 85, 190, 223 }, { 36, 139, 200 }, { 5, 90, 146 }, + { 1, 60, 103 }, { 1, 38, 65 }, { 1, 18, 30 } + }, { // Band 3 + { 72, 202, 223 }, { 23, 141, 199 }, { 2, 86, 140 }, + { 1, 56, 97 }, { 1, 36, 61 }, { 1, 16, 27 } + }, { // Band 4 + { 55, 218, 225 }, { 13, 145, 200 }, { 1, 86, 141 }, + { 1, 57, 99 }, { 1, 35, 61 }, { 1, 13, 22 } + }, { // Band 5 + { 15, 235, 212 }, { 1, 132, 184 }, { 1, 84, 139 }, + { 1, 57, 97 }, { 1, 34, 56 }, { 1, 14, 23 } + } + } + }, { // UV plane + { // Intra + { // Band 0 + { 181, 21, 201 }, { 61, 37, 123 }, { 10, 38, 71 } + }, { // Band 1 + { 47, 106, 172 }, { 95, 104, 173 }, { 42, 93, 159 }, + { 18, 77, 131 }, { 4, 50, 81 }, { 1, 17, 23 } + }, { // Band 2 + { 62, 147, 199 }, { 44, 130, 189 }, { 28, 102, 154 }, + { 18, 75, 115 }, { 2, 44, 65 }, { 1, 12, 19 } + }, { // Band 3 + { 55, 153, 210 }, { 24, 130, 194 }, { 3, 93, 146 }, + { 1, 61, 97 }, { 1, 31, 50 }, { 1, 10, 16 } + }, { // Band 4 + { 49, 186, 223 }, { 17, 148, 204 }, { 1, 96, 142 }, + { 1, 53, 83 }, { 1, 26, 44 }, { 1, 11, 17 } + }, { // Band 5 + { 13, 217, 212 }, { 2, 136, 180 }, { 1, 78, 124 }, + { 1, 50, 83 }, { 1, 29, 49 }, { 1, 14, 23 } + } + }, { // Inter + { // Band 0 + { 197, 13, 247 }, { 82, 17, 222 }, { 25, 17, 162 } + }, { // Band 1 + { 126, 186, 247 }, { 234, 191, 243 }, { 176, 177, 234 }, + { 104, 158, 220 }, { 66, 128, 186 }, { 55, 90, 137 } + }, { // Band 2 + { 111, 197, 242 }, { 46, 158, 219 }, { 9, 104, 171 }, + { 2, 65, 125 }, { 1, 44, 80 }, { 1, 17, 91 } + }, { // Band 3 + { 104, 208, 245 }, { 39, 168, 224 }, { 3, 109, 162 }, + { 1, 79, 124 }, { 1, 50, 102 }, { 1, 43, 102 } + }, { // Band 4 + { 84, 220, 246 }, { 31, 177, 231 }, { 2, 115, 180 }, + { 1, 79, 134 }, { 1, 55, 77 }, { 1, 60, 79 } + }, { // Band 5 + { 43, 243, 240 }, { 8, 180, 217 }, { 1, 115, 166 }, + { 1, 84, 121 }, { 1, 51, 67 }, { 1, 16, 6 } + } + } + } +}; + +static void extend_to_full_distribution(vpx_prob *probs, vpx_prob p) { + assert(p != 0); + memcpy(probs, vp9_pareto8_full[p - 1], MODEL_NODES * sizeof(vpx_prob)); +} + +void vp9_model_to_full_probs(const vpx_prob *model, vpx_prob *full) { + if (full != model) + memcpy(full, model, sizeof(vpx_prob) * UNCONSTRAINED_NODES); + extend_to_full_distribution(&full[UNCONSTRAINED_NODES], model[PIVOT_NODE]); +} + +void vp9_default_coef_probs(VP9_COMMON *cm) { + vp9_copy(cm->fc->coef_probs[TX_4X4], default_coef_probs_4x4); + vp9_copy(cm->fc->coef_probs[TX_8X8], default_coef_probs_8x8); + vp9_copy(cm->fc->coef_probs[TX_16X16], default_coef_probs_16x16); + vp9_copy(cm->fc->coef_probs[TX_32X32], default_coef_probs_32x32); +} + +#define COEF_COUNT_SAT 24 +#define COEF_MAX_UPDATE_FACTOR 112 +#define COEF_COUNT_SAT_KEY 24 +#define COEF_MAX_UPDATE_FACTOR_KEY 112 +#define COEF_COUNT_SAT_AFTER_KEY 24 +#define COEF_MAX_UPDATE_FACTOR_AFTER_KEY 128 + +static void adapt_coef_probs(VP9_COMMON *cm, TX_SIZE tx_size, + unsigned int count_sat, + unsigned int update_factor) { + const FRAME_CONTEXT *pre_fc = &cm->frame_contexts[cm->frame_context_idx]; + vp9_coeff_probs_model *const probs = cm->fc->coef_probs[tx_size]; + const vp9_coeff_probs_model *const pre_probs = pre_fc->coef_probs[tx_size]; + vp9_coeff_count_model *counts = cm->counts.coef[tx_size]; + unsigned int (*eob_counts)[REF_TYPES][COEF_BANDS][COEFF_CONTEXTS] = + cm->counts.eob_branch[tx_size]; + int i, j, k, l, m; + + for (i = 0; i < PLANE_TYPES; ++i) + for (j = 0; j < REF_TYPES; ++j) + for (k = 0; k < COEF_BANDS; ++k) + for (l = 0; l < BAND_COEFF_CONTEXTS(k); ++l) { + const int n0 = counts[i][j][k][l][ZERO_TOKEN]; + const int n1 = counts[i][j][k][l][ONE_TOKEN]; + const int n2 = counts[i][j][k][l][TWO_TOKEN]; + const int neob = counts[i][j][k][l][EOB_MODEL_TOKEN]; + const unsigned int branch_ct[UNCONSTRAINED_NODES][2] = { + { neob, eob_counts[i][j][k][l] - neob }, + { n0, n1 + n2 }, + { n1, n2 } + }; + for (m = 0; m < UNCONSTRAINED_NODES; ++m) + probs[i][j][k][l][m] = merge_probs(pre_probs[i][j][k][l][m], + branch_ct[m], + count_sat, update_factor); + } +} + +void vp9_adapt_coef_probs(VP9_COMMON *cm) { + TX_SIZE t; + unsigned int count_sat, update_factor; + + if (frame_is_intra_only(cm)) { + update_factor = COEF_MAX_UPDATE_FACTOR_KEY; + count_sat = COEF_COUNT_SAT_KEY; + } else if (cm->last_frame_type == KEY_FRAME) { + update_factor = COEF_MAX_UPDATE_FACTOR_AFTER_KEY; /* adapt quickly */ + count_sat = COEF_COUNT_SAT_AFTER_KEY; + } else { + update_factor = COEF_MAX_UPDATE_FACTOR; + count_sat = COEF_COUNT_SAT; + } + for (t = TX_4X4; t <= TX_32X32; t++) + adapt_coef_probs(cm, t, count_sat, update_factor); +} diff --git a/thirdparty/libvpx/vp9/common/vp9_entropy.h b/thirdparty/libvpx/vp9/common/vp9_entropy.h new file mode 100644 index 0000000000..63b3bff5d9 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_entropy.h @@ -0,0 +1,200 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_COMMON_VP9_ENTROPY_H_ +#define VP9_COMMON_VP9_ENTROPY_H_ + +#include "vpx/vpx_integer.h" +#include "vpx_dsp/prob.h" + +#include "vp9/common/vp9_common.h" +#include "vp9/common/vp9_enums.h" + +#ifdef __cplusplus +extern "C" { +#endif + +#define DIFF_UPDATE_PROB 252 + +// Coefficient token alphabet +#define ZERO_TOKEN 0 // 0 Extra Bits 0+0 +#define ONE_TOKEN 1 // 1 Extra Bits 0+1 +#define TWO_TOKEN 2 // 2 Extra Bits 0+1 +#define THREE_TOKEN 3 // 3 Extra Bits 0+1 +#define FOUR_TOKEN 4 // 4 Extra Bits 0+1 +#define CATEGORY1_TOKEN 5 // 5-6 Extra Bits 1+1 +#define CATEGORY2_TOKEN 6 // 7-10 Extra Bits 2+1 +#define CATEGORY3_TOKEN 7 // 11-18 Extra Bits 3+1 +#define CATEGORY4_TOKEN 8 // 19-34 Extra Bits 4+1 +#define CATEGORY5_TOKEN 9 // 35-66 Extra Bits 5+1 +#define CATEGORY6_TOKEN 10 // 67+ Extra Bits 14+1 +#define EOB_TOKEN 11 // EOB Extra Bits 0+0 + +#define ENTROPY_TOKENS 12 + +#define ENTROPY_NODES 11 + +DECLARE_ALIGNED(16, extern const uint8_t, vp9_pt_energy_class[ENTROPY_TOKENS]); + +#define CAT1_MIN_VAL 5 +#define CAT2_MIN_VAL 7 +#define CAT3_MIN_VAL 11 +#define CAT4_MIN_VAL 19 +#define CAT5_MIN_VAL 35 +#define CAT6_MIN_VAL 67 + +// Extra bit probabilities. +DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat1_prob[1]); +DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat2_prob[2]); +DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat3_prob[3]); +DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat4_prob[4]); +DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat5_prob[5]); +DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat6_prob[14]); + +#if CONFIG_VP9_HIGHBITDEPTH +DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat1_prob_high10[1]); +DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat2_prob_high10[2]); +DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat3_prob_high10[3]); +DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat4_prob_high10[4]); +DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat5_prob_high10[5]); +DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat6_prob_high10[16]); +DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat1_prob_high12[1]); +DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat2_prob_high12[2]); +DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat3_prob_high12[3]); +DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat4_prob_high12[4]); +DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat5_prob_high12[5]); +DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat6_prob_high12[18]); +#endif // CONFIG_VP9_HIGHBITDEPTH + +#define EOB_MODEL_TOKEN 3 + +#define DCT_MAX_VALUE 16384 +#if CONFIG_VP9_HIGHBITDEPTH +#define DCT_MAX_VALUE_HIGH10 65536 +#define DCT_MAX_VALUE_HIGH12 262144 +#endif // CONFIG_VP9_HIGHBITDEPTH + +/* Coefficients are predicted via a 3-dimensional probability table. */ + +#define REF_TYPES 2 // intra=0, inter=1 + +/* Middle dimension reflects the coefficient position within the transform. */ +#define COEF_BANDS 6 + +/* Inside dimension is measure of nearby complexity, that reflects the energy + of nearby coefficients are nonzero. For the first coefficient (DC, unless + block type is 0), we look at the (already encoded) blocks above and to the + left of the current block. The context index is then the number (0,1,or 2) + of these blocks having nonzero coefficients. + After decoding a coefficient, the measure is determined by the size of the + most recently decoded coefficient. + Note that the intuitive meaning of this measure changes as coefficients + are decoded, e.g., prior to the first token, a zero means that my neighbors + are empty while, after the first token, because of the use of end-of-block, + a zero means we just decoded a zero and hence guarantees that a non-zero + coefficient will appear later in this block. However, this shift + in meaning is perfectly OK because our context depends also on the + coefficient band (and since zigzag positions 0, 1, and 2 are in + distinct bands). */ + +#define COEFF_CONTEXTS 6 +#define BAND_COEFF_CONTEXTS(band) ((band) == 0 ? 3 : COEFF_CONTEXTS) + +// #define ENTROPY_STATS + +typedef unsigned int vp9_coeff_count[REF_TYPES][COEF_BANDS][COEFF_CONTEXTS] + [ENTROPY_TOKENS]; +typedef unsigned int vp9_coeff_stats[REF_TYPES][COEF_BANDS][COEFF_CONTEXTS] + [ENTROPY_NODES][2]; + +#define SUBEXP_PARAM 4 /* Subexponential code parameter */ +#define MODULUS_PARAM 13 /* Modulus parameter */ + +struct VP9Common; +void vp9_default_coef_probs(struct VP9Common *cm); +void vp9_adapt_coef_probs(struct VP9Common *cm); + +// This is the index in the scan order beyond which all coefficients for +// 8x8 transform and above are in the top band. +// This macro is currently unused but may be used by certain implementations +#define MAXBAND_INDEX 21 + +DECLARE_ALIGNED(16, extern const uint8_t, vp9_coefband_trans_8x8plus[1024]); +DECLARE_ALIGNED(16, extern const uint8_t, vp9_coefband_trans_4x4[16]); + +static INLINE const uint8_t *get_band_translate(TX_SIZE tx_size) { + return tx_size == TX_4X4 ? vp9_coefband_trans_4x4 + : vp9_coefband_trans_8x8plus; +} + +// 128 lists of probabilities are stored for the following ONE node probs: +// 1, 3, 5, 7, ..., 253, 255 +// In between probabilities are interpolated linearly + +#define COEFF_PROB_MODELS 255 + +#define UNCONSTRAINED_NODES 3 + +#define PIVOT_NODE 2 // which node is pivot + +#define MODEL_NODES (ENTROPY_NODES - UNCONSTRAINED_NODES) +extern const vpx_tree_index vp9_coef_con_tree[TREE_SIZE(ENTROPY_TOKENS)]; +extern const vpx_prob vp9_pareto8_full[COEFF_PROB_MODELS][MODEL_NODES]; + +typedef vpx_prob vp9_coeff_probs_model[REF_TYPES][COEF_BANDS] + [COEFF_CONTEXTS][UNCONSTRAINED_NODES]; + +typedef unsigned int vp9_coeff_count_model[REF_TYPES][COEF_BANDS] + [COEFF_CONTEXTS] + [UNCONSTRAINED_NODES + 1]; + +void vp9_model_to_full_probs(const vpx_prob *model, vpx_prob *full); + +typedef char ENTROPY_CONTEXT; + +static INLINE int combine_entropy_contexts(ENTROPY_CONTEXT a, + ENTROPY_CONTEXT b) { + return (a != 0) + (b != 0); +} + +static INLINE int get_entropy_context(TX_SIZE tx_size, const ENTROPY_CONTEXT *a, + const ENTROPY_CONTEXT *l) { + ENTROPY_CONTEXT above_ec = 0, left_ec = 0; + + switch (tx_size) { + case TX_4X4: + above_ec = a[0] != 0; + left_ec = l[0] != 0; + break; + case TX_8X8: + above_ec = !!*(const uint16_t *)a; + left_ec = !!*(const uint16_t *)l; + break; + case TX_16X16: + above_ec = !!*(const uint32_t *)a; + left_ec = !!*(const uint32_t *)l; + break; + case TX_32X32: + above_ec = !!*(const uint64_t *)a; + left_ec = !!*(const uint64_t *)l; + break; + default: + assert(0 && "Invalid transform size."); + break; + } + + return combine_entropy_contexts(above_ec, left_ec); +} + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_ENTROPY_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_entropymode.c b/thirdparty/libvpx/vp9/common/vp9_entropymode.c new file mode 100644 index 0000000000..670348bafd --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_entropymode.c @@ -0,0 +1,469 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "vpx_mem/vpx_mem.h" + +#include "vp9/common/vp9_onyxc_int.h" +#include "vp9/common/vp9_seg_common.h" + +const vpx_prob vp9_kf_y_mode_prob[INTRA_MODES][INTRA_MODES][INTRA_MODES - 1] = { + { // above = dc + { 137, 30, 42, 148, 151, 207, 70, 52, 91 }, // left = dc + { 92, 45, 102, 136, 116, 180, 74, 90, 100 }, // left = v + { 73, 32, 19, 187, 222, 215, 46, 34, 100 }, // left = h + { 91, 30, 32, 116, 121, 186, 93, 86, 94 }, // left = d45 + { 72, 35, 36, 149, 68, 206, 68, 63, 105 }, // left = d135 + { 73, 31, 28, 138, 57, 124, 55, 122, 151 }, // left = d117 + { 67, 23, 21, 140, 126, 197, 40, 37, 171 }, // left = d153 + { 86, 27, 28, 128, 154, 212, 45, 43, 53 }, // left = d207 + { 74, 32, 27, 107, 86, 160, 63, 134, 102 }, // left = d63 + { 59, 67, 44, 140, 161, 202, 78, 67, 119 } // left = tm + }, { // above = v + { 63, 36, 126, 146, 123, 158, 60, 90, 96 }, // left = dc + { 43, 46, 168, 134, 107, 128, 69, 142, 92 }, // left = v + { 44, 29, 68, 159, 201, 177, 50, 57, 77 }, // left = h + { 58, 38, 76, 114, 97, 172, 78, 133, 92 }, // left = d45 + { 46, 41, 76, 140, 63, 184, 69, 112, 57 }, // left = d135 + { 38, 32, 85, 140, 46, 112, 54, 151, 133 }, // left = d117 + { 39, 27, 61, 131, 110, 175, 44, 75, 136 }, // left = d153 + { 52, 30, 74, 113, 130, 175, 51, 64, 58 }, // left = d207 + { 47, 35, 80, 100, 74, 143, 64, 163, 74 }, // left = d63 + { 36, 61, 116, 114, 128, 162, 80, 125, 82 } // left = tm + }, { // above = h + { 82, 26, 26, 171, 208, 204, 44, 32, 105 }, // left = dc + { 55, 44, 68, 166, 179, 192, 57, 57, 108 }, // left = v + { 42, 26, 11, 199, 241, 228, 23, 15, 85 }, // left = h + { 68, 42, 19, 131, 160, 199, 55, 52, 83 }, // left = d45 + { 58, 50, 25, 139, 115, 232, 39, 52, 118 }, // left = d135 + { 50, 35, 33, 153, 104, 162, 64, 59, 131 }, // left = d117 + { 44, 24, 16, 150, 177, 202, 33, 19, 156 }, // left = d153 + { 55, 27, 12, 153, 203, 218, 26, 27, 49 }, // left = d207 + { 53, 49, 21, 110, 116, 168, 59, 80, 76 }, // left = d63 + { 38, 72, 19, 168, 203, 212, 50, 50, 107 } // left = tm + }, { // above = d45 + { 103, 26, 36, 129, 132, 201, 83, 80, 93 }, // left = dc + { 59, 38, 83, 112, 103, 162, 98, 136, 90 }, // left = v + { 62, 30, 23, 158, 200, 207, 59, 57, 50 }, // left = h + { 67, 30, 29, 84, 86, 191, 102, 91, 59 }, // left = d45 + { 60, 32, 33, 112, 71, 220, 64, 89, 104 }, // left = d135 + { 53, 26, 34, 130, 56, 149, 84, 120, 103 }, // left = d117 + { 53, 21, 23, 133, 109, 210, 56, 77, 172 }, // left = d153 + { 77, 19, 29, 112, 142, 228, 55, 66, 36 }, // left = d207 + { 61, 29, 29, 93, 97, 165, 83, 175, 162 }, // left = d63 + { 47, 47, 43, 114, 137, 181, 100, 99, 95 } // left = tm + }, { // above = d135 + { 69, 23, 29, 128, 83, 199, 46, 44, 101 }, // left = dc + { 53, 40, 55, 139, 69, 183, 61, 80, 110 }, // left = v + { 40, 29, 19, 161, 180, 207, 43, 24, 91 }, // left = h + { 60, 34, 19, 105, 61, 198, 53, 64, 89 }, // left = d45 + { 52, 31, 22, 158, 40, 209, 58, 62, 89 }, // left = d135 + { 44, 31, 29, 147, 46, 158, 56, 102, 198 }, // left = d117 + { 35, 19, 12, 135, 87, 209, 41, 45, 167 }, // left = d153 + { 55, 25, 21, 118, 95, 215, 38, 39, 66 }, // left = d207 + { 51, 38, 25, 113, 58, 164, 70, 93, 97 }, // left = d63 + { 47, 54, 34, 146, 108, 203, 72, 103, 151 } // left = tm + }, { // above = d117 + { 64, 19, 37, 156, 66, 138, 49, 95, 133 }, // left = dc + { 46, 27, 80, 150, 55, 124, 55, 121, 135 }, // left = v + { 36, 23, 27, 165, 149, 166, 54, 64, 118 }, // left = h + { 53, 21, 36, 131, 63, 163, 60, 109, 81 }, // left = d45 + { 40, 26, 35, 154, 40, 185, 51, 97, 123 }, // left = d135 + { 35, 19, 34, 179, 19, 97, 48, 129, 124 }, // left = d117 + { 36, 20, 26, 136, 62, 164, 33, 77, 154 }, // left = d153 + { 45, 18, 32, 130, 90, 157, 40, 79, 91 }, // left = d207 + { 45, 26, 28, 129, 45, 129, 49, 147, 123 }, // left = d63 + { 38, 44, 51, 136, 74, 162, 57, 97, 121 } // left = tm + }, { // above = d153 + { 75, 17, 22, 136, 138, 185, 32, 34, 166 }, // left = dc + { 56, 39, 58, 133, 117, 173, 48, 53, 187 }, // left = v + { 35, 21, 12, 161, 212, 207, 20, 23, 145 }, // left = h + { 56, 29, 19, 117, 109, 181, 55, 68, 112 }, // left = d45 + { 47, 29, 17, 153, 64, 220, 59, 51, 114 }, // left = d135 + { 46, 16, 24, 136, 76, 147, 41, 64, 172 }, // left = d117 + { 34, 17, 11, 108, 152, 187, 13, 15, 209 }, // left = d153 + { 51, 24, 14, 115, 133, 209, 32, 26, 104 }, // left = d207 + { 55, 30, 18, 122, 79, 179, 44, 88, 116 }, // left = d63 + { 37, 49, 25, 129, 168, 164, 41, 54, 148 } // left = tm + }, { // above = d207 + { 82, 22, 32, 127, 143, 213, 39, 41, 70 }, // left = dc + { 62, 44, 61, 123, 105, 189, 48, 57, 64 }, // left = v + { 47, 25, 17, 175, 222, 220, 24, 30, 86 }, // left = h + { 68, 36, 17, 106, 102, 206, 59, 74, 74 }, // left = d45 + { 57, 39, 23, 151, 68, 216, 55, 63, 58 }, // left = d135 + { 49, 30, 35, 141, 70, 168, 82, 40, 115 }, // left = d117 + { 51, 25, 15, 136, 129, 202, 38, 35, 139 }, // left = d153 + { 68, 26, 16, 111, 141, 215, 29, 28, 28 }, // left = d207 + { 59, 39, 19, 114, 75, 180, 77, 104, 42 }, // left = d63 + { 40, 61, 26, 126, 152, 206, 61, 59, 93 } // left = tm + }, { // above = d63 + { 78, 23, 39, 111, 117, 170, 74, 124, 94 }, // left = dc + { 48, 34, 86, 101, 92, 146, 78, 179, 134 }, // left = v + { 47, 22, 24, 138, 187, 178, 68, 69, 59 }, // left = h + { 56, 25, 33, 105, 112, 187, 95, 177, 129 }, // left = d45 + { 48, 31, 27, 114, 63, 183, 82, 116, 56 }, // left = d135 + { 43, 28, 37, 121, 63, 123, 61, 192, 169 }, // left = d117 + { 42, 17, 24, 109, 97, 177, 56, 76, 122 }, // left = d153 + { 58, 18, 28, 105, 139, 182, 70, 92, 63 }, // left = d207 + { 46, 23, 32, 74, 86, 150, 67, 183, 88 }, // left = d63 + { 36, 38, 48, 92, 122, 165, 88, 137, 91 } // left = tm + }, { // above = tm + { 65, 70, 60, 155, 159, 199, 61, 60, 81 }, // left = dc + { 44, 78, 115, 132, 119, 173, 71, 112, 93 }, // left = v + { 39, 38, 21, 184, 227, 206, 42, 32, 64 }, // left = h + { 58, 47, 36, 124, 137, 193, 80, 82, 78 }, // left = d45 + { 49, 50, 35, 144, 95, 205, 63, 78, 59 }, // left = d135 + { 41, 53, 52, 148, 71, 142, 65, 128, 51 }, // left = d117 + { 40, 36, 28, 143, 143, 202, 40, 55, 137 }, // left = d153 + { 52, 34, 29, 129, 183, 227, 42, 35, 43 }, // left = d207 + { 42, 44, 44, 104, 105, 164, 64, 130, 80 }, // left = d63 + { 43, 81, 53, 140, 169, 204, 68, 84, 72 } // left = tm + } +}; + +const vpx_prob vp9_kf_uv_mode_prob[INTRA_MODES][INTRA_MODES - 1] = { + { 144, 11, 54, 157, 195, 130, 46, 58, 108 }, // y = dc + { 118, 15, 123, 148, 131, 101, 44, 93, 131 }, // y = v + { 113, 12, 23, 188, 226, 142, 26, 32, 125 }, // y = h + { 120, 11, 50, 123, 163, 135, 64, 77, 103 }, // y = d45 + { 113, 9, 36, 155, 111, 157, 32, 44, 161 }, // y = d135 + { 116, 9, 55, 176, 76, 96, 37, 61, 149 }, // y = d117 + { 115, 9, 28, 141, 161, 167, 21, 25, 193 }, // y = d153 + { 120, 12, 32, 145, 195, 142, 32, 38, 86 }, // y = d207 + { 116, 12, 64, 120, 140, 125, 49, 115, 121 }, // y = d63 + { 102, 19, 66, 162, 182, 122, 35, 59, 128 } // y = tm +}; + +static const vpx_prob default_if_y_probs[BLOCK_SIZE_GROUPS][INTRA_MODES - 1] = { + { 65, 32, 18, 144, 162, 194, 41, 51, 98 }, // block_size < 8x8 + { 132, 68, 18, 165, 217, 196, 45, 40, 78 }, // block_size < 16x16 + { 173, 80, 19, 176, 240, 193, 64, 35, 46 }, // block_size < 32x32 + { 221, 135, 38, 194, 248, 121, 96, 85, 29 } // block_size >= 32x32 +}; + +static const vpx_prob default_if_uv_probs[INTRA_MODES][INTRA_MODES - 1] = { + { 120, 7, 76, 176, 208, 126, 28, 54, 103 }, // y = dc + { 48, 12, 154, 155, 139, 90, 34, 117, 119 }, // y = v + { 67, 6, 25, 204, 243, 158, 13, 21, 96 }, // y = h + { 97, 5, 44, 131, 176, 139, 48, 68, 97 }, // y = d45 + { 83, 5, 42, 156, 111, 152, 26, 49, 152 }, // y = d135 + { 80, 5, 58, 178, 74, 83, 33, 62, 145 }, // y = d117 + { 86, 5, 32, 154, 192, 168, 14, 22, 163 }, // y = d153 + { 85, 5, 32, 156, 216, 148, 19, 29, 73 }, // y = d207 + { 77, 7, 64, 116, 132, 122, 37, 126, 120 }, // y = d63 + { 101, 21, 107, 181, 192, 103, 19, 67, 125 } // y = tm +}; + +const vpx_prob vp9_kf_partition_probs[PARTITION_CONTEXTS] + [PARTITION_TYPES - 1] = { + // 8x8 -> 4x4 + { 158, 97, 94 }, // a/l both not split + { 93, 24, 99 }, // a split, l not split + { 85, 119, 44 }, // l split, a not split + { 62, 59, 67 }, // a/l both split + // 16x16 -> 8x8 + { 149, 53, 53 }, // a/l both not split + { 94, 20, 48 }, // a split, l not split + { 83, 53, 24 }, // l split, a not split + { 52, 18, 18 }, // a/l both split + // 32x32 -> 16x16 + { 150, 40, 39 }, // a/l both not split + { 78, 12, 26 }, // a split, l not split + { 67, 33, 11 }, // l split, a not split + { 24, 7, 5 }, // a/l both split + // 64x64 -> 32x32 + { 174, 35, 49 }, // a/l both not split + { 68, 11, 27 }, // a split, l not split + { 57, 15, 9 }, // l split, a not split + { 12, 3, 3 }, // a/l both split +}; + +static const vpx_prob default_partition_probs[PARTITION_CONTEXTS] + [PARTITION_TYPES - 1] = { + // 8x8 -> 4x4 + { 199, 122, 141 }, // a/l both not split + { 147, 63, 159 }, // a split, l not split + { 148, 133, 118 }, // l split, a not split + { 121, 104, 114 }, // a/l both split + // 16x16 -> 8x8 + { 174, 73, 87 }, // a/l both not split + { 92, 41, 83 }, // a split, l not split + { 82, 99, 50 }, // l split, a not split + { 53, 39, 39 }, // a/l both split + // 32x32 -> 16x16 + { 177, 58, 59 }, // a/l both not split + { 68, 26, 63 }, // a split, l not split + { 52, 79, 25 }, // l split, a not split + { 17, 14, 12 }, // a/l both split + // 64x64 -> 32x32 + { 222, 34, 30 }, // a/l both not split + { 72, 16, 44 }, // a split, l not split + { 58, 32, 12 }, // l split, a not split + { 10, 7, 6 }, // a/l both split +}; + +static const vpx_prob default_inter_mode_probs[INTER_MODE_CONTEXTS] + [INTER_MODES - 1] = { + {2, 173, 34}, // 0 = both zero mv + {7, 145, 85}, // 1 = one zero mv + one a predicted mv + {7, 166, 63}, // 2 = two predicted mvs + {7, 94, 66}, // 3 = one predicted/zero and one new mv + {8, 64, 46}, // 4 = two new mvs + {17, 81, 31}, // 5 = one intra neighbour + x + {25, 29, 30}, // 6 = two intra neighbours +}; + +/* Array indices are identical to previously-existing INTRAMODECONTEXTNODES. */ +const vpx_tree_index vp9_intra_mode_tree[TREE_SIZE(INTRA_MODES)] = { + -DC_PRED, 2, /* 0 = DC_NODE */ + -TM_PRED, 4, /* 1 = TM_NODE */ + -V_PRED, 6, /* 2 = V_NODE */ + 8, 12, /* 3 = COM_NODE */ + -H_PRED, 10, /* 4 = H_NODE */ + -D135_PRED, -D117_PRED, /* 5 = D135_NODE */ + -D45_PRED, 14, /* 6 = D45_NODE */ + -D63_PRED, 16, /* 7 = D63_NODE */ + -D153_PRED, -D207_PRED /* 8 = D153_NODE */ +}; + +const vpx_tree_index vp9_inter_mode_tree[TREE_SIZE(INTER_MODES)] = { + -INTER_OFFSET(ZEROMV), 2, + -INTER_OFFSET(NEARESTMV), 4, + -INTER_OFFSET(NEARMV), -INTER_OFFSET(NEWMV) +}; + +const vpx_tree_index vp9_partition_tree[TREE_SIZE(PARTITION_TYPES)] = { + -PARTITION_NONE, 2, + -PARTITION_HORZ, 4, + -PARTITION_VERT, -PARTITION_SPLIT +}; + +static const vpx_prob default_intra_inter_p[INTRA_INTER_CONTEXTS] = { + 9, 102, 187, 225 +}; + +static const vpx_prob default_comp_inter_p[COMP_INTER_CONTEXTS] = { + 239, 183, 119, 96, 41 +}; + +static const vpx_prob default_comp_ref_p[REF_CONTEXTS] = { + 50, 126, 123, 221, 226 +}; + +static const vpx_prob default_single_ref_p[REF_CONTEXTS][2] = { + { 33, 16 }, + { 77, 74 }, + { 142, 142 }, + { 172, 170 }, + { 238, 247 } +}; + +static const struct tx_probs default_tx_probs = { + { { 3, 136, 37 }, + { 5, 52, 13 } }, + + { { 20, 152 }, + { 15, 101 } }, + + { { 100 }, + { 66 } } +}; + +void tx_counts_to_branch_counts_32x32(const unsigned int *tx_count_32x32p, + unsigned int (*ct_32x32p)[2]) { + ct_32x32p[0][0] = tx_count_32x32p[TX_4X4]; + ct_32x32p[0][1] = tx_count_32x32p[TX_8X8] + + tx_count_32x32p[TX_16X16] + + tx_count_32x32p[TX_32X32]; + ct_32x32p[1][0] = tx_count_32x32p[TX_8X8]; + ct_32x32p[1][1] = tx_count_32x32p[TX_16X16] + + tx_count_32x32p[TX_32X32]; + ct_32x32p[2][0] = tx_count_32x32p[TX_16X16]; + ct_32x32p[2][1] = tx_count_32x32p[TX_32X32]; +} + +void tx_counts_to_branch_counts_16x16(const unsigned int *tx_count_16x16p, + unsigned int (*ct_16x16p)[2]) { + ct_16x16p[0][0] = tx_count_16x16p[TX_4X4]; + ct_16x16p[0][1] = tx_count_16x16p[TX_8X8] + tx_count_16x16p[TX_16X16]; + ct_16x16p[1][0] = tx_count_16x16p[TX_8X8]; + ct_16x16p[1][1] = tx_count_16x16p[TX_16X16]; +} + +void tx_counts_to_branch_counts_8x8(const unsigned int *tx_count_8x8p, + unsigned int (*ct_8x8p)[2]) { + ct_8x8p[0][0] = tx_count_8x8p[TX_4X4]; + ct_8x8p[0][1] = tx_count_8x8p[TX_8X8]; +} + +static const vpx_prob default_skip_probs[SKIP_CONTEXTS] = { + 192, 128, 64 +}; + +static const vpx_prob default_switchable_interp_prob[SWITCHABLE_FILTER_CONTEXTS] + [SWITCHABLE_FILTERS - 1] = { + { 235, 162, }, + { 36, 255, }, + { 34, 3, }, + { 149, 144, }, +}; + +static void init_mode_probs(FRAME_CONTEXT *fc) { + vp9_copy(fc->uv_mode_prob, default_if_uv_probs); + vp9_copy(fc->y_mode_prob, default_if_y_probs); + vp9_copy(fc->switchable_interp_prob, default_switchable_interp_prob); + vp9_copy(fc->partition_prob, default_partition_probs); + vp9_copy(fc->intra_inter_prob, default_intra_inter_p); + vp9_copy(fc->comp_inter_prob, default_comp_inter_p); + vp9_copy(fc->comp_ref_prob, default_comp_ref_p); + vp9_copy(fc->single_ref_prob, default_single_ref_p); + fc->tx_probs = default_tx_probs; + vp9_copy(fc->skip_probs, default_skip_probs); + vp9_copy(fc->inter_mode_probs, default_inter_mode_probs); +} + +const vpx_tree_index vp9_switchable_interp_tree + [TREE_SIZE(SWITCHABLE_FILTERS)] = { + -EIGHTTAP, 2, + -EIGHTTAP_SMOOTH, -EIGHTTAP_SHARP +}; + +void vp9_adapt_mode_probs(VP9_COMMON *cm) { + int i, j; + FRAME_CONTEXT *fc = cm->fc; + const FRAME_CONTEXT *pre_fc = &cm->frame_contexts[cm->frame_context_idx]; + const FRAME_COUNTS *counts = &cm->counts; + + for (i = 0; i < INTRA_INTER_CONTEXTS; i++) + fc->intra_inter_prob[i] = mode_mv_merge_probs(pre_fc->intra_inter_prob[i], + counts->intra_inter[i]); + for (i = 0; i < COMP_INTER_CONTEXTS; i++) + fc->comp_inter_prob[i] = mode_mv_merge_probs(pre_fc->comp_inter_prob[i], + counts->comp_inter[i]); + for (i = 0; i < REF_CONTEXTS; i++) + fc->comp_ref_prob[i] = mode_mv_merge_probs(pre_fc->comp_ref_prob[i], + counts->comp_ref[i]); + for (i = 0; i < REF_CONTEXTS; i++) + for (j = 0; j < 2; j++) + fc->single_ref_prob[i][j] = mode_mv_merge_probs( + pre_fc->single_ref_prob[i][j], counts->single_ref[i][j]); + + for (i = 0; i < INTER_MODE_CONTEXTS; i++) + vpx_tree_merge_probs(vp9_inter_mode_tree, pre_fc->inter_mode_probs[i], + counts->inter_mode[i], fc->inter_mode_probs[i]); + + for (i = 0; i < BLOCK_SIZE_GROUPS; i++) + vpx_tree_merge_probs(vp9_intra_mode_tree, pre_fc->y_mode_prob[i], + counts->y_mode[i], fc->y_mode_prob[i]); + + for (i = 0; i < INTRA_MODES; ++i) + vpx_tree_merge_probs(vp9_intra_mode_tree, pre_fc->uv_mode_prob[i], + counts->uv_mode[i], fc->uv_mode_prob[i]); + + for (i = 0; i < PARTITION_CONTEXTS; i++) + vpx_tree_merge_probs(vp9_partition_tree, pre_fc->partition_prob[i], + counts->partition[i], fc->partition_prob[i]); + + if (cm->interp_filter == SWITCHABLE) { + for (i = 0; i < SWITCHABLE_FILTER_CONTEXTS; i++) + vpx_tree_merge_probs(vp9_switchable_interp_tree, + pre_fc->switchable_interp_prob[i], + counts->switchable_interp[i], + fc->switchable_interp_prob[i]); + } + + if (cm->tx_mode == TX_MODE_SELECT) { + int j; + unsigned int branch_ct_8x8p[TX_SIZES - 3][2]; + unsigned int branch_ct_16x16p[TX_SIZES - 2][2]; + unsigned int branch_ct_32x32p[TX_SIZES - 1][2]; + + for (i = 0; i < TX_SIZE_CONTEXTS; ++i) { + tx_counts_to_branch_counts_8x8(counts->tx.p8x8[i], branch_ct_8x8p); + for (j = 0; j < TX_SIZES - 3; ++j) + fc->tx_probs.p8x8[i][j] = mode_mv_merge_probs( + pre_fc->tx_probs.p8x8[i][j], branch_ct_8x8p[j]); + + tx_counts_to_branch_counts_16x16(counts->tx.p16x16[i], branch_ct_16x16p); + for (j = 0; j < TX_SIZES - 2; ++j) + fc->tx_probs.p16x16[i][j] = mode_mv_merge_probs( + pre_fc->tx_probs.p16x16[i][j], branch_ct_16x16p[j]); + + tx_counts_to_branch_counts_32x32(counts->tx.p32x32[i], branch_ct_32x32p); + for (j = 0; j < TX_SIZES - 1; ++j) + fc->tx_probs.p32x32[i][j] = mode_mv_merge_probs( + pre_fc->tx_probs.p32x32[i][j], branch_ct_32x32p[j]); + } + } + + for (i = 0; i < SKIP_CONTEXTS; ++i) + fc->skip_probs[i] = mode_mv_merge_probs( + pre_fc->skip_probs[i], counts->skip[i]); +} + +static void set_default_lf_deltas(struct loopfilter *lf) { + lf->mode_ref_delta_enabled = 1; + lf->mode_ref_delta_update = 1; + + lf->ref_deltas[INTRA_FRAME] = 1; + lf->ref_deltas[LAST_FRAME] = 0; + lf->ref_deltas[GOLDEN_FRAME] = -1; + lf->ref_deltas[ALTREF_FRAME] = -1; + + lf->mode_deltas[0] = 0; + lf->mode_deltas[1] = 0; +} + +void vp9_setup_past_independence(VP9_COMMON *cm) { + // Reset the segment feature data to the default stats: + // Features disabled, 0, with delta coding (Default state). + struct loopfilter *const lf = &cm->lf; + + int i; + vp9_clearall_segfeatures(&cm->seg); + cm->seg.abs_delta = SEGMENT_DELTADATA; + + if (cm->last_frame_seg_map && !cm->frame_parallel_decode) + memset(cm->last_frame_seg_map, 0, (cm->mi_rows * cm->mi_cols)); + + if (cm->current_frame_seg_map) + memset(cm->current_frame_seg_map, 0, (cm->mi_rows * cm->mi_cols)); + + // Reset the mode ref deltas for loop filter + vp9_zero(lf->last_ref_deltas); + vp9_zero(lf->last_mode_deltas); + set_default_lf_deltas(lf); + + // To force update of the sharpness + lf->last_sharpness_level = -1; + + vp9_default_coef_probs(cm); + init_mode_probs(cm->fc); + vp9_init_mv_probs(cm); + cm->fc->initialized = 1; + + if (cm->frame_type == KEY_FRAME || + cm->error_resilient_mode || cm->reset_frame_context == 3) { + // Reset all frame contexts. + for (i = 0; i < FRAME_CONTEXTS; ++i) + cm->frame_contexts[i] = *cm->fc; + } else if (cm->reset_frame_context == 2) { + // Reset only the frame context specified in the frame header. + cm->frame_contexts[cm->frame_context_idx] = *cm->fc; + } + + // prev_mip will only be allocated in encoder. + if (frame_is_intra_only(cm) && cm->prev_mip && !cm->frame_parallel_decode) + memset(cm->prev_mip, 0, + cm->mi_stride * (cm->mi_rows + 1) * sizeof(*cm->prev_mip)); + + vp9_zero(cm->ref_frame_sign_bias); + + cm->frame_context_idx = 0; +} diff --git a/thirdparty/libvpx/vp9/common/vp9_entropymode.h b/thirdparty/libvpx/vp9/common/vp9_entropymode.h new file mode 100644 index 0000000000..0285be1557 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_entropymode.h @@ -0,0 +1,107 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_COMMON_VP9_ENTROPYMODE_H_ +#define VP9_COMMON_VP9_ENTROPYMODE_H_ + +#include "vp9/common/vp9_entropy.h" +#include "vp9/common/vp9_entropymv.h" +#include "vp9/common/vp9_filter.h" +#include "vpx_dsp/vpx_filter.h" + +#ifdef __cplusplus +extern "C" { +#endif + +#define BLOCK_SIZE_GROUPS 4 + +#define TX_SIZE_CONTEXTS 2 + +#define INTER_OFFSET(mode) ((mode) - NEARESTMV) + +struct VP9Common; + +struct tx_probs { + vpx_prob p32x32[TX_SIZE_CONTEXTS][TX_SIZES - 1]; + vpx_prob p16x16[TX_SIZE_CONTEXTS][TX_SIZES - 2]; + vpx_prob p8x8[TX_SIZE_CONTEXTS][TX_SIZES - 3]; +}; + +struct tx_counts { + unsigned int p32x32[TX_SIZE_CONTEXTS][TX_SIZES]; + unsigned int p16x16[TX_SIZE_CONTEXTS][TX_SIZES - 1]; + unsigned int p8x8[TX_SIZE_CONTEXTS][TX_SIZES - 2]; + unsigned int tx_totals[TX_SIZES]; +}; + +typedef struct frame_contexts { + vpx_prob y_mode_prob[BLOCK_SIZE_GROUPS][INTRA_MODES - 1]; + vpx_prob uv_mode_prob[INTRA_MODES][INTRA_MODES - 1]; + vpx_prob partition_prob[PARTITION_CONTEXTS][PARTITION_TYPES - 1]; + vp9_coeff_probs_model coef_probs[TX_SIZES][PLANE_TYPES]; + vpx_prob switchable_interp_prob[SWITCHABLE_FILTER_CONTEXTS] + [SWITCHABLE_FILTERS - 1]; + vpx_prob inter_mode_probs[INTER_MODE_CONTEXTS][INTER_MODES - 1]; + vpx_prob intra_inter_prob[INTRA_INTER_CONTEXTS]; + vpx_prob comp_inter_prob[COMP_INTER_CONTEXTS]; + vpx_prob single_ref_prob[REF_CONTEXTS][2]; + vpx_prob comp_ref_prob[REF_CONTEXTS]; + struct tx_probs tx_probs; + vpx_prob skip_probs[SKIP_CONTEXTS]; + nmv_context nmvc; + int initialized; +} FRAME_CONTEXT; + +typedef struct FRAME_COUNTS { + unsigned int y_mode[BLOCK_SIZE_GROUPS][INTRA_MODES]; + unsigned int uv_mode[INTRA_MODES][INTRA_MODES]; + unsigned int partition[PARTITION_CONTEXTS][PARTITION_TYPES]; + vp9_coeff_count_model coef[TX_SIZES][PLANE_TYPES]; + unsigned int eob_branch[TX_SIZES][PLANE_TYPES][REF_TYPES] + [COEF_BANDS][COEFF_CONTEXTS]; + unsigned int switchable_interp[SWITCHABLE_FILTER_CONTEXTS] + [SWITCHABLE_FILTERS]; + unsigned int inter_mode[INTER_MODE_CONTEXTS][INTER_MODES]; + unsigned int intra_inter[INTRA_INTER_CONTEXTS][2]; + unsigned int comp_inter[COMP_INTER_CONTEXTS][2]; + unsigned int single_ref[REF_CONTEXTS][2][2]; + unsigned int comp_ref[REF_CONTEXTS][2]; + struct tx_counts tx; + unsigned int skip[SKIP_CONTEXTS][2]; + nmv_context_counts mv; +} FRAME_COUNTS; + +extern const vpx_prob vp9_kf_uv_mode_prob[INTRA_MODES][INTRA_MODES - 1]; +extern const vpx_prob vp9_kf_y_mode_prob[INTRA_MODES][INTRA_MODES] + [INTRA_MODES - 1]; +extern const vpx_prob vp9_kf_partition_probs[PARTITION_CONTEXTS] + [PARTITION_TYPES - 1]; +extern const vpx_tree_index vp9_intra_mode_tree[TREE_SIZE(INTRA_MODES)]; +extern const vpx_tree_index vp9_inter_mode_tree[TREE_SIZE(INTER_MODES)]; +extern const vpx_tree_index vp9_partition_tree[TREE_SIZE(PARTITION_TYPES)]; +extern const vpx_tree_index vp9_switchable_interp_tree + [TREE_SIZE(SWITCHABLE_FILTERS)]; + +void vp9_setup_past_independence(struct VP9Common *cm); + +void vp9_adapt_mode_probs(struct VP9Common *cm); + +void tx_counts_to_branch_counts_32x32(const unsigned int *tx_count_32x32p, + unsigned int (*ct_32x32p)[2]); +void tx_counts_to_branch_counts_16x16(const unsigned int *tx_count_16x16p, + unsigned int (*ct_16x16p)[2]); +void tx_counts_to_branch_counts_8x8(const unsigned int *tx_count_8x8p, + unsigned int (*ct_8x8p)[2]); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_ENTROPYMODE_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_entropymv.c b/thirdparty/libvpx/vp9/common/vp9_entropymv.c new file mode 100644 index 0000000000..566ae91cf7 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_entropymv.c @@ -0,0 +1,210 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "vp9/common/vp9_onyxc_int.h" +#include "vp9/common/vp9_entropymv.h" + +const vpx_tree_index vp9_mv_joint_tree[TREE_SIZE(MV_JOINTS)] = { + -MV_JOINT_ZERO, 2, + -MV_JOINT_HNZVZ, 4, + -MV_JOINT_HZVNZ, -MV_JOINT_HNZVNZ +}; + +const vpx_tree_index vp9_mv_class_tree[TREE_SIZE(MV_CLASSES)] = { + -MV_CLASS_0, 2, + -MV_CLASS_1, 4, + 6, 8, + -MV_CLASS_2, -MV_CLASS_3, + 10, 12, + -MV_CLASS_4, -MV_CLASS_5, + -MV_CLASS_6, 14, + 16, 18, + -MV_CLASS_7, -MV_CLASS_8, + -MV_CLASS_9, -MV_CLASS_10, +}; + +const vpx_tree_index vp9_mv_class0_tree[TREE_SIZE(CLASS0_SIZE)] = { + -0, -1, +}; + +const vpx_tree_index vp9_mv_fp_tree[TREE_SIZE(MV_FP_SIZE)] = { + -0, 2, + -1, 4, + -2, -3 +}; + +static const nmv_context default_nmv_context = { + {32, 64, 96}, + { + { // Vertical component + 128, // sign + {224, 144, 192, 168, 192, 176, 192, 198, 198, 245}, // class + {216}, // class0 + {136, 140, 148, 160, 176, 192, 224, 234, 234, 240}, // bits + {{128, 128, 64}, {96, 112, 64}}, // class0_fp + {64, 96, 64}, // fp + 160, // class0_hp bit + 128, // hp + }, + { // Horizontal component + 128, // sign + {216, 128, 176, 160, 176, 176, 192, 198, 198, 208}, // class + {208}, // class0 + {136, 140, 148, 160, 176, 192, 224, 234, 234, 240}, // bits + {{128, 128, 64}, {96, 112, 64}}, // class0_fp + {64, 96, 64}, // fp + 160, // class0_hp bit + 128, // hp + } + }, +}; + +static const uint8_t log_in_base_2[] = { + 0, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, + 4, 4, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 6, 6, + 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, + 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, + 6, 6, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, + 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, + 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, + 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, + 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, + 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 8, 8, + 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, + 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, + 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, + 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, + 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, + 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, + 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, + 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, + 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, + 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, + 8, 8, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 10 +}; + +static INLINE int mv_class_base(MV_CLASS_TYPE c) { + return c ? CLASS0_SIZE << (c + 2) : 0; +} + +MV_CLASS_TYPE vp9_get_mv_class(int z, int *offset) { + const MV_CLASS_TYPE c = (z >= CLASS0_SIZE * 4096) ? + MV_CLASS_10 : (MV_CLASS_TYPE)log_in_base_2[z >> 3]; + if (offset) + *offset = z - mv_class_base(c); + return c; +} + +static void inc_mv_component(int v, nmv_component_counts *comp_counts, + int incr, int usehp) { + int s, z, c, o, d, e, f; + assert(v != 0); /* should not be zero */ + s = v < 0; + comp_counts->sign[s] += incr; + z = (s ? -v : v) - 1; /* magnitude - 1 */ + + c = vp9_get_mv_class(z, &o); + comp_counts->classes[c] += incr; + + d = (o >> 3); /* int mv data */ + f = (o >> 1) & 3; /* fractional pel mv data */ + e = (o & 1); /* high precision mv data */ + + if (c == MV_CLASS_0) { + comp_counts->class0[d] += incr; + comp_counts->class0_fp[d][f] += incr; + comp_counts->class0_hp[e] += usehp * incr; + } else { + int i; + int b = c + CLASS0_BITS - 1; // number of bits + for (i = 0; i < b; ++i) + comp_counts->bits[i][((d >> i) & 1)] += incr; + comp_counts->fp[f] += incr; + comp_counts->hp[e] += usehp * incr; + } +} + +void vp9_inc_mv(const MV *mv, nmv_context_counts *counts) { + if (counts != NULL) { + const MV_JOINT_TYPE j = vp9_get_mv_joint(mv); + ++counts->joints[j]; + + if (mv_joint_vertical(j)) { + inc_mv_component(mv->row, &counts->comps[0], 1, 1); + } + + if (mv_joint_horizontal(j)) { + inc_mv_component(mv->col, &counts->comps[1], 1, 1); + } + } +} + +void vp9_adapt_mv_probs(VP9_COMMON *cm, int allow_hp) { + int i, j; + + nmv_context *fc = &cm->fc->nmvc; + const nmv_context *pre_fc = &cm->frame_contexts[cm->frame_context_idx].nmvc; + const nmv_context_counts *counts = &cm->counts.mv; + + vpx_tree_merge_probs(vp9_mv_joint_tree, pre_fc->joints, counts->joints, + fc->joints); + + for (i = 0; i < 2; ++i) { + nmv_component *comp = &fc->comps[i]; + const nmv_component *pre_comp = &pre_fc->comps[i]; + const nmv_component_counts *c = &counts->comps[i]; + + comp->sign = mode_mv_merge_probs(pre_comp->sign, c->sign); + vpx_tree_merge_probs(vp9_mv_class_tree, pre_comp->classes, c->classes, + comp->classes); + vpx_tree_merge_probs(vp9_mv_class0_tree, pre_comp->class0, c->class0, + comp->class0); + + for (j = 0; j < MV_OFFSET_BITS; ++j) + comp->bits[j] = mode_mv_merge_probs(pre_comp->bits[j], c->bits[j]); + + for (j = 0; j < CLASS0_SIZE; ++j) + vpx_tree_merge_probs(vp9_mv_fp_tree, pre_comp->class0_fp[j], + c->class0_fp[j], comp->class0_fp[j]); + + vpx_tree_merge_probs(vp9_mv_fp_tree, pre_comp->fp, c->fp, comp->fp); + + if (allow_hp) { + comp->class0_hp = mode_mv_merge_probs(pre_comp->class0_hp, c->class0_hp); + comp->hp = mode_mv_merge_probs(pre_comp->hp, c->hp); + } + } +} + +void vp9_init_mv_probs(VP9_COMMON *cm) { + cm->fc->nmvc = default_nmv_context; +} diff --git a/thirdparty/libvpx/vp9/common/vp9_entropymv.h b/thirdparty/libvpx/vp9/common/vp9_entropymv.h new file mode 100644 index 0000000000..2f05ad44b6 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_entropymv.h @@ -0,0 +1,140 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP9_COMMON_VP9_ENTROPYMV_H_ +#define VP9_COMMON_VP9_ENTROPYMV_H_ + +#include "./vpx_config.h" + +#include "vpx_dsp/prob.h" + +#include "vp9/common/vp9_mv.h" + +#ifdef __cplusplus +extern "C" { +#endif + +struct VP9Common; + +void vp9_init_mv_probs(struct VP9Common *cm); + +void vp9_adapt_mv_probs(struct VP9Common *cm, int usehp); + +// Integer pel reference mv threshold for use of high-precision 1/8 mv +#define COMPANDED_MVREF_THRESH 8 + +static INLINE int use_mv_hp(const MV *ref) { + return (abs(ref->row) >> 3) < COMPANDED_MVREF_THRESH && + (abs(ref->col) >> 3) < COMPANDED_MVREF_THRESH; +} + +#define MV_UPDATE_PROB 252 + +/* Symbols for coding which components are zero jointly */ +#define MV_JOINTS 4 +typedef enum { + MV_JOINT_ZERO = 0, /* Zero vector */ + MV_JOINT_HNZVZ = 1, /* Vert zero, hor nonzero */ + MV_JOINT_HZVNZ = 2, /* Hor zero, vert nonzero */ + MV_JOINT_HNZVNZ = 3, /* Both components nonzero */ +} MV_JOINT_TYPE; + +static INLINE int mv_joint_vertical(MV_JOINT_TYPE type) { + return type == MV_JOINT_HZVNZ || type == MV_JOINT_HNZVNZ; +} + +static INLINE int mv_joint_horizontal(MV_JOINT_TYPE type) { + return type == MV_JOINT_HNZVZ || type == MV_JOINT_HNZVNZ; +} + +/* Symbols for coding magnitude class of nonzero components */ +#define MV_CLASSES 11 +typedef enum { + MV_CLASS_0 = 0, /* (0, 2] integer pel */ + MV_CLASS_1 = 1, /* (2, 4] integer pel */ + MV_CLASS_2 = 2, /* (4, 8] integer pel */ + MV_CLASS_3 = 3, /* (8, 16] integer pel */ + MV_CLASS_4 = 4, /* (16, 32] integer pel */ + MV_CLASS_5 = 5, /* (32, 64] integer pel */ + MV_CLASS_6 = 6, /* (64, 128] integer pel */ + MV_CLASS_7 = 7, /* (128, 256] integer pel */ + MV_CLASS_8 = 8, /* (256, 512] integer pel */ + MV_CLASS_9 = 9, /* (512, 1024] integer pel */ + MV_CLASS_10 = 10, /* (1024,2048] integer pel */ +} MV_CLASS_TYPE; + +#define CLASS0_BITS 1 /* bits at integer precision for class 0 */ +#define CLASS0_SIZE (1 << CLASS0_BITS) +#define MV_OFFSET_BITS (MV_CLASSES + CLASS0_BITS - 2) +#define MV_FP_SIZE 4 + +#define MV_MAX_BITS (MV_CLASSES + CLASS0_BITS + 2) +#define MV_MAX ((1 << MV_MAX_BITS) - 1) +#define MV_VALS ((MV_MAX << 1) + 1) + +#define MV_IN_USE_BITS 14 +#define MV_UPP ((1 << MV_IN_USE_BITS) - 1) +#define MV_LOW (-(1 << MV_IN_USE_BITS)) + +extern const vpx_tree_index vp9_mv_joint_tree[]; +extern const vpx_tree_index vp9_mv_class_tree[]; +extern const vpx_tree_index vp9_mv_class0_tree[]; +extern const vpx_tree_index vp9_mv_fp_tree[]; + +typedef struct { + vpx_prob sign; + vpx_prob classes[MV_CLASSES - 1]; + vpx_prob class0[CLASS0_SIZE - 1]; + vpx_prob bits[MV_OFFSET_BITS]; + vpx_prob class0_fp[CLASS0_SIZE][MV_FP_SIZE - 1]; + vpx_prob fp[MV_FP_SIZE - 1]; + vpx_prob class0_hp; + vpx_prob hp; +} nmv_component; + +typedef struct { + vpx_prob joints[MV_JOINTS - 1]; + nmv_component comps[2]; +} nmv_context; + +static INLINE MV_JOINT_TYPE vp9_get_mv_joint(const MV *mv) { + if (mv->row == 0) { + return mv->col == 0 ? MV_JOINT_ZERO : MV_JOINT_HNZVZ; + } else { + return mv->col == 0 ? MV_JOINT_HZVNZ : MV_JOINT_HNZVNZ; + } +} + +MV_CLASS_TYPE vp9_get_mv_class(int z, int *offset); + +typedef struct { + unsigned int sign[2]; + unsigned int classes[MV_CLASSES]; + unsigned int class0[CLASS0_SIZE]; + unsigned int bits[MV_OFFSET_BITS][2]; + unsigned int class0_fp[CLASS0_SIZE][MV_FP_SIZE]; + unsigned int fp[MV_FP_SIZE]; + unsigned int class0_hp[2]; + unsigned int hp[2]; +} nmv_component_counts; + +typedef struct { + unsigned int joints[MV_JOINTS]; + nmv_component_counts comps[2]; +} nmv_context_counts; + +void vp9_inc_mv(const MV *mv, nmv_context_counts *mvctx); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_ENTROPYMV_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_enums.h b/thirdparty/libvpx/vp9/common/vp9_enums.h new file mode 100644 index 0000000000..d089f23f97 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_enums.h @@ -0,0 +1,147 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_COMMON_VP9_ENUMS_H_ +#define VP9_COMMON_VP9_ENUMS_H_ + +#include "./vpx_config.h" +#include "vpx/vpx_integer.h" + +#ifdef __cplusplus +extern "C" { +#endif + +#define MI_SIZE_LOG2 3 +#define MI_BLOCK_SIZE_LOG2 (6 - MI_SIZE_LOG2) // 64 = 2^6 + +#define MI_SIZE (1 << MI_SIZE_LOG2) // pixels per mi-unit +#define MI_BLOCK_SIZE (1 << MI_BLOCK_SIZE_LOG2) // mi-units per max block + +#define MI_MASK (MI_BLOCK_SIZE - 1) + +// Bitstream profiles indicated by 2-3 bits in the uncompressed header. +// 00: Profile 0. 8-bit 4:2:0 only. +// 10: Profile 1. 8-bit 4:4:4, 4:2:2, and 4:4:0. +// 01: Profile 2. 10-bit and 12-bit color only, with 4:2:0 sampling. +// 110: Profile 3. 10-bit and 12-bit color only, with 4:2:2/4:4:4/4:4:0 +// sampling. +// 111: Undefined profile. +typedef enum BITSTREAM_PROFILE { + PROFILE_0, + PROFILE_1, + PROFILE_2, + PROFILE_3, + MAX_PROFILES +} BITSTREAM_PROFILE; + +#define BLOCK_4X4 0 +#define BLOCK_4X8 1 +#define BLOCK_8X4 2 +#define BLOCK_8X8 3 +#define BLOCK_8X16 4 +#define BLOCK_16X8 5 +#define BLOCK_16X16 6 +#define BLOCK_16X32 7 +#define BLOCK_32X16 8 +#define BLOCK_32X32 9 +#define BLOCK_32X64 10 +#define BLOCK_64X32 11 +#define BLOCK_64X64 12 +#define BLOCK_SIZES 13 +#define BLOCK_INVALID BLOCK_SIZES +typedef uint8_t BLOCK_SIZE; + +typedef enum PARTITION_TYPE { + PARTITION_NONE, + PARTITION_HORZ, + PARTITION_VERT, + PARTITION_SPLIT, + PARTITION_TYPES, + PARTITION_INVALID = PARTITION_TYPES +} PARTITION_TYPE; + +typedef char PARTITION_CONTEXT; +#define PARTITION_PLOFFSET 4 // number of probability models per block size +#define PARTITION_CONTEXTS (4 * PARTITION_PLOFFSET) + +// block transform size +typedef uint8_t TX_SIZE; +#define TX_4X4 ((TX_SIZE)0) // 4x4 transform +#define TX_8X8 ((TX_SIZE)1) // 8x8 transform +#define TX_16X16 ((TX_SIZE)2) // 16x16 transform +#define TX_32X32 ((TX_SIZE)3) // 32x32 transform +#define TX_SIZES ((TX_SIZE)4) + +// frame transform mode +typedef enum { + ONLY_4X4 = 0, // only 4x4 transform used + ALLOW_8X8 = 1, // allow block transform size up to 8x8 + ALLOW_16X16 = 2, // allow block transform size up to 16x16 + ALLOW_32X32 = 3, // allow block transform size up to 32x32 + TX_MODE_SELECT = 4, // transform specified for each block + TX_MODES = 5, +} TX_MODE; + +typedef enum { + DCT_DCT = 0, // DCT in both horizontal and vertical + ADST_DCT = 1, // ADST in vertical, DCT in horizontal + DCT_ADST = 2, // DCT in vertical, ADST in horizontal + ADST_ADST = 3, // ADST in both directions + TX_TYPES = 4 +} TX_TYPE; + +typedef enum { + VP9_LAST_FLAG = 1 << 0, + VP9_GOLD_FLAG = 1 << 1, + VP9_ALT_FLAG = 1 << 2, +} VP9_REFFRAME; + +typedef enum { + PLANE_TYPE_Y = 0, + PLANE_TYPE_UV = 1, + PLANE_TYPES +} PLANE_TYPE; + +#define DC_PRED 0 // Average of above and left pixels +#define V_PRED 1 // Vertical +#define H_PRED 2 // Horizontal +#define D45_PRED 3 // Directional 45 deg = round(arctan(1/1) * 180/pi) +#define D135_PRED 4 // Directional 135 deg = 180 - 45 +#define D117_PRED 5 // Directional 117 deg = 180 - 63 +#define D153_PRED 6 // Directional 153 deg = 180 - 27 +#define D207_PRED 7 // Directional 207 deg = 180 + 27 +#define D63_PRED 8 // Directional 63 deg = round(arctan(2/1) * 180/pi) +#define TM_PRED 9 // True-motion +#define NEARESTMV 10 +#define NEARMV 11 +#define ZEROMV 12 +#define NEWMV 13 +#define MB_MODE_COUNT 14 +typedef uint8_t PREDICTION_MODE; + +#define INTRA_MODES (TM_PRED + 1) + +#define INTER_MODES (1 + NEWMV - NEARESTMV) + +#define SKIP_CONTEXTS 3 +#define INTER_MODE_CONTEXTS 7 + +/* Segment Feature Masks */ +#define MAX_MV_REF_CANDIDATES 2 + +#define INTRA_INTER_CONTEXTS 4 +#define COMP_INTER_CONTEXTS 5 +#define REF_CONTEXTS 5 + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_ENUMS_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_filter.c b/thirdparty/libvpx/vp9/common/vp9_filter.c new file mode 100644 index 0000000000..4b2198fc40 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_filter.c @@ -0,0 +1,104 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <assert.h> + +#include "vp9/common/vp9_filter.h" + +DECLARE_ALIGNED(256, static const InterpKernel, + bilinear_filters[SUBPEL_SHIFTS]) = { + { 0, 0, 0, 128, 0, 0, 0, 0 }, + { 0, 0, 0, 120, 8, 0, 0, 0 }, + { 0, 0, 0, 112, 16, 0, 0, 0 }, + { 0, 0, 0, 104, 24, 0, 0, 0 }, + { 0, 0, 0, 96, 32, 0, 0, 0 }, + { 0, 0, 0, 88, 40, 0, 0, 0 }, + { 0, 0, 0, 80, 48, 0, 0, 0 }, + { 0, 0, 0, 72, 56, 0, 0, 0 }, + { 0, 0, 0, 64, 64, 0, 0, 0 }, + { 0, 0, 0, 56, 72, 0, 0, 0 }, + { 0, 0, 0, 48, 80, 0, 0, 0 }, + { 0, 0, 0, 40, 88, 0, 0, 0 }, + { 0, 0, 0, 32, 96, 0, 0, 0 }, + { 0, 0, 0, 24, 104, 0, 0, 0 }, + { 0, 0, 0, 16, 112, 0, 0, 0 }, + { 0, 0, 0, 8, 120, 0, 0, 0 } +}; + +// Lagrangian interpolation filter +DECLARE_ALIGNED(256, static const InterpKernel, + sub_pel_filters_8[SUBPEL_SHIFTS]) = { + { 0, 0, 0, 128, 0, 0, 0, 0}, + { 0, 1, -5, 126, 8, -3, 1, 0}, + { -1, 3, -10, 122, 18, -6, 2, 0}, + { -1, 4, -13, 118, 27, -9, 3, -1}, + { -1, 4, -16, 112, 37, -11, 4, -1}, + { -1, 5, -18, 105, 48, -14, 4, -1}, + { -1, 5, -19, 97, 58, -16, 5, -1}, + { -1, 6, -19, 88, 68, -18, 5, -1}, + { -1, 6, -19, 78, 78, -19, 6, -1}, + { -1, 5, -18, 68, 88, -19, 6, -1}, + { -1, 5, -16, 58, 97, -19, 5, -1}, + { -1, 4, -14, 48, 105, -18, 5, -1}, + { -1, 4, -11, 37, 112, -16, 4, -1}, + { -1, 3, -9, 27, 118, -13, 4, -1}, + { 0, 2, -6, 18, 122, -10, 3, -1}, + { 0, 1, -3, 8, 126, -5, 1, 0} +}; + +// DCT based filter +DECLARE_ALIGNED(256, static const InterpKernel, + sub_pel_filters_8s[SUBPEL_SHIFTS]) = { + {0, 0, 0, 128, 0, 0, 0, 0}, + {-1, 3, -7, 127, 8, -3, 1, 0}, + {-2, 5, -13, 125, 17, -6, 3, -1}, + {-3, 7, -17, 121, 27, -10, 5, -2}, + {-4, 9, -20, 115, 37, -13, 6, -2}, + {-4, 10, -23, 108, 48, -16, 8, -3}, + {-4, 10, -24, 100, 59, -19, 9, -3}, + {-4, 11, -24, 90, 70, -21, 10, -4}, + {-4, 11, -23, 80, 80, -23, 11, -4}, + {-4, 10, -21, 70, 90, -24, 11, -4}, + {-3, 9, -19, 59, 100, -24, 10, -4}, + {-3, 8, -16, 48, 108, -23, 10, -4}, + {-2, 6, -13, 37, 115, -20, 9, -4}, + {-2, 5, -10, 27, 121, -17, 7, -3}, + {-1, 3, -6, 17, 125, -13, 5, -2}, + {0, 1, -3, 8, 127, -7, 3, -1} +}; + +// freqmultiplier = 0.5 +DECLARE_ALIGNED(256, static const InterpKernel, + sub_pel_filters_8lp[SUBPEL_SHIFTS]) = { + { 0, 0, 0, 128, 0, 0, 0, 0}, + {-3, -1, 32, 64, 38, 1, -3, 0}, + {-2, -2, 29, 63, 41, 2, -3, 0}, + {-2, -2, 26, 63, 43, 4, -4, 0}, + {-2, -3, 24, 62, 46, 5, -4, 0}, + {-2, -3, 21, 60, 49, 7, -4, 0}, + {-1, -4, 18, 59, 51, 9, -4, 0}, + {-1, -4, 16, 57, 53, 12, -4, -1}, + {-1, -4, 14, 55, 55, 14, -4, -1}, + {-1, -4, 12, 53, 57, 16, -4, -1}, + { 0, -4, 9, 51, 59, 18, -4, -1}, + { 0, -4, 7, 49, 60, 21, -3, -2}, + { 0, -4, 5, 46, 62, 24, -3, -2}, + { 0, -4, 4, 43, 63, 26, -2, -2}, + { 0, -3, 2, 41, 63, 29, -2, -2}, + { 0, -3, 1, 38, 64, 32, -1, -3} +}; + + +const InterpKernel *vp9_filter_kernels[4] = { + sub_pel_filters_8, + sub_pel_filters_8lp, + sub_pel_filters_8s, + bilinear_filters +}; diff --git a/thirdparty/libvpx/vp9/common/vp9_filter.h b/thirdparty/libvpx/vp9/common/vp9_filter.h new file mode 100644 index 0000000000..efa24bc67b --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_filter.h @@ -0,0 +1,42 @@ +/* + * Copyright (c) 2011 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_COMMON_VP9_FILTER_H_ +#define VP9_COMMON_VP9_FILTER_H_ + +#include "./vpx_config.h" +#include "vpx/vpx_integer.h" +#include "vpx_dsp/vpx_filter.h" +#include "vpx_ports/mem.h" + + +#ifdef __cplusplus +extern "C" { +#endif + +#define EIGHTTAP 0 +#define EIGHTTAP_SMOOTH 1 +#define EIGHTTAP_SHARP 2 +#define SWITCHABLE_FILTERS 3 /* Number of switchable filters */ +#define BILINEAR 3 +// The codec can operate in four possible inter prediction filter mode: +// 8-tap, 8-tap-smooth, 8-tap-sharp, and switching between the three. +#define SWITCHABLE_FILTER_CONTEXTS (SWITCHABLE_FILTERS + 1) +#define SWITCHABLE 4 /* should be the last one */ + +typedef uint8_t INTERP_FILTER; + +extern const InterpKernel *vp9_filter_kernels[4]; + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_FILTER_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_frame_buffers.c b/thirdparty/libvpx/vp9/common/vp9_frame_buffers.c new file mode 100644 index 0000000000..0f41d66985 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_frame_buffers.c @@ -0,0 +1,86 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <assert.h> + +#include "vp9/common/vp9_frame_buffers.h" +#include "vpx_mem/vpx_mem.h" + +int vp9_alloc_internal_frame_buffers(InternalFrameBufferList *list) { + assert(list != NULL); + vp9_free_internal_frame_buffers(list); + + list->num_internal_frame_buffers = + VP9_MAXIMUM_REF_BUFFERS + VPX_MAXIMUM_WORK_BUFFERS; + list->int_fb = + (InternalFrameBuffer *)vpx_calloc(list->num_internal_frame_buffers, + sizeof(*list->int_fb)); + return (list->int_fb == NULL); +} + +void vp9_free_internal_frame_buffers(InternalFrameBufferList *list) { + int i; + + assert(list != NULL); + + for (i = 0; i < list->num_internal_frame_buffers; ++i) { + vpx_free(list->int_fb[i].data); + list->int_fb[i].data = NULL; + } + vpx_free(list->int_fb); + list->int_fb = NULL; +} + +int vp9_get_frame_buffer(void *cb_priv, size_t min_size, + vpx_codec_frame_buffer_t *fb) { + int i; + InternalFrameBufferList *const int_fb_list = + (InternalFrameBufferList *)cb_priv; + if (int_fb_list == NULL) + return -1; + + // Find a free frame buffer. + for (i = 0; i < int_fb_list->num_internal_frame_buffers; ++i) { + if (!int_fb_list->int_fb[i].in_use) + break; + } + + if (i == int_fb_list->num_internal_frame_buffers) + return -1; + + if (int_fb_list->int_fb[i].size < min_size) { + int_fb_list->int_fb[i].data = + (uint8_t *)vpx_realloc(int_fb_list->int_fb[i].data, min_size); + if (!int_fb_list->int_fb[i].data) + return -1; + + // This memset is needed for fixing valgrind error from C loop filter + // due to access uninitialized memory in frame border. It could be + // removed if border is totally removed. + memset(int_fb_list->int_fb[i].data, 0, min_size); + int_fb_list->int_fb[i].size = min_size; + } + + fb->data = int_fb_list->int_fb[i].data; + fb->size = int_fb_list->int_fb[i].size; + int_fb_list->int_fb[i].in_use = 1; + + // Set the frame buffer's private data to point at the internal frame buffer. + fb->priv = &int_fb_list->int_fb[i]; + return 0; +} + +int vp9_release_frame_buffer(void *cb_priv, vpx_codec_frame_buffer_t *fb) { + InternalFrameBuffer *const int_fb = (InternalFrameBuffer *)fb->priv; + (void)cb_priv; + if (int_fb) + int_fb->in_use = 0; + return 0; +} diff --git a/thirdparty/libvpx/vp9/common/vp9_frame_buffers.h b/thirdparty/libvpx/vp9/common/vp9_frame_buffers.h new file mode 100644 index 0000000000..e2cfe61b66 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_frame_buffers.h @@ -0,0 +1,53 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_COMMON_VP9_FRAME_BUFFERS_H_ +#define VP9_COMMON_VP9_FRAME_BUFFERS_H_ + +#include "vpx/vpx_frame_buffer.h" +#include "vpx/vpx_integer.h" + +#ifdef __cplusplus +extern "C" { +#endif + +typedef struct InternalFrameBuffer { + uint8_t *data; + size_t size; + int in_use; +} InternalFrameBuffer; + +typedef struct InternalFrameBufferList { + int num_internal_frame_buffers; + InternalFrameBuffer *int_fb; +} InternalFrameBufferList; + +// Initializes |list|. Returns 0 on success. +int vp9_alloc_internal_frame_buffers(InternalFrameBufferList *list); + +// Free any data allocated to the frame buffers. +void vp9_free_internal_frame_buffers(InternalFrameBufferList *list); + +// Callback used by libvpx to request an external frame buffer. |cb_priv| +// Callback private data, which points to an InternalFrameBufferList. +// |min_size| is the minimum size in bytes needed to decode the next frame. +// |fb| pointer to the frame buffer. +int vp9_get_frame_buffer(void *cb_priv, size_t min_size, + vpx_codec_frame_buffer_t *fb); + +// Callback used by libvpx when there are no references to the frame buffer. +// |cb_priv| is not used. |fb| pointer to the frame buffer. +int vp9_release_frame_buffer(void *cb_priv, vpx_codec_frame_buffer_t *fb); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_FRAME_BUFFERS_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_idct.c b/thirdparty/libvpx/vp9/common/vp9_idct.c new file mode 100644 index 0000000000..1b420143bb --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_idct.c @@ -0,0 +1,405 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <math.h> + +#include "./vp9_rtcd.h" +#include "./vpx_dsp_rtcd.h" +#include "vp9/common/vp9_blockd.h" +#include "vp9/common/vp9_idct.h" +#include "vpx_dsp/inv_txfm.h" +#include "vpx_ports/mem.h" + +void vp9_iht4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int stride, + int tx_type) { + const transform_2d IHT_4[] = { + { idct4_c, idct4_c }, // DCT_DCT = 0 + { iadst4_c, idct4_c }, // ADST_DCT = 1 + { idct4_c, iadst4_c }, // DCT_ADST = 2 + { iadst4_c, iadst4_c } // ADST_ADST = 3 + }; + + int i, j; + tran_low_t out[4 * 4]; + tran_low_t *outptr = out; + tran_low_t temp_in[4], temp_out[4]; + + // inverse transform row vectors + for (i = 0; i < 4; ++i) { + IHT_4[tx_type].rows(input, outptr); + input += 4; + outptr += 4; + } + + // inverse transform column vectors + for (i = 0; i < 4; ++i) { + for (j = 0; j < 4; ++j) + temp_in[j] = out[j * 4 + i]; + IHT_4[tx_type].cols(temp_in, temp_out); + for (j = 0; j < 4; ++j) { + dest[j * stride + i] = clip_pixel_add(dest[j * stride + i], + ROUND_POWER_OF_TWO(temp_out[j], 4)); + } + } +} + +static const transform_2d IHT_8[] = { + { idct8_c, idct8_c }, // DCT_DCT = 0 + { iadst8_c, idct8_c }, // ADST_DCT = 1 + { idct8_c, iadst8_c }, // DCT_ADST = 2 + { iadst8_c, iadst8_c } // ADST_ADST = 3 +}; + +void vp9_iht8x8_64_add_c(const tran_low_t *input, uint8_t *dest, int stride, + int tx_type) { + int i, j; + tran_low_t out[8 * 8]; + tran_low_t *outptr = out; + tran_low_t temp_in[8], temp_out[8]; + const transform_2d ht = IHT_8[tx_type]; + + // inverse transform row vectors + for (i = 0; i < 8; ++i) { + ht.rows(input, outptr); + input += 8; + outptr += 8; + } + + // inverse transform column vectors + for (i = 0; i < 8; ++i) { + for (j = 0; j < 8; ++j) + temp_in[j] = out[j * 8 + i]; + ht.cols(temp_in, temp_out); + for (j = 0; j < 8; ++j) { + dest[j * stride + i] = clip_pixel_add(dest[j * stride + i], + ROUND_POWER_OF_TWO(temp_out[j], 5)); + } + } +} + +static const transform_2d IHT_16[] = { + { idct16_c, idct16_c }, // DCT_DCT = 0 + { iadst16_c, idct16_c }, // ADST_DCT = 1 + { idct16_c, iadst16_c }, // DCT_ADST = 2 + { iadst16_c, iadst16_c } // ADST_ADST = 3 +}; + +void vp9_iht16x16_256_add_c(const tran_low_t *input, uint8_t *dest, int stride, + int tx_type) { + int i, j; + tran_low_t out[16 * 16]; + tran_low_t *outptr = out; + tran_low_t temp_in[16], temp_out[16]; + const transform_2d ht = IHT_16[tx_type]; + + // Rows + for (i = 0; i < 16; ++i) { + ht.rows(input, outptr); + input += 16; + outptr += 16; + } + + // Columns + for (i = 0; i < 16; ++i) { + for (j = 0; j < 16; ++j) + temp_in[j] = out[j * 16 + i]; + ht.cols(temp_in, temp_out); + for (j = 0; j < 16; ++j) { + dest[j * stride + i] = clip_pixel_add(dest[j * stride + i], + ROUND_POWER_OF_TWO(temp_out[j], 6)); + } + } +} + +// idct +void vp9_idct4x4_add(const tran_low_t *input, uint8_t *dest, int stride, + int eob) { + if (eob > 1) + vpx_idct4x4_16_add(input, dest, stride); + else + vpx_idct4x4_1_add(input, dest, stride); +} + + +void vp9_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride, + int eob) { + if (eob > 1) + vpx_iwht4x4_16_add(input, dest, stride); + else + vpx_iwht4x4_1_add(input, dest, stride); +} + +void vp9_idct8x8_add(const tran_low_t *input, uint8_t *dest, int stride, + int eob) { + // If dc is 1, then input[0] is the reconstructed value, do not need + // dequantization. Also, when dc is 1, dc is counted in eobs, namely eobs >=1. + + // The calculation can be simplified if there are not many non-zero dct + // coefficients. Use eobs to decide what to do. + // TODO(yunqingwang): "eobs = 1" case is also handled in vp9_short_idct8x8_c. + // Combine that with code here. + if (eob == 1) + // DC only DCT coefficient + vpx_idct8x8_1_add(input, dest, stride); + else if (eob <= 12) + vpx_idct8x8_12_add(input, dest, stride); + else + vpx_idct8x8_64_add(input, dest, stride); +} + +void vp9_idct16x16_add(const tran_low_t *input, uint8_t *dest, int stride, + int eob) { + /* The calculation can be simplified if there are not many non-zero dct + * coefficients. Use eobs to separate different cases. */ + if (eob == 1) + /* DC only DCT coefficient. */ + vpx_idct16x16_1_add(input, dest, stride); + else if (eob <= 10) + vpx_idct16x16_10_add(input, dest, stride); + else + vpx_idct16x16_256_add(input, dest, stride); +} + +void vp9_idct32x32_add(const tran_low_t *input, uint8_t *dest, int stride, + int eob) { + if (eob == 1) + vpx_idct32x32_1_add(input, dest, stride); + else if (eob <= 34) + // non-zero coeff only in upper-left 8x8 + vpx_idct32x32_34_add(input, dest, stride); + else if (eob <= 135) + // non-zero coeff only in upper-left 16x16 + vpx_idct32x32_135_add(input, dest, stride); + else + vpx_idct32x32_1024_add(input, dest, stride); +} + +// iht +void vp9_iht4x4_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest, + int stride, int eob) { + if (tx_type == DCT_DCT) + vp9_idct4x4_add(input, dest, stride, eob); + else + vp9_iht4x4_16_add(input, dest, stride, tx_type); +} + +void vp9_iht8x8_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest, + int stride, int eob) { + if (tx_type == DCT_DCT) { + vp9_idct8x8_add(input, dest, stride, eob); + } else { + vp9_iht8x8_64_add(input, dest, stride, tx_type); + } +} + +void vp9_iht16x16_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest, + int stride, int eob) { + if (tx_type == DCT_DCT) { + vp9_idct16x16_add(input, dest, stride, eob); + } else { + vp9_iht16x16_256_add(input, dest, stride, tx_type); + } +} + +#if CONFIG_VP9_HIGHBITDEPTH +void vp9_highbd_iht4x4_16_add_c(const tran_low_t *input, uint8_t *dest8, + int stride, int tx_type, int bd) { + const highbd_transform_2d IHT_4[] = { + { vpx_highbd_idct4_c, vpx_highbd_idct4_c }, // DCT_DCT = 0 + { vpx_highbd_iadst4_c, vpx_highbd_idct4_c }, // ADST_DCT = 1 + { vpx_highbd_idct4_c, vpx_highbd_iadst4_c }, // DCT_ADST = 2 + { vpx_highbd_iadst4_c, vpx_highbd_iadst4_c } // ADST_ADST = 3 + }; + uint16_t *dest = CONVERT_TO_SHORTPTR(dest8); + + int i, j; + tran_low_t out[4 * 4]; + tran_low_t *outptr = out; + tran_low_t temp_in[4], temp_out[4]; + + // Inverse transform row vectors. + for (i = 0; i < 4; ++i) { + IHT_4[tx_type].rows(input, outptr, bd); + input += 4; + outptr += 4; + } + + // Inverse transform column vectors. + for (i = 0; i < 4; ++i) { + for (j = 0; j < 4; ++j) + temp_in[j] = out[j * 4 + i]; + IHT_4[tx_type].cols(temp_in, temp_out, bd); + for (j = 0; j < 4; ++j) { + dest[j * stride + i] = highbd_clip_pixel_add( + dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 4), bd); + } + } +} + +static const highbd_transform_2d HIGH_IHT_8[] = { + { vpx_highbd_idct8_c, vpx_highbd_idct8_c }, // DCT_DCT = 0 + { vpx_highbd_iadst8_c, vpx_highbd_idct8_c }, // ADST_DCT = 1 + { vpx_highbd_idct8_c, vpx_highbd_iadst8_c }, // DCT_ADST = 2 + { vpx_highbd_iadst8_c, vpx_highbd_iadst8_c } // ADST_ADST = 3 +}; + +void vp9_highbd_iht8x8_64_add_c(const tran_low_t *input, uint8_t *dest8, + int stride, int tx_type, int bd) { + int i, j; + tran_low_t out[8 * 8]; + tran_low_t *outptr = out; + tran_low_t temp_in[8], temp_out[8]; + const highbd_transform_2d ht = HIGH_IHT_8[tx_type]; + uint16_t *dest = CONVERT_TO_SHORTPTR(dest8); + + // Inverse transform row vectors. + for (i = 0; i < 8; ++i) { + ht.rows(input, outptr, bd); + input += 8; + outptr += 8; + } + + // Inverse transform column vectors. + for (i = 0; i < 8; ++i) { + for (j = 0; j < 8; ++j) + temp_in[j] = out[j * 8 + i]; + ht.cols(temp_in, temp_out, bd); + for (j = 0; j < 8; ++j) { + dest[j * stride + i] = highbd_clip_pixel_add( + dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 5), bd); + } + } +} + +static const highbd_transform_2d HIGH_IHT_16[] = { + { vpx_highbd_idct16_c, vpx_highbd_idct16_c }, // DCT_DCT = 0 + { vpx_highbd_iadst16_c, vpx_highbd_idct16_c }, // ADST_DCT = 1 + { vpx_highbd_idct16_c, vpx_highbd_iadst16_c }, // DCT_ADST = 2 + { vpx_highbd_iadst16_c, vpx_highbd_iadst16_c } // ADST_ADST = 3 +}; + +void vp9_highbd_iht16x16_256_add_c(const tran_low_t *input, uint8_t *dest8, + int stride, int tx_type, int bd) { + int i, j; + tran_low_t out[16 * 16]; + tran_low_t *outptr = out; + tran_low_t temp_in[16], temp_out[16]; + const highbd_transform_2d ht = HIGH_IHT_16[tx_type]; + uint16_t *dest = CONVERT_TO_SHORTPTR(dest8); + + // Rows + for (i = 0; i < 16; ++i) { + ht.rows(input, outptr, bd); + input += 16; + outptr += 16; + } + + // Columns + for (i = 0; i < 16; ++i) { + for (j = 0; j < 16; ++j) + temp_in[j] = out[j * 16 + i]; + ht.cols(temp_in, temp_out, bd); + for (j = 0; j < 16; ++j) { + dest[j * stride + i] = highbd_clip_pixel_add( + dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 6), bd); + } + } +} + +// idct +void vp9_highbd_idct4x4_add(const tran_low_t *input, uint8_t *dest, int stride, + int eob, int bd) { + if (eob > 1) + vpx_highbd_idct4x4_16_add(input, dest, stride, bd); + else + vpx_highbd_idct4x4_1_add(input, dest, stride, bd); +} + + +void vp9_highbd_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride, + int eob, int bd) { + if (eob > 1) + vpx_highbd_iwht4x4_16_add(input, dest, stride, bd); + else + vpx_highbd_iwht4x4_1_add(input, dest, stride, bd); +} + +void vp9_highbd_idct8x8_add(const tran_low_t *input, uint8_t *dest, int stride, + int eob, int bd) { + // If dc is 1, then input[0] is the reconstructed value, do not need + // dequantization. Also, when dc is 1, dc is counted in eobs, namely eobs >=1. + + // The calculation can be simplified if there are not many non-zero dct + // coefficients. Use eobs to decide what to do. + // TODO(yunqingwang): "eobs = 1" case is also handled in vp9_short_idct8x8_c. + // Combine that with code here. + // DC only DCT coefficient + if (eob == 1) { + vpx_highbd_idct8x8_1_add(input, dest, stride, bd); + } else if (eob <= 10) { + vpx_highbd_idct8x8_10_add(input, dest, stride, bd); + } else { + vpx_highbd_idct8x8_64_add(input, dest, stride, bd); + } +} + +void vp9_highbd_idct16x16_add(const tran_low_t *input, uint8_t *dest, + int stride, int eob, int bd) { + // The calculation can be simplified if there are not many non-zero dct + // coefficients. Use eobs to separate different cases. + // DC only DCT coefficient. + if (eob == 1) { + vpx_highbd_idct16x16_1_add(input, dest, stride, bd); + } else if (eob <= 10) { + vpx_highbd_idct16x16_10_add(input, dest, stride, bd); + } else { + vpx_highbd_idct16x16_256_add(input, dest, stride, bd); + } +} + +void vp9_highbd_idct32x32_add(const tran_low_t *input, uint8_t *dest, + int stride, int eob, int bd) { + // Non-zero coeff only in upper-left 8x8 + if (eob == 1) { + vpx_highbd_idct32x32_1_add(input, dest, stride, bd); + } else if (eob <= 34) { + vpx_highbd_idct32x32_34_add(input, dest, stride, bd); + } else { + vpx_highbd_idct32x32_1024_add(input, dest, stride, bd); + } +} + +// iht +void vp9_highbd_iht4x4_add(TX_TYPE tx_type, const tran_low_t *input, + uint8_t *dest, int stride, int eob, int bd) { + if (tx_type == DCT_DCT) + vp9_highbd_idct4x4_add(input, dest, stride, eob, bd); + else + vp9_highbd_iht4x4_16_add(input, dest, stride, tx_type, bd); +} + +void vp9_highbd_iht8x8_add(TX_TYPE tx_type, const tran_low_t *input, + uint8_t *dest, int stride, int eob, int bd) { + if (tx_type == DCT_DCT) { + vp9_highbd_idct8x8_add(input, dest, stride, eob, bd); + } else { + vp9_highbd_iht8x8_64_add(input, dest, stride, tx_type, bd); + } +} + +void vp9_highbd_iht16x16_add(TX_TYPE tx_type, const tran_low_t *input, + uint8_t *dest, int stride, int eob, int bd) { + if (tx_type == DCT_DCT) { + vp9_highbd_idct16x16_add(input, dest, stride, eob, bd); + } else { + vp9_highbd_iht16x16_256_add(input, dest, stride, tx_type, bd); + } +} +#endif // CONFIG_VP9_HIGHBITDEPTH diff --git a/thirdparty/libvpx/vp9/common/vp9_idct.h b/thirdparty/libvpx/vp9/common/vp9_idct.h new file mode 100644 index 0000000000..b5a3fbf362 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_idct.h @@ -0,0 +1,81 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_COMMON_VP9_IDCT_H_ +#define VP9_COMMON_VP9_IDCT_H_ + +#include <assert.h> + +#include "./vpx_config.h" +#include "vp9/common/vp9_common.h" +#include "vp9/common/vp9_enums.h" +#include "vpx_dsp/inv_txfm.h" +#include "vpx_dsp/txfm_common.h" +#include "vpx_ports/mem.h" + +#ifdef __cplusplus +extern "C" { +#endif + +typedef void (*transform_1d)(const tran_low_t*, tran_low_t*); + +typedef struct { + transform_1d cols, rows; // vertical and horizontal +} transform_2d; + +#if CONFIG_VP9_HIGHBITDEPTH +typedef void (*highbd_transform_1d)(const tran_low_t*, tran_low_t*, int bd); + +typedef struct { + highbd_transform_1d cols, rows; // vertical and horizontal +} highbd_transform_2d; +#endif // CONFIG_VP9_HIGHBITDEPTH + +void vp9_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride, + int eob); +void vp9_idct4x4_add(const tran_low_t *input, uint8_t *dest, int stride, + int eob); +void vp9_idct8x8_add(const tran_low_t *input, uint8_t *dest, int stride, + int eob); +void vp9_idct16x16_add(const tran_low_t *input, uint8_t *dest, int stride, + int eob); +void vp9_idct32x32_add(const tran_low_t *input, uint8_t *dest, int stride, + int eob); + +void vp9_iht4x4_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest, + int stride, int eob); +void vp9_iht8x8_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest, + int stride, int eob); +void vp9_iht16x16_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest, + int stride, int eob); + +#if CONFIG_VP9_HIGHBITDEPTH +void vp9_highbd_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride, + int eob, int bd); +void vp9_highbd_idct4x4_add(const tran_low_t *input, uint8_t *dest, int stride, + int eob, int bd); +void vp9_highbd_idct8x8_add(const tran_low_t *input, uint8_t *dest, int stride, + int eob, int bd); +void vp9_highbd_idct16x16_add(const tran_low_t *input, uint8_t *dest, + int stride, int eob, int bd); +void vp9_highbd_idct32x32_add(const tran_low_t *input, uint8_t *dest, + int stride, int eob, int bd); +void vp9_highbd_iht4x4_add(TX_TYPE tx_type, const tran_low_t *input, + uint8_t *dest, int stride, int eob, int bd); +void vp9_highbd_iht8x8_add(TX_TYPE tx_type, const tran_low_t *input, + uint8_t *dest, int stride, int eob, int bd); +void vp9_highbd_iht16x16_add(TX_TYPE tx_type, const tran_low_t *input, + uint8_t *dest, int stride, int eob, int bd); +#endif // CONFIG_VP9_HIGHBITDEPTH +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_IDCT_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_loopfilter.c b/thirdparty/libvpx/vp9/common/vp9_loopfilter.c new file mode 100644 index 0000000000..183dec4e71 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_loopfilter.c @@ -0,0 +1,1697 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "./vpx_config.h" +#include "./vpx_dsp_rtcd.h" +#include "vp9/common/vp9_loopfilter.h" +#include "vp9/common/vp9_onyxc_int.h" +#include "vp9/common/vp9_reconinter.h" +#include "vpx_dsp/vpx_dsp_common.h" +#include "vpx_mem/vpx_mem.h" +#include "vpx_ports/mem.h" + +#include "vp9/common/vp9_seg_common.h" + +// 64 bit masks for left transform size. Each 1 represents a position where +// we should apply a loop filter across the left border of an 8x8 block +// boundary. +// +// In the case of TX_16X16-> ( in low order byte first we end up with +// a mask that looks like this +// +// 10101010 +// 10101010 +// 10101010 +// 10101010 +// 10101010 +// 10101010 +// 10101010 +// 10101010 +// +// A loopfilter should be applied to every other 8x8 horizontally. +static const uint64_t left_64x64_txform_mask[TX_SIZES]= { + 0xffffffffffffffffULL, // TX_4X4 + 0xffffffffffffffffULL, // TX_8x8 + 0x5555555555555555ULL, // TX_16x16 + 0x1111111111111111ULL, // TX_32x32 +}; + +// 64 bit masks for above transform size. Each 1 represents a position where +// we should apply a loop filter across the top border of an 8x8 block +// boundary. +// +// In the case of TX_32x32 -> ( in low order byte first we end up with +// a mask that looks like this +// +// 11111111 +// 00000000 +// 00000000 +// 00000000 +// 11111111 +// 00000000 +// 00000000 +// 00000000 +// +// A loopfilter should be applied to every other 4 the row vertically. +static const uint64_t above_64x64_txform_mask[TX_SIZES]= { + 0xffffffffffffffffULL, // TX_4X4 + 0xffffffffffffffffULL, // TX_8x8 + 0x00ff00ff00ff00ffULL, // TX_16x16 + 0x000000ff000000ffULL, // TX_32x32 +}; + +// 64 bit masks for prediction sizes (left). Each 1 represents a position +// where left border of an 8x8 block. These are aligned to the right most +// appropriate bit, and then shifted into place. +// +// In the case of TX_16x32 -> ( low order byte first ) we end up with +// a mask that looks like this : +// +// 10000000 +// 10000000 +// 10000000 +// 10000000 +// 00000000 +// 00000000 +// 00000000 +// 00000000 +static const uint64_t left_prediction_mask[BLOCK_SIZES] = { + 0x0000000000000001ULL, // BLOCK_4X4, + 0x0000000000000001ULL, // BLOCK_4X8, + 0x0000000000000001ULL, // BLOCK_8X4, + 0x0000000000000001ULL, // BLOCK_8X8, + 0x0000000000000101ULL, // BLOCK_8X16, + 0x0000000000000001ULL, // BLOCK_16X8, + 0x0000000000000101ULL, // BLOCK_16X16, + 0x0000000001010101ULL, // BLOCK_16X32, + 0x0000000000000101ULL, // BLOCK_32X16, + 0x0000000001010101ULL, // BLOCK_32X32, + 0x0101010101010101ULL, // BLOCK_32X64, + 0x0000000001010101ULL, // BLOCK_64X32, + 0x0101010101010101ULL, // BLOCK_64X64 +}; + +// 64 bit mask to shift and set for each prediction size. +static const uint64_t above_prediction_mask[BLOCK_SIZES] = { + 0x0000000000000001ULL, // BLOCK_4X4 + 0x0000000000000001ULL, // BLOCK_4X8 + 0x0000000000000001ULL, // BLOCK_8X4 + 0x0000000000000001ULL, // BLOCK_8X8 + 0x0000000000000001ULL, // BLOCK_8X16, + 0x0000000000000003ULL, // BLOCK_16X8 + 0x0000000000000003ULL, // BLOCK_16X16 + 0x0000000000000003ULL, // BLOCK_16X32, + 0x000000000000000fULL, // BLOCK_32X16, + 0x000000000000000fULL, // BLOCK_32X32, + 0x000000000000000fULL, // BLOCK_32X64, + 0x00000000000000ffULL, // BLOCK_64X32, + 0x00000000000000ffULL, // BLOCK_64X64 +}; +// 64 bit mask to shift and set for each prediction size. A bit is set for +// each 8x8 block that would be in the left most block of the given block +// size in the 64x64 block. +static const uint64_t size_mask[BLOCK_SIZES] = { + 0x0000000000000001ULL, // BLOCK_4X4 + 0x0000000000000001ULL, // BLOCK_4X8 + 0x0000000000000001ULL, // BLOCK_8X4 + 0x0000000000000001ULL, // BLOCK_8X8 + 0x0000000000000101ULL, // BLOCK_8X16, + 0x0000000000000003ULL, // BLOCK_16X8 + 0x0000000000000303ULL, // BLOCK_16X16 + 0x0000000003030303ULL, // BLOCK_16X32, + 0x0000000000000f0fULL, // BLOCK_32X16, + 0x000000000f0f0f0fULL, // BLOCK_32X32, + 0x0f0f0f0f0f0f0f0fULL, // BLOCK_32X64, + 0x00000000ffffffffULL, // BLOCK_64X32, + 0xffffffffffffffffULL, // BLOCK_64X64 +}; + +// These are used for masking the left and above borders. +static const uint64_t left_border = 0x1111111111111111ULL; +static const uint64_t above_border = 0x000000ff000000ffULL; + +// 16 bit masks for uv transform sizes. +static const uint16_t left_64x64_txform_mask_uv[TX_SIZES]= { + 0xffff, // TX_4X4 + 0xffff, // TX_8x8 + 0x5555, // TX_16x16 + 0x1111, // TX_32x32 +}; + +static const uint16_t above_64x64_txform_mask_uv[TX_SIZES]= { + 0xffff, // TX_4X4 + 0xffff, // TX_8x8 + 0x0f0f, // TX_16x16 + 0x000f, // TX_32x32 +}; + +// 16 bit left mask to shift and set for each uv prediction size. +static const uint16_t left_prediction_mask_uv[BLOCK_SIZES] = { + 0x0001, // BLOCK_4X4, + 0x0001, // BLOCK_4X8, + 0x0001, // BLOCK_8X4, + 0x0001, // BLOCK_8X8, + 0x0001, // BLOCK_8X16, + 0x0001, // BLOCK_16X8, + 0x0001, // BLOCK_16X16, + 0x0011, // BLOCK_16X32, + 0x0001, // BLOCK_32X16, + 0x0011, // BLOCK_32X32, + 0x1111, // BLOCK_32X64 + 0x0011, // BLOCK_64X32, + 0x1111, // BLOCK_64X64 +}; +// 16 bit above mask to shift and set for uv each prediction size. +static const uint16_t above_prediction_mask_uv[BLOCK_SIZES] = { + 0x0001, // BLOCK_4X4 + 0x0001, // BLOCK_4X8 + 0x0001, // BLOCK_8X4 + 0x0001, // BLOCK_8X8 + 0x0001, // BLOCK_8X16, + 0x0001, // BLOCK_16X8 + 0x0001, // BLOCK_16X16 + 0x0001, // BLOCK_16X32, + 0x0003, // BLOCK_32X16, + 0x0003, // BLOCK_32X32, + 0x0003, // BLOCK_32X64, + 0x000f, // BLOCK_64X32, + 0x000f, // BLOCK_64X64 +}; + +// 64 bit mask to shift and set for each uv prediction size +static const uint16_t size_mask_uv[BLOCK_SIZES] = { + 0x0001, // BLOCK_4X4 + 0x0001, // BLOCK_4X8 + 0x0001, // BLOCK_8X4 + 0x0001, // BLOCK_8X8 + 0x0001, // BLOCK_8X16, + 0x0001, // BLOCK_16X8 + 0x0001, // BLOCK_16X16 + 0x0011, // BLOCK_16X32, + 0x0003, // BLOCK_32X16, + 0x0033, // BLOCK_32X32, + 0x3333, // BLOCK_32X64, + 0x00ff, // BLOCK_64X32, + 0xffff, // BLOCK_64X64 +}; +static const uint16_t left_border_uv = 0x1111; +static const uint16_t above_border_uv = 0x000f; + +static const int mode_lf_lut[MB_MODE_COUNT] = { + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // INTRA_MODES + 1, 1, 0, 1 // INTER_MODES (ZEROMV == 0) +}; + +static void update_sharpness(loop_filter_info_n *lfi, int sharpness_lvl) { + int lvl; + + // For each possible value for the loop filter fill out limits + for (lvl = 0; lvl <= MAX_LOOP_FILTER; lvl++) { + // Set loop filter parameters that control sharpness. + int block_inside_limit = lvl >> ((sharpness_lvl > 0) + (sharpness_lvl > 4)); + + if (sharpness_lvl > 0) { + if (block_inside_limit > (9 - sharpness_lvl)) + block_inside_limit = (9 - sharpness_lvl); + } + + if (block_inside_limit < 1) + block_inside_limit = 1; + + memset(lfi->lfthr[lvl].lim, block_inside_limit, SIMD_WIDTH); + memset(lfi->lfthr[lvl].mblim, (2 * (lvl + 2) + block_inside_limit), + SIMD_WIDTH); + } +} + +static uint8_t get_filter_level(const loop_filter_info_n *lfi_n, + const MODE_INFO *mi) { + return lfi_n->lvl[mi->segment_id][mi->ref_frame[0]] + [mode_lf_lut[mi->mode]]; +} + +void vp9_loop_filter_init(VP9_COMMON *cm) { + loop_filter_info_n *lfi = &cm->lf_info; + struct loopfilter *lf = &cm->lf; + int lvl; + + // init limits for given sharpness + update_sharpness(lfi, lf->sharpness_level); + lf->last_sharpness_level = lf->sharpness_level; + + // init hev threshold const vectors + for (lvl = 0; lvl <= MAX_LOOP_FILTER; lvl++) + memset(lfi->lfthr[lvl].hev_thr, (lvl >> 4), SIMD_WIDTH); +} + +void vp9_loop_filter_frame_init(VP9_COMMON *cm, int default_filt_lvl) { + int seg_id; + // n_shift is the multiplier for lf_deltas + // the multiplier is 1 for when filter_lvl is between 0 and 31; + // 2 when filter_lvl is between 32 and 63 + const int scale = 1 << (default_filt_lvl >> 5); + loop_filter_info_n *const lfi = &cm->lf_info; + struct loopfilter *const lf = &cm->lf; + const struct segmentation *const seg = &cm->seg; + + // update limits if sharpness has changed + if (lf->last_sharpness_level != lf->sharpness_level) { + update_sharpness(lfi, lf->sharpness_level); + lf->last_sharpness_level = lf->sharpness_level; + } + + for (seg_id = 0; seg_id < MAX_SEGMENTS; seg_id++) { + int lvl_seg = default_filt_lvl; + if (segfeature_active(seg, seg_id, SEG_LVL_ALT_LF)) { + const int data = get_segdata(seg, seg_id, SEG_LVL_ALT_LF); + lvl_seg = clamp(seg->abs_delta == SEGMENT_ABSDATA ? + data : default_filt_lvl + data, + 0, MAX_LOOP_FILTER); + } + + if (!lf->mode_ref_delta_enabled) { + // we could get rid of this if we assume that deltas are set to + // zero when not in use; encoder always uses deltas + memset(lfi->lvl[seg_id], lvl_seg, sizeof(lfi->lvl[seg_id])); + } else { + int ref, mode; + const int intra_lvl = lvl_seg + lf->ref_deltas[INTRA_FRAME] * scale; + lfi->lvl[seg_id][INTRA_FRAME][0] = clamp(intra_lvl, 0, MAX_LOOP_FILTER); + + for (ref = LAST_FRAME; ref < MAX_REF_FRAMES; ++ref) { + for (mode = 0; mode < MAX_MODE_LF_DELTAS; ++mode) { + const int inter_lvl = lvl_seg + lf->ref_deltas[ref] * scale + + lf->mode_deltas[mode] * scale; + lfi->lvl[seg_id][ref][mode] = clamp(inter_lvl, 0, MAX_LOOP_FILTER); + } + } + } + } +} + +static void filter_selectively_vert_row2(int subsampling_factor, + uint8_t *s, int pitch, + unsigned int mask_16x16, + unsigned int mask_8x8, + unsigned int mask_4x4, + unsigned int mask_4x4_int, + const loop_filter_thresh *lfthr, + const uint8_t *lfl) { + const int dual_mask_cutoff = subsampling_factor ? 0xff : 0xffff; + const int lfl_forward = subsampling_factor ? 4 : 8; + const unsigned int dual_one = 1 | (1 << lfl_forward); + unsigned int mask; + uint8_t *ss[2]; + ss[0] = s; + + for (mask = + (mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int) & dual_mask_cutoff; + mask; mask = (mask & ~dual_one) >> 1) { + if (mask & dual_one) { + const loop_filter_thresh *lfis[2]; + lfis[0] = lfthr + *lfl; + lfis[1] = lfthr + *(lfl + lfl_forward); + ss[1] = ss[0] + 8 * pitch; + + if (mask_16x16 & dual_one) { + if ((mask_16x16 & dual_one) == dual_one) { + vpx_lpf_vertical_16_dual(ss[0], pitch, lfis[0]->mblim, lfis[0]->lim, + lfis[0]->hev_thr); + } else { + const loop_filter_thresh *lfi = lfis[!(mask_16x16 & 1)]; + vpx_lpf_vertical_16(ss[!(mask_16x16 & 1)], pitch, lfi->mblim, + lfi->lim, lfi->hev_thr); + } + } + + if (mask_8x8 & dual_one) { + if ((mask_8x8 & dual_one) == dual_one) { + vpx_lpf_vertical_8_dual(ss[0], pitch, lfis[0]->mblim, lfis[0]->lim, + lfis[0]->hev_thr, lfis[1]->mblim, + lfis[1]->lim, lfis[1]->hev_thr); + } else { + const loop_filter_thresh *lfi = lfis[!(mask_8x8 & 1)]; + vpx_lpf_vertical_8(ss[!(mask_8x8 & 1)], pitch, lfi->mblim, lfi->lim, + lfi->hev_thr); + } + } + + if (mask_4x4 & dual_one) { + if ((mask_4x4 & dual_one) == dual_one) { + vpx_lpf_vertical_4_dual(ss[0], pitch, lfis[0]->mblim, lfis[0]->lim, + lfis[0]->hev_thr, lfis[1]->mblim, + lfis[1]->lim, lfis[1]->hev_thr); + } else { + const loop_filter_thresh *lfi = lfis[!(mask_4x4 & 1)]; + vpx_lpf_vertical_4(ss[!(mask_4x4 & 1)], pitch, lfi->mblim, lfi->lim, + lfi->hev_thr); + } + } + + if (mask_4x4_int & dual_one) { + if ((mask_4x4_int & dual_one) == dual_one) { + vpx_lpf_vertical_4_dual(ss[0] + 4, pitch, lfis[0]->mblim, + lfis[0]->lim, lfis[0]->hev_thr, + lfis[1]->mblim, lfis[1]->lim, + lfis[1]->hev_thr); + } else { + const loop_filter_thresh *lfi = lfis[!(mask_4x4_int & 1)]; + vpx_lpf_vertical_4(ss[!(mask_4x4_int & 1)] + 4, pitch, lfi->mblim, + lfi->lim, lfi->hev_thr); + } + } + } + + ss[0] += 8; + lfl += 1; + mask_16x16 >>= 1; + mask_8x8 >>= 1; + mask_4x4 >>= 1; + mask_4x4_int >>= 1; + } +} + +#if CONFIG_VP9_HIGHBITDEPTH +static void highbd_filter_selectively_vert_row2(int subsampling_factor, + uint16_t *s, int pitch, + unsigned int mask_16x16, + unsigned int mask_8x8, + unsigned int mask_4x4, + unsigned int mask_4x4_int, + const loop_filter_thresh *lfthr, + const uint8_t *lfl, int bd) { + const int dual_mask_cutoff = subsampling_factor ? 0xff : 0xffff; + const int lfl_forward = subsampling_factor ? 4 : 8; + const unsigned int dual_one = 1 | (1 << lfl_forward); + unsigned int mask; + uint16_t *ss[2]; + ss[0] = s; + + for (mask = + (mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int) & dual_mask_cutoff; + mask; mask = (mask & ~dual_one) >> 1) { + if (mask & dual_one) { + const loop_filter_thresh *lfis[2]; + lfis[0] = lfthr + *lfl; + lfis[1] = lfthr + *(lfl + lfl_forward); + ss[1] = ss[0] + 8 * pitch; + + if (mask_16x16 & dual_one) { + if ((mask_16x16 & dual_one) == dual_one) { + vpx_highbd_lpf_vertical_16_dual(ss[0], pitch, lfis[0]->mblim, + lfis[0]->lim, lfis[0]->hev_thr, bd); + } else { + const loop_filter_thresh *lfi = lfis[!(mask_16x16 & 1)]; + vpx_highbd_lpf_vertical_16(ss[!(mask_16x16 & 1)], pitch, lfi->mblim, + lfi->lim, lfi->hev_thr, bd); + } + } + + if (mask_8x8 & dual_one) { + if ((mask_8x8 & dual_one) == dual_one) { + vpx_highbd_lpf_vertical_8_dual(ss[0], pitch, lfis[0]->mblim, + lfis[0]->lim, lfis[0]->hev_thr, + lfis[1]->mblim, lfis[1]->lim, + lfis[1]->hev_thr, bd); + } else { + const loop_filter_thresh *lfi = lfis[!(mask_8x8 & 1)]; + vpx_highbd_lpf_vertical_8(ss[!(mask_8x8 & 1)], pitch, lfi->mblim, + lfi->lim, lfi->hev_thr, bd); + } + } + + if (mask_4x4 & dual_one) { + if ((mask_4x4 & dual_one) == dual_one) { + vpx_highbd_lpf_vertical_4_dual(ss[0], pitch, lfis[0]->mblim, + lfis[0]->lim, lfis[0]->hev_thr, + lfis[1]->mblim, lfis[1]->lim, + lfis[1]->hev_thr, bd); + } else { + const loop_filter_thresh *lfi = lfis[!(mask_4x4 & 1)]; + vpx_highbd_lpf_vertical_4(ss[!(mask_4x4 & 1)], pitch, lfi->mblim, + lfi->lim, lfi->hev_thr, bd); + } + } + + if (mask_4x4_int & dual_one) { + if ((mask_4x4_int & dual_one) == dual_one) { + vpx_highbd_lpf_vertical_4_dual(ss[0] + 4, pitch, lfis[0]->mblim, + lfis[0]->lim, lfis[0]->hev_thr, + lfis[1]->mblim, lfis[1]->lim, + lfis[1]->hev_thr, bd); + } else { + const loop_filter_thresh *lfi = lfis[!(mask_4x4_int & 1)]; + vpx_highbd_lpf_vertical_4(ss[!(mask_4x4_int & 1)] + 4, pitch, + lfi->mblim, lfi->lim, lfi->hev_thr, bd); + } + } + } + + ss[0] += 8; + lfl += 1; + mask_16x16 >>= 1; + mask_8x8 >>= 1; + mask_4x4 >>= 1; + mask_4x4_int >>= 1; + } +} +#endif // CONFIG_VP9_HIGHBITDEPTH + +static void filter_selectively_horiz(uint8_t *s, int pitch, + unsigned int mask_16x16, + unsigned int mask_8x8, + unsigned int mask_4x4, + unsigned int mask_4x4_int, + const loop_filter_thresh *lfthr, + const uint8_t *lfl) { + unsigned int mask; + int count; + + for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int; + mask; mask >>= count) { + count = 1; + if (mask & 1) { + const loop_filter_thresh *lfi = lfthr + *lfl; + + if (mask_16x16 & 1) { + if ((mask_16x16 & 3) == 3) { + vpx_lpf_horizontal_edge_16(s, pitch, lfi->mblim, lfi->lim, + lfi->hev_thr); + count = 2; + } else { + vpx_lpf_horizontal_edge_8(s, pitch, lfi->mblim, lfi->lim, + lfi->hev_thr); + } + } else if (mask_8x8 & 1) { + if ((mask_8x8 & 3) == 3) { + // Next block's thresholds. + const loop_filter_thresh *lfin = lfthr + *(lfl + 1); + + vpx_lpf_horizontal_8_dual(s, pitch, lfi->mblim, lfi->lim, + lfi->hev_thr, lfin->mblim, lfin->lim, + lfin->hev_thr); + + if ((mask_4x4_int & 3) == 3) { + vpx_lpf_horizontal_4_dual(s + 4 * pitch, pitch, lfi->mblim, + lfi->lim, lfi->hev_thr, lfin->mblim, + lfin->lim, lfin->hev_thr); + } else { + if (mask_4x4_int & 1) + vpx_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim, + lfi->hev_thr); + else if (mask_4x4_int & 2) + vpx_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim, + lfin->lim, lfin->hev_thr); + } + count = 2; + } else { + vpx_lpf_horizontal_8(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr); + + if (mask_4x4_int & 1) + vpx_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim, + lfi->hev_thr); + } + } else if (mask_4x4 & 1) { + if ((mask_4x4 & 3) == 3) { + // Next block's thresholds. + const loop_filter_thresh *lfin = lfthr + *(lfl + 1); + + vpx_lpf_horizontal_4_dual(s, pitch, lfi->mblim, lfi->lim, + lfi->hev_thr, lfin->mblim, lfin->lim, + lfin->hev_thr); + if ((mask_4x4_int & 3) == 3) { + vpx_lpf_horizontal_4_dual(s + 4 * pitch, pitch, lfi->mblim, + lfi->lim, lfi->hev_thr, lfin->mblim, + lfin->lim, lfin->hev_thr); + } else { + if (mask_4x4_int & 1) + vpx_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim, + lfi->hev_thr); + else if (mask_4x4_int & 2) + vpx_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim, + lfin->lim, lfin->hev_thr); + } + count = 2; + } else { + vpx_lpf_horizontal_4(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr); + + if (mask_4x4_int & 1) + vpx_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim, + lfi->hev_thr); + } + } else { + vpx_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim, + lfi->hev_thr); + } + } + s += 8 * count; + lfl += count; + mask_16x16 >>= count; + mask_8x8 >>= count; + mask_4x4 >>= count; + mask_4x4_int >>= count; + } +} + +#if CONFIG_VP9_HIGHBITDEPTH +static void highbd_filter_selectively_horiz(uint16_t *s, int pitch, + unsigned int mask_16x16, + unsigned int mask_8x8, + unsigned int mask_4x4, + unsigned int mask_4x4_int, + const loop_filter_thresh *lfthr, + const uint8_t *lfl, int bd) { + unsigned int mask; + int count; + + for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int; + mask; mask >>= count) { + count = 1; + if (mask & 1) { + const loop_filter_thresh *lfi = lfthr + *lfl; + + if (mask_16x16 & 1) { + if ((mask_16x16 & 3) == 3) { + vpx_highbd_lpf_horizontal_edge_16(s, pitch, lfi->mblim, lfi->lim, + lfi->hev_thr, bd); + count = 2; + } else { + vpx_highbd_lpf_horizontal_edge_8(s, pitch, lfi->mblim, lfi->lim, + lfi->hev_thr, bd); + } + } else if (mask_8x8 & 1) { + if ((mask_8x8 & 3) == 3) { + // Next block's thresholds. + const loop_filter_thresh *lfin = lfthr + *(lfl + 1); + + vpx_highbd_lpf_horizontal_8_dual(s, pitch, lfi->mblim, lfi->lim, + lfi->hev_thr, lfin->mblim, lfin->lim, + lfin->hev_thr, bd); + + if ((mask_4x4_int & 3) == 3) { + vpx_highbd_lpf_horizontal_4_dual(s + 4 * pitch, pitch, lfi->mblim, + lfi->lim, lfi->hev_thr, + lfin->mblim, lfin->lim, + lfin->hev_thr, bd); + } else { + if (mask_4x4_int & 1) { + vpx_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, + lfi->lim, lfi->hev_thr, bd); + } else if (mask_4x4_int & 2) { + vpx_highbd_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim, + lfin->lim, lfin->hev_thr, bd); + } + } + count = 2; + } else { + vpx_highbd_lpf_horizontal_8(s, pitch, lfi->mblim, lfi->lim, + lfi->hev_thr, bd); + + if (mask_4x4_int & 1) { + vpx_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, + lfi->lim, lfi->hev_thr, bd); + } + } + } else if (mask_4x4 & 1) { + if ((mask_4x4 & 3) == 3) { + // Next block's thresholds. + const loop_filter_thresh *lfin = lfthr + *(lfl + 1); + + vpx_highbd_lpf_horizontal_4_dual(s, pitch, lfi->mblim, lfi->lim, + lfi->hev_thr, lfin->mblim, lfin->lim, + lfin->hev_thr, bd); + if ((mask_4x4_int & 3) == 3) { + vpx_highbd_lpf_horizontal_4_dual(s + 4 * pitch, pitch, lfi->mblim, + lfi->lim, lfi->hev_thr, + lfin->mblim, lfin->lim, + lfin->hev_thr, bd); + } else { + if (mask_4x4_int & 1) { + vpx_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, + lfi->lim, lfi->hev_thr, bd); + } else if (mask_4x4_int & 2) { + vpx_highbd_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim, + lfin->lim, lfin->hev_thr, bd); + } + } + count = 2; + } else { + vpx_highbd_lpf_horizontal_4(s, pitch, lfi->mblim, lfi->lim, + lfi->hev_thr, bd); + + if (mask_4x4_int & 1) { + vpx_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, + lfi->lim, lfi->hev_thr, bd); + } + } + } else { + vpx_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim, + lfi->hev_thr, bd); + } + } + s += 8 * count; + lfl += count; + mask_16x16 >>= count; + mask_8x8 >>= count; + mask_4x4 >>= count; + mask_4x4_int >>= count; + } +} +#endif // CONFIG_VP9_HIGHBITDEPTH + +// This function ors into the current lfm structure, where to do loop +// filters for the specific mi we are looking at. It uses information +// including the block_size_type (32x16, 32x32, etc.), the transform size, +// whether there were any coefficients encoded, and the loop filter strength +// block we are currently looking at. Shift is used to position the +// 1's we produce. +static void build_masks(const loop_filter_info_n *const lfi_n, + const MODE_INFO *mi, const int shift_y, + const int shift_uv, + LOOP_FILTER_MASK *lfm) { + const BLOCK_SIZE block_size = mi->sb_type; + const TX_SIZE tx_size_y = mi->tx_size; + const TX_SIZE tx_size_uv = get_uv_tx_size_impl(tx_size_y, block_size, 1, 1); + const int filter_level = get_filter_level(lfi_n, mi); + uint64_t *const left_y = &lfm->left_y[tx_size_y]; + uint64_t *const above_y = &lfm->above_y[tx_size_y]; + uint64_t *const int_4x4_y = &lfm->int_4x4_y; + uint16_t *const left_uv = &lfm->left_uv[tx_size_uv]; + uint16_t *const above_uv = &lfm->above_uv[tx_size_uv]; + uint16_t *const int_4x4_uv = &lfm->int_4x4_uv; + int i; + + // If filter level is 0 we don't loop filter. + if (!filter_level) { + return; + } else { + const int w = num_8x8_blocks_wide_lookup[block_size]; + const int h = num_8x8_blocks_high_lookup[block_size]; + int index = shift_y; + for (i = 0; i < h; i++) { + memset(&lfm->lfl_y[index], filter_level, w); + index += 8; + } + } + + // These set 1 in the current block size for the block size edges. + // For instance if the block size is 32x16, we'll set: + // above = 1111 + // 0000 + // and + // left = 1000 + // = 1000 + // NOTE : In this example the low bit is left most ( 1000 ) is stored as + // 1, not 8... + // + // U and V set things on a 16 bit scale. + // + *above_y |= above_prediction_mask[block_size] << shift_y; + *above_uv |= above_prediction_mask_uv[block_size] << shift_uv; + *left_y |= left_prediction_mask[block_size] << shift_y; + *left_uv |= left_prediction_mask_uv[block_size] << shift_uv; + + // If the block has no coefficients and is not intra we skip applying + // the loop filter on block edges. + if (mi->skip && is_inter_block(mi)) + return; + + // Here we are adding a mask for the transform size. The transform + // size mask is set to be correct for a 64x64 prediction block size. We + // mask to match the size of the block we are working on and then shift it + // into place.. + *above_y |= (size_mask[block_size] & + above_64x64_txform_mask[tx_size_y]) << shift_y; + *above_uv |= (size_mask_uv[block_size] & + above_64x64_txform_mask_uv[tx_size_uv]) << shift_uv; + + *left_y |= (size_mask[block_size] & + left_64x64_txform_mask[tx_size_y]) << shift_y; + *left_uv |= (size_mask_uv[block_size] & + left_64x64_txform_mask_uv[tx_size_uv]) << shift_uv; + + // Here we are trying to determine what to do with the internal 4x4 block + // boundaries. These differ from the 4x4 boundaries on the outside edge of + // an 8x8 in that the internal ones can be skipped and don't depend on + // the prediction block size. + if (tx_size_y == TX_4X4) + *int_4x4_y |= size_mask[block_size] << shift_y; + + if (tx_size_uv == TX_4X4) + *int_4x4_uv |= (size_mask_uv[block_size] & 0xffff) << shift_uv; +} + +// This function does the same thing as the one above with the exception that +// it only affects the y masks. It exists because for blocks < 16x16 in size, +// we only update u and v masks on the first block. +static void build_y_mask(const loop_filter_info_n *const lfi_n, + const MODE_INFO *mi, const int shift_y, + LOOP_FILTER_MASK *lfm) { + const BLOCK_SIZE block_size = mi->sb_type; + const TX_SIZE tx_size_y = mi->tx_size; + const int filter_level = get_filter_level(lfi_n, mi); + uint64_t *const left_y = &lfm->left_y[tx_size_y]; + uint64_t *const above_y = &lfm->above_y[tx_size_y]; + uint64_t *const int_4x4_y = &lfm->int_4x4_y; + int i; + + if (!filter_level) { + return; + } else { + const int w = num_8x8_blocks_wide_lookup[block_size]; + const int h = num_8x8_blocks_high_lookup[block_size]; + int index = shift_y; + for (i = 0; i < h; i++) { + memset(&lfm->lfl_y[index], filter_level, w); + index += 8; + } + } + + *above_y |= above_prediction_mask[block_size] << shift_y; + *left_y |= left_prediction_mask[block_size] << shift_y; + + if (mi->skip && is_inter_block(mi)) + return; + + *above_y |= (size_mask[block_size] & + above_64x64_txform_mask[tx_size_y]) << shift_y; + + *left_y |= (size_mask[block_size] & + left_64x64_txform_mask[tx_size_y]) << shift_y; + + if (tx_size_y == TX_4X4) + *int_4x4_y |= size_mask[block_size] << shift_y; +} + +void vp9_adjust_mask(VP9_COMMON *const cm, const int mi_row, + const int mi_col, LOOP_FILTER_MASK *lfm) { + int i; + + // The largest loopfilter we have is 16x16 so we use the 16x16 mask + // for 32x32 transforms also. + lfm->left_y[TX_16X16] |= lfm->left_y[TX_32X32]; + lfm->above_y[TX_16X16] |= lfm->above_y[TX_32X32]; + lfm->left_uv[TX_16X16] |= lfm->left_uv[TX_32X32]; + lfm->above_uv[TX_16X16] |= lfm->above_uv[TX_32X32]; + + // We do at least 8 tap filter on every 32x32 even if the transform size + // is 4x4. So if the 4x4 is set on a border pixel add it to the 8x8 and + // remove it from the 4x4. + lfm->left_y[TX_8X8] |= lfm->left_y[TX_4X4] & left_border; + lfm->left_y[TX_4X4] &= ~left_border; + lfm->above_y[TX_8X8] |= lfm->above_y[TX_4X4] & above_border; + lfm->above_y[TX_4X4] &= ~above_border; + lfm->left_uv[TX_8X8] |= lfm->left_uv[TX_4X4] & left_border_uv; + lfm->left_uv[TX_4X4] &= ~left_border_uv; + lfm->above_uv[TX_8X8] |= lfm->above_uv[TX_4X4] & above_border_uv; + lfm->above_uv[TX_4X4] &= ~above_border_uv; + + // We do some special edge handling. + if (mi_row + MI_BLOCK_SIZE > cm->mi_rows) { + const uint64_t rows = cm->mi_rows - mi_row; + + // Each pixel inside the border gets a 1, + const uint64_t mask_y = (((uint64_t) 1 << (rows << 3)) - 1); + const uint16_t mask_uv = (((uint16_t) 1 << (((rows + 1) >> 1) << 2)) - 1); + + // Remove values completely outside our border. + for (i = 0; i < TX_32X32; i++) { + lfm->left_y[i] &= mask_y; + lfm->above_y[i] &= mask_y; + lfm->left_uv[i] &= mask_uv; + lfm->above_uv[i] &= mask_uv; + } + lfm->int_4x4_y &= mask_y; + lfm->int_4x4_uv &= mask_uv; + + // We don't apply a wide loop filter on the last uv block row. If set + // apply the shorter one instead. + if (rows == 1) { + lfm->above_uv[TX_8X8] |= lfm->above_uv[TX_16X16]; + lfm->above_uv[TX_16X16] = 0; + } + if (rows == 5) { + lfm->above_uv[TX_8X8] |= lfm->above_uv[TX_16X16] & 0xff00; + lfm->above_uv[TX_16X16] &= ~(lfm->above_uv[TX_16X16] & 0xff00); + } + } + + if (mi_col + MI_BLOCK_SIZE > cm->mi_cols) { + const uint64_t columns = cm->mi_cols - mi_col; + + // Each pixel inside the border gets a 1, the multiply copies the border + // to where we need it. + const uint64_t mask_y = (((1 << columns) - 1)) * 0x0101010101010101ULL; + const uint16_t mask_uv = ((1 << ((columns + 1) >> 1)) - 1) * 0x1111; + + // Internal edges are not applied on the last column of the image so + // we mask 1 more for the internal edges + const uint16_t mask_uv_int = ((1 << (columns >> 1)) - 1) * 0x1111; + + // Remove the bits outside the image edge. + for (i = 0; i < TX_32X32; i++) { + lfm->left_y[i] &= mask_y; + lfm->above_y[i] &= mask_y; + lfm->left_uv[i] &= mask_uv; + lfm->above_uv[i] &= mask_uv; + } + lfm->int_4x4_y &= mask_y; + lfm->int_4x4_uv &= mask_uv_int; + + // We don't apply a wide loop filter on the last uv column. If set + // apply the shorter one instead. + if (columns == 1) { + lfm->left_uv[TX_8X8] |= lfm->left_uv[TX_16X16]; + lfm->left_uv[TX_16X16] = 0; + } + if (columns == 5) { + lfm->left_uv[TX_8X8] |= (lfm->left_uv[TX_16X16] & 0xcccc); + lfm->left_uv[TX_16X16] &= ~(lfm->left_uv[TX_16X16] & 0xcccc); + } + } + // We don't apply a loop filter on the first column in the image, mask that + // out. + if (mi_col == 0) { + for (i = 0; i < TX_32X32; i++) { + lfm->left_y[i] &= 0xfefefefefefefefeULL; + lfm->left_uv[i] &= 0xeeee; + } + } + + // Assert if we try to apply 2 different loop filters at the same position. + assert(!(lfm->left_y[TX_16X16] & lfm->left_y[TX_8X8])); + assert(!(lfm->left_y[TX_16X16] & lfm->left_y[TX_4X4])); + assert(!(lfm->left_y[TX_8X8] & lfm->left_y[TX_4X4])); + assert(!(lfm->int_4x4_y & lfm->left_y[TX_16X16])); + assert(!(lfm->left_uv[TX_16X16]&lfm->left_uv[TX_8X8])); + assert(!(lfm->left_uv[TX_16X16] & lfm->left_uv[TX_4X4])); + assert(!(lfm->left_uv[TX_8X8] & lfm->left_uv[TX_4X4])); + assert(!(lfm->int_4x4_uv & lfm->left_uv[TX_16X16])); + assert(!(lfm->above_y[TX_16X16] & lfm->above_y[TX_8X8])); + assert(!(lfm->above_y[TX_16X16] & lfm->above_y[TX_4X4])); + assert(!(lfm->above_y[TX_8X8] & lfm->above_y[TX_4X4])); + assert(!(lfm->int_4x4_y & lfm->above_y[TX_16X16])); + assert(!(lfm->above_uv[TX_16X16] & lfm->above_uv[TX_8X8])); + assert(!(lfm->above_uv[TX_16X16] & lfm->above_uv[TX_4X4])); + assert(!(lfm->above_uv[TX_8X8] & lfm->above_uv[TX_4X4])); + assert(!(lfm->int_4x4_uv & lfm->above_uv[TX_16X16])); +} + +// This function sets up the bit masks for the entire 64x64 region represented +// by mi_row, mi_col. +void vp9_setup_mask(VP9_COMMON *const cm, const int mi_row, const int mi_col, + MODE_INFO **mi, const int mode_info_stride, + LOOP_FILTER_MASK *lfm) { + int idx_32, idx_16, idx_8; + const loop_filter_info_n *const lfi_n = &cm->lf_info; + MODE_INFO **mip = mi; + MODE_INFO **mip2 = mi; + + // These are offsets to the next mi in the 64x64 block. It is what gets + // added to the mi ptr as we go through each loop. It helps us to avoid + // setting up special row and column counters for each index. The last step + // brings us out back to the starting position. + const int offset_32[] = {4, (mode_info_stride << 2) - 4, 4, + -(mode_info_stride << 2) - 4}; + const int offset_16[] = {2, (mode_info_stride << 1) - 2, 2, + -(mode_info_stride << 1) - 2}; + const int offset[] = {1, mode_info_stride - 1, 1, -mode_info_stride - 1}; + + // Following variables represent shifts to position the current block + // mask over the appropriate block. A shift of 36 to the left will move + // the bits for the final 32 by 32 block in the 64x64 up 4 rows and left + // 4 rows to the appropriate spot. + const int shift_32_y[] = {0, 4, 32, 36}; + const int shift_16_y[] = {0, 2, 16, 18}; + const int shift_8_y[] = {0, 1, 8, 9}; + const int shift_32_uv[] = {0, 2, 8, 10}; + const int shift_16_uv[] = {0, 1, 4, 5}; + const int max_rows = (mi_row + MI_BLOCK_SIZE > cm->mi_rows ? + cm->mi_rows - mi_row : MI_BLOCK_SIZE); + const int max_cols = (mi_col + MI_BLOCK_SIZE > cm->mi_cols ? + cm->mi_cols - mi_col : MI_BLOCK_SIZE); + + vp9_zero(*lfm); + assert(mip[0] != NULL); + + switch (mip[0]->sb_type) { + case BLOCK_64X64: + build_masks(lfi_n, mip[0] , 0, 0, lfm); + break; + case BLOCK_64X32: + build_masks(lfi_n, mip[0], 0, 0, lfm); + mip2 = mip + mode_info_stride * 4; + if (4 >= max_rows) + break; + build_masks(lfi_n, mip2[0], 32, 8, lfm); + break; + case BLOCK_32X64: + build_masks(lfi_n, mip[0], 0, 0, lfm); + mip2 = mip + 4; + if (4 >= max_cols) + break; + build_masks(lfi_n, mip2[0], 4, 2, lfm); + break; + default: + for (idx_32 = 0; idx_32 < 4; mip += offset_32[idx_32], ++idx_32) { + const int shift_y = shift_32_y[idx_32]; + const int shift_uv = shift_32_uv[idx_32]; + const int mi_32_col_offset = ((idx_32 & 1) << 2); + const int mi_32_row_offset = ((idx_32 >> 1) << 2); + if (mi_32_col_offset >= max_cols || mi_32_row_offset >= max_rows) + continue; + switch (mip[0]->sb_type) { + case BLOCK_32X32: + build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm); + break; + case BLOCK_32X16: + build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm); + if (mi_32_row_offset + 2 >= max_rows) + continue; + mip2 = mip + mode_info_stride * 2; + build_masks(lfi_n, mip2[0], shift_y + 16, shift_uv + 4, lfm); + break; + case BLOCK_16X32: + build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm); + if (mi_32_col_offset + 2 >= max_cols) + continue; + mip2 = mip + 2; + build_masks(lfi_n, mip2[0], shift_y + 2, shift_uv + 1, lfm); + break; + default: + for (idx_16 = 0; idx_16 < 4; mip += offset_16[idx_16], ++idx_16) { + const int shift_y = shift_32_y[idx_32] + shift_16_y[idx_16]; + const int shift_uv = shift_32_uv[idx_32] + shift_16_uv[idx_16]; + const int mi_16_col_offset = mi_32_col_offset + + ((idx_16 & 1) << 1); + const int mi_16_row_offset = mi_32_row_offset + + ((idx_16 >> 1) << 1); + + if (mi_16_col_offset >= max_cols || mi_16_row_offset >= max_rows) + continue; + + switch (mip[0]->sb_type) { + case BLOCK_16X16: + build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm); + break; + case BLOCK_16X8: + build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm); + if (mi_16_row_offset + 1 >= max_rows) + continue; + mip2 = mip + mode_info_stride; + build_y_mask(lfi_n, mip2[0], shift_y+8, lfm); + break; + case BLOCK_8X16: + build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm); + if (mi_16_col_offset +1 >= max_cols) + continue; + mip2 = mip + 1; + build_y_mask(lfi_n, mip2[0], shift_y+1, lfm); + break; + default: { + const int shift_y = shift_32_y[idx_32] + + shift_16_y[idx_16] + + shift_8_y[0]; + build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm); + mip += offset[0]; + for (idx_8 = 1; idx_8 < 4; mip += offset[idx_8], ++idx_8) { + const int shift_y = shift_32_y[idx_32] + + shift_16_y[idx_16] + + shift_8_y[idx_8]; + const int mi_8_col_offset = mi_16_col_offset + + ((idx_8 & 1)); + const int mi_8_row_offset = mi_16_row_offset + + ((idx_8 >> 1)); + + if (mi_8_col_offset >= max_cols || + mi_8_row_offset >= max_rows) + continue; + build_y_mask(lfi_n, mip[0], shift_y, lfm); + } + break; + } + } + } + break; + } + } + break; + } +} + +static void filter_selectively_vert(uint8_t *s, int pitch, + unsigned int mask_16x16, + unsigned int mask_8x8, + unsigned int mask_4x4, + unsigned int mask_4x4_int, + const loop_filter_thresh *lfthr, + const uint8_t *lfl) { + unsigned int mask; + + for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int; + mask; mask >>= 1) { + const loop_filter_thresh *lfi = lfthr + *lfl; + + if (mask & 1) { + if (mask_16x16 & 1) { + vpx_lpf_vertical_16(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr); + } else if (mask_8x8 & 1) { + vpx_lpf_vertical_8(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr); + } else if (mask_4x4 & 1) { + vpx_lpf_vertical_4(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr); + } + } + if (mask_4x4_int & 1) + vpx_lpf_vertical_4(s + 4, pitch, lfi->mblim, lfi->lim, lfi->hev_thr); + s += 8; + lfl += 1; + mask_16x16 >>= 1; + mask_8x8 >>= 1; + mask_4x4 >>= 1; + mask_4x4_int >>= 1; + } +} + +#if CONFIG_VP9_HIGHBITDEPTH +static void highbd_filter_selectively_vert(uint16_t *s, int pitch, + unsigned int mask_16x16, + unsigned int mask_8x8, + unsigned int mask_4x4, + unsigned int mask_4x4_int, + const loop_filter_thresh *lfthr, + const uint8_t *lfl, int bd) { + unsigned int mask; + + for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int; + mask; mask >>= 1) { + const loop_filter_thresh *lfi = lfthr + *lfl; + + if (mask & 1) { + if (mask_16x16 & 1) { + vpx_highbd_lpf_vertical_16(s, pitch, lfi->mblim, lfi->lim, + lfi->hev_thr, bd); + } else if (mask_8x8 & 1) { + vpx_highbd_lpf_vertical_8(s, pitch, lfi->mblim, lfi->lim, + lfi->hev_thr, bd); + } else if (mask_4x4 & 1) { + vpx_highbd_lpf_vertical_4(s, pitch, lfi->mblim, lfi->lim, + lfi->hev_thr, bd); + } + } + if (mask_4x4_int & 1) + vpx_highbd_lpf_vertical_4(s + 4, pitch, lfi->mblim, lfi->lim, + lfi->hev_thr, bd); + s += 8; + lfl += 1; + mask_16x16 >>= 1; + mask_8x8 >>= 1; + mask_4x4 >>= 1; + mask_4x4_int >>= 1; + } +} +#endif // CONFIG_VP9_HIGHBITDEPTH + +void vp9_filter_block_plane_non420(VP9_COMMON *cm, + struct macroblockd_plane *plane, + MODE_INFO **mi_8x8, + int mi_row, int mi_col) { + const int ss_x = plane->subsampling_x; + const int ss_y = plane->subsampling_y; + const int row_step = 1 << ss_y; + const int col_step = 1 << ss_x; + const int row_step_stride = cm->mi_stride * row_step; + struct buf_2d *const dst = &plane->dst; + uint8_t* const dst0 = dst->buf; + unsigned int mask_16x16[MI_BLOCK_SIZE] = {0}; + unsigned int mask_8x8[MI_BLOCK_SIZE] = {0}; + unsigned int mask_4x4[MI_BLOCK_SIZE] = {0}; + unsigned int mask_4x4_int[MI_BLOCK_SIZE] = {0}; + uint8_t lfl[MI_BLOCK_SIZE * MI_BLOCK_SIZE]; + int r, c; + + for (r = 0; r < MI_BLOCK_SIZE && mi_row + r < cm->mi_rows; r += row_step) { + unsigned int mask_16x16_c = 0; + unsigned int mask_8x8_c = 0; + unsigned int mask_4x4_c = 0; + unsigned int border_mask; + + // Determine the vertical edges that need filtering + for (c = 0; c < MI_BLOCK_SIZE && mi_col + c < cm->mi_cols; c += col_step) { + const MODE_INFO *mi = mi_8x8[c]; + const BLOCK_SIZE sb_type = mi[0].sb_type; + const int skip_this = mi[0].skip && is_inter_block(mi); + // left edge of current unit is block/partition edge -> no skip + const int block_edge_left = (num_4x4_blocks_wide_lookup[sb_type] > 1) ? + !(c & (num_8x8_blocks_wide_lookup[sb_type] - 1)) : 1; + const int skip_this_c = skip_this && !block_edge_left; + // top edge of current unit is block/partition edge -> no skip + const int block_edge_above = (num_4x4_blocks_high_lookup[sb_type] > 1) ? + !(r & (num_8x8_blocks_high_lookup[sb_type] - 1)) : 1; + const int skip_this_r = skip_this && !block_edge_above; + const TX_SIZE tx_size = get_uv_tx_size(mi, plane); + const int skip_border_4x4_c = ss_x && mi_col + c == cm->mi_cols - 1; + const int skip_border_4x4_r = ss_y && mi_row + r == cm->mi_rows - 1; + + // Filter level can vary per MI + if (!(lfl[(r << 3) + (c >> ss_x)] = + get_filter_level(&cm->lf_info, mi))) + continue; + + // Build masks based on the transform size of each block + if (tx_size == TX_32X32) { + if (!skip_this_c && ((c >> ss_x) & 3) == 0) { + if (!skip_border_4x4_c) + mask_16x16_c |= 1 << (c >> ss_x); + else + mask_8x8_c |= 1 << (c >> ss_x); + } + if (!skip_this_r && ((r >> ss_y) & 3) == 0) { + if (!skip_border_4x4_r) + mask_16x16[r] |= 1 << (c >> ss_x); + else + mask_8x8[r] |= 1 << (c >> ss_x); + } + } else if (tx_size == TX_16X16) { + if (!skip_this_c && ((c >> ss_x) & 1) == 0) { + if (!skip_border_4x4_c) + mask_16x16_c |= 1 << (c >> ss_x); + else + mask_8x8_c |= 1 << (c >> ss_x); + } + if (!skip_this_r && ((r >> ss_y) & 1) == 0) { + if (!skip_border_4x4_r) + mask_16x16[r] |= 1 << (c >> ss_x); + else + mask_8x8[r] |= 1 << (c >> ss_x); + } + } else { + // force 8x8 filtering on 32x32 boundaries + if (!skip_this_c) { + if (tx_size == TX_8X8 || ((c >> ss_x) & 3) == 0) + mask_8x8_c |= 1 << (c >> ss_x); + else + mask_4x4_c |= 1 << (c >> ss_x); + } + + if (!skip_this_r) { + if (tx_size == TX_8X8 || ((r >> ss_y) & 3) == 0) + mask_8x8[r] |= 1 << (c >> ss_x); + else + mask_4x4[r] |= 1 << (c >> ss_x); + } + + if (!skip_this && tx_size < TX_8X8 && !skip_border_4x4_c) + mask_4x4_int[r] |= 1 << (c >> ss_x); + } + } + + // Disable filtering on the leftmost column + border_mask = ~(mi_col == 0); +#if CONFIG_VP9_HIGHBITDEPTH + if (cm->use_highbitdepth) { + highbd_filter_selectively_vert(CONVERT_TO_SHORTPTR(dst->buf), + dst->stride, + mask_16x16_c & border_mask, + mask_8x8_c & border_mask, + mask_4x4_c & border_mask, + mask_4x4_int[r], + cm->lf_info.lfthr, &lfl[r << 3], + (int)cm->bit_depth); + } else { +#endif // CONFIG_VP9_HIGHBITDEPTH + filter_selectively_vert(dst->buf, dst->stride, + mask_16x16_c & border_mask, + mask_8x8_c & border_mask, + mask_4x4_c & border_mask, + mask_4x4_int[r], + cm->lf_info.lfthr, &lfl[r << 3]); +#if CONFIG_VP9_HIGHBITDEPTH + } +#endif // CONFIG_VP9_HIGHBITDEPTH + dst->buf += 8 * dst->stride; + mi_8x8 += row_step_stride; + } + + // Now do horizontal pass + dst->buf = dst0; + for (r = 0; r < MI_BLOCK_SIZE && mi_row + r < cm->mi_rows; r += row_step) { + const int skip_border_4x4_r = ss_y && mi_row + r == cm->mi_rows - 1; + const unsigned int mask_4x4_int_r = skip_border_4x4_r ? 0 : mask_4x4_int[r]; + + unsigned int mask_16x16_r; + unsigned int mask_8x8_r; + unsigned int mask_4x4_r; + + if (mi_row + r == 0) { + mask_16x16_r = 0; + mask_8x8_r = 0; + mask_4x4_r = 0; + } else { + mask_16x16_r = mask_16x16[r]; + mask_8x8_r = mask_8x8[r]; + mask_4x4_r = mask_4x4[r]; + } +#if CONFIG_VP9_HIGHBITDEPTH + if (cm->use_highbitdepth) { + highbd_filter_selectively_horiz(CONVERT_TO_SHORTPTR(dst->buf), + dst->stride, + mask_16x16_r, + mask_8x8_r, + mask_4x4_r, + mask_4x4_int_r, + cm->lf_info.lfthr, &lfl[r << 3], + (int)cm->bit_depth); + } else { +#endif // CONFIG_VP9_HIGHBITDEPTH + filter_selectively_horiz(dst->buf, dst->stride, + mask_16x16_r, + mask_8x8_r, + mask_4x4_r, + mask_4x4_int_r, + cm->lf_info.lfthr, &lfl[r << 3]); +#if CONFIG_VP9_HIGHBITDEPTH + } +#endif // CONFIG_VP9_HIGHBITDEPTH + dst->buf += 8 * dst->stride; + } +} + +void vp9_filter_block_plane_ss00(VP9_COMMON *const cm, + struct macroblockd_plane *const plane, + int mi_row, + LOOP_FILTER_MASK *lfm) { + struct buf_2d *const dst = &plane->dst; + uint8_t *const dst0 = dst->buf; + int r; + uint64_t mask_16x16 = lfm->left_y[TX_16X16]; + uint64_t mask_8x8 = lfm->left_y[TX_8X8]; + uint64_t mask_4x4 = lfm->left_y[TX_4X4]; + uint64_t mask_4x4_int = lfm->int_4x4_y; + + assert(plane->subsampling_x == 0 && plane->subsampling_y == 0); + + // Vertical pass: do 2 rows at one time + for (r = 0; r < MI_BLOCK_SIZE && mi_row + r < cm->mi_rows; r += 2) { + // Disable filtering on the leftmost column. +#if CONFIG_VP9_HIGHBITDEPTH + if (cm->use_highbitdepth) { + highbd_filter_selectively_vert_row2(plane->subsampling_x, + CONVERT_TO_SHORTPTR(dst->buf), + dst->stride, + (unsigned int)mask_16x16, + (unsigned int)mask_8x8, + (unsigned int)mask_4x4, + (unsigned int)mask_4x4_int, + cm->lf_info.lfthr, + &lfm->lfl_y[r << 3], + (int)cm->bit_depth); + } else { +#endif // CONFIG_VP9_HIGHBITDEPTH + filter_selectively_vert_row2(plane->subsampling_x, dst->buf, dst->stride, + (unsigned int)mask_16x16, + (unsigned int)mask_8x8, + (unsigned int)mask_4x4, + (unsigned int)mask_4x4_int, + cm->lf_info.lfthr, &lfm->lfl_y[r << 3]); +#if CONFIG_VP9_HIGHBITDEPTH + } +#endif // CONFIG_VP9_HIGHBITDEPTH + dst->buf += 16 * dst->stride; + mask_16x16 >>= 16; + mask_8x8 >>= 16; + mask_4x4 >>= 16; + mask_4x4_int >>= 16; + } + + // Horizontal pass + dst->buf = dst0; + mask_16x16 = lfm->above_y[TX_16X16]; + mask_8x8 = lfm->above_y[TX_8X8]; + mask_4x4 = lfm->above_y[TX_4X4]; + mask_4x4_int = lfm->int_4x4_y; + + for (r = 0; r < MI_BLOCK_SIZE && mi_row + r < cm->mi_rows; r++) { + unsigned int mask_16x16_r; + unsigned int mask_8x8_r; + unsigned int mask_4x4_r; + + if (mi_row + r == 0) { + mask_16x16_r = 0; + mask_8x8_r = 0; + mask_4x4_r = 0; + } else { + mask_16x16_r = mask_16x16 & 0xff; + mask_8x8_r = mask_8x8 & 0xff; + mask_4x4_r = mask_4x4 & 0xff; + } + +#if CONFIG_VP9_HIGHBITDEPTH + if (cm->use_highbitdepth) { + highbd_filter_selectively_horiz(CONVERT_TO_SHORTPTR(dst->buf), + dst->stride, mask_16x16_r, mask_8x8_r, + mask_4x4_r, mask_4x4_int & 0xff, + cm->lf_info.lfthr, &lfm->lfl_y[r << 3], + (int)cm->bit_depth); + } else { +#endif // CONFIG_VP9_HIGHBITDEPTH + filter_selectively_horiz(dst->buf, dst->stride, mask_16x16_r, mask_8x8_r, + mask_4x4_r, mask_4x4_int & 0xff, + cm->lf_info.lfthr, &lfm->lfl_y[r << 3]); +#if CONFIG_VP9_HIGHBITDEPTH + } +#endif // CONFIG_VP9_HIGHBITDEPTH + + dst->buf += 8 * dst->stride; + mask_16x16 >>= 8; + mask_8x8 >>= 8; + mask_4x4 >>= 8; + mask_4x4_int >>= 8; + } +} + +void vp9_filter_block_plane_ss11(VP9_COMMON *const cm, + struct macroblockd_plane *const plane, + int mi_row, + LOOP_FILTER_MASK *lfm) { + struct buf_2d *const dst = &plane->dst; + uint8_t *const dst0 = dst->buf; + int r, c; + uint8_t lfl_uv[16]; + + uint16_t mask_16x16 = lfm->left_uv[TX_16X16]; + uint16_t mask_8x8 = lfm->left_uv[TX_8X8]; + uint16_t mask_4x4 = lfm->left_uv[TX_4X4]; + uint16_t mask_4x4_int = lfm->int_4x4_uv; + + assert(plane->subsampling_x == 1 && plane->subsampling_y == 1); + + // Vertical pass: do 2 rows at one time + for (r = 0; r < MI_BLOCK_SIZE && mi_row + r < cm->mi_rows; r += 4) { + for (c = 0; c < (MI_BLOCK_SIZE >> 1); c++) { + lfl_uv[(r << 1) + c] = lfm->lfl_y[(r << 3) + (c << 1)]; + lfl_uv[((r + 2) << 1) + c] = lfm->lfl_y[((r + 2) << 3) + (c << 1)]; + } + + // Disable filtering on the leftmost column. +#if CONFIG_VP9_HIGHBITDEPTH + if (cm->use_highbitdepth) { + highbd_filter_selectively_vert_row2(plane->subsampling_x, + CONVERT_TO_SHORTPTR(dst->buf), + dst->stride, + (unsigned int)mask_16x16, + (unsigned int)mask_8x8, + (unsigned int)mask_4x4, + (unsigned int)mask_4x4_int, + cm->lf_info.lfthr, &lfl_uv[r << 1], + (int)cm->bit_depth); + } else { +#endif // CONFIG_VP9_HIGHBITDEPTH + filter_selectively_vert_row2(plane->subsampling_x, dst->buf, dst->stride, + (unsigned int)mask_16x16, + (unsigned int)mask_8x8, + (unsigned int)mask_4x4, + (unsigned int)mask_4x4_int, + cm->lf_info.lfthr, &lfl_uv[r << 1]); +#if CONFIG_VP9_HIGHBITDEPTH + } +#endif // CONFIG_VP9_HIGHBITDEPTH + + dst->buf += 16 * dst->stride; + mask_16x16 >>= 8; + mask_8x8 >>= 8; + mask_4x4 >>= 8; + mask_4x4_int >>= 8; + } + + // Horizontal pass + dst->buf = dst0; + mask_16x16 = lfm->above_uv[TX_16X16]; + mask_8x8 = lfm->above_uv[TX_8X8]; + mask_4x4 = lfm->above_uv[TX_4X4]; + mask_4x4_int = lfm->int_4x4_uv; + + for (r = 0; r < MI_BLOCK_SIZE && mi_row + r < cm->mi_rows; r += 2) { + const int skip_border_4x4_r = mi_row + r == cm->mi_rows - 1; + const unsigned int mask_4x4_int_r = + skip_border_4x4_r ? 0 : (mask_4x4_int & 0xf); + unsigned int mask_16x16_r; + unsigned int mask_8x8_r; + unsigned int mask_4x4_r; + + if (mi_row + r == 0) { + mask_16x16_r = 0; + mask_8x8_r = 0; + mask_4x4_r = 0; + } else { + mask_16x16_r = mask_16x16 & 0xf; + mask_8x8_r = mask_8x8 & 0xf; + mask_4x4_r = mask_4x4 & 0xf; + } + +#if CONFIG_VP9_HIGHBITDEPTH + if (cm->use_highbitdepth) { + highbd_filter_selectively_horiz(CONVERT_TO_SHORTPTR(dst->buf), + dst->stride, mask_16x16_r, mask_8x8_r, + mask_4x4_r, mask_4x4_int_r, + cm->lf_info.lfthr, &lfl_uv[r << 1], + (int)cm->bit_depth); + } else { +#endif // CONFIG_VP9_HIGHBITDEPTH + filter_selectively_horiz(dst->buf, dst->stride, mask_16x16_r, mask_8x8_r, + mask_4x4_r, mask_4x4_int_r, cm->lf_info.lfthr, + &lfl_uv[r << 1]); +#if CONFIG_VP9_HIGHBITDEPTH + } +#endif // CONFIG_VP9_HIGHBITDEPTH + + dst->buf += 8 * dst->stride; + mask_16x16 >>= 4; + mask_8x8 >>= 4; + mask_4x4 >>= 4; + mask_4x4_int >>= 4; + } +} + +static void loop_filter_rows(YV12_BUFFER_CONFIG *frame_buffer, VP9_COMMON *cm, + struct macroblockd_plane planes[MAX_MB_PLANE], + int start, int stop, int y_only) { + const int num_planes = y_only ? 1 : MAX_MB_PLANE; + enum lf_path path; + int mi_row, mi_col; + + if (y_only) + path = LF_PATH_444; + else if (planes[1].subsampling_y == 1 && planes[1].subsampling_x == 1) + path = LF_PATH_420; + else if (planes[1].subsampling_y == 0 && planes[1].subsampling_x == 0) + path = LF_PATH_444; + else + path = LF_PATH_SLOW; + + for (mi_row = start; mi_row < stop; mi_row += MI_BLOCK_SIZE) { + MODE_INFO **mi = cm->mi_grid_visible + mi_row * cm->mi_stride; + LOOP_FILTER_MASK *lfm = get_lfm(&cm->lf, mi_row, 0); + + for (mi_col = 0; mi_col < cm->mi_cols; mi_col += MI_BLOCK_SIZE, ++lfm) { + int plane; + + vp9_setup_dst_planes(planes, frame_buffer, mi_row, mi_col); + + // TODO(jimbankoski): For 444 only need to do y mask. + vp9_adjust_mask(cm, mi_row, mi_col, lfm); + + vp9_filter_block_plane_ss00(cm, &planes[0], mi_row, lfm); + for (plane = 1; plane < num_planes; ++plane) { + switch (path) { + case LF_PATH_420: + vp9_filter_block_plane_ss11(cm, &planes[plane], mi_row, lfm); + break; + case LF_PATH_444: + vp9_filter_block_plane_ss00(cm, &planes[plane], mi_row, lfm); + break; + case LF_PATH_SLOW: + vp9_filter_block_plane_non420(cm, &planes[plane], mi + mi_col, + mi_row, mi_col); + break; + } + } + } + } +} + +void vp9_loop_filter_frame(YV12_BUFFER_CONFIG *frame, + VP9_COMMON *cm, MACROBLOCKD *xd, + int frame_filter_level, + int y_only, int partial_frame) { + int start_mi_row, end_mi_row, mi_rows_to_filter; + if (!frame_filter_level) return; + start_mi_row = 0; + mi_rows_to_filter = cm->mi_rows; + if (partial_frame && cm->mi_rows > 8) { + start_mi_row = cm->mi_rows >> 1; + start_mi_row &= 0xfffffff8; + mi_rows_to_filter = VPXMAX(cm->mi_rows / 8, 8); + } + end_mi_row = start_mi_row + mi_rows_to_filter; + loop_filter_rows(frame, cm, xd->plane, start_mi_row, end_mi_row, y_only); +} + +// Used by the encoder to build the loopfilter masks. +// TODO(slavarnway): Do the encoder the same way the decoder does it and +// build the masks in line as part of the encode process. +void vp9_build_mask_frame(VP9_COMMON *cm, int frame_filter_level, + int partial_frame) { + int start_mi_row, end_mi_row, mi_rows_to_filter; + int mi_col, mi_row; + if (!frame_filter_level) return; + start_mi_row = 0; + mi_rows_to_filter = cm->mi_rows; + if (partial_frame && cm->mi_rows > 8) { + start_mi_row = cm->mi_rows >> 1; + start_mi_row &= 0xfffffff8; + mi_rows_to_filter = VPXMAX(cm->mi_rows / 8, 8); + } + end_mi_row = start_mi_row + mi_rows_to_filter; + + vp9_loop_filter_frame_init(cm, frame_filter_level); + + for (mi_row = start_mi_row; mi_row < end_mi_row; mi_row += MI_BLOCK_SIZE) { + MODE_INFO **mi = cm->mi_grid_visible + mi_row * cm->mi_stride; + for (mi_col = 0; mi_col < cm->mi_cols; mi_col += MI_BLOCK_SIZE) { + // vp9_setup_mask() zeros lfm + vp9_setup_mask(cm, mi_row, mi_col, mi + mi_col, cm->mi_stride, + get_lfm(&cm->lf, mi_row, mi_col)); + } + } +} + +// 8x8 blocks in a superblock. A "1" represents the first block in a 16x16 +// or greater area. +static const uint8_t first_block_in_16x16[8][8] = { + {1, 0, 1, 0, 1, 0, 1, 0}, + {0, 0, 0, 0, 0, 0, 0, 0}, + {1, 0, 1, 0, 1, 0, 1, 0}, + {0, 0, 0, 0, 0, 0, 0, 0}, + {1, 0, 1, 0, 1, 0, 1, 0}, + {0, 0, 0, 0, 0, 0, 0, 0}, + {1, 0, 1, 0, 1, 0, 1, 0}, + {0, 0, 0, 0, 0, 0, 0, 0} +}; + +// This function sets up the bit masks for a block represented +// by mi_row, mi_col in a 64x64 region. +// TODO(SJL): This function only works for yv12. +void vp9_build_mask(VP9_COMMON *cm, const MODE_INFO *mi, int mi_row, + int mi_col, int bw, int bh) { + const BLOCK_SIZE block_size = mi->sb_type; + const TX_SIZE tx_size_y = mi->tx_size; + const loop_filter_info_n *const lfi_n = &cm->lf_info; + const int filter_level = get_filter_level(lfi_n, mi); + const TX_SIZE tx_size_uv = get_uv_tx_size_impl(tx_size_y, block_size, 1, 1); + LOOP_FILTER_MASK *const lfm = get_lfm(&cm->lf, mi_row, mi_col); + uint64_t *const left_y = &lfm->left_y[tx_size_y]; + uint64_t *const above_y = &lfm->above_y[tx_size_y]; + uint64_t *const int_4x4_y = &lfm->int_4x4_y; + uint16_t *const left_uv = &lfm->left_uv[tx_size_uv]; + uint16_t *const above_uv = &lfm->above_uv[tx_size_uv]; + uint16_t *const int_4x4_uv = &lfm->int_4x4_uv; + const int row_in_sb = (mi_row & 7); + const int col_in_sb = (mi_col & 7); + const int shift_y = col_in_sb + (row_in_sb << 3); + const int shift_uv = (col_in_sb >> 1) + ((row_in_sb >> 1) << 2); + const int build_uv = first_block_in_16x16[row_in_sb][col_in_sb]; + + if (!filter_level) { + return; + } else { + int index = shift_y; + int i; + for (i = 0; i < bh; i++) { + memset(&lfm->lfl_y[index], filter_level, bw); + index += 8; + } + } + + // These set 1 in the current block size for the block size edges. + // For instance if the block size is 32x16, we'll set: + // above = 1111 + // 0000 + // and + // left = 1000 + // = 1000 + // NOTE : In this example the low bit is left most ( 1000 ) is stored as + // 1, not 8... + // + // U and V set things on a 16 bit scale. + // + *above_y |= above_prediction_mask[block_size] << shift_y; + *left_y |= left_prediction_mask[block_size] << shift_y; + + if (build_uv) { + *above_uv |= above_prediction_mask_uv[block_size] << shift_uv; + *left_uv |= left_prediction_mask_uv[block_size] << shift_uv; + } + + // If the block has no coefficients and is not intra we skip applying + // the loop filter on block edges. + if (mi->skip && is_inter_block(mi)) + return; + + // Add a mask for the transform size. The transform size mask is set to + // be correct for a 64x64 prediction block size. Mask to match the size of + // the block we are working on and then shift it into place. + *above_y |= (size_mask[block_size] & + above_64x64_txform_mask[tx_size_y]) << shift_y; + *left_y |= (size_mask[block_size] & + left_64x64_txform_mask[tx_size_y]) << shift_y; + + if (build_uv) { + *above_uv |= (size_mask_uv[block_size] & + above_64x64_txform_mask_uv[tx_size_uv]) << shift_uv; + + *left_uv |= (size_mask_uv[block_size] & + left_64x64_txform_mask_uv[tx_size_uv]) << shift_uv; + } + + // Try to determine what to do with the internal 4x4 block boundaries. These + // differ from the 4x4 boundaries on the outside edge of an 8x8 in that the + // internal ones can be skipped and don't depend on the prediction block size. + if (tx_size_y == TX_4X4) + *int_4x4_y |= size_mask[block_size] << shift_y; + + if (build_uv && tx_size_uv == TX_4X4) + *int_4x4_uv |= (size_mask_uv[block_size] & 0xffff) << shift_uv; +} + +void vp9_loop_filter_data_reset( + LFWorkerData *lf_data, YV12_BUFFER_CONFIG *frame_buffer, + struct VP9Common *cm, const struct macroblockd_plane planes[MAX_MB_PLANE]) { + lf_data->frame_buffer = frame_buffer; + lf_data->cm = cm; + lf_data->start = 0; + lf_data->stop = 0; + lf_data->y_only = 0; + memcpy(lf_data->planes, planes, sizeof(lf_data->planes)); +} + +void vp9_reset_lfm(VP9_COMMON *const cm) { + if (cm->lf.filter_level) { + memset(cm->lf.lfm, 0, + ((cm->mi_rows + (MI_BLOCK_SIZE - 1)) >> 3) * cm->lf.lfm_stride * + sizeof(*cm->lf.lfm)); + } +} + +int vp9_loop_filter_worker(LFWorkerData *const lf_data, void *unused) { + (void)unused; + loop_filter_rows(lf_data->frame_buffer, lf_data->cm, lf_data->planes, + lf_data->start, lf_data->stop, lf_data->y_only); + return 1; +} diff --git a/thirdparty/libvpx/vp9/common/vp9_loopfilter.h b/thirdparty/libvpx/vp9/common/vp9_loopfilter.h new file mode 100644 index 0000000000..fca8830fa1 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_loopfilter.h @@ -0,0 +1,166 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_COMMON_VP9_LOOPFILTER_H_ +#define VP9_COMMON_VP9_LOOPFILTER_H_ + +#include "vpx_ports/mem.h" +#include "./vpx_config.h" + +#include "vp9/common/vp9_blockd.h" +#include "vp9/common/vp9_seg_common.h" + +#ifdef __cplusplus +extern "C" { +#endif + +#define MAX_LOOP_FILTER 63 +#define MAX_SHARPNESS 7 + +#define SIMD_WIDTH 16 + +#define MAX_REF_LF_DELTAS 4 +#define MAX_MODE_LF_DELTAS 2 + +enum lf_path { + LF_PATH_420, + LF_PATH_444, + LF_PATH_SLOW, +}; + +// Need to align this structure so when it is declared and +// passed it can be loaded into vector registers. +typedef struct { + DECLARE_ALIGNED(SIMD_WIDTH, uint8_t, mblim[SIMD_WIDTH]); + DECLARE_ALIGNED(SIMD_WIDTH, uint8_t, lim[SIMD_WIDTH]); + DECLARE_ALIGNED(SIMD_WIDTH, uint8_t, hev_thr[SIMD_WIDTH]); +} loop_filter_thresh; + +typedef struct { + loop_filter_thresh lfthr[MAX_LOOP_FILTER + 1]; + uint8_t lvl[MAX_SEGMENTS][MAX_REF_FRAMES][MAX_MODE_LF_DELTAS]; +} loop_filter_info_n; + +// This structure holds bit masks for all 8x8 blocks in a 64x64 region. +// Each 1 bit represents a position in which we want to apply the loop filter. +// Left_ entries refer to whether we apply a filter on the border to the +// left of the block. Above_ entries refer to whether or not to apply a +// filter on the above border. Int_ entries refer to whether or not to +// apply borders on the 4x4 edges within the 8x8 block that each bit +// represents. +// Since each transform is accompanied by a potentially different type of +// loop filter there is a different entry in the array for each transform size. +typedef struct { + uint64_t left_y[TX_SIZES]; + uint64_t above_y[TX_SIZES]; + uint64_t int_4x4_y; + uint16_t left_uv[TX_SIZES]; + uint16_t above_uv[TX_SIZES]; + uint16_t int_4x4_uv; + uint8_t lfl_y[64]; +} LOOP_FILTER_MASK; + +struct loopfilter { + int filter_level; + int last_filt_level; + + int sharpness_level; + int last_sharpness_level; + + uint8_t mode_ref_delta_enabled; + uint8_t mode_ref_delta_update; + + // 0 = Intra, Last, GF, ARF + signed char ref_deltas[MAX_REF_LF_DELTAS]; + signed char last_ref_deltas[MAX_REF_LF_DELTAS]; + + // 0 = ZERO_MV, MV + signed char mode_deltas[MAX_MODE_LF_DELTAS]; + signed char last_mode_deltas[MAX_MODE_LF_DELTAS]; + + LOOP_FILTER_MASK *lfm; + int lfm_stride; +}; + +/* assorted loopfilter functions which get used elsewhere */ +struct VP9Common; +struct macroblockd; +struct VP9LfSyncData; + +// This function sets up the bit masks for the entire 64x64 region represented +// by mi_row, mi_col. +void vp9_setup_mask(struct VP9Common *const cm, + const int mi_row, const int mi_col, + MODE_INFO **mi_8x8, const int mode_info_stride, + LOOP_FILTER_MASK *lfm); + +void vp9_filter_block_plane_ss00(struct VP9Common *const cm, + struct macroblockd_plane *const plane, + int mi_row, + LOOP_FILTER_MASK *lfm); + +void vp9_filter_block_plane_ss11(struct VP9Common *const cm, + struct macroblockd_plane *const plane, + int mi_row, + LOOP_FILTER_MASK *lfm); + +void vp9_filter_block_plane_non420(struct VP9Common *cm, + struct macroblockd_plane *plane, + MODE_INFO **mi_8x8, + int mi_row, int mi_col); + +void vp9_loop_filter_init(struct VP9Common *cm); + +// Update the loop filter for the current frame. +// This should be called before vp9_loop_filter_frame(), vp9_build_mask_frame() +// calls this function directly. +void vp9_loop_filter_frame_init(struct VP9Common *cm, int default_filt_lvl); + +void vp9_loop_filter_frame(YV12_BUFFER_CONFIG *frame, + struct VP9Common *cm, + struct macroblockd *mbd, + int filter_level, + int y_only, int partial_frame); + +// Get the superblock lfm for a given mi_row, mi_col. +static INLINE LOOP_FILTER_MASK *get_lfm(const struct loopfilter *lf, + const int mi_row, const int mi_col) { + return &lf->lfm[(mi_col >> 3) + ((mi_row >> 3) * lf->lfm_stride)]; +} + +void vp9_build_mask(struct VP9Common *cm, const MODE_INFO *mi, int mi_row, + int mi_col, int bw, int bh); +void vp9_adjust_mask(struct VP9Common *const cm, const int mi_row, + const int mi_col, LOOP_FILTER_MASK *lfm); +void vp9_build_mask_frame(struct VP9Common *cm, int frame_filter_level, + int partial_frame); +void vp9_reset_lfm(struct VP9Common *const cm); + +typedef struct LoopFilterWorkerData { + YV12_BUFFER_CONFIG *frame_buffer; + struct VP9Common *cm; + struct macroblockd_plane planes[MAX_MB_PLANE]; + + int start; + int stop; + int y_only; +} LFWorkerData; + +void vp9_loop_filter_data_reset( + LFWorkerData *lf_data, YV12_BUFFER_CONFIG *frame_buffer, + struct VP9Common *cm, const struct macroblockd_plane planes[MAX_MB_PLANE]); + +// Operates on the rows described by 'lf_data'. +int vp9_loop_filter_worker(LFWorkerData *const lf_data, void *unused); +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_LOOPFILTER_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_mv.h b/thirdparty/libvpx/vp9/common/vp9_mv.h new file mode 100644 index 0000000000..5d89da8c25 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_mv.h @@ -0,0 +1,55 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_COMMON_VP9_MV_H_ +#define VP9_COMMON_VP9_MV_H_ + +#include "vpx/vpx_integer.h" + +#include "vp9/common/vp9_common.h" + +#ifdef __cplusplus +extern "C" { +#endif + +typedef struct mv { + int16_t row; + int16_t col; +} MV; + +typedef union int_mv { + uint32_t as_int; + MV as_mv; +} int_mv; /* facilitates faster equality tests and copies */ + +typedef struct mv32 { + int32_t row; + int32_t col; +} MV32; + +static INLINE int is_zero_mv(const MV *mv) { + return *((const uint32_t *)mv) == 0; +} + +static INLINE int is_equal_mv(const MV *a, const MV *b) { + return *((const uint32_t *)a) == *((const uint32_t *)b); +} + +static INLINE void clamp_mv(MV *mv, int min_col, int max_col, + int min_row, int max_row) { + mv->col = clamp(mv->col, min_col, max_col); + mv->row = clamp(mv->row, min_row, max_row); +} + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_MV_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_mvref_common.c b/thirdparty/libvpx/vp9/common/vp9_mvref_common.c new file mode 100644 index 0000000000..0eb01a51ba --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_mvref_common.c @@ -0,0 +1,201 @@ + +/* + * Copyright (c) 2012 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "vp9/common/vp9_mvref_common.h" + +// This function searches the neighborhood of a given MB/SB +// to try and find candidate reference vectors. +static void find_mv_refs_idx(const VP9_COMMON *cm, const MACROBLOCKD *xd, + MODE_INFO *mi, MV_REFERENCE_FRAME ref_frame, + int_mv *mv_ref_list, + int block, int mi_row, int mi_col, + uint8_t *mode_context) { + const int *ref_sign_bias = cm->ref_frame_sign_bias; + int i, refmv_count = 0; + const POSITION *const mv_ref_search = mv_ref_blocks[mi->sb_type]; + int different_ref_found = 0; + int context_counter = 0; + const MV_REF *const prev_frame_mvs = cm->use_prev_frame_mvs ? + cm->prev_frame->mvs + mi_row * cm->mi_cols + mi_col : NULL; + const TileInfo *const tile = &xd->tile; + + // Blank the reference vector list + memset(mv_ref_list, 0, sizeof(*mv_ref_list) * MAX_MV_REF_CANDIDATES); + + // The nearest 2 blocks are treated differently + // if the size < 8x8 we get the mv from the bmi substructure, + // and we also need to keep a mode count. + for (i = 0; i < 2; ++i) { + const POSITION *const mv_ref = &mv_ref_search[i]; + if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) { + const MODE_INFO *const candidate_mi = xd->mi[mv_ref->col + mv_ref->row * + xd->mi_stride]; + // Keep counts for entropy encoding. + context_counter += mode_2_counter[candidate_mi->mode]; + different_ref_found = 1; + + if (candidate_mi->ref_frame[0] == ref_frame) + ADD_MV_REF_LIST(get_sub_block_mv(candidate_mi, 0, mv_ref->col, block), + refmv_count, mv_ref_list, Done); + else if (candidate_mi->ref_frame[1] == ref_frame) + ADD_MV_REF_LIST(get_sub_block_mv(candidate_mi, 1, mv_ref->col, block), + refmv_count, mv_ref_list, Done); + } + } + + // Check the rest of the neighbors in much the same way + // as before except we don't need to keep track of sub blocks or + // mode counts. + for (; i < MVREF_NEIGHBOURS; ++i) { + const POSITION *const mv_ref = &mv_ref_search[i]; + if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) { + const MODE_INFO *const candidate_mi = + xd->mi[mv_ref->col + mv_ref->row * xd->mi_stride]; + different_ref_found = 1; + + if (candidate_mi->ref_frame[0] == ref_frame) + ADD_MV_REF_LIST(candidate_mi->mv[0], refmv_count, mv_ref_list, Done); + else if (candidate_mi->ref_frame[1] == ref_frame) + ADD_MV_REF_LIST(candidate_mi->mv[1], refmv_count, mv_ref_list, Done); + } + } + + // Check the last frame's mode and mv info. + if (cm->use_prev_frame_mvs) { + if (prev_frame_mvs->ref_frame[0] == ref_frame) { + ADD_MV_REF_LIST(prev_frame_mvs->mv[0], refmv_count, mv_ref_list, Done); + } else if (prev_frame_mvs->ref_frame[1] == ref_frame) { + ADD_MV_REF_LIST(prev_frame_mvs->mv[1], refmv_count, mv_ref_list, Done); + } + } + + // Since we couldn't find 2 mvs from the same reference frame + // go back through the neighbors and find motion vectors from + // different reference frames. + if (different_ref_found) { + for (i = 0; i < MVREF_NEIGHBOURS; ++i) { + const POSITION *mv_ref = &mv_ref_search[i]; + if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) { + const MODE_INFO *const candidate_mi = + xd->mi[mv_ref->col + mv_ref->row * xd->mi_stride]; + + // If the candidate is INTRA we don't want to consider its mv. + IF_DIFF_REF_FRAME_ADD_MV(candidate_mi, ref_frame, ref_sign_bias, + refmv_count, mv_ref_list, Done); + } + } + } + + // Since we still don't have a candidate we'll try the last frame. + if (cm->use_prev_frame_mvs) { + if (prev_frame_mvs->ref_frame[0] != ref_frame && + prev_frame_mvs->ref_frame[0] > INTRA_FRAME) { + int_mv mv = prev_frame_mvs->mv[0]; + if (ref_sign_bias[prev_frame_mvs->ref_frame[0]] != + ref_sign_bias[ref_frame]) { + mv.as_mv.row *= -1; + mv.as_mv.col *= -1; + } + ADD_MV_REF_LIST(mv, refmv_count, mv_ref_list, Done); + } + + if (prev_frame_mvs->ref_frame[1] > INTRA_FRAME && + prev_frame_mvs->ref_frame[1] != ref_frame && + prev_frame_mvs->mv[1].as_int != prev_frame_mvs->mv[0].as_int) { + int_mv mv = prev_frame_mvs->mv[1]; + if (ref_sign_bias[prev_frame_mvs->ref_frame[1]] != + ref_sign_bias[ref_frame]) { + mv.as_mv.row *= -1; + mv.as_mv.col *= -1; + } + ADD_MV_REF_LIST(mv, refmv_count, mv_ref_list, Done); + } + } + + Done: + + mode_context[ref_frame] = counter_to_context[context_counter]; + + // Clamp vectors + for (i = 0; i < MAX_MV_REF_CANDIDATES; ++i) + clamp_mv_ref(&mv_ref_list[i].as_mv, xd); +} + +void vp9_find_mv_refs(const VP9_COMMON *cm, const MACROBLOCKD *xd, + MODE_INFO *mi, MV_REFERENCE_FRAME ref_frame, + int_mv *mv_ref_list, + int mi_row, int mi_col, + uint8_t *mode_context) { + find_mv_refs_idx(cm, xd, mi, ref_frame, mv_ref_list, -1, + mi_row, mi_col, mode_context); +} + +void vp9_find_best_ref_mvs(MACROBLOCKD *xd, int allow_hp, + int_mv *mvlist, int_mv *nearest_mv, + int_mv *near_mv) { + int i; + // Make sure all the candidates are properly clamped etc + for (i = 0; i < MAX_MV_REF_CANDIDATES; ++i) { + lower_mv_precision(&mvlist[i].as_mv, allow_hp); + clamp_mv2(&mvlist[i].as_mv, xd); + } + *nearest_mv = mvlist[0]; + *near_mv = mvlist[1]; +} + +void vp9_append_sub8x8_mvs_for_idx(VP9_COMMON *cm, MACROBLOCKD *xd, + int block, int ref, int mi_row, int mi_col, + int_mv *nearest_mv, int_mv *near_mv, + uint8_t *mode_context) { + int_mv mv_list[MAX_MV_REF_CANDIDATES]; + MODE_INFO *const mi = xd->mi[0]; + b_mode_info *bmi = mi->bmi; + int n; + + assert(MAX_MV_REF_CANDIDATES == 2); + + find_mv_refs_idx(cm, xd, mi, mi->ref_frame[ref], mv_list, block, + mi_row, mi_col, mode_context); + + near_mv->as_int = 0; + switch (block) { + case 0: + nearest_mv->as_int = mv_list[0].as_int; + near_mv->as_int = mv_list[1].as_int; + break; + case 1: + case 2: + nearest_mv->as_int = bmi[0].as_mv[ref].as_int; + for (n = 0; n < MAX_MV_REF_CANDIDATES; ++n) + if (nearest_mv->as_int != mv_list[n].as_int) { + near_mv->as_int = mv_list[n].as_int; + break; + } + break; + case 3: { + int_mv candidates[2 + MAX_MV_REF_CANDIDATES]; + candidates[0] = bmi[1].as_mv[ref]; + candidates[1] = bmi[0].as_mv[ref]; + candidates[2] = mv_list[0]; + candidates[3] = mv_list[1]; + + nearest_mv->as_int = bmi[2].as_mv[ref].as_int; + for (n = 0; n < 2 + MAX_MV_REF_CANDIDATES; ++n) + if (nearest_mv->as_int != candidates[n].as_int) { + near_mv->as_int = candidates[n].as_int; + break; + } + break; + } + default: + assert(0 && "Invalid block index."); + } +} diff --git a/thirdparty/libvpx/vp9/common/vp9_mvref_common.h b/thirdparty/libvpx/vp9/common/vp9_mvref_common.h new file mode 100644 index 0000000000..4380843e24 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_mvref_common.h @@ -0,0 +1,241 @@ +/* + * Copyright (c) 2012 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ +#ifndef VP9_COMMON_VP9_MVREF_COMMON_H_ +#define VP9_COMMON_VP9_MVREF_COMMON_H_ + +#include "vp9/common/vp9_onyxc_int.h" +#include "vp9/common/vp9_blockd.h" + +#ifdef __cplusplus +extern "C" { +#endif + +#define LEFT_TOP_MARGIN ((VP9_ENC_BORDER_IN_PIXELS - VP9_INTERP_EXTEND) << 3) +#define RIGHT_BOTTOM_MARGIN ((VP9_ENC_BORDER_IN_PIXELS -\ + VP9_INTERP_EXTEND) << 3) + +#define MVREF_NEIGHBOURS 8 + +typedef struct position { + int row; + int col; +} POSITION; + +typedef enum { + BOTH_ZERO = 0, + ZERO_PLUS_PREDICTED = 1, + BOTH_PREDICTED = 2, + NEW_PLUS_NON_INTRA = 3, + BOTH_NEW = 4, + INTRA_PLUS_NON_INTRA = 5, + BOTH_INTRA = 6, + INVALID_CASE = 9 +} motion_vector_context; + +// This is used to figure out a context for the ref blocks. The code flattens +// an array that would have 3 possible counts (0, 1 & 2) for 3 choices by +// adding 9 for each intra block, 3 for each zero mv and 1 for each new +// motion vector. This single number is then converted into a context +// with a single lookup ( counter_to_context ). +static const int mode_2_counter[MB_MODE_COUNT] = { + 9, // DC_PRED + 9, // V_PRED + 9, // H_PRED + 9, // D45_PRED + 9, // D135_PRED + 9, // D117_PRED + 9, // D153_PRED + 9, // D207_PRED + 9, // D63_PRED + 9, // TM_PRED + 0, // NEARESTMV + 0, // NEARMV + 3, // ZEROMV + 1, // NEWMV +}; + +// There are 3^3 different combinations of 3 counts that can be either 0,1 or +// 2. However the actual count can never be greater than 2 so the highest +// counter we need is 18. 9 is an invalid counter that's never used. +static const int counter_to_context[19] = { + BOTH_PREDICTED, // 0 + NEW_PLUS_NON_INTRA, // 1 + BOTH_NEW, // 2 + ZERO_PLUS_PREDICTED, // 3 + NEW_PLUS_NON_INTRA, // 4 + INVALID_CASE, // 5 + BOTH_ZERO, // 6 + INVALID_CASE, // 7 + INVALID_CASE, // 8 + INTRA_PLUS_NON_INTRA, // 9 + INTRA_PLUS_NON_INTRA, // 10 + INVALID_CASE, // 11 + INTRA_PLUS_NON_INTRA, // 12 + INVALID_CASE, // 13 + INVALID_CASE, // 14 + INVALID_CASE, // 15 + INVALID_CASE, // 16 + INVALID_CASE, // 17 + BOTH_INTRA // 18 +}; + +static const POSITION mv_ref_blocks[BLOCK_SIZES][MVREF_NEIGHBOURS] = { + // 4X4 + {{-1, 0}, {0, -1}, {-1, -1}, {-2, 0}, {0, -2}, {-2, -1}, {-1, -2}, {-2, -2}}, + // 4X8 + {{-1, 0}, {0, -1}, {-1, -1}, {-2, 0}, {0, -2}, {-2, -1}, {-1, -2}, {-2, -2}}, + // 8X4 + {{-1, 0}, {0, -1}, {-1, -1}, {-2, 0}, {0, -2}, {-2, -1}, {-1, -2}, {-2, -2}}, + // 8X8 + {{-1, 0}, {0, -1}, {-1, -1}, {-2, 0}, {0, -2}, {-2, -1}, {-1, -2}, {-2, -2}}, + // 8X16 + {{0, -1}, {-1, 0}, {1, -1}, {-1, -1}, {0, -2}, {-2, 0}, {-2, -1}, {-1, -2}}, + // 16X8 + {{-1, 0}, {0, -1}, {-1, 1}, {-1, -1}, {-2, 0}, {0, -2}, {-1, -2}, {-2, -1}}, + // 16X16 + {{-1, 0}, {0, -1}, {-1, 1}, {1, -1}, {-1, -1}, {-3, 0}, {0, -3}, {-3, -3}}, + // 16X32 + {{0, -1}, {-1, 0}, {2, -1}, {-1, -1}, {-1, 1}, {0, -3}, {-3, 0}, {-3, -3}}, + // 32X16 + {{-1, 0}, {0, -1}, {-1, 2}, {-1, -1}, {1, -1}, {-3, 0}, {0, -3}, {-3, -3}}, + // 32X32 + {{-1, 1}, {1, -1}, {-1, 2}, {2, -1}, {-1, -1}, {-3, 0}, {0, -3}, {-3, -3}}, + // 32X64 + {{0, -1}, {-1, 0}, {4, -1}, {-1, 2}, {-1, -1}, {0, -3}, {-3, 0}, {2, -1}}, + // 64X32 + {{-1, 0}, {0, -1}, {-1, 4}, {2, -1}, {-1, -1}, {-3, 0}, {0, -3}, {-1, 2}}, + // 64X64 + {{-1, 3}, {3, -1}, {-1, 4}, {4, -1}, {-1, -1}, {-1, 0}, {0, -1}, {-1, 6}} +}; + +static const int idx_n_column_to_subblock[4][2] = { + {1, 2}, + {1, 3}, + {3, 2}, + {3, 3} +}; + +// clamp_mv_ref +#define MV_BORDER (16 << 3) // Allow 16 pels in 1/8th pel units + +static INLINE void clamp_mv_ref(MV *mv, const MACROBLOCKD *xd) { + clamp_mv(mv, xd->mb_to_left_edge - MV_BORDER, + xd->mb_to_right_edge + MV_BORDER, + xd->mb_to_top_edge - MV_BORDER, + xd->mb_to_bottom_edge + MV_BORDER); +} + +// This function returns either the appropriate sub block or block's mv +// on whether the block_size < 8x8 and we have check_sub_blocks set. +static INLINE int_mv get_sub_block_mv(const MODE_INFO *candidate, int which_mv, + int search_col, int block_idx) { + return block_idx >= 0 && candidate->sb_type < BLOCK_8X8 + ? candidate->bmi[idx_n_column_to_subblock[block_idx][search_col == 0]] + .as_mv[which_mv] + : candidate->mv[which_mv]; +} + + +// Performs mv sign inversion if indicated by the reference frame combination. +static INLINE int_mv scale_mv(const MODE_INFO *mi, int ref, + const MV_REFERENCE_FRAME this_ref_frame, + const int *ref_sign_bias) { + int_mv mv = mi->mv[ref]; + if (ref_sign_bias[mi->ref_frame[ref]] != ref_sign_bias[this_ref_frame]) { + mv.as_mv.row *= -1; + mv.as_mv.col *= -1; + } + return mv; +} + +// This macro is used to add a motion vector mv_ref list if it isn't +// already in the list. If it's the second motion vector it will also +// skip all additional processing and jump to Done! +#define ADD_MV_REF_LIST(mv, refmv_count, mv_ref_list, Done) \ + do { \ + if (refmv_count) { \ + if ((mv).as_int != (mv_ref_list)[0].as_int) { \ + (mv_ref_list)[(refmv_count)] = (mv); \ + goto Done; \ + } \ + } else { \ + (mv_ref_list)[(refmv_count)++] = (mv); \ + } \ + } while (0) + +// If either reference frame is different, not INTRA, and they +// are different from each other scale and add the mv to our list. +#define IF_DIFF_REF_FRAME_ADD_MV(mbmi, ref_frame, ref_sign_bias, refmv_count, \ + mv_ref_list, Done) \ + do { \ + if (is_inter_block(mbmi)) { \ + if ((mbmi)->ref_frame[0] != ref_frame) \ + ADD_MV_REF_LIST(scale_mv((mbmi), 0, ref_frame, ref_sign_bias), \ + refmv_count, mv_ref_list, Done); \ + if (has_second_ref(mbmi) && \ + (mbmi)->ref_frame[1] != ref_frame && \ + (mbmi)->mv[1].as_int != (mbmi)->mv[0].as_int) \ + ADD_MV_REF_LIST(scale_mv((mbmi), 1, ref_frame, ref_sign_bias), \ + refmv_count, mv_ref_list, Done); \ + } \ + } while (0) + + +// Checks that the given mi_row, mi_col and search point +// are inside the borders of the tile. +static INLINE int is_inside(const TileInfo *const tile, + int mi_col, int mi_row, int mi_rows, + const POSITION *mi_pos) { + return !(mi_row + mi_pos->row < 0 || + mi_col + mi_pos->col < tile->mi_col_start || + mi_row + mi_pos->row >= mi_rows || + mi_col + mi_pos->col >= tile->mi_col_end); +} + +// TODO(jingning): this mv clamping function should be block size dependent. +static INLINE void clamp_mv2(MV *mv, const MACROBLOCKD *xd) { + clamp_mv(mv, xd->mb_to_left_edge - LEFT_TOP_MARGIN, + xd->mb_to_right_edge + RIGHT_BOTTOM_MARGIN, + xd->mb_to_top_edge - LEFT_TOP_MARGIN, + xd->mb_to_bottom_edge + RIGHT_BOTTOM_MARGIN); +} + +static INLINE void lower_mv_precision(MV *mv, int allow_hp) { + const int use_hp = allow_hp && use_mv_hp(mv); + if (!use_hp) { + if (mv->row & 1) + mv->row += (mv->row > 0 ? -1 : 1); + if (mv->col & 1) + mv->col += (mv->col > 0 ? -1 : 1); + } +} + +typedef void (*find_mv_refs_sync)(void *const data, int mi_row); +void vp9_find_mv_refs(const VP9_COMMON *cm, const MACROBLOCKD *xd, + MODE_INFO *mi, MV_REFERENCE_FRAME ref_frame, + int_mv *mv_ref_list, int mi_row, int mi_col, + uint8_t *mode_context); + +// check a list of motion vectors by sad score using a number rows of pixels +// above and a number cols of pixels in the left to select the one with best +// score to use as ref motion vector +void vp9_find_best_ref_mvs(MACROBLOCKD *xd, int allow_hp, + int_mv *mvlist, int_mv *nearest_mv, int_mv *near_mv); + +void vp9_append_sub8x8_mvs_for_idx(VP9_COMMON *cm, MACROBLOCKD *xd, + int block, int ref, int mi_row, int mi_col, + int_mv *nearest_mv, int_mv *near_mv, + uint8_t *mode_context); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_MVREF_COMMON_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_onyxc_int.h b/thirdparty/libvpx/vp9/common/vp9_onyxc_int.h new file mode 100644 index 0000000000..3fd935e628 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_onyxc_int.h @@ -0,0 +1,446 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_COMMON_VP9_ONYXC_INT_H_ +#define VP9_COMMON_VP9_ONYXC_INT_H_ + +#include "./vpx_config.h" +#include "vpx/internal/vpx_codec_internal.h" +#include "vpx_util/vpx_thread.h" +#include "./vp9_rtcd.h" +#include "vp9/common/vp9_alloccommon.h" +#include "vp9/common/vp9_loopfilter.h" +#include "vp9/common/vp9_entropymv.h" +#include "vp9/common/vp9_entropy.h" +#include "vp9/common/vp9_entropymode.h" +#include "vp9/common/vp9_frame_buffers.h" +#include "vp9/common/vp9_quant_common.h" +#include "vp9/common/vp9_tile_common.h" + +#if CONFIG_VP9_POSTPROC +#include "vp9/common/vp9_postproc.h" +#endif + +#ifdef __cplusplus +extern "C" { +#endif + +#define REFS_PER_FRAME 3 + +#define REF_FRAMES_LOG2 3 +#define REF_FRAMES (1 << REF_FRAMES_LOG2) + +// 4 scratch frames for the new frames to support a maximum of 4 cores decoding +// in parallel, 3 for scaled references on the encoder. +// TODO(hkuang): Add ondemand frame buffers instead of hardcoding the number +// of framebuffers. +// TODO(jkoleszar): These 3 extra references could probably come from the +// normal reference pool. +#define FRAME_BUFFERS (REF_FRAMES + 7) + +#define FRAME_CONTEXTS_LOG2 2 +#define FRAME_CONTEXTS (1 << FRAME_CONTEXTS_LOG2) + +#define NUM_PING_PONG_BUFFERS 2 + +extern const struct { + PARTITION_CONTEXT above; + PARTITION_CONTEXT left; +} partition_context_lookup[BLOCK_SIZES]; + + +typedef enum { + SINGLE_REFERENCE = 0, + COMPOUND_REFERENCE = 1, + REFERENCE_MODE_SELECT = 2, + REFERENCE_MODES = 3, +} REFERENCE_MODE; + +typedef struct { + int_mv mv[2]; + MV_REFERENCE_FRAME ref_frame[2]; +} MV_REF; + +typedef struct { + int ref_count; + MV_REF *mvs; + int mi_rows; + int mi_cols; + vpx_codec_frame_buffer_t raw_frame_buffer; + YV12_BUFFER_CONFIG buf; + + // The Following variables will only be used in frame parallel decode. + + // frame_worker_owner indicates which FrameWorker owns this buffer. NULL means + // that no FrameWorker owns, or is decoding, this buffer. + VPxWorker *frame_worker_owner; + + // row and col indicate which position frame has been decoded to in real + // pixel unit. They are reset to -1 when decoding begins and set to INT_MAX + // when the frame is fully decoded. + int row; + int col; +} RefCntBuffer; + +typedef struct BufferPool { + // Protect BufferPool from being accessed by several FrameWorkers at + // the same time during frame parallel decode. + // TODO(hkuang): Try to use atomic variable instead of locking the whole pool. +#if CONFIG_MULTITHREAD + pthread_mutex_t pool_mutex; +#endif + + // Private data associated with the frame buffer callbacks. + void *cb_priv; + + vpx_get_frame_buffer_cb_fn_t get_fb_cb; + vpx_release_frame_buffer_cb_fn_t release_fb_cb; + + RefCntBuffer frame_bufs[FRAME_BUFFERS]; + + // Frame buffers allocated internally by the codec. + InternalFrameBufferList int_frame_buffers; +} BufferPool; + +typedef struct VP9Common { + struct vpx_internal_error_info error; + vpx_color_space_t color_space; + vpx_color_range_t color_range; + int width; + int height; + int render_width; + int render_height; + int last_width; + int last_height; + + // TODO(jkoleszar): this implies chroma ss right now, but could vary per + // plane. Revisit as part of the future change to YV12_BUFFER_CONFIG to + // support additional planes. + int subsampling_x; + int subsampling_y; + +#if CONFIG_VP9_HIGHBITDEPTH + int use_highbitdepth; // Marks if we need to use 16bit frame buffers. +#endif + + YV12_BUFFER_CONFIG *frame_to_show; + RefCntBuffer *prev_frame; + + // TODO(hkuang): Combine this with cur_buf in macroblockd. + RefCntBuffer *cur_frame; + + int ref_frame_map[REF_FRAMES]; /* maps fb_idx to reference slot */ + + // Prepare ref_frame_map for the next frame. + // Only used in frame parallel decode. + int next_ref_frame_map[REF_FRAMES]; + + // TODO(jkoleszar): could expand active_ref_idx to 4, with 0 as intra, and + // roll new_fb_idx into it. + + // Each frame can reference REFS_PER_FRAME buffers + RefBuffer frame_refs[REFS_PER_FRAME]; + + int new_fb_idx; + +#if CONFIG_VP9_POSTPROC + YV12_BUFFER_CONFIG post_proc_buffer; + YV12_BUFFER_CONFIG post_proc_buffer_int; +#endif + + FRAME_TYPE last_frame_type; /* last frame's frame type for motion search.*/ + FRAME_TYPE frame_type; + + int show_frame; + int last_show_frame; + int show_existing_frame; + + // Flag signaling that the frame is encoded using only INTRA modes. + uint8_t intra_only; + uint8_t last_intra_only; + + int allow_high_precision_mv; + + // Flag signaling that the frame context should be reset to default values. + // 0 or 1 implies don't reset, 2 reset just the context specified in the + // frame header, 3 reset all contexts. + int reset_frame_context; + + // MBs, mb_rows/cols is in 16-pixel units; mi_rows/cols is in + // MODE_INFO (8-pixel) units. + int MBs; + int mb_rows, mi_rows; + int mb_cols, mi_cols; + int mi_stride; + + /* profile settings */ + TX_MODE tx_mode; + + int base_qindex; + int y_dc_delta_q; + int uv_dc_delta_q; + int uv_ac_delta_q; + int16_t y_dequant[MAX_SEGMENTS][2]; + int16_t uv_dequant[MAX_SEGMENTS][2]; + + /* We allocate a MODE_INFO struct for each macroblock, together with + an extra row on top and column on the left to simplify prediction. */ + int mi_alloc_size; + MODE_INFO *mip; /* Base of allocated array */ + MODE_INFO *mi; /* Corresponds to upper left visible macroblock */ + + // TODO(agrange): Move prev_mi into encoder structure. + // prev_mip and prev_mi will only be allocated in VP9 encoder. + MODE_INFO *prev_mip; /* MODE_INFO array 'mip' from last decoded frame */ + MODE_INFO *prev_mi; /* 'mi' from last frame (points into prev_mip) */ + + // Separate mi functions between encoder and decoder. + int (*alloc_mi)(struct VP9Common *cm, int mi_size); + void (*free_mi)(struct VP9Common *cm); + void (*setup_mi)(struct VP9Common *cm); + + // Grid of pointers to 8x8 MODE_INFO structs. Any 8x8 not in the visible + // area will be NULL. + MODE_INFO **mi_grid_base; + MODE_INFO **mi_grid_visible; + MODE_INFO **prev_mi_grid_base; + MODE_INFO **prev_mi_grid_visible; + + // Whether to use previous frame's motion vectors for prediction. + int use_prev_frame_mvs; + + // Persistent mb segment id map used in prediction. + int seg_map_idx; + int prev_seg_map_idx; + + uint8_t *seg_map_array[NUM_PING_PONG_BUFFERS]; + uint8_t *last_frame_seg_map; + uint8_t *current_frame_seg_map; + int seg_map_alloc_size; + + INTERP_FILTER interp_filter; + + loop_filter_info_n lf_info; + + int refresh_frame_context; /* Two state 0 = NO, 1 = YES */ + + int ref_frame_sign_bias[MAX_REF_FRAMES]; /* Two state 0, 1 */ + + struct loopfilter lf; + struct segmentation seg; + + // TODO(hkuang): Remove this as it is the same as frame_parallel_decode + // in pbi. + int frame_parallel_decode; // frame-based threading. + + // Context probabilities for reference frame prediction + MV_REFERENCE_FRAME comp_fixed_ref; + MV_REFERENCE_FRAME comp_var_ref[2]; + REFERENCE_MODE reference_mode; + + FRAME_CONTEXT *fc; /* this frame entropy */ + FRAME_CONTEXT *frame_contexts; // FRAME_CONTEXTS + unsigned int frame_context_idx; /* Context to use/update */ + FRAME_COUNTS counts; + + unsigned int current_video_frame; + BITSTREAM_PROFILE profile; + + // VPX_BITS_8 in profile 0 or 1, VPX_BITS_10 or VPX_BITS_12 in profile 2 or 3. + vpx_bit_depth_t bit_depth; + vpx_bit_depth_t dequant_bit_depth; // bit_depth of current dequantizer + +#if CONFIG_VP9_POSTPROC + struct postproc_state postproc_state; +#endif + + int error_resilient_mode; + int frame_parallel_decoding_mode; + + int log2_tile_cols, log2_tile_rows; + int byte_alignment; + int skip_loop_filter; + + // Private data associated with the frame buffer callbacks. + void *cb_priv; + vpx_get_frame_buffer_cb_fn_t get_fb_cb; + vpx_release_frame_buffer_cb_fn_t release_fb_cb; + + // Handles memory for the codec. + InternalFrameBufferList int_frame_buffers; + + // External BufferPool passed from outside. + BufferPool *buffer_pool; + + PARTITION_CONTEXT *above_seg_context; + ENTROPY_CONTEXT *above_context; + int above_context_alloc_cols; +} VP9_COMMON; + +// TODO(hkuang): Don't need to lock the whole pool after implementing atomic +// frame reference count. +void lock_buffer_pool(BufferPool *const pool); +void unlock_buffer_pool(BufferPool *const pool); + +static INLINE YV12_BUFFER_CONFIG *get_ref_frame(VP9_COMMON *cm, int index) { + if (index < 0 || index >= REF_FRAMES) + return NULL; + if (cm->ref_frame_map[index] < 0) + return NULL; + assert(cm->ref_frame_map[index] < FRAME_BUFFERS); + return &cm->buffer_pool->frame_bufs[cm->ref_frame_map[index]].buf; +} + +static INLINE YV12_BUFFER_CONFIG *get_frame_new_buffer(VP9_COMMON *cm) { + return &cm->buffer_pool->frame_bufs[cm->new_fb_idx].buf; +} + +static INLINE int get_free_fb(VP9_COMMON *cm) { + RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs; + int i; + + lock_buffer_pool(cm->buffer_pool); + for (i = 0; i < FRAME_BUFFERS; ++i) + if (frame_bufs[i].ref_count == 0) + break; + + if (i != FRAME_BUFFERS) { + frame_bufs[i].ref_count = 1; + } else { + // Reset i to be INVALID_IDX to indicate no free buffer found. + i = INVALID_IDX; + } + + unlock_buffer_pool(cm->buffer_pool); + return i; +} + +static INLINE void ref_cnt_fb(RefCntBuffer *bufs, int *idx, int new_idx) { + const int ref_index = *idx; + + if (ref_index >= 0 && bufs[ref_index].ref_count > 0) + bufs[ref_index].ref_count--; + + *idx = new_idx; + + bufs[new_idx].ref_count++; +} + +static INLINE int mi_cols_aligned_to_sb(int n_mis) { + return ALIGN_POWER_OF_TWO(n_mis, MI_BLOCK_SIZE_LOG2); +} + +static INLINE int frame_is_intra_only(const VP9_COMMON *const cm) { + return cm->frame_type == KEY_FRAME || cm->intra_only; +} + +static INLINE void set_partition_probs(const VP9_COMMON *const cm, + MACROBLOCKD *const xd) { + xd->partition_probs = + frame_is_intra_only(cm) ? + &vp9_kf_partition_probs[0] : + (const vpx_prob (*)[PARTITION_TYPES - 1])cm->fc->partition_prob; +} + +static INLINE void vp9_init_macroblockd(VP9_COMMON *cm, MACROBLOCKD *xd, + tran_low_t *dqcoeff) { + int i; + + for (i = 0; i < MAX_MB_PLANE; ++i) { + xd->plane[i].dqcoeff = dqcoeff; + xd->above_context[i] = cm->above_context + + i * sizeof(*cm->above_context) * 2 * mi_cols_aligned_to_sb(cm->mi_cols); + + if (get_plane_type(i) == PLANE_TYPE_Y) { + memcpy(xd->plane[i].seg_dequant, cm->y_dequant, sizeof(cm->y_dequant)); + } else { + memcpy(xd->plane[i].seg_dequant, cm->uv_dequant, sizeof(cm->uv_dequant)); + } + xd->fc = cm->fc; + } + + xd->above_seg_context = cm->above_seg_context; + xd->mi_stride = cm->mi_stride; + xd->error_info = &cm->error; + + set_partition_probs(cm, xd); +} + +static INLINE const vpx_prob* get_partition_probs(const MACROBLOCKD *xd, + int ctx) { + return xd->partition_probs[ctx]; +} + +static INLINE void set_skip_context(MACROBLOCKD *xd, int mi_row, int mi_col) { + const int above_idx = mi_col * 2; + const int left_idx = (mi_row * 2) & 15; + int i; + for (i = 0; i < MAX_MB_PLANE; ++i) { + struct macroblockd_plane *const pd = &xd->plane[i]; + pd->above_context = &xd->above_context[i][above_idx >> pd->subsampling_x]; + pd->left_context = &xd->left_context[i][left_idx >> pd->subsampling_y]; + } +} + +static INLINE int calc_mi_size(int len) { + // len is in mi units. + return len + MI_BLOCK_SIZE; +} + +static INLINE void set_mi_row_col(MACROBLOCKD *xd, const TileInfo *const tile, + int mi_row, int bh, + int mi_col, int bw, + int mi_rows, int mi_cols) { + xd->mb_to_top_edge = -((mi_row * MI_SIZE) * 8); + xd->mb_to_bottom_edge = ((mi_rows - bh - mi_row) * MI_SIZE) * 8; + xd->mb_to_left_edge = -((mi_col * MI_SIZE) * 8); + xd->mb_to_right_edge = ((mi_cols - bw - mi_col) * MI_SIZE) * 8; + + // Are edges available for intra prediction? + xd->above_mi = (mi_row != 0) ? xd->mi[-xd->mi_stride] : NULL; + xd->left_mi = (mi_col > tile->mi_col_start) ? xd->mi[-1] : NULL; +} + +static INLINE void update_partition_context(MACROBLOCKD *xd, + int mi_row, int mi_col, + BLOCK_SIZE subsize, + BLOCK_SIZE bsize) { + PARTITION_CONTEXT *const above_ctx = xd->above_seg_context + mi_col; + PARTITION_CONTEXT *const left_ctx = xd->left_seg_context + (mi_row & MI_MASK); + + // num_4x4_blocks_wide_lookup[bsize] / 2 + const int bs = num_8x8_blocks_wide_lookup[bsize]; + + // update the partition context at the end notes. set partition bits + // of block sizes larger than the current one to be one, and partition + // bits of smaller block sizes to be zero. + memset(above_ctx, partition_context_lookup[subsize].above, bs); + memset(left_ctx, partition_context_lookup[subsize].left, bs); +} + +static INLINE int partition_plane_context(const MACROBLOCKD *xd, + int mi_row, int mi_col, + BLOCK_SIZE bsize) { + const PARTITION_CONTEXT *above_ctx = xd->above_seg_context + mi_col; + const PARTITION_CONTEXT *left_ctx = xd->left_seg_context + (mi_row & MI_MASK); + const int bsl = mi_width_log2_lookup[bsize]; + int above = (*above_ctx >> bsl) & 1 , left = (*left_ctx >> bsl) & 1; + + assert(b_width_log2_lookup[bsize] == b_height_log2_lookup[bsize]); + assert(bsl >= 0); + + return (left * 2 + above) + bsl * PARTITION_PLOFFSET; +} + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_ONYXC_INT_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_ppflags.h b/thirdparty/libvpx/vp9/common/vp9_ppflags.h new file mode 100644 index 0000000000..12b989f43a --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_ppflags.h @@ -0,0 +1,43 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_COMMON_VP9_PPFLAGS_H_ +#define VP9_COMMON_VP9_PPFLAGS_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +enum { + VP9D_NOFILTERING = 0, + VP9D_DEBLOCK = 1 << 0, + VP9D_DEMACROBLOCK = 1 << 1, + VP9D_ADDNOISE = 1 << 2, + VP9D_DEBUG_TXT_FRAME_INFO = 1 << 3, + VP9D_DEBUG_TXT_MBLK_MODES = 1 << 4, + VP9D_DEBUG_TXT_DC_DIFF = 1 << 5, + VP9D_DEBUG_TXT_RATE_INFO = 1 << 6, + VP9D_DEBUG_DRAW_MV = 1 << 7, + VP9D_DEBUG_CLR_BLK_MODES = 1 << 8, + VP9D_DEBUG_CLR_FRM_REF_BLKS = 1 << 9, + VP9D_MFQE = 1 << 10 +}; + +typedef struct { + int post_proc_flag; + int deblocking_level; + int noise_level; +} vp9_ppflags_t; + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_PPFLAGS_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_pred_common.c b/thirdparty/libvpx/vp9/common/vp9_pred_common.c new file mode 100644 index 0000000000..8f90e70e73 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_pred_common.c @@ -0,0 +1,314 @@ + +/* + * Copyright (c) 2012 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "vp9/common/vp9_common.h" +#include "vp9/common/vp9_pred_common.h" +#include "vp9/common/vp9_seg_common.h" + +// Returns a context number for the given MB prediction signal +int vp9_get_pred_context_switchable_interp(const MACROBLOCKD *xd) { + // Note: + // The mode info data structure has a one element border above and to the + // left of the entries corresponding to real macroblocks. + // The prediction flags in these dummy entries are initialized to 0. + const MODE_INFO *const left_mi = xd->left_mi; + const int left_type = left_mi && is_inter_block(left_mi) ? + left_mi->interp_filter : SWITCHABLE_FILTERS; + const MODE_INFO *const above_mi = xd->above_mi; + const int above_type = above_mi && is_inter_block(above_mi) ? + above_mi->interp_filter : SWITCHABLE_FILTERS; + + if (left_type == above_type) + return left_type; + else if (left_type == SWITCHABLE_FILTERS) + return above_type; + else if (above_type == SWITCHABLE_FILTERS) + return left_type; + else + return SWITCHABLE_FILTERS; +} + +int vp9_get_reference_mode_context(const VP9_COMMON *cm, + const MACROBLOCKD *xd) { + int ctx; + const MODE_INFO *const above_mi = xd->above_mi; + const MODE_INFO *const left_mi = xd->left_mi; + const int has_above = !!above_mi; + const int has_left = !!left_mi; + // Note: + // The mode info data structure has a one element border above and to the + // left of the entries corresponding to real macroblocks. + // The prediction flags in these dummy entries are initialized to 0. + if (has_above && has_left) { // both edges available + if (!has_second_ref(above_mi) && !has_second_ref(left_mi)) + // neither edge uses comp pred (0/1) + ctx = (above_mi->ref_frame[0] == cm->comp_fixed_ref) ^ + (left_mi->ref_frame[0] == cm->comp_fixed_ref); + else if (!has_second_ref(above_mi)) + // one of two edges uses comp pred (2/3) + ctx = 2 + (above_mi->ref_frame[0] == cm->comp_fixed_ref || + !is_inter_block(above_mi)); + else if (!has_second_ref(left_mi)) + // one of two edges uses comp pred (2/3) + ctx = 2 + (left_mi->ref_frame[0] == cm->comp_fixed_ref || + !is_inter_block(left_mi)); + else // both edges use comp pred (4) + ctx = 4; + } else if (has_above || has_left) { // one edge available + const MODE_INFO *edge_mi = has_above ? above_mi : left_mi; + + if (!has_second_ref(edge_mi)) + // edge does not use comp pred (0/1) + ctx = edge_mi->ref_frame[0] == cm->comp_fixed_ref; + else + // edge uses comp pred (3) + ctx = 3; + } else { // no edges available (1) + ctx = 1; + } + assert(ctx >= 0 && ctx < COMP_INTER_CONTEXTS); + return ctx; +} + +// Returns a context number for the given MB prediction signal +int vp9_get_pred_context_comp_ref_p(const VP9_COMMON *cm, + const MACROBLOCKD *xd) { + int pred_context; + const MODE_INFO *const above_mi = xd->above_mi; + const MODE_INFO *const left_mi = xd->left_mi; + const int above_in_image = !!above_mi; + const int left_in_image = !!left_mi; + + // Note: + // The mode info data structure has a one element border above and to the + // left of the entries corresponding to real macroblocks. + // The prediction flags in these dummy entries are initialized to 0. + const int fix_ref_idx = cm->ref_frame_sign_bias[cm->comp_fixed_ref]; + const int var_ref_idx = !fix_ref_idx; + + if (above_in_image && left_in_image) { // both edges available + const int above_intra = !is_inter_block(above_mi); + const int left_intra = !is_inter_block(left_mi); + + if (above_intra && left_intra) { // intra/intra (2) + pred_context = 2; + } else if (above_intra || left_intra) { // intra/inter + const MODE_INFO *edge_mi = above_intra ? left_mi : above_mi; + + if (!has_second_ref(edge_mi)) // single pred (1/3) + pred_context = 1 + 2 * (edge_mi->ref_frame[0] != cm->comp_var_ref[1]); + else // comp pred (1/3) + pred_context = 1 + 2 * (edge_mi->ref_frame[var_ref_idx] + != cm->comp_var_ref[1]); + } else { // inter/inter + const int l_sg = !has_second_ref(left_mi); + const int a_sg = !has_second_ref(above_mi); + const MV_REFERENCE_FRAME vrfa = a_sg ? above_mi->ref_frame[0] + : above_mi->ref_frame[var_ref_idx]; + const MV_REFERENCE_FRAME vrfl = l_sg ? left_mi->ref_frame[0] + : left_mi->ref_frame[var_ref_idx]; + + if (vrfa == vrfl && cm->comp_var_ref[1] == vrfa) { + pred_context = 0; + } else if (l_sg && a_sg) { // single/single + if ((vrfa == cm->comp_fixed_ref && vrfl == cm->comp_var_ref[0]) || + (vrfl == cm->comp_fixed_ref && vrfa == cm->comp_var_ref[0])) + pred_context = 4; + else if (vrfa == vrfl) + pred_context = 3; + else + pred_context = 1; + } else if (l_sg || a_sg) { // single/comp + const MV_REFERENCE_FRAME vrfc = l_sg ? vrfa : vrfl; + const MV_REFERENCE_FRAME rfs = a_sg ? vrfa : vrfl; + if (vrfc == cm->comp_var_ref[1] && rfs != cm->comp_var_ref[1]) + pred_context = 1; + else if (rfs == cm->comp_var_ref[1] && vrfc != cm->comp_var_ref[1]) + pred_context = 2; + else + pred_context = 4; + } else if (vrfa == vrfl) { // comp/comp + pred_context = 4; + } else { + pred_context = 2; + } + } + } else if (above_in_image || left_in_image) { // one edge available + const MODE_INFO *edge_mi = above_in_image ? above_mi : left_mi; + + if (!is_inter_block(edge_mi)) { + pred_context = 2; + } else { + if (has_second_ref(edge_mi)) + pred_context = 4 * (edge_mi->ref_frame[var_ref_idx] + != cm->comp_var_ref[1]); + else + pred_context = 3 * (edge_mi->ref_frame[0] != cm->comp_var_ref[1]); + } + } else { // no edges available (2) + pred_context = 2; + } + assert(pred_context >= 0 && pred_context < REF_CONTEXTS); + + return pred_context; +} + +int vp9_get_pred_context_single_ref_p1(const MACROBLOCKD *xd) { + int pred_context; + const MODE_INFO *const above_mi = xd->above_mi; + const MODE_INFO *const left_mi = xd->left_mi; + const int has_above = !!above_mi; + const int has_left = !!left_mi; + // Note: + // The mode info data structure has a one element border above and to the + // left of the entries corresponding to real macroblocks. + // The prediction flags in these dummy entries are initialized to 0. + if (has_above && has_left) { // both edges available + const int above_intra = !is_inter_block(above_mi); + const int left_intra = !is_inter_block(left_mi); + + if (above_intra && left_intra) { // intra/intra + pred_context = 2; + } else if (above_intra || left_intra) { // intra/inter or inter/intra + const MODE_INFO *edge_mi = above_intra ? left_mi : above_mi; + if (!has_second_ref(edge_mi)) + pred_context = 4 * (edge_mi->ref_frame[0] == LAST_FRAME); + else + pred_context = 1 + (edge_mi->ref_frame[0] == LAST_FRAME || + edge_mi->ref_frame[1] == LAST_FRAME); + } else { // inter/inter + const int above_has_second = has_second_ref(above_mi); + const int left_has_second = has_second_ref(left_mi); + const MV_REFERENCE_FRAME above0 = above_mi->ref_frame[0]; + const MV_REFERENCE_FRAME above1 = above_mi->ref_frame[1]; + const MV_REFERENCE_FRAME left0 = left_mi->ref_frame[0]; + const MV_REFERENCE_FRAME left1 = left_mi->ref_frame[1]; + + if (above_has_second && left_has_second) { + pred_context = 1 + (above0 == LAST_FRAME || above1 == LAST_FRAME || + left0 == LAST_FRAME || left1 == LAST_FRAME); + } else if (above_has_second || left_has_second) { + const MV_REFERENCE_FRAME rfs = !above_has_second ? above0 : left0; + const MV_REFERENCE_FRAME crf1 = above_has_second ? above0 : left0; + const MV_REFERENCE_FRAME crf2 = above_has_second ? above1 : left1; + + if (rfs == LAST_FRAME) + pred_context = 3 + (crf1 == LAST_FRAME || crf2 == LAST_FRAME); + else + pred_context = (crf1 == LAST_FRAME || crf2 == LAST_FRAME); + } else { + pred_context = 2 * (above0 == LAST_FRAME) + 2 * (left0 == LAST_FRAME); + } + } + } else if (has_above || has_left) { // one edge available + const MODE_INFO *edge_mi = has_above ? above_mi : left_mi; + if (!is_inter_block(edge_mi)) { // intra + pred_context = 2; + } else { // inter + if (!has_second_ref(edge_mi)) + pred_context = 4 * (edge_mi->ref_frame[0] == LAST_FRAME); + else + pred_context = 1 + (edge_mi->ref_frame[0] == LAST_FRAME || + edge_mi->ref_frame[1] == LAST_FRAME); + } + } else { // no edges available + pred_context = 2; + } + + assert(pred_context >= 0 && pred_context < REF_CONTEXTS); + return pred_context; +} + +int vp9_get_pred_context_single_ref_p2(const MACROBLOCKD *xd) { + int pred_context; + const MODE_INFO *const above_mi = xd->above_mi; + const MODE_INFO *const left_mi = xd->left_mi; + const int has_above = !!above_mi; + const int has_left = !!left_mi; + + // Note: + // The mode info data structure has a one element border above and to the + // left of the entries corresponding to real macroblocks. + // The prediction flags in these dummy entries are initialized to 0. + if (has_above && has_left) { // both edges available + const int above_intra = !is_inter_block(above_mi); + const int left_intra = !is_inter_block(left_mi); + + if (above_intra && left_intra) { // intra/intra + pred_context = 2; + } else if (above_intra || left_intra) { // intra/inter or inter/intra + const MODE_INFO *edge_mi = above_intra ? left_mi : above_mi; + if (!has_second_ref(edge_mi)) { + if (edge_mi->ref_frame[0] == LAST_FRAME) + pred_context = 3; + else + pred_context = 4 * (edge_mi->ref_frame[0] == GOLDEN_FRAME); + } else { + pred_context = 1 + 2 * (edge_mi->ref_frame[0] == GOLDEN_FRAME || + edge_mi->ref_frame[1] == GOLDEN_FRAME); + } + } else { // inter/inter + const int above_has_second = has_second_ref(above_mi); + const int left_has_second = has_second_ref(left_mi); + const MV_REFERENCE_FRAME above0 = above_mi->ref_frame[0]; + const MV_REFERENCE_FRAME above1 = above_mi->ref_frame[1]; + const MV_REFERENCE_FRAME left0 = left_mi->ref_frame[0]; + const MV_REFERENCE_FRAME left1 = left_mi->ref_frame[1]; + + if (above_has_second && left_has_second) { + if (above0 == left0 && above1 == left1) + pred_context = 3 * (above0 == GOLDEN_FRAME || + above1 == GOLDEN_FRAME || + left0 == GOLDEN_FRAME || + left1 == GOLDEN_FRAME); + else + pred_context = 2; + } else if (above_has_second || left_has_second) { + const MV_REFERENCE_FRAME rfs = !above_has_second ? above0 : left0; + const MV_REFERENCE_FRAME crf1 = above_has_second ? above0 : left0; + const MV_REFERENCE_FRAME crf2 = above_has_second ? above1 : left1; + + if (rfs == GOLDEN_FRAME) + pred_context = 3 + (crf1 == GOLDEN_FRAME || crf2 == GOLDEN_FRAME); + else if (rfs == ALTREF_FRAME) + pred_context = crf1 == GOLDEN_FRAME || crf2 == GOLDEN_FRAME; + else + pred_context = 1 + 2 * (crf1 == GOLDEN_FRAME || crf2 == GOLDEN_FRAME); + } else { + if (above0 == LAST_FRAME && left0 == LAST_FRAME) { + pred_context = 3; + } else if (above0 == LAST_FRAME || left0 == LAST_FRAME) { + const MV_REFERENCE_FRAME edge0 = (above0 == LAST_FRAME) ? left0 + : above0; + pred_context = 4 * (edge0 == GOLDEN_FRAME); + } else { + pred_context = 2 * (above0 == GOLDEN_FRAME) + + 2 * (left0 == GOLDEN_FRAME); + } + } + } + } else if (has_above || has_left) { // one edge available + const MODE_INFO *edge_mi = has_above ? above_mi : left_mi; + + if (!is_inter_block(edge_mi) || + (edge_mi->ref_frame[0] == LAST_FRAME && !has_second_ref(edge_mi))) + pred_context = 2; + else if (!has_second_ref(edge_mi)) + pred_context = 4 * (edge_mi->ref_frame[0] == GOLDEN_FRAME); + else + pred_context = 3 * (edge_mi->ref_frame[0] == GOLDEN_FRAME || + edge_mi->ref_frame[1] == GOLDEN_FRAME); + } else { // no edges available (2) + pred_context = 2; + } + assert(pred_context >= 0 && pred_context < REF_CONTEXTS); + return pred_context; +} diff --git a/thirdparty/libvpx/vp9/common/vp9_pred_common.h b/thirdparty/libvpx/vp9/common/vp9_pred_common.h new file mode 100644 index 0000000000..f3c676e953 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_pred_common.h @@ -0,0 +1,192 @@ +/* + * Copyright (c) 2012 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_COMMON_VP9_PRED_COMMON_H_ +#define VP9_COMMON_VP9_PRED_COMMON_H_ + +#include "vp9/common/vp9_blockd.h" +#include "vp9/common/vp9_onyxc_int.h" +#include "vpx_dsp/vpx_dsp_common.h" + +#ifdef __cplusplus +extern "C" { +#endif + +static INLINE int get_segment_id(const VP9_COMMON *cm, + const uint8_t *segment_ids, + BLOCK_SIZE bsize, int mi_row, int mi_col) { + const int mi_offset = mi_row * cm->mi_cols + mi_col; + const int bw = num_8x8_blocks_wide_lookup[bsize]; + const int bh = num_8x8_blocks_high_lookup[bsize]; + const int xmis = VPXMIN(cm->mi_cols - mi_col, bw); + const int ymis = VPXMIN(cm->mi_rows - mi_row, bh); + int x, y, segment_id = MAX_SEGMENTS; + + for (y = 0; y < ymis; ++y) + for (x = 0; x < xmis; ++x) + segment_id = + VPXMIN(segment_id, segment_ids[mi_offset + y * cm->mi_cols + x]); + + assert(segment_id >= 0 && segment_id < MAX_SEGMENTS); + return segment_id; +} + +static INLINE int vp9_get_pred_context_seg_id(const MACROBLOCKD *xd) { + const MODE_INFO *const above_mi = xd->above_mi; + const MODE_INFO *const left_mi = xd->left_mi; + const int above_sip = (above_mi != NULL) ? + above_mi->seg_id_predicted : 0; + const int left_sip = (left_mi != NULL) ? left_mi->seg_id_predicted : 0; + + return above_sip + left_sip; +} + +static INLINE vpx_prob vp9_get_pred_prob_seg_id(const struct segmentation *seg, + const MACROBLOCKD *xd) { + return seg->pred_probs[vp9_get_pred_context_seg_id(xd)]; +} + +static INLINE int vp9_get_skip_context(const MACROBLOCKD *xd) { + const MODE_INFO *const above_mi = xd->above_mi; + const MODE_INFO *const left_mi = xd->left_mi; + const int above_skip = (above_mi != NULL) ? above_mi->skip : 0; + const int left_skip = (left_mi != NULL) ? left_mi->skip : 0; + return above_skip + left_skip; +} + +static INLINE vpx_prob vp9_get_skip_prob(const VP9_COMMON *cm, + const MACROBLOCKD *xd) { + return cm->fc->skip_probs[vp9_get_skip_context(xd)]; +} + +int vp9_get_pred_context_switchable_interp(const MACROBLOCKD *xd); + +// The mode info data structure has a one element border above and to the +// left of the entries corresponding to real macroblocks. +// The prediction flags in these dummy entries are initialized to 0. +// 0 - inter/inter, inter/--, --/inter, --/-- +// 1 - intra/inter, inter/intra +// 2 - intra/--, --/intra +// 3 - intra/intra +static INLINE int get_intra_inter_context(const MACROBLOCKD *xd) { + const MODE_INFO *const above_mi = xd->above_mi; + const MODE_INFO *const left_mi = xd->left_mi; + const int has_above = !!above_mi; + const int has_left = !!left_mi; + + if (has_above && has_left) { // both edges available + const int above_intra = !is_inter_block(above_mi); + const int left_intra = !is_inter_block(left_mi); + return left_intra && above_intra ? 3 : left_intra || above_intra; + } else if (has_above || has_left) { // one edge available + return 2 * !is_inter_block(has_above ? above_mi : left_mi); + } + return 0; +} + +static INLINE vpx_prob vp9_get_intra_inter_prob(const VP9_COMMON *cm, + const MACROBLOCKD *xd) { + return cm->fc->intra_inter_prob[get_intra_inter_context(xd)]; +} + +int vp9_get_reference_mode_context(const VP9_COMMON *cm, const MACROBLOCKD *xd); + +static INLINE vpx_prob vp9_get_reference_mode_prob(const VP9_COMMON *cm, + const MACROBLOCKD *xd) { + return cm->fc->comp_inter_prob[vp9_get_reference_mode_context(cm, xd)]; +} + +int vp9_get_pred_context_comp_ref_p(const VP9_COMMON *cm, + const MACROBLOCKD *xd); + +static INLINE vpx_prob vp9_get_pred_prob_comp_ref_p(const VP9_COMMON *cm, + const MACROBLOCKD *xd) { + const int pred_context = vp9_get_pred_context_comp_ref_p(cm, xd); + return cm->fc->comp_ref_prob[pred_context]; +} + +int vp9_get_pred_context_single_ref_p1(const MACROBLOCKD *xd); + +static INLINE vpx_prob vp9_get_pred_prob_single_ref_p1(const VP9_COMMON *cm, + const MACROBLOCKD *xd) { + return cm->fc->single_ref_prob[vp9_get_pred_context_single_ref_p1(xd)][0]; +} + +int vp9_get_pred_context_single_ref_p2(const MACROBLOCKD *xd); + +static INLINE vpx_prob vp9_get_pred_prob_single_ref_p2(const VP9_COMMON *cm, + const MACROBLOCKD *xd) { + return cm->fc->single_ref_prob[vp9_get_pred_context_single_ref_p2(xd)][1]; +} + +// Returns a context number for the given MB prediction signal +// The mode info data structure has a one element border above and to the +// left of the entries corresponding to real blocks. +// The prediction flags in these dummy entries are initialized to 0. +static INLINE int get_tx_size_context(const MACROBLOCKD *xd) { + const int max_tx_size = max_txsize_lookup[xd->mi[0]->sb_type]; + const MODE_INFO *const above_mi = xd->above_mi; + const MODE_INFO *const left_mi = xd->left_mi; + const int has_above = !!above_mi; + const int has_left = !!left_mi; + int above_ctx = (has_above && !above_mi->skip) ? (int)above_mi->tx_size + : max_tx_size; + int left_ctx = (has_left && !left_mi->skip) ? (int)left_mi->tx_size + : max_tx_size; + if (!has_left) + left_ctx = above_ctx; + + if (!has_above) + above_ctx = left_ctx; + + return (above_ctx + left_ctx) > max_tx_size; +} + +static INLINE const vpx_prob *get_tx_probs(TX_SIZE max_tx_size, int ctx, + const struct tx_probs *tx_probs) { + switch (max_tx_size) { + case TX_8X8: + return tx_probs->p8x8[ctx]; + case TX_16X16: + return tx_probs->p16x16[ctx]; + case TX_32X32: + return tx_probs->p32x32[ctx]; + default: + assert(0 && "Invalid max_tx_size."); + return NULL; + } +} + +static INLINE const vpx_prob *get_tx_probs2(TX_SIZE max_tx_size, + const MACROBLOCKD *xd, + const struct tx_probs *tx_probs) { + return get_tx_probs(max_tx_size, get_tx_size_context(xd), tx_probs); +} + +static INLINE unsigned int *get_tx_counts(TX_SIZE max_tx_size, int ctx, + struct tx_counts *tx_counts) { + switch (max_tx_size) { + case TX_8X8: + return tx_counts->p8x8[ctx]; + case TX_16X16: + return tx_counts->p16x16[ctx]; + case TX_32X32: + return tx_counts->p32x32[ctx]; + default: + assert(0 && "Invalid max_tx_size."); + return NULL; + } +} + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_PRED_COMMON_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_quant_common.c b/thirdparty/libvpx/vp9/common/vp9_quant_common.c new file mode 100644 index 0000000000..d83f3c1a2f --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_quant_common.c @@ -0,0 +1,278 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "vp9/common/vp9_common.h" +#include "vp9/common/vp9_quant_common.h" +#include "vp9/common/vp9_seg_common.h" + +static const int16_t dc_qlookup[QINDEX_RANGE] = { + 4, 8, 8, 9, 10, 11, 12, 12, + 13, 14, 15, 16, 17, 18, 19, 19, + 20, 21, 22, 23, 24, 25, 26, 26, + 27, 28, 29, 30, 31, 32, 32, 33, + 34, 35, 36, 37, 38, 38, 39, 40, + 41, 42, 43, 43, 44, 45, 46, 47, + 48, 48, 49, 50, 51, 52, 53, 53, + 54, 55, 56, 57, 57, 58, 59, 60, + 61, 62, 62, 63, 64, 65, 66, 66, + 67, 68, 69, 70, 70, 71, 72, 73, + 74, 74, 75, 76, 77, 78, 78, 79, + 80, 81, 81, 82, 83, 84, 85, 85, + 87, 88, 90, 92, 93, 95, 96, 98, + 99, 101, 102, 104, 105, 107, 108, 110, + 111, 113, 114, 116, 117, 118, 120, 121, + 123, 125, 127, 129, 131, 134, 136, 138, + 140, 142, 144, 146, 148, 150, 152, 154, + 156, 158, 161, 164, 166, 169, 172, 174, + 177, 180, 182, 185, 187, 190, 192, 195, + 199, 202, 205, 208, 211, 214, 217, 220, + 223, 226, 230, 233, 237, 240, 243, 247, + 250, 253, 257, 261, 265, 269, 272, 276, + 280, 284, 288, 292, 296, 300, 304, 309, + 313, 317, 322, 326, 330, 335, 340, 344, + 349, 354, 359, 364, 369, 374, 379, 384, + 389, 395, 400, 406, 411, 417, 423, 429, + 435, 441, 447, 454, 461, 467, 475, 482, + 489, 497, 505, 513, 522, 530, 539, 549, + 559, 569, 579, 590, 602, 614, 626, 640, + 654, 668, 684, 700, 717, 736, 755, 775, + 796, 819, 843, 869, 896, 925, 955, 988, + 1022, 1058, 1098, 1139, 1184, 1232, 1282, 1336, +}; + +#if CONFIG_VP9_HIGHBITDEPTH +static const int16_t dc_qlookup_10[QINDEX_RANGE] = { + 4, 9, 10, 13, 15, 17, 20, 22, + 25, 28, 31, 34, 37, 40, 43, 47, + 50, 53, 57, 60, 64, 68, 71, 75, + 78, 82, 86, 90, 93, 97, 101, 105, + 109, 113, 116, 120, 124, 128, 132, 136, + 140, 143, 147, 151, 155, 159, 163, 166, + 170, 174, 178, 182, 185, 189, 193, 197, + 200, 204, 208, 212, 215, 219, 223, 226, + 230, 233, 237, 241, 244, 248, 251, 255, + 259, 262, 266, 269, 273, 276, 280, 283, + 287, 290, 293, 297, 300, 304, 307, 310, + 314, 317, 321, 324, 327, 331, 334, 337, + 343, 350, 356, 362, 369, 375, 381, 387, + 394, 400, 406, 412, 418, 424, 430, 436, + 442, 448, 454, 460, 466, 472, 478, 484, + 490, 499, 507, 516, 525, 533, 542, 550, + 559, 567, 576, 584, 592, 601, 609, 617, + 625, 634, 644, 655, 666, 676, 687, 698, + 708, 718, 729, 739, 749, 759, 770, 782, + 795, 807, 819, 831, 844, 856, 868, 880, + 891, 906, 920, 933, 947, 961, 975, 988, + 1001, 1015, 1030, 1045, 1061, 1076, 1090, 1105, + 1120, 1137, 1153, 1170, 1186, 1202, 1218, 1236, + 1253, 1271, 1288, 1306, 1323, 1342, 1361, 1379, + 1398, 1416, 1436, 1456, 1476, 1496, 1516, 1537, + 1559, 1580, 1601, 1624, 1647, 1670, 1692, 1717, + 1741, 1766, 1791, 1817, 1844, 1871, 1900, 1929, + 1958, 1990, 2021, 2054, 2088, 2123, 2159, 2197, + 2236, 2276, 2319, 2363, 2410, 2458, 2508, 2561, + 2616, 2675, 2737, 2802, 2871, 2944, 3020, 3102, + 3188, 3280, 3375, 3478, 3586, 3702, 3823, 3953, + 4089, 4236, 4394, 4559, 4737, 4929, 5130, 5347, +}; + +static const int16_t dc_qlookup_12[QINDEX_RANGE] = { + 4, 12, 18, 25, 33, 41, 50, 60, + 70, 80, 91, 103, 115, 127, 140, 153, + 166, 180, 194, 208, 222, 237, 251, 266, + 281, 296, 312, 327, 343, 358, 374, 390, + 405, 421, 437, 453, 469, 484, 500, 516, + 532, 548, 564, 580, 596, 611, 627, 643, + 659, 674, 690, 706, 721, 737, 752, 768, + 783, 798, 814, 829, 844, 859, 874, 889, + 904, 919, 934, 949, 964, 978, 993, 1008, + 1022, 1037, 1051, 1065, 1080, 1094, 1108, 1122, + 1136, 1151, 1165, 1179, 1192, 1206, 1220, 1234, + 1248, 1261, 1275, 1288, 1302, 1315, 1329, 1342, + 1368, 1393, 1419, 1444, 1469, 1494, 1519, 1544, + 1569, 1594, 1618, 1643, 1668, 1692, 1717, 1741, + 1765, 1789, 1814, 1838, 1862, 1885, 1909, 1933, + 1957, 1992, 2027, 2061, 2096, 2130, 2165, 2199, + 2233, 2267, 2300, 2334, 2367, 2400, 2434, 2467, + 2499, 2532, 2575, 2618, 2661, 2704, 2746, 2788, + 2830, 2872, 2913, 2954, 2995, 3036, 3076, 3127, + 3177, 3226, 3275, 3324, 3373, 3421, 3469, 3517, + 3565, 3621, 3677, 3733, 3788, 3843, 3897, 3951, + 4005, 4058, 4119, 4181, 4241, 4301, 4361, 4420, + 4479, 4546, 4612, 4677, 4742, 4807, 4871, 4942, + 5013, 5083, 5153, 5222, 5291, 5367, 5442, 5517, + 5591, 5665, 5745, 5825, 5905, 5984, 6063, 6149, + 6234, 6319, 6404, 6495, 6587, 6678, 6769, 6867, + 6966, 7064, 7163, 7269, 7376, 7483, 7599, 7715, + 7832, 7958, 8085, 8214, 8352, 8492, 8635, 8788, + 8945, 9104, 9275, 9450, 9639, 9832, 10031, 10245, + 10465, 10702, 10946, 11210, 11482, 11776, 12081, 12409, + 12750, 13118, 13501, 13913, 14343, 14807, 15290, 15812, + 16356, 16943, 17575, 18237, 18949, 19718, 20521, 21387, +}; +#endif + +static const int16_t ac_qlookup[QINDEX_RANGE] = { + 4, 8, 9, 10, 11, 12, 13, 14, + 15, 16, 17, 18, 19, 20, 21, 22, + 23, 24, 25, 26, 27, 28, 29, 30, + 31, 32, 33, 34, 35, 36, 37, 38, + 39, 40, 41, 42, 43, 44, 45, 46, + 47, 48, 49, 50, 51, 52, 53, 54, + 55, 56, 57, 58, 59, 60, 61, 62, + 63, 64, 65, 66, 67, 68, 69, 70, + 71, 72, 73, 74, 75, 76, 77, 78, + 79, 80, 81, 82, 83, 84, 85, 86, + 87, 88, 89, 90, 91, 92, 93, 94, + 95, 96, 97, 98, 99, 100, 101, 102, + 104, 106, 108, 110, 112, 114, 116, 118, + 120, 122, 124, 126, 128, 130, 132, 134, + 136, 138, 140, 142, 144, 146, 148, 150, + 152, 155, 158, 161, 164, 167, 170, 173, + 176, 179, 182, 185, 188, 191, 194, 197, + 200, 203, 207, 211, 215, 219, 223, 227, + 231, 235, 239, 243, 247, 251, 255, 260, + 265, 270, 275, 280, 285, 290, 295, 300, + 305, 311, 317, 323, 329, 335, 341, 347, + 353, 359, 366, 373, 380, 387, 394, 401, + 408, 416, 424, 432, 440, 448, 456, 465, + 474, 483, 492, 501, 510, 520, 530, 540, + 550, 560, 571, 582, 593, 604, 615, 627, + 639, 651, 663, 676, 689, 702, 715, 729, + 743, 757, 771, 786, 801, 816, 832, 848, + 864, 881, 898, 915, 933, 951, 969, 988, + 1007, 1026, 1046, 1066, 1087, 1108, 1129, 1151, + 1173, 1196, 1219, 1243, 1267, 1292, 1317, 1343, + 1369, 1396, 1423, 1451, 1479, 1508, 1537, 1567, + 1597, 1628, 1660, 1692, 1725, 1759, 1793, 1828, +}; + +#if CONFIG_VP9_HIGHBITDEPTH +static const int16_t ac_qlookup_10[QINDEX_RANGE] = { + 4, 9, 11, 13, 16, 18, 21, 24, + 27, 30, 33, 37, 40, 44, 48, 51, + 55, 59, 63, 67, 71, 75, 79, 83, + 88, 92, 96, 100, 105, 109, 114, 118, + 122, 127, 131, 136, 140, 145, 149, 154, + 158, 163, 168, 172, 177, 181, 186, 190, + 195, 199, 204, 208, 213, 217, 222, 226, + 231, 235, 240, 244, 249, 253, 258, 262, + 267, 271, 275, 280, 284, 289, 293, 297, + 302, 306, 311, 315, 319, 324, 328, 332, + 337, 341, 345, 349, 354, 358, 362, 367, + 371, 375, 379, 384, 388, 392, 396, 401, + 409, 417, 425, 433, 441, 449, 458, 466, + 474, 482, 490, 498, 506, 514, 523, 531, + 539, 547, 555, 563, 571, 579, 588, 596, + 604, 616, 628, 640, 652, 664, 676, 688, + 700, 713, 725, 737, 749, 761, 773, 785, + 797, 809, 825, 841, 857, 873, 889, 905, + 922, 938, 954, 970, 986, 1002, 1018, 1038, + 1058, 1078, 1098, 1118, 1138, 1158, 1178, 1198, + 1218, 1242, 1266, 1290, 1314, 1338, 1362, 1386, + 1411, 1435, 1463, 1491, 1519, 1547, 1575, 1603, + 1631, 1663, 1695, 1727, 1759, 1791, 1823, 1859, + 1895, 1931, 1967, 2003, 2039, 2079, 2119, 2159, + 2199, 2239, 2283, 2327, 2371, 2415, 2459, 2507, + 2555, 2603, 2651, 2703, 2755, 2807, 2859, 2915, + 2971, 3027, 3083, 3143, 3203, 3263, 3327, 3391, + 3455, 3523, 3591, 3659, 3731, 3803, 3876, 3952, + 4028, 4104, 4184, 4264, 4348, 4432, 4516, 4604, + 4692, 4784, 4876, 4972, 5068, 5168, 5268, 5372, + 5476, 5584, 5692, 5804, 5916, 6032, 6148, 6268, + 6388, 6512, 6640, 6768, 6900, 7036, 7172, 7312, +}; + +static const int16_t ac_qlookup_12[QINDEX_RANGE] = { + 4, 13, 19, 27, 35, 44, 54, 64, + 75, 87, 99, 112, 126, 139, 154, 168, + 183, 199, 214, 230, 247, 263, 280, 297, + 314, 331, 349, 366, 384, 402, 420, 438, + 456, 475, 493, 511, 530, 548, 567, 586, + 604, 623, 642, 660, 679, 698, 716, 735, + 753, 772, 791, 809, 828, 846, 865, 884, + 902, 920, 939, 957, 976, 994, 1012, 1030, + 1049, 1067, 1085, 1103, 1121, 1139, 1157, 1175, + 1193, 1211, 1229, 1246, 1264, 1282, 1299, 1317, + 1335, 1352, 1370, 1387, 1405, 1422, 1440, 1457, + 1474, 1491, 1509, 1526, 1543, 1560, 1577, 1595, + 1627, 1660, 1693, 1725, 1758, 1791, 1824, 1856, + 1889, 1922, 1954, 1987, 2020, 2052, 2085, 2118, + 2150, 2183, 2216, 2248, 2281, 2313, 2346, 2378, + 2411, 2459, 2508, 2556, 2605, 2653, 2701, 2750, + 2798, 2847, 2895, 2943, 2992, 3040, 3088, 3137, + 3185, 3234, 3298, 3362, 3426, 3491, 3555, 3619, + 3684, 3748, 3812, 3876, 3941, 4005, 4069, 4149, + 4230, 4310, 4390, 4470, 4550, 4631, 4711, 4791, + 4871, 4967, 5064, 5160, 5256, 5352, 5448, 5544, + 5641, 5737, 5849, 5961, 6073, 6185, 6297, 6410, + 6522, 6650, 6778, 6906, 7034, 7162, 7290, 7435, + 7579, 7723, 7867, 8011, 8155, 8315, 8475, 8635, + 8795, 8956, 9132, 9308, 9484, 9660, 9836, 10028, + 10220, 10412, 10604, 10812, 11020, 11228, 11437, 11661, + 11885, 12109, 12333, 12573, 12813, 13053, 13309, 13565, + 13821, 14093, 14365, 14637, 14925, 15213, 15502, 15806, + 16110, 16414, 16734, 17054, 17390, 17726, 18062, 18414, + 18766, 19134, 19502, 19886, 20270, 20670, 21070, 21486, + 21902, 22334, 22766, 23214, 23662, 24126, 24590, 25070, + 25551, 26047, 26559, 27071, 27599, 28143, 28687, 29247, +}; +#endif + +int16_t vp9_dc_quant(int qindex, int delta, vpx_bit_depth_t bit_depth) { +#if CONFIG_VP9_HIGHBITDEPTH + switch (bit_depth) { + case VPX_BITS_8: + return dc_qlookup[clamp(qindex + delta, 0, MAXQ)]; + case VPX_BITS_10: + return dc_qlookup_10[clamp(qindex + delta, 0, MAXQ)]; + case VPX_BITS_12: + return dc_qlookup_12[clamp(qindex + delta, 0, MAXQ)]; + default: + assert(0 && "bit_depth should be VPX_BITS_8, VPX_BITS_10 or VPX_BITS_12"); + return -1; + } +#else + (void) bit_depth; + return dc_qlookup[clamp(qindex + delta, 0, MAXQ)]; +#endif +} + +int16_t vp9_ac_quant(int qindex, int delta, vpx_bit_depth_t bit_depth) { +#if CONFIG_VP9_HIGHBITDEPTH + switch (bit_depth) { + case VPX_BITS_8: + return ac_qlookup[clamp(qindex + delta, 0, MAXQ)]; + case VPX_BITS_10: + return ac_qlookup_10[clamp(qindex + delta, 0, MAXQ)]; + case VPX_BITS_12: + return ac_qlookup_12[clamp(qindex + delta, 0, MAXQ)]; + default: + assert(0 && "bit_depth should be VPX_BITS_8, VPX_BITS_10 or VPX_BITS_12"); + return -1; + } +#else + (void) bit_depth; + return ac_qlookup[clamp(qindex + delta, 0, MAXQ)]; +#endif +} + +int vp9_get_qindex(const struct segmentation *seg, int segment_id, + int base_qindex) { + if (segfeature_active(seg, segment_id, SEG_LVL_ALT_Q)) { + const int data = get_segdata(seg, segment_id, SEG_LVL_ALT_Q); + const int seg_qindex = seg->abs_delta == SEGMENT_ABSDATA ? + data : base_qindex + data; + return clamp(seg_qindex, 0, MAXQ); + } else { + return base_qindex; + } +} + diff --git a/thirdparty/libvpx/vp9/common/vp9_quant_common.h b/thirdparty/libvpx/vp9/common/vp9_quant_common.h new file mode 100644 index 0000000000..4bae4a8967 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_quant_common.h @@ -0,0 +1,36 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_COMMON_VP9_QUANT_COMMON_H_ +#define VP9_COMMON_VP9_QUANT_COMMON_H_ + +#include "vpx/vpx_codec.h" +#include "vp9/common/vp9_seg_common.h" + +#ifdef __cplusplus +extern "C" { +#endif + +#define MINQ 0 +#define MAXQ 255 +#define QINDEX_RANGE (MAXQ - MINQ + 1) +#define QINDEX_BITS 8 + +int16_t vp9_dc_quant(int qindex, int delta, vpx_bit_depth_t bit_depth); +int16_t vp9_ac_quant(int qindex, int delta, vpx_bit_depth_t bit_depth); + +int vp9_get_qindex(const struct segmentation *seg, int segment_id, + int base_qindex); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_QUANT_COMMON_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_reconinter.c b/thirdparty/libvpx/vp9/common/vp9_reconinter.c new file mode 100644 index 0000000000..84718e9703 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_reconinter.c @@ -0,0 +1,309 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <assert.h> + +#include "./vpx_scale_rtcd.h" +#include "./vpx_config.h" + +#include "vpx/vpx_integer.h" + +#include "vp9/common/vp9_blockd.h" +#include "vp9/common/vp9_reconinter.h" +#include "vp9/common/vp9_reconintra.h" + +#if CONFIG_VP9_HIGHBITDEPTH +void vp9_highbd_build_inter_predictor(const uint8_t *src, int src_stride, + uint8_t *dst, int dst_stride, + const MV *src_mv, + const struct scale_factors *sf, + int w, int h, int ref, + const InterpKernel *kernel, + enum mv_precision precision, + int x, int y, int bd) { + const int is_q4 = precision == MV_PRECISION_Q4; + const MV mv_q4 = { is_q4 ? src_mv->row : src_mv->row * 2, + is_q4 ? src_mv->col : src_mv->col * 2 }; + MV32 mv = vp9_scale_mv(&mv_q4, x, y, sf); + const int subpel_x = mv.col & SUBPEL_MASK; + const int subpel_y = mv.row & SUBPEL_MASK; + + src += (mv.row >> SUBPEL_BITS) * src_stride + (mv.col >> SUBPEL_BITS); + + highbd_inter_predictor(src, src_stride, dst, dst_stride, subpel_x, subpel_y, + sf, w, h, ref, kernel, sf->x_step_q4, sf->y_step_q4, + bd); +} +#endif // CONFIG_VP9_HIGHBITDEPTH + +void vp9_build_inter_predictor(const uint8_t *src, int src_stride, + uint8_t *dst, int dst_stride, + const MV *src_mv, + const struct scale_factors *sf, + int w, int h, int ref, + const InterpKernel *kernel, + enum mv_precision precision, + int x, int y) { + const int is_q4 = precision == MV_PRECISION_Q4; + const MV mv_q4 = { is_q4 ? src_mv->row : src_mv->row * 2, + is_q4 ? src_mv->col : src_mv->col * 2 }; + MV32 mv = vp9_scale_mv(&mv_q4, x, y, sf); + const int subpel_x = mv.col & SUBPEL_MASK; + const int subpel_y = mv.row & SUBPEL_MASK; + + src += (mv.row >> SUBPEL_BITS) * src_stride + (mv.col >> SUBPEL_BITS); + + inter_predictor(src, src_stride, dst, dst_stride, subpel_x, subpel_y, + sf, w, h, ref, kernel, sf->x_step_q4, sf->y_step_q4); +} + +static INLINE int round_mv_comp_q4(int value) { + return (value < 0 ? value - 2 : value + 2) / 4; +} + +static MV mi_mv_pred_q4(const MODE_INFO *mi, int idx) { + MV res = { round_mv_comp_q4(mi->bmi[0].as_mv[idx].as_mv.row + + mi->bmi[1].as_mv[idx].as_mv.row + + mi->bmi[2].as_mv[idx].as_mv.row + + mi->bmi[3].as_mv[idx].as_mv.row), + round_mv_comp_q4(mi->bmi[0].as_mv[idx].as_mv.col + + mi->bmi[1].as_mv[idx].as_mv.col + + mi->bmi[2].as_mv[idx].as_mv.col + + mi->bmi[3].as_mv[idx].as_mv.col) }; + return res; +} + +static INLINE int round_mv_comp_q2(int value) { + return (value < 0 ? value - 1 : value + 1) / 2; +} + +static MV mi_mv_pred_q2(const MODE_INFO *mi, int idx, int block0, int block1) { + MV res = { round_mv_comp_q2(mi->bmi[block0].as_mv[idx].as_mv.row + + mi->bmi[block1].as_mv[idx].as_mv.row), + round_mv_comp_q2(mi->bmi[block0].as_mv[idx].as_mv.col + + mi->bmi[block1].as_mv[idx].as_mv.col) }; + return res; +} + +// TODO(jkoleszar): yet another mv clamping function :-( +MV clamp_mv_to_umv_border_sb(const MACROBLOCKD *xd, const MV *src_mv, + int bw, int bh, int ss_x, int ss_y) { + // If the MV points so far into the UMV border that no visible pixels + // are used for reconstruction, the subpel part of the MV can be + // discarded and the MV limited to 16 pixels with equivalent results. + const int spel_left = (VP9_INTERP_EXTEND + bw) << SUBPEL_BITS; + const int spel_right = spel_left - SUBPEL_SHIFTS; + const int spel_top = (VP9_INTERP_EXTEND + bh) << SUBPEL_BITS; + const int spel_bottom = spel_top - SUBPEL_SHIFTS; + MV clamped_mv = { + src_mv->row * (1 << (1 - ss_y)), + src_mv->col * (1 << (1 - ss_x)) + }; + assert(ss_x <= 1); + assert(ss_y <= 1); + + clamp_mv(&clamped_mv, + xd->mb_to_left_edge * (1 << (1 - ss_x)) - spel_left, + xd->mb_to_right_edge * (1 << (1 - ss_x)) + spel_right, + xd->mb_to_top_edge * (1 << (1 - ss_y)) - spel_top, + xd->mb_to_bottom_edge * (1 << (1 - ss_y)) + spel_bottom); + + return clamped_mv; +} + +MV average_split_mvs(const struct macroblockd_plane *pd, + const MODE_INFO *mi, int ref, int block) { + const int ss_idx = ((pd->subsampling_x > 0) << 1) | (pd->subsampling_y > 0); + MV res = {0, 0}; + switch (ss_idx) { + case 0: + res = mi->bmi[block].as_mv[ref].as_mv; + break; + case 1: + res = mi_mv_pred_q2(mi, ref, block, block + 2); + break; + case 2: + res = mi_mv_pred_q2(mi, ref, block, block + 1); + break; + case 3: + res = mi_mv_pred_q4(mi, ref); + break; + default: + assert(ss_idx <= 3 && ss_idx >= 0); + } + return res; +} + +static void build_inter_predictors(MACROBLOCKD *xd, int plane, int block, + int bw, int bh, + int x, int y, int w, int h, + int mi_x, int mi_y) { + struct macroblockd_plane *const pd = &xd->plane[plane]; + const MODE_INFO *mi = xd->mi[0]; + const int is_compound = has_second_ref(mi); + const InterpKernel *kernel = vp9_filter_kernels[mi->interp_filter]; + int ref; + + for (ref = 0; ref < 1 + is_compound; ++ref) { + const struct scale_factors *const sf = &xd->block_refs[ref]->sf; + struct buf_2d *const pre_buf = &pd->pre[ref]; + struct buf_2d *const dst_buf = &pd->dst; + uint8_t *const dst = dst_buf->buf + dst_buf->stride * y + x; + const MV mv = mi->sb_type < BLOCK_8X8 + ? average_split_mvs(pd, mi, ref, block) + : mi->mv[ref].as_mv; + + // TODO(jkoleszar): This clamping is done in the incorrect place for the + // scaling case. It needs to be done on the scaled MV, not the pre-scaling + // MV. Note however that it performs the subsampling aware scaling so + // that the result is always q4. + // mv_precision precision is MV_PRECISION_Q4. + const MV mv_q4 = clamp_mv_to_umv_border_sb(xd, &mv, bw, bh, + pd->subsampling_x, + pd->subsampling_y); + + uint8_t *pre; + MV32 scaled_mv; + int xs, ys, subpel_x, subpel_y; + const int is_scaled = vp9_is_scaled(sf); + + if (is_scaled) { + // Co-ordinate of containing block to pixel precision. + const int x_start = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x)); + const int y_start = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y)); +#if CONFIG_BETTER_HW_COMPATIBILITY + assert(xd->mi[0]->sb_type != BLOCK_4X8 && + xd->mi[0]->sb_type != BLOCK_8X4); + assert(mv_q4.row == mv.row * (1 << (1 - pd->subsampling_y)) && + mv_q4.col == mv.col * (1 << (1 - pd->subsampling_x))); +#endif + if (plane == 0) + pre_buf->buf = xd->block_refs[ref]->buf->y_buffer; + else if (plane == 1) + pre_buf->buf = xd->block_refs[ref]->buf->u_buffer; + else + pre_buf->buf = xd->block_refs[ref]->buf->v_buffer; + + pre_buf->buf += scaled_buffer_offset(x_start + x, y_start + y, + pre_buf->stride, sf); + pre = pre_buf->buf; + scaled_mv = vp9_scale_mv(&mv_q4, mi_x + x, mi_y + y, sf); + xs = sf->x_step_q4; + ys = sf->y_step_q4; + } else { + pre = pre_buf->buf + (y * pre_buf->stride + x); + scaled_mv.row = mv_q4.row; + scaled_mv.col = mv_q4.col; + xs = ys = 16; + } + subpel_x = scaled_mv.col & SUBPEL_MASK; + subpel_y = scaled_mv.row & SUBPEL_MASK; + pre += (scaled_mv.row >> SUBPEL_BITS) * pre_buf->stride + + (scaled_mv.col >> SUBPEL_BITS); + +#if CONFIG_VP9_HIGHBITDEPTH + if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) { + highbd_inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride, + subpel_x, subpel_y, sf, w, h, ref, kernel, xs, ys, + xd->bd); + } else { + inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride, + subpel_x, subpel_y, sf, w, h, ref, kernel, xs, ys); + } +#else + inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride, + subpel_x, subpel_y, sf, w, h, ref, kernel, xs, ys); +#endif // CONFIG_VP9_HIGHBITDEPTH + } +} + +static void build_inter_predictors_for_planes(MACROBLOCKD *xd, BLOCK_SIZE bsize, + int mi_row, int mi_col, + int plane_from, int plane_to) { + int plane; + const int mi_x = mi_col * MI_SIZE; + const int mi_y = mi_row * MI_SIZE; + for (plane = plane_from; plane <= plane_to; ++plane) { + const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, + &xd->plane[plane]); + const int num_4x4_w = num_4x4_blocks_wide_lookup[plane_bsize]; + const int num_4x4_h = num_4x4_blocks_high_lookup[plane_bsize]; + const int bw = 4 * num_4x4_w; + const int bh = 4 * num_4x4_h; + + if (xd->mi[0]->sb_type < BLOCK_8X8) { + int i = 0, x, y; + assert(bsize == BLOCK_8X8); + for (y = 0; y < num_4x4_h; ++y) + for (x = 0; x < num_4x4_w; ++x) + build_inter_predictors(xd, plane, i++, bw, bh, + 4 * x, 4 * y, 4, 4, mi_x, mi_y); + } else { + build_inter_predictors(xd, plane, 0, bw, bh, + 0, 0, bw, bh, mi_x, mi_y); + } + } +} + +void vp9_build_inter_predictors_sby(MACROBLOCKD *xd, int mi_row, int mi_col, + BLOCK_SIZE bsize) { + build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, 0, 0); +} + +void vp9_build_inter_predictors_sbp(MACROBLOCKD *xd, int mi_row, int mi_col, + BLOCK_SIZE bsize, int plane) { + build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, plane, plane); +} + +void vp9_build_inter_predictors_sbuv(MACROBLOCKD *xd, int mi_row, int mi_col, + BLOCK_SIZE bsize) { + build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, 1, + MAX_MB_PLANE - 1); +} + +void vp9_build_inter_predictors_sb(MACROBLOCKD *xd, int mi_row, int mi_col, + BLOCK_SIZE bsize) { + build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, 0, + MAX_MB_PLANE - 1); +} + +void vp9_setup_dst_planes(struct macroblockd_plane planes[MAX_MB_PLANE], + const YV12_BUFFER_CONFIG *src, + int mi_row, int mi_col) { + uint8_t *const buffers[MAX_MB_PLANE] = { src->y_buffer, src->u_buffer, + src->v_buffer}; + const int strides[MAX_MB_PLANE] = { src->y_stride, src->uv_stride, + src->uv_stride}; + int i; + + for (i = 0; i < MAX_MB_PLANE; ++i) { + struct macroblockd_plane *const pd = &planes[i]; + setup_pred_plane(&pd->dst, buffers[i], strides[i], mi_row, mi_col, NULL, + pd->subsampling_x, pd->subsampling_y); + } +} + +void vp9_setup_pre_planes(MACROBLOCKD *xd, int idx, + const YV12_BUFFER_CONFIG *src, + int mi_row, int mi_col, + const struct scale_factors *sf) { + if (src != NULL) { + int i; + uint8_t *const buffers[MAX_MB_PLANE] = { src->y_buffer, src->u_buffer, + src->v_buffer}; + const int strides[MAX_MB_PLANE] = { src->y_stride, src->uv_stride, + src->uv_stride}; + for (i = 0; i < MAX_MB_PLANE; ++i) { + struct macroblockd_plane *const pd = &xd->plane[i]; + setup_pred_plane(&pd->pre[idx], buffers[i], strides[i], mi_row, mi_col, + sf, pd->subsampling_x, pd->subsampling_y); + } + } +} diff --git a/thirdparty/libvpx/vp9/common/vp9_reconinter.h b/thirdparty/libvpx/vp9/common/vp9_reconinter.h new file mode 100644 index 0000000000..07745e3aaa --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_reconinter.h @@ -0,0 +1,119 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_COMMON_VP9_RECONINTER_H_ +#define VP9_COMMON_VP9_RECONINTER_H_ + +#include "vp9/common/vp9_filter.h" +#include "vp9/common/vp9_onyxc_int.h" +#include "vpx/vpx_integer.h" +#include "vpx_dsp/vpx_filter.h" + +#ifdef __cplusplus +extern "C" { +#endif + +static INLINE void inter_predictor(const uint8_t *src, int src_stride, + uint8_t *dst, int dst_stride, + const int subpel_x, + const int subpel_y, + const struct scale_factors *sf, + int w, int h, int ref, + const InterpKernel *kernel, + int xs, int ys) { + sf->predict[subpel_x != 0][subpel_y != 0][ref]( + src, src_stride, dst, dst_stride, + kernel[subpel_x], xs, kernel[subpel_y], ys, w, h); +} + +#if CONFIG_VP9_HIGHBITDEPTH +static INLINE void highbd_inter_predictor(const uint8_t *src, int src_stride, + uint8_t *dst, int dst_stride, + const int subpel_x, + const int subpel_y, + const struct scale_factors *sf, + int w, int h, int ref, + const InterpKernel *kernel, + int xs, int ys, int bd) { + sf->highbd_predict[subpel_x != 0][subpel_y != 0][ref]( + src, src_stride, dst, dst_stride, + kernel[subpel_x], xs, kernel[subpel_y], ys, w, h, bd); +} +#endif // CONFIG_VP9_HIGHBITDEPTH + +MV average_split_mvs(const struct macroblockd_plane *pd, const MODE_INFO *mi, + int ref, int block); + +MV clamp_mv_to_umv_border_sb(const MACROBLOCKD *xd, const MV *src_mv, + int bw, int bh, int ss_x, int ss_y); + +void vp9_build_inter_predictors_sby(MACROBLOCKD *xd, int mi_row, int mi_col, + BLOCK_SIZE bsize); + +void vp9_build_inter_predictors_sbp(MACROBLOCKD *xd, int mi_row, int mi_col, + BLOCK_SIZE bsize, int plane); + +void vp9_build_inter_predictors_sbuv(MACROBLOCKD *xd, int mi_row, int mi_col, + BLOCK_SIZE bsize); + +void vp9_build_inter_predictors_sb(MACROBLOCKD *xd, int mi_row, int mi_col, + BLOCK_SIZE bsize); + +void vp9_build_inter_predictor(const uint8_t *src, int src_stride, + uint8_t *dst, int dst_stride, + const MV *mv_q3, + const struct scale_factors *sf, + int w, int h, int do_avg, + const InterpKernel *kernel, + enum mv_precision precision, + int x, int y); + +#if CONFIG_VP9_HIGHBITDEPTH +void vp9_highbd_build_inter_predictor(const uint8_t *src, int src_stride, + uint8_t *dst, int dst_stride, + const MV *mv_q3, + const struct scale_factors *sf, + int w, int h, int do_avg, + const InterpKernel *kernel, + enum mv_precision precision, + int x, int y, int bd); +#endif + +static INLINE int scaled_buffer_offset(int x_offset, int y_offset, int stride, + const struct scale_factors *sf) { + const int x = sf ? sf->scale_value_x(x_offset, sf) : x_offset; + const int y = sf ? sf->scale_value_y(y_offset, sf) : y_offset; + return y * stride + x; +} + +static INLINE void setup_pred_plane(struct buf_2d *dst, + uint8_t *src, int stride, + int mi_row, int mi_col, + const struct scale_factors *scale, + int subsampling_x, int subsampling_y) { + const int x = (MI_SIZE * mi_col) >> subsampling_x; + const int y = (MI_SIZE * mi_row) >> subsampling_y; + dst->buf = src + scaled_buffer_offset(x, y, stride, scale); + dst->stride = stride; +} + +void vp9_setup_dst_planes(struct macroblockd_plane planes[MAX_MB_PLANE], + const YV12_BUFFER_CONFIG *src, + int mi_row, int mi_col); + +void vp9_setup_pre_planes(MACROBLOCKD *xd, int idx, + const YV12_BUFFER_CONFIG *src, int mi_row, int mi_col, + const struct scale_factors *sf); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_RECONINTER_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_reconintra.c b/thirdparty/libvpx/vp9/common/vp9_reconintra.c new file mode 100644 index 0000000000..445785835a --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_reconintra.c @@ -0,0 +1,445 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "./vpx_config.h" +#include "./vpx_dsp_rtcd.h" + +#if CONFIG_VP9_HIGHBITDEPTH +#include "vpx_dsp/vpx_dsp_common.h" +#endif // CONFIG_VP9_HIGHBITDEPTH +#include "vpx_mem/vpx_mem.h" +#include "vpx_ports/mem.h" +#include "vpx_ports/vpx_once.h" + +#include "vp9/common/vp9_reconintra.h" +#include "vp9/common/vp9_onyxc_int.h" + +const TX_TYPE intra_mode_to_tx_type_lookup[INTRA_MODES] = { + DCT_DCT, // DC + ADST_DCT, // V + DCT_ADST, // H + DCT_DCT, // D45 + ADST_ADST, // D135 + ADST_DCT, // D117 + DCT_ADST, // D153 + DCT_ADST, // D207 + ADST_DCT, // D63 + ADST_ADST, // TM +}; + +enum { + NEED_LEFT = 1 << 1, + NEED_ABOVE = 1 << 2, + NEED_ABOVERIGHT = 1 << 3, +}; + +static const uint8_t extend_modes[INTRA_MODES] = { + NEED_ABOVE | NEED_LEFT, // DC + NEED_ABOVE, // V + NEED_LEFT, // H + NEED_ABOVERIGHT, // D45 + NEED_LEFT | NEED_ABOVE, // D135 + NEED_LEFT | NEED_ABOVE, // D117 + NEED_LEFT | NEED_ABOVE, // D153 + NEED_LEFT, // D207 + NEED_ABOVERIGHT, // D63 + NEED_LEFT | NEED_ABOVE, // TM +}; + +typedef void (*intra_pred_fn)(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left); + +static intra_pred_fn pred[INTRA_MODES][TX_SIZES]; +static intra_pred_fn dc_pred[2][2][TX_SIZES]; + +#if CONFIG_VP9_HIGHBITDEPTH +typedef void (*intra_high_pred_fn)(uint16_t *dst, ptrdiff_t stride, + const uint16_t *above, const uint16_t *left, + int bd); +static intra_high_pred_fn pred_high[INTRA_MODES][4]; +static intra_high_pred_fn dc_pred_high[2][2][4]; +#endif // CONFIG_VP9_HIGHBITDEPTH + +static void vp9_init_intra_predictors_internal(void) { +#define INIT_ALL_SIZES(p, type) \ + p[TX_4X4] = vpx_##type##_predictor_4x4; \ + p[TX_8X8] = vpx_##type##_predictor_8x8; \ + p[TX_16X16] = vpx_##type##_predictor_16x16; \ + p[TX_32X32] = vpx_##type##_predictor_32x32 + + INIT_ALL_SIZES(pred[V_PRED], v); + INIT_ALL_SIZES(pred[H_PRED], h); + INIT_ALL_SIZES(pred[D207_PRED], d207); + INIT_ALL_SIZES(pred[D45_PRED], d45); + INIT_ALL_SIZES(pred[D63_PRED], d63); + INIT_ALL_SIZES(pred[D117_PRED], d117); + INIT_ALL_SIZES(pred[D135_PRED], d135); + INIT_ALL_SIZES(pred[D153_PRED], d153); + INIT_ALL_SIZES(pred[TM_PRED], tm); + + INIT_ALL_SIZES(dc_pred[0][0], dc_128); + INIT_ALL_SIZES(dc_pred[0][1], dc_top); + INIT_ALL_SIZES(dc_pred[1][0], dc_left); + INIT_ALL_SIZES(dc_pred[1][1], dc); + +#if CONFIG_VP9_HIGHBITDEPTH + INIT_ALL_SIZES(pred_high[V_PRED], highbd_v); + INIT_ALL_SIZES(pred_high[H_PRED], highbd_h); + INIT_ALL_SIZES(pred_high[D207_PRED], highbd_d207); + INIT_ALL_SIZES(pred_high[D45_PRED], highbd_d45); + INIT_ALL_SIZES(pred_high[D63_PRED], highbd_d63); + INIT_ALL_SIZES(pred_high[D117_PRED], highbd_d117); + INIT_ALL_SIZES(pred_high[D135_PRED], highbd_d135); + INIT_ALL_SIZES(pred_high[D153_PRED], highbd_d153); + INIT_ALL_SIZES(pred_high[TM_PRED], highbd_tm); + + INIT_ALL_SIZES(dc_pred_high[0][0], highbd_dc_128); + INIT_ALL_SIZES(dc_pred_high[0][1], highbd_dc_top); + INIT_ALL_SIZES(dc_pred_high[1][0], highbd_dc_left); + INIT_ALL_SIZES(dc_pred_high[1][1], highbd_dc); +#endif // CONFIG_VP9_HIGHBITDEPTH + +#undef intra_pred_allsizes +} + +#if CONFIG_VP9_HIGHBITDEPTH +static void build_intra_predictors_high(const MACROBLOCKD *xd, + const uint8_t *ref8, + int ref_stride, + uint8_t *dst8, + int dst_stride, + PREDICTION_MODE mode, + TX_SIZE tx_size, + int up_available, + int left_available, + int right_available, + int x, int y, + int plane, int bd) { + int i; + uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); + uint16_t *ref = CONVERT_TO_SHORTPTR(ref8); + DECLARE_ALIGNED(16, uint16_t, left_col[32]); + DECLARE_ALIGNED(16, uint16_t, above_data[64 + 16]); + uint16_t *above_row = above_data + 16; + const uint16_t *const_above_row = above_row; + const int bs = 4 << tx_size; + int frame_width, frame_height; + int x0, y0; + const struct macroblockd_plane *const pd = &xd->plane[plane]; + const int need_left = extend_modes[mode] & NEED_LEFT; + const int need_above = extend_modes[mode] & NEED_ABOVE; + const int need_aboveright = extend_modes[mode] & NEED_ABOVERIGHT; + int base = 128 << (bd - 8); + // 127 127 127 .. 127 127 127 127 127 127 + // 129 A B .. Y Z + // 129 C D .. W X + // 129 E F .. U V + // 129 G H .. S T T T T T + // For 10 bit and 12 bit, 127 and 129 are replaced by base -1 and base + 1. + + // Get current frame pointer, width and height. + if (plane == 0) { + frame_width = xd->cur_buf->y_width; + frame_height = xd->cur_buf->y_height; + } else { + frame_width = xd->cur_buf->uv_width; + frame_height = xd->cur_buf->uv_height; + } + + // Get block position in current frame. + x0 = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x)) + x; + y0 = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y)) + y; + + // NEED_LEFT + if (need_left) { + if (left_available) { + if (xd->mb_to_bottom_edge < 0) { + /* slower path if the block needs border extension */ + if (y0 + bs <= frame_height) { + for (i = 0; i < bs; ++i) + left_col[i] = ref[i * ref_stride - 1]; + } else { + const int extend_bottom = frame_height - y0; + for (i = 0; i < extend_bottom; ++i) + left_col[i] = ref[i * ref_stride - 1]; + for (; i < bs; ++i) + left_col[i] = ref[(extend_bottom - 1) * ref_stride - 1]; + } + } else { + /* faster path if the block does not need extension */ + for (i = 0; i < bs; ++i) + left_col[i] = ref[i * ref_stride - 1]; + } + } else { + vpx_memset16(left_col, base + 1, bs); + } + } + + // NEED_ABOVE + if (need_above) { + if (up_available) { + const uint16_t *above_ref = ref - ref_stride; + if (xd->mb_to_right_edge < 0) { + /* slower path if the block needs border extension */ + if (x0 + bs <= frame_width) { + memcpy(above_row, above_ref, bs * sizeof(above_row[0])); + } else if (x0 <= frame_width) { + const int r = frame_width - x0; + memcpy(above_row, above_ref, r * sizeof(above_row[0])); + vpx_memset16(above_row + r, above_row[r - 1], x0 + bs - frame_width); + } + } else { + /* faster path if the block does not need extension */ + if (bs == 4 && right_available && left_available) { + const_above_row = above_ref; + } else { + memcpy(above_row, above_ref, bs * sizeof(above_row[0])); + } + } + above_row[-1] = left_available ? above_ref[-1] : (base + 1); + } else { + vpx_memset16(above_row, base - 1, bs); + above_row[-1] = base - 1; + } + } + + // NEED_ABOVERIGHT + if (need_aboveright) { + if (up_available) { + const uint16_t *above_ref = ref - ref_stride; + if (xd->mb_to_right_edge < 0) { + /* slower path if the block needs border extension */ + if (x0 + 2 * bs <= frame_width) { + if (right_available && bs == 4) { + memcpy(above_row, above_ref, 2 * bs * sizeof(above_row[0])); + } else { + memcpy(above_row, above_ref, bs * sizeof(above_row[0])); + vpx_memset16(above_row + bs, above_row[bs - 1], bs); + } + } else if (x0 + bs <= frame_width) { + const int r = frame_width - x0; + if (right_available && bs == 4) { + memcpy(above_row, above_ref, r * sizeof(above_row[0])); + vpx_memset16(above_row + r, above_row[r - 1], + x0 + 2 * bs - frame_width); + } else { + memcpy(above_row, above_ref, bs * sizeof(above_row[0])); + vpx_memset16(above_row + bs, above_row[bs - 1], bs); + } + } else if (x0 <= frame_width) { + const int r = frame_width - x0; + memcpy(above_row, above_ref, r * sizeof(above_row[0])); + vpx_memset16(above_row + r, above_row[r - 1], + x0 + 2 * bs - frame_width); + } + above_row[-1] = left_available ? above_ref[-1] : (base + 1); + } else { + /* faster path if the block does not need extension */ + if (bs == 4 && right_available && left_available) { + const_above_row = above_ref; + } else { + memcpy(above_row, above_ref, bs * sizeof(above_row[0])); + if (bs == 4 && right_available) + memcpy(above_row + bs, above_ref + bs, bs * sizeof(above_row[0])); + else + vpx_memset16(above_row + bs, above_row[bs - 1], bs); + above_row[-1] = left_available ? above_ref[-1] : (base + 1); + } + } + } else { + vpx_memset16(above_row, base - 1, bs * 2); + above_row[-1] = base - 1; + } + } + + // predict + if (mode == DC_PRED) { + dc_pred_high[left_available][up_available][tx_size](dst, dst_stride, + const_above_row, + left_col, xd->bd); + } else { + pred_high[mode][tx_size](dst, dst_stride, const_above_row, left_col, + xd->bd); + } +} +#endif // CONFIG_VP9_HIGHBITDEPTH + +static void build_intra_predictors(const MACROBLOCKD *xd, const uint8_t *ref, + int ref_stride, uint8_t *dst, int dst_stride, + PREDICTION_MODE mode, TX_SIZE tx_size, + int up_available, int left_available, + int right_available, int x, int y, + int plane) { + int i; + DECLARE_ALIGNED(16, uint8_t, left_col[32]); + DECLARE_ALIGNED(16, uint8_t, above_data[64 + 16]); + uint8_t *above_row = above_data + 16; + const uint8_t *const_above_row = above_row; + const int bs = 4 << tx_size; + int frame_width, frame_height; + int x0, y0; + const struct macroblockd_plane *const pd = &xd->plane[plane]; + + // 127 127 127 .. 127 127 127 127 127 127 + // 129 A B .. Y Z + // 129 C D .. W X + // 129 E F .. U V + // 129 G H .. S T T T T T + // .. + + // Get current frame pointer, width and height. + if (plane == 0) { + frame_width = xd->cur_buf->y_width; + frame_height = xd->cur_buf->y_height; + } else { + frame_width = xd->cur_buf->uv_width; + frame_height = xd->cur_buf->uv_height; + } + + // Get block position in current frame. + x0 = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x)) + x; + y0 = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y)) + y; + + // NEED_LEFT + if (extend_modes[mode] & NEED_LEFT) { + if (left_available) { + if (xd->mb_to_bottom_edge < 0) { + /* slower path if the block needs border extension */ + if (y0 + bs <= frame_height) { + for (i = 0; i < bs; ++i) + left_col[i] = ref[i * ref_stride - 1]; + } else { + const int extend_bottom = frame_height - y0; + for (i = 0; i < extend_bottom; ++i) + left_col[i] = ref[i * ref_stride - 1]; + for (; i < bs; ++i) + left_col[i] = ref[(extend_bottom - 1) * ref_stride - 1]; + } + } else { + /* faster path if the block does not need extension */ + for (i = 0; i < bs; ++i) + left_col[i] = ref[i * ref_stride - 1]; + } + } else { + memset(left_col, 129, bs); + } + } + + // NEED_ABOVE + if (extend_modes[mode] & NEED_ABOVE) { + if (up_available) { + const uint8_t *above_ref = ref - ref_stride; + if (xd->mb_to_right_edge < 0) { + /* slower path if the block needs border extension */ + if (x0 + bs <= frame_width) { + memcpy(above_row, above_ref, bs); + } else if (x0 <= frame_width) { + const int r = frame_width - x0; + memcpy(above_row, above_ref, r); + memset(above_row + r, above_row[r - 1], x0 + bs - frame_width); + } + } else { + /* faster path if the block does not need extension */ + if (bs == 4 && right_available && left_available) { + const_above_row = above_ref; + } else { + memcpy(above_row, above_ref, bs); + } + } + above_row[-1] = left_available ? above_ref[-1] : 129; + } else { + memset(above_row, 127, bs); + above_row[-1] = 127; + } + } + + // NEED_ABOVERIGHT + if (extend_modes[mode] & NEED_ABOVERIGHT) { + if (up_available) { + const uint8_t *above_ref = ref - ref_stride; + if (xd->mb_to_right_edge < 0) { + /* slower path if the block needs border extension */ + if (x0 + 2 * bs <= frame_width) { + if (right_available && bs == 4) { + memcpy(above_row, above_ref, 2 * bs); + } else { + memcpy(above_row, above_ref, bs); + memset(above_row + bs, above_row[bs - 1], bs); + } + } else if (x0 + bs <= frame_width) { + const int r = frame_width - x0; + if (right_available && bs == 4) { + memcpy(above_row, above_ref, r); + memset(above_row + r, above_row[r - 1], x0 + 2 * bs - frame_width); + } else { + memcpy(above_row, above_ref, bs); + memset(above_row + bs, above_row[bs - 1], bs); + } + } else if (x0 <= frame_width) { + const int r = frame_width - x0; + memcpy(above_row, above_ref, r); + memset(above_row + r, above_row[r - 1], x0 + 2 * bs - frame_width); + } + } else { + /* faster path if the block does not need extension */ + if (bs == 4 && right_available && left_available) { + const_above_row = above_ref; + } else { + memcpy(above_row, above_ref, bs); + if (bs == 4 && right_available) + memcpy(above_row + bs, above_ref + bs, bs); + else + memset(above_row + bs, above_row[bs - 1], bs); + } + } + above_row[-1] = left_available ? above_ref[-1] : 129; + } else { + memset(above_row, 127, bs * 2); + above_row[-1] = 127; + } + } + + // predict + if (mode == DC_PRED) { + dc_pred[left_available][up_available][tx_size](dst, dst_stride, + const_above_row, left_col); + } else { + pred[mode][tx_size](dst, dst_stride, const_above_row, left_col); + } +} + +void vp9_predict_intra_block(const MACROBLOCKD *xd, int bwl_in, + TX_SIZE tx_size, PREDICTION_MODE mode, + const uint8_t *ref, int ref_stride, + uint8_t *dst, int dst_stride, + int aoff, int loff, int plane) { + const int bw = (1 << bwl_in); + const int txw = (1 << tx_size); + const int have_top = loff || (xd->above_mi != NULL); + const int have_left = aoff || (xd->left_mi != NULL); + const int have_right = (aoff + txw) < bw; + const int x = aoff * 4; + const int y = loff * 4; + +#if CONFIG_VP9_HIGHBITDEPTH + if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) { + build_intra_predictors_high(xd, ref, ref_stride, dst, dst_stride, mode, + tx_size, have_top, have_left, have_right, + x, y, plane, xd->bd); + return; + } +#endif + build_intra_predictors(xd, ref, ref_stride, dst, dst_stride, mode, tx_size, + have_top, have_left, have_right, x, y, plane); +} + +void vp9_init_intra_predictors(void) { + once(vp9_init_intra_predictors_internal); +} diff --git a/thirdparty/libvpx/vp9/common/vp9_reconintra.h b/thirdparty/libvpx/vp9/common/vp9_reconintra.h new file mode 100644 index 0000000000..de453808b7 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_reconintra.h @@ -0,0 +1,32 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_COMMON_VP9_RECONINTRA_H_ +#define VP9_COMMON_VP9_RECONINTRA_H_ + +#include "vpx/vpx_integer.h" +#include "vp9/common/vp9_blockd.h" + +#ifdef __cplusplus +extern "C" { +#endif + +void vp9_init_intra_predictors(void); + +void vp9_predict_intra_block(const MACROBLOCKD *xd, int bwl_in, + TX_SIZE tx_size, PREDICTION_MODE mode, + const uint8_t *ref, int ref_stride, + uint8_t *dst, int dst_stride, + int aoff, int loff, int plane); +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_RECONINTRA_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_rtcd.c b/thirdparty/libvpx/vp9/common/vp9_rtcd.c new file mode 100644 index 0000000000..2dfa09f50e --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_rtcd.c @@ -0,0 +1,19 @@ +/* + * Copyright (c) 2011 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ +#include "./vpx_config.h" +#define RTCD_C +#include "./vp9_rtcd.h" +#include "vpx_ports/vpx_once.h" + +void vp9_rtcd() { + // TODO(JBB): Remove this once, by insuring that both the encoder and + // decoder setup functions are protected by once(); + once(setup_rtcd_internal); +} diff --git a/thirdparty/libvpx/vp9/common/vp9_scale.c b/thirdparty/libvpx/vp9/common/vp9_scale.c new file mode 100644 index 0000000000..b763b925b3 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_scale.c @@ -0,0 +1,175 @@ +/* + * Copyright (c) 2013 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "./vpx_dsp_rtcd.h" +#include "vp9/common/vp9_filter.h" +#include "vp9/common/vp9_scale.h" +#include "vpx_dsp/vpx_filter.h" + +static INLINE int scaled_x(int val, const struct scale_factors *sf) { + return (int)((int64_t)val * sf->x_scale_fp >> REF_SCALE_SHIFT); +} + +static INLINE int scaled_y(int val, const struct scale_factors *sf) { + return (int)((int64_t)val * sf->y_scale_fp >> REF_SCALE_SHIFT); +} + +static int unscaled_value(int val, const struct scale_factors *sf) { + (void) sf; + return val; +} + +static int get_fixed_point_scale_factor(int other_size, int this_size) { + // Calculate scaling factor once for each reference frame + // and use fixed point scaling factors in decoding and encoding routines. + // Hardware implementations can calculate scale factor in device driver + // and use multiplication and shifting on hardware instead of division. + return (other_size << REF_SCALE_SHIFT) / this_size; +} + +MV32 vp9_scale_mv(const MV *mv, int x, int y, const struct scale_factors *sf) { + const int x_off_q4 = scaled_x(x << SUBPEL_BITS, sf) & SUBPEL_MASK; + const int y_off_q4 = scaled_y(y << SUBPEL_BITS, sf) & SUBPEL_MASK; + const MV32 res = { + scaled_y(mv->row, sf) + y_off_q4, + scaled_x(mv->col, sf) + x_off_q4 + }; + return res; +} + +#if CONFIG_VP9_HIGHBITDEPTH +void vp9_setup_scale_factors_for_frame(struct scale_factors *sf, + int other_w, int other_h, + int this_w, int this_h, + int use_highbd) { +#else +void vp9_setup_scale_factors_for_frame(struct scale_factors *sf, + int other_w, int other_h, + int this_w, int this_h) { +#endif + if (!valid_ref_frame_size(other_w, other_h, this_w, this_h)) { + sf->x_scale_fp = REF_INVALID_SCALE; + sf->y_scale_fp = REF_INVALID_SCALE; + return; + } + + sf->x_scale_fp = get_fixed_point_scale_factor(other_w, this_w); + sf->y_scale_fp = get_fixed_point_scale_factor(other_h, this_h); + sf->x_step_q4 = scaled_x(16, sf); + sf->y_step_q4 = scaled_y(16, sf); + + if (vp9_is_scaled(sf)) { + sf->scale_value_x = scaled_x; + sf->scale_value_y = scaled_y; + } else { + sf->scale_value_x = unscaled_value; + sf->scale_value_y = unscaled_value; + } + + // TODO(agrange): Investigate the best choice of functions to use here + // for EIGHTTAP_SMOOTH. Since it is not interpolating, need to choose what + // to do at full-pel offsets. The current selection, where the filter is + // applied in one direction only, and not at all for 0,0, seems to give the + // best quality, but it may be worth trying an additional mode that does + // do the filtering on full-pel. + + if (sf->x_step_q4 == 16) { + if (sf->y_step_q4 == 16) { + // No scaling in either direction. + sf->predict[0][0][0] = vpx_convolve_copy; + sf->predict[0][0][1] = vpx_convolve_avg; + sf->predict[0][1][0] = vpx_convolve8_vert; + sf->predict[0][1][1] = vpx_convolve8_avg_vert; + sf->predict[1][0][0] = vpx_convolve8_horiz; + sf->predict[1][0][1] = vpx_convolve8_avg_horiz; + } else { + // No scaling in x direction. Must always scale in the y direction. + sf->predict[0][0][0] = vpx_scaled_vert; + sf->predict[0][0][1] = vpx_scaled_avg_vert; + sf->predict[0][1][0] = vpx_scaled_vert; + sf->predict[0][1][1] = vpx_scaled_avg_vert; + sf->predict[1][0][0] = vpx_scaled_2d; + sf->predict[1][0][1] = vpx_scaled_avg_2d; + } + } else { + if (sf->y_step_q4 == 16) { + // No scaling in the y direction. Must always scale in the x direction. + sf->predict[0][0][0] = vpx_scaled_horiz; + sf->predict[0][0][1] = vpx_scaled_avg_horiz; + sf->predict[0][1][0] = vpx_scaled_2d; + sf->predict[0][1][1] = vpx_scaled_avg_2d; + sf->predict[1][0][0] = vpx_scaled_horiz; + sf->predict[1][0][1] = vpx_scaled_avg_horiz; + } else { + // Must always scale in both directions. + sf->predict[0][0][0] = vpx_scaled_2d; + sf->predict[0][0][1] = vpx_scaled_avg_2d; + sf->predict[0][1][0] = vpx_scaled_2d; + sf->predict[0][1][1] = vpx_scaled_avg_2d; + sf->predict[1][0][0] = vpx_scaled_2d; + sf->predict[1][0][1] = vpx_scaled_avg_2d; + } + } + + // 2D subpel motion always gets filtered in both directions + + if ((sf->x_step_q4 != 16) || (sf->y_step_q4 != 16)) { + sf->predict[1][1][0] = vpx_scaled_2d; + sf->predict[1][1][1] = vpx_scaled_avg_2d; + } else { + sf->predict[1][1][0] = vpx_convolve8; + sf->predict[1][1][1] = vpx_convolve8_avg; + } + +#if CONFIG_VP9_HIGHBITDEPTH + if (use_highbd) { + if (sf->x_step_q4 == 16) { + if (sf->y_step_q4 == 16) { + // No scaling in either direction. + sf->highbd_predict[0][0][0] = vpx_highbd_convolve_copy; + sf->highbd_predict[0][0][1] = vpx_highbd_convolve_avg; + sf->highbd_predict[0][1][0] = vpx_highbd_convolve8_vert; + sf->highbd_predict[0][1][1] = vpx_highbd_convolve8_avg_vert; + sf->highbd_predict[1][0][0] = vpx_highbd_convolve8_horiz; + sf->highbd_predict[1][0][1] = vpx_highbd_convolve8_avg_horiz; + } else { + // No scaling in x direction. Must always scale in the y direction. + sf->highbd_predict[0][0][0] = vpx_highbd_convolve8_vert; + sf->highbd_predict[0][0][1] = vpx_highbd_convolve8_avg_vert; + sf->highbd_predict[0][1][0] = vpx_highbd_convolve8_vert; + sf->highbd_predict[0][1][1] = vpx_highbd_convolve8_avg_vert; + sf->highbd_predict[1][0][0] = vpx_highbd_convolve8; + sf->highbd_predict[1][0][1] = vpx_highbd_convolve8_avg; + } + } else { + if (sf->y_step_q4 == 16) { + // No scaling in the y direction. Must always scale in the x direction. + sf->highbd_predict[0][0][0] = vpx_highbd_convolve8_horiz; + sf->highbd_predict[0][0][1] = vpx_highbd_convolve8_avg_horiz; + sf->highbd_predict[0][1][0] = vpx_highbd_convolve8; + sf->highbd_predict[0][1][1] = vpx_highbd_convolve8_avg; + sf->highbd_predict[1][0][0] = vpx_highbd_convolve8_horiz; + sf->highbd_predict[1][0][1] = vpx_highbd_convolve8_avg_horiz; + } else { + // Must always scale in both directions. + sf->highbd_predict[0][0][0] = vpx_highbd_convolve8; + sf->highbd_predict[0][0][1] = vpx_highbd_convolve8_avg; + sf->highbd_predict[0][1][0] = vpx_highbd_convolve8; + sf->highbd_predict[0][1][1] = vpx_highbd_convolve8_avg; + sf->highbd_predict[1][0][0] = vpx_highbd_convolve8; + sf->highbd_predict[1][0][1] = vpx_highbd_convolve8_avg; + } + } + // 2D subpel motion always gets filtered in both directions. + sf->highbd_predict[1][1][0] = vpx_highbd_convolve8; + sf->highbd_predict[1][1][1] = vpx_highbd_convolve8_avg; + } +#endif +} diff --git a/thirdparty/libvpx/vp9/common/vp9_scale.h b/thirdparty/libvpx/vp9/common/vp9_scale.h new file mode 100644 index 0000000000..5e91041079 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_scale.h @@ -0,0 +1,75 @@ +/* + * Copyright (c) 2013 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_COMMON_VP9_SCALE_H_ +#define VP9_COMMON_VP9_SCALE_H_ + +#include "vp9/common/vp9_mv.h" +#include "vpx_dsp/vpx_convolve.h" + +#ifdef __cplusplus +extern "C" { +#endif + +#define REF_SCALE_SHIFT 14 +#define REF_NO_SCALE (1 << REF_SCALE_SHIFT) +#define REF_INVALID_SCALE -1 + +struct scale_factors { + int x_scale_fp; // horizontal fixed point scale factor + int y_scale_fp; // vertical fixed point scale factor + int x_step_q4; + int y_step_q4; + + int (*scale_value_x)(int val, const struct scale_factors *sf); + int (*scale_value_y)(int val, const struct scale_factors *sf); + + convolve_fn_t predict[2][2][2]; // horiz, vert, avg +#if CONFIG_VP9_HIGHBITDEPTH + highbd_convolve_fn_t highbd_predict[2][2][2]; // horiz, vert, avg +#endif +}; + +MV32 vp9_scale_mv(const MV *mv, int x, int y, const struct scale_factors *sf); + +#if CONFIG_VP9_HIGHBITDEPTH +void vp9_setup_scale_factors_for_frame(struct scale_factors *sf, + int other_w, int other_h, + int this_w, int this_h, + int use_high); +#else +void vp9_setup_scale_factors_for_frame(struct scale_factors *sf, + int other_w, int other_h, + int this_w, int this_h); +#endif + +static INLINE int vp9_is_valid_scale(const struct scale_factors *sf) { + return sf->x_scale_fp != REF_INVALID_SCALE && + sf->y_scale_fp != REF_INVALID_SCALE; +} + +static INLINE int vp9_is_scaled(const struct scale_factors *sf) { + return vp9_is_valid_scale(sf) && + (sf->x_scale_fp != REF_NO_SCALE || sf->y_scale_fp != REF_NO_SCALE); +} + +static INLINE int valid_ref_frame_size(int ref_width, int ref_height, + int this_width, int this_height) { + return 2 * this_width >= ref_width && + 2 * this_height >= ref_height && + this_width <= 16 * ref_width && + this_height <= 16 * ref_height; +} + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_SCALE_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_scan.c b/thirdparty/libvpx/vp9/common/vp9_scan.c new file mode 100644 index 0000000000..8b8b09f4a3 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_scan.c @@ -0,0 +1,725 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <assert.h> + +#include "vp9/common/vp9_scan.h" + +DECLARE_ALIGNED(16, static const int16_t, default_scan_4x4[16]) = { + 0, 4, 1, 5, + 8, 2, 12, 9, + 3, 6, 13, 10, + 7, 14, 11, 15, +}; + +DECLARE_ALIGNED(16, static const int16_t, col_scan_4x4[16]) = { + 0, 4, 8, 1, + 12, 5, 9, 2, + 13, 6, 10, 3, + 7, 14, 11, 15, +}; + +DECLARE_ALIGNED(16, static const int16_t, row_scan_4x4[16]) = { + 0, 1, 4, 2, + 5, 3, 6, 8, + 9, 7, 12, 10, + 13, 11, 14, 15, +}; + +DECLARE_ALIGNED(16, static const int16_t, default_scan_8x8[64]) = { + 0, 8, 1, 16, 9, 2, 17, 24, + 10, 3, 18, 25, 32, 11, 4, 26, + 33, 19, 40, 12, 34, 27, 5, 41, + 20, 48, 13, 35, 42, 28, 21, 6, + 49, 56, 36, 43, 29, 7, 14, 50, + 57, 44, 22, 37, 15, 51, 58, 30, + 45, 23, 52, 59, 38, 31, 60, 53, + 46, 39, 61, 54, 47, 62, 55, 63, +}; + +DECLARE_ALIGNED(16, static const int16_t, col_scan_8x8[64]) = { + 0, 8, 16, 1, 24, 9, 32, 17, + 2, 40, 25, 10, 33, 18, 48, 3, + 26, 41, 11, 56, 19, 34, 4, 49, + 27, 42, 12, 35, 20, 57, 50, 28, + 5, 43, 13, 36, 58, 51, 21, 44, + 6, 29, 59, 37, 14, 52, 22, 7, + 45, 60, 30, 15, 38, 53, 23, 46, + 31, 61, 39, 54, 47, 62, 55, 63, +}; + +DECLARE_ALIGNED(16, static const int16_t, row_scan_8x8[64]) = { + 0, 1, 2, 8, 9, 3, 16, 10, + 4, 17, 11, 24, 5, 18, 25, 12, + 19, 26, 32, 6, 13, 20, 33, 27, + 7, 34, 40, 21, 28, 41, 14, 35, + 48, 42, 29, 36, 49, 22, 43, 15, + 56, 37, 50, 44, 30, 57, 23, 51, + 58, 45, 38, 52, 31, 59, 53, 46, + 60, 39, 61, 47, 54, 55, 62, 63, +}; + +DECLARE_ALIGNED(16, static const int16_t, default_scan_16x16[256]) = { + 0, 16, 1, 32, 17, 2, 48, 33, 18, 3, 64, 34, 49, 19, 65, 80, + 50, 4, 35, 66, 20, 81, 96, 51, 5, 36, 82, 97, 67, 112, 21, 52, + 98, 37, 83, 113, 6, 68, 128, 53, 22, 99, 114, 84, 7, 129, 38, 69, + 100, 115, 144, 130, 85, 54, 23, 8, 145, 39, 70, 116, 101, 131, 160, 146, + 55, 86, 24, 71, 132, 117, 161, 40, 9, 102, 147, 176, 162, 87, 56, 25, + 133, 118, 177, 148, 72, 103, 41, 163, 10, 192, 178, 88, 57, 134, 149, 119, + 26, 164, 73, 104, 193, 42, 179, 208, 11, 135, 89, 165, 120, 150, 58, 194, + 180, 27, 74, 209, 105, 151, 136, 43, 90, 224, 166, 195, 181, 121, 210, 59, + 12, 152, 106, 167, 196, 75, 137, 225, 211, 240, 182, 122, 91, 28, 197, 13, + 226, 168, 183, 153, 44, 212, 138, 107, 241, 60, 29, 123, 198, 184, 227, 169, + 242, 76, 213, 154, 45, 92, 14, 199, 139, 61, 228, 214, 170, 185, 243, 108, + 77, 155, 30, 15, 200, 229, 124, 215, 244, 93, 46, 186, 171, 201, 109, 140, + 230, 62, 216, 245, 31, 125, 78, 156, 231, 47, 187, 202, 217, 94, 246, 141, + 63, 232, 172, 110, 247, 157, 79, 218, 203, 126, 233, 188, 248, 95, 173, 142, + 219, 111, 249, 234, 158, 127, 189, 204, 250, 235, 143, 174, 220, 205, 159, + 251, + 190, 221, 175, 236, 237, 191, 206, 252, 222, 253, 207, 238, 223, 254, 239, + 255, +}; + +DECLARE_ALIGNED(16, static const int16_t, col_scan_16x16[256]) = { + 0, 16, 32, 48, 1, 64, 17, 80, 33, 96, 49, 2, 65, 112, 18, 81, + 34, 128, 50, 97, 3, 66, 144, 19, 113, 35, 82, 160, 98, 51, 129, 4, + 67, 176, 20, 114, 145, 83, 36, 99, 130, 52, 192, 5, 161, 68, 115, 21, + 146, 84, 208, 177, 37, 131, 100, 53, 162, 224, 69, 6, 116, 193, 147, 85, + 22, 240, 132, 38, 178, 101, 163, 54, 209, 117, 70, 7, 148, 194, 86, 179, + 225, 23, 133, 39, 164, 8, 102, 210, 241, 55, 195, 118, 149, 71, 180, 24, + 87, 226, 134, 165, 211, 40, 103, 56, 72, 150, 196, 242, 119, 9, 181, 227, + 88, 166, 25, 135, 41, 104, 212, 57, 151, 197, 120, 73, 243, 182, 136, 167, + 213, 89, 10, 228, 105, 152, 198, 26, 42, 121, 183, 244, 168, 58, 137, 229, + 74, 214, 90, 153, 199, 184, 11, 106, 245, 27, 122, 230, 169, 43, 215, 59, + 200, 138, 185, 246, 75, 12, 91, 154, 216, 231, 107, 28, 44, 201, 123, 170, + 60, 247, 232, 76, 139, 13, 92, 217, 186, 248, 155, 108, 29, 124, 45, 202, + 233, 171, 61, 14, 77, 140, 15, 249, 93, 30, 187, 156, 218, 46, 109, 125, + 62, 172, 78, 203, 31, 141, 234, 94, 47, 188, 63, 157, 110, 250, 219, 79, + 126, 204, 173, 142, 95, 189, 111, 235, 158, 220, 251, 127, 174, 143, 205, + 236, + 159, 190, 221, 252, 175, 206, 237, 191, 253, 222, 238, 207, 254, 223, 239, + 255, +}; + +DECLARE_ALIGNED(16, static const int16_t, row_scan_16x16[256]) = { + 0, 1, 2, 16, 3, 17, 4, 18, 32, 5, 33, 19, 6, 34, 48, 20, + 49, 7, 35, 21, 50, 64, 8, 36, 65, 22, 51, 37, 80, 9, 66, 52, + 23, 38, 81, 67, 10, 53, 24, 82, 68, 96, 39, 11, 54, 83, 97, 69, + 25, 98, 84, 40, 112, 55, 12, 70, 99, 113, 85, 26, 41, 56, 114, 100, + 13, 71, 128, 86, 27, 115, 101, 129, 42, 57, 72, 116, 14, 87, 130, 102, + 144, 73, 131, 117, 28, 58, 15, 88, 43, 145, 103, 132, 146, 118, 74, 160, + 89, 133, 104, 29, 59, 147, 119, 44, 161, 148, 90, 105, 134, 162, 120, 176, + 75, 135, 149, 30, 60, 163, 177, 45, 121, 91, 106, 164, 178, 150, 192, 136, + 165, 179, 31, 151, 193, 76, 122, 61, 137, 194, 107, 152, 180, 208, 46, 166, + 167, 195, 92, 181, 138, 209, 123, 153, 224, 196, 77, 168, 210, 182, 240, 108, + 197, 62, 154, 225, 183, 169, 211, 47, 139, 93, 184, 226, 212, 241, 198, 170, + 124, 155, 199, 78, 213, 185, 109, 227, 200, 63, 228, 242, 140, 214, 171, 186, + 156, 229, 243, 125, 94, 201, 244, 215, 216, 230, 141, 187, 202, 79, 172, 110, + 157, 245, 217, 231, 95, 246, 232, 126, 203, 247, 233, 173, 218, 142, 111, + 158, + 188, 248, 127, 234, 219, 249, 189, 204, 143, 174, 159, 250, 235, 205, 220, + 175, + 190, 251, 221, 191, 206, 236, 207, 237, 252, 222, 253, 223, 238, 239, 254, + 255, +}; + +DECLARE_ALIGNED(16, static const int16_t, default_scan_32x32[1024]) = { + 0, 32, 1, 64, 33, 2, 96, 65, 34, 128, 3, 97, 66, 160, + 129, 35, 98, 4, 67, 130, 161, 192, 36, 99, 224, 5, 162, 193, + 68, 131, 37, 100, + 225, 194, 256, 163, 69, 132, 6, 226, 257, 288, 195, 101, 164, 38, + 258, 7, 227, 289, 133, 320, 70, 196, 165, 290, 259, 228, 39, 321, + 102, 352, 8, 197, + 71, 134, 322, 291, 260, 353, 384, 229, 166, 103, 40, 354, 323, 292, + 135, 385, 198, 261, 72, 9, 416, 167, 386, 355, 230, 324, 104, 293, + 41, 417, 199, 136, + 262, 387, 448, 325, 356, 10, 73, 418, 231, 168, 449, 294, 388, 105, + 419, 263, 42, 200, 357, 450, 137, 480, 74, 326, 232, 11, 389, 169, + 295, 420, 106, 451, + 481, 358, 264, 327, 201, 43, 138, 512, 482, 390, 296, 233, 170, 421, + 75, 452, 359, 12, 513, 265, 483, 328, 107, 202, 514, 544, 422, 391, + 453, 139, 44, 234, + 484, 297, 360, 171, 76, 515, 545, 266, 329, 454, 13, 423, 203, 108, + 546, 485, 576, 298, 235, 140, 361, 330, 172, 547, 45, 455, 267, 577, + 486, 77, 204, 362, + 608, 14, 299, 578, 109, 236, 487, 609, 331, 141, 579, 46, 15, 173, + 610, 363, 78, 205, 16, 110, 237, 611, 142, 47, 174, 79, 206, 17, + 111, 238, 48, 143, + 80, 175, 112, 207, 49, 18, 239, 81, 113, 19, 50, 82, 114, 51, + 83, 115, 640, 516, 392, 268, 144, 20, 672, 641, 548, 517, 424, + 393, 300, 269, 176, 145, + 52, 21, 704, 673, 642, 580, 549, 518, 456, 425, 394, 332, 301, + 270, 208, 177, 146, 84, 53, 22, 736, 705, 674, 643, 612, 581, + 550, 519, 488, 457, 426, 395, + 364, 333, 302, 271, 240, 209, 178, 147, 116, 85, 54, 23, 737, + 706, 675, 613, 582, 551, 489, 458, 427, 365, 334, 303, 241, + 210, 179, 117, 86, 55, 738, 707, + 614, 583, 490, 459, 366, 335, 242, 211, 118, 87, 739, 615, 491, + 367, 243, 119, 768, 644, 520, 396, 272, 148, 24, 800, 769, 676, + 645, 552, 521, 428, 397, 304, + 273, 180, 149, 56, 25, 832, 801, 770, 708, 677, 646, 584, 553, + 522, 460, 429, 398, 336, 305, 274, 212, 181, 150, 88, 57, 26, + 864, 833, 802, 771, 740, 709, + 678, 647, 616, 585, 554, 523, 492, 461, 430, 399, 368, 337, 306, + 275, 244, 213, 182, 151, 120, 89, 58, 27, 865, 834, 803, 741, + 710, 679, 617, 586, 555, 493, + 462, 431, 369, 338, 307, 245, 214, 183, 121, 90, 59, 866, 835, + 742, 711, 618, 587, 494, 463, 370, 339, 246, 215, 122, 91, 867, + 743, 619, 495, 371, 247, 123, + 896, 772, 648, 524, 400, 276, 152, 28, 928, 897, 804, 773, 680, + 649, 556, 525, 432, 401, 308, 277, 184, 153, 60, 29, 960, 929, + 898, 836, 805, 774, 712, 681, + 650, 588, 557, 526, 464, 433, 402, 340, 309, 278, 216, 185, 154, + 92, 61, 30, 992, 961, 930, 899, 868, 837, 806, 775, 744, 713, 682, + 651, 620, 589, 558, 527, + 496, 465, 434, 403, 372, 341, 310, 279, 248, 217, 186, 155, 124, + 93, 62, 31, 993, 962, 931, 869, 838, 807, 745, 714, 683, 621, 590, + 559, 497, 466, 435, 373, + 342, 311, 249, 218, 187, 125, 94, 63, 994, 963, 870, 839, 746, 715, + 622, 591, 498, 467, 374, 343, 250, 219, 126, 95, 995, 871, 747, 623, + 499, 375, 251, 127, + 900, 776, 652, 528, 404, 280, 156, 932, 901, 808, 777, 684, 653, 560, + 529, 436, 405, 312, 281, 188, 157, 964, 933, 902, 840, 809, 778, 716, + 685, 654, 592, 561, + 530, 468, 437, 406, 344, 313, 282, 220, 189, 158, 996, 965, 934, 903, + 872, 841, 810, 779, 748, 717, 686, 655, 624, 593, 562, 531, 500, 469, + 438, 407, 376, 345, + 314, 283, 252, 221, 190, 159, 997, 966, 935, 873, 842, 811, 749, 718, + 687, 625, 594, 563, 501, 470, 439, 377, 346, 315, 253, 222, 191, 998, + 967, 874, 843, 750, + 719, 626, 595, 502, 471, 378, 347, 254, 223, 999, 875, 751, 627, 503, + 379, 255, 904, 780, 656, 532, 408, 284, 936, 905, 812, 781, 688, 657, + 564, 533, 440, 409, + 316, 285, 968, 937, 906, 844, 813, 782, 720, 689, 658, 596, 565, 534, + 472, 441, 410, 348, 317, 286, 1000, 969, 938, 907, 876, 845, 814, 783, + 752, 721, 690, 659, + 628, 597, 566, 535, 504, 473, 442, 411, 380, 349, 318, 287, 1001, 970, + 939, 877, 846, 815, 753, 722, 691, 629, 598, 567, 505, 474, 443, 381, + 350, 319, 1002, 971, + 878, 847, 754, 723, 630, 599, 506, 475, 382, 351, 1003, 879, 755, 631, + 507, 383, 908, 784, 660, 536, 412, 940, 909, 816, 785, 692, 661, 568, + 537, 444, 413, 972, + 941, 910, 848, 817, 786, 724, 693, 662, 600, 569, 538, 476, 445, 414, + 1004, 973, 942, 911, 880, 849, 818, 787, 756, 725, 694, 663, 632, 601, + 570, 539, 508, 477, + 446, 415, 1005, 974, 943, 881, 850, 819, 757, 726, 695, 633, 602, 571, + 509, 478, 447, 1006, 975, 882, 851, 758, 727, 634, 603, 510, 479, + 1007, 883, 759, 635, 511, + 912, 788, 664, 540, 944, 913, 820, 789, 696, 665, 572, 541, 976, 945, + 914, 852, 821, 790, 728, 697, 666, 604, 573, 542, 1008, 977, 946, 915, + 884, 853, 822, 791, + 760, 729, 698, 667, 636, 605, 574, 543, 1009, 978, 947, 885, 854, 823, + 761, 730, 699, 637, 606, 575, 1010, 979, 886, 855, 762, 731, 638, 607, + 1011, 887, 763, 639, + 916, 792, 668, 948, 917, 824, 793, 700, 669, 980, 949, 918, 856, 825, + 794, 732, 701, 670, 1012, 981, 950, 919, 888, 857, 826, 795, 764, 733, + 702, 671, 1013, 982, + 951, 889, 858, 827, 765, 734, 703, 1014, 983, 890, 859, 766, 735, 1015, + 891, 767, 920, 796, 952, 921, 828, 797, 984, 953, 922, 860, 829, 798, + 1016, 985, 954, 923, + 892, 861, 830, 799, 1017, 986, 955, 893, 862, 831, 1018, 987, 894, 863, + 1019, 895, 924, 956, 925, 988, 957, 926, 1020, 989, 958, 927, 1021, + 990, 959, 1022, 991, 1023, +}; + +// Neighborhood 2-tuples for various scans and blocksizes, +// in {top, left} order for each position in corresponding scan order. +DECLARE_ALIGNED(16, static const int16_t, + default_scan_4x4_neighbors[17 * MAX_NEIGHBORS]) = { + 0, 0, 0, 0, 0, 0, 1, 4, 4, 4, 1, 1, 8, 8, 5, 8, 2, 2, 2, 5, 9, 12, 6, 9, + 3, 6, 10, 13, 7, 10, 11, 14, 0, 0, +}; + +DECLARE_ALIGNED(16, static const int16_t, + col_scan_4x4_neighbors[17 * MAX_NEIGHBORS]) = { + 0, 0, 0, 0, 4, 4, 0, 0, 8, 8, 1, 1, 5, 5, 1, 1, 9, 9, 2, 2, 6, 6, 2, 2, 3, + 3, 10, 10, 7, 7, 11, 11, 0, 0, +}; + +DECLARE_ALIGNED(16, static const int16_t, + row_scan_4x4_neighbors[17 * MAX_NEIGHBORS]) = { + 0, 0, 0, 0, 0, 0, 1, 1, 4, 4, 2, 2, 5, 5, 4, 4, 8, 8, 6, 6, 8, 8, 9, 9, 12, + 12, 10, 10, 13, 13, 14, 14, 0, 0, +}; + +DECLARE_ALIGNED(16, static const int16_t, + col_scan_8x8_neighbors[65 * MAX_NEIGHBORS]) = { + 0, 0, 0, 0, 8, 8, 0, 0, 16, 16, 1, 1, 24, 24, 9, 9, 1, 1, 32, 32, 17, 17, 2, + 2, 25, 25, 10, 10, 40, 40, 2, 2, 18, 18, 33, 33, 3, 3, 48, 48, 11, 11, 26, + 26, 3, 3, 41, 41, 19, 19, 34, 34, 4, 4, 27, 27, 12, 12, 49, 49, 42, 42, 20, + 20, 4, 4, 35, 35, 5, 5, 28, 28, 50, 50, 43, 43, 13, 13, 36, 36, 5, 5, 21, 21, + 51, 51, 29, 29, 6, 6, 44, 44, 14, 14, 6, 6, 37, 37, 52, 52, 22, 22, 7, 7, 30, + 30, 45, 45, 15, 15, 38, 38, 23, 23, 53, 53, 31, 31, 46, 46, 39, 39, 54, 54, + 47, 47, 55, 55, 0, 0, +}; + +DECLARE_ALIGNED(16, static const int16_t, + row_scan_8x8_neighbors[65 * MAX_NEIGHBORS]) = { + 0, 0, 0, 0, 1, 1, 0, 0, 8, 8, 2, 2, 8, 8, 9, 9, 3, 3, 16, 16, 10, 10, 16, 16, + 4, 4, 17, 17, 24, 24, 11, 11, 18, 18, 25, 25, 24, 24, 5, 5, 12, 12, 19, 19, + 32, 32, 26, 26, 6, 6, 33, 33, 32, 32, 20, 20, 27, 27, 40, 40, 13, 13, 34, 34, + 40, 40, 41, 41, 28, 28, 35, 35, 48, 48, 21, 21, 42, 42, 14, 14, 48, 48, 36, + 36, 49, 49, 43, 43, 29, 29, 56, 56, 22, 22, 50, 50, 57, 57, 44, 44, 37, 37, + 51, 51, 30, 30, 58, 58, 52, 52, 45, 45, 59, 59, 38, 38, 60, 60, 46, 46, 53, + 53, 54, 54, 61, 61, 62, 62, 0, 0, +}; + +DECLARE_ALIGNED(16, static const int16_t, + default_scan_8x8_neighbors[65 * MAX_NEIGHBORS]) = { + 0, 0, 0, 0, 0, 0, 8, 8, 1, 8, 1, 1, 9, 16, 16, 16, 2, 9, 2, 2, 10, 17, 17, + 24, 24, 24, 3, 10, 3, 3, 18, 25, 25, 32, 11, 18, 32, 32, 4, 11, 26, 33, 19, + 26, 4, 4, 33, 40, 12, 19, 40, 40, 5, 12, 27, 34, 34, 41, 20, 27, 13, 20, 5, + 5, 41, 48, 48, 48, 28, 35, 35, 42, 21, 28, 6, 6, 6, 13, 42, 49, 49, 56, 36, + 43, 14, 21, 29, 36, 7, 14, 43, 50, 50, 57, 22, 29, 37, 44, 15, 22, 44, 51, + 51, 58, 30, 37, 23, 30, 52, 59, 45, 52, 38, 45, 31, 38, 53, 60, 46, 53, 39, + 46, 54, 61, 47, 54, 55, 62, 0, 0, +}; + +DECLARE_ALIGNED(16, static const int16_t, + col_scan_16x16_neighbors[257 * MAX_NEIGHBORS]) = { + 0, 0, 0, 0, 16, 16, 32, 32, 0, 0, 48, 48, 1, 1, 64, 64, + 17, 17, 80, 80, 33, 33, 1, 1, 49, 49, 96, 96, 2, 2, 65, 65, + 18, 18, 112, 112, 34, 34, 81, 81, 2, 2, 50, 50, 128, 128, 3, 3, + 97, 97, 19, 19, 66, 66, 144, 144, 82, 82, 35, 35, 113, 113, 3, 3, + 51, 51, 160, 160, 4, 4, 98, 98, 129, 129, 67, 67, 20, 20, 83, 83, + 114, 114, 36, 36, 176, 176, 4, 4, 145, 145, 52, 52, 99, 99, 5, 5, + 130, 130, 68, 68, 192, 192, 161, 161, 21, 21, 115, 115, 84, 84, 37, 37, + 146, 146, 208, 208, 53, 53, 5, 5, 100, 100, 177, 177, 131, 131, 69, 69, + 6, 6, 224, 224, 116, 116, 22, 22, 162, 162, 85, 85, 147, 147, 38, 38, + 193, 193, 101, 101, 54, 54, 6, 6, 132, 132, 178, 178, 70, 70, 163, 163, + 209, 209, 7, 7, 117, 117, 23, 23, 148, 148, 7, 7, 86, 86, 194, 194, + 225, 225, 39, 39, 179, 179, 102, 102, 133, 133, 55, 55, 164, 164, 8, 8, + 71, 71, 210, 210, 118, 118, 149, 149, 195, 195, 24, 24, 87, 87, 40, 40, + 56, 56, 134, 134, 180, 180, 226, 226, 103, 103, 8, 8, 165, 165, 211, 211, + 72, 72, 150, 150, 9, 9, 119, 119, 25, 25, 88, 88, 196, 196, 41, 41, + 135, 135, 181, 181, 104, 104, 57, 57, 227, 227, 166, 166, 120, 120, 151, 151, + 197, 197, 73, 73, 9, 9, 212, 212, 89, 89, 136, 136, 182, 182, 10, 10, + 26, 26, 105, 105, 167, 167, 228, 228, 152, 152, 42, 42, 121, 121, 213, 213, + 58, 58, 198, 198, 74, 74, 137, 137, 183, 183, 168, 168, 10, 10, 90, 90, + 229, 229, 11, 11, 106, 106, 214, 214, 153, 153, 27, 27, 199, 199, 43, 43, + 184, 184, 122, 122, 169, 169, 230, 230, 59, 59, 11, 11, 75, 75, 138, 138, + 200, 200, 215, 215, 91, 91, 12, 12, 28, 28, 185, 185, 107, 107, 154, 154, + 44, 44, 231, 231, 216, 216, 60, 60, 123, 123, 12, 12, 76, 76, 201, 201, + 170, 170, 232, 232, 139, 139, 92, 92, 13, 13, 108, 108, 29, 29, 186, 186, + 217, 217, 155, 155, 45, 45, 13, 13, 61, 61, 124, 124, 14, 14, 233, 233, + 77, 77, 14, 14, 171, 171, 140, 140, 202, 202, 30, 30, 93, 93, 109, 109, + 46, 46, 156, 156, 62, 62, 187, 187, 15, 15, 125, 125, 218, 218, 78, 78, + 31, 31, 172, 172, 47, 47, 141, 141, 94, 94, 234, 234, 203, 203, 63, 63, + 110, 110, 188, 188, 157, 157, 126, 126, 79, 79, 173, 173, 95, 95, 219, 219, + 142, 142, 204, 204, 235, 235, 111, 111, 158, 158, 127, 127, 189, 189, 220, + 220, 143, 143, 174, 174, 205, 205, 236, 236, 159, 159, 190, 190, 221, 221, + 175, 175, 237, 237, 206, 206, 222, 222, 191, 191, 238, 238, 207, 207, 223, + 223, 239, 239, 0, 0, +}; + +DECLARE_ALIGNED(16, static const int16_t, + row_scan_16x16_neighbors[257 * MAX_NEIGHBORS]) = { + 0, 0, 0, 0, 1, 1, 0, 0, 2, 2, 16, 16, 3, 3, 17, 17, + 16, 16, 4, 4, 32, 32, 18, 18, 5, 5, 33, 33, 32, 32, 19, 19, + 48, 48, 6, 6, 34, 34, 20, 20, 49, 49, 48, 48, 7, 7, 35, 35, + 64, 64, 21, 21, 50, 50, 36, 36, 64, 64, 8, 8, 65, 65, 51, 51, + 22, 22, 37, 37, 80, 80, 66, 66, 9, 9, 52, 52, 23, 23, 81, 81, + 67, 67, 80, 80, 38, 38, 10, 10, 53, 53, 82, 82, 96, 96, 68, 68, + 24, 24, 97, 97, 83, 83, 39, 39, 96, 96, 54, 54, 11, 11, 69, 69, + 98, 98, 112, 112, 84, 84, 25, 25, 40, 40, 55, 55, 113, 113, 99, 99, + 12, 12, 70, 70, 112, 112, 85, 85, 26, 26, 114, 114, 100, 100, 128, 128, + 41, 41, 56, 56, 71, 71, 115, 115, 13, 13, 86, 86, 129, 129, 101, 101, + 128, 128, 72, 72, 130, 130, 116, 116, 27, 27, 57, 57, 14, 14, 87, 87, + 42, 42, 144, 144, 102, 102, 131, 131, 145, 145, 117, 117, 73, 73, 144, 144, + 88, 88, 132, 132, 103, 103, 28, 28, 58, 58, 146, 146, 118, 118, 43, 43, + 160, 160, 147, 147, 89, 89, 104, 104, 133, 133, 161, 161, 119, 119, 160, 160, + 74, 74, 134, 134, 148, 148, 29, 29, 59, 59, 162, 162, 176, 176, 44, 44, + 120, 120, 90, 90, 105, 105, 163, 163, 177, 177, 149, 149, 176, 176, 135, 135, + 164, 164, 178, 178, 30, 30, 150, 150, 192, 192, 75, 75, 121, 121, 60, 60, + 136, 136, 193, 193, 106, 106, 151, 151, 179, 179, 192, 192, 45, 45, 165, 165, + 166, 166, 194, 194, 91, 91, 180, 180, 137, 137, 208, 208, 122, 122, 152, 152, + 208, 208, 195, 195, 76, 76, 167, 167, 209, 209, 181, 181, 224, 224, 107, 107, + 196, 196, 61, 61, 153, 153, 224, 224, 182, 182, 168, 168, 210, 210, 46, 46, + 138, 138, 92, 92, 183, 183, 225, 225, 211, 211, 240, 240, 197, 197, 169, 169, + 123, 123, 154, 154, 198, 198, 77, 77, 212, 212, 184, 184, 108, 108, 226, 226, + 199, 199, 62, 62, 227, 227, 241, 241, 139, 139, 213, 213, 170, 170, 185, 185, + 155, 155, 228, 228, 242, 242, 124, 124, 93, 93, 200, 200, 243, 243, 214, 214, + 215, 215, 229, 229, 140, 140, 186, 186, 201, 201, 78, 78, 171, 171, 109, 109, + 156, 156, 244, 244, 216, 216, 230, 230, 94, 94, 245, 245, 231, 231, 125, 125, + 202, 202, 246, 246, 232, 232, 172, 172, 217, 217, 141, 141, 110, 110, 157, + 157, 187, 187, 247, 247, 126, 126, 233, 233, 218, 218, 248, 248, 188, 188, + 203, 203, 142, 142, 173, 173, 158, 158, 249, 249, 234, 234, 204, 204, 219, + 219, 174, 174, 189, 189, 250, 250, 220, 220, 190, 190, 205, 205, 235, 235, + 206, 206, 236, 236, 251, 251, 221, 221, 252, 252, 222, 222, 237, 237, 238, + 238, 253, 253, 254, 254, 0, 0, +}; + +DECLARE_ALIGNED(16, static const int16_t, + default_scan_16x16_neighbors[257 * MAX_NEIGHBORS]) = { + 0, 0, 0, 0, 0, 0, 16, 16, 1, 16, 1, 1, 32, 32, 17, 32, + 2, 17, 2, 2, 48, 48, 18, 33, 33, 48, 3, 18, 49, 64, 64, 64, + 34, 49, 3, 3, 19, 34, 50, 65, 4, 19, 65, 80, 80, 80, 35, 50, + 4, 4, 20, 35, 66, 81, 81, 96, 51, 66, 96, 96, 5, 20, 36, 51, + 82, 97, 21, 36, 67, 82, 97, 112, 5, 5, 52, 67, 112, 112, 37, 52, + 6, 21, 83, 98, 98, 113, 68, 83, 6, 6, 113, 128, 22, 37, 53, 68, + 84, 99, 99, 114, 128, 128, 114, 129, 69, 84, 38, 53, 7, 22, 7, 7, + 129, 144, 23, 38, 54, 69, 100, 115, 85, 100, 115, 130, 144, 144, 130, 145, + 39, 54, 70, 85, 8, 23, 55, 70, 116, 131, 101, 116, 145, 160, 24, 39, + 8, 8, 86, 101, 131, 146, 160, 160, 146, 161, 71, 86, 40, 55, 9, 24, + 117, 132, 102, 117, 161, 176, 132, 147, 56, 71, 87, 102, 25, 40, 147, 162, + 9, 9, 176, 176, 162, 177, 72, 87, 41, 56, 118, 133, 133, 148, 103, 118, + 10, 25, 148, 163, 57, 72, 88, 103, 177, 192, 26, 41, 163, 178, 192, 192, + 10, 10, 119, 134, 73, 88, 149, 164, 104, 119, 134, 149, 42, 57, 178, 193, + 164, 179, 11, 26, 58, 73, 193, 208, 89, 104, 135, 150, 120, 135, 27, 42, + 74, 89, 208, 208, 150, 165, 179, 194, 165, 180, 105, 120, 194, 209, 43, 58, + 11, 11, 136, 151, 90, 105, 151, 166, 180, 195, 59, 74, 121, 136, 209, 224, + 195, 210, 224, 224, 166, 181, 106, 121, 75, 90, 12, 27, 181, 196, 12, 12, + 210, 225, 152, 167, 167, 182, 137, 152, 28, 43, 196, 211, 122, 137, 91, 106, + 225, 240, 44, 59, 13, 28, 107, 122, 182, 197, 168, 183, 211, 226, 153, 168, + 226, 241, 60, 75, 197, 212, 138, 153, 29, 44, 76, 91, 13, 13, 183, 198, + 123, 138, 45, 60, 212, 227, 198, 213, 154, 169, 169, 184, 227, 242, 92, 107, + 61, 76, 139, 154, 14, 29, 14, 14, 184, 199, 213, 228, 108, 123, 199, 214, + 228, 243, 77, 92, 30, 45, 170, 185, 155, 170, 185, 200, 93, 108, 124, 139, + 214, 229, 46, 61, 200, 215, 229, 244, 15, 30, 109, 124, 62, 77, 140, 155, + 215, 230, 31, 46, 171, 186, 186, 201, 201, 216, 78, 93, 230, 245, 125, 140, + 47, 62, 216, 231, 156, 171, 94, 109, 231, 246, 141, 156, 63, 78, 202, 217, + 187, 202, 110, 125, 217, 232, 172, 187, 232, 247, 79, 94, 157, 172, 126, 141, + 203, 218, 95, 110, 233, 248, 218, 233, 142, 157, 111, 126, 173, 188, 188, 203, + 234, 249, 219, 234, 127, 142, 158, 173, 204, 219, 189, 204, 143, 158, 235, + 250, 174, 189, 205, 220, 159, 174, 220, 235, 221, 236, 175, 190, 190, 205, + 236, 251, 206, 221, 237, 252, 191, 206, 222, 237, 207, 222, 238, 253, 223, + 238, 239, 254, 0, 0, +}; + +DECLARE_ALIGNED(16, static const int16_t, + default_scan_32x32_neighbors[1025 * MAX_NEIGHBORS]) = { + 0, 0, 0, 0, 0, 0, 32, 32, 1, 32, 1, 1, 64, 64, 33, 64, + 2, 33, 96, 96, 2, 2, 65, 96, 34, 65, 128, 128, 97, 128, 3, 34, + 66, 97, 3, 3, 35, 66, 98, 129, 129, 160, 160, 160, 4, 35, 67, 98, + 192, 192, 4, 4, 130, 161, 161, 192, 36, 67, 99, 130, 5, 36, 68, 99, + 193, 224, 162, 193, 224, 224, 131, 162, 37, 68, 100, 131, 5, 5, 194, 225, + 225, 256, 256, 256, 163, 194, 69, 100, 132, 163, 6, 37, 226, 257, 6, 6, + 195, 226, 257, 288, 101, 132, 288, 288, 38, 69, 164, 195, 133, 164, 258, 289, + 227, 258, 196, 227, 7, 38, 289, 320, 70, 101, 320, 320, 7, 7, 165, 196, + 39, 70, 102, 133, 290, 321, 259, 290, 228, 259, 321, 352, 352, 352, 197, 228, + 134, 165, 71, 102, 8, 39, 322, 353, 291, 322, 260, 291, 103, 134, 353, 384, + 166, 197, 229, 260, 40, 71, 8, 8, 384, 384, 135, 166, 354, 385, 323, 354, + 198, 229, 292, 323, 72, 103, 261, 292, 9, 40, 385, 416, 167, 198, 104, 135, + 230, 261, 355, 386, 416, 416, 293, 324, 324, 355, 9, 9, 41, 72, 386, 417, + 199, 230, 136, 167, 417, 448, 262, 293, 356, 387, 73, 104, 387, 418, 231, 262, + 10, 41, 168, 199, 325, 356, 418, 449, 105, 136, 448, 448, 42, 73, 294, 325, + 200, 231, 10, 10, 357, 388, 137, 168, 263, 294, 388, 419, 74, 105, 419, 450, + 449, 480, 326, 357, 232, 263, 295, 326, 169, 200, 11, 42, 106, 137, 480, 480, + 450, 481, 358, 389, 264, 295, 201, 232, 138, 169, 389, 420, 43, 74, 420, 451, + 327, 358, 11, 11, 481, 512, 233, 264, 451, 482, 296, 327, 75, 106, 170, 201, + 482, 513, 512, 512, 390, 421, 359, 390, 421, 452, 107, 138, 12, 43, 202, 233, + 452, 483, 265, 296, 328, 359, 139, 170, 44, 75, 483, 514, 513, 544, 234, 265, + 297, 328, 422, 453, 12, 12, 391, 422, 171, 202, 76, 107, 514, 545, 453, 484, + 544, 544, 266, 297, 203, 234, 108, 139, 329, 360, 298, 329, 140, 171, 515, + 546, 13, 44, 423, 454, 235, 266, 545, 576, 454, 485, 45, 76, 172, 203, 330, + 361, 576, 576, 13, 13, 267, 298, 546, 577, 77, 108, 204, 235, 455, 486, 577, + 608, 299, 330, 109, 140, 547, 578, 14, 45, 14, 14, 141, 172, 578, 609, 331, + 362, 46, 77, 173, 204, 15, 15, 78, 109, 205, 236, 579, 610, 110, 141, 15, 46, + 142, 173, 47, 78, 174, 205, 16, 16, 79, 110, 206, 237, 16, 47, 111, 142, + 48, 79, 143, 174, 80, 111, 175, 206, 17, 48, 17, 17, 207, 238, 49, 80, + 81, 112, 18, 18, 18, 49, 50, 81, 82, 113, 19, 50, 51, 82, 83, 114, 608, 608, + 484, 515, 360, 391, 236, 267, 112, 143, 19, 19, 640, 640, 609, 640, 516, 547, + 485, 516, 392, 423, 361, 392, 268, 299, 237, 268, 144, 175, 113, 144, 20, 51, + 20, 20, 672, 672, 641, 672, 610, 641, 548, 579, 517, 548, 486, 517, 424, 455, + 393, 424, 362, 393, 300, 331, 269, 300, 238, 269, 176, 207, 145, 176, 114, + 145, 52, 83, 21, 52, 21, 21, 704, 704, 673, 704, 642, 673, 611, 642, 580, + 611, 549, 580, 518, 549, 487, 518, 456, 487, 425, 456, 394, 425, 363, 394, + 332, 363, 301, 332, 270, 301, 239, 270, 208, 239, 177, 208, 146, 177, 115, + 146, 84, 115, 53, 84, 22, 53, 22, 22, 705, 736, 674, 705, 643, 674, 581, 612, + 550, 581, 519, 550, 457, 488, 426, 457, 395, 426, 333, 364, 302, 333, 271, + 302, 209, 240, 178, 209, 147, 178, 85, 116, 54, 85, 23, 54, 706, 737, 675, + 706, 582, 613, 551, 582, 458, 489, 427, 458, 334, 365, 303, 334, 210, 241, + 179, 210, 86, 117, 55, 86, 707, 738, 583, 614, 459, 490, 335, 366, 211, 242, + 87, 118, 736, 736, 612, 643, 488, 519, 364, 395, 240, 271, 116, 147, 23, 23, + 768, 768, 737, 768, 644, 675, 613, 644, 520, 551, 489, 520, 396, 427, 365, + 396, 272, 303, 241, 272, 148, 179, 117, 148, 24, 55, 24, 24, 800, 800, 769, + 800, 738, 769, 676, 707, 645, 676, 614, 645, 552, 583, 521, 552, 490, 521, + 428, 459, 397, 428, 366, 397, 304, 335, 273, 304, 242, 273, 180, 211, 149, + 180, 118, 149, 56, 87, 25, 56, 25, 25, 832, 832, 801, 832, 770, 801, 739, + 770, 708, 739, 677, 708, 646, 677, 615, 646, 584, 615, 553, 584, 522, 553, + 491, 522, 460, 491, 429, 460, 398, 429, 367, 398, 336, 367, 305, 336, 274, + 305, 243, 274, 212, 243, 181, 212, 150, 181, 119, 150, 88, 119, 57, 88, 26, + 57, 26, 26, 833, 864, 802, 833, 771, 802, 709, 740, 678, 709, 647, 678, 585, + 616, 554, 585, 523, 554, 461, 492, 430, 461, 399, 430, 337, 368, 306, 337, + 275, 306, 213, 244, 182, 213, 151, 182, 89, 120, 58, 89, 27, 58, 834, 865, + 803, 834, 710, 741, 679, 710, 586, 617, 555, 586, 462, 493, 431, 462, 338, + 369, 307, 338, 214, 245, 183, 214, 90, 121, 59, 90, 835, 866, 711, 742, 587, + 618, 463, 494, 339, 370, 215, 246, 91, 122, 864, 864, 740, 771, 616, 647, + 492, 523, 368, 399, 244, 275, 120, 151, 27, 27, 896, 896, 865, 896, 772, 803, + 741, 772, 648, 679, 617, 648, 524, 555, 493, 524, 400, 431, 369, 400, 276, + 307, 245, 276, 152, 183, 121, 152, 28, 59, 28, 28, 928, 928, 897, 928, 866, + 897, 804, 835, 773, 804, 742, 773, 680, 711, 649, 680, 618, 649, 556, 587, + 525, 556, 494, 525, 432, 463, 401, 432, 370, 401, 308, 339, 277, 308, 246, + 277, 184, 215, 153, 184, 122, 153, 60, 91, 29, 60, 29, 29, 960, 960, 929, + 960, 898, 929, 867, 898, 836, 867, 805, 836, 774, 805, 743, 774, 712, 743, + 681, 712, 650, 681, 619, 650, 588, 619, 557, 588, 526, 557, 495, 526, 464, + 495, 433, 464, 402, 433, 371, 402, 340, 371, 309, 340, 278, 309, 247, 278, + 216, 247, 185, 216, 154, 185, 123, 154, 92, 123, 61, 92, 30, 61, 30, 30, + 961, 992, 930, 961, 899, 930, 837, 868, 806, 837, 775, 806, 713, 744, 682, + 713, 651, 682, 589, 620, 558, 589, 527, 558, 465, 496, 434, 465, 403, 434, + 341, 372, 310, 341, 279, 310, 217, 248, 186, 217, 155, 186, 93, 124, 62, 93, + 31, 62, 962, 993, 931, 962, 838, 869, 807, 838, 714, 745, 683, 714, 590, 621, + 559, 590, 466, 497, 435, 466, 342, 373, 311, 342, 218, 249, 187, 218, 94, + 125, 63, 94, 963, 994, 839, 870, 715, 746, 591, 622, 467, 498, 343, 374, 219, + 250, 95, 126, 868, 899, 744, 775, 620, 651, 496, 527, 372, 403, 248, 279, + 124, 155, 900, 931, 869, 900, 776, 807, 745, 776, 652, 683, 621, 652, 528, + 559, 497, 528, 404, 435, 373, 404, 280, 311, 249, 280, 156, 187, 125, 156, + 932, 963, 901, 932, 870, 901, 808, 839, 777, 808, 746, 777, 684, 715, 653, + 684, 622, 653, 560, 591, 529, 560, 498, 529, 436, 467, 405, 436, 374, 405, + 312, 343, 281, 312, 250, 281, 188, 219, 157, 188, 126, 157, 964, 995, 933, + 964, 902, 933, 871, 902, 840, 871, 809, 840, 778, 809, 747, 778, 716, 747, + 685, 716, 654, 685, 623, 654, 592, 623, 561, 592, 530, 561, 499, 530, 468, + 499, 437, 468, 406, 437, 375, 406, 344, 375, 313, 344, 282, 313, 251, 282, + 220, 251, 189, 220, 158, 189, 127, 158, 965, 996, 934, 965, 903, 934, 841, + 872, 810, 841, 779, 810, 717, 748, 686, 717, 655, 686, 593, 624, 562, 593, + 531, 562, 469, 500, 438, 469, 407, 438, 345, 376, 314, 345, 283, 314, 221, + 252, 190, 221, 159, 190, 966, 997, 935, 966, 842, 873, 811, 842, 718, 749, + 687, 718, 594, 625, 563, 594, 470, 501, 439, 470, 346, 377, 315, 346, 222, + 253, 191, 222, 967, 998, 843, 874, 719, 750, 595, 626, 471, 502, 347, 378, + 223, 254, 872, 903, 748, 779, 624, 655, 500, 531, 376, 407, 252, 283, 904, + 935, 873, 904, 780, 811, 749, 780, 656, 687, 625, 656, 532, 563, 501, 532, + 408, 439, 377, 408, 284, 315, 253, 284, 936, 967, 905, 936, 874, 905, 812, + 843, 781, 812, 750, 781, 688, 719, 657, 688, 626, 657, 564, 595, 533, 564, + 502, 533, 440, 471, 409, 440, 378, 409, 316, 347, 285, 316, 254, 285, 968, + 999, 937, 968, 906, 937, 875, 906, 844, 875, 813, 844, 782, 813, 751, 782, + 720, 751, 689, 720, 658, 689, 627, 658, 596, 627, 565, 596, 534, 565, 503, + 534, 472, 503, 441, 472, 410, 441, 379, 410, 348, 379, 317, 348, 286, 317, + 255, 286, 969, 1000, 938, 969, 907, 938, 845, 876, 814, 845, 783, 814, 721, + 752, 690, 721, 659, 690, 597, 628, 566, 597, 535, 566, 473, 504, 442, 473, + 411, 442, 349, 380, 318, 349, 287, 318, 970, 1001, 939, 970, 846, 877, 815, + 846, 722, 753, 691, 722, 598, 629, 567, 598, 474, 505, 443, 474, 350, 381, + 319, 350, 971, 1002, 847, 878, 723, 754, 599, 630, 475, 506, 351, 382, 876, + 907, 752, 783, 628, 659, 504, 535, 380, 411, 908, 939, 877, 908, 784, 815, + 753, 784, 660, 691, 629, 660, 536, 567, 505, 536, 412, 443, 381, 412, 940, + 971, 909, 940, 878, 909, 816, 847, 785, 816, 754, 785, 692, 723, 661, 692, + 630, 661, 568, 599, 537, 568, 506, 537, 444, 475, 413, 444, 382, 413, 972, + 1003, 941, 972, 910, 941, 879, 910, 848, 879, 817, 848, 786, 817, 755, 786, + 724, 755, 693, 724, 662, 693, 631, 662, 600, 631, 569, 600, 538, 569, 507, + 538, 476, 507, 445, 476, 414, 445, 383, 414, 973, 1004, 942, 973, 911, 942, + 849, 880, 818, 849, 787, 818, 725, 756, 694, 725, 663, 694, 601, 632, 570, + 601, 539, 570, 477, 508, 446, 477, 415, 446, 974, 1005, 943, 974, 850, 881, + 819, 850, 726, 757, 695, 726, 602, 633, 571, 602, 478, 509, 447, 478, 975, + 1006, 851, 882, 727, 758, 603, 634, 479, 510, 880, 911, 756, 787, 632, 663, + 508, 539, 912, 943, 881, 912, 788, 819, 757, 788, 664, 695, 633, 664, 540, + 571, 509, 540, 944, 975, 913, 944, 882, 913, 820, 851, 789, 820, 758, 789, + 696, 727, 665, 696, 634, 665, 572, 603, 541, 572, 510, 541, 976, 1007, 945, + 976, 914, 945, 883, 914, 852, 883, 821, 852, 790, 821, 759, 790, 728, 759, + 697, 728, 666, 697, 635, 666, 604, 635, 573, 604, 542, 573, 511, 542, 977, + 1008, 946, 977, 915, 946, 853, 884, 822, 853, 791, 822, 729, 760, 698, 729, + 667, 698, 605, 636, 574, 605, 543, 574, 978, 1009, 947, 978, 854, 885, 823, + 854, 730, 761, 699, 730, 606, 637, 575, 606, 979, 1010, 855, 886, 731, 762, + 607, 638, 884, 915, 760, 791, 636, 667, 916, 947, 885, 916, 792, 823, 761, + 792, 668, 699, 637, 668, 948, 979, 917, 948, 886, 917, 824, 855, 793, 824, + 762, 793, 700, 731, 669, 700, 638, 669, 980, 1011, 949, 980, 918, 949, 887, + 918, 856, 887, 825, 856, 794, 825, 763, 794, 732, 763, 701, 732, 670, 701, + 639, 670, 981, 1012, 950, 981, 919, 950, 857, 888, 826, 857, 795, 826, 733, + 764, 702, 733, 671, 702, 982, 1013, 951, 982, 858, 889, 827, 858, 734, 765, + 703, 734, 983, 1014, 859, 890, 735, 766, 888, 919, 764, 795, 920, 951, 889, + 920, 796, 827, 765, 796, 952, 983, 921, 952, 890, 921, 828, 859, 797, 828, + 766, 797, 984, 1015, 953, 984, 922, 953, 891, 922, 860, 891, 829, 860, 798, + 829, 767, 798, 985, 1016, 954, 985, 923, 954, 861, 892, 830, 861, 799, 830, + 986, 1017, 955, 986, 862, 893, 831, 862, 987, 1018, 863, 894, 892, 923, 924, + 955, 893, 924, 956, 987, 925, 956, 894, 925, 988, 1019, 957, 988, 926, 957, + 895, 926, 989, 1020, 958, 989, 927, 958, 990, 1021, 959, 990, 991, 1022, 0, 0, +}; + +DECLARE_ALIGNED(16, static const int16_t, vp9_default_iscan_4x4[16]) = { + 0, 2, 5, 8, 1, 3, 9, 12, 4, 7, 11, 14, 6, 10, 13, 15, +}; + +DECLARE_ALIGNED(16, static const int16_t, vp9_col_iscan_4x4[16]) = { + 0, 3, 7, 11, 1, 5, 9, 12, 2, 6, 10, 14, 4, 8, 13, 15, +}; + +DECLARE_ALIGNED(16, static const int16_t, vp9_row_iscan_4x4[16]) = { + 0, 1, 3, 5, 2, 4, 6, 9, 7, 8, 11, 13, 10, 12, 14, 15, +}; + +DECLARE_ALIGNED(16, static const int16_t, vp9_col_iscan_8x8[64]) = { + 0, 3, 8, 15, 22, 32, 40, 47, 1, 5, 11, 18, 26, 34, 44, 51, + 2, 7, 13, 20, 28, 38, 46, 54, 4, 10, 16, 24, 31, 41, 50, 56, + 6, 12, 21, 27, 35, 43, 52, 58, 9, 17, 25, 33, 39, 48, 55, 60, + 14, 23, 30, 37, 45, 53, 59, 62, 19, 29, 36, 42, 49, 57, 61, 63, +}; + +DECLARE_ALIGNED(16, static const int16_t, vp9_row_iscan_8x8[64]) = { + 0, 1, 2, 5, 8, 12, 19, 24, 3, 4, 7, 10, 15, 20, 30, 39, + 6, 9, 13, 16, 21, 27, 37, 46, 11, 14, 17, 23, 28, 34, 44, 52, + 18, 22, 25, 31, 35, 41, 50, 57, 26, 29, 33, 38, 43, 49, 55, 59, + 32, 36, 42, 47, 51, 54, 60, 61, 40, 45, 48, 53, 56, 58, 62, 63, +}; + +DECLARE_ALIGNED(16, static const int16_t, vp9_default_iscan_8x8[64]) = { + 0, 2, 5, 9, 14, 22, 31, 37, 1, 4, 8, 13, 19, 26, 38, 44, + 3, 6, 10, 17, 24, 30, 42, 49, 7, 11, 15, 21, 29, 36, 47, 53, + 12, 16, 20, 27, 34, 43, 52, 57, 18, 23, 28, 35, 41, 48, 56, 60, + 25, 32, 39, 45, 50, 55, 59, 62, 33, 40, 46, 51, 54, 58, 61, 63, +}; + +DECLARE_ALIGNED(16, static const int16_t, vp9_col_iscan_16x16[256]) = { + 0, 4, 11, 20, 31, 43, 59, 75, 85, 109, 130, 150, 165, 181, 195, 198, + 1, 6, 14, 23, 34, 47, 64, 81, 95, 114, 135, 153, 171, 188, 201, 212, + 2, 8, 16, 25, 38, 52, 67, 83, 101, 116, 136, 157, 172, 190, 205, 216, + 3, 10, 18, 29, 41, 55, 71, 89, 103, 119, 141, 159, 176, 194, 208, 218, + 5, 12, 21, 32, 45, 58, 74, 93, 104, 123, 144, 164, 179, 196, 210, 223, + 7, 15, 26, 37, 49, 63, 78, 96, 112, 129, 146, 166, 182, 200, 215, 228, + 9, 19, 28, 39, 54, 69, 86, 102, 117, 132, 151, 170, 187, 206, 220, 230, + 13, 24, 35, 46, 60, 73, 91, 108, 122, 137, 154, 174, 189, 207, 224, 235, + 17, 30, 40, 53, 66, 82, 98, 115, 126, 142, 161, 180, 197, 213, 227, 237, + 22, 36, 48, 62, 76, 92, 105, 120, 133, 147, 167, 186, 203, 219, 232, 240, + 27, 44, 56, 70, 84, 99, 113, 127, 140, 156, 175, 193, 209, 226, 236, 244, + 33, 51, 68, 79, 94, 110, 125, 138, 149, 162, 184, 202, 217, 229, 241, 247, + 42, 61, 77, 90, 106, 121, 134, 148, 160, 173, 191, 211, 225, 238, 245, 251, + 50, 72, 87, 100, 118, 128, 145, 158, 168, 183, 204, 222, 233, 242, 249, 253, + 57, 80, 97, 111, 131, 143, 155, 169, 178, 192, 214, 231, 239, 246, 250, 254, + 65, 88, 107, 124, 139, 152, 163, 177, 185, 199, 221, 234, 243, 248, 252, 255, +}; + +DECLARE_ALIGNED(16, static const int16_t, vp9_row_iscan_16x16[256]) = { + 0, 1, 2, 4, 6, 9, 12, 17, 22, 29, 36, 43, 54, 64, 76, 86, + 3, 5, 7, 11, 15, 19, 25, 32, 38, 48, 59, 68, 84, 99, 115, 130, + 8, 10, 13, 18, 23, 27, 33, 42, 51, 60, 72, 88, 103, 119, 142, 167, + 14, 16, 20, 26, 31, 37, 44, 53, 61, 73, 85, 100, 116, 135, 161, 185, + 21, 24, 30, 35, 40, 47, 55, 65, 74, 81, 94, 112, 133, 154, 179, 205, + 28, 34, 39, 45, 50, 58, 67, 77, 87, 96, 106, 121, 146, 169, 196, 212, + 41, 46, 49, 56, 63, 70, 79, 90, 98, 107, 122, 138, 159, 182, 207, 222, + 52, 57, 62, 69, 75, 83, 93, 102, 110, 120, 134, 150, 176, 195, 215, 226, + 66, 71, 78, 82, 91, 97, 108, 113, 127, 136, 148, 168, 188, 202, 221, 232, + 80, 89, 92, 101, 105, 114, 125, 131, 139, 151, 162, 177, 192, 208, 223, 234, + 95, 104, 109, 117, 123, 128, 143, 144, 155, 165, 175, 190, 206, 219, 233, 239, + 111, 118, 124, 129, 140, 147, 157, 164, 170, 181, 191, 203, 224, 230, 240, + 243, 126, 132, 137, 145, 153, 160, 174, 178, 184, 197, 204, 216, 231, 237, + 244, 246, 141, 149, 156, 166, 172, 180, 189, 199, 200, 210, 220, 228, 238, + 242, 249, 251, 152, 163, 171, 183, 186, 193, 201, 211, 214, 218, 227, 236, + 245, 247, 252, 253, 158, 173, 187, 194, 198, 209, 213, 217, 225, 229, 235, + 241, 248, 250, 254, 255, +}; + +DECLARE_ALIGNED(16, static const int16_t, vp9_default_iscan_16x16[256]) = { + 0, 2, 5, 9, 17, 24, 36, 44, 55, 72, 88, 104, 128, 143, 166, 179, + 1, 4, 8, 13, 20, 30, 40, 54, 66, 79, 96, 113, 141, 154, 178, 196, + 3, 7, 11, 18, 25, 33, 46, 57, 71, 86, 101, 119, 148, 164, 186, 201, + 6, 12, 16, 23, 31, 39, 53, 64, 78, 92, 110, 127, 153, 169, 193, 208, + 10, 14, 19, 28, 37, 47, 58, 67, 84, 98, 114, 133, 161, 176, 198, 214, + 15, 21, 26, 34, 43, 52, 65, 77, 91, 106, 120, 140, 165, 185, 205, 221, + 22, 27, 32, 41, 48, 60, 73, 85, 99, 116, 130, 151, 175, 190, 211, 225, + 29, 35, 42, 49, 59, 69, 81, 95, 108, 125, 139, 155, 182, 197, 217, 229, + 38, 45, 51, 61, 68, 80, 93, 105, 118, 134, 150, 168, 191, 207, 223, 234, + 50, 56, 63, 74, 83, 94, 109, 117, 129, 147, 163, 177, 199, 213, 228, 238, + 62, 70, 76, 87, 97, 107, 122, 131, 145, 159, 172, 188, 210, 222, 235, 242, + 75, 82, 90, 102, 112, 124, 138, 146, 157, 173, 187, 202, 219, 230, 240, 245, + 89, 100, 111, 123, 132, 142, 156, 167, 180, 189, 203, 216, 231, 237, 246, 250, + 103, 115, 126, 136, 149, 162, 171, 183, 194, 204, 215, 224, 236, 241, 248, + 252, 121, 135, 144, 158, 170, 181, 192, 200, 209, 218, 227, 233, 243, 244, + 251, 254, 137, 152, 160, 174, 184, 195, 206, 212, 220, 226, 232, 239, 247, + 249, 253, 255, +}; + +DECLARE_ALIGNED(16, static const int16_t, vp9_default_iscan_32x32[1024]) = { + 0, 2, 5, 10, 17, 25, 38, 47, 62, 83, 101, 121, 145, 170, 193, 204, + 210, 219, 229, 233, 245, 257, 275, 299, 342, 356, 377, 405, 455, 471, 495, + 527, 1, 4, 8, 15, 22, 30, 45, 58, 74, 92, 112, 133, 158, 184, 203, 215, 222, + 228, 234, 237, 256, 274, 298, 317, 355, 376, 404, 426, 470, 494, 526, 551, + 3, 7, 12, 18, 28, 36, 52, 64, 82, 102, 118, 142, 164, 189, 208, 217, 224, + 231, 235, 238, 273, 297, 316, 329, 375, 403, 425, 440, 493, 525, 550, 567, + 6, 11, 16, 23, 31, 43, 60, 73, 90, 109, 126, 150, 173, 196, 211, 220, 226, + 232, 236, 239, 296, 315, 328, 335, 402, 424, 439, 447, 524, 549, 566, 575, + 9, 14, 19, 29, 37, 50, 65, 78, 95, 116, 134, 157, 179, 201, 214, 223, 244, + 255, 272, 295, 341, 354, 374, 401, 454, 469, 492, 523, 582, 596, 617, 645, + 13, 20, 26, 35, 44, 54, 72, 85, 105, 123, 140, 163, 182, 205, 216, 225, + 254, 271, 294, 314, 353, 373, 400, 423, 468, 491, 522, 548, 595, 616, 644, + 666, 21, 27, 33, 42, 53, 63, 80, 94, 113, 132, 151, 172, 190, 209, 218, 227, + 270, 293, 313, 327, 372, 399, 422, 438, 490, 521, 547, 565, 615, 643, 665, + 680, 24, 32, 39, 48, 57, 71, 88, 104, 120, 139, 159, 178, 197, 212, 221, 230, + 292, 312, 326, 334, 398, 421, 437, 446, 520, 546, 564, 574, 642, 664, 679, + 687, 34, 40, 46, 56, 68, 81, 96, 111, 130, 147, 167, 186, 243, 253, 269, 291, + 340, 352, 371, 397, 453, 467, 489, 519, 581, 594, 614, 641, 693, 705, 723, + 747, 41, 49, 55, 67, 77, 91, 107, 124, 138, 161, 177, 194, 252, 268, 290, + 311, 351, 370, 396, 420, 466, 488, 518, 545, 593, 613, 640, 663, 704, 722, + 746, 765, 51, 59, 66, 76, 89, 99, 119, 131, 149, 168, 181, 200, 267, 289, + 310, 325, 369, 395, 419, 436, 487, 517, 544, 563, 612, 639, 662, 678, 721, + 745, 764, 777, 61, 69, 75, 87, 100, 114, 129, 144, 162, 180, 191, 207, 288, + 309, 324, 333, 394, 418, 435, 445, 516, 543, 562, 573, 638, 661, 677, 686, + 744, 763, 776, 783, 70, 79, 86, 97, 108, 122, 137, 155, 242, 251, 266, 287, + 339, 350, 368, 393, 452, 465, 486, 515, 580, 592, 611, 637, 692, 703, 720, + 743, 788, 798, 813, 833, 84, 93, 103, 110, 125, 141, 154, 171, 250, 265, 286, + 308, 349, 367, 392, 417, 464, 485, 514, 542, 591, 610, 636, 660, 702, 719, + 742, 762, 797, 812, 832, 848, 98, 106, 115, 127, 143, 156, 169, 185, 264, + 285, 307, 323, 366, 391, 416, 434, 484, 513, 541, 561, 609, 635, 659, 676, + 718, 741, 761, 775, 811, 831, 847, 858, 117, 128, 136, 148, 160, 175, 188, + 198, 284, 306, 322, 332, 390, 415, 433, 444, 512, 540, 560, 572, 634, 658, + 675, 685, 740, 760, 774, 782, 830, 846, 857, 863, 135, 146, 152, 165, 241, + 249, 263, 283, 338, 348, 365, 389, 451, 463, 483, 511, 579, 590, 608, 633, + 691, 701, 717, 739, 787, 796, 810, 829, 867, 875, 887, 903, 153, 166, 174, + 183, 248, 262, 282, 305, 347, 364, 388, 414, 462, 482, 510, 539, 589, 607, + 632, 657, 700, 716, 738, 759, 795, 809, 828, 845, 874, 886, 902, 915, 176, + 187, 195, 202, 261, 281, 304, 321, 363, 387, 413, 432, 481, 509, 538, 559, + 606, 631, 656, 674, 715, 737, 758, 773, 808, 827, 844, 856, 885, 901, 914, + 923, 192, 199, 206, 213, 280, 303, 320, 331, 386, 412, 431, 443, 508, 537, + 558, 571, 630, 655, 673, 684, 736, 757, 772, 781, 826, 843, 855, 862, 900, + 913, 922, 927, 240, 247, 260, 279, 337, 346, 362, 385, 450, 461, 480, 507, + 578, 588, 605, 629, 690, 699, 714, 735, 786, 794, 807, 825, 866, 873, 884, + 899, 930, 936, 945, 957, 246, 259, 278, 302, 345, 361, 384, 411, 460, 479, + 506, 536, 587, 604, 628, 654, 698, 713, 734, 756, 793, 806, 824, 842, 872, + 883, 898, 912, 935, 944, 956, 966, 258, 277, 301, 319, 360, 383, 410, 430, + 478, 505, 535, 557, 603, 627, 653, 672, 712, 733, 755, 771, 805, 823, 841, + 854, 882, 897, 911, 921, 943, 955, 965, 972, 276, 300, 318, 330, 382, 409, + 429, 442, 504, 534, 556, 570, 626, 652, 671, 683, 732, 754, 770, 780, 822, + 840, 853, 861, 896, 910, 920, 926, 954, 964, 971, 975, 336, 344, 359, 381, + 449, 459, 477, 503, 577, 586, 602, 625, 689, 697, 711, 731, 785, 792, 804, + 821, 865, 871, 881, 895, 929, 934, 942, 953, 977, 981, 987, 995, 343, 358, + 380, 408, 458, 476, 502, 533, 585, 601, 624, 651, 696, 710, 730, 753, 791, + 803, 820, 839, 870, 880, 894, 909, 933, 941, 952, 963, 980, 986, 994, 1001, + 357, 379, 407, 428, 475, 501, 532, 555, 600, 623, 650, 670, 709, 729, 752, + 769, 802, 819, 838, 852, 879, 893, 908, 919, 940, 951, 962, 970, 985, 993, + 1000, 1005, 378, 406, 427, 441, 500, 531, 554, 569, 622, 649, 669, 682, 728, + 751, 768, 779, 818, 837, 851, 860, 892, 907, 918, 925, 950, 961, 969, 974, + 992, 999, 1004, 1007, 448, 457, 474, 499, 576, 584, 599, 621, 688, 695, 708, + 727, 784, 790, 801, 817, 864, 869, 878, 891, 928, 932, 939, 949, 976, 979, + 984, 991, 1008, 1010, 1013, 1017, 456, 473, 498, 530, 583, 598, 620, 648, + 694, 707, 726, 750, 789, 800, 816, 836, 868, 877, 890, 906, 931, 938, 948, + 960, 978, 983, 990, 998, 1009, 1012, 1016, 1020, 472, 497, 529, 553, 597, + 619, 647, 668, 706, 725, 749, 767, 799, 815, 835, 850, 876, 889, 905, 917, + 937, 947, 959, 968, 982, 989, 997, 1003, 1011, 1015, 1019, 1022, 496, 528, + 552, 568, 618, 646, 667, 681, 724, 748, 766, 778, 814, 834, 849, 859, 888, + 904, 916, 924, 946, 958, 967, 973, 988, 996, 1002, 1006, 1014, 1018, 1021, + 1023, +}; + +const scan_order vp9_default_scan_orders[TX_SIZES] = { + {default_scan_4x4, vp9_default_iscan_4x4, default_scan_4x4_neighbors}, + {default_scan_8x8, vp9_default_iscan_8x8, default_scan_8x8_neighbors}, + {default_scan_16x16, vp9_default_iscan_16x16, default_scan_16x16_neighbors}, + {default_scan_32x32, vp9_default_iscan_32x32, default_scan_32x32_neighbors}, +}; + +const scan_order vp9_scan_orders[TX_SIZES][TX_TYPES] = { + { // TX_4X4 + {default_scan_4x4, vp9_default_iscan_4x4, default_scan_4x4_neighbors}, + {row_scan_4x4, vp9_row_iscan_4x4, row_scan_4x4_neighbors}, + {col_scan_4x4, vp9_col_iscan_4x4, col_scan_4x4_neighbors}, + {default_scan_4x4, vp9_default_iscan_4x4, default_scan_4x4_neighbors} + }, { // TX_8X8 + {default_scan_8x8, vp9_default_iscan_8x8, default_scan_8x8_neighbors}, + {row_scan_8x8, vp9_row_iscan_8x8, row_scan_8x8_neighbors}, + {col_scan_8x8, vp9_col_iscan_8x8, col_scan_8x8_neighbors}, + {default_scan_8x8, vp9_default_iscan_8x8, default_scan_8x8_neighbors} + }, { // TX_16X16 + {default_scan_16x16, vp9_default_iscan_16x16, default_scan_16x16_neighbors}, + {row_scan_16x16, vp9_row_iscan_16x16, row_scan_16x16_neighbors}, + {col_scan_16x16, vp9_col_iscan_16x16, col_scan_16x16_neighbors}, + {default_scan_16x16, vp9_default_iscan_16x16, default_scan_16x16_neighbors} + }, { // TX_32X32 + {default_scan_32x32, vp9_default_iscan_32x32, default_scan_32x32_neighbors}, + {default_scan_32x32, vp9_default_iscan_32x32, default_scan_32x32_neighbors}, + {default_scan_32x32, vp9_default_iscan_32x32, default_scan_32x32_neighbors}, + {default_scan_32x32, vp9_default_iscan_32x32, default_scan_32x32_neighbors}, + } +}; diff --git a/thirdparty/libvpx/vp9/common/vp9_scan.h b/thirdparty/libvpx/vp9/common/vp9_scan.h new file mode 100644 index 0000000000..4c1ee8107c --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_scan.h @@ -0,0 +1,57 @@ +/* + * Copyright (c) 2013 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_COMMON_VP9_SCAN_H_ +#define VP9_COMMON_VP9_SCAN_H_ + +#include "vpx/vpx_integer.h" +#include "vpx_ports/mem.h" + +#include "vp9/common/vp9_enums.h" +#include "vp9/common/vp9_blockd.h" + +#ifdef __cplusplus +extern "C" { +#endif + +#define MAX_NEIGHBORS 2 + +typedef struct { + const int16_t *scan; + const int16_t *iscan; + const int16_t *neighbors; +} scan_order; + +extern const scan_order vp9_default_scan_orders[TX_SIZES]; +extern const scan_order vp9_scan_orders[TX_SIZES][TX_TYPES]; + +static INLINE int get_coef_context(const int16_t *neighbors, + const uint8_t *token_cache, int c) { + return (1 + token_cache[neighbors[MAX_NEIGHBORS * c + 0]] + + token_cache[neighbors[MAX_NEIGHBORS * c + 1]]) >> 1; +} + +static INLINE const scan_order *get_scan(const MACROBLOCKD *xd, TX_SIZE tx_size, + PLANE_TYPE type, int block_idx) { + const MODE_INFO *const mi = xd->mi[0]; + + if (is_inter_block(mi) || type != PLANE_TYPE_Y || xd->lossless) { + return &vp9_default_scan_orders[tx_size]; + } else { + const PREDICTION_MODE mode = get_y_mode(mi, block_idx); + return &vp9_scan_orders[tx_size][intra_mode_to_tx_type_lookup[mode]]; + } +} + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_SCAN_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_seg_common.c b/thirdparty/libvpx/vp9/common/vp9_seg_common.c new file mode 100644 index 0000000000..7af61629a0 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_seg_common.c @@ -0,0 +1,64 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <assert.h> + +#include "vp9/common/vp9_blockd.h" +#include "vp9/common/vp9_loopfilter.h" +#include "vp9/common/vp9_seg_common.h" +#include "vp9/common/vp9_quant_common.h" + +static const int seg_feature_data_signed[SEG_LVL_MAX] = { 1, 1, 0, 0 }; + +static const int seg_feature_data_max[SEG_LVL_MAX] = { + MAXQ, MAX_LOOP_FILTER, 3, 0 }; + +// These functions provide access to new segment level features. +// Eventually these function may be "optimized out" but for the moment, +// the coding mechanism is still subject to change so these provide a +// convenient single point of change. + +void vp9_clearall_segfeatures(struct segmentation *seg) { + vp9_zero(seg->feature_data); + vp9_zero(seg->feature_mask); + seg->aq_av_offset = 0; +} + +void vp9_enable_segfeature(struct segmentation *seg, int segment_id, + SEG_LVL_FEATURES feature_id) { + seg->feature_mask[segment_id] |= 1 << feature_id; +} + +int vp9_seg_feature_data_max(SEG_LVL_FEATURES feature_id) { + return seg_feature_data_max[feature_id]; +} + +int vp9_is_segfeature_signed(SEG_LVL_FEATURES feature_id) { + return seg_feature_data_signed[feature_id]; +} + +void vp9_set_segdata(struct segmentation *seg, int segment_id, + SEG_LVL_FEATURES feature_id, int seg_data) { + assert(seg_data <= seg_feature_data_max[feature_id]); + if (seg_data < 0) { + assert(seg_feature_data_signed[feature_id]); + assert(-seg_data <= seg_feature_data_max[feature_id]); + } + + seg->feature_data[segment_id][feature_id] = seg_data; +} + +const vpx_tree_index vp9_segment_tree[TREE_SIZE(MAX_SEGMENTS)] = { + 2, 4, 6, 8, 10, 12, + 0, -1, -2, -3, -4, -5, -6, -7 +}; + + +// TBD? Functions to read and write segment data with range / validity checking diff --git a/thirdparty/libvpx/vp9/common/vp9_seg_common.h b/thirdparty/libvpx/vp9/common/vp9_seg_common.h new file mode 100644 index 0000000000..99a9440c17 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_seg_common.h @@ -0,0 +1,87 @@ +/* + * Copyright (c) 2012 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_COMMON_VP9_SEG_COMMON_H_ +#define VP9_COMMON_VP9_SEG_COMMON_H_ + +#include "vpx_dsp/prob.h" + +#ifdef __cplusplus +extern "C" { +#endif + +#define SEGMENT_DELTADATA 0 +#define SEGMENT_ABSDATA 1 + +#define MAX_SEGMENTS 8 +#define SEG_TREE_PROBS (MAX_SEGMENTS-1) + +#define PREDICTION_PROBS 3 + +// Segment level features. +typedef enum { + SEG_LVL_ALT_Q = 0, // Use alternate Quantizer .... + SEG_LVL_ALT_LF = 1, // Use alternate loop filter value... + SEG_LVL_REF_FRAME = 2, // Optional Segment reference frame + SEG_LVL_SKIP = 3, // Optional Segment (0,0) + skip mode + SEG_LVL_MAX = 4 // Number of features supported +} SEG_LVL_FEATURES; + + +struct segmentation { + uint8_t enabled; + uint8_t update_map; + uint8_t update_data; + uint8_t abs_delta; + uint8_t temporal_update; + + vpx_prob tree_probs[SEG_TREE_PROBS]; + vpx_prob pred_probs[PREDICTION_PROBS]; + + int16_t feature_data[MAX_SEGMENTS][SEG_LVL_MAX]; + uint32_t feature_mask[MAX_SEGMENTS]; + int aq_av_offset; +}; + +static INLINE int segfeature_active(const struct segmentation *seg, + int segment_id, + SEG_LVL_FEATURES feature_id) { + return seg->enabled && + (seg->feature_mask[segment_id] & (1 << feature_id)); +} + +void vp9_clearall_segfeatures(struct segmentation *seg); + +void vp9_enable_segfeature(struct segmentation *seg, + int segment_id, + SEG_LVL_FEATURES feature_id); + +int vp9_seg_feature_data_max(SEG_LVL_FEATURES feature_id); + +int vp9_is_segfeature_signed(SEG_LVL_FEATURES feature_id); + +void vp9_set_segdata(struct segmentation *seg, + int segment_id, + SEG_LVL_FEATURES feature_id, + int seg_data); + +static INLINE int get_segdata(const struct segmentation *seg, int segment_id, + SEG_LVL_FEATURES feature_id) { + return seg->feature_data[segment_id][feature_id]; +} + +extern const vpx_tree_index vp9_segment_tree[TREE_SIZE(MAX_SEGMENTS)]; + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_SEG_COMMON_H_ + diff --git a/thirdparty/libvpx/vp9/common/vp9_thread_common.c b/thirdparty/libvpx/vp9/common/vp9_thread_common.c new file mode 100644 index 0000000000..db78d6be89 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_thread_common.c @@ -0,0 +1,435 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "./vpx_config.h" +#include "vpx_dsp/vpx_dsp_common.h" +#include "vpx_mem/vpx_mem.h" +#include "vp9/common/vp9_entropymode.h" +#include "vp9/common/vp9_thread_common.h" +#include "vp9/common/vp9_reconinter.h" +#include "vp9/common/vp9_loopfilter.h" + +#if CONFIG_MULTITHREAD +static INLINE void mutex_lock(pthread_mutex_t *const mutex) { + const int kMaxTryLocks = 4000; + int locked = 0; + int i; + + for (i = 0; i < kMaxTryLocks; ++i) { + if (!pthread_mutex_trylock(mutex)) { + locked = 1; + break; + } + } + + if (!locked) + pthread_mutex_lock(mutex); +} +#endif // CONFIG_MULTITHREAD + +static INLINE void sync_read(VP9LfSync *const lf_sync, int r, int c) { +#if CONFIG_MULTITHREAD + const int nsync = lf_sync->sync_range; + + if (r && !(c & (nsync - 1))) { + pthread_mutex_t *const mutex = &lf_sync->mutex_[r - 1]; + mutex_lock(mutex); + + while (c > lf_sync->cur_sb_col[r - 1] - nsync) { + pthread_cond_wait(&lf_sync->cond_[r - 1], mutex); + } + pthread_mutex_unlock(mutex); + } +#else + (void)lf_sync; + (void)r; + (void)c; +#endif // CONFIG_MULTITHREAD +} + +static INLINE void sync_write(VP9LfSync *const lf_sync, int r, int c, + const int sb_cols) { +#if CONFIG_MULTITHREAD + const int nsync = lf_sync->sync_range; + int cur; + // Only signal when there are enough filtered SB for next row to run. + int sig = 1; + + if (c < sb_cols - 1) { + cur = c; + if (c % nsync) + sig = 0; + } else { + cur = sb_cols + nsync; + } + + if (sig) { + mutex_lock(&lf_sync->mutex_[r]); + + lf_sync->cur_sb_col[r] = cur; + + pthread_cond_signal(&lf_sync->cond_[r]); + pthread_mutex_unlock(&lf_sync->mutex_[r]); + } +#else + (void)lf_sync; + (void)r; + (void)c; + (void)sb_cols; +#endif // CONFIG_MULTITHREAD +} + +// Implement row loopfiltering for each thread. +static INLINE +void thread_loop_filter_rows(const YV12_BUFFER_CONFIG *const frame_buffer, + VP9_COMMON *const cm, + struct macroblockd_plane planes[MAX_MB_PLANE], + int start, int stop, int y_only, + VP9LfSync *const lf_sync) { + const int num_planes = y_only ? 1 : MAX_MB_PLANE; + const int sb_cols = mi_cols_aligned_to_sb(cm->mi_cols) >> MI_BLOCK_SIZE_LOG2; + int mi_row, mi_col; + enum lf_path path; + if (y_only) + path = LF_PATH_444; + else if (planes[1].subsampling_y == 1 && planes[1].subsampling_x == 1) + path = LF_PATH_420; + else if (planes[1].subsampling_y == 0 && planes[1].subsampling_x == 0) + path = LF_PATH_444; + else + path = LF_PATH_SLOW; + + for (mi_row = start; mi_row < stop; + mi_row += lf_sync->num_workers * MI_BLOCK_SIZE) { + MODE_INFO **const mi = cm->mi_grid_visible + mi_row * cm->mi_stride; + LOOP_FILTER_MASK *lfm = get_lfm(&cm->lf, mi_row, 0); + + for (mi_col = 0; mi_col < cm->mi_cols; mi_col += MI_BLOCK_SIZE, ++lfm) { + const int r = mi_row >> MI_BLOCK_SIZE_LOG2; + const int c = mi_col >> MI_BLOCK_SIZE_LOG2; + int plane; + + sync_read(lf_sync, r, c); + + vp9_setup_dst_planes(planes, frame_buffer, mi_row, mi_col); + + vp9_adjust_mask(cm, mi_row, mi_col, lfm); + + vp9_filter_block_plane_ss00(cm, &planes[0], mi_row, lfm); + for (plane = 1; plane < num_planes; ++plane) { + switch (path) { + case LF_PATH_420: + vp9_filter_block_plane_ss11(cm, &planes[plane], mi_row, lfm); + break; + case LF_PATH_444: + vp9_filter_block_plane_ss00(cm, &planes[plane], mi_row, lfm); + break; + case LF_PATH_SLOW: + vp9_filter_block_plane_non420(cm, &planes[plane], mi + mi_col, + mi_row, mi_col); + break; + } + } + + sync_write(lf_sync, r, c, sb_cols); + } + } +} + +// Row-based multi-threaded loopfilter hook +static int loop_filter_row_worker(VP9LfSync *const lf_sync, + LFWorkerData *const lf_data) { + thread_loop_filter_rows(lf_data->frame_buffer, lf_data->cm, lf_data->planes, + lf_data->start, lf_data->stop, lf_data->y_only, + lf_sync); + return 1; +} + +static void loop_filter_rows_mt(YV12_BUFFER_CONFIG *frame, + VP9_COMMON *cm, + struct macroblockd_plane planes[MAX_MB_PLANE], + int start, int stop, int y_only, + VPxWorker *workers, int nworkers, + VP9LfSync *lf_sync) { + const VPxWorkerInterface *const winterface = vpx_get_worker_interface(); + // Number of superblock rows and cols + const int sb_rows = mi_cols_aligned_to_sb(cm->mi_rows) >> MI_BLOCK_SIZE_LOG2; + // Decoder may allocate more threads than number of tiles based on user's + // input. + const int tile_cols = 1 << cm->log2_tile_cols; + const int num_workers = VPXMIN(nworkers, tile_cols); + int i; + + if (!lf_sync->sync_range || sb_rows != lf_sync->rows || + num_workers > lf_sync->num_workers) { + vp9_loop_filter_dealloc(lf_sync); + vp9_loop_filter_alloc(lf_sync, cm, sb_rows, cm->width, num_workers); + } + + // Initialize cur_sb_col to -1 for all SB rows. + memset(lf_sync->cur_sb_col, -1, sizeof(*lf_sync->cur_sb_col) * sb_rows); + + // Set up loopfilter thread data. + // The decoder is capping num_workers because it has been observed that using + // more threads on the loopfilter than there are cores will hurt performance + // on Android. This is because the system will only schedule the tile decode + // workers on cores equal to the number of tile columns. Then if the decoder + // tries to use more threads for the loopfilter, it will hurt performance + // because of contention. If the multithreading code changes in the future + // then the number of workers used by the loopfilter should be revisited. + for (i = 0; i < num_workers; ++i) { + VPxWorker *const worker = &workers[i]; + LFWorkerData *const lf_data = &lf_sync->lfdata[i]; + + worker->hook = (VPxWorkerHook)loop_filter_row_worker; + worker->data1 = lf_sync; + worker->data2 = lf_data; + + // Loopfilter data + vp9_loop_filter_data_reset(lf_data, frame, cm, planes); + lf_data->start = start + i * MI_BLOCK_SIZE; + lf_data->stop = stop; + lf_data->y_only = y_only; + + // Start loopfiltering + if (i == num_workers - 1) { + winterface->execute(worker); + } else { + winterface->launch(worker); + } + } + + // Wait till all rows are finished + for (i = 0; i < num_workers; ++i) { + winterface->sync(&workers[i]); + } +} + +void vp9_loop_filter_frame_mt(YV12_BUFFER_CONFIG *frame, + VP9_COMMON *cm, + struct macroblockd_plane planes[MAX_MB_PLANE], + int frame_filter_level, + int y_only, int partial_frame, + VPxWorker *workers, int num_workers, + VP9LfSync *lf_sync) { + int start_mi_row, end_mi_row, mi_rows_to_filter; + + if (!frame_filter_level) return; + + start_mi_row = 0; + mi_rows_to_filter = cm->mi_rows; + if (partial_frame && cm->mi_rows > 8) { + start_mi_row = cm->mi_rows >> 1; + start_mi_row &= 0xfffffff8; + mi_rows_to_filter = VPXMAX(cm->mi_rows / 8, 8); + } + end_mi_row = start_mi_row + mi_rows_to_filter; + vp9_loop_filter_frame_init(cm, frame_filter_level); + + loop_filter_rows_mt(frame, cm, planes, start_mi_row, end_mi_row, + y_only, workers, num_workers, lf_sync); +} + +// Set up nsync by width. +static INLINE int get_sync_range(int width) { + // nsync numbers are picked by testing. For example, for 4k + // video, using 4 gives best performance. + if (width < 640) + return 1; + else if (width <= 1280) + return 2; + else if (width <= 4096) + return 4; + else + return 8; +} + +// Allocate memory for lf row synchronization +void vp9_loop_filter_alloc(VP9LfSync *lf_sync, VP9_COMMON *cm, int rows, + int width, int num_workers) { + lf_sync->rows = rows; +#if CONFIG_MULTITHREAD + { + int i; + + CHECK_MEM_ERROR(cm, lf_sync->mutex_, + vpx_malloc(sizeof(*lf_sync->mutex_) * rows)); + if (lf_sync->mutex_) { + for (i = 0; i < rows; ++i) { + pthread_mutex_init(&lf_sync->mutex_[i], NULL); + } + } + + CHECK_MEM_ERROR(cm, lf_sync->cond_, + vpx_malloc(sizeof(*lf_sync->cond_) * rows)); + if (lf_sync->cond_) { + for (i = 0; i < rows; ++i) { + pthread_cond_init(&lf_sync->cond_[i], NULL); + } + } + } +#endif // CONFIG_MULTITHREAD + + CHECK_MEM_ERROR(cm, lf_sync->lfdata, + vpx_malloc(num_workers * sizeof(*lf_sync->lfdata))); + lf_sync->num_workers = num_workers; + + CHECK_MEM_ERROR(cm, lf_sync->cur_sb_col, + vpx_malloc(sizeof(*lf_sync->cur_sb_col) * rows)); + + // Set up nsync. + lf_sync->sync_range = get_sync_range(width); +} + +// Deallocate lf synchronization related mutex and data +void vp9_loop_filter_dealloc(VP9LfSync *lf_sync) { + if (lf_sync != NULL) { +#if CONFIG_MULTITHREAD + int i; + + if (lf_sync->mutex_ != NULL) { + for (i = 0; i < lf_sync->rows; ++i) { + pthread_mutex_destroy(&lf_sync->mutex_[i]); + } + vpx_free(lf_sync->mutex_); + } + if (lf_sync->cond_ != NULL) { + for (i = 0; i < lf_sync->rows; ++i) { + pthread_cond_destroy(&lf_sync->cond_[i]); + } + vpx_free(lf_sync->cond_); + } +#endif // CONFIG_MULTITHREAD + vpx_free(lf_sync->lfdata); + vpx_free(lf_sync->cur_sb_col); + // clear the structure as the source of this call may be a resize in which + // case this call will be followed by an _alloc() which may fail. + vp9_zero(*lf_sync); + } +} + +// Accumulate frame counts. +void vp9_accumulate_frame_counts(FRAME_COUNTS *accum, + const FRAME_COUNTS *counts, int is_dec) { + int i, j, k, l, m; + + for (i = 0; i < BLOCK_SIZE_GROUPS; i++) + for (j = 0; j < INTRA_MODES; j++) + accum->y_mode[i][j] += counts->y_mode[i][j]; + + for (i = 0; i < INTRA_MODES; i++) + for (j = 0; j < INTRA_MODES; j++) + accum->uv_mode[i][j] += counts->uv_mode[i][j]; + + for (i = 0; i < PARTITION_CONTEXTS; i++) + for (j = 0; j < PARTITION_TYPES; j++) + accum->partition[i][j] += counts->partition[i][j]; + + if (is_dec) { + int n; + for (i = 0; i < TX_SIZES; i++) + for (j = 0; j < PLANE_TYPES; j++) + for (k = 0; k < REF_TYPES; k++) + for (l = 0; l < COEF_BANDS; l++) + for (m = 0; m < COEFF_CONTEXTS; m++) { + accum->eob_branch[i][j][k][l][m] += + counts->eob_branch[i][j][k][l][m]; + for (n = 0; n < UNCONSTRAINED_NODES + 1; n++) + accum->coef[i][j][k][l][m][n] += + counts->coef[i][j][k][l][m][n]; + } + } else { + for (i = 0; i < TX_SIZES; i++) + for (j = 0; j < PLANE_TYPES; j++) + for (k = 0; k < REF_TYPES; k++) + for (l = 0; l < COEF_BANDS; l++) + for (m = 0; m < COEFF_CONTEXTS; m++) + accum->eob_branch[i][j][k][l][m] += + counts->eob_branch[i][j][k][l][m]; + // In the encoder, coef is only updated at frame + // level, so not need to accumulate it here. + // for (n = 0; n < UNCONSTRAINED_NODES + 1; n++) + // accum->coef[i][j][k][l][m][n] += + // counts->coef[i][j][k][l][m][n]; + } + + for (i = 0; i < SWITCHABLE_FILTER_CONTEXTS; i++) + for (j = 0; j < SWITCHABLE_FILTERS; j++) + accum->switchable_interp[i][j] += counts->switchable_interp[i][j]; + + for (i = 0; i < INTER_MODE_CONTEXTS; i++) + for (j = 0; j < INTER_MODES; j++) + accum->inter_mode[i][j] += counts->inter_mode[i][j]; + + for (i = 0; i < INTRA_INTER_CONTEXTS; i++) + for (j = 0; j < 2; j++) + accum->intra_inter[i][j] += counts->intra_inter[i][j]; + + for (i = 0; i < COMP_INTER_CONTEXTS; i++) + for (j = 0; j < 2; j++) + accum->comp_inter[i][j] += counts->comp_inter[i][j]; + + for (i = 0; i < REF_CONTEXTS; i++) + for (j = 0; j < 2; j++) + for (k = 0; k < 2; k++) + accum->single_ref[i][j][k] += counts->single_ref[i][j][k]; + + for (i = 0; i < REF_CONTEXTS; i++) + for (j = 0; j < 2; j++) + accum->comp_ref[i][j] += counts->comp_ref[i][j]; + + for (i = 0; i < TX_SIZE_CONTEXTS; i++) { + for (j = 0; j < TX_SIZES; j++) + accum->tx.p32x32[i][j] += counts->tx.p32x32[i][j]; + + for (j = 0; j < TX_SIZES - 1; j++) + accum->tx.p16x16[i][j] += counts->tx.p16x16[i][j]; + + for (j = 0; j < TX_SIZES - 2; j++) + accum->tx.p8x8[i][j] += counts->tx.p8x8[i][j]; + } + + for (i = 0; i < TX_SIZES; i++) + accum->tx.tx_totals[i] += counts->tx.tx_totals[i]; + + for (i = 0; i < SKIP_CONTEXTS; i++) + for (j = 0; j < 2; j++) + accum->skip[i][j] += counts->skip[i][j]; + + for (i = 0; i < MV_JOINTS; i++) + accum->mv.joints[i] += counts->mv.joints[i]; + + for (k = 0; k < 2; k++) { + nmv_component_counts *const comps = &accum->mv.comps[k]; + const nmv_component_counts *const comps_t = &counts->mv.comps[k]; + + for (i = 0; i < 2; i++) { + comps->sign[i] += comps_t->sign[i]; + comps->class0_hp[i] += comps_t->class0_hp[i]; + comps->hp[i] += comps_t->hp[i]; + } + + for (i = 0; i < MV_CLASSES; i++) + comps->classes[i] += comps_t->classes[i]; + + for (i = 0; i < CLASS0_SIZE; i++) { + comps->class0[i] += comps_t->class0[i]; + for (j = 0; j < MV_FP_SIZE; j++) + comps->class0_fp[i][j] += comps_t->class0_fp[i][j]; + } + + for (i = 0; i < MV_OFFSET_BITS; i++) + for (j = 0; j < 2; j++) + comps->bits[i][j] += comps_t->bits[i][j]; + + for (i = 0; i < MV_FP_SIZE; i++) + comps->fp[i] += comps_t->fp[i]; + } +} diff --git a/thirdparty/libvpx/vp9/common/vp9_thread_common.h b/thirdparty/libvpx/vp9/common/vp9_thread_common.h new file mode 100644 index 0000000000..b3b60c253f --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_thread_common.h @@ -0,0 +1,65 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_COMMON_VP9_THREAD_COMMON_H_ +#define VP9_COMMON_VP9_THREAD_COMMON_H_ +#include "./vpx_config.h" +#include "vp9/common/vp9_loopfilter.h" +#include "vpx_util/vpx_thread.h" + +#ifdef __cplusplus +extern "C" { +#endif + +struct VP9Common; +struct FRAME_COUNTS; + +// Loopfilter row synchronization +typedef struct VP9LfSyncData { +#if CONFIG_MULTITHREAD + pthread_mutex_t *mutex_; + pthread_cond_t *cond_; +#endif + // Allocate memory to store the loop-filtered superblock index in each row. + int *cur_sb_col; + // The optimal sync_range for different resolution and platform should be + // determined by testing. Currently, it is chosen to be a power-of-2 number. + int sync_range; + int rows; + + // Row-based parallel loopfilter data + LFWorkerData *lfdata; + int num_workers; +} VP9LfSync; + +// Allocate memory for loopfilter row synchronization. +void vp9_loop_filter_alloc(VP9LfSync *lf_sync, struct VP9Common *cm, int rows, + int width, int num_workers); + +// Deallocate loopfilter synchronization related mutex and data. +void vp9_loop_filter_dealloc(VP9LfSync *lf_sync); + +// Multi-threaded loopfilter that uses the tile threads. +void vp9_loop_filter_frame_mt(YV12_BUFFER_CONFIG *frame, + struct VP9Common *cm, + struct macroblockd_plane planes[MAX_MB_PLANE], + int frame_filter_level, + int y_only, int partial_frame, + VPxWorker *workers, int num_workers, + VP9LfSync *lf_sync); + +void vp9_accumulate_frame_counts(struct FRAME_COUNTS *accum, + const struct FRAME_COUNTS *counts, int is_dec); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_THREAD_COMMON_H_ diff --git a/thirdparty/libvpx/vp9/common/vp9_tile_common.c b/thirdparty/libvpx/vp9/common/vp9_tile_common.c new file mode 100644 index 0000000000..9fcb97c854 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_tile_common.c @@ -0,0 +1,59 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "vp9/common/vp9_tile_common.h" +#include "vp9/common/vp9_onyxc_int.h" +#include "vpx_dsp/vpx_dsp_common.h" + +#define MIN_TILE_WIDTH_B64 4 +#define MAX_TILE_WIDTH_B64 64 + +static int get_tile_offset(int idx, int mis, int log2) { + const int sb_cols = mi_cols_aligned_to_sb(mis) >> MI_BLOCK_SIZE_LOG2; + const int offset = ((idx * sb_cols) >> log2) << MI_BLOCK_SIZE_LOG2; + return VPXMIN(offset, mis); +} + +void vp9_tile_set_row(TileInfo *tile, const VP9_COMMON *cm, int row) { + tile->mi_row_start = get_tile_offset(row, cm->mi_rows, cm->log2_tile_rows); + tile->mi_row_end = get_tile_offset(row + 1, cm->mi_rows, cm->log2_tile_rows); +} + +void vp9_tile_set_col(TileInfo *tile, const VP9_COMMON *cm, int col) { + tile->mi_col_start = get_tile_offset(col, cm->mi_cols, cm->log2_tile_cols); + tile->mi_col_end = get_tile_offset(col + 1, cm->mi_cols, cm->log2_tile_cols); +} + +void vp9_tile_init(TileInfo *tile, const VP9_COMMON *cm, int row, int col) { + vp9_tile_set_row(tile, cm, row); + vp9_tile_set_col(tile, cm, col); +} + +static int get_min_log2_tile_cols(const int sb64_cols) { + int min_log2 = 0; + while ((MAX_TILE_WIDTH_B64 << min_log2) < sb64_cols) + ++min_log2; + return min_log2; +} + +static int get_max_log2_tile_cols(const int sb64_cols) { + int max_log2 = 1; + while ((sb64_cols >> max_log2) >= MIN_TILE_WIDTH_B64) + ++max_log2; + return max_log2 - 1; +} + +void vp9_get_tile_n_bits(int mi_cols, + int *min_log2_tile_cols, int *max_log2_tile_cols) { + const int sb64_cols = mi_cols_aligned_to_sb(mi_cols) >> MI_BLOCK_SIZE_LOG2; + *min_log2_tile_cols = get_min_log2_tile_cols(sb64_cols); + *max_log2_tile_cols = get_max_log2_tile_cols(sb64_cols); + assert(*min_log2_tile_cols <= *max_log2_tile_cols); +} diff --git a/thirdparty/libvpx/vp9/common/vp9_tile_common.h b/thirdparty/libvpx/vp9/common/vp9_tile_common.h new file mode 100644 index 0000000000..ae58805de1 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/vp9_tile_common.h @@ -0,0 +1,40 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_COMMON_VP9_TILE_COMMON_H_ +#define VP9_COMMON_VP9_TILE_COMMON_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +struct VP9Common; + +typedef struct TileInfo { + int mi_row_start, mi_row_end; + int mi_col_start, mi_col_end; +} TileInfo; + +// initializes 'tile->mi_(row|col)_(start|end)' for (row, col) based on +// 'cm->log2_tile_(rows|cols)' & 'cm->mi_(rows|cols)' +void vp9_tile_init(TileInfo *tile, const struct VP9Common *cm, + int row, int col); + +void vp9_tile_set_row(TileInfo *tile, const struct VP9Common *cm, int row); +void vp9_tile_set_col(TileInfo *tile, const struct VP9Common *cm, int col); + +void vp9_get_tile_n_bits(int mi_cols, + int *min_log2_tile_cols, int *max_log2_tile_cols); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_COMMON_VP9_TILE_COMMON_H_ diff --git a/thirdparty/libvpx/vp9/common/x86/vp9_idct_intrin_sse2.c b/thirdparty/libvpx/vp9/common/x86/vp9_idct_intrin_sse2.c new file mode 100644 index 0000000000..1c77b57ff1 --- /dev/null +++ b/thirdparty/libvpx/vp9/common/x86/vp9_idct_intrin_sse2.c @@ -0,0 +1,181 @@ +/* + * Copyright (c) 2012 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "./vp9_rtcd.h" +#include "vpx_dsp/x86/inv_txfm_sse2.h" +#include "vpx_dsp/x86/txfm_common_sse2.h" +#include "vpx_ports/mem.h" + +void vp9_iht4x4_16_add_sse2(const tran_low_t *input, uint8_t *dest, int stride, + int tx_type) { + __m128i in[2]; + const __m128i zero = _mm_setzero_si128(); + const __m128i eight = _mm_set1_epi16(8); + + in[0] = load_input_data(input); + in[1] = load_input_data(input + 8); + + switch (tx_type) { + case 0: // DCT_DCT + idct4_sse2(in); + idct4_sse2(in); + break; + case 1: // ADST_DCT + idct4_sse2(in); + iadst4_sse2(in); + break; + case 2: // DCT_ADST + iadst4_sse2(in); + idct4_sse2(in); + break; + case 3: // ADST_ADST + iadst4_sse2(in); + iadst4_sse2(in); + break; + default: + assert(0); + break; + } + + // Final round and shift + in[0] = _mm_add_epi16(in[0], eight); + in[1] = _mm_add_epi16(in[1], eight); + + in[0] = _mm_srai_epi16(in[0], 4); + in[1] = _mm_srai_epi16(in[1], 4); + + // Reconstruction and Store + { + __m128i d0 = _mm_cvtsi32_si128(*(const int *)(dest)); + __m128i d2 = _mm_cvtsi32_si128(*(const int *)(dest + stride * 2)); + d0 = _mm_unpacklo_epi32(d0, + _mm_cvtsi32_si128(*(const int *)(dest + stride))); + d2 = _mm_unpacklo_epi32( + d2, _mm_cvtsi32_si128(*(const int *)(dest + stride * 3))); + d0 = _mm_unpacklo_epi8(d0, zero); + d2 = _mm_unpacklo_epi8(d2, zero); + d0 = _mm_add_epi16(d0, in[0]); + d2 = _mm_add_epi16(d2, in[1]); + d0 = _mm_packus_epi16(d0, d2); + // store result[0] + *(int *)dest = _mm_cvtsi128_si32(d0); + // store result[1] + d0 = _mm_srli_si128(d0, 4); + *(int *)(dest + stride) = _mm_cvtsi128_si32(d0); + // store result[2] + d0 = _mm_srli_si128(d0, 4); + *(int *)(dest + stride * 2) = _mm_cvtsi128_si32(d0); + // store result[3] + d0 = _mm_srli_si128(d0, 4); + *(int *)(dest + stride * 3) = _mm_cvtsi128_si32(d0); + } +} + +void vp9_iht8x8_64_add_sse2(const tran_low_t *input, uint8_t *dest, int stride, + int tx_type) { + __m128i in[8]; + const __m128i zero = _mm_setzero_si128(); + const __m128i final_rounding = _mm_set1_epi16(1 << 4); + + // load input data + in[0] = load_input_data(input); + in[1] = load_input_data(input + 8 * 1); + in[2] = load_input_data(input + 8 * 2); + in[3] = load_input_data(input + 8 * 3); + in[4] = load_input_data(input + 8 * 4); + in[5] = load_input_data(input + 8 * 5); + in[6] = load_input_data(input + 8 * 6); + in[7] = load_input_data(input + 8 * 7); + + switch (tx_type) { + case 0: // DCT_DCT + idct8_sse2(in); + idct8_sse2(in); + break; + case 1: // ADST_DCT + idct8_sse2(in); + iadst8_sse2(in); + break; + case 2: // DCT_ADST + iadst8_sse2(in); + idct8_sse2(in); + break; + case 3: // ADST_ADST + iadst8_sse2(in); + iadst8_sse2(in); + break; + default: + assert(0); + break; + } + + // Final rounding and shift + in[0] = _mm_adds_epi16(in[0], final_rounding); + in[1] = _mm_adds_epi16(in[1], final_rounding); + in[2] = _mm_adds_epi16(in[2], final_rounding); + in[3] = _mm_adds_epi16(in[3], final_rounding); + in[4] = _mm_adds_epi16(in[4], final_rounding); + in[5] = _mm_adds_epi16(in[5], final_rounding); + in[6] = _mm_adds_epi16(in[6], final_rounding); + in[7] = _mm_adds_epi16(in[7], final_rounding); + + in[0] = _mm_srai_epi16(in[0], 5); + in[1] = _mm_srai_epi16(in[1], 5); + in[2] = _mm_srai_epi16(in[2], 5); + in[3] = _mm_srai_epi16(in[3], 5); + in[4] = _mm_srai_epi16(in[4], 5); + in[5] = _mm_srai_epi16(in[5], 5); + in[6] = _mm_srai_epi16(in[6], 5); + in[7] = _mm_srai_epi16(in[7], 5); + + RECON_AND_STORE(dest + 0 * stride, in[0]); + RECON_AND_STORE(dest + 1 * stride, in[1]); + RECON_AND_STORE(dest + 2 * stride, in[2]); + RECON_AND_STORE(dest + 3 * stride, in[3]); + RECON_AND_STORE(dest + 4 * stride, in[4]); + RECON_AND_STORE(dest + 5 * stride, in[5]); + RECON_AND_STORE(dest + 6 * stride, in[6]); + RECON_AND_STORE(dest + 7 * stride, in[7]); +} + +void vp9_iht16x16_256_add_sse2(const tran_low_t *input, uint8_t *dest, + int stride, int tx_type) { + __m128i in0[16], in1[16]; + + load_buffer_8x16(input, in0); + input += 8; + load_buffer_8x16(input, in1); + + switch (tx_type) { + case 0: // DCT_DCT + idct16_sse2(in0, in1); + idct16_sse2(in0, in1); + break; + case 1: // ADST_DCT + idct16_sse2(in0, in1); + iadst16_sse2(in0, in1); + break; + case 2: // DCT_ADST + iadst16_sse2(in0, in1); + idct16_sse2(in0, in1); + break; + case 3: // ADST_ADST + iadst16_sse2(in0, in1); + iadst16_sse2(in0, in1); + break; + default: + assert(0); + break; + } + + write_buffer_8x16(dest, in0, stride); + dest += 8; + write_buffer_8x16(dest, in1, stride); +} diff --git a/thirdparty/libvpx/vp9/decoder/vp9_decodeframe.c b/thirdparty/libvpx/vp9/decoder/vp9_decodeframe.c new file mode 100644 index 0000000000..d63912932c --- /dev/null +++ b/thirdparty/libvpx/vp9/decoder/vp9_decodeframe.c @@ -0,0 +1,2271 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <assert.h> +#include <stdlib.h> // qsort() + +#include "./vp9_rtcd.h" +#include "./vpx_dsp_rtcd.h" +#include "./vpx_scale_rtcd.h" + +#include "vpx_dsp/bitreader_buffer.h" +#include "vpx_dsp/bitreader.h" +#include "vpx_dsp/vpx_dsp_common.h" +#include "vpx_mem/vpx_mem.h" +#include "vpx_ports/mem.h" +#include "vpx_ports/mem_ops.h" +#include "vpx_scale/vpx_scale.h" +#include "vpx_util/vpx_thread.h" + +#include "vp9/common/vp9_alloccommon.h" +#include "vp9/common/vp9_common.h" +#include "vp9/common/vp9_entropy.h" +#include "vp9/common/vp9_entropymode.h" +#include "vp9/common/vp9_idct.h" +#include "vp9/common/vp9_thread_common.h" +#include "vp9/common/vp9_pred_common.h" +#include "vp9/common/vp9_quant_common.h" +#include "vp9/common/vp9_reconintra.h" +#include "vp9/common/vp9_reconinter.h" +#include "vp9/common/vp9_seg_common.h" +#include "vp9/common/vp9_tile_common.h" + +#include "vp9/decoder/vp9_decodeframe.h" +#include "vp9/decoder/vp9_detokenize.h" +#include "vp9/decoder/vp9_decodemv.h" +#include "vp9/decoder/vp9_decoder.h" +#include "vp9/decoder/vp9_dsubexp.h" + +#define MAX_VP9_HEADER_SIZE 80 + +static int is_compound_reference_allowed(const VP9_COMMON *cm) { + int i; + for (i = 1; i < REFS_PER_FRAME; ++i) + if (cm->ref_frame_sign_bias[i + 1] != cm->ref_frame_sign_bias[1]) + return 1; + + return 0; +} + +static void setup_compound_reference_mode(VP9_COMMON *cm) { + if (cm->ref_frame_sign_bias[LAST_FRAME] == + cm->ref_frame_sign_bias[GOLDEN_FRAME]) { + cm->comp_fixed_ref = ALTREF_FRAME; + cm->comp_var_ref[0] = LAST_FRAME; + cm->comp_var_ref[1] = GOLDEN_FRAME; + } else if (cm->ref_frame_sign_bias[LAST_FRAME] == + cm->ref_frame_sign_bias[ALTREF_FRAME]) { + cm->comp_fixed_ref = GOLDEN_FRAME; + cm->comp_var_ref[0] = LAST_FRAME; + cm->comp_var_ref[1] = ALTREF_FRAME; + } else { + cm->comp_fixed_ref = LAST_FRAME; + cm->comp_var_ref[0] = GOLDEN_FRAME; + cm->comp_var_ref[1] = ALTREF_FRAME; + } +} + +static int read_is_valid(const uint8_t *start, size_t len, const uint8_t *end) { + return len != 0 && len <= (size_t)(end - start); +} + +static int decode_unsigned_max(struct vpx_read_bit_buffer *rb, int max) { + const int data = vpx_rb_read_literal(rb, get_unsigned_bits(max)); + return data > max ? max : data; +} + +static TX_MODE read_tx_mode(vpx_reader *r) { + TX_MODE tx_mode = vpx_read_literal(r, 2); + if (tx_mode == ALLOW_32X32) + tx_mode += vpx_read_bit(r); + return tx_mode; +} + +static void read_tx_mode_probs(struct tx_probs *tx_probs, vpx_reader *r) { + int i, j; + + for (i = 0; i < TX_SIZE_CONTEXTS; ++i) + for (j = 0; j < TX_SIZES - 3; ++j) + vp9_diff_update_prob(r, &tx_probs->p8x8[i][j]); + + for (i = 0; i < TX_SIZE_CONTEXTS; ++i) + for (j = 0; j < TX_SIZES - 2; ++j) + vp9_diff_update_prob(r, &tx_probs->p16x16[i][j]); + + for (i = 0; i < TX_SIZE_CONTEXTS; ++i) + for (j = 0; j < TX_SIZES - 1; ++j) + vp9_diff_update_prob(r, &tx_probs->p32x32[i][j]); +} + +static void read_switchable_interp_probs(FRAME_CONTEXT *fc, vpx_reader *r) { + int i, j; + for (j = 0; j < SWITCHABLE_FILTER_CONTEXTS; ++j) + for (i = 0; i < SWITCHABLE_FILTERS - 1; ++i) + vp9_diff_update_prob(r, &fc->switchable_interp_prob[j][i]); +} + +static void read_inter_mode_probs(FRAME_CONTEXT *fc, vpx_reader *r) { + int i, j; + for (i = 0; i < INTER_MODE_CONTEXTS; ++i) + for (j = 0; j < INTER_MODES - 1; ++j) + vp9_diff_update_prob(r, &fc->inter_mode_probs[i][j]); +} + +static REFERENCE_MODE read_frame_reference_mode(const VP9_COMMON *cm, + vpx_reader *r) { + if (is_compound_reference_allowed(cm)) { + return vpx_read_bit(r) ? (vpx_read_bit(r) ? REFERENCE_MODE_SELECT + : COMPOUND_REFERENCE) + : SINGLE_REFERENCE; + } else { + return SINGLE_REFERENCE; + } +} + +static void read_frame_reference_mode_probs(VP9_COMMON *cm, vpx_reader *r) { + FRAME_CONTEXT *const fc = cm->fc; + int i; + + if (cm->reference_mode == REFERENCE_MODE_SELECT) + for (i = 0; i < COMP_INTER_CONTEXTS; ++i) + vp9_diff_update_prob(r, &fc->comp_inter_prob[i]); + + if (cm->reference_mode != COMPOUND_REFERENCE) + for (i = 0; i < REF_CONTEXTS; ++i) { + vp9_diff_update_prob(r, &fc->single_ref_prob[i][0]); + vp9_diff_update_prob(r, &fc->single_ref_prob[i][1]); + } + + if (cm->reference_mode != SINGLE_REFERENCE) + for (i = 0; i < REF_CONTEXTS; ++i) + vp9_diff_update_prob(r, &fc->comp_ref_prob[i]); +} + +static void update_mv_probs(vpx_prob *p, int n, vpx_reader *r) { + int i; + for (i = 0; i < n; ++i) + if (vpx_read(r, MV_UPDATE_PROB)) + p[i] = (vpx_read_literal(r, 7) << 1) | 1; +} + +static void read_mv_probs(nmv_context *ctx, int allow_hp, vpx_reader *r) { + int i, j; + + update_mv_probs(ctx->joints, MV_JOINTS - 1, r); + + for (i = 0; i < 2; ++i) { + nmv_component *const comp_ctx = &ctx->comps[i]; + update_mv_probs(&comp_ctx->sign, 1, r); + update_mv_probs(comp_ctx->classes, MV_CLASSES - 1, r); + update_mv_probs(comp_ctx->class0, CLASS0_SIZE - 1, r); + update_mv_probs(comp_ctx->bits, MV_OFFSET_BITS, r); + } + + for (i = 0; i < 2; ++i) { + nmv_component *const comp_ctx = &ctx->comps[i]; + for (j = 0; j < CLASS0_SIZE; ++j) + update_mv_probs(comp_ctx->class0_fp[j], MV_FP_SIZE - 1, r); + update_mv_probs(comp_ctx->fp, 3, r); + } + + if (allow_hp) { + for (i = 0; i < 2; ++i) { + nmv_component *const comp_ctx = &ctx->comps[i]; + update_mv_probs(&comp_ctx->class0_hp, 1, r); + update_mv_probs(&comp_ctx->hp, 1, r); + } + } +} + +static void inverse_transform_block_inter(MACROBLOCKD* xd, int plane, + const TX_SIZE tx_size, + uint8_t *dst, int stride, + int eob) { + struct macroblockd_plane *const pd = &xd->plane[plane]; + tran_low_t *const dqcoeff = pd->dqcoeff; + assert(eob > 0); +#if CONFIG_VP9_HIGHBITDEPTH + if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) { + if (xd->lossless) { + vp9_highbd_iwht4x4_add(dqcoeff, dst, stride, eob, xd->bd); + } else { + switch (tx_size) { + case TX_4X4: + vp9_highbd_idct4x4_add(dqcoeff, dst, stride, eob, xd->bd); + break; + case TX_8X8: + vp9_highbd_idct8x8_add(dqcoeff, dst, stride, eob, xd->bd); + break; + case TX_16X16: + vp9_highbd_idct16x16_add(dqcoeff, dst, stride, eob, xd->bd); + break; + case TX_32X32: + vp9_highbd_idct32x32_add(dqcoeff, dst, stride, eob, xd->bd); + break; + default: + assert(0 && "Invalid transform size"); + } + } + } else { + if (xd->lossless) { + vp9_iwht4x4_add(dqcoeff, dst, stride, eob); + } else { + switch (tx_size) { + case TX_4X4: + vp9_idct4x4_add(dqcoeff, dst, stride, eob); + break; + case TX_8X8: + vp9_idct8x8_add(dqcoeff, dst, stride, eob); + break; + case TX_16X16: + vp9_idct16x16_add(dqcoeff, dst, stride, eob); + break; + case TX_32X32: + vp9_idct32x32_add(dqcoeff, dst, stride, eob); + break; + default: + assert(0 && "Invalid transform size"); + return; + } + } + } +#else + if (xd->lossless) { + vp9_iwht4x4_add(dqcoeff, dst, stride, eob); + } else { + switch (tx_size) { + case TX_4X4: + vp9_idct4x4_add(dqcoeff, dst, stride, eob); + break; + case TX_8X8: + vp9_idct8x8_add(dqcoeff, dst, stride, eob); + break; + case TX_16X16: + vp9_idct16x16_add(dqcoeff, dst, stride, eob); + break; + case TX_32X32: + vp9_idct32x32_add(dqcoeff, dst, stride, eob); + break; + default: + assert(0 && "Invalid transform size"); + return; + } + } +#endif // CONFIG_VP9_HIGHBITDEPTH + + if (eob == 1) { + dqcoeff[0] = 0; + } else { + if (tx_size <= TX_16X16 && eob <= 10) + memset(dqcoeff, 0, 4 * (4 << tx_size) * sizeof(dqcoeff[0])); + else if (tx_size == TX_32X32 && eob <= 34) + memset(dqcoeff, 0, 256 * sizeof(dqcoeff[0])); + else + memset(dqcoeff, 0, (16 << (tx_size << 1)) * sizeof(dqcoeff[0])); + } +} + +static void inverse_transform_block_intra(MACROBLOCKD* xd, int plane, + const TX_TYPE tx_type, + const TX_SIZE tx_size, + uint8_t *dst, int stride, + int eob) { + struct macroblockd_plane *const pd = &xd->plane[plane]; + tran_low_t *const dqcoeff = pd->dqcoeff; + assert(eob > 0); +#if CONFIG_VP9_HIGHBITDEPTH + if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) { + if (xd->lossless) { + vp9_highbd_iwht4x4_add(dqcoeff, dst, stride, eob, xd->bd); + } else { + switch (tx_size) { + case TX_4X4: + vp9_highbd_iht4x4_add(tx_type, dqcoeff, dst, stride, eob, xd->bd); + break; + case TX_8X8: + vp9_highbd_iht8x8_add(tx_type, dqcoeff, dst, stride, eob, xd->bd); + break; + case TX_16X16: + vp9_highbd_iht16x16_add(tx_type, dqcoeff, dst, stride, eob, xd->bd); + break; + case TX_32X32: + vp9_highbd_idct32x32_add(dqcoeff, dst, stride, eob, xd->bd); + break; + default: + assert(0 && "Invalid transform size"); + } + } + } else { + if (xd->lossless) { + vp9_iwht4x4_add(dqcoeff, dst, stride, eob); + } else { + switch (tx_size) { + case TX_4X4: + vp9_iht4x4_add(tx_type, dqcoeff, dst, stride, eob); + break; + case TX_8X8: + vp9_iht8x8_add(tx_type, dqcoeff, dst, stride, eob); + break; + case TX_16X16: + vp9_iht16x16_add(tx_type, dqcoeff, dst, stride, eob); + break; + case TX_32X32: + vp9_idct32x32_add(dqcoeff, dst, stride, eob); + break; + default: + assert(0 && "Invalid transform size"); + return; + } + } + } +#else + if (xd->lossless) { + vp9_iwht4x4_add(dqcoeff, dst, stride, eob); + } else { + switch (tx_size) { + case TX_4X4: + vp9_iht4x4_add(tx_type, dqcoeff, dst, stride, eob); + break; + case TX_8X8: + vp9_iht8x8_add(tx_type, dqcoeff, dst, stride, eob); + break; + case TX_16X16: + vp9_iht16x16_add(tx_type, dqcoeff, dst, stride, eob); + break; + case TX_32X32: + vp9_idct32x32_add(dqcoeff, dst, stride, eob); + break; + default: + assert(0 && "Invalid transform size"); + return; + } + } +#endif // CONFIG_VP9_HIGHBITDEPTH + + if (eob == 1) { + dqcoeff[0] = 0; + } else { + if (tx_type == DCT_DCT && tx_size <= TX_16X16 && eob <= 10) + memset(dqcoeff, 0, 4 * (4 << tx_size) * sizeof(dqcoeff[0])); + else if (tx_size == TX_32X32 && eob <= 34) + memset(dqcoeff, 0, 256 * sizeof(dqcoeff[0])); + else + memset(dqcoeff, 0, (16 << (tx_size << 1)) * sizeof(dqcoeff[0])); + } +} + +static void predict_and_reconstruct_intra_block(MACROBLOCKD *const xd, + vpx_reader *r, + MODE_INFO *const mi, + int plane, + int row, int col, + TX_SIZE tx_size) { + struct macroblockd_plane *const pd = &xd->plane[plane]; + PREDICTION_MODE mode = (plane == 0) ? mi->mode : mi->uv_mode; + uint8_t *dst; + dst = &pd->dst.buf[4 * row * pd->dst.stride + 4 * col]; + + if (mi->sb_type < BLOCK_8X8) + if (plane == 0) + mode = xd->mi[0]->bmi[(row << 1) + col].as_mode; + + vp9_predict_intra_block(xd, pd->n4_wl, tx_size, mode, + dst, pd->dst.stride, dst, pd->dst.stride, + col, row, plane); + + if (!mi->skip) { + const TX_TYPE tx_type = (plane || xd->lossless) ? + DCT_DCT : intra_mode_to_tx_type_lookup[mode]; + const scan_order *sc = (plane || xd->lossless) ? + &vp9_default_scan_orders[tx_size] : &vp9_scan_orders[tx_size][tx_type]; + const int eob = vp9_decode_block_tokens(xd, plane, sc, col, row, tx_size, + r, mi->segment_id); + if (eob > 0) { + inverse_transform_block_intra(xd, plane, tx_type, tx_size, + dst, pd->dst.stride, eob); + } + } +} + +static int reconstruct_inter_block(MACROBLOCKD *const xd, vpx_reader *r, + MODE_INFO *const mi, int plane, + int row, int col, TX_SIZE tx_size) { + struct macroblockd_plane *const pd = &xd->plane[plane]; + const scan_order *sc = &vp9_default_scan_orders[tx_size]; + const int eob = vp9_decode_block_tokens(xd, plane, sc, col, row, tx_size, r, + mi->segment_id); + + if (eob > 0) { + inverse_transform_block_inter( + xd, plane, tx_size, &pd->dst.buf[4 * row * pd->dst.stride + 4 * col], + pd->dst.stride, eob); + } + return eob; +} + +static void build_mc_border(const uint8_t *src, int src_stride, + uint8_t *dst, int dst_stride, + int x, int y, int b_w, int b_h, int w, int h) { + // Get a pointer to the start of the real data for this row. + const uint8_t *ref_row = src - x - y * src_stride; + + if (y >= h) + ref_row += (h - 1) * src_stride; + else if (y > 0) + ref_row += y * src_stride; + + do { + int right = 0, copy; + int left = x < 0 ? -x : 0; + + if (left > b_w) + left = b_w; + + if (x + b_w > w) + right = x + b_w - w; + + if (right > b_w) + right = b_w; + + copy = b_w - left - right; + + if (left) + memset(dst, ref_row[0], left); + + if (copy) + memcpy(dst + left, ref_row + x + left, copy); + + if (right) + memset(dst + left + copy, ref_row[w - 1], right); + + dst += dst_stride; + ++y; + + if (y > 0 && y < h) + ref_row += src_stride; + } while (--b_h); +} + +#if CONFIG_VP9_HIGHBITDEPTH +static void high_build_mc_border(const uint8_t *src8, int src_stride, + uint16_t *dst, int dst_stride, + int x, int y, int b_w, int b_h, + int w, int h) { + // Get a pointer to the start of the real data for this row. + const uint16_t *src = CONVERT_TO_SHORTPTR(src8); + const uint16_t *ref_row = src - x - y * src_stride; + + if (y >= h) + ref_row += (h - 1) * src_stride; + else if (y > 0) + ref_row += y * src_stride; + + do { + int right = 0, copy; + int left = x < 0 ? -x : 0; + + if (left > b_w) + left = b_w; + + if (x + b_w > w) + right = x + b_w - w; + + if (right > b_w) + right = b_w; + + copy = b_w - left - right; + + if (left) + vpx_memset16(dst, ref_row[0], left); + + if (copy) + memcpy(dst + left, ref_row + x + left, copy * sizeof(uint16_t)); + + if (right) + vpx_memset16(dst + left + copy, ref_row[w - 1], right); + + dst += dst_stride; + ++y; + + if (y > 0 && y < h) + ref_row += src_stride; + } while (--b_h); +} +#endif // CONFIG_VP9_HIGHBITDEPTH + +#if CONFIG_VP9_HIGHBITDEPTH +static void extend_and_predict(const uint8_t *buf_ptr1, int pre_buf_stride, + int x0, int y0, int b_w, int b_h, + int frame_width, int frame_height, + int border_offset, + uint8_t *const dst, int dst_buf_stride, + int subpel_x, int subpel_y, + const InterpKernel *kernel, + const struct scale_factors *sf, + MACROBLOCKD *xd, + int w, int h, int ref, int xs, int ys) { + DECLARE_ALIGNED(16, uint16_t, mc_buf_high[80 * 2 * 80 * 2]); + const uint8_t *buf_ptr; + + if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) { + high_build_mc_border(buf_ptr1, pre_buf_stride, mc_buf_high, b_w, + x0, y0, b_w, b_h, frame_width, frame_height); + buf_ptr = CONVERT_TO_BYTEPTR(mc_buf_high) + border_offset; + } else { + build_mc_border(buf_ptr1, pre_buf_stride, (uint8_t *)mc_buf_high, b_w, + x0, y0, b_w, b_h, frame_width, frame_height); + buf_ptr = ((uint8_t *)mc_buf_high) + border_offset; + } + + if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) { + highbd_inter_predictor(buf_ptr, b_w, dst, dst_buf_stride, subpel_x, + subpel_y, sf, w, h, ref, kernel, xs, ys, xd->bd); + } else { + inter_predictor(buf_ptr, b_w, dst, dst_buf_stride, subpel_x, + subpel_y, sf, w, h, ref, kernel, xs, ys); + } +} +#else +static void extend_and_predict(const uint8_t *buf_ptr1, int pre_buf_stride, + int x0, int y0, int b_w, int b_h, + int frame_width, int frame_height, + int border_offset, + uint8_t *const dst, int dst_buf_stride, + int subpel_x, int subpel_y, + const InterpKernel *kernel, + const struct scale_factors *sf, + int w, int h, int ref, int xs, int ys) { + DECLARE_ALIGNED(16, uint8_t, mc_buf[80 * 2 * 80 * 2]); + const uint8_t *buf_ptr; + + build_mc_border(buf_ptr1, pre_buf_stride, mc_buf, b_w, + x0, y0, b_w, b_h, frame_width, frame_height); + buf_ptr = mc_buf + border_offset; + + inter_predictor(buf_ptr, b_w, dst, dst_buf_stride, subpel_x, + subpel_y, sf, w, h, ref, kernel, xs, ys); +} +#endif // CONFIG_VP9_HIGHBITDEPTH + +static void dec_build_inter_predictors(VPxWorker *const worker, MACROBLOCKD *xd, + int plane, int bw, int bh, int x, + int y, int w, int h, int mi_x, int mi_y, + const InterpKernel *kernel, + const struct scale_factors *sf, + struct buf_2d *pre_buf, + struct buf_2d *dst_buf, const MV* mv, + RefCntBuffer *ref_frame_buf, + int is_scaled, int ref) { + struct macroblockd_plane *const pd = &xd->plane[plane]; + uint8_t *const dst = dst_buf->buf + dst_buf->stride * y + x; + MV32 scaled_mv; + int xs, ys, x0, y0, x0_16, y0_16, frame_width, frame_height, + buf_stride, subpel_x, subpel_y; + uint8_t *ref_frame, *buf_ptr; + + // Get reference frame pointer, width and height. + if (plane == 0) { + frame_width = ref_frame_buf->buf.y_crop_width; + frame_height = ref_frame_buf->buf.y_crop_height; + ref_frame = ref_frame_buf->buf.y_buffer; + } else { + frame_width = ref_frame_buf->buf.uv_crop_width; + frame_height = ref_frame_buf->buf.uv_crop_height; + ref_frame = plane == 1 ? ref_frame_buf->buf.u_buffer + : ref_frame_buf->buf.v_buffer; + } + + if (is_scaled) { + const MV mv_q4 = clamp_mv_to_umv_border_sb(xd, mv, bw, bh, + pd->subsampling_x, + pd->subsampling_y); + // Co-ordinate of containing block to pixel precision. + int x_start = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x)); + int y_start = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y)); +#if CONFIG_BETTER_HW_COMPATIBILITY + assert(xd->mi[0]->sb_type != BLOCK_4X8 && + xd->mi[0]->sb_type != BLOCK_8X4); + assert(mv_q4.row == mv->row * (1 << (1 - pd->subsampling_y)) && + mv_q4.col == mv->col * (1 << (1 - pd->subsampling_x))); +#endif + // Co-ordinate of the block to 1/16th pixel precision. + x0_16 = (x_start + x) << SUBPEL_BITS; + y0_16 = (y_start + y) << SUBPEL_BITS; + + // Co-ordinate of current block in reference frame + // to 1/16th pixel precision. + x0_16 = sf->scale_value_x(x0_16, sf); + y0_16 = sf->scale_value_y(y0_16, sf); + + // Map the top left corner of the block into the reference frame. + x0 = sf->scale_value_x(x_start + x, sf); + y0 = sf->scale_value_y(y_start + y, sf); + + // Scale the MV and incorporate the sub-pixel offset of the block + // in the reference frame. + scaled_mv = vp9_scale_mv(&mv_q4, mi_x + x, mi_y + y, sf); + xs = sf->x_step_q4; + ys = sf->y_step_q4; + } else { + // Co-ordinate of containing block to pixel precision. + x0 = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x)) + x; + y0 = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y)) + y; + + // Co-ordinate of the block to 1/16th pixel precision. + x0_16 = x0 << SUBPEL_BITS; + y0_16 = y0 << SUBPEL_BITS; + + scaled_mv.row = mv->row * (1 << (1 - pd->subsampling_y)); + scaled_mv.col = mv->col * (1 << (1 - pd->subsampling_x)); + xs = ys = 16; + } + subpel_x = scaled_mv.col & SUBPEL_MASK; + subpel_y = scaled_mv.row & SUBPEL_MASK; + + // Calculate the top left corner of the best matching block in the + // reference frame. + x0 += scaled_mv.col >> SUBPEL_BITS; + y0 += scaled_mv.row >> SUBPEL_BITS; + x0_16 += scaled_mv.col; + y0_16 += scaled_mv.row; + + // Get reference block pointer. + buf_ptr = ref_frame + y0 * pre_buf->stride + x0; + buf_stride = pre_buf->stride; + + // Do border extension if there is motion or the + // width/height is not a multiple of 8 pixels. + if (is_scaled || scaled_mv.col || scaled_mv.row || + (frame_width & 0x7) || (frame_height & 0x7)) { + int y1 = ((y0_16 + (h - 1) * ys) >> SUBPEL_BITS) + 1; + + // Get reference block bottom right horizontal coordinate. + int x1 = ((x0_16 + (w - 1) * xs) >> SUBPEL_BITS) + 1; + int x_pad = 0, y_pad = 0; + + if (subpel_x || (sf->x_step_q4 != SUBPEL_SHIFTS)) { + x0 -= VP9_INTERP_EXTEND - 1; + x1 += VP9_INTERP_EXTEND; + x_pad = 1; + } + + if (subpel_y || (sf->y_step_q4 != SUBPEL_SHIFTS)) { + y0 -= VP9_INTERP_EXTEND - 1; + y1 += VP9_INTERP_EXTEND; + y_pad = 1; + } + + // Wait until reference block is ready. Pad 7 more pixels as last 7 + // pixels of each superblock row can be changed by next superblock row. + if (worker != NULL) + vp9_frameworker_wait(worker, ref_frame_buf, + VPXMAX(0, (y1 + 7)) << (plane == 0 ? 0 : 1)); + + // Skip border extension if block is inside the frame. + if (x0 < 0 || x0 > frame_width - 1 || x1 < 0 || x1 > frame_width - 1 || + y0 < 0 || y0 > frame_height - 1 || y1 < 0 || y1 > frame_height - 1) { + // Extend the border. + const uint8_t *const buf_ptr1 = ref_frame + y0 * buf_stride + x0; + const int b_w = x1 - x0 + 1; + const int b_h = y1 - y0 + 1; + const int border_offset = y_pad * 3 * b_w + x_pad * 3; + + extend_and_predict(buf_ptr1, buf_stride, x0, y0, b_w, b_h, + frame_width, frame_height, border_offset, + dst, dst_buf->stride, + subpel_x, subpel_y, + kernel, sf, +#if CONFIG_VP9_HIGHBITDEPTH + xd, +#endif + w, h, ref, xs, ys); + return; + } + } else { + // Wait until reference block is ready. Pad 7 more pixels as last 7 + // pixels of each superblock row can be changed by next superblock row. + if (worker != NULL) { + const int y1 = (y0_16 + (h - 1) * ys) >> SUBPEL_BITS; + vp9_frameworker_wait(worker, ref_frame_buf, + VPXMAX(0, (y1 + 7)) << (plane == 0 ? 0 : 1)); + } + } +#if CONFIG_VP9_HIGHBITDEPTH + if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) { + highbd_inter_predictor(buf_ptr, buf_stride, dst, dst_buf->stride, subpel_x, + subpel_y, sf, w, h, ref, kernel, xs, ys, xd->bd); + } else { + inter_predictor(buf_ptr, buf_stride, dst, dst_buf->stride, subpel_x, + subpel_y, sf, w, h, ref, kernel, xs, ys); + } +#else + inter_predictor(buf_ptr, buf_stride, dst, dst_buf->stride, subpel_x, + subpel_y, sf, w, h, ref, kernel, xs, ys); +#endif // CONFIG_VP9_HIGHBITDEPTH +} + +static void dec_build_inter_predictors_sb(VP9Decoder *const pbi, + MACROBLOCKD *xd, + int mi_row, int mi_col) { + int plane; + const int mi_x = mi_col * MI_SIZE; + const int mi_y = mi_row * MI_SIZE; + const MODE_INFO *mi = xd->mi[0]; + const InterpKernel *kernel = vp9_filter_kernels[mi->interp_filter]; + const BLOCK_SIZE sb_type = mi->sb_type; + const int is_compound = has_second_ref(mi); + int ref; + int is_scaled; + VPxWorker *const fwo = pbi->frame_parallel_decode ? + pbi->frame_worker_owner : NULL; + + for (ref = 0; ref < 1 + is_compound; ++ref) { + const MV_REFERENCE_FRAME frame = mi->ref_frame[ref]; + RefBuffer *ref_buf = &pbi->common.frame_refs[frame - LAST_FRAME]; + const struct scale_factors *const sf = &ref_buf->sf; + const int idx = ref_buf->idx; + BufferPool *const pool = pbi->common.buffer_pool; + RefCntBuffer *const ref_frame_buf = &pool->frame_bufs[idx]; + + if (!vp9_is_valid_scale(sf)) + vpx_internal_error(xd->error_info, VPX_CODEC_UNSUP_BITSTREAM, + "Reference frame has invalid dimensions"); + + is_scaled = vp9_is_scaled(sf); + vp9_setup_pre_planes(xd, ref, ref_buf->buf, mi_row, mi_col, + is_scaled ? sf : NULL); + xd->block_refs[ref] = ref_buf; + + if (sb_type < BLOCK_8X8) { + for (plane = 0; plane < MAX_MB_PLANE; ++plane) { + struct macroblockd_plane *const pd = &xd->plane[plane]; + struct buf_2d *const dst_buf = &pd->dst; + const int num_4x4_w = pd->n4_w; + const int num_4x4_h = pd->n4_h; + const int n4w_x4 = 4 * num_4x4_w; + const int n4h_x4 = 4 * num_4x4_h; + struct buf_2d *const pre_buf = &pd->pre[ref]; + int i = 0, x, y; + for (y = 0; y < num_4x4_h; ++y) { + for (x = 0; x < num_4x4_w; ++x) { + const MV mv = average_split_mvs(pd, mi, ref, i++); + dec_build_inter_predictors(fwo, xd, plane, n4w_x4, n4h_x4, + 4 * x, 4 * y, 4, 4, mi_x, mi_y, kernel, + sf, pre_buf, dst_buf, &mv, + ref_frame_buf, is_scaled, ref); + } + } + } + } else { + const MV mv = mi->mv[ref].as_mv; + for (plane = 0; plane < MAX_MB_PLANE; ++plane) { + struct macroblockd_plane *const pd = &xd->plane[plane]; + struct buf_2d *const dst_buf = &pd->dst; + const int num_4x4_w = pd->n4_w; + const int num_4x4_h = pd->n4_h; + const int n4w_x4 = 4 * num_4x4_w; + const int n4h_x4 = 4 * num_4x4_h; + struct buf_2d *const pre_buf = &pd->pre[ref]; + dec_build_inter_predictors(fwo, xd, plane, n4w_x4, n4h_x4, + 0, 0, n4w_x4, n4h_x4, mi_x, mi_y, kernel, + sf, pre_buf, dst_buf, &mv, + ref_frame_buf, is_scaled, ref); + } + } + } +} + +static INLINE TX_SIZE dec_get_uv_tx_size(const MODE_INFO *mi, + int n4_wl, int n4_hl) { + // get minimum log2 num4x4s dimension + const int x = VPXMIN(n4_wl, n4_hl); + return VPXMIN(mi->tx_size, x); +} + +static INLINE void dec_reset_skip_context(MACROBLOCKD *xd) { + int i; + for (i = 0; i < MAX_MB_PLANE; i++) { + struct macroblockd_plane *const pd = &xd->plane[i]; + memset(pd->above_context, 0, sizeof(ENTROPY_CONTEXT) * pd->n4_w); + memset(pd->left_context, 0, sizeof(ENTROPY_CONTEXT) * pd->n4_h); + } +} + +static void set_plane_n4(MACROBLOCKD *const xd, int bw, int bh, int bwl, + int bhl) { + int i; + for (i = 0; i < MAX_MB_PLANE; i++) { + xd->plane[i].n4_w = (bw << 1) >> xd->plane[i].subsampling_x; + xd->plane[i].n4_h = (bh << 1) >> xd->plane[i].subsampling_y; + xd->plane[i].n4_wl = bwl - xd->plane[i].subsampling_x; + xd->plane[i].n4_hl = bhl - xd->plane[i].subsampling_y; + } +} + +static MODE_INFO *set_offsets(VP9_COMMON *const cm, MACROBLOCKD *const xd, + BLOCK_SIZE bsize, int mi_row, int mi_col, + int bw, int bh, int x_mis, int y_mis, + int bwl, int bhl) { + const int offset = mi_row * cm->mi_stride + mi_col; + int x, y; + const TileInfo *const tile = &xd->tile; + + xd->mi = cm->mi_grid_visible + offset; + xd->mi[0] = &cm->mi[offset]; + // TODO(slavarnway): Generate sb_type based on bwl and bhl, instead of + // passing bsize from decode_partition(). + xd->mi[0]->sb_type = bsize; + for (y = 0; y < y_mis; ++y) + for (x = !y; x < x_mis; ++x) { + xd->mi[y * cm->mi_stride + x] = xd->mi[0]; + } + + set_plane_n4(xd, bw, bh, bwl, bhl); + + set_skip_context(xd, mi_row, mi_col); + + // Distance of Mb to the various image edges. These are specified to 8th pel + // as they are always compared to values that are in 1/8th pel units + set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols); + + vp9_setup_dst_planes(xd->plane, get_frame_new_buffer(cm), mi_row, mi_col); + return xd->mi[0]; +} + +static void decode_block(VP9Decoder *const pbi, MACROBLOCKD *const xd, + int mi_row, int mi_col, + vpx_reader *r, BLOCK_SIZE bsize, + int bwl, int bhl) { + VP9_COMMON *const cm = &pbi->common; + const int less8x8 = bsize < BLOCK_8X8; + const int bw = 1 << (bwl - 1); + const int bh = 1 << (bhl - 1); + const int x_mis = VPXMIN(bw, cm->mi_cols - mi_col); + const int y_mis = VPXMIN(bh, cm->mi_rows - mi_row); + + MODE_INFO *mi = set_offsets(cm, xd, bsize, mi_row, mi_col, + bw, bh, x_mis, y_mis, bwl, bhl); + + if (bsize >= BLOCK_8X8 && (cm->subsampling_x || cm->subsampling_y)) { + const BLOCK_SIZE uv_subsize = + ss_size_lookup[bsize][cm->subsampling_x][cm->subsampling_y]; + if (uv_subsize == BLOCK_INVALID) + vpx_internal_error(xd->error_info, + VPX_CODEC_CORRUPT_FRAME, "Invalid block size."); + } + + vp9_read_mode_info(pbi, xd, mi_row, mi_col, r, x_mis, y_mis); + + if (mi->skip) { + dec_reset_skip_context(xd); + } + + if (!is_inter_block(mi)) { + int plane; + for (plane = 0; plane < MAX_MB_PLANE; ++plane) { + const struct macroblockd_plane *const pd = &xd->plane[plane]; + const TX_SIZE tx_size = + plane ? dec_get_uv_tx_size(mi, pd->n4_wl, pd->n4_hl) + : mi->tx_size; + const int num_4x4_w = pd->n4_w; + const int num_4x4_h = pd->n4_h; + const int step = (1 << tx_size); + int row, col; + const int max_blocks_wide = num_4x4_w + (xd->mb_to_right_edge >= 0 ? + 0 : xd->mb_to_right_edge >> (5 + pd->subsampling_x)); + const int max_blocks_high = num_4x4_h + (xd->mb_to_bottom_edge >= 0 ? + 0 : xd->mb_to_bottom_edge >> (5 + pd->subsampling_y)); + + xd->max_blocks_wide = xd->mb_to_right_edge >= 0 ? 0 : max_blocks_wide; + xd->max_blocks_high = xd->mb_to_bottom_edge >= 0 ? 0 : max_blocks_high; + + for (row = 0; row < max_blocks_high; row += step) + for (col = 0; col < max_blocks_wide; col += step) + predict_and_reconstruct_intra_block(xd, r, mi, plane, + row, col, tx_size); + } + } else { + // Prediction + dec_build_inter_predictors_sb(pbi, xd, mi_row, mi_col); + + // Reconstruction + if (!mi->skip) { + int eobtotal = 0; + int plane; + + for (plane = 0; plane < MAX_MB_PLANE; ++plane) { + const struct macroblockd_plane *const pd = &xd->plane[plane]; + const TX_SIZE tx_size = + plane ? dec_get_uv_tx_size(mi, pd->n4_wl, pd->n4_hl) + : mi->tx_size; + const int num_4x4_w = pd->n4_w; + const int num_4x4_h = pd->n4_h; + const int step = (1 << tx_size); + int row, col; + const int max_blocks_wide = num_4x4_w + (xd->mb_to_right_edge >= 0 ? + 0 : xd->mb_to_right_edge >> (5 + pd->subsampling_x)); + const int max_blocks_high = num_4x4_h + (xd->mb_to_bottom_edge >= 0 ? + 0 : xd->mb_to_bottom_edge >> (5 + pd->subsampling_y)); + + xd->max_blocks_wide = xd->mb_to_right_edge >= 0 ? 0 : max_blocks_wide; + xd->max_blocks_high = xd->mb_to_bottom_edge >= 0 ? 0 : max_blocks_high; + + for (row = 0; row < max_blocks_high; row += step) + for (col = 0; col < max_blocks_wide; col += step) + eobtotal += reconstruct_inter_block(xd, r, mi, plane, row, col, + tx_size); + } + + if (!less8x8 && eobtotal == 0) + mi->skip = 1; // skip loopfilter + } + } + + xd->corrupted |= vpx_reader_has_error(r); + + if (cm->lf.filter_level) { + vp9_build_mask(cm, mi, mi_row, mi_col, bw, bh); + } +} + +static INLINE int dec_partition_plane_context(const MACROBLOCKD *xd, + int mi_row, int mi_col, + int bsl) { + const PARTITION_CONTEXT *above_ctx = xd->above_seg_context + mi_col; + const PARTITION_CONTEXT *left_ctx = xd->left_seg_context + (mi_row & MI_MASK); + int above = (*above_ctx >> bsl) & 1 , left = (*left_ctx >> bsl) & 1; + +// assert(bsl >= 0); + + return (left * 2 + above) + bsl * PARTITION_PLOFFSET; +} + +static INLINE void dec_update_partition_context(MACROBLOCKD *xd, + int mi_row, int mi_col, + BLOCK_SIZE subsize, + int bw) { + PARTITION_CONTEXT *const above_ctx = xd->above_seg_context + mi_col; + PARTITION_CONTEXT *const left_ctx = xd->left_seg_context + (mi_row & MI_MASK); + + // update the partition context at the end notes. set partition bits + // of block sizes larger than the current one to be one, and partition + // bits of smaller block sizes to be zero. + memset(above_ctx, partition_context_lookup[subsize].above, bw); + memset(left_ctx, partition_context_lookup[subsize].left, bw); +} + +static PARTITION_TYPE read_partition(MACROBLOCKD *xd, int mi_row, int mi_col, + vpx_reader *r, + int has_rows, int has_cols, int bsl) { + const int ctx = dec_partition_plane_context(xd, mi_row, mi_col, bsl); + const vpx_prob *const probs = get_partition_probs(xd, ctx); + FRAME_COUNTS *counts = xd->counts; + PARTITION_TYPE p; + + if (has_rows && has_cols) + p = (PARTITION_TYPE)vpx_read_tree(r, vp9_partition_tree, probs); + else if (!has_rows && has_cols) + p = vpx_read(r, probs[1]) ? PARTITION_SPLIT : PARTITION_HORZ; + else if (has_rows && !has_cols) + p = vpx_read(r, probs[2]) ? PARTITION_SPLIT : PARTITION_VERT; + else + p = PARTITION_SPLIT; + + if (counts) + ++counts->partition[ctx][p]; + + return p; +} + +// TODO(slavarnway): eliminate bsize and subsize in future commits +static void decode_partition(VP9Decoder *const pbi, MACROBLOCKD *const xd, + int mi_row, int mi_col, + vpx_reader* r, BLOCK_SIZE bsize, int n4x4_l2) { + VP9_COMMON *const cm = &pbi->common; + const int n8x8_l2 = n4x4_l2 - 1; + const int num_8x8_wh = 1 << n8x8_l2; + const int hbs = num_8x8_wh >> 1; + PARTITION_TYPE partition; + BLOCK_SIZE subsize; + const int has_rows = (mi_row + hbs) < cm->mi_rows; + const int has_cols = (mi_col + hbs) < cm->mi_cols; + + if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols) + return; + + partition = read_partition(xd, mi_row, mi_col, r, has_rows, has_cols, + n8x8_l2); + subsize = subsize_lookup[partition][bsize]; // get_subsize(bsize, partition); + if (!hbs) { + // calculate bmode block dimensions (log 2) + xd->bmode_blocks_wl = 1 >> !!(partition & PARTITION_VERT); + xd->bmode_blocks_hl = 1 >> !!(partition & PARTITION_HORZ); + decode_block(pbi, xd, mi_row, mi_col, r, subsize, 1, 1); + } else { + switch (partition) { + case PARTITION_NONE: + decode_block(pbi, xd, mi_row, mi_col, r, subsize, n4x4_l2, n4x4_l2); + break; + case PARTITION_HORZ: + decode_block(pbi, xd, mi_row, mi_col, r, subsize, n4x4_l2, n8x8_l2); + if (has_rows) + decode_block(pbi, xd, mi_row + hbs, mi_col, r, subsize, n4x4_l2, + n8x8_l2); + break; + case PARTITION_VERT: + decode_block(pbi, xd, mi_row, mi_col, r, subsize, n8x8_l2, n4x4_l2); + if (has_cols) + decode_block(pbi, xd, mi_row, mi_col + hbs, r, subsize, n8x8_l2, + n4x4_l2); + break; + case PARTITION_SPLIT: + decode_partition(pbi, xd, mi_row, mi_col, r, subsize, n8x8_l2); + decode_partition(pbi, xd, mi_row, mi_col + hbs, r, subsize, n8x8_l2); + decode_partition(pbi, xd, mi_row + hbs, mi_col, r, subsize, n8x8_l2); + decode_partition(pbi, xd, mi_row + hbs, mi_col + hbs, r, subsize, + n8x8_l2); + break; + default: + assert(0 && "Invalid partition type"); + } + } + + // update partition context + if (bsize >= BLOCK_8X8 && + (bsize == BLOCK_8X8 || partition != PARTITION_SPLIT)) + dec_update_partition_context(xd, mi_row, mi_col, subsize, num_8x8_wh); +} + +static void setup_token_decoder(const uint8_t *data, + const uint8_t *data_end, + size_t read_size, + struct vpx_internal_error_info *error_info, + vpx_reader *r, + vpx_decrypt_cb decrypt_cb, + void *decrypt_state) { + // Validate the calculated partition length. If the buffer + // described by the partition can't be fully read, then restrict + // it to the portion that can be (for EC mode) or throw an error. + if (!read_is_valid(data, read_size, data_end)) + vpx_internal_error(error_info, VPX_CODEC_CORRUPT_FRAME, + "Truncated packet or corrupt tile length"); + + if (vpx_reader_init(r, data, read_size, decrypt_cb, decrypt_state)) + vpx_internal_error(error_info, VPX_CODEC_MEM_ERROR, + "Failed to allocate bool decoder %d", 1); +} + +static void read_coef_probs_common(vp9_coeff_probs_model *coef_probs, + vpx_reader *r) { + int i, j, k, l, m; + + if (vpx_read_bit(r)) + for (i = 0; i < PLANE_TYPES; ++i) + for (j = 0; j < REF_TYPES; ++j) + for (k = 0; k < COEF_BANDS; ++k) + for (l = 0; l < BAND_COEFF_CONTEXTS(k); ++l) + for (m = 0; m < UNCONSTRAINED_NODES; ++m) + vp9_diff_update_prob(r, &coef_probs[i][j][k][l][m]); +} + +static void read_coef_probs(FRAME_CONTEXT *fc, TX_MODE tx_mode, + vpx_reader *r) { + const TX_SIZE max_tx_size = tx_mode_to_biggest_tx_size[tx_mode]; + TX_SIZE tx_size; + for (tx_size = TX_4X4; tx_size <= max_tx_size; ++tx_size) + read_coef_probs_common(fc->coef_probs[tx_size], r); +} + +static void setup_segmentation(struct segmentation *seg, + struct vpx_read_bit_buffer *rb) { + int i, j; + + seg->update_map = 0; + seg->update_data = 0; + + seg->enabled = vpx_rb_read_bit(rb); + if (!seg->enabled) + return; + + // Segmentation map update + seg->update_map = vpx_rb_read_bit(rb); + if (seg->update_map) { + for (i = 0; i < SEG_TREE_PROBS; i++) + seg->tree_probs[i] = vpx_rb_read_bit(rb) ? vpx_rb_read_literal(rb, 8) + : MAX_PROB; + + seg->temporal_update = vpx_rb_read_bit(rb); + if (seg->temporal_update) { + for (i = 0; i < PREDICTION_PROBS; i++) + seg->pred_probs[i] = vpx_rb_read_bit(rb) ? vpx_rb_read_literal(rb, 8) + : MAX_PROB; + } else { + for (i = 0; i < PREDICTION_PROBS; i++) + seg->pred_probs[i] = MAX_PROB; + } + } + + // Segmentation data update + seg->update_data = vpx_rb_read_bit(rb); + if (seg->update_data) { + seg->abs_delta = vpx_rb_read_bit(rb); + + vp9_clearall_segfeatures(seg); + + for (i = 0; i < MAX_SEGMENTS; i++) { + for (j = 0; j < SEG_LVL_MAX; j++) { + int data = 0; + const int feature_enabled = vpx_rb_read_bit(rb); + if (feature_enabled) { + vp9_enable_segfeature(seg, i, j); + data = decode_unsigned_max(rb, vp9_seg_feature_data_max(j)); + if (vp9_is_segfeature_signed(j)) + data = vpx_rb_read_bit(rb) ? -data : data; + } + vp9_set_segdata(seg, i, j, data); + } + } + } +} + +static void setup_loopfilter(struct loopfilter *lf, + struct vpx_read_bit_buffer *rb) { + lf->filter_level = vpx_rb_read_literal(rb, 6); + lf->sharpness_level = vpx_rb_read_literal(rb, 3); + + // Read in loop filter deltas applied at the MB level based on mode or ref + // frame. + lf->mode_ref_delta_update = 0; + + lf->mode_ref_delta_enabled = vpx_rb_read_bit(rb); + if (lf->mode_ref_delta_enabled) { + lf->mode_ref_delta_update = vpx_rb_read_bit(rb); + if (lf->mode_ref_delta_update) { + int i; + + for (i = 0; i < MAX_REF_LF_DELTAS; i++) + if (vpx_rb_read_bit(rb)) + lf->ref_deltas[i] = vpx_rb_read_signed_literal(rb, 6); + + for (i = 0; i < MAX_MODE_LF_DELTAS; i++) + if (vpx_rb_read_bit(rb)) + lf->mode_deltas[i] = vpx_rb_read_signed_literal(rb, 6); + } + } +} + +static INLINE int read_delta_q(struct vpx_read_bit_buffer *rb) { + return vpx_rb_read_bit(rb) ? vpx_rb_read_signed_literal(rb, 4) : 0; +} + +static void setup_quantization(VP9_COMMON *const cm, MACROBLOCKD *const xd, + struct vpx_read_bit_buffer *rb) { + cm->base_qindex = vpx_rb_read_literal(rb, QINDEX_BITS); + cm->y_dc_delta_q = read_delta_q(rb); + cm->uv_dc_delta_q = read_delta_q(rb); + cm->uv_ac_delta_q = read_delta_q(rb); + cm->dequant_bit_depth = cm->bit_depth; + xd->lossless = cm->base_qindex == 0 && + cm->y_dc_delta_q == 0 && + cm->uv_dc_delta_q == 0 && + cm->uv_ac_delta_q == 0; + +#if CONFIG_VP9_HIGHBITDEPTH + xd->bd = (int)cm->bit_depth; +#endif +} + +static void setup_segmentation_dequant(VP9_COMMON *const cm) { + // Build y/uv dequant values based on segmentation. + if (cm->seg.enabled) { + int i; + for (i = 0; i < MAX_SEGMENTS; ++i) { + const int qindex = vp9_get_qindex(&cm->seg, i, cm->base_qindex); + cm->y_dequant[i][0] = vp9_dc_quant(qindex, cm->y_dc_delta_q, + cm->bit_depth); + cm->y_dequant[i][1] = vp9_ac_quant(qindex, 0, cm->bit_depth); + cm->uv_dequant[i][0] = vp9_dc_quant(qindex, cm->uv_dc_delta_q, + cm->bit_depth); + cm->uv_dequant[i][1] = vp9_ac_quant(qindex, cm->uv_ac_delta_q, + cm->bit_depth); + } + } else { + const int qindex = cm->base_qindex; + // When segmentation is disabled, only the first value is used. The + // remaining are don't cares. + cm->y_dequant[0][0] = vp9_dc_quant(qindex, cm->y_dc_delta_q, cm->bit_depth); + cm->y_dequant[0][1] = vp9_ac_quant(qindex, 0, cm->bit_depth); + cm->uv_dequant[0][0] = vp9_dc_quant(qindex, cm->uv_dc_delta_q, + cm->bit_depth); + cm->uv_dequant[0][1] = vp9_ac_quant(qindex, cm->uv_ac_delta_q, + cm->bit_depth); + } +} + +static INTERP_FILTER read_interp_filter(struct vpx_read_bit_buffer *rb) { + const INTERP_FILTER literal_to_filter[] = { EIGHTTAP_SMOOTH, + EIGHTTAP, + EIGHTTAP_SHARP, + BILINEAR }; + return vpx_rb_read_bit(rb) ? SWITCHABLE + : literal_to_filter[vpx_rb_read_literal(rb, 2)]; +} + +static void setup_render_size(VP9_COMMON *cm, struct vpx_read_bit_buffer *rb) { + cm->render_width = cm->width; + cm->render_height = cm->height; + if (vpx_rb_read_bit(rb)) + vp9_read_frame_size(rb, &cm->render_width, &cm->render_height); +} + +static void resize_mv_buffer(VP9_COMMON *cm) { + vpx_free(cm->cur_frame->mvs); + cm->cur_frame->mi_rows = cm->mi_rows; + cm->cur_frame->mi_cols = cm->mi_cols; + CHECK_MEM_ERROR(cm, cm->cur_frame->mvs, + (MV_REF *)vpx_calloc(cm->mi_rows * cm->mi_cols, + sizeof(*cm->cur_frame->mvs))); +} + +static void resize_context_buffers(VP9_COMMON *cm, int width, int height) { +#if CONFIG_SIZE_LIMIT + if (width > DECODE_WIDTH_LIMIT || height > DECODE_HEIGHT_LIMIT) + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Dimensions of %dx%d beyond allowed size of %dx%d.", + width, height, DECODE_WIDTH_LIMIT, DECODE_HEIGHT_LIMIT); +#endif + if (cm->width != width || cm->height != height) { + const int new_mi_rows = + ALIGN_POWER_OF_TWO(height, MI_SIZE_LOG2) >> MI_SIZE_LOG2; + const int new_mi_cols = + ALIGN_POWER_OF_TWO(width, MI_SIZE_LOG2) >> MI_SIZE_LOG2; + + // Allocations in vp9_alloc_context_buffers() depend on individual + // dimensions as well as the overall size. + if (new_mi_cols > cm->mi_cols || new_mi_rows > cm->mi_rows) { + if (vp9_alloc_context_buffers(cm, width, height)) + vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR, + "Failed to allocate context buffers"); + } else { + vp9_set_mb_mi(cm, width, height); + } + vp9_init_context_buffers(cm); + cm->width = width; + cm->height = height; + } + if (cm->cur_frame->mvs == NULL || cm->mi_rows > cm->cur_frame->mi_rows || + cm->mi_cols > cm->cur_frame->mi_cols) { + resize_mv_buffer(cm); + } +} + +static void setup_frame_size(VP9_COMMON *cm, struct vpx_read_bit_buffer *rb) { + int width, height; + BufferPool *const pool = cm->buffer_pool; + vp9_read_frame_size(rb, &width, &height); + resize_context_buffers(cm, width, height); + setup_render_size(cm, rb); + + lock_buffer_pool(pool); + if (vpx_realloc_frame_buffer( + get_frame_new_buffer(cm), cm->width, cm->height, + cm->subsampling_x, cm->subsampling_y, +#if CONFIG_VP9_HIGHBITDEPTH + cm->use_highbitdepth, +#endif + VP9_DEC_BORDER_IN_PIXELS, + cm->byte_alignment, + &pool->frame_bufs[cm->new_fb_idx].raw_frame_buffer, pool->get_fb_cb, + pool->cb_priv)) { + unlock_buffer_pool(pool); + vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR, + "Failed to allocate frame buffer"); + } + unlock_buffer_pool(pool); + + pool->frame_bufs[cm->new_fb_idx].buf.subsampling_x = cm->subsampling_x; + pool->frame_bufs[cm->new_fb_idx].buf.subsampling_y = cm->subsampling_y; + pool->frame_bufs[cm->new_fb_idx].buf.bit_depth = (unsigned int)cm->bit_depth; + pool->frame_bufs[cm->new_fb_idx].buf.color_space = cm->color_space; + pool->frame_bufs[cm->new_fb_idx].buf.color_range = cm->color_range; + pool->frame_bufs[cm->new_fb_idx].buf.render_width = cm->render_width; + pool->frame_bufs[cm->new_fb_idx].buf.render_height = cm->render_height; +} + +static INLINE int valid_ref_frame_img_fmt(vpx_bit_depth_t ref_bit_depth, + int ref_xss, int ref_yss, + vpx_bit_depth_t this_bit_depth, + int this_xss, int this_yss) { + return ref_bit_depth == this_bit_depth && ref_xss == this_xss && + ref_yss == this_yss; +} + +static void setup_frame_size_with_refs(VP9_COMMON *cm, + struct vpx_read_bit_buffer *rb) { + int width, height; + int found = 0, i; + int has_valid_ref_frame = 0; + BufferPool *const pool = cm->buffer_pool; + for (i = 0; i < REFS_PER_FRAME; ++i) { + if (vpx_rb_read_bit(rb)) { + if (cm->frame_refs[i].idx != INVALID_IDX) { + YV12_BUFFER_CONFIG *const buf = cm->frame_refs[i].buf; + width = buf->y_crop_width; + height = buf->y_crop_height; + found = 1; + break; + } else { + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Failed to decode frame size"); + } + } + } + + if (!found) + vp9_read_frame_size(rb, &width, &height); + + if (width <= 0 || height <= 0) + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Invalid frame size"); + + // Check to make sure at least one of frames that this frame references + // has valid dimensions. + for (i = 0; i < REFS_PER_FRAME; ++i) { + RefBuffer *const ref_frame = &cm->frame_refs[i]; + has_valid_ref_frame |= (ref_frame->idx != INVALID_IDX && + valid_ref_frame_size(ref_frame->buf->y_crop_width, + ref_frame->buf->y_crop_height, + width, height)); + } + if (!has_valid_ref_frame) + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Referenced frame has invalid size"); + for (i = 0; i < REFS_PER_FRAME; ++i) { + RefBuffer *const ref_frame = &cm->frame_refs[i]; + if (ref_frame->idx == INVALID_IDX || + !valid_ref_frame_img_fmt(ref_frame->buf->bit_depth, + ref_frame->buf->subsampling_x, + ref_frame->buf->subsampling_y, + cm->bit_depth, + cm->subsampling_x, + cm->subsampling_y)) + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Referenced frame has incompatible color format"); + } + + resize_context_buffers(cm, width, height); + setup_render_size(cm, rb); + + lock_buffer_pool(pool); + if (vpx_realloc_frame_buffer( + get_frame_new_buffer(cm), cm->width, cm->height, + cm->subsampling_x, cm->subsampling_y, +#if CONFIG_VP9_HIGHBITDEPTH + cm->use_highbitdepth, +#endif + VP9_DEC_BORDER_IN_PIXELS, + cm->byte_alignment, + &pool->frame_bufs[cm->new_fb_idx].raw_frame_buffer, pool->get_fb_cb, + pool->cb_priv)) { + unlock_buffer_pool(pool); + vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR, + "Failed to allocate frame buffer"); + } + unlock_buffer_pool(pool); + + pool->frame_bufs[cm->new_fb_idx].buf.subsampling_x = cm->subsampling_x; + pool->frame_bufs[cm->new_fb_idx].buf.subsampling_y = cm->subsampling_y; + pool->frame_bufs[cm->new_fb_idx].buf.bit_depth = (unsigned int)cm->bit_depth; + pool->frame_bufs[cm->new_fb_idx].buf.color_space = cm->color_space; + pool->frame_bufs[cm->new_fb_idx].buf.color_range = cm->color_range; + pool->frame_bufs[cm->new_fb_idx].buf.render_width = cm->render_width; + pool->frame_bufs[cm->new_fb_idx].buf.render_height = cm->render_height; +} + +static void setup_tile_info(VP9_COMMON *cm, struct vpx_read_bit_buffer *rb) { + int min_log2_tile_cols, max_log2_tile_cols, max_ones; + vp9_get_tile_n_bits(cm->mi_cols, &min_log2_tile_cols, &max_log2_tile_cols); + + // columns + max_ones = max_log2_tile_cols - min_log2_tile_cols; + cm->log2_tile_cols = min_log2_tile_cols; + while (max_ones-- && vpx_rb_read_bit(rb)) + cm->log2_tile_cols++; + + if (cm->log2_tile_cols > 6) + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Invalid number of tile columns"); + + // rows + cm->log2_tile_rows = vpx_rb_read_bit(rb); + if (cm->log2_tile_rows) + cm->log2_tile_rows += vpx_rb_read_bit(rb); +} + +// Reads the next tile returning its size and adjusting '*data' accordingly +// based on 'is_last'. +static void get_tile_buffer(const uint8_t *const data_end, + int is_last, + struct vpx_internal_error_info *error_info, + const uint8_t **data, + vpx_decrypt_cb decrypt_cb, void *decrypt_state, + TileBuffer *buf) { + size_t size; + + if (!is_last) { + if (!read_is_valid(*data, 4, data_end)) + vpx_internal_error(error_info, VPX_CODEC_CORRUPT_FRAME, + "Truncated packet or corrupt tile length"); + + if (decrypt_cb) { + uint8_t be_data[4]; + decrypt_cb(decrypt_state, *data, be_data, 4); + size = mem_get_be32(be_data); + } else { + size = mem_get_be32(*data); + } + *data += 4; + + if (size > (size_t)(data_end - *data)) + vpx_internal_error(error_info, VPX_CODEC_CORRUPT_FRAME, + "Truncated packet or corrupt tile size"); + } else { + size = data_end - *data; + } + + buf->data = *data; + buf->size = size; + + *data += size; +} + +static void get_tile_buffers(VP9Decoder *pbi, + const uint8_t *data, const uint8_t *data_end, + int tile_cols, int tile_rows, + TileBuffer (*tile_buffers)[1 << 6]) { + int r, c; + + for (r = 0; r < tile_rows; ++r) { + for (c = 0; c < tile_cols; ++c) { + const int is_last = (r == tile_rows - 1) && (c == tile_cols - 1); + TileBuffer *const buf = &tile_buffers[r][c]; + buf->col = c; + get_tile_buffer(data_end, is_last, &pbi->common.error, &data, + pbi->decrypt_cb, pbi->decrypt_state, buf); + } + } +} + +static const uint8_t *decode_tiles(VP9Decoder *pbi, + const uint8_t *data, + const uint8_t *data_end) { + VP9_COMMON *const cm = &pbi->common; + const VPxWorkerInterface *const winterface = vpx_get_worker_interface(); + const int aligned_cols = mi_cols_aligned_to_sb(cm->mi_cols); + const int tile_cols = 1 << cm->log2_tile_cols; + const int tile_rows = 1 << cm->log2_tile_rows; + TileBuffer tile_buffers[4][1 << 6]; + int tile_row, tile_col; + int mi_row, mi_col; + TileWorkerData *tile_data = NULL; + + if (cm->lf.filter_level && !cm->skip_loop_filter && + pbi->lf_worker.data1 == NULL) { + CHECK_MEM_ERROR(cm, pbi->lf_worker.data1, + vpx_memalign(32, sizeof(LFWorkerData))); + pbi->lf_worker.hook = (VPxWorkerHook)vp9_loop_filter_worker; + if (pbi->max_threads > 1 && !winterface->reset(&pbi->lf_worker)) { + vpx_internal_error(&cm->error, VPX_CODEC_ERROR, + "Loop filter thread creation failed"); + } + } + + if (cm->lf.filter_level && !cm->skip_loop_filter) { + LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1; + // Be sure to sync as we might be resuming after a failed frame decode. + winterface->sync(&pbi->lf_worker); + vp9_loop_filter_data_reset(lf_data, get_frame_new_buffer(cm), cm, + pbi->mb.plane); + } + + assert(tile_rows <= 4); + assert(tile_cols <= (1 << 6)); + + // Note: this memset assumes above_context[0], [1] and [2] + // are allocated as part of the same buffer. + memset(cm->above_context, 0, + sizeof(*cm->above_context) * MAX_MB_PLANE * 2 * aligned_cols); + + memset(cm->above_seg_context, 0, + sizeof(*cm->above_seg_context) * aligned_cols); + + vp9_reset_lfm(cm); + + get_tile_buffers(pbi, data, data_end, tile_cols, tile_rows, tile_buffers); + + // Load all tile information into tile_data. + for (tile_row = 0; tile_row < tile_rows; ++tile_row) { + for (tile_col = 0; tile_col < tile_cols; ++tile_col) { + const TileBuffer *const buf = &tile_buffers[tile_row][tile_col]; + tile_data = pbi->tile_worker_data + tile_cols * tile_row + tile_col; + tile_data->xd = pbi->mb; + tile_data->xd.corrupted = 0; + tile_data->xd.counts = + cm->frame_parallel_decoding_mode ? NULL : &cm->counts; + vp9_zero(tile_data->dqcoeff); + vp9_tile_init(&tile_data->xd.tile, cm, tile_row, tile_col); + setup_token_decoder(buf->data, data_end, buf->size, &cm->error, + &tile_data->bit_reader, pbi->decrypt_cb, + pbi->decrypt_state); + vp9_init_macroblockd(cm, &tile_data->xd, tile_data->dqcoeff); + } + } + + for (tile_row = 0; tile_row < tile_rows; ++tile_row) { + TileInfo tile; + vp9_tile_set_row(&tile, cm, tile_row); + for (mi_row = tile.mi_row_start; mi_row < tile.mi_row_end; + mi_row += MI_BLOCK_SIZE) { + for (tile_col = 0; tile_col < tile_cols; ++tile_col) { + const int col = pbi->inv_tile_order ? + tile_cols - tile_col - 1 : tile_col; + tile_data = pbi->tile_worker_data + tile_cols * tile_row + col; + vp9_tile_set_col(&tile, cm, col); + vp9_zero(tile_data->xd.left_context); + vp9_zero(tile_data->xd.left_seg_context); + for (mi_col = tile.mi_col_start; mi_col < tile.mi_col_end; + mi_col += MI_BLOCK_SIZE) { + decode_partition(pbi, &tile_data->xd, mi_row, + mi_col, &tile_data->bit_reader, BLOCK_64X64, 4); + } + pbi->mb.corrupted |= tile_data->xd.corrupted; + if (pbi->mb.corrupted) + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Failed to decode tile data"); + } + // Loopfilter one row. + if (cm->lf.filter_level && !cm->skip_loop_filter) { + const int lf_start = mi_row - MI_BLOCK_SIZE; + LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1; + + // delay the loopfilter by 1 macroblock row. + if (lf_start < 0) continue; + + // decoding has completed: finish up the loop filter in this thread. + if (mi_row + MI_BLOCK_SIZE >= cm->mi_rows) continue; + + winterface->sync(&pbi->lf_worker); + lf_data->start = lf_start; + lf_data->stop = mi_row; + if (pbi->max_threads > 1) { + winterface->launch(&pbi->lf_worker); + } else { + winterface->execute(&pbi->lf_worker); + } + } + // After loopfiltering, the last 7 row pixels in each superblock row may + // still be changed by the longest loopfilter of the next superblock + // row. + if (pbi->frame_parallel_decode) + vp9_frameworker_broadcast(pbi->cur_buf, + mi_row << MI_BLOCK_SIZE_LOG2); + } + } + + // Loopfilter remaining rows in the frame. + if (cm->lf.filter_level && !cm->skip_loop_filter) { + LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1; + winterface->sync(&pbi->lf_worker); + lf_data->start = lf_data->stop; + lf_data->stop = cm->mi_rows; + winterface->execute(&pbi->lf_worker); + } + + // Get last tile data. + tile_data = pbi->tile_worker_data + tile_cols * tile_rows - 1; + + if (pbi->frame_parallel_decode) + vp9_frameworker_broadcast(pbi->cur_buf, INT_MAX); + return vpx_reader_find_end(&tile_data->bit_reader); +} + +// On entry 'tile_data->data_end' points to the end of the input frame, on exit +// it is updated to reflect the bitreader position of the final tile column if +// present in the tile buffer group or NULL otherwise. +static int tile_worker_hook(TileWorkerData *const tile_data, + VP9Decoder *const pbi) { + TileInfo *volatile tile = &tile_data->xd.tile; + const int final_col = (1 << pbi->common.log2_tile_cols) - 1; + const uint8_t *volatile bit_reader_end = NULL; + volatile int n = tile_data->buf_start; + tile_data->error_info.setjmp = 1; + + if (setjmp(tile_data->error_info.jmp)) { + tile_data->error_info.setjmp = 0; + tile_data->xd.corrupted = 1; + tile_data->data_end = NULL; + return 0; + } + + tile_data->xd.error_info = &tile_data->error_info; + tile_data->xd.corrupted = 0; + + do { + int mi_row, mi_col; + const TileBuffer *const buf = pbi->tile_buffers + n; + vp9_zero(tile_data->dqcoeff); + vp9_tile_init(tile, &pbi->common, 0, buf->col); + setup_token_decoder(buf->data, tile_data->data_end, buf->size, + &tile_data->error_info, &tile_data->bit_reader, + pbi->decrypt_cb, pbi->decrypt_state); + vp9_init_macroblockd(&pbi->common, &tile_data->xd, tile_data->dqcoeff); + + for (mi_row = tile->mi_row_start; mi_row < tile->mi_row_end; + mi_row += MI_BLOCK_SIZE) { + vp9_zero(tile_data->xd.left_context); + vp9_zero(tile_data->xd.left_seg_context); + for (mi_col = tile->mi_col_start; mi_col < tile->mi_col_end; + mi_col += MI_BLOCK_SIZE) { + decode_partition(pbi, &tile_data->xd, mi_row, mi_col, + &tile_data->bit_reader, BLOCK_64X64, 4); + } + } + + if (buf->col == final_col) { + bit_reader_end = vpx_reader_find_end(&tile_data->bit_reader); + } + } while (!tile_data->xd.corrupted && ++n <= tile_data->buf_end); + + tile_data->data_end = bit_reader_end; + return !tile_data->xd.corrupted; +} + +// sorts in descending order +static int compare_tile_buffers(const void *a, const void *b) { + const TileBuffer *const buf1 = (const TileBuffer*)a; + const TileBuffer *const buf2 = (const TileBuffer*)b; + return (int)(buf2->size - buf1->size); +} + +static const uint8_t *decode_tiles_mt(VP9Decoder *pbi, + const uint8_t *data, + const uint8_t *data_end) { + VP9_COMMON *const cm = &pbi->common; + const VPxWorkerInterface *const winterface = vpx_get_worker_interface(); + const uint8_t *bit_reader_end = NULL; + const int aligned_mi_cols = mi_cols_aligned_to_sb(cm->mi_cols); + const int tile_cols = 1 << cm->log2_tile_cols; + const int tile_rows = 1 << cm->log2_tile_rows; + const int num_workers = VPXMIN(pbi->max_threads, tile_cols); + int n; + + assert(tile_cols <= (1 << 6)); + assert(tile_rows == 1); + (void)tile_rows; + + if (pbi->num_tile_workers == 0) { + const int num_threads = pbi->max_threads; + CHECK_MEM_ERROR(cm, pbi->tile_workers, + vpx_malloc(num_threads * sizeof(*pbi->tile_workers))); + for (n = 0; n < num_threads; ++n) { + VPxWorker *const worker = &pbi->tile_workers[n]; + ++pbi->num_tile_workers; + + winterface->init(worker); + if (n < num_threads - 1 && !winterface->reset(worker)) { + vpx_internal_error(&cm->error, VPX_CODEC_ERROR, + "Tile decoder thread creation failed"); + } + } + } + + // Reset tile decoding hook + for (n = 0; n < num_workers; ++n) { + VPxWorker *const worker = &pbi->tile_workers[n]; + TileWorkerData *const tile_data = + &pbi->tile_worker_data[n + pbi->total_tiles]; + winterface->sync(worker); + tile_data->xd = pbi->mb; + tile_data->xd.counts = + cm->frame_parallel_decoding_mode ? NULL : &tile_data->counts; + worker->hook = (VPxWorkerHook)tile_worker_hook; + worker->data1 = tile_data; + worker->data2 = pbi; + } + + // Note: this memset assumes above_context[0], [1] and [2] + // are allocated as part of the same buffer. + memset(cm->above_context, 0, + sizeof(*cm->above_context) * MAX_MB_PLANE * 2 * aligned_mi_cols); + memset(cm->above_seg_context, 0, + sizeof(*cm->above_seg_context) * aligned_mi_cols); + + vp9_reset_lfm(cm); + + // Load tile data into tile_buffers + get_tile_buffers(pbi, data, data_end, tile_cols, tile_rows, + &pbi->tile_buffers); + + // Sort the buffers based on size in descending order. + qsort(pbi->tile_buffers, tile_cols, sizeof(pbi->tile_buffers[0]), + compare_tile_buffers); + + if (num_workers == tile_cols) { + // Rearrange the tile buffers such that the largest, and + // presumably the most difficult, tile will be decoded in the main thread. + // This should help minimize the number of instances where the main thread + // is waiting for a worker to complete. + const TileBuffer largest = pbi->tile_buffers[0]; + memmove(pbi->tile_buffers, pbi->tile_buffers + 1, + (tile_cols - 1) * sizeof(pbi->tile_buffers[0])); + pbi->tile_buffers[tile_cols - 1] = largest; + } else { + int start = 0, end = tile_cols - 2; + TileBuffer tmp; + + // Interleave the tiles to distribute the load between threads, assuming a + // larger tile implies it is more difficult to decode. + while (start < end) { + tmp = pbi->tile_buffers[start]; + pbi->tile_buffers[start] = pbi->tile_buffers[end]; + pbi->tile_buffers[end] = tmp; + start += 2; + end -= 2; + } + } + + // Initialize thread frame counts. + if (!cm->frame_parallel_decoding_mode) { + for (n = 0; n < num_workers; ++n) { + TileWorkerData *const tile_data = + (TileWorkerData*)pbi->tile_workers[n].data1; + vp9_zero(tile_data->counts); + } + } + + { + const int base = tile_cols / num_workers; + const int remain = tile_cols % num_workers; + int buf_start = 0; + + for (n = 0; n < num_workers; ++n) { + const int count = base + (remain + n) / num_workers; + VPxWorker *const worker = &pbi->tile_workers[n]; + TileWorkerData *const tile_data = (TileWorkerData*)worker->data1; + + tile_data->buf_start = buf_start; + tile_data->buf_end = buf_start + count - 1; + tile_data->data_end = data_end; + buf_start += count; + + worker->had_error = 0; + if (n == num_workers - 1) { + assert(tile_data->buf_end == tile_cols - 1); + winterface->execute(worker); + } else { + winterface->launch(worker); + } + } + + for (; n > 0; --n) { + VPxWorker *const worker = &pbi->tile_workers[n - 1]; + TileWorkerData *const tile_data = (TileWorkerData*)worker->data1; + // TODO(jzern): The tile may have specific error data associated with + // its vpx_internal_error_info which could be propagated to the main info + // in cm. Additionally once the threads have been synced and an error is + // detected, there's no point in continuing to decode tiles. + pbi->mb.corrupted |= !winterface->sync(worker); + if (!bit_reader_end) bit_reader_end = tile_data->data_end; + } + } + + // Accumulate thread frame counts. + if (!cm->frame_parallel_decoding_mode) { + for (n = 0; n < num_workers; ++n) { + TileWorkerData *const tile_data = + (TileWorkerData*)pbi->tile_workers[n].data1; + vp9_accumulate_frame_counts(&cm->counts, &tile_data->counts, 1); + } + } + + assert(bit_reader_end || pbi->mb.corrupted); + return bit_reader_end; +} + +static void error_handler(void *data) { + VP9_COMMON *const cm = (VP9_COMMON *)data; + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, "Truncated packet"); +} + +static void read_bitdepth_colorspace_sampling( + VP9_COMMON *cm, struct vpx_read_bit_buffer *rb) { + if (cm->profile >= PROFILE_2) { + cm->bit_depth = vpx_rb_read_bit(rb) ? VPX_BITS_12 : VPX_BITS_10; +#if CONFIG_VP9_HIGHBITDEPTH + cm->use_highbitdepth = 1; +#endif + } else { + cm->bit_depth = VPX_BITS_8; +#if CONFIG_VP9_HIGHBITDEPTH + cm->use_highbitdepth = 0; +#endif + } + cm->color_space = vpx_rb_read_literal(rb, 3); + if (cm->color_space != VPX_CS_SRGB) { + cm->color_range = (vpx_color_range_t)vpx_rb_read_bit(rb); + if (cm->profile == PROFILE_1 || cm->profile == PROFILE_3) { + cm->subsampling_x = vpx_rb_read_bit(rb); + cm->subsampling_y = vpx_rb_read_bit(rb); + if (cm->subsampling_x == 1 && cm->subsampling_y == 1) + vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, + "4:2:0 color not supported in profile 1 or 3"); + if (vpx_rb_read_bit(rb)) + vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, + "Reserved bit set"); + } else { + cm->subsampling_y = cm->subsampling_x = 1; + } + } else { + cm->color_range = VPX_CR_FULL_RANGE; + if (cm->profile == PROFILE_1 || cm->profile == PROFILE_3) { + // Note if colorspace is SRGB then 4:4:4 chroma sampling is assumed. + // 4:2:2 or 4:4:0 chroma sampling is not allowed. + cm->subsampling_y = cm->subsampling_x = 0; + if (vpx_rb_read_bit(rb)) + vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, + "Reserved bit set"); + } else { + vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, + "4:4:4 color not supported in profile 0 or 2"); + } + } +} + +static size_t read_uncompressed_header(VP9Decoder *pbi, + struct vpx_read_bit_buffer *rb) { + VP9_COMMON *const cm = &pbi->common; + BufferPool *const pool = cm->buffer_pool; + RefCntBuffer *const frame_bufs = pool->frame_bufs; + int i, mask, ref_index = 0; + size_t sz; + + cm->last_frame_type = cm->frame_type; + cm->last_intra_only = cm->intra_only; + + if (vpx_rb_read_literal(rb, 2) != VP9_FRAME_MARKER) + vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, + "Invalid frame marker"); + + cm->profile = vp9_read_profile(rb); +#if CONFIG_VP9_HIGHBITDEPTH + if (cm->profile >= MAX_PROFILES) + vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, + "Unsupported bitstream profile"); +#else + if (cm->profile >= PROFILE_2) + vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, + "Unsupported bitstream profile"); +#endif + + cm->show_existing_frame = vpx_rb_read_bit(rb); + if (cm->show_existing_frame) { + // Show an existing frame directly. + const int frame_to_show = cm->ref_frame_map[vpx_rb_read_literal(rb, 3)]; + lock_buffer_pool(pool); + if (frame_to_show < 0 || frame_bufs[frame_to_show].ref_count < 1) { + unlock_buffer_pool(pool); + vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, + "Buffer %d does not contain a decoded frame", + frame_to_show); + } + + ref_cnt_fb(frame_bufs, &cm->new_fb_idx, frame_to_show); + unlock_buffer_pool(pool); + pbi->refresh_frame_flags = 0; + cm->lf.filter_level = 0; + cm->show_frame = 1; + + if (pbi->frame_parallel_decode) { + for (i = 0; i < REF_FRAMES; ++i) + cm->next_ref_frame_map[i] = cm->ref_frame_map[i]; + } + return 0; + } + + cm->frame_type = (FRAME_TYPE) vpx_rb_read_bit(rb); + cm->show_frame = vpx_rb_read_bit(rb); + cm->error_resilient_mode = vpx_rb_read_bit(rb); + + if (cm->frame_type == KEY_FRAME) { + if (!vp9_read_sync_code(rb)) + vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, + "Invalid frame sync code"); + + read_bitdepth_colorspace_sampling(cm, rb); + pbi->refresh_frame_flags = (1 << REF_FRAMES) - 1; + + for (i = 0; i < REFS_PER_FRAME; ++i) { + cm->frame_refs[i].idx = INVALID_IDX; + cm->frame_refs[i].buf = NULL; + } + + setup_frame_size(cm, rb); + if (pbi->need_resync) { + memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map)); + pbi->need_resync = 0; + } + } else { + cm->intra_only = cm->show_frame ? 0 : vpx_rb_read_bit(rb); + + cm->reset_frame_context = cm->error_resilient_mode ? + 0 : vpx_rb_read_literal(rb, 2); + + if (cm->intra_only) { + if (!vp9_read_sync_code(rb)) + vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, + "Invalid frame sync code"); + if (cm->profile > PROFILE_0) { + read_bitdepth_colorspace_sampling(cm, rb); + } else { + // NOTE: The intra-only frame header does not include the specification + // of either the color format or color sub-sampling in profile 0. VP9 + // specifies that the default color format should be YUV 4:2:0 in this + // case (normative). + cm->color_space = VPX_CS_BT_601; + cm->color_range = VPX_CR_STUDIO_RANGE; + cm->subsampling_y = cm->subsampling_x = 1; + cm->bit_depth = VPX_BITS_8; +#if CONFIG_VP9_HIGHBITDEPTH + cm->use_highbitdepth = 0; +#endif + } + + pbi->refresh_frame_flags = vpx_rb_read_literal(rb, REF_FRAMES); + setup_frame_size(cm, rb); + if (pbi->need_resync) { + memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map)); + pbi->need_resync = 0; + } + } else if (pbi->need_resync != 1) { /* Skip if need resync */ + pbi->refresh_frame_flags = vpx_rb_read_literal(rb, REF_FRAMES); + for (i = 0; i < REFS_PER_FRAME; ++i) { + const int ref = vpx_rb_read_literal(rb, REF_FRAMES_LOG2); + const int idx = cm->ref_frame_map[ref]; + RefBuffer *const ref_frame = &cm->frame_refs[i]; + ref_frame->idx = idx; + ref_frame->buf = &frame_bufs[idx].buf; + cm->ref_frame_sign_bias[LAST_FRAME + i] = vpx_rb_read_bit(rb); + } + + setup_frame_size_with_refs(cm, rb); + + cm->allow_high_precision_mv = vpx_rb_read_bit(rb); + cm->interp_filter = read_interp_filter(rb); + + for (i = 0; i < REFS_PER_FRAME; ++i) { + RefBuffer *const ref_buf = &cm->frame_refs[i]; +#if CONFIG_VP9_HIGHBITDEPTH + vp9_setup_scale_factors_for_frame(&ref_buf->sf, + ref_buf->buf->y_crop_width, + ref_buf->buf->y_crop_height, + cm->width, cm->height, + cm->use_highbitdepth); +#else + vp9_setup_scale_factors_for_frame(&ref_buf->sf, + ref_buf->buf->y_crop_width, + ref_buf->buf->y_crop_height, + cm->width, cm->height); +#endif + } + } + } +#if CONFIG_VP9_HIGHBITDEPTH + get_frame_new_buffer(cm)->bit_depth = cm->bit_depth; +#endif + get_frame_new_buffer(cm)->color_space = cm->color_space; + get_frame_new_buffer(cm)->color_range = cm->color_range; + get_frame_new_buffer(cm)->render_width = cm->render_width; + get_frame_new_buffer(cm)->render_height = cm->render_height; + + if (pbi->need_resync) { + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Keyframe / intra-only frame required to reset decoder" + " state"); + } + + if (!cm->error_resilient_mode) { + cm->refresh_frame_context = vpx_rb_read_bit(rb); + cm->frame_parallel_decoding_mode = vpx_rb_read_bit(rb); + if (!cm->frame_parallel_decoding_mode) + vp9_zero(cm->counts); + } else { + cm->refresh_frame_context = 0; + cm->frame_parallel_decoding_mode = 1; + } + + // This flag will be overridden by the call to vp9_setup_past_independence + // below, forcing the use of context 0 for those frame types. + cm->frame_context_idx = vpx_rb_read_literal(rb, FRAME_CONTEXTS_LOG2); + + // Generate next_ref_frame_map. + lock_buffer_pool(pool); + for (mask = pbi->refresh_frame_flags; mask; mask >>= 1) { + if (mask & 1) { + cm->next_ref_frame_map[ref_index] = cm->new_fb_idx; + ++frame_bufs[cm->new_fb_idx].ref_count; + } else { + cm->next_ref_frame_map[ref_index] = cm->ref_frame_map[ref_index]; + } + // Current thread holds the reference frame. + if (cm->ref_frame_map[ref_index] >= 0) + ++frame_bufs[cm->ref_frame_map[ref_index]].ref_count; + ++ref_index; + } + + for (; ref_index < REF_FRAMES; ++ref_index) { + cm->next_ref_frame_map[ref_index] = cm->ref_frame_map[ref_index]; + // Current thread holds the reference frame. + if (cm->ref_frame_map[ref_index] >= 0) + ++frame_bufs[cm->ref_frame_map[ref_index]].ref_count; + } + unlock_buffer_pool(pool); + pbi->hold_ref_buf = 1; + + if (frame_is_intra_only(cm) || cm->error_resilient_mode) + vp9_setup_past_independence(cm); + + setup_loopfilter(&cm->lf, rb); + setup_quantization(cm, &pbi->mb, rb); + setup_segmentation(&cm->seg, rb); + setup_segmentation_dequant(cm); + + setup_tile_info(cm, rb); + sz = vpx_rb_read_literal(rb, 16); + + if (sz == 0) + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Invalid header size"); + + return sz; +} + +static int read_compressed_header(VP9Decoder *pbi, const uint8_t *data, + size_t partition_size) { + VP9_COMMON *const cm = &pbi->common; + MACROBLOCKD *const xd = &pbi->mb; + FRAME_CONTEXT *const fc = cm->fc; + vpx_reader r; + int k; + + if (vpx_reader_init(&r, data, partition_size, pbi->decrypt_cb, + pbi->decrypt_state)) + vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR, + "Failed to allocate bool decoder 0"); + + cm->tx_mode = xd->lossless ? ONLY_4X4 : read_tx_mode(&r); + if (cm->tx_mode == TX_MODE_SELECT) + read_tx_mode_probs(&fc->tx_probs, &r); + read_coef_probs(fc, cm->tx_mode, &r); + + for (k = 0; k < SKIP_CONTEXTS; ++k) + vp9_diff_update_prob(&r, &fc->skip_probs[k]); + + if (!frame_is_intra_only(cm)) { + nmv_context *const nmvc = &fc->nmvc; + int i, j; + + read_inter_mode_probs(fc, &r); + + if (cm->interp_filter == SWITCHABLE) + read_switchable_interp_probs(fc, &r); + + for (i = 0; i < INTRA_INTER_CONTEXTS; i++) + vp9_diff_update_prob(&r, &fc->intra_inter_prob[i]); + + cm->reference_mode = read_frame_reference_mode(cm, &r); + if (cm->reference_mode != SINGLE_REFERENCE) + setup_compound_reference_mode(cm); + read_frame_reference_mode_probs(cm, &r); + + for (j = 0; j < BLOCK_SIZE_GROUPS; j++) + for (i = 0; i < INTRA_MODES - 1; ++i) + vp9_diff_update_prob(&r, &fc->y_mode_prob[j][i]); + + for (j = 0; j < PARTITION_CONTEXTS; ++j) + for (i = 0; i < PARTITION_TYPES - 1; ++i) + vp9_diff_update_prob(&r, &fc->partition_prob[j][i]); + + read_mv_probs(nmvc, cm->allow_high_precision_mv, &r); + } + + return vpx_reader_has_error(&r); +} + +static struct vpx_read_bit_buffer *init_read_bit_buffer( + VP9Decoder *pbi, + struct vpx_read_bit_buffer *rb, + const uint8_t *data, + const uint8_t *data_end, + uint8_t clear_data[MAX_VP9_HEADER_SIZE]) { + rb->bit_offset = 0; + rb->error_handler = error_handler; + rb->error_handler_data = &pbi->common; + if (pbi->decrypt_cb) { + const int n = (int)VPXMIN(MAX_VP9_HEADER_SIZE, data_end - data); + pbi->decrypt_cb(pbi->decrypt_state, data, clear_data, n); + rb->bit_buffer = clear_data; + rb->bit_buffer_end = clear_data + n; + } else { + rb->bit_buffer = data; + rb->bit_buffer_end = data_end; + } + return rb; +} + +//------------------------------------------------------------------------------ + +int vp9_read_sync_code(struct vpx_read_bit_buffer *const rb) { + return vpx_rb_read_literal(rb, 8) == VP9_SYNC_CODE_0 && + vpx_rb_read_literal(rb, 8) == VP9_SYNC_CODE_1 && + vpx_rb_read_literal(rb, 8) == VP9_SYNC_CODE_2; +} + +void vp9_read_frame_size(struct vpx_read_bit_buffer *rb, + int *width, int *height) { + *width = vpx_rb_read_literal(rb, 16) + 1; + *height = vpx_rb_read_literal(rb, 16) + 1; +} + +BITSTREAM_PROFILE vp9_read_profile(struct vpx_read_bit_buffer *rb) { + int profile = vpx_rb_read_bit(rb); + profile |= vpx_rb_read_bit(rb) << 1; + if (profile > 2) + profile += vpx_rb_read_bit(rb); + return (BITSTREAM_PROFILE) profile; +} + +void vp9_decode_frame(VP9Decoder *pbi, + const uint8_t *data, const uint8_t *data_end, + const uint8_t **p_data_end) { + VP9_COMMON *const cm = &pbi->common; + MACROBLOCKD *const xd = &pbi->mb; + struct vpx_read_bit_buffer rb; + int context_updated = 0; + uint8_t clear_data[MAX_VP9_HEADER_SIZE]; + const size_t first_partition_size = read_uncompressed_header(pbi, + init_read_bit_buffer(pbi, &rb, data, data_end, clear_data)); + const int tile_rows = 1 << cm->log2_tile_rows; + const int tile_cols = 1 << cm->log2_tile_cols; + YV12_BUFFER_CONFIG *const new_fb = get_frame_new_buffer(cm); + xd->cur_buf = new_fb; + + if (!first_partition_size) { + // showing a frame directly + *p_data_end = data + (cm->profile <= PROFILE_2 ? 1 : 2); + return; + } + + data += vpx_rb_bytes_read(&rb); + if (!read_is_valid(data, first_partition_size, data_end)) + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Truncated packet or corrupt header length"); + + cm->use_prev_frame_mvs = !cm->error_resilient_mode && + cm->width == cm->last_width && + cm->height == cm->last_height && + !cm->last_intra_only && + cm->last_show_frame && + (cm->last_frame_type != KEY_FRAME); + + vp9_setup_block_planes(xd, cm->subsampling_x, cm->subsampling_y); + + *cm->fc = cm->frame_contexts[cm->frame_context_idx]; + if (!cm->fc->initialized) + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Uninitialized entropy context."); + + xd->corrupted = 0; + new_fb->corrupted = read_compressed_header(pbi, data, first_partition_size); + if (new_fb->corrupted) + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Decode failed. Frame data header is corrupted."); + + if (cm->lf.filter_level && !cm->skip_loop_filter) { + vp9_loop_filter_frame_init(cm, cm->lf.filter_level); + } + + // If encoded in frame parallel mode, frame context is ready after decoding + // the frame header. + if (pbi->frame_parallel_decode && cm->frame_parallel_decoding_mode) { + VPxWorker *const worker = pbi->frame_worker_owner; + FrameWorkerData *const frame_worker_data = worker->data1; + if (cm->refresh_frame_context) { + context_updated = 1; + cm->frame_contexts[cm->frame_context_idx] = *cm->fc; + } + vp9_frameworker_lock_stats(worker); + pbi->cur_buf->row = -1; + pbi->cur_buf->col = -1; + frame_worker_data->frame_context_ready = 1; + // Signal the main thread that context is ready. + vp9_frameworker_signal_stats(worker); + vp9_frameworker_unlock_stats(worker); + } + + if (pbi->tile_worker_data == NULL || + (tile_cols * tile_rows) != pbi->total_tiles) { + const int num_tile_workers = tile_cols * tile_rows + + ((pbi->max_threads > 1) ? pbi->max_threads : 0); + const size_t twd_size = num_tile_workers * sizeof(*pbi->tile_worker_data); + // Ensure tile data offsets will be properly aligned. This may fail on + // platforms without DECLARE_ALIGNED(). + assert((sizeof(*pbi->tile_worker_data) % 16) == 0); + vpx_free(pbi->tile_worker_data); + CHECK_MEM_ERROR(cm, pbi->tile_worker_data, vpx_memalign(32, twd_size)); + pbi->total_tiles = tile_rows * tile_cols; + } + + if (pbi->max_threads > 1 && tile_rows == 1 && tile_cols > 1) { + // Multi-threaded tile decoder + *p_data_end = decode_tiles_mt(pbi, data + first_partition_size, data_end); + if (!xd->corrupted) { + if (!cm->skip_loop_filter) { + // If multiple threads are used to decode tiles, then we use those + // threads to do parallel loopfiltering. + vp9_loop_filter_frame_mt(new_fb, cm, pbi->mb.plane, + cm->lf.filter_level, 0, 0, pbi->tile_workers, + pbi->num_tile_workers, &pbi->lf_row_sync); + } + } else { + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Decode failed. Frame data is corrupted."); + } + } else { + *p_data_end = decode_tiles(pbi, data + first_partition_size, data_end); + } + + if (!xd->corrupted) { + if (!cm->error_resilient_mode && !cm->frame_parallel_decoding_mode) { + vp9_adapt_coef_probs(cm); + + if (!frame_is_intra_only(cm)) { + vp9_adapt_mode_probs(cm); + vp9_adapt_mv_probs(cm, cm->allow_high_precision_mv); + } + } + } else { + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Decode failed. Frame data is corrupted."); + } + + // Non frame parallel update frame context here. + if (cm->refresh_frame_context && !context_updated) + cm->frame_contexts[cm->frame_context_idx] = *cm->fc; +} diff --git a/thirdparty/libvpx/vp9/decoder/vp9_decodeframe.h b/thirdparty/libvpx/vp9/decoder/vp9_decodeframe.h new file mode 100644 index 0000000000..ce33cbdbd9 --- /dev/null +++ b/thirdparty/libvpx/vp9/decoder/vp9_decodeframe.h @@ -0,0 +1,37 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP9_DECODER_VP9_DECODEFRAME_H_ +#define VP9_DECODER_VP9_DECODEFRAME_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +#include "vp9/common/vp9_enums.h" + +struct VP9Decoder; +struct vpx_read_bit_buffer; + +int vp9_read_sync_code(struct vpx_read_bit_buffer *const rb); +void vp9_read_frame_size(struct vpx_read_bit_buffer *rb, + int *width, int *height); +BITSTREAM_PROFILE vp9_read_profile(struct vpx_read_bit_buffer *rb); + +void vp9_decode_frame(struct VP9Decoder *pbi, + const uint8_t *data, const uint8_t *data_end, + const uint8_t **p_data_end); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_DECODER_VP9_DECODEFRAME_H_ diff --git a/thirdparty/libvpx/vp9/decoder/vp9_decodemv.c b/thirdparty/libvpx/vp9/decoder/vp9_decodemv.c new file mode 100644 index 0000000000..ffc6839ad1 --- /dev/null +++ b/thirdparty/libvpx/vp9/decoder/vp9_decodemv.c @@ -0,0 +1,911 @@ +/* + Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <assert.h> + +#include "vp9/common/vp9_common.h" +#include "vp9/common/vp9_entropy.h" +#include "vp9/common/vp9_entropymode.h" +#include "vp9/common/vp9_entropymv.h" +#include "vp9/common/vp9_mvref_common.h" +#include "vp9/common/vp9_pred_common.h" +#include "vp9/common/vp9_reconinter.h" +#include "vp9/common/vp9_seg_common.h" + +#include "vp9/decoder/vp9_decodemv.h" +#include "vp9/decoder/vp9_decodeframe.h" + +#include "vpx_dsp/vpx_dsp_common.h" + +static PREDICTION_MODE read_intra_mode(vpx_reader *r, const vpx_prob *p) { + return (PREDICTION_MODE)vpx_read_tree(r, vp9_intra_mode_tree, p); +} + +static PREDICTION_MODE read_intra_mode_y(VP9_COMMON *cm, MACROBLOCKD *xd, + vpx_reader *r, int size_group) { + const PREDICTION_MODE y_mode = + read_intra_mode(r, cm->fc->y_mode_prob[size_group]); + FRAME_COUNTS *counts = xd->counts; + if (counts) + ++counts->y_mode[size_group][y_mode]; + return y_mode; +} + +static PREDICTION_MODE read_intra_mode_uv(VP9_COMMON *cm, MACROBLOCKD *xd, + vpx_reader *r, + PREDICTION_MODE y_mode) { + const PREDICTION_MODE uv_mode = read_intra_mode(r, + cm->fc->uv_mode_prob[y_mode]); + FRAME_COUNTS *counts = xd->counts; + if (counts) + ++counts->uv_mode[y_mode][uv_mode]; + return uv_mode; +} + +static PREDICTION_MODE read_inter_mode(VP9_COMMON *cm, MACROBLOCKD *xd, + vpx_reader *r, int ctx) { + const int mode = vpx_read_tree(r, vp9_inter_mode_tree, + cm->fc->inter_mode_probs[ctx]); + FRAME_COUNTS *counts = xd->counts; + if (counts) + ++counts->inter_mode[ctx][mode]; + + return NEARESTMV + mode; +} + +static int read_segment_id(vpx_reader *r, const struct segmentation *seg) { + return vpx_read_tree(r, vp9_segment_tree, seg->tree_probs); +} + +static TX_SIZE read_selected_tx_size(VP9_COMMON *cm, MACROBLOCKD *xd, + TX_SIZE max_tx_size, vpx_reader *r) { + FRAME_COUNTS *counts = xd->counts; + const int ctx = get_tx_size_context(xd); + const vpx_prob *tx_probs = get_tx_probs(max_tx_size, ctx, &cm->fc->tx_probs); + int tx_size = vpx_read(r, tx_probs[0]); + if (tx_size != TX_4X4 && max_tx_size >= TX_16X16) { + tx_size += vpx_read(r, tx_probs[1]); + if (tx_size != TX_8X8 && max_tx_size >= TX_32X32) + tx_size += vpx_read(r, tx_probs[2]); + } + + if (counts) + ++get_tx_counts(max_tx_size, ctx, &counts->tx)[tx_size]; + return (TX_SIZE)tx_size; +} + +static INLINE TX_SIZE read_tx_size(VP9_COMMON *cm, MACROBLOCKD *xd, + int allow_select, vpx_reader *r) { + TX_MODE tx_mode = cm->tx_mode; + BLOCK_SIZE bsize = xd->mi[0]->sb_type; + const TX_SIZE max_tx_size = max_txsize_lookup[bsize]; + if (allow_select && tx_mode == TX_MODE_SELECT && bsize >= BLOCK_8X8) + return read_selected_tx_size(cm, xd, max_tx_size, r); + else + return VPXMIN(max_tx_size, tx_mode_to_biggest_tx_size[tx_mode]); +} + +static int dec_get_segment_id(const VP9_COMMON *cm, const uint8_t *segment_ids, + int mi_offset, int x_mis, int y_mis) { + int x, y, segment_id = INT_MAX; + + for (y = 0; y < y_mis; y++) + for (x = 0; x < x_mis; x++) + segment_id = + VPXMIN(segment_id, segment_ids[mi_offset + y * cm->mi_cols + x]); + + assert(segment_id >= 0 && segment_id < MAX_SEGMENTS); + return segment_id; +} + +static void set_segment_id(VP9_COMMON *cm, int mi_offset, + int x_mis, int y_mis, int segment_id) { + int x, y; + + assert(segment_id >= 0 && segment_id < MAX_SEGMENTS); + + for (y = 0; y < y_mis; y++) + for (x = 0; x < x_mis; x++) + cm->current_frame_seg_map[mi_offset + y * cm->mi_cols + x] = segment_id; +} + +static void copy_segment_id(const VP9_COMMON *cm, + const uint8_t *last_segment_ids, + uint8_t *current_segment_ids, + int mi_offset, int x_mis, int y_mis) { + int x, y; + + for (y = 0; y < y_mis; y++) + for (x = 0; x < x_mis; x++) + current_segment_ids[mi_offset + y * cm->mi_cols + x] = last_segment_ids ? + last_segment_ids[mi_offset + y * cm->mi_cols + x] : 0; +} + +static int read_intra_segment_id(VP9_COMMON *const cm, int mi_offset, + int x_mis, int y_mis, + vpx_reader *r) { + struct segmentation *const seg = &cm->seg; + int segment_id; + + if (!seg->enabled) + return 0; // Default for disabled segmentation + + if (!seg->update_map) { + copy_segment_id(cm, cm->last_frame_seg_map, cm->current_frame_seg_map, + mi_offset, x_mis, y_mis); + return 0; + } + + segment_id = read_segment_id(r, seg); + set_segment_id(cm, mi_offset, x_mis, y_mis, segment_id); + return segment_id; +} + +static int read_inter_segment_id(VP9_COMMON *const cm, MACROBLOCKD *const xd, + int mi_row, int mi_col, vpx_reader *r, + int x_mis, int y_mis) { + struct segmentation *const seg = &cm->seg; + MODE_INFO *const mi = xd->mi[0]; + int predicted_segment_id, segment_id; + const int mi_offset = mi_row * cm->mi_cols + mi_col; + + if (!seg->enabled) + return 0; // Default for disabled segmentation + + predicted_segment_id = cm->last_frame_seg_map ? + dec_get_segment_id(cm, cm->last_frame_seg_map, mi_offset, x_mis, y_mis) : + 0; + + if (!seg->update_map) { + copy_segment_id(cm, cm->last_frame_seg_map, cm->current_frame_seg_map, + mi_offset, x_mis, y_mis); + return predicted_segment_id; + } + + if (seg->temporal_update) { + const vpx_prob pred_prob = vp9_get_pred_prob_seg_id(seg, xd); + mi->seg_id_predicted = vpx_read(r, pred_prob); + segment_id = mi->seg_id_predicted ? predicted_segment_id + : read_segment_id(r, seg); + } else { + segment_id = read_segment_id(r, seg); + } + set_segment_id(cm, mi_offset, x_mis, y_mis, segment_id); + return segment_id; +} + +static int read_skip(VP9_COMMON *cm, const MACROBLOCKD *xd, + int segment_id, vpx_reader *r) { + if (segfeature_active(&cm->seg, segment_id, SEG_LVL_SKIP)) { + return 1; + } else { + const int ctx = vp9_get_skip_context(xd); + const int skip = vpx_read(r, cm->fc->skip_probs[ctx]); + FRAME_COUNTS *counts = xd->counts; + if (counts) + ++counts->skip[ctx][skip]; + return skip; + } +} + +static void read_intra_frame_mode_info(VP9_COMMON *const cm, + MACROBLOCKD *const xd, + int mi_row, int mi_col, vpx_reader *r, + int x_mis, int y_mis) { + MODE_INFO *const mi = xd->mi[0]; + const MODE_INFO *above_mi = xd->above_mi; + const MODE_INFO *left_mi = xd->left_mi; + const BLOCK_SIZE bsize = mi->sb_type; + int i; + const int mi_offset = mi_row * cm->mi_cols + mi_col; + + mi->segment_id = read_intra_segment_id(cm, mi_offset, x_mis, y_mis, r); + mi->skip = read_skip(cm, xd, mi->segment_id, r); + mi->tx_size = read_tx_size(cm, xd, 1, r); + mi->ref_frame[0] = INTRA_FRAME; + mi->ref_frame[1] = NONE; + + switch (bsize) { + case BLOCK_4X4: + for (i = 0; i < 4; ++i) + mi->bmi[i].as_mode = + read_intra_mode(r, get_y_mode_probs(mi, above_mi, left_mi, i)); + mi->mode = mi->bmi[3].as_mode; + break; + case BLOCK_4X8: + mi->bmi[0].as_mode = mi->bmi[2].as_mode = + read_intra_mode(r, get_y_mode_probs(mi, above_mi, left_mi, 0)); + mi->bmi[1].as_mode = mi->bmi[3].as_mode = mi->mode = + read_intra_mode(r, get_y_mode_probs(mi, above_mi, left_mi, 1)); + break; + case BLOCK_8X4: + mi->bmi[0].as_mode = mi->bmi[1].as_mode = + read_intra_mode(r, get_y_mode_probs(mi, above_mi, left_mi, 0)); + mi->bmi[2].as_mode = mi->bmi[3].as_mode = mi->mode = + read_intra_mode(r, get_y_mode_probs(mi, above_mi, left_mi, 2)); + break; + default: + mi->mode = read_intra_mode(r, + get_y_mode_probs(mi, above_mi, left_mi, 0)); + } + + mi->uv_mode = read_intra_mode(r, vp9_kf_uv_mode_prob[mi->mode]); +} + +static int read_mv_component(vpx_reader *r, + const nmv_component *mvcomp, int usehp) { + int mag, d, fr, hp; + const int sign = vpx_read(r, mvcomp->sign); + const int mv_class = vpx_read_tree(r, vp9_mv_class_tree, mvcomp->classes); + const int class0 = mv_class == MV_CLASS_0; + + // Integer part + if (class0) { + d = vpx_read_tree(r, vp9_mv_class0_tree, mvcomp->class0); + mag = 0; + } else { + int i; + const int n = mv_class + CLASS0_BITS - 1; // number of bits + + d = 0; + for (i = 0; i < n; ++i) + d |= vpx_read(r, mvcomp->bits[i]) << i; + mag = CLASS0_SIZE << (mv_class + 2); + } + + // Fractional part + fr = vpx_read_tree(r, vp9_mv_fp_tree, class0 ? mvcomp->class0_fp[d] + : mvcomp->fp); + + // High precision part (if hp is not used, the default value of the hp is 1) + hp = usehp ? vpx_read(r, class0 ? mvcomp->class0_hp : mvcomp->hp) + : 1; + + // Result + mag += ((d << 3) | (fr << 1) | hp) + 1; + return sign ? -mag : mag; +} + +static INLINE void read_mv(vpx_reader *r, MV *mv, const MV *ref, + const nmv_context *ctx, + nmv_context_counts *counts, int allow_hp) { + const MV_JOINT_TYPE joint_type = + (MV_JOINT_TYPE)vpx_read_tree(r, vp9_mv_joint_tree, ctx->joints); + const int use_hp = allow_hp && use_mv_hp(ref); + MV diff = {0, 0}; + + if (mv_joint_vertical(joint_type)) + diff.row = read_mv_component(r, &ctx->comps[0], use_hp); + + if (mv_joint_horizontal(joint_type)) + diff.col = read_mv_component(r, &ctx->comps[1], use_hp); + + vp9_inc_mv(&diff, counts); + + mv->row = ref->row + diff.row; + mv->col = ref->col + diff.col; +} + +static REFERENCE_MODE read_block_reference_mode(VP9_COMMON *cm, + const MACROBLOCKD *xd, + vpx_reader *r) { + if (cm->reference_mode == REFERENCE_MODE_SELECT) { + const int ctx = vp9_get_reference_mode_context(cm, xd); + const REFERENCE_MODE mode = + (REFERENCE_MODE)vpx_read(r, cm->fc->comp_inter_prob[ctx]); + FRAME_COUNTS *counts = xd->counts; + if (counts) + ++counts->comp_inter[ctx][mode]; + return mode; // SINGLE_REFERENCE or COMPOUND_REFERENCE + } else { + return cm->reference_mode; + } +} + +// Read the referncence frame +static void read_ref_frames(VP9_COMMON *const cm, MACROBLOCKD *const xd, + vpx_reader *r, + int segment_id, MV_REFERENCE_FRAME ref_frame[2]) { + FRAME_CONTEXT *const fc = cm->fc; + FRAME_COUNTS *counts = xd->counts; + + if (segfeature_active(&cm->seg, segment_id, SEG_LVL_REF_FRAME)) { + ref_frame[0] = (MV_REFERENCE_FRAME)get_segdata(&cm->seg, segment_id, + SEG_LVL_REF_FRAME); + ref_frame[1] = NONE; + } else { + const REFERENCE_MODE mode = read_block_reference_mode(cm, xd, r); + // FIXME(rbultje) I'm pretty sure this breaks segmentation ref frame coding + if (mode == COMPOUND_REFERENCE) { + const int idx = cm->ref_frame_sign_bias[cm->comp_fixed_ref]; + const int ctx = vp9_get_pred_context_comp_ref_p(cm, xd); + const int bit = vpx_read(r, fc->comp_ref_prob[ctx]); + if (counts) + ++counts->comp_ref[ctx][bit]; + ref_frame[idx] = cm->comp_fixed_ref; + ref_frame[!idx] = cm->comp_var_ref[bit]; + } else if (mode == SINGLE_REFERENCE) { + const int ctx0 = vp9_get_pred_context_single_ref_p1(xd); + const int bit0 = vpx_read(r, fc->single_ref_prob[ctx0][0]); + if (counts) + ++counts->single_ref[ctx0][0][bit0]; + if (bit0) { + const int ctx1 = vp9_get_pred_context_single_ref_p2(xd); + const int bit1 = vpx_read(r, fc->single_ref_prob[ctx1][1]); + if (counts) + ++counts->single_ref[ctx1][1][bit1]; + ref_frame[0] = bit1 ? ALTREF_FRAME : GOLDEN_FRAME; + } else { + ref_frame[0] = LAST_FRAME; + } + + ref_frame[1] = NONE; + } else { + assert(0 && "Invalid prediction mode."); + } + } +} + +// TODO(slavarnway): Move this decoder version of +// vp9_get_pred_context_switchable_interp() to vp9_pred_common.h and update the +// encoder. +// +// Returns a context number for the given MB prediction signal +static int dec_get_pred_context_switchable_interp(const MACROBLOCKD *xd) { + // Note: + // The mode info data structure has a one element border above and to the + // left of the entries corresponding to real macroblocks. + // The prediction flags in these dummy entries are initialized to 0. + const MODE_INFO *const left_mi = xd->left_mi; + const int left_type = left_mi ? left_mi->interp_filter : SWITCHABLE_FILTERS; + const MODE_INFO *const above_mi = xd->above_mi; + const int above_type = above_mi ? above_mi->interp_filter + : SWITCHABLE_FILTERS; + + if (left_type == above_type) + return left_type; + else if (left_type == SWITCHABLE_FILTERS) + return above_type; + else if (above_type == SWITCHABLE_FILTERS) + return left_type; + else + return SWITCHABLE_FILTERS; +} + +static INLINE INTERP_FILTER read_switchable_interp_filter( + VP9_COMMON *const cm, MACROBLOCKD *const xd, + vpx_reader *r) { + const int ctx = dec_get_pred_context_switchable_interp(xd); + const INTERP_FILTER type = + (INTERP_FILTER)vpx_read_tree(r, vp9_switchable_interp_tree, + cm->fc->switchable_interp_prob[ctx]); + FRAME_COUNTS *counts = xd->counts; + if (counts) + ++counts->switchable_interp[ctx][type]; + return type; +} + +static void read_intra_block_mode_info(VP9_COMMON *const cm, + MACROBLOCKD *const xd, MODE_INFO *mi, + vpx_reader *r) { + const BLOCK_SIZE bsize = mi->sb_type; + int i; + + switch (bsize) { + case BLOCK_4X4: + for (i = 0; i < 4; ++i) + mi->bmi[i].as_mode = read_intra_mode_y(cm, xd, r, 0); + mi->mode = mi->bmi[3].as_mode; + break; + case BLOCK_4X8: + mi->bmi[0].as_mode = mi->bmi[2].as_mode = read_intra_mode_y(cm, xd, + r, 0); + mi->bmi[1].as_mode = mi->bmi[3].as_mode = mi->mode = + read_intra_mode_y(cm, xd, r, 0); + break; + case BLOCK_8X4: + mi->bmi[0].as_mode = mi->bmi[1].as_mode = read_intra_mode_y(cm, xd, + r, 0); + mi->bmi[2].as_mode = mi->bmi[3].as_mode = mi->mode = + read_intra_mode_y(cm, xd, r, 0); + break; + default: + mi->mode = read_intra_mode_y(cm, xd, r, size_group_lookup[bsize]); + } + + mi->uv_mode = read_intra_mode_uv(cm, xd, r, mi->mode); + + // Initialize interp_filter here so we do not have to check for inter block + // modes in dec_get_pred_context_switchable_interp() + mi->interp_filter = SWITCHABLE_FILTERS; + + mi->ref_frame[0] = INTRA_FRAME; + mi->ref_frame[1] = NONE; +} + +static INLINE int is_mv_valid(const MV *mv) { + return mv->row > MV_LOW && mv->row < MV_UPP && + mv->col > MV_LOW && mv->col < MV_UPP; +} + +static INLINE void copy_mv_pair(int_mv *dst, const int_mv *src) { + memcpy(dst, src, sizeof(*dst) * 2); +} + +static INLINE void zero_mv_pair(int_mv *dst) { + memset(dst, 0, sizeof(*dst) * 2); +} + +static INLINE int assign_mv(VP9_COMMON *cm, MACROBLOCKD *xd, + PREDICTION_MODE mode, + int_mv mv[2], int_mv ref_mv[2], + int_mv near_nearest_mv[2], + int is_compound, int allow_hp, vpx_reader *r) { + int i; + int ret = 1; + + switch (mode) { + case NEWMV: { + FRAME_COUNTS *counts = xd->counts; + nmv_context_counts *const mv_counts = counts ? &counts->mv : NULL; + for (i = 0; i < 1 + is_compound; ++i) { + read_mv(r, &mv[i].as_mv, &ref_mv[i].as_mv, &cm->fc->nmvc, mv_counts, + allow_hp); + ret = ret && is_mv_valid(&mv[i].as_mv); + } + break; + } + case NEARMV: + case NEARESTMV: { + copy_mv_pair(mv, near_nearest_mv); + break; + } + case ZEROMV: { + zero_mv_pair(mv); + break; + } + default: { + return 0; + } + } + return ret; +} + +static int read_is_inter_block(VP9_COMMON *const cm, MACROBLOCKD *const xd, + int segment_id, vpx_reader *r) { + if (segfeature_active(&cm->seg, segment_id, SEG_LVL_REF_FRAME)) { + return get_segdata(&cm->seg, segment_id, SEG_LVL_REF_FRAME) != INTRA_FRAME; + } else { + const int ctx = get_intra_inter_context(xd); + const int is_inter = vpx_read(r, cm->fc->intra_inter_prob[ctx]); + FRAME_COUNTS *counts = xd->counts; + if (counts) + ++counts->intra_inter[ctx][is_inter]; + return is_inter; + } +} + +static void dec_find_best_ref_mvs(int allow_hp, int_mv *mvlist, int_mv *best_mv, + int refmv_count) { + int i; + + // Make sure all the candidates are properly clamped etc + for (i = 0; i < refmv_count; ++i) { + lower_mv_precision(&mvlist[i].as_mv, allow_hp); + *best_mv = mvlist[i]; + } +} + +static void fpm_sync(void *const data, int mi_row) { + VP9Decoder *const pbi = (VP9Decoder *)data; + vp9_frameworker_wait(pbi->frame_worker_owner, pbi->common.prev_frame, + mi_row << MI_BLOCK_SIZE_LOG2); +} + +// This macro is used to add a motion vector mv_ref list if it isn't +// already in the list. If it's the second motion vector or early_break +// it will also skip all additional processing and jump to Done! +#define ADD_MV_REF_LIST_EB(mv, refmv_count, mv_ref_list, Done) \ + do { \ + if (refmv_count) { \ + if ((mv).as_int != (mv_ref_list)[0].as_int) { \ + (mv_ref_list)[(refmv_count)] = (mv); \ + refmv_count++; \ + goto Done; \ + } \ + } else { \ + (mv_ref_list)[(refmv_count)++] = (mv); \ + if (early_break) \ + goto Done; \ + } \ + } while (0) + +// If either reference frame is different, not INTRA, and they +// are different from each other scale and add the mv to our list. +#define IF_DIFF_REF_FRAME_ADD_MV_EB(mbmi, ref_frame, ref_sign_bias, \ + refmv_count, mv_ref_list, Done) \ + do { \ + if (is_inter_block(mbmi)) { \ + if ((mbmi)->ref_frame[0] != ref_frame) \ + ADD_MV_REF_LIST_EB(scale_mv((mbmi), 0, ref_frame, ref_sign_bias), \ + refmv_count, mv_ref_list, Done); \ + if (has_second_ref(mbmi) && \ + (mbmi)->ref_frame[1] != ref_frame && \ + (mbmi)->mv[1].as_int != (mbmi)->mv[0].as_int) \ + ADD_MV_REF_LIST_EB(scale_mv((mbmi), 1, ref_frame, ref_sign_bias), \ + refmv_count, mv_ref_list, Done); \ + } \ + } while (0) + +// This function searches the neighborhood of a given MB/SB +// to try and find candidate reference vectors. +static int dec_find_mv_refs(const VP9_COMMON *cm, const MACROBLOCKD *xd, + PREDICTION_MODE mode, MV_REFERENCE_FRAME ref_frame, + const POSITION *const mv_ref_search, + int_mv *mv_ref_list, + int mi_row, int mi_col, int block, int is_sub8x8, + find_mv_refs_sync sync, void *const data) { + const int *ref_sign_bias = cm->ref_frame_sign_bias; + int i, refmv_count = 0; + int different_ref_found = 0; + const MV_REF *const prev_frame_mvs = cm->use_prev_frame_mvs ? + cm->prev_frame->mvs + mi_row * cm->mi_cols + mi_col : NULL; + const TileInfo *const tile = &xd->tile; + // If mode is nearestmv or newmv (uses nearestmv as a reference) then stop + // searching after the first mv is found. + const int early_break = (mode != NEARMV); + + // Blank the reference vector list + memset(mv_ref_list, 0, sizeof(*mv_ref_list) * MAX_MV_REF_CANDIDATES); + + i = 0; + if (is_sub8x8) { + // If the size < 8x8 we get the mv from the bmi substructure for the + // nearest two blocks. + for (i = 0; i < 2; ++i) { + const POSITION *const mv_ref = &mv_ref_search[i]; + if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) { + const MODE_INFO *const candidate_mi = + xd->mi[mv_ref->col + mv_ref->row * xd->mi_stride]; + different_ref_found = 1; + + if (candidate_mi->ref_frame[0] == ref_frame) + ADD_MV_REF_LIST_EB( + get_sub_block_mv(candidate_mi, 0, mv_ref->col, block), + refmv_count, mv_ref_list, Done); + else if (candidate_mi->ref_frame[1] == ref_frame) + ADD_MV_REF_LIST_EB( + get_sub_block_mv(candidate_mi, 1, mv_ref->col, block), + refmv_count, mv_ref_list, Done); + } + } + } + + // Check the rest of the neighbors in much the same way + // as before except we don't need to keep track of sub blocks or + // mode counts. + for (; i < MVREF_NEIGHBOURS; ++i) { + const POSITION *const mv_ref = &mv_ref_search[i]; + if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) { + const MODE_INFO *const candidate = + xd->mi[mv_ref->col + mv_ref->row * xd->mi_stride]; + different_ref_found = 1; + + if (candidate->ref_frame[0] == ref_frame) + ADD_MV_REF_LIST_EB(candidate->mv[0], refmv_count, mv_ref_list, Done); + else if (candidate->ref_frame[1] == ref_frame) + ADD_MV_REF_LIST_EB(candidate->mv[1], refmv_count, mv_ref_list, Done); + } + } + + // TODO(hkuang): Remove this sync after fixing pthread_cond_broadcast + // on windows platform. The sync here is unnecessary if use_prev_frame_mvs + // is 0. But after removing it, there will be hang in the unit test on windows + // due to several threads waiting for a thread's signal. +#if defined(_WIN32) && !HAVE_PTHREAD_H + if (cm->frame_parallel_decode && sync != NULL) { + sync(data, mi_row); + } +#endif + + // Check the last frame's mode and mv info. + if (prev_frame_mvs) { + // Synchronize here for frame parallel decode if sync function is provided. + if (cm->frame_parallel_decode && sync != NULL) { + sync(data, mi_row); + } + + if (prev_frame_mvs->ref_frame[0] == ref_frame) { + ADD_MV_REF_LIST_EB(prev_frame_mvs->mv[0], refmv_count, mv_ref_list, Done); + } else if (prev_frame_mvs->ref_frame[1] == ref_frame) { + ADD_MV_REF_LIST_EB(prev_frame_mvs->mv[1], refmv_count, mv_ref_list, Done); + } + } + + // Since we couldn't find 2 mvs from the same reference frame + // go back through the neighbors and find motion vectors from + // different reference frames. + if (different_ref_found) { + for (i = 0; i < MVREF_NEIGHBOURS; ++i) { + const POSITION *mv_ref = &mv_ref_search[i]; + if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) { + const MODE_INFO *const candidate = + xd->mi[mv_ref->col + mv_ref->row * xd->mi_stride]; + + // If the candidate is INTRA we don't want to consider its mv. + IF_DIFF_REF_FRAME_ADD_MV_EB(candidate, ref_frame, ref_sign_bias, + refmv_count, mv_ref_list, Done); + } + } + } + + // Since we still don't have a candidate we'll try the last frame. + if (prev_frame_mvs) { + if (prev_frame_mvs->ref_frame[0] != ref_frame && + prev_frame_mvs->ref_frame[0] > INTRA_FRAME) { + int_mv mv = prev_frame_mvs->mv[0]; + if (ref_sign_bias[prev_frame_mvs->ref_frame[0]] != + ref_sign_bias[ref_frame]) { + mv.as_mv.row *= -1; + mv.as_mv.col *= -1; + } + ADD_MV_REF_LIST_EB(mv, refmv_count, mv_ref_list, Done); + } + + if (prev_frame_mvs->ref_frame[1] > INTRA_FRAME && + prev_frame_mvs->ref_frame[1] != ref_frame && + prev_frame_mvs->mv[1].as_int != prev_frame_mvs->mv[0].as_int) { + int_mv mv = prev_frame_mvs->mv[1]; + if (ref_sign_bias[prev_frame_mvs->ref_frame[1]] != + ref_sign_bias[ref_frame]) { + mv.as_mv.row *= -1; + mv.as_mv.col *= -1; + } + ADD_MV_REF_LIST_EB(mv, refmv_count, mv_ref_list, Done); + } + } + + if (mode == NEARMV) + refmv_count = MAX_MV_REF_CANDIDATES; + else + // we only care about the nearestmv for the remaining modes + refmv_count = 1; + + Done: + // Clamp vectors + for (i = 0; i < refmv_count; ++i) + clamp_mv_ref(&mv_ref_list[i].as_mv, xd); + + return refmv_count; +} + +static void append_sub8x8_mvs_for_idx(VP9_COMMON *cm, MACROBLOCKD *xd, + const POSITION *const mv_ref_search, + PREDICTION_MODE b_mode, int block, + int ref, int mi_row, int mi_col, + int_mv *best_sub8x8) { + int_mv mv_list[MAX_MV_REF_CANDIDATES]; + MODE_INFO *const mi = xd->mi[0]; + b_mode_info *bmi = mi->bmi; + int n; + int refmv_count; + + assert(MAX_MV_REF_CANDIDATES == 2); + + refmv_count = dec_find_mv_refs(cm, xd, b_mode, mi->ref_frame[ref], + mv_ref_search, mv_list, mi_row, mi_col, block, + 1, NULL, NULL); + + switch (block) { + case 0: + best_sub8x8->as_int = mv_list[refmv_count - 1].as_int; + break; + case 1: + case 2: + if (b_mode == NEARESTMV) { + best_sub8x8->as_int = bmi[0].as_mv[ref].as_int; + } else { + best_sub8x8->as_int = 0; + for (n = 0; n < refmv_count; ++n) + if (bmi[0].as_mv[ref].as_int != mv_list[n].as_int) { + best_sub8x8->as_int = mv_list[n].as_int; + break; + } + } + break; + case 3: + if (b_mode == NEARESTMV) { + best_sub8x8->as_int = bmi[2].as_mv[ref].as_int; + } else { + int_mv candidates[2 + MAX_MV_REF_CANDIDATES]; + candidates[0] = bmi[1].as_mv[ref]; + candidates[1] = bmi[0].as_mv[ref]; + candidates[2] = mv_list[0]; + candidates[3] = mv_list[1]; + best_sub8x8->as_int = 0; + for (n = 0; n < 2 + MAX_MV_REF_CANDIDATES; ++n) + if (bmi[2].as_mv[ref].as_int != candidates[n].as_int) { + best_sub8x8->as_int = candidates[n].as_int; + break; + } + } + break; + default: + assert(0 && "Invalid block index."); + } +} + +static uint8_t get_mode_context(const VP9_COMMON *cm, const MACROBLOCKD *xd, + const POSITION *const mv_ref_search, + int mi_row, int mi_col) { + int i; + int context_counter = 0; + const TileInfo *const tile = &xd->tile; + + // Get mode count from nearest 2 blocks + for (i = 0; i < 2; ++i) { + const POSITION *const mv_ref = &mv_ref_search[i]; + if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) { + const MODE_INFO *const candidate = + xd->mi[mv_ref->col + mv_ref->row * xd->mi_stride]; + // Keep counts for entropy encoding. + context_counter += mode_2_counter[candidate->mode]; + } + } + + return counter_to_context[context_counter]; +} + +static void read_inter_block_mode_info(VP9Decoder *const pbi, + MACROBLOCKD *const xd, + MODE_INFO *const mi, + int mi_row, int mi_col, vpx_reader *r) { + VP9_COMMON *const cm = &pbi->common; + const BLOCK_SIZE bsize = mi->sb_type; + const int allow_hp = cm->allow_high_precision_mv; + int_mv best_ref_mvs[2]; + int ref, is_compound; + uint8_t inter_mode_ctx; + const POSITION *const mv_ref_search = mv_ref_blocks[bsize]; + + read_ref_frames(cm, xd, r, mi->segment_id, mi->ref_frame); + is_compound = has_second_ref(mi); + inter_mode_ctx = get_mode_context(cm, xd, mv_ref_search, mi_row, mi_col); + + if (segfeature_active(&cm->seg, mi->segment_id, SEG_LVL_SKIP)) { + mi->mode = ZEROMV; + if (bsize < BLOCK_8X8) { + vpx_internal_error(xd->error_info, VPX_CODEC_UNSUP_BITSTREAM, + "Invalid usage of segement feature on small blocks"); + return; + } + } else { + if (bsize >= BLOCK_8X8) + mi->mode = read_inter_mode(cm, xd, r, inter_mode_ctx); + else + // Sub 8x8 blocks use the nearestmv as a ref_mv if the b_mode is NEWMV. + // Setting mode to NEARESTMV forces the search to stop after the nearestmv + // has been found. After b_modes have been read, mode will be overwritten + // by the last b_mode. + mi->mode = NEARESTMV; + + if (mi->mode != ZEROMV) { + for (ref = 0; ref < 1 + is_compound; ++ref) { + int_mv tmp_mvs[MAX_MV_REF_CANDIDATES]; + const MV_REFERENCE_FRAME frame = mi->ref_frame[ref]; + int refmv_count; + + refmv_count = dec_find_mv_refs(cm, xd, mi->mode, frame, mv_ref_search, + tmp_mvs, mi_row, mi_col, -1, 0, + fpm_sync, (void *)pbi); + + dec_find_best_ref_mvs(allow_hp, tmp_mvs, &best_ref_mvs[ref], + refmv_count); + } + } + } + + mi->interp_filter = (cm->interp_filter == SWITCHABLE) + ? read_switchable_interp_filter(cm, xd, r) + : cm->interp_filter; + + if (bsize < BLOCK_8X8) { + const int num_4x4_w = 1 << xd->bmode_blocks_wl; + const int num_4x4_h = 1 << xd->bmode_blocks_hl; + int idx, idy; + PREDICTION_MODE b_mode; + int_mv best_sub8x8[2]; + for (idy = 0; idy < 2; idy += num_4x4_h) { + for (idx = 0; idx < 2; idx += num_4x4_w) { + const int j = idy * 2 + idx; + b_mode = read_inter_mode(cm, xd, r, inter_mode_ctx); + + if (b_mode == NEARESTMV || b_mode == NEARMV) { + for (ref = 0; ref < 1 + is_compound; ++ref) + append_sub8x8_mvs_for_idx(cm, xd, mv_ref_search, b_mode, j, ref, + mi_row, mi_col, &best_sub8x8[ref]); + } + + if (!assign_mv(cm, xd, b_mode, mi->bmi[j].as_mv, best_ref_mvs, + best_sub8x8, is_compound, allow_hp, r)) { + xd->corrupted |= 1; + break; + } + + if (num_4x4_h == 2) + mi->bmi[j + 2] = mi->bmi[j]; + if (num_4x4_w == 2) + mi->bmi[j + 1] = mi->bmi[j]; + } + } + + mi->mode = b_mode; + + copy_mv_pair(mi->mv, mi->bmi[3].as_mv); + } else { + xd->corrupted |= !assign_mv(cm, xd, mi->mode, mi->mv, best_ref_mvs, + best_ref_mvs, is_compound, allow_hp, r); + } +} + +static void read_inter_frame_mode_info(VP9Decoder *const pbi, + MACROBLOCKD *const xd, + int mi_row, int mi_col, vpx_reader *r, + int x_mis, int y_mis) { + VP9_COMMON *const cm = &pbi->common; + MODE_INFO *const mi = xd->mi[0]; + int inter_block; + + mi->segment_id = read_inter_segment_id(cm, xd, mi_row, mi_col, r, x_mis, + y_mis); + mi->skip = read_skip(cm, xd, mi->segment_id, r); + inter_block = read_is_inter_block(cm, xd, mi->segment_id, r); + mi->tx_size = read_tx_size(cm, xd, !mi->skip || !inter_block, r); + + if (inter_block) + read_inter_block_mode_info(pbi, xd, mi, mi_row, mi_col, r); + else + read_intra_block_mode_info(cm, xd, mi, r); +} + +static INLINE void copy_ref_frame_pair(MV_REFERENCE_FRAME *dst, + const MV_REFERENCE_FRAME *src) { + memcpy(dst, src, sizeof(*dst) * 2); +} + +void vp9_read_mode_info(VP9Decoder *const pbi, MACROBLOCKD *xd, + int mi_row, int mi_col, vpx_reader *r, + int x_mis, int y_mis) { + VP9_COMMON *const cm = &pbi->common; + MODE_INFO *const mi = xd->mi[0]; + MV_REF* frame_mvs = cm->cur_frame->mvs + mi_row * cm->mi_cols + mi_col; + int w, h; + + if (frame_is_intra_only(cm)) { + read_intra_frame_mode_info(cm, xd, mi_row, mi_col, r, x_mis, y_mis); + } else { + read_inter_frame_mode_info(pbi, xd, mi_row, mi_col, r, x_mis, y_mis); + + for (h = 0; h < y_mis; ++h) { + for (w = 0; w < x_mis; ++w) { + MV_REF *const mv = frame_mvs + w; + copy_ref_frame_pair(mv->ref_frame, mi->ref_frame); + copy_mv_pair(mv->mv, mi->mv); + } + frame_mvs += cm->mi_cols; + } + } +#if CONFIG_BETTER_HW_COMPATIBILITY && CONFIG_VP9_HIGHBITDEPTH + if ((xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) && + (xd->above_mi == NULL || xd->left_mi == NULL) && + !is_inter_block(mi) && need_top_left[mi->uv_mode]) + assert(0); +#endif // CONFIG_BETTER_HW_COMPATIBILITY && CONFIG_VP9_HIGHBITDEPTH +} diff --git a/thirdparty/libvpx/vp9/decoder/vp9_decodemv.h b/thirdparty/libvpx/vp9/decoder/vp9_decodemv.h new file mode 100644 index 0000000000..45569ec81f --- /dev/null +++ b/thirdparty/libvpx/vp9/decoder/vp9_decodemv.h @@ -0,0 +1,30 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_DECODER_VP9_DECODEMV_H_ +#define VP9_DECODER_VP9_DECODEMV_H_ + +#include "vpx_dsp/bitreader.h" + +#include "vp9/decoder/vp9_decoder.h" + +#ifdef __cplusplus +extern "C" { +#endif + +void vp9_read_mode_info(VP9Decoder *const pbi, MACROBLOCKD *xd, + int mi_row, int mi_col, vpx_reader *r, + int x_mis, int y_mis); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_DECODER_VP9_DECODEMV_H_ diff --git a/thirdparty/libvpx/vp9/decoder/vp9_decoder.c b/thirdparty/libvpx/vp9/decoder/vp9_decoder.c new file mode 100644 index 0000000000..935c04f3aa --- /dev/null +++ b/thirdparty/libvpx/vp9/decoder/vp9_decoder.c @@ -0,0 +1,518 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <assert.h> +#include <limits.h> +#include <stdio.h> + +#include "./vp9_rtcd.h" +#include "./vpx_dsp_rtcd.h" +#include "./vpx_scale_rtcd.h" + +#include "vpx_mem/vpx_mem.h" +#include "vpx_ports/system_state.h" +#include "vpx_ports/vpx_once.h" +#include "vpx_ports/vpx_timer.h" +#include "vpx_scale/vpx_scale.h" +#include "vpx_util/vpx_thread.h" + +#include "vp9/common/vp9_alloccommon.h" +#include "vp9/common/vp9_loopfilter.h" +#include "vp9/common/vp9_onyxc_int.h" +#if CONFIG_VP9_POSTPROC +#include "vp9/common/vp9_postproc.h" +#endif +#include "vp9/common/vp9_quant_common.h" +#include "vp9/common/vp9_reconintra.h" + +#include "vp9/decoder/vp9_decodeframe.h" +#include "vp9/decoder/vp9_decoder.h" +#include "vp9/decoder/vp9_detokenize.h" + +static void initialize_dec(void) { + static volatile int init_done = 0; + + if (!init_done) { + vp9_rtcd(); + vpx_dsp_rtcd(); + vpx_scale_rtcd(); + vp9_init_intra_predictors(); + init_done = 1; + } +} + +static void vp9_dec_setup_mi(VP9_COMMON *cm) { + cm->mi = cm->mip + cm->mi_stride + 1; + cm->mi_grid_visible = cm->mi_grid_base + cm->mi_stride + 1; + memset(cm->mi_grid_base, 0, + cm->mi_stride * (cm->mi_rows + 1) * sizeof(*cm->mi_grid_base)); +} + +static int vp9_dec_alloc_mi(VP9_COMMON *cm, int mi_size) { + cm->mip = vpx_calloc(mi_size, sizeof(*cm->mip)); + if (!cm->mip) + return 1; + cm->mi_alloc_size = mi_size; + cm->mi_grid_base = (MODE_INFO **)vpx_calloc(mi_size, sizeof(MODE_INFO*)); + if (!cm->mi_grid_base) + return 1; + return 0; +} + +static void vp9_dec_free_mi(VP9_COMMON *cm) { + vpx_free(cm->mip); + cm->mip = NULL; + vpx_free(cm->mi_grid_base); + cm->mi_grid_base = NULL; +} + +VP9Decoder *vp9_decoder_create(BufferPool *const pool) { + VP9Decoder *volatile const pbi = vpx_memalign(32, sizeof(*pbi)); + VP9_COMMON *volatile const cm = pbi ? &pbi->common : NULL; + + if (!cm) + return NULL; + + vp9_zero(*pbi); + + if (setjmp(cm->error.jmp)) { + cm->error.setjmp = 0; + vp9_decoder_remove(pbi); + return NULL; + } + + cm->error.setjmp = 1; + + CHECK_MEM_ERROR(cm, cm->fc, + (FRAME_CONTEXT *)vpx_calloc(1, sizeof(*cm->fc))); + CHECK_MEM_ERROR(cm, cm->frame_contexts, + (FRAME_CONTEXT *)vpx_calloc(FRAME_CONTEXTS, + sizeof(*cm->frame_contexts))); + + pbi->need_resync = 1; + once(initialize_dec); + + // Initialize the references to not point to any frame buffers. + memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map)); + memset(&cm->next_ref_frame_map, -1, sizeof(cm->next_ref_frame_map)); + + cm->current_video_frame = 0; + pbi->ready_for_new_data = 1; + pbi->common.buffer_pool = pool; + + cm->bit_depth = VPX_BITS_8; + cm->dequant_bit_depth = VPX_BITS_8; + + cm->alloc_mi = vp9_dec_alloc_mi; + cm->free_mi = vp9_dec_free_mi; + cm->setup_mi = vp9_dec_setup_mi; + + vp9_loop_filter_init(cm); + + cm->error.setjmp = 0; + + vpx_get_worker_interface()->init(&pbi->lf_worker); + + return pbi; +} + +void vp9_decoder_remove(VP9Decoder *pbi) { + int i; + + if (!pbi) + return; + + vpx_get_worker_interface()->end(&pbi->lf_worker); + vpx_free(pbi->lf_worker.data1); + + for (i = 0; i < pbi->num_tile_workers; ++i) { + VPxWorker *const worker = &pbi->tile_workers[i]; + vpx_get_worker_interface()->end(worker); + } + + vpx_free(pbi->tile_worker_data); + vpx_free(pbi->tile_workers); + + if (pbi->num_tile_workers > 0) { + vp9_loop_filter_dealloc(&pbi->lf_row_sync); + } + + vpx_free(pbi); +} + +static int equal_dimensions(const YV12_BUFFER_CONFIG *a, + const YV12_BUFFER_CONFIG *b) { + return a->y_height == b->y_height && a->y_width == b->y_width && + a->uv_height == b->uv_height && a->uv_width == b->uv_width; +} + +vpx_codec_err_t vp9_copy_reference_dec(VP9Decoder *pbi, + VP9_REFFRAME ref_frame_flag, + YV12_BUFFER_CONFIG *sd) { + VP9_COMMON *cm = &pbi->common; + + /* TODO(jkoleszar): The decoder doesn't have any real knowledge of what the + * encoder is using the frame buffers for. This is just a stub to keep the + * vpxenc --test-decode functionality working, and will be replaced in a + * later commit that adds VP9-specific controls for this functionality. + */ + if (ref_frame_flag == VP9_LAST_FLAG) { + const YV12_BUFFER_CONFIG *const cfg = get_ref_frame(cm, 0); + if (cfg == NULL) { + vpx_internal_error(&cm->error, VPX_CODEC_ERROR, + "No 'last' reference frame"); + return VPX_CODEC_ERROR; + } + if (!equal_dimensions(cfg, sd)) + vpx_internal_error(&cm->error, VPX_CODEC_ERROR, + "Incorrect buffer dimensions"); + else + vp8_yv12_copy_frame(cfg, sd); + } else { + vpx_internal_error(&cm->error, VPX_CODEC_ERROR, + "Invalid reference frame"); + } + + return cm->error.error_code; +} + + +vpx_codec_err_t vp9_set_reference_dec(VP9_COMMON *cm, + VP9_REFFRAME ref_frame_flag, + YV12_BUFFER_CONFIG *sd) { + RefBuffer *ref_buf = NULL; + RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs; + + // TODO(jkoleszar): The decoder doesn't have any real knowledge of what the + // encoder is using the frame buffers for. This is just a stub to keep the + // vpxenc --test-decode functionality working, and will be replaced in a + // later commit that adds VP9-specific controls for this functionality. + if (ref_frame_flag == VP9_LAST_FLAG) { + ref_buf = &cm->frame_refs[0]; + } else if (ref_frame_flag == VP9_GOLD_FLAG) { + ref_buf = &cm->frame_refs[1]; + } else if (ref_frame_flag == VP9_ALT_FLAG) { + ref_buf = &cm->frame_refs[2]; + } else { + vpx_internal_error(&cm->error, VPX_CODEC_ERROR, + "Invalid reference frame"); + return cm->error.error_code; + } + + if (!equal_dimensions(ref_buf->buf, sd)) { + vpx_internal_error(&cm->error, VPX_CODEC_ERROR, + "Incorrect buffer dimensions"); + } else { + int *ref_fb_ptr = &ref_buf->idx; + + // Find an empty frame buffer. + const int free_fb = get_free_fb(cm); + if (cm->new_fb_idx == INVALID_IDX) { + vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR, + "Unable to find free frame buffer"); + return cm->error.error_code; + } + + // Decrease ref_count since it will be increased again in + // ref_cnt_fb() below. + --frame_bufs[free_fb].ref_count; + + // Manage the reference counters and copy image. + ref_cnt_fb(frame_bufs, ref_fb_ptr, free_fb); + ref_buf->buf = &frame_bufs[*ref_fb_ptr].buf; + vp8_yv12_copy_frame(sd, ref_buf->buf); + } + + return cm->error.error_code; +} + +/* If any buffer updating is signaled it should be done here. */ +static void swap_frame_buffers(VP9Decoder *pbi) { + int ref_index = 0, mask; + VP9_COMMON *const cm = &pbi->common; + BufferPool *const pool = cm->buffer_pool; + RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs; + + lock_buffer_pool(pool); + for (mask = pbi->refresh_frame_flags; mask; mask >>= 1) { + const int old_idx = cm->ref_frame_map[ref_index]; + // Current thread releases the holding of reference frame. + decrease_ref_count(old_idx, frame_bufs, pool); + + // Release the reference frame in reference map. + if (mask & 1) { + decrease_ref_count(old_idx, frame_bufs, pool); + } + cm->ref_frame_map[ref_index] = cm->next_ref_frame_map[ref_index]; + ++ref_index; + } + + // Current thread releases the holding of reference frame. + for (; ref_index < REF_FRAMES && !cm->show_existing_frame; ++ref_index) { + const int old_idx = cm->ref_frame_map[ref_index]; + decrease_ref_count(old_idx, frame_bufs, pool); + cm->ref_frame_map[ref_index] = cm->next_ref_frame_map[ref_index]; + } + unlock_buffer_pool(pool); + pbi->hold_ref_buf = 0; + cm->frame_to_show = get_frame_new_buffer(cm); + + if (!pbi->frame_parallel_decode || !cm->show_frame) { + lock_buffer_pool(pool); + --frame_bufs[cm->new_fb_idx].ref_count; + unlock_buffer_pool(pool); + } + + // Invalidate these references until the next frame starts. + for (ref_index = 0; ref_index < 3; ref_index++) + cm->frame_refs[ref_index].idx = -1; +} + +int vp9_receive_compressed_data(VP9Decoder *pbi, + size_t size, const uint8_t **psource) { + VP9_COMMON *volatile const cm = &pbi->common; + BufferPool *volatile const pool = cm->buffer_pool; + RefCntBuffer *volatile const frame_bufs = cm->buffer_pool->frame_bufs; + const uint8_t *source = *psource; + int retcode = 0; + cm->error.error_code = VPX_CODEC_OK; + + if (size == 0) { + // This is used to signal that we are missing frames. + // We do not know if the missing frame(s) was supposed to update + // any of the reference buffers, but we act conservative and + // mark only the last buffer as corrupted. + // + // TODO(jkoleszar): Error concealment is undefined and non-normative + // at this point, but if it becomes so, [0] may not always be the correct + // thing to do here. + if (cm->frame_refs[0].idx > 0) { + assert(cm->frame_refs[0].buf != NULL); + cm->frame_refs[0].buf->corrupted = 1; + } + } + + pbi->ready_for_new_data = 0; + + // Check if the previous frame was a frame without any references to it. + // Release frame buffer if not decoding in frame parallel mode. + if (!pbi->frame_parallel_decode && cm->new_fb_idx >= 0 + && frame_bufs[cm->new_fb_idx].ref_count == 0) + pool->release_fb_cb(pool->cb_priv, + &frame_bufs[cm->new_fb_idx].raw_frame_buffer); + // Find a free frame buffer. Return error if can not find any. + cm->new_fb_idx = get_free_fb(cm); + if (cm->new_fb_idx == INVALID_IDX) { + vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR, + "Unable to find free frame buffer"); + return cm->error.error_code; + } + + // Assign a MV array to the frame buffer. + cm->cur_frame = &pool->frame_bufs[cm->new_fb_idx]; + + pbi->hold_ref_buf = 0; + if (pbi->frame_parallel_decode) { + VPxWorker *const worker = pbi->frame_worker_owner; + vp9_frameworker_lock_stats(worker); + frame_bufs[cm->new_fb_idx].frame_worker_owner = worker; + // Reset decoding progress. + pbi->cur_buf = &frame_bufs[cm->new_fb_idx]; + pbi->cur_buf->row = -1; + pbi->cur_buf->col = -1; + vp9_frameworker_unlock_stats(worker); + } else { + pbi->cur_buf = &frame_bufs[cm->new_fb_idx]; + } + + + if (setjmp(cm->error.jmp)) { + const VPxWorkerInterface *const winterface = vpx_get_worker_interface(); + int i; + + cm->error.setjmp = 0; + pbi->ready_for_new_data = 1; + + // Synchronize all threads immediately as a subsequent decode call may + // cause a resize invalidating some allocations. + winterface->sync(&pbi->lf_worker); + for (i = 0; i < pbi->num_tile_workers; ++i) { + winterface->sync(&pbi->tile_workers[i]); + } + + lock_buffer_pool(pool); + // Release all the reference buffers if worker thread is holding them. + if (pbi->hold_ref_buf == 1) { + int ref_index = 0, mask; + for (mask = pbi->refresh_frame_flags; mask; mask >>= 1) { + const int old_idx = cm->ref_frame_map[ref_index]; + // Current thread releases the holding of reference frame. + decrease_ref_count(old_idx, frame_bufs, pool); + + // Release the reference frame in reference map. + if (mask & 1) { + decrease_ref_count(old_idx, frame_bufs, pool); + } + ++ref_index; + } + + // Current thread releases the holding of reference frame. + for (; ref_index < REF_FRAMES && !cm->show_existing_frame; ++ref_index) { + const int old_idx = cm->ref_frame_map[ref_index]; + decrease_ref_count(old_idx, frame_bufs, pool); + } + pbi->hold_ref_buf = 0; + } + // Release current frame. + decrease_ref_count(cm->new_fb_idx, frame_bufs, pool); + unlock_buffer_pool(pool); + + vpx_clear_system_state(); + return -1; + } + + cm->error.setjmp = 1; + vp9_decode_frame(pbi, source, source + size, psource); + + swap_frame_buffers(pbi); + + vpx_clear_system_state(); + + if (!cm->show_existing_frame) { + cm->last_show_frame = cm->show_frame; + cm->prev_frame = cm->cur_frame; + if (cm->seg.enabled && !pbi->frame_parallel_decode) + vp9_swap_current_and_last_seg_map(cm); + } + + // Update progress in frame parallel decode. + if (pbi->frame_parallel_decode) { + // Need to lock the mutex here as another thread may + // be accessing this buffer. + VPxWorker *const worker = pbi->frame_worker_owner; + FrameWorkerData *const frame_worker_data = worker->data1; + vp9_frameworker_lock_stats(worker); + + if (cm->show_frame) { + cm->current_video_frame++; + } + frame_worker_data->frame_decoded = 1; + frame_worker_data->frame_context_ready = 1; + vp9_frameworker_signal_stats(worker); + vp9_frameworker_unlock_stats(worker); + } else { + cm->last_width = cm->width; + cm->last_height = cm->height; + if (cm->show_frame) { + cm->current_video_frame++; + } + } + + cm->error.setjmp = 0; + return retcode; +} + +int vp9_get_raw_frame(VP9Decoder *pbi, YV12_BUFFER_CONFIG *sd, + vp9_ppflags_t *flags) { + VP9_COMMON *const cm = &pbi->common; + int ret = -1; +#if !CONFIG_VP9_POSTPROC + (void)*flags; +#endif + + if (pbi->ready_for_new_data == 1) + return ret; + + pbi->ready_for_new_data = 1; + + /* no raw frame to show!!! */ + if (!cm->show_frame) + return ret; + + pbi->ready_for_new_data = 1; + +#if CONFIG_VP9_POSTPROC + if (!cm->show_existing_frame) { + ret = vp9_post_proc_frame(cm, sd, flags); + } else { + *sd = *cm->frame_to_show; + ret = 0; + } +#else + *sd = *cm->frame_to_show; + ret = 0; +#endif /*!CONFIG_POSTPROC*/ + vpx_clear_system_state(); + return ret; +} + +vpx_codec_err_t vp9_parse_superframe_index(const uint8_t *data, + size_t data_sz, + uint32_t sizes[8], int *count, + vpx_decrypt_cb decrypt_cb, + void *decrypt_state) { + // A chunk ending with a byte matching 0xc0 is an invalid chunk unless + // it is a super frame index. If the last byte of real video compression + // data is 0xc0 the encoder must add a 0 byte. If we have the marker but + // not the associated matching marker byte at the front of the index we have + // an invalid bitstream and need to return an error. + + uint8_t marker; + + assert(data_sz); + marker = read_marker(decrypt_cb, decrypt_state, data + data_sz - 1); + *count = 0; + + if ((marker & 0xe0) == 0xc0) { + const uint32_t frames = (marker & 0x7) + 1; + const uint32_t mag = ((marker >> 3) & 0x3) + 1; + const size_t index_sz = 2 + mag * frames; + + // This chunk is marked as having a superframe index but doesn't have + // enough data for it, thus it's an invalid superframe index. + if (data_sz < index_sz) + return VPX_CODEC_CORRUPT_FRAME; + + { + const uint8_t marker2 = read_marker(decrypt_cb, decrypt_state, + data + data_sz - index_sz); + + // This chunk is marked as having a superframe index but doesn't have + // the matching marker byte at the front of the index therefore it's an + // invalid chunk. + if (marker != marker2) + return VPX_CODEC_CORRUPT_FRAME; + } + + { + // Found a valid superframe index. + uint32_t i, j; + const uint8_t *x = &data[data_sz - index_sz + 1]; + + // Frames has a maximum of 8 and mag has a maximum of 4. + uint8_t clear_buffer[32]; + assert(sizeof(clear_buffer) >= frames * mag); + if (decrypt_cb) { + decrypt_cb(decrypt_state, x, clear_buffer, frames * mag); + x = clear_buffer; + } + + for (i = 0; i < frames; ++i) { + uint32_t this_sz = 0; + + for (j = 0; j < mag; ++j) + this_sz |= ((uint32_t)(*x++)) << (j * 8); + sizes[i] = this_sz; + } + *count = frames; + } + } + return VPX_CODEC_OK; +} diff --git a/thirdparty/libvpx/vp9/decoder/vp9_decoder.h b/thirdparty/libvpx/vp9/decoder/vp9_decoder.h new file mode 100644 index 0000000000..7111a36d37 --- /dev/null +++ b/thirdparty/libvpx/vp9/decoder/vp9_decoder.h @@ -0,0 +1,137 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_DECODER_VP9_DECODER_H_ +#define VP9_DECODER_VP9_DECODER_H_ + +#include "./vpx_config.h" + +#include "vpx/vpx_codec.h" +#include "vpx_dsp/bitreader.h" +#include "vpx_scale/yv12config.h" +#include "vpx_util/vpx_thread.h" + +#include "vp9/common/vp9_thread_common.h" +#include "vp9/common/vp9_onyxc_int.h" +#include "vp9/common/vp9_ppflags.h" +#include "vp9/decoder/vp9_dthread.h" + +#ifdef __cplusplus +extern "C" { +#endif + +typedef struct TileBuffer { + const uint8_t *data; + size_t size; + int col; // only used with multi-threaded decoding +} TileBuffer; + +typedef struct TileWorkerData { + const uint8_t *data_end; + int buf_start, buf_end; // pbi->tile_buffers to decode, inclusive + vpx_reader bit_reader; + FRAME_COUNTS counts; + DECLARE_ALIGNED(16, MACROBLOCKD, xd); + /* dqcoeff are shared by all the planes. So planes must be decoded serially */ + DECLARE_ALIGNED(16, tran_low_t, dqcoeff[32 * 32]); + struct vpx_internal_error_info error_info; +} TileWorkerData; + +typedef struct VP9Decoder { + DECLARE_ALIGNED(16, MACROBLOCKD, mb); + + DECLARE_ALIGNED(16, VP9_COMMON, common); + + int ready_for_new_data; + + int refresh_frame_flags; + + int frame_parallel_decode; // frame-based threading. + + // TODO(hkuang): Combine this with cur_buf in macroblockd as they are + // the same. + RefCntBuffer *cur_buf; // Current decoding frame buffer. + + VPxWorker *frame_worker_owner; // frame_worker that owns this pbi. + VPxWorker lf_worker; + VPxWorker *tile_workers; + TileWorkerData *tile_worker_data; + TileBuffer tile_buffers[64]; + int num_tile_workers; + int total_tiles; + + VP9LfSync lf_row_sync; + + vpx_decrypt_cb decrypt_cb; + void *decrypt_state; + + int max_threads; + int inv_tile_order; + int need_resync; // wait for key/intra-only frame. + int hold_ref_buf; // hold the reference buffer. +} VP9Decoder; + +int vp9_receive_compressed_data(struct VP9Decoder *pbi, + size_t size, const uint8_t **dest); + +int vp9_get_raw_frame(struct VP9Decoder *pbi, YV12_BUFFER_CONFIG *sd, + vp9_ppflags_t *flags); + +vpx_codec_err_t vp9_copy_reference_dec(struct VP9Decoder *pbi, + VP9_REFFRAME ref_frame_flag, + YV12_BUFFER_CONFIG *sd); + +vpx_codec_err_t vp9_set_reference_dec(VP9_COMMON *cm, + VP9_REFFRAME ref_frame_flag, + YV12_BUFFER_CONFIG *sd); + +static INLINE uint8_t read_marker(vpx_decrypt_cb decrypt_cb, + void *decrypt_state, + const uint8_t *data) { + if (decrypt_cb) { + uint8_t marker; + decrypt_cb(decrypt_state, data, &marker, 1); + return marker; + } + return *data; +} + +// This function is exposed for use in tests, as well as the inlined function +// "read_marker". +vpx_codec_err_t vp9_parse_superframe_index(const uint8_t *data, + size_t data_sz, + uint32_t sizes[8], int *count, + vpx_decrypt_cb decrypt_cb, + void *decrypt_state); + +struct VP9Decoder *vp9_decoder_create(BufferPool *const pool); + +void vp9_decoder_remove(struct VP9Decoder *pbi); + +static INLINE void decrease_ref_count(int idx, RefCntBuffer *const frame_bufs, + BufferPool *const pool) { + if (idx >= 0 && frame_bufs[idx].ref_count > 0) { + --frame_bufs[idx].ref_count; + // A worker may only get a free framebuffer index when calling get_free_fb. + // But the private buffer is not set up until finish decoding header. + // So any error happens during decoding header, the frame_bufs will not + // have valid priv buffer. + if (frame_bufs[idx].ref_count == 0 && + frame_bufs[idx].raw_frame_buffer.priv) { + pool->release_fb_cb(pool->cb_priv, &frame_bufs[idx].raw_frame_buffer); + } + } +} + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_DECODER_VP9_DECODER_H_ diff --git a/thirdparty/libvpx/vp9/decoder/vp9_detokenize.c b/thirdparty/libvpx/vp9/decoder/vp9_detokenize.c new file mode 100644 index 0000000000..47dc107fe2 --- /dev/null +++ b/thirdparty/libvpx/vp9/decoder/vp9_detokenize.c @@ -0,0 +1,224 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "vpx_mem/vpx_mem.h" +#include "vpx_ports/mem.h" + +#include "vp9/common/vp9_blockd.h" +#include "vp9/common/vp9_common.h" +#include "vp9/common/vp9_entropy.h" +#if CONFIG_COEFFICIENT_RANGE_CHECKING +#include "vp9/common/vp9_idct.h" +#endif + +#include "vp9/decoder/vp9_detokenize.h" + +#define EOB_CONTEXT_NODE 0 +#define ZERO_CONTEXT_NODE 1 +#define ONE_CONTEXT_NODE 2 + +#define INCREMENT_COUNT(token) \ + do { \ + if (counts) \ + ++coef_counts[band][ctx][token]; \ + } while (0) + +static INLINE int read_coeff(const vpx_prob *probs, int n, vpx_reader *r) { + int i, val = 0; + for (i = 0; i < n; ++i) + val = (val << 1) | vpx_read(r, probs[i]); + return val; +} + +static int decode_coefs(const MACROBLOCKD *xd, + PLANE_TYPE type, + tran_low_t *dqcoeff, TX_SIZE tx_size, const int16_t *dq, + int ctx, const int16_t *scan, const int16_t *nb, + vpx_reader *r) { + FRAME_COUNTS *counts = xd->counts; + const int max_eob = 16 << (tx_size << 1); + const FRAME_CONTEXT *const fc = xd->fc; + const int ref = is_inter_block(xd->mi[0]); + int band, c = 0; + const vpx_prob (*coef_probs)[COEFF_CONTEXTS][UNCONSTRAINED_NODES] = + fc->coef_probs[tx_size][type][ref]; + const vpx_prob *prob; + unsigned int (*coef_counts)[COEFF_CONTEXTS][UNCONSTRAINED_NODES + 1]; + unsigned int (*eob_branch_count)[COEFF_CONTEXTS]; + uint8_t token_cache[32 * 32]; + const uint8_t *band_translate = get_band_translate(tx_size); + const int dq_shift = (tx_size == TX_32X32); + int v, token; + int16_t dqv = dq[0]; + const uint8_t *const cat6_prob = +#if CONFIG_VP9_HIGHBITDEPTH + (xd->bd == VPX_BITS_12) ? vp9_cat6_prob_high12 : + (xd->bd == VPX_BITS_10) ? vp9_cat6_prob_high12 + 2 : +#endif // CONFIG_VP9_HIGHBITDEPTH + vp9_cat6_prob; + const int cat6_bits = +#if CONFIG_VP9_HIGHBITDEPTH + (xd->bd == VPX_BITS_12) ? 18 : + (xd->bd == VPX_BITS_10) ? 16 : +#endif // CONFIG_VP9_HIGHBITDEPTH + 14; + + if (counts) { + coef_counts = counts->coef[tx_size][type][ref]; + eob_branch_count = counts->eob_branch[tx_size][type][ref]; + } + + while (c < max_eob) { + int val = -1; + band = *band_translate++; + prob = coef_probs[band][ctx]; + if (counts) + ++eob_branch_count[band][ctx]; + if (!vpx_read(r, prob[EOB_CONTEXT_NODE])) { + INCREMENT_COUNT(EOB_MODEL_TOKEN); + break; + } + + while (!vpx_read(r, prob[ZERO_CONTEXT_NODE])) { + INCREMENT_COUNT(ZERO_TOKEN); + dqv = dq[1]; + token_cache[scan[c]] = 0; + ++c; + if (c >= max_eob) + return c; // zero tokens at the end (no eob token) + ctx = get_coef_context(nb, token_cache, c); + band = *band_translate++; + prob = coef_probs[band][ctx]; + } + + if (!vpx_read(r, prob[ONE_CONTEXT_NODE])) { + INCREMENT_COUNT(ONE_TOKEN); + token = ONE_TOKEN; + val = 1; + } else { + INCREMENT_COUNT(TWO_TOKEN); + token = vpx_read_tree(r, vp9_coef_con_tree, + vp9_pareto8_full[prob[PIVOT_NODE] - 1]); + switch (token) { + case TWO_TOKEN: + case THREE_TOKEN: + case FOUR_TOKEN: + val = token; + break; + case CATEGORY1_TOKEN: + val = CAT1_MIN_VAL + read_coeff(vp9_cat1_prob, 1, r); + break; + case CATEGORY2_TOKEN: + val = CAT2_MIN_VAL + read_coeff(vp9_cat2_prob, 2, r); + break; + case CATEGORY3_TOKEN: + val = CAT3_MIN_VAL + read_coeff(vp9_cat3_prob, 3, r); + break; + case CATEGORY4_TOKEN: + val = CAT4_MIN_VAL + read_coeff(vp9_cat4_prob, 4, r); + break; + case CATEGORY5_TOKEN: + val = CAT5_MIN_VAL + read_coeff(vp9_cat5_prob, 5, r); + break; + case CATEGORY6_TOKEN: + val = CAT6_MIN_VAL + read_coeff(cat6_prob, cat6_bits, r); + break; + } + } + v = (val * dqv) >> dq_shift; +#if CONFIG_COEFFICIENT_RANGE_CHECKING +#if CONFIG_VP9_HIGHBITDEPTH + dqcoeff[scan[c]] = highbd_check_range((vpx_read_bit(r) ? -v : v), + xd->bd); +#else + dqcoeff[scan[c]] = check_range(vpx_read_bit(r) ? -v : v); +#endif // CONFIG_VP9_HIGHBITDEPTH +#else + dqcoeff[scan[c]] = vpx_read_bit(r) ? -v : v; +#endif // CONFIG_COEFFICIENT_RANGE_CHECKING + token_cache[scan[c]] = vp9_pt_energy_class[token]; + ++c; + ctx = get_coef_context(nb, token_cache, c); + dqv = dq[1]; + } + + return c; +} + +static void get_ctx_shift(MACROBLOCKD *xd, int *ctx_shift_a, int *ctx_shift_l, + int x, int y, unsigned int tx_size_in_blocks) { + if (xd->max_blocks_wide) { + if (tx_size_in_blocks + x > xd->max_blocks_wide) + *ctx_shift_a = (tx_size_in_blocks - (xd->max_blocks_wide - x)) * 8; + } + if (xd->max_blocks_high) { + if (tx_size_in_blocks + y > xd->max_blocks_high) + *ctx_shift_l = (tx_size_in_blocks - (xd->max_blocks_high - y)) * 8; + } +} + +int vp9_decode_block_tokens(MACROBLOCKD *xd, int plane, const scan_order *sc, + int x, int y, TX_SIZE tx_size, vpx_reader *r, + int seg_id) { + struct macroblockd_plane *const pd = &xd->plane[plane]; + const int16_t *const dequant = pd->seg_dequant[seg_id]; + int eob; + ENTROPY_CONTEXT *a = pd->above_context + x; + ENTROPY_CONTEXT *l = pd->left_context + y; + int ctx; + int ctx_shift_a = 0; + int ctx_shift_l = 0; + + switch (tx_size) { + case TX_4X4: + ctx = a[0] != 0; + ctx += l[0] != 0; + eob = decode_coefs(xd, get_plane_type(plane), pd->dqcoeff, tx_size, + dequant, ctx, sc->scan, sc->neighbors, r); + a[0] = l[0] = (eob > 0); + break; + case TX_8X8: + get_ctx_shift(xd, &ctx_shift_a, &ctx_shift_l, x, y, 1 << TX_8X8); + ctx = !!*(const uint16_t *)a; + ctx += !!*(const uint16_t *)l; + eob = decode_coefs(xd, get_plane_type(plane), pd->dqcoeff, tx_size, + dequant, ctx, sc->scan, sc->neighbors, r); + *(uint16_t *)a = ((eob > 0) * 0x0101) >> ctx_shift_a; + *(uint16_t *)l = ((eob > 0) * 0x0101) >> ctx_shift_l; + break; + case TX_16X16: + get_ctx_shift(xd, &ctx_shift_a, &ctx_shift_l, x, y, 1 << TX_16X16); + ctx = !!*(const uint32_t *)a; + ctx += !!*(const uint32_t *)l; + eob = decode_coefs(xd, get_plane_type(plane), pd->dqcoeff, tx_size, + dequant, ctx, sc->scan, sc->neighbors, r); + *(uint32_t *)a = ((eob > 0) * 0x01010101) >> ctx_shift_a; + *(uint32_t *)l = ((eob > 0) * 0x01010101) >> ctx_shift_l; + break; + case TX_32X32: + get_ctx_shift(xd, &ctx_shift_a, &ctx_shift_l, x, y, 1 << TX_32X32); + // NOTE: casting to uint64_t here is safe because the default memory + // alignment is at least 8 bytes and the TX_32X32 is aligned on 8 byte + // boundaries. + ctx = !!*(const uint64_t *)a; + ctx += !!*(const uint64_t *)l; + eob = decode_coefs(xd, get_plane_type(plane), pd->dqcoeff, tx_size, + dequant, ctx, sc->scan, sc->neighbors, r); + *(uint64_t *)a = ((eob > 0) * 0x0101010101010101ULL) >> ctx_shift_a; + *(uint64_t *)l = ((eob > 0) * 0x0101010101010101ULL) >> ctx_shift_l; + break; + default: + assert(0 && "Invalid transform size."); + eob = 0; + break; + } + + return eob; +} diff --git a/thirdparty/libvpx/vp9/decoder/vp9_detokenize.h b/thirdparty/libvpx/vp9/decoder/vp9_detokenize.h new file mode 100644 index 0000000000..d242d4466e --- /dev/null +++ b/thirdparty/libvpx/vp9/decoder/vp9_detokenize.h @@ -0,0 +1,33 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP9_DECODER_VP9_DETOKENIZE_H_ +#define VP9_DECODER_VP9_DETOKENIZE_H_ + +#include "vpx_dsp/bitreader.h" +#include "vp9/decoder/vp9_decoder.h" +#include "vp9/common/vp9_scan.h" + +#ifdef __cplusplus +extern "C" { +#endif + +int vp9_decode_block_tokens(MACROBLOCKD *xd, + int plane, const scan_order *sc, + int x, int y, + TX_SIZE tx_size, vpx_reader *r, + int seg_id); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_DECODER_VP9_DETOKENIZE_H_ diff --git a/thirdparty/libvpx/vp9/decoder/vp9_dsubexp.c b/thirdparty/libvpx/vp9/decoder/vp9_dsubexp.c new file mode 100644 index 0000000000..05b38538ae --- /dev/null +++ b/thirdparty/libvpx/vp9/decoder/vp9_dsubexp.c @@ -0,0 +1,76 @@ +/* + Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <assert.h> + +#include "vp9/common/vp9_entropy.h" + +#include "vp9/decoder/vp9_dsubexp.h" + +static int inv_recenter_nonneg(int v, int m) { + if (v > 2 * m) + return v; + + return (v & 1) ? m - ((v + 1) >> 1) : m + (v >> 1); +} + +static int decode_uniform(vpx_reader *r) { + const int l = 8; + const int m = (1 << l) - 191; + const int v = vpx_read_literal(r, l - 1); + return v < m ? v : (v << 1) - m + vpx_read_bit(r); +} + +static int inv_remap_prob(int v, int m) { + static uint8_t inv_map_table[MAX_PROB] = { + 7, 20, 33, 46, 59, 72, 85, 98, 111, 124, 137, 150, 163, 176, 189, + 202, 215, 228, 241, 254, 1, 2, 3, 4, 5, 6, 8, 9, 10, 11, + 12, 13, 14, 15, 16, 17, 18, 19, 21, 22, 23, 24, 25, 26, 27, + 28, 29, 30, 31, 32, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, + 44, 45, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 60, + 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 73, 74, 75, 76, + 77, 78, 79, 80, 81, 82, 83, 84, 86, 87, 88, 89, 90, 91, 92, + 93, 94, 95, 96, 97, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, + 109, 110, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 123, 125, + 126, 127, 128, 129, 130, 131, 132, 133, 134, 135, 136, 138, 139, 140, 141, + 142, 143, 144, 145, 146, 147, 148, 149, 151, 152, 153, 154, 155, 156, 157, + 158, 159, 160, 161, 162, 164, 165, 166, 167, 168, 169, 170, 171, 172, 173, + 174, 175, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 188, 190, + 191, 192, 193, 194, 195, 196, 197, 198, 199, 200, 201, 203, 204, 205, 206, + 207, 208, 209, 210, 211, 212, 213, 214, 216, 217, 218, 219, 220, 221, 222, + 223, 224, 225, 226, 227, 229, 230, 231, 232, 233, 234, 235, 236, 237, 238, + 239, 240, 242, 243, 244, 245, 246, 247, 248, 249, 250, 251, 252, 253, 253 + }; + assert(v < (int)(sizeof(inv_map_table) / sizeof(inv_map_table[0]))); + v = inv_map_table[v]; + m--; + if ((m << 1) <= MAX_PROB) { + return 1 + inv_recenter_nonneg(v, m); + } else { + return MAX_PROB - inv_recenter_nonneg(v, MAX_PROB - 1 - m); + } +} + +static int decode_term_subexp(vpx_reader *r) { + if (!vpx_read_bit(r)) + return vpx_read_literal(r, 4); + if (!vpx_read_bit(r)) + return vpx_read_literal(r, 4) + 16; + if (!vpx_read_bit(r)) + return vpx_read_literal(r, 5) + 32; + return decode_uniform(r) + 64; +} + +void vp9_diff_update_prob(vpx_reader *r, vpx_prob* p) { + if (vpx_read(r, DIFF_UPDATE_PROB)) { + const int delp = decode_term_subexp(r); + *p = (vpx_prob)inv_remap_prob(delp, *p); + } +} diff --git a/thirdparty/libvpx/vp9/decoder/vp9_dsubexp.h b/thirdparty/libvpx/vp9/decoder/vp9_dsubexp.h new file mode 100644 index 0000000000..a8bcc70be9 --- /dev/null +++ b/thirdparty/libvpx/vp9/decoder/vp9_dsubexp.h @@ -0,0 +1,27 @@ +/* + * Copyright (c) 2013 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VP9_DECODER_VP9_DSUBEXP_H_ +#define VP9_DECODER_VP9_DSUBEXP_H_ + +#include "vpx_dsp/bitreader.h" + +#ifdef __cplusplus +extern "C" { +#endif + +void vp9_diff_update_prob(vpx_reader *r, vpx_prob* p); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_DECODER_VP9_DSUBEXP_H_ diff --git a/thirdparty/libvpx/vp9/decoder/vp9_dthread.c b/thirdparty/libvpx/vp9/decoder/vp9_dthread.c new file mode 100644 index 0000000000..14a71448fe --- /dev/null +++ b/thirdparty/libvpx/vp9/decoder/vp9_dthread.c @@ -0,0 +1,189 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "./vpx_config.h" +#include "vpx_mem/vpx_mem.h" +#include "vp9/common/vp9_reconinter.h" +#include "vp9/decoder/vp9_dthread.h" +#include "vp9/decoder/vp9_decoder.h" + +// #define DEBUG_THREAD + +// TODO(hkuang): Clean up all the #ifdef in this file. +void vp9_frameworker_lock_stats(VPxWorker *const worker) { +#if CONFIG_MULTITHREAD + FrameWorkerData *const worker_data = worker->data1; + pthread_mutex_lock(&worker_data->stats_mutex); +#else + (void)worker; +#endif +} + +void vp9_frameworker_unlock_stats(VPxWorker *const worker) { +#if CONFIG_MULTITHREAD + FrameWorkerData *const worker_data = worker->data1; + pthread_mutex_unlock(&worker_data->stats_mutex); +#else + (void)worker; +#endif +} + +void vp9_frameworker_signal_stats(VPxWorker *const worker) { +#if CONFIG_MULTITHREAD + FrameWorkerData *const worker_data = worker->data1; + +// TODO(hkuang): Fix the pthread_cond_broadcast in windows wrapper. +#if defined(_WIN32) && !HAVE_PTHREAD_H + pthread_cond_signal(&worker_data->stats_cond); +#else + pthread_cond_broadcast(&worker_data->stats_cond); +#endif + +#else + (void)worker; +#endif +} + +// This macro prevents thread_sanitizer from reporting known concurrent writes. +#if defined(__has_feature) +#if __has_feature(thread_sanitizer) +#define BUILDING_WITH_TSAN +#endif +#endif + +// TODO(hkuang): Remove worker parameter as it is only used in debug code. +void vp9_frameworker_wait(VPxWorker *const worker, RefCntBuffer *const ref_buf, + int row) { +#if CONFIG_MULTITHREAD + if (!ref_buf) + return; + +#ifndef BUILDING_WITH_TSAN + // The following line of code will get harmless tsan error but it is the key + // to get best performance. + if (ref_buf->row >= row && ref_buf->buf.corrupted != 1) return; +#endif + + { + // Find the worker thread that owns the reference frame. If the reference + // frame has been fully decoded, it may not have owner. + VPxWorker *const ref_worker = ref_buf->frame_worker_owner; + FrameWorkerData *const ref_worker_data = + (FrameWorkerData *)ref_worker->data1; + const VP9Decoder *const pbi = ref_worker_data->pbi; + +#ifdef DEBUG_THREAD + { + FrameWorkerData *const worker_data = (FrameWorkerData *)worker->data1; + printf("%d %p worker is waiting for %d %p worker (%d) ref %d \r\n", + worker_data->worker_id, worker, ref_worker_data->worker_id, + ref_buf->frame_worker_owner, row, ref_buf->row); + } +#endif + + vp9_frameworker_lock_stats(ref_worker); + while (ref_buf->row < row && pbi->cur_buf == ref_buf && + ref_buf->buf.corrupted != 1) { + pthread_cond_wait(&ref_worker_data->stats_cond, + &ref_worker_data->stats_mutex); + } + + if (ref_buf->buf.corrupted == 1) { + FrameWorkerData *const worker_data = (FrameWorkerData *)worker->data1; + vp9_frameworker_unlock_stats(ref_worker); + vpx_internal_error(&worker_data->pbi->common.error, + VPX_CODEC_CORRUPT_FRAME, + "Worker %p failed to decode frame", worker); + } + vp9_frameworker_unlock_stats(ref_worker); + } +#else + (void)worker; + (void)ref_buf; + (void)row; + (void)ref_buf; +#endif // CONFIG_MULTITHREAD +} + +void vp9_frameworker_broadcast(RefCntBuffer *const buf, int row) { +#if CONFIG_MULTITHREAD + VPxWorker *worker = buf->frame_worker_owner; + +#ifdef DEBUG_THREAD + { + FrameWorkerData *const worker_data = (FrameWorkerData *)worker->data1; + printf("%d %p worker decode to (%d) \r\n", worker_data->worker_id, + buf->frame_worker_owner, row); + } +#endif + + vp9_frameworker_lock_stats(worker); + buf->row = row; + vp9_frameworker_signal_stats(worker); + vp9_frameworker_unlock_stats(worker); +#else + (void)buf; + (void)row; +#endif // CONFIG_MULTITHREAD +} + +void vp9_frameworker_copy_context(VPxWorker *const dst_worker, + VPxWorker *const src_worker) { +#if CONFIG_MULTITHREAD + FrameWorkerData *const src_worker_data = (FrameWorkerData *)src_worker->data1; + FrameWorkerData *const dst_worker_data = (FrameWorkerData *)dst_worker->data1; + VP9_COMMON *const src_cm = &src_worker_data->pbi->common; + VP9_COMMON *const dst_cm = &dst_worker_data->pbi->common; + int i; + + // Wait until source frame's context is ready. + vp9_frameworker_lock_stats(src_worker); + while (!src_worker_data->frame_context_ready) { + pthread_cond_wait(&src_worker_data->stats_cond, + &src_worker_data->stats_mutex); + } + + dst_cm->last_frame_seg_map = src_cm->seg.enabled ? + src_cm->current_frame_seg_map : src_cm->last_frame_seg_map; + dst_worker_data->pbi->need_resync = src_worker_data->pbi->need_resync; + vp9_frameworker_unlock_stats(src_worker); + + dst_cm->bit_depth = src_cm->bit_depth; +#if CONFIG_VP9_HIGHBITDEPTH + dst_cm->use_highbitdepth = src_cm->use_highbitdepth; +#endif + dst_cm->prev_frame = src_cm->show_existing_frame ? + src_cm->prev_frame : src_cm->cur_frame; + dst_cm->last_width = !src_cm->show_existing_frame ? + src_cm->width : src_cm->last_width; + dst_cm->last_height = !src_cm->show_existing_frame ? + src_cm->height : src_cm->last_height; + dst_cm->subsampling_x = src_cm->subsampling_x; + dst_cm->subsampling_y = src_cm->subsampling_y; + dst_cm->frame_type = src_cm->frame_type; + dst_cm->last_show_frame = !src_cm->show_existing_frame ? + src_cm->show_frame : src_cm->last_show_frame; + for (i = 0; i < REF_FRAMES; ++i) + dst_cm->ref_frame_map[i] = src_cm->next_ref_frame_map[i]; + + memcpy(dst_cm->lf_info.lfthr, src_cm->lf_info.lfthr, + (MAX_LOOP_FILTER + 1) * sizeof(loop_filter_thresh)); + dst_cm->lf.last_sharpness_level = src_cm->lf.sharpness_level; + dst_cm->lf.filter_level = src_cm->lf.filter_level; + memcpy(dst_cm->lf.ref_deltas, src_cm->lf.ref_deltas, MAX_REF_LF_DELTAS); + memcpy(dst_cm->lf.mode_deltas, src_cm->lf.mode_deltas, MAX_MODE_LF_DELTAS); + dst_cm->seg = src_cm->seg; + memcpy(dst_cm->frame_contexts, src_cm->frame_contexts, + FRAME_CONTEXTS * sizeof(dst_cm->frame_contexts[0])); +#else + (void) dst_worker; + (void) src_worker; +#endif // CONFIG_MULTITHREAD +} diff --git a/thirdparty/libvpx/vp9/decoder/vp9_dthread.h b/thirdparty/libvpx/vp9/decoder/vp9_dthread.h new file mode 100644 index 0000000000..ba7c38a511 --- /dev/null +++ b/thirdparty/libvpx/vp9/decoder/vp9_dthread.h @@ -0,0 +1,74 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_DECODER_VP9_DTHREAD_H_ +#define VP9_DECODER_VP9_DTHREAD_H_ + +#include "./vpx_config.h" +#include "vpx_util/vpx_thread.h" +#include "vpx/internal/vpx_codec_internal.h" + +#ifdef __cplusplus +extern "C" { +#endif + +struct VP9Common; +struct VP9Decoder; + +// WorkerData for the FrameWorker thread. It contains all the information of +// the worker and decode structures for decoding a frame. +typedef struct FrameWorkerData { + struct VP9Decoder *pbi; + const uint8_t *data; + const uint8_t *data_end; + size_t data_size; + void *user_priv; + int result; + int worker_id; + int received_frame; + + // scratch_buffer is used in frame parallel mode only. + // It is used to make a copy of the compressed data. + uint8_t *scratch_buffer; + size_t scratch_buffer_size; + +#if CONFIG_MULTITHREAD + pthread_mutex_t stats_mutex; + pthread_cond_t stats_cond; +#endif + + int frame_context_ready; // Current frame's context is ready to read. + int frame_decoded; // Finished decoding current frame. +} FrameWorkerData; + +void vp9_frameworker_lock_stats(VPxWorker *const worker); +void vp9_frameworker_unlock_stats(VPxWorker *const worker); +void vp9_frameworker_signal_stats(VPxWorker *const worker); + +// Wait until ref_buf has been decoded to row in real pixel unit. +// Note: worker may already finish decoding ref_buf and release it in order to +// start decoding next frame. So need to check whether worker is still decoding +// ref_buf. +void vp9_frameworker_wait(VPxWorker *const worker, RefCntBuffer *const ref_buf, + int row); + +// FrameWorker broadcasts its decoding progress so other workers that are +// waiting on it can resume decoding. +void vp9_frameworker_broadcast(RefCntBuffer *const buf, int row); + +// Copy necessary decoding context from src worker to dst worker. +void vp9_frameworker_copy_context(VPxWorker *const dst_worker, + VPxWorker *const src_worker); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP9_DECODER_VP9_DTHREAD_H_ diff --git a/thirdparty/libvpx/vp9/vp9_dx_iface.c b/thirdparty/libvpx/vp9/vp9_dx_iface.c new file mode 100644 index 0000000000..6531e2c618 --- /dev/null +++ b/thirdparty/libvpx/vp9/vp9_dx_iface.c @@ -0,0 +1,1093 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <stdlib.h> +#include <string.h> + +#include "./vpx_config.h" +#include "./vpx_version.h" + +#include "vpx/internal/vpx_codec_internal.h" +#include "vpx/vp8dx.h" +#include "vpx/vpx_decoder.h" +#include "vpx_dsp/bitreader_buffer.h" +#include "vpx_dsp/vpx_dsp_common.h" +#include "vpx_util/vpx_thread.h" + +#include "vp9/common/vp9_alloccommon.h" +#include "vp9/common/vp9_frame_buffers.h" + +#include "vp9/decoder/vp9_decodeframe.h" + +#include "vp9/vp9_dx_iface.h" +#include "vp9/vp9_iface_common.h" + +#define VP9_CAP_POSTPROC (CONFIG_VP9_POSTPROC ? VPX_CODEC_CAP_POSTPROC : 0) + +static vpx_codec_err_t decoder_init(vpx_codec_ctx_t *ctx, + vpx_codec_priv_enc_mr_cfg_t *data) { + // This function only allocates space for the vpx_codec_alg_priv_t + // structure. More memory may be required at the time the stream + // information becomes known. + (void)data; + + if (!ctx->priv) { + vpx_codec_alg_priv_t *const priv = + (vpx_codec_alg_priv_t *)vpx_calloc(1, sizeof(*priv)); + if (priv == NULL) + return VPX_CODEC_MEM_ERROR; + + ctx->priv = (vpx_codec_priv_t *)priv; + ctx->priv->init_flags = ctx->init_flags; + priv->si.sz = sizeof(priv->si); + priv->flushed = 0; + // Only do frame parallel decode when threads > 1. + priv->frame_parallel_decode = + (ctx->config.dec && (ctx->config.dec->threads > 1) && + (ctx->init_flags & VPX_CODEC_USE_FRAME_THREADING)) ? 1 : 0; + if (ctx->config.dec) { + priv->cfg = *ctx->config.dec; + ctx->config.dec = &priv->cfg; + } + } + + return VPX_CODEC_OK; +} + +static vpx_codec_err_t decoder_destroy(vpx_codec_alg_priv_t *ctx) { + if (ctx->frame_workers != NULL) { + int i; + for (i = 0; i < ctx->num_frame_workers; ++i) { + VPxWorker *const worker = &ctx->frame_workers[i]; + FrameWorkerData *const frame_worker_data = + (FrameWorkerData *)worker->data1; + vpx_get_worker_interface()->end(worker); + vp9_remove_common(&frame_worker_data->pbi->common); +#if CONFIG_VP9_POSTPROC + vp9_free_postproc_buffers(&frame_worker_data->pbi->common); +#endif + vp9_decoder_remove(frame_worker_data->pbi); + vpx_free(frame_worker_data->scratch_buffer); +#if CONFIG_MULTITHREAD + pthread_mutex_destroy(&frame_worker_data->stats_mutex); + pthread_cond_destroy(&frame_worker_data->stats_cond); +#endif + vpx_free(frame_worker_data); + } +#if CONFIG_MULTITHREAD + pthread_mutex_destroy(&ctx->buffer_pool->pool_mutex); +#endif + } + + if (ctx->buffer_pool) { + vp9_free_ref_frame_buffers(ctx->buffer_pool); + vp9_free_internal_frame_buffers(&ctx->buffer_pool->int_frame_buffers); + } + + vpx_free(ctx->frame_workers); + vpx_free(ctx->buffer_pool); + vpx_free(ctx); + return VPX_CODEC_OK; +} + +static int parse_bitdepth_colorspace_sampling( + BITSTREAM_PROFILE profile, struct vpx_read_bit_buffer *rb) { + vpx_color_space_t color_space; + if (profile >= PROFILE_2) + rb->bit_offset += 1; // Bit-depth 10 or 12. + color_space = (vpx_color_space_t)vpx_rb_read_literal(rb, 3); + if (color_space != VPX_CS_SRGB) { + rb->bit_offset += 1; // [16,235] (including xvycc) vs [0,255] range. + if (profile == PROFILE_1 || profile == PROFILE_3) { + rb->bit_offset += 2; // subsampling x/y. + rb->bit_offset += 1; // unused. + } + } else { + if (profile == PROFILE_1 || profile == PROFILE_3) { + rb->bit_offset += 1; // unused + } else { + // RGB is only available in version 1. + return 0; + } + } + return 1; +} + +static vpx_codec_err_t decoder_peek_si_internal(const uint8_t *data, + unsigned int data_sz, + vpx_codec_stream_info_t *si, + int *is_intra_only, + vpx_decrypt_cb decrypt_cb, + void *decrypt_state) { + int intra_only_flag = 0; + uint8_t clear_buffer[10]; + + if (data + data_sz <= data) + return VPX_CODEC_INVALID_PARAM; + + si->is_kf = 0; + si->w = si->h = 0; + + if (decrypt_cb) { + data_sz = VPXMIN(sizeof(clear_buffer), data_sz); + decrypt_cb(decrypt_state, data, clear_buffer, data_sz); + data = clear_buffer; + } + + // A maximum of 6 bits are needed to read the frame marker, profile and + // show_existing_frame. + if (data_sz < 1) + return VPX_CODEC_UNSUP_BITSTREAM; + + { + int show_frame; + int error_resilient; + struct vpx_read_bit_buffer rb = { data, data + data_sz, 0, NULL, NULL }; + const int frame_marker = vpx_rb_read_literal(&rb, 2); + const BITSTREAM_PROFILE profile = vp9_read_profile(&rb); + + if (frame_marker != VP9_FRAME_MARKER) + return VPX_CODEC_UNSUP_BITSTREAM; + + if (profile >= MAX_PROFILES) + return VPX_CODEC_UNSUP_BITSTREAM; + + if (vpx_rb_read_bit(&rb)) { // show an existing frame + // If profile is > 2 and show_existing_frame is true, then at least 1 more + // byte (6+3=9 bits) is needed. + if (profile > 2 && data_sz < 2) + return VPX_CODEC_UNSUP_BITSTREAM; + vpx_rb_read_literal(&rb, 3); // Frame buffer to show. + return VPX_CODEC_OK; + } + + // For the rest of the function, a maximum of 9 more bytes are needed + // (computed by taking the maximum possible bits needed in each case). Note + // that this has to be updated if we read any more bits in this function. + if (data_sz < 10) + return VPX_CODEC_UNSUP_BITSTREAM; + + si->is_kf = !vpx_rb_read_bit(&rb); + show_frame = vpx_rb_read_bit(&rb); + error_resilient = vpx_rb_read_bit(&rb); + + if (si->is_kf) { + if (!vp9_read_sync_code(&rb)) + return VPX_CODEC_UNSUP_BITSTREAM; + + if (!parse_bitdepth_colorspace_sampling(profile, &rb)) + return VPX_CODEC_UNSUP_BITSTREAM; + vp9_read_frame_size(&rb, (int *)&si->w, (int *)&si->h); + } else { + intra_only_flag = show_frame ? 0 : vpx_rb_read_bit(&rb); + + rb.bit_offset += error_resilient ? 0 : 2; // reset_frame_context + + if (intra_only_flag) { + if (!vp9_read_sync_code(&rb)) + return VPX_CODEC_UNSUP_BITSTREAM; + if (profile > PROFILE_0) { + if (!parse_bitdepth_colorspace_sampling(profile, &rb)) + return VPX_CODEC_UNSUP_BITSTREAM; + } + rb.bit_offset += REF_FRAMES; // refresh_frame_flags + vp9_read_frame_size(&rb, (int *)&si->w, (int *)&si->h); + } + } + } + if (is_intra_only != NULL) + *is_intra_only = intra_only_flag; + return VPX_CODEC_OK; +} + +static vpx_codec_err_t decoder_peek_si(const uint8_t *data, + unsigned int data_sz, + vpx_codec_stream_info_t *si) { + return decoder_peek_si_internal(data, data_sz, si, NULL, NULL, NULL); +} + +static vpx_codec_err_t decoder_get_si(vpx_codec_alg_priv_t *ctx, + vpx_codec_stream_info_t *si) { + const size_t sz = (si->sz >= sizeof(vp9_stream_info_t)) + ? sizeof(vp9_stream_info_t) + : sizeof(vpx_codec_stream_info_t); + memcpy(si, &ctx->si, sz); + si->sz = (unsigned int)sz; + + return VPX_CODEC_OK; +} + +static void set_error_detail(vpx_codec_alg_priv_t *ctx, + const char *const error) { + ctx->base.err_detail = error; +} + +static vpx_codec_err_t update_error_state(vpx_codec_alg_priv_t *ctx, + const struct vpx_internal_error_info *error) { + if (error->error_code) + set_error_detail(ctx, error->has_detail ? error->detail : NULL); + + return error->error_code; +} + +static void init_buffer_callbacks(vpx_codec_alg_priv_t *ctx) { + int i; + + for (i = 0; i < ctx->num_frame_workers; ++i) { + VPxWorker *const worker = &ctx->frame_workers[i]; + FrameWorkerData *const frame_worker_data = (FrameWorkerData *)worker->data1; + VP9_COMMON *const cm = &frame_worker_data->pbi->common; + BufferPool *const pool = cm->buffer_pool; + + cm->new_fb_idx = INVALID_IDX; + cm->byte_alignment = ctx->byte_alignment; + cm->skip_loop_filter = ctx->skip_loop_filter; + + if (ctx->get_ext_fb_cb != NULL && ctx->release_ext_fb_cb != NULL) { + pool->get_fb_cb = ctx->get_ext_fb_cb; + pool->release_fb_cb = ctx->release_ext_fb_cb; + pool->cb_priv = ctx->ext_priv; + } else { + pool->get_fb_cb = vp9_get_frame_buffer; + pool->release_fb_cb = vp9_release_frame_buffer; + + if (vp9_alloc_internal_frame_buffers(&pool->int_frame_buffers)) + vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR, + "Failed to initialize internal frame buffers"); + + pool->cb_priv = &pool->int_frame_buffers; + } + } +} + +static void set_default_ppflags(vp8_postproc_cfg_t *cfg) { + cfg->post_proc_flag = VP8_DEBLOCK | VP8_DEMACROBLOCK; + cfg->deblocking_level = 4; + cfg->noise_level = 0; +} + +static void set_ppflags(const vpx_codec_alg_priv_t *ctx, + vp9_ppflags_t *flags) { + flags->post_proc_flag = + ctx->postproc_cfg.post_proc_flag; + + flags->deblocking_level = ctx->postproc_cfg.deblocking_level; + flags->noise_level = ctx->postproc_cfg.noise_level; +} + +static int frame_worker_hook(void *arg1, void *arg2) { + FrameWorkerData *const frame_worker_data = (FrameWorkerData *)arg1; + const uint8_t *data = frame_worker_data->data; + (void)arg2; + + frame_worker_data->result = + vp9_receive_compressed_data(frame_worker_data->pbi, + frame_worker_data->data_size, + &data); + frame_worker_data->data_end = data; + + if (frame_worker_data->pbi->frame_parallel_decode) { + // In frame parallel decoding, a worker thread must successfully decode all + // the compressed data. + if (frame_worker_data->result != 0 || + frame_worker_data->data + frame_worker_data->data_size - 1 > data) { + VPxWorker *const worker = frame_worker_data->pbi->frame_worker_owner; + BufferPool *const pool = frame_worker_data->pbi->common.buffer_pool; + // Signal all the other threads that are waiting for this frame. + vp9_frameworker_lock_stats(worker); + frame_worker_data->frame_context_ready = 1; + lock_buffer_pool(pool); + frame_worker_data->pbi->cur_buf->buf.corrupted = 1; + unlock_buffer_pool(pool); + frame_worker_data->pbi->need_resync = 1; + vp9_frameworker_signal_stats(worker); + vp9_frameworker_unlock_stats(worker); + return 0; + } + } else if (frame_worker_data->result != 0) { + // Check decode result in serial decode. + frame_worker_data->pbi->cur_buf->buf.corrupted = 1; + frame_worker_data->pbi->need_resync = 1; + } + return !frame_worker_data->result; +} + +static vpx_codec_err_t init_decoder(vpx_codec_alg_priv_t *ctx) { + int i; + const VPxWorkerInterface *const winterface = vpx_get_worker_interface(); + + ctx->last_show_frame = -1; + ctx->next_submit_worker_id = 0; + ctx->last_submit_worker_id = 0; + ctx->next_output_worker_id = 0; + ctx->frame_cache_read = 0; + ctx->frame_cache_write = 0; + ctx->num_cache_frames = 0; + ctx->need_resync = 1; + ctx->num_frame_workers = + (ctx->frame_parallel_decode == 1) ? ctx->cfg.threads: 1; + if (ctx->num_frame_workers > MAX_DECODE_THREADS) + ctx->num_frame_workers = MAX_DECODE_THREADS; + ctx->available_threads = ctx->num_frame_workers; + ctx->flushed = 0; + + ctx->buffer_pool = (BufferPool *)vpx_calloc(1, sizeof(BufferPool)); + if (ctx->buffer_pool == NULL) + return VPX_CODEC_MEM_ERROR; + +#if CONFIG_MULTITHREAD + if (pthread_mutex_init(&ctx->buffer_pool->pool_mutex, NULL)) { + set_error_detail(ctx, "Failed to allocate buffer pool mutex"); + return VPX_CODEC_MEM_ERROR; + } +#endif + + ctx->frame_workers = (VPxWorker *) + vpx_malloc(ctx->num_frame_workers * sizeof(*ctx->frame_workers)); + if (ctx->frame_workers == NULL) { + set_error_detail(ctx, "Failed to allocate frame_workers"); + return VPX_CODEC_MEM_ERROR; + } + + for (i = 0; i < ctx->num_frame_workers; ++i) { + VPxWorker *const worker = &ctx->frame_workers[i]; + FrameWorkerData *frame_worker_data = NULL; + winterface->init(worker); + worker->data1 = vpx_memalign(32, sizeof(FrameWorkerData)); + if (worker->data1 == NULL) { + set_error_detail(ctx, "Failed to allocate frame_worker_data"); + return VPX_CODEC_MEM_ERROR; + } + frame_worker_data = (FrameWorkerData *)worker->data1; + frame_worker_data->pbi = vp9_decoder_create(ctx->buffer_pool); + if (frame_worker_data->pbi == NULL) { + set_error_detail(ctx, "Failed to allocate frame_worker_data"); + return VPX_CODEC_MEM_ERROR; + } + frame_worker_data->pbi->frame_worker_owner = worker; + frame_worker_data->worker_id = i; + frame_worker_data->scratch_buffer = NULL; + frame_worker_data->scratch_buffer_size = 0; + frame_worker_data->frame_context_ready = 0; + frame_worker_data->received_frame = 0; +#if CONFIG_MULTITHREAD + if (pthread_mutex_init(&frame_worker_data->stats_mutex, NULL)) { + set_error_detail(ctx, "Failed to allocate frame_worker_data mutex"); + return VPX_CODEC_MEM_ERROR; + } + + if (pthread_cond_init(&frame_worker_data->stats_cond, NULL)) { + set_error_detail(ctx, "Failed to allocate frame_worker_data cond"); + return VPX_CODEC_MEM_ERROR; + } +#endif + // If decoding in serial mode, FrameWorker thread could create tile worker + // thread or loopfilter thread. + frame_worker_data->pbi->max_threads = + (ctx->frame_parallel_decode == 0) ? ctx->cfg.threads : 0; + + frame_worker_data->pbi->inv_tile_order = ctx->invert_tile_order; + frame_worker_data->pbi->frame_parallel_decode = ctx->frame_parallel_decode; + frame_worker_data->pbi->common.frame_parallel_decode = + ctx->frame_parallel_decode; + worker->hook = (VPxWorkerHook)frame_worker_hook; + if (!winterface->reset(worker)) { + set_error_detail(ctx, "Frame Worker thread creation failed"); + return VPX_CODEC_MEM_ERROR; + } + } + + // If postprocessing was enabled by the application and a + // configuration has not been provided, default it. + if (!ctx->postproc_cfg_set && + (ctx->base.init_flags & VPX_CODEC_USE_POSTPROC)) + set_default_ppflags(&ctx->postproc_cfg); + + init_buffer_callbacks(ctx); + + return VPX_CODEC_OK; +} + +static INLINE void check_resync(vpx_codec_alg_priv_t *const ctx, + const VP9Decoder *const pbi) { + // Clear resync flag if worker got a key frame or intra only frame. + if (ctx->need_resync == 1 && pbi->need_resync == 0 && + (pbi->common.intra_only || pbi->common.frame_type == KEY_FRAME)) + ctx->need_resync = 0; +} + +static vpx_codec_err_t decode_one(vpx_codec_alg_priv_t *ctx, + const uint8_t **data, unsigned int data_sz, + void *user_priv, int64_t deadline) { + const VPxWorkerInterface *const winterface = vpx_get_worker_interface(); + (void)deadline; + + // Determine the stream parameters. Note that we rely on peek_si to + // validate that we have a buffer that does not wrap around the top + // of the heap. + if (!ctx->si.h) { + int is_intra_only = 0; + const vpx_codec_err_t res = + decoder_peek_si_internal(*data, data_sz, &ctx->si, &is_intra_only, + ctx->decrypt_cb, ctx->decrypt_state); + if (res != VPX_CODEC_OK) + return res; + + if (!ctx->si.is_kf && !is_intra_only) + return VPX_CODEC_ERROR; + } + + if (!ctx->frame_parallel_decode) { + VPxWorker *const worker = ctx->frame_workers; + FrameWorkerData *const frame_worker_data = (FrameWorkerData *)worker->data1; + frame_worker_data->data = *data; + frame_worker_data->data_size = data_sz; + frame_worker_data->user_priv = user_priv; + frame_worker_data->received_frame = 1; + + // Set these even if already initialized. The caller may have changed the + // decrypt config between frames. + frame_worker_data->pbi->decrypt_cb = ctx->decrypt_cb; + frame_worker_data->pbi->decrypt_state = ctx->decrypt_state; + + worker->had_error = 0; + winterface->execute(worker); + + // Update data pointer after decode. + *data = frame_worker_data->data_end; + + if (worker->had_error) + return update_error_state(ctx, &frame_worker_data->pbi->common.error); + + check_resync(ctx, frame_worker_data->pbi); + } else { + VPxWorker *const worker = &ctx->frame_workers[ctx->next_submit_worker_id]; + FrameWorkerData *const frame_worker_data = (FrameWorkerData *)worker->data1; + // Copy context from last worker thread to next worker thread. + if (ctx->next_submit_worker_id != ctx->last_submit_worker_id) + vp9_frameworker_copy_context( + &ctx->frame_workers[ctx->next_submit_worker_id], + &ctx->frame_workers[ctx->last_submit_worker_id]); + + frame_worker_data->pbi->ready_for_new_data = 0; + // Copy the compressed data into worker's internal buffer. + // TODO(hkuang): Will all the workers allocate the same size + // as the size of the first intra frame be better? This will + // avoid too many deallocate and allocate. + if (frame_worker_data->scratch_buffer_size < data_sz) { + frame_worker_data->scratch_buffer = + (uint8_t *)vpx_realloc(frame_worker_data->scratch_buffer, data_sz); + if (frame_worker_data->scratch_buffer == NULL) { + set_error_detail(ctx, "Failed to reallocate scratch buffer"); + return VPX_CODEC_MEM_ERROR; + } + frame_worker_data->scratch_buffer_size = data_sz; + } + frame_worker_data->data_size = data_sz; + memcpy(frame_worker_data->scratch_buffer, *data, data_sz); + + frame_worker_data->frame_decoded = 0; + frame_worker_data->frame_context_ready = 0; + frame_worker_data->received_frame = 1; + frame_worker_data->data = frame_worker_data->scratch_buffer; + frame_worker_data->user_priv = user_priv; + + if (ctx->next_submit_worker_id != ctx->last_submit_worker_id) + ctx->last_submit_worker_id = + (ctx->last_submit_worker_id + 1) % ctx->num_frame_workers; + + ctx->next_submit_worker_id = + (ctx->next_submit_worker_id + 1) % ctx->num_frame_workers; + --ctx->available_threads; + worker->had_error = 0; + winterface->launch(worker); + } + + return VPX_CODEC_OK; +} + +static void wait_worker_and_cache_frame(vpx_codec_alg_priv_t *ctx) { + YV12_BUFFER_CONFIG sd; + vp9_ppflags_t flags = {0, 0, 0}; + const VPxWorkerInterface *const winterface = vpx_get_worker_interface(); + VPxWorker *const worker = &ctx->frame_workers[ctx->next_output_worker_id]; + FrameWorkerData *const frame_worker_data = (FrameWorkerData *)worker->data1; + ctx->next_output_worker_id = + (ctx->next_output_worker_id + 1) % ctx->num_frame_workers; + // TODO(hkuang): Add worker error handling here. + winterface->sync(worker); + frame_worker_data->received_frame = 0; + ++ctx->available_threads; + + check_resync(ctx, frame_worker_data->pbi); + + if (vp9_get_raw_frame(frame_worker_data->pbi, &sd, &flags) == 0) { + VP9_COMMON *const cm = &frame_worker_data->pbi->common; + RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs; + ctx->frame_cache[ctx->frame_cache_write].fb_idx = cm->new_fb_idx; + yuvconfig2image(&ctx->frame_cache[ctx->frame_cache_write].img, &sd, + frame_worker_data->user_priv); + ctx->frame_cache[ctx->frame_cache_write].img.fb_priv = + frame_bufs[cm->new_fb_idx].raw_frame_buffer.priv; + ctx->frame_cache_write = + (ctx->frame_cache_write + 1) % FRAME_CACHE_SIZE; + ++ctx->num_cache_frames; + } +} + +static vpx_codec_err_t decoder_decode(vpx_codec_alg_priv_t *ctx, + const uint8_t *data, unsigned int data_sz, + void *user_priv, long deadline) { + const uint8_t *data_start = data; + const uint8_t * const data_end = data + data_sz; + vpx_codec_err_t res; + uint32_t frame_sizes[8]; + int frame_count; + + if (data == NULL && data_sz == 0) { + ctx->flushed = 1; + return VPX_CODEC_OK; + } + + // Reset flushed when receiving a valid frame. + ctx->flushed = 0; + + // Initialize the decoder workers on the first frame. + if (ctx->frame_workers == NULL) { + const vpx_codec_err_t res = init_decoder(ctx); + if (res != VPX_CODEC_OK) + return res; + } + + res = vp9_parse_superframe_index(data, data_sz, frame_sizes, &frame_count, + ctx->decrypt_cb, ctx->decrypt_state); + if (res != VPX_CODEC_OK) + return res; + + if (ctx->frame_parallel_decode) { + // Decode in frame parallel mode. When decoding in this mode, the frame + // passed to the decoder must be either a normal frame or a superframe with + // superframe index so the decoder could get each frame's start position + // in the superframe. + if (frame_count > 0) { + int i; + + for (i = 0; i < frame_count; ++i) { + const uint8_t *data_start_copy = data_start; + const uint32_t frame_size = frame_sizes[i]; + if (data_start < data + || frame_size > (uint32_t) (data_end - data_start)) { + set_error_detail(ctx, "Invalid frame size in index"); + return VPX_CODEC_CORRUPT_FRAME; + } + + if (ctx->available_threads == 0) { + // No more threads for decoding. Wait until the next output worker + // finishes decoding. Then copy the decoded frame into cache. + if (ctx->num_cache_frames < FRAME_CACHE_SIZE) { + wait_worker_and_cache_frame(ctx); + } else { + // TODO(hkuang): Add unit test to test this path. + set_error_detail(ctx, "Frame output cache is full."); + return VPX_CODEC_ERROR; + } + } + + res = decode_one(ctx, &data_start_copy, frame_size, user_priv, + deadline); + if (res != VPX_CODEC_OK) + return res; + data_start += frame_size; + } + } else { + if (ctx->available_threads == 0) { + // No more threads for decoding. Wait until the next output worker + // finishes decoding. Then copy the decoded frame into cache. + if (ctx->num_cache_frames < FRAME_CACHE_SIZE) { + wait_worker_and_cache_frame(ctx); + } else { + // TODO(hkuang): Add unit test to test this path. + set_error_detail(ctx, "Frame output cache is full."); + return VPX_CODEC_ERROR; + } + } + + res = decode_one(ctx, &data, data_sz, user_priv, deadline); + if (res != VPX_CODEC_OK) + return res; + } + } else { + // Decode in serial mode. + if (frame_count > 0) { + int i; + + for (i = 0; i < frame_count; ++i) { + const uint8_t *data_start_copy = data_start; + const uint32_t frame_size = frame_sizes[i]; + vpx_codec_err_t res; + if (data_start < data + || frame_size > (uint32_t) (data_end - data_start)) { + set_error_detail(ctx, "Invalid frame size in index"); + return VPX_CODEC_CORRUPT_FRAME; + } + + res = decode_one(ctx, &data_start_copy, frame_size, user_priv, + deadline); + if (res != VPX_CODEC_OK) + return res; + + data_start += frame_size; + } + } else { + while (data_start < data_end) { + const uint32_t frame_size = (uint32_t) (data_end - data_start); + const vpx_codec_err_t res = decode_one(ctx, &data_start, frame_size, + user_priv, deadline); + if (res != VPX_CODEC_OK) + return res; + + // Account for suboptimal termination by the encoder. + while (data_start < data_end) { + const uint8_t marker = read_marker(ctx->decrypt_cb, + ctx->decrypt_state, data_start); + if (marker) + break; + ++data_start; + } + } + } + } + + return res; +} + +static void release_last_output_frame(vpx_codec_alg_priv_t *ctx) { + RefCntBuffer *const frame_bufs = ctx->buffer_pool->frame_bufs; + // Decrease reference count of last output frame in frame parallel mode. + if (ctx->frame_parallel_decode && ctx->last_show_frame >= 0) { + BufferPool *const pool = ctx->buffer_pool; + lock_buffer_pool(pool); + decrease_ref_count(ctx->last_show_frame, frame_bufs, pool); + unlock_buffer_pool(pool); + } +} + +static vpx_image_t *decoder_get_frame(vpx_codec_alg_priv_t *ctx, + vpx_codec_iter_t *iter) { + vpx_image_t *img = NULL; + + // Only return frame when all the cpu are busy or + // application fluhsed the decoder in frame parallel decode. + if (ctx->frame_parallel_decode && ctx->available_threads > 0 && + !ctx->flushed) { + return NULL; + } + + // Output the frames in the cache first. + if (ctx->num_cache_frames > 0) { + release_last_output_frame(ctx); + ctx->last_show_frame = ctx->frame_cache[ctx->frame_cache_read].fb_idx; + if (ctx->need_resync) + return NULL; + img = &ctx->frame_cache[ctx->frame_cache_read].img; + ctx->frame_cache_read = (ctx->frame_cache_read + 1) % FRAME_CACHE_SIZE; + --ctx->num_cache_frames; + return img; + } + + // iter acts as a flip flop, so an image is only returned on the first + // call to get_frame. + if (*iter == NULL && ctx->frame_workers != NULL) { + do { + YV12_BUFFER_CONFIG sd; + vp9_ppflags_t flags = {0, 0, 0}; + const VPxWorkerInterface *const winterface = vpx_get_worker_interface(); + VPxWorker *const worker = + &ctx->frame_workers[ctx->next_output_worker_id]; + FrameWorkerData *const frame_worker_data = + (FrameWorkerData *)worker->data1; + ctx->next_output_worker_id = + (ctx->next_output_worker_id + 1) % ctx->num_frame_workers; + if (ctx->base.init_flags & VPX_CODEC_USE_POSTPROC) + set_ppflags(ctx, &flags); + // Wait for the frame from worker thread. + if (winterface->sync(worker)) { + // Check if worker has received any frames. + if (frame_worker_data->received_frame == 1) { + ++ctx->available_threads; + frame_worker_data->received_frame = 0; + check_resync(ctx, frame_worker_data->pbi); + } + if (vp9_get_raw_frame(frame_worker_data->pbi, &sd, &flags) == 0) { + VP9_COMMON *const cm = &frame_worker_data->pbi->common; + RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs; + release_last_output_frame(ctx); + ctx->last_show_frame = frame_worker_data->pbi->common.new_fb_idx; + if (ctx->need_resync) + return NULL; + yuvconfig2image(&ctx->img, &sd, frame_worker_data->user_priv); + ctx->img.fb_priv = frame_bufs[cm->new_fb_idx].raw_frame_buffer.priv; + img = &ctx->img; + return img; + } + } else { + // Decoding failed. Release the worker thread. + frame_worker_data->received_frame = 0; + ++ctx->available_threads; + ctx->need_resync = 1; + if (ctx->flushed != 1) + return NULL; + } + } while (ctx->next_output_worker_id != ctx->next_submit_worker_id); + } + return NULL; +} + +static vpx_codec_err_t decoder_set_fb_fn( + vpx_codec_alg_priv_t *ctx, + vpx_get_frame_buffer_cb_fn_t cb_get, + vpx_release_frame_buffer_cb_fn_t cb_release, void *cb_priv) { + if (cb_get == NULL || cb_release == NULL) { + return VPX_CODEC_INVALID_PARAM; + } else if (ctx->frame_workers == NULL) { + // If the decoder has already been initialized, do not accept changes to + // the frame buffer functions. + ctx->get_ext_fb_cb = cb_get; + ctx->release_ext_fb_cb = cb_release; + ctx->ext_priv = cb_priv; + return VPX_CODEC_OK; + } + + return VPX_CODEC_ERROR; +} + +static vpx_codec_err_t ctrl_set_reference(vpx_codec_alg_priv_t *ctx, + va_list args) { + vpx_ref_frame_t *const data = va_arg(args, vpx_ref_frame_t *); + + // Only support this function in serial decode. + if (ctx->frame_parallel_decode) { + set_error_detail(ctx, "Not supported in frame parallel decode"); + return VPX_CODEC_INCAPABLE; + } + + if (data) { + vpx_ref_frame_t *const frame = (vpx_ref_frame_t *)data; + YV12_BUFFER_CONFIG sd; + VPxWorker *const worker = ctx->frame_workers; + FrameWorkerData *const frame_worker_data = (FrameWorkerData *)worker->data1; + image2yuvconfig(&frame->img, &sd); + return vp9_set_reference_dec(&frame_worker_data->pbi->common, + (VP9_REFFRAME)frame->frame_type, &sd); + } else { + return VPX_CODEC_INVALID_PARAM; + } +} + +static vpx_codec_err_t ctrl_copy_reference(vpx_codec_alg_priv_t *ctx, + va_list args) { + vpx_ref_frame_t *data = va_arg(args, vpx_ref_frame_t *); + + // Only support this function in serial decode. + if (ctx->frame_parallel_decode) { + set_error_detail(ctx, "Not supported in frame parallel decode"); + return VPX_CODEC_INCAPABLE; + } + + if (data) { + vpx_ref_frame_t *frame = (vpx_ref_frame_t *) data; + YV12_BUFFER_CONFIG sd; + VPxWorker *const worker = ctx->frame_workers; + FrameWorkerData *const frame_worker_data = (FrameWorkerData *)worker->data1; + image2yuvconfig(&frame->img, &sd); + return vp9_copy_reference_dec(frame_worker_data->pbi, + (VP9_REFFRAME)frame->frame_type, &sd); + } else { + return VPX_CODEC_INVALID_PARAM; + } +} + +static vpx_codec_err_t ctrl_get_reference(vpx_codec_alg_priv_t *ctx, + va_list args) { + vp9_ref_frame_t *data = va_arg(args, vp9_ref_frame_t *); + + // Only support this function in serial decode. + if (ctx->frame_parallel_decode) { + set_error_detail(ctx, "Not supported in frame parallel decode"); + return VPX_CODEC_INCAPABLE; + } + + if (data) { + YV12_BUFFER_CONFIG* fb; + VPxWorker *const worker = ctx->frame_workers; + FrameWorkerData *const frame_worker_data = (FrameWorkerData *)worker->data1; + fb = get_ref_frame(&frame_worker_data->pbi->common, data->idx); + if (fb == NULL) return VPX_CODEC_ERROR; + yuvconfig2image(&data->img, fb, NULL); + return VPX_CODEC_OK; + } else { + return VPX_CODEC_INVALID_PARAM; + } +} + +static vpx_codec_err_t ctrl_set_postproc(vpx_codec_alg_priv_t *ctx, + va_list args) { +#if CONFIG_VP9_POSTPROC + vp8_postproc_cfg_t *data = va_arg(args, vp8_postproc_cfg_t *); + + if (data) { + ctx->postproc_cfg_set = 1; + ctx->postproc_cfg = *((vp8_postproc_cfg_t *)data); + return VPX_CODEC_OK; + } else { + return VPX_CODEC_INVALID_PARAM; + } +#else + (void)ctx; + (void)args; + return VPX_CODEC_INCAPABLE; +#endif +} + +static vpx_codec_err_t ctrl_set_dbg_options(vpx_codec_alg_priv_t *ctx, + va_list args) { + (void)ctx; + (void)args; + return VPX_CODEC_INCAPABLE; +} + +static vpx_codec_err_t ctrl_get_last_ref_updates(vpx_codec_alg_priv_t *ctx, + va_list args) { + int *const update_info = va_arg(args, int *); + + // Only support this function in serial decode. + if (ctx->frame_parallel_decode) { + set_error_detail(ctx, "Not supported in frame parallel decode"); + return VPX_CODEC_INCAPABLE; + } + + if (update_info) { + if (ctx->frame_workers) { + VPxWorker *const worker = ctx->frame_workers; + FrameWorkerData *const frame_worker_data = + (FrameWorkerData *)worker->data1; + *update_info = frame_worker_data->pbi->refresh_frame_flags; + return VPX_CODEC_OK; + } else { + return VPX_CODEC_ERROR; + } + } + + return VPX_CODEC_INVALID_PARAM; +} + +static vpx_codec_err_t ctrl_get_frame_corrupted(vpx_codec_alg_priv_t *ctx, + va_list args) { + int *corrupted = va_arg(args, int *); + + if (corrupted) { + if (ctx->frame_workers) { + VPxWorker *const worker = ctx->frame_workers; + FrameWorkerData *const frame_worker_data = + (FrameWorkerData *)worker->data1; + RefCntBuffer *const frame_bufs = + frame_worker_data->pbi->common.buffer_pool->frame_bufs; + if (frame_worker_data->pbi->common.frame_to_show == NULL) + return VPX_CODEC_ERROR; + if (ctx->last_show_frame >= 0) + *corrupted = frame_bufs[ctx->last_show_frame].buf.corrupted; + return VPX_CODEC_OK; + } else { + return VPX_CODEC_ERROR; + } + } + + return VPX_CODEC_INVALID_PARAM; +} + +static vpx_codec_err_t ctrl_get_frame_size(vpx_codec_alg_priv_t *ctx, + va_list args) { + int *const frame_size = va_arg(args, int *); + + // Only support this function in serial decode. + if (ctx->frame_parallel_decode) { + set_error_detail(ctx, "Not supported in frame parallel decode"); + return VPX_CODEC_INCAPABLE; + } + + if (frame_size) { + if (ctx->frame_workers) { + VPxWorker *const worker = ctx->frame_workers; + FrameWorkerData *const frame_worker_data = + (FrameWorkerData *)worker->data1; + const VP9_COMMON *const cm = &frame_worker_data->pbi->common; + frame_size[0] = cm->width; + frame_size[1] = cm->height; + return VPX_CODEC_OK; + } else { + return VPX_CODEC_ERROR; + } + } + + return VPX_CODEC_INVALID_PARAM; +} + +static vpx_codec_err_t ctrl_get_render_size(vpx_codec_alg_priv_t *ctx, + va_list args) { + int *const render_size = va_arg(args, int *); + + // Only support this function in serial decode. + if (ctx->frame_parallel_decode) { + set_error_detail(ctx, "Not supported in frame parallel decode"); + return VPX_CODEC_INCAPABLE; + } + + if (render_size) { + if (ctx->frame_workers) { + VPxWorker *const worker = ctx->frame_workers; + FrameWorkerData *const frame_worker_data = + (FrameWorkerData *)worker->data1; + const VP9_COMMON *const cm = &frame_worker_data->pbi->common; + render_size[0] = cm->render_width; + render_size[1] = cm->render_height; + return VPX_CODEC_OK; + } else { + return VPX_CODEC_ERROR; + } + } + + return VPX_CODEC_INVALID_PARAM; +} + +static vpx_codec_err_t ctrl_get_bit_depth(vpx_codec_alg_priv_t *ctx, + va_list args) { + unsigned int *const bit_depth = va_arg(args, unsigned int *); + VPxWorker *const worker = &ctx->frame_workers[ctx->next_output_worker_id]; + + if (bit_depth) { + if (worker) { + FrameWorkerData *const frame_worker_data = + (FrameWorkerData *)worker->data1; + const VP9_COMMON *const cm = &frame_worker_data->pbi->common; + *bit_depth = cm->bit_depth; + return VPX_CODEC_OK; + } else { + return VPX_CODEC_ERROR; + } + } + + return VPX_CODEC_INVALID_PARAM; +} + +static vpx_codec_err_t ctrl_set_invert_tile_order(vpx_codec_alg_priv_t *ctx, + va_list args) { + ctx->invert_tile_order = va_arg(args, int); + return VPX_CODEC_OK; +} + +static vpx_codec_err_t ctrl_set_decryptor(vpx_codec_alg_priv_t *ctx, + va_list args) { + vpx_decrypt_init *init = va_arg(args, vpx_decrypt_init *); + ctx->decrypt_cb = init ? init->decrypt_cb : NULL; + ctx->decrypt_state = init ? init->decrypt_state : NULL; + return VPX_CODEC_OK; +} + +static vpx_codec_err_t ctrl_set_byte_alignment(vpx_codec_alg_priv_t *ctx, + va_list args) { + const int legacy_byte_alignment = 0; + const int min_byte_alignment = 32; + const int max_byte_alignment = 1024; + const int byte_alignment = va_arg(args, int); + + if (byte_alignment != legacy_byte_alignment && + (byte_alignment < min_byte_alignment || + byte_alignment > max_byte_alignment || + (byte_alignment & (byte_alignment - 1)) != 0)) + return VPX_CODEC_INVALID_PARAM; + + ctx->byte_alignment = byte_alignment; + if (ctx->frame_workers) { + VPxWorker *const worker = ctx->frame_workers; + FrameWorkerData *const frame_worker_data = + (FrameWorkerData *)worker->data1; + frame_worker_data->pbi->common.byte_alignment = byte_alignment; + } + return VPX_CODEC_OK; +} + +static vpx_codec_err_t ctrl_set_skip_loop_filter(vpx_codec_alg_priv_t *ctx, + va_list args) { + ctx->skip_loop_filter = va_arg(args, int); + + if (ctx->frame_workers) { + VPxWorker *const worker = ctx->frame_workers; + FrameWorkerData *const frame_worker_data = (FrameWorkerData *)worker->data1; + frame_worker_data->pbi->common.skip_loop_filter = ctx->skip_loop_filter; + } + + return VPX_CODEC_OK; +} + +static vpx_codec_ctrl_fn_map_t decoder_ctrl_maps[] = { + {VP8_COPY_REFERENCE, ctrl_copy_reference}, + + // Setters + {VP8_SET_REFERENCE, ctrl_set_reference}, + {VP8_SET_POSTPROC, ctrl_set_postproc}, + {VP8_SET_DBG_COLOR_REF_FRAME, ctrl_set_dbg_options}, + {VP8_SET_DBG_COLOR_MB_MODES, ctrl_set_dbg_options}, + {VP8_SET_DBG_COLOR_B_MODES, ctrl_set_dbg_options}, + {VP8_SET_DBG_DISPLAY_MV, ctrl_set_dbg_options}, + {VP9_INVERT_TILE_DECODE_ORDER, ctrl_set_invert_tile_order}, + {VPXD_SET_DECRYPTOR, ctrl_set_decryptor}, + {VP9_SET_BYTE_ALIGNMENT, ctrl_set_byte_alignment}, + {VP9_SET_SKIP_LOOP_FILTER, ctrl_set_skip_loop_filter}, + + // Getters + {VP8D_GET_LAST_REF_UPDATES, ctrl_get_last_ref_updates}, + {VP8D_GET_FRAME_CORRUPTED, ctrl_get_frame_corrupted}, + {VP9_GET_REFERENCE, ctrl_get_reference}, + {VP9D_GET_DISPLAY_SIZE, ctrl_get_render_size}, + {VP9D_GET_BIT_DEPTH, ctrl_get_bit_depth}, + {VP9D_GET_FRAME_SIZE, ctrl_get_frame_size}, + + { -1, NULL}, +}; + +#ifndef VERSION_STRING +#define VERSION_STRING +#endif +CODEC_INTERFACE(vpx_codec_vp9_dx) = { + "WebM Project VP9 Decoder" VERSION_STRING, + VPX_CODEC_INTERNAL_ABI_VERSION, + VPX_CODEC_CAP_DECODER | VP9_CAP_POSTPROC | + VPX_CODEC_CAP_EXTERNAL_FRAME_BUFFER, // vpx_codec_caps_t + decoder_init, // vpx_codec_init_fn_t + decoder_destroy, // vpx_codec_destroy_fn_t + decoder_ctrl_maps, // vpx_codec_ctrl_fn_map_t + { // NOLINT + decoder_peek_si, // vpx_codec_peek_si_fn_t + decoder_get_si, // vpx_codec_get_si_fn_t + decoder_decode, // vpx_codec_decode_fn_t + decoder_get_frame, // vpx_codec_frame_get_fn_t + decoder_set_fb_fn, // vpx_codec_set_fb_fn_t + }, + { // NOLINT + 0, + NULL, // vpx_codec_enc_cfg_map_t + NULL, // vpx_codec_encode_fn_t + NULL, // vpx_codec_get_cx_data_fn_t + NULL, // vpx_codec_enc_config_set_fn_t + NULL, // vpx_codec_get_global_headers_fn_t + NULL, // vpx_codec_get_preview_frame_fn_t + NULL // vpx_codec_enc_mr_get_mem_loc_fn_t + } +}; diff --git a/thirdparty/libvpx/vp9/vp9_dx_iface.h b/thirdparty/libvpx/vp9/vp9_dx_iface.h new file mode 100644 index 0000000000..e0e948e16c --- /dev/null +++ b/thirdparty/libvpx/vp9/vp9_dx_iface.h @@ -0,0 +1,65 @@ +/* + * Copyright (c) 2015 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP9_VP9_DX_IFACE_H_ +#define VP9_VP9_DX_IFACE_H_ + +#include "vp9/decoder/vp9_decoder.h" + +typedef vpx_codec_stream_info_t vp9_stream_info_t; + +// This limit is due to framebuffer numbers. +// TODO(hkuang): Remove this limit after implementing ondemand framebuffers. +#define FRAME_CACHE_SIZE 6 // Cache maximum 6 decoded frames. + +typedef struct cache_frame { + int fb_idx; + vpx_image_t img; +} cache_frame; + +struct vpx_codec_alg_priv { + vpx_codec_priv_t base; + vpx_codec_dec_cfg_t cfg; + vp9_stream_info_t si; + int postproc_cfg_set; + vp8_postproc_cfg_t postproc_cfg; + vpx_decrypt_cb decrypt_cb; + void *decrypt_state; + vpx_image_t img; + int img_avail; + int flushed; + int invert_tile_order; + int last_show_frame; // Index of last output frame. + int byte_alignment; + int skip_loop_filter; + + // Frame parallel related. + int frame_parallel_decode; // frame-based threading. + VPxWorker *frame_workers; + int num_frame_workers; + int next_submit_worker_id; + int last_submit_worker_id; + int next_output_worker_id; + int available_threads; + cache_frame frame_cache[FRAME_CACHE_SIZE]; + int frame_cache_write; + int frame_cache_read; + int num_cache_frames; + int need_resync; // wait for key/intra-only frame + // BufferPool that holds all reference frames. Shared by all the FrameWorkers. + BufferPool *buffer_pool; + + // External frame buffer info to save for VP9 common. + void *ext_priv; // Private data associated with the external frame buffers. + vpx_get_frame_buffer_cb_fn_t get_ext_fb_cb; + vpx_release_frame_buffer_cb_fn_t release_ext_fb_cb; +}; + +#endif // VP9_VP9_DX_IFACE_H_ diff --git a/thirdparty/libvpx/vp9/vp9_iface_common.h b/thirdparty/libvpx/vp9/vp9_iface_common.h new file mode 100644 index 0000000000..938d4224ba --- /dev/null +++ b/thirdparty/libvpx/vp9/vp9_iface_common.h @@ -0,0 +1,136 @@ +/* + * Copyright (c) 2013 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ +#ifndef VP9_VP9_IFACE_COMMON_H_ +#define VP9_VP9_IFACE_COMMON_H_ + +#include "vpx_ports/mem.h" + +static void yuvconfig2image(vpx_image_t *img, const YV12_BUFFER_CONFIG *yv12, + void *user_priv) { + /** vpx_img_wrap() doesn't allow specifying independent strides for + * the Y, U, and V planes, nor other alignment adjustments that + * might be representable by a YV12_BUFFER_CONFIG, so we just + * initialize all the fields.*/ + int bps; + if (!yv12->subsampling_y) { + if (!yv12->subsampling_x) { + img->fmt = VPX_IMG_FMT_I444; + bps = 24; + } else { + img->fmt = VPX_IMG_FMT_I422; + bps = 16; + } + } else { + if (!yv12->subsampling_x) { + img->fmt = VPX_IMG_FMT_I440; + bps = 16; + } else { + img->fmt = VPX_IMG_FMT_I420; + bps = 12; + } + } + img->cs = yv12->color_space; + img->range = yv12->color_range; + img->bit_depth = 8; + img->w = yv12->y_stride; + img->h = ALIGN_POWER_OF_TWO(yv12->y_height + 2 * VP9_ENC_BORDER_IN_PIXELS, 3); + img->d_w = yv12->y_crop_width; + img->d_h = yv12->y_crop_height; + img->r_w = yv12->render_width; + img->r_h = yv12->render_height; + img->x_chroma_shift = yv12->subsampling_x; + img->y_chroma_shift = yv12->subsampling_y; + img->planes[VPX_PLANE_Y] = yv12->y_buffer; + img->planes[VPX_PLANE_U] = yv12->u_buffer; + img->planes[VPX_PLANE_V] = yv12->v_buffer; + img->planes[VPX_PLANE_ALPHA] = NULL; + img->stride[VPX_PLANE_Y] = yv12->y_stride; + img->stride[VPX_PLANE_U] = yv12->uv_stride; + img->stride[VPX_PLANE_V] = yv12->uv_stride; + img->stride[VPX_PLANE_ALPHA] = yv12->y_stride; +#if CONFIG_VP9_HIGHBITDEPTH + if (yv12->flags & YV12_FLAG_HIGHBITDEPTH) { + // vpx_image_t uses byte strides and a pointer to the first byte + // of the image. + img->fmt = (vpx_img_fmt_t)(img->fmt | VPX_IMG_FMT_HIGHBITDEPTH); + img->bit_depth = yv12->bit_depth; + img->planes[VPX_PLANE_Y] = (uint8_t*)CONVERT_TO_SHORTPTR(yv12->y_buffer); + img->planes[VPX_PLANE_U] = (uint8_t*)CONVERT_TO_SHORTPTR(yv12->u_buffer); + img->planes[VPX_PLANE_V] = (uint8_t*)CONVERT_TO_SHORTPTR(yv12->v_buffer); + img->planes[VPX_PLANE_ALPHA] = NULL; + img->stride[VPX_PLANE_Y] = 2 * yv12->y_stride; + img->stride[VPX_PLANE_U] = 2 * yv12->uv_stride; + img->stride[VPX_PLANE_V] = 2 * yv12->uv_stride; + img->stride[VPX_PLANE_ALPHA] = 2 * yv12->y_stride; + } +#endif // CONFIG_VP9_HIGHBITDEPTH + img->bps = bps; + img->user_priv = user_priv; + img->img_data = yv12->buffer_alloc; + img->img_data_owner = 0; + img->self_allocd = 0; +} + +static vpx_codec_err_t image2yuvconfig(const vpx_image_t *img, + YV12_BUFFER_CONFIG *yv12) { + yv12->y_buffer = img->planes[VPX_PLANE_Y]; + yv12->u_buffer = img->planes[VPX_PLANE_U]; + yv12->v_buffer = img->planes[VPX_PLANE_V]; + + yv12->y_crop_width = img->d_w; + yv12->y_crop_height = img->d_h; + yv12->render_width = img->r_w; + yv12->render_height = img->r_h; + yv12->y_width = img->d_w; + yv12->y_height = img->d_h; + + yv12->uv_width = img->x_chroma_shift == 1 ? (1 + yv12->y_width) / 2 + : yv12->y_width; + yv12->uv_height = img->y_chroma_shift == 1 ? (1 + yv12->y_height) / 2 + : yv12->y_height; + yv12->uv_crop_width = yv12->uv_width; + yv12->uv_crop_height = yv12->uv_height; + + yv12->y_stride = img->stride[VPX_PLANE_Y]; + yv12->uv_stride = img->stride[VPX_PLANE_U]; + yv12->color_space = img->cs; + yv12->color_range = img->range; + +#if CONFIG_VP9_HIGHBITDEPTH + if (img->fmt & VPX_IMG_FMT_HIGHBITDEPTH) { + // In vpx_image_t + // planes point to uint8 address of start of data + // stride counts uint8s to reach next row + // In YV12_BUFFER_CONFIG + // y_buffer, u_buffer, v_buffer point to uint16 address of data + // stride and border counts in uint16s + // This means that all the address calculations in the main body of code + // should work correctly. + // However, before we do any pixel operations we need to cast the address + // to a uint16 ponter and double its value. + yv12->y_buffer = CONVERT_TO_BYTEPTR(yv12->y_buffer); + yv12->u_buffer = CONVERT_TO_BYTEPTR(yv12->u_buffer); + yv12->v_buffer = CONVERT_TO_BYTEPTR(yv12->v_buffer); + yv12->y_stride >>= 1; + yv12->uv_stride >>= 1; + yv12->flags = YV12_FLAG_HIGHBITDEPTH; + } else { + yv12->flags = 0; + } + yv12->border = (yv12->y_stride - img->w) / 2; +#else + yv12->border = (img->stride[VPX_PLANE_Y] - img->w) / 2; +#endif // CONFIG_VP9_HIGHBITDEPTH + yv12->subsampling_x = img->x_chroma_shift; + yv12->subsampling_y = img->y_chroma_shift; + return VPX_CODEC_OK; +} + +#endif // VP9_VP9_IFACE_COMMON_H_ diff --git a/thirdparty/libvpx/vpx/internal/vpx_codec_internal.h b/thirdparty/libvpx/vpx/internal/vpx_codec_internal.h new file mode 100644 index 0000000000..7380fcc7e2 --- /dev/null +++ b/thirdparty/libvpx/vpx/internal/vpx_codec_internal.h @@ -0,0 +1,445 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +/*!\file + * \brief Describes the decoder algorithm interface for algorithm + * implementations. + * + * This file defines the private structures and data types that are only + * relevant to implementing an algorithm, as opposed to using it. + * + * To create a decoder algorithm class, an interface structure is put + * into the global namespace: + * <pre> + * my_codec.c: + * vpx_codec_iface_t my_codec = { + * "My Codec v1.0", + * VPX_CODEC_ALG_ABI_VERSION, + * ... + * }; + * </pre> + * + * An application instantiates a specific decoder instance by using + * vpx_codec_init() and a pointer to the algorithm's interface structure: + * <pre> + * my_app.c: + * extern vpx_codec_iface_t my_codec; + * { + * vpx_codec_ctx_t algo; + * res = vpx_codec_init(&algo, &my_codec); + * } + * </pre> + * + * Once initialized, the instance is manged using other functions from + * the vpx_codec_* family. + */ +#ifndef VPX_INTERNAL_VPX_CODEC_INTERNAL_H_ +#define VPX_INTERNAL_VPX_CODEC_INTERNAL_H_ +#include "../vpx_decoder.h" +#include "../vpx_encoder.h" +#include <stdarg.h> + +#ifdef __cplusplus +extern "C" { +#endif + +/*!\brief Current ABI version number + * + * \internal + * If this file is altered in any way that changes the ABI, this value + * must be bumped. Examples include, but are not limited to, changing + * types, removing or reassigning enums, adding/removing/rearranging + * fields to structures + */ +#define VPX_CODEC_INTERNAL_ABI_VERSION (5) /**<\hideinitializer*/ + +typedef struct vpx_codec_alg_priv vpx_codec_alg_priv_t; +typedef struct vpx_codec_priv_enc_mr_cfg vpx_codec_priv_enc_mr_cfg_t; + +/*!\brief init function pointer prototype + * + * Performs algorithm-specific initialization of the decoder context. This + * function is called by the generic vpx_codec_init() wrapper function, so + * plugins implementing this interface may trust the input parameters to be + * properly initialized. + * + * \param[in] ctx Pointer to this instance's context + * \retval #VPX_CODEC_OK + * The input stream was recognized and decoder initialized. + * \retval #VPX_CODEC_MEM_ERROR + * Memory operation failed. + */ +typedef vpx_codec_err_t (*vpx_codec_init_fn_t)(vpx_codec_ctx_t *ctx, + vpx_codec_priv_enc_mr_cfg_t *data); + +/*!\brief destroy function pointer prototype + * + * Performs algorithm-specific destruction of the decoder context. This + * function is called by the generic vpx_codec_destroy() wrapper function, + * so plugins implementing this interface may trust the input parameters + * to be properly initialized. + * + * \param[in] ctx Pointer to this instance's context + * \retval #VPX_CODEC_OK + * The input stream was recognized and decoder initialized. + * \retval #VPX_CODEC_MEM_ERROR + * Memory operation failed. + */ +typedef vpx_codec_err_t (*vpx_codec_destroy_fn_t)(vpx_codec_alg_priv_t *ctx); + +/*!\brief parse stream info function pointer prototype + * + * Performs high level parsing of the bitstream. This function is called by the + * generic vpx_codec_peek_stream_info() wrapper function, so plugins + * implementing this interface may trust the input parameters to be properly + * initialized. + * + * \param[in] data Pointer to a block of data to parse + * \param[in] data_sz Size of the data buffer + * \param[in,out] si Pointer to stream info to update. The size member + * \ref MUST be properly initialized, but \ref MAY be + * clobbered by the algorithm. This parameter \ref MAY + * be NULL. + * + * \retval #VPX_CODEC_OK + * Bitstream is parsable and stream information updated + */ +typedef vpx_codec_err_t (*vpx_codec_peek_si_fn_t)(const uint8_t *data, + unsigned int data_sz, + vpx_codec_stream_info_t *si); + +/*!\brief Return information about the current stream. + * + * Returns information about the stream that has been parsed during decoding. + * + * \param[in] ctx Pointer to this instance's context + * \param[in,out] si Pointer to stream info to update. The size member + * \ref MUST be properly initialized, but \ref MAY be + * clobbered by the algorithm. This parameter \ref MAY + * be NULL. + * + * \retval #VPX_CODEC_OK + * Bitstream is parsable and stream information updated + */ +typedef vpx_codec_err_t (*vpx_codec_get_si_fn_t)(vpx_codec_alg_priv_t *ctx, + vpx_codec_stream_info_t *si); + +/*!\brief control function pointer prototype + * + * This function is used to exchange algorithm specific data with the decoder + * instance. This can be used to implement features specific to a particular + * algorithm. + * + * This function is called by the generic vpx_codec_control() wrapper + * function, so plugins implementing this interface may trust the input + * parameters to be properly initialized. However, this interface does not + * provide type safety for the exchanged data or assign meanings to the + * control codes. Those details should be specified in the algorithm's + * header file. In particular, the ctrl_id parameter is guaranteed to exist + * in the algorithm's control mapping table, and the data parameter may be NULL. + * + * + * \param[in] ctx Pointer to this instance's context + * \param[in] ctrl_id Algorithm specific control identifier + * \param[in,out] data Data to exchange with algorithm instance. + * + * \retval #VPX_CODEC_OK + * The internal state data was deserialized. + */ +typedef vpx_codec_err_t (*vpx_codec_control_fn_t)(vpx_codec_alg_priv_t *ctx, + va_list ap); + +/*!\brief control function pointer mapping + * + * This structure stores the mapping between control identifiers and + * implementing functions. Each algorithm provides a list of these + * mappings. This list is searched by the vpx_codec_control() wrapper + * function to determine which function to invoke. The special + * value {0, NULL} is used to indicate end-of-list, and must be + * present. The special value {0, <non-null>} can be used as a catch-all + * mapping. This implies that ctrl_id values chosen by the algorithm + * \ref MUST be non-zero. + */ +typedef const struct vpx_codec_ctrl_fn_map { + int ctrl_id; + vpx_codec_control_fn_t fn; +} vpx_codec_ctrl_fn_map_t; + +/*!\brief decode data function pointer prototype + * + * Processes a buffer of coded data. If the processing results in a new + * decoded frame becoming available, #VPX_CODEC_CB_PUT_SLICE and + * #VPX_CODEC_CB_PUT_FRAME events are generated as appropriate. This + * function is called by the generic vpx_codec_decode() wrapper function, + * so plugins implementing this interface may trust the input parameters + * to be properly initialized. + * + * \param[in] ctx Pointer to this instance's context + * \param[in] data Pointer to this block of new coded data. If + * NULL, a #VPX_CODEC_CB_PUT_FRAME event is posted + * for the previously decoded frame. + * \param[in] data_sz Size of the coded data, in bytes. + * + * \return Returns #VPX_CODEC_OK if the coded data was processed completely + * and future pictures can be decoded without error. Otherwise, + * see the descriptions of the other error codes in ::vpx_codec_err_t + * for recoverability capabilities. + */ +typedef vpx_codec_err_t (*vpx_codec_decode_fn_t)(vpx_codec_alg_priv_t *ctx, + const uint8_t *data, + unsigned int data_sz, + void *user_priv, + long deadline); + +/*!\brief Decoded frames iterator + * + * Iterates over a list of the frames available for display. The iterator + * storage should be initialized to NULL to start the iteration. Iteration is + * complete when this function returns NULL. + * + * The list of available frames becomes valid upon completion of the + * vpx_codec_decode call, and remains valid until the next call to vpx_codec_decode. + * + * \param[in] ctx Pointer to this instance's context + * \param[in out] iter Iterator storage, initialized to NULL + * + * \return Returns a pointer to an image, if one is ready for display. Frames + * produced will always be in PTS (presentation time stamp) order. + */ +typedef vpx_image_t *(*vpx_codec_get_frame_fn_t)(vpx_codec_alg_priv_t *ctx, + vpx_codec_iter_t *iter); + +/*!\brief Pass in external frame buffers for the decoder to use. + * + * Registers functions to be called when libvpx needs a frame buffer + * to decode the current frame and a function to be called when libvpx does + * not internally reference the frame buffer. This set function must + * be called before the first call to decode or libvpx will assume the + * default behavior of allocating frame buffers internally. + * + * \param[in] ctx Pointer to this instance's context + * \param[in] cb_get Pointer to the get callback function + * \param[in] cb_release Pointer to the release callback function + * \param[in] cb_priv Callback's private data + * + * \retval #VPX_CODEC_OK + * External frame buffers will be used by libvpx. + * \retval #VPX_CODEC_INVALID_PARAM + * One or more of the callbacks were NULL. + * \retval #VPX_CODEC_ERROR + * Decoder context not initialized, or algorithm not capable of + * using external frame buffers. + * + * \note + * When decoding VP9, the application may be required to pass in at least + * #VP9_MAXIMUM_REF_BUFFERS + #VPX_MAXIMUM_WORK_BUFFERS external frame + * buffers. + */ +typedef vpx_codec_err_t (*vpx_codec_set_fb_fn_t)( + vpx_codec_alg_priv_t *ctx, + vpx_get_frame_buffer_cb_fn_t cb_get, + vpx_release_frame_buffer_cb_fn_t cb_release, void *cb_priv); + + +typedef vpx_codec_err_t (*vpx_codec_encode_fn_t)(vpx_codec_alg_priv_t *ctx, + const vpx_image_t *img, + vpx_codec_pts_t pts, + unsigned long duration, + vpx_enc_frame_flags_t flags, + unsigned long deadline); +typedef const vpx_codec_cx_pkt_t *(*vpx_codec_get_cx_data_fn_t)(vpx_codec_alg_priv_t *ctx, + vpx_codec_iter_t *iter); + +typedef vpx_codec_err_t +(*vpx_codec_enc_config_set_fn_t)(vpx_codec_alg_priv_t *ctx, + const vpx_codec_enc_cfg_t *cfg); +typedef vpx_fixed_buf_t * +(*vpx_codec_get_global_headers_fn_t)(vpx_codec_alg_priv_t *ctx); + +typedef vpx_image_t * +(*vpx_codec_get_preview_frame_fn_t)(vpx_codec_alg_priv_t *ctx); + +typedef vpx_codec_err_t +(*vpx_codec_enc_mr_get_mem_loc_fn_t)(const vpx_codec_enc_cfg_t *cfg, + void **mem_loc); + +/*!\brief usage configuration mapping + * + * This structure stores the mapping between usage identifiers and + * configuration structures. Each algorithm provides a list of these + * mappings. This list is searched by the vpx_codec_enc_config_default() + * wrapper function to determine which config to return. The special value + * {-1, {0}} is used to indicate end-of-list, and must be present. At least + * one mapping must be present, in addition to the end-of-list. + * + */ +typedef const struct vpx_codec_enc_cfg_map { + int usage; + vpx_codec_enc_cfg_t cfg; +} vpx_codec_enc_cfg_map_t; + +/*!\brief Decoder algorithm interface interface + * + * All decoders \ref MUST expose a variable of this type. + */ +struct vpx_codec_iface { + const char *name; /**< Identification String */ + int abi_version; /**< Implemented ABI version */ + vpx_codec_caps_t caps; /**< Decoder capabilities */ + vpx_codec_init_fn_t init; /**< \copydoc ::vpx_codec_init_fn_t */ + vpx_codec_destroy_fn_t destroy; /**< \copydoc ::vpx_codec_destroy_fn_t */ + vpx_codec_ctrl_fn_map_t *ctrl_maps; /**< \copydoc ::vpx_codec_ctrl_fn_map_t */ + struct vpx_codec_dec_iface { + vpx_codec_peek_si_fn_t peek_si; /**< \copydoc ::vpx_codec_peek_si_fn_t */ + vpx_codec_get_si_fn_t get_si; /**< \copydoc ::vpx_codec_get_si_fn_t */ + vpx_codec_decode_fn_t decode; /**< \copydoc ::vpx_codec_decode_fn_t */ + vpx_codec_get_frame_fn_t get_frame; /**< \copydoc ::vpx_codec_get_frame_fn_t */ + vpx_codec_set_fb_fn_t set_fb_fn; /**< \copydoc ::vpx_codec_set_fb_fn_t */ + } dec; + struct vpx_codec_enc_iface { + int cfg_map_count; + vpx_codec_enc_cfg_map_t *cfg_maps; /**< \copydoc ::vpx_codec_enc_cfg_map_t */ + vpx_codec_encode_fn_t encode; /**< \copydoc ::vpx_codec_encode_fn_t */ + vpx_codec_get_cx_data_fn_t get_cx_data; /**< \copydoc ::vpx_codec_get_cx_data_fn_t */ + vpx_codec_enc_config_set_fn_t cfg_set; /**< \copydoc ::vpx_codec_enc_config_set_fn_t */ + vpx_codec_get_global_headers_fn_t get_glob_hdrs; /**< \copydoc ::vpx_codec_get_global_headers_fn_t */ + vpx_codec_get_preview_frame_fn_t get_preview; /**< \copydoc ::vpx_codec_get_preview_frame_fn_t */ + vpx_codec_enc_mr_get_mem_loc_fn_t mr_get_mem_loc; /**< \copydoc ::vpx_codec_enc_mr_get_mem_loc_fn_t */ + } enc; +}; + +/*!\brief Callback function pointer / user data pair storage */ +typedef struct vpx_codec_priv_cb_pair { + union { + vpx_codec_put_frame_cb_fn_t put_frame; + vpx_codec_put_slice_cb_fn_t put_slice; + } u; + void *user_priv; +} vpx_codec_priv_cb_pair_t; + + +/*!\brief Instance private storage + * + * This structure is allocated by the algorithm's init function. It can be + * extended in one of two ways. First, a second, algorithm specific structure + * can be allocated and the priv member pointed to it. Alternatively, this + * structure can be made the first member of the algorithm specific structure, + * and the pointer cast to the proper type. + */ +struct vpx_codec_priv { + const char *err_detail; + vpx_codec_flags_t init_flags; + struct { + vpx_codec_priv_cb_pair_t put_frame_cb; + vpx_codec_priv_cb_pair_t put_slice_cb; + } dec; + struct { + vpx_fixed_buf_t cx_data_dst_buf; + unsigned int cx_data_pad_before; + unsigned int cx_data_pad_after; + vpx_codec_cx_pkt_t cx_data_pkt; + unsigned int total_encoders; + } enc; +}; + +/* + * Multi-resolution encoding internal configuration + */ +struct vpx_codec_priv_enc_mr_cfg +{ + unsigned int mr_total_resolutions; + unsigned int mr_encoder_id; + struct vpx_rational mr_down_sampling_factor; + void* mr_low_res_mode_info; +}; + +#undef VPX_CTRL_USE_TYPE +#define VPX_CTRL_USE_TYPE(id, typ) \ + static VPX_INLINE typ id##__value(va_list args) {return va_arg(args, typ);} + +#undef VPX_CTRL_USE_TYPE_DEPRECATED +#define VPX_CTRL_USE_TYPE_DEPRECATED(id, typ) \ + static VPX_INLINE typ id##__value(va_list args) {return va_arg(args, typ);} + +#define CAST(id, arg) id##__value(arg) + +/* CODEC_INTERFACE convenience macro + * + * By convention, each codec interface is a struct with extern linkage, where + * the symbol is suffixed with _algo. A getter function is also defined to + * return a pointer to the struct, since in some cases it's easier to work + * with text symbols than data symbols (see issue #169). This function has + * the same name as the struct, less the _algo suffix. The CODEC_INTERFACE + * macro is provided to define this getter function automatically. + */ +#define CODEC_INTERFACE(id)\ + vpx_codec_iface_t* id(void) { return &id##_algo; }\ + vpx_codec_iface_t id##_algo + + +/* Internal Utility Functions + * + * The following functions are intended to be used inside algorithms as + * utilities for manipulating vpx_codec_* data structures. + */ +struct vpx_codec_pkt_list { + unsigned int cnt; + unsigned int max; + struct vpx_codec_cx_pkt pkts[1]; +}; + +#define vpx_codec_pkt_list_decl(n)\ + union {struct vpx_codec_pkt_list head;\ + struct {struct vpx_codec_pkt_list head;\ + struct vpx_codec_cx_pkt pkts[n];} alloc;} + +#define vpx_codec_pkt_list_init(m)\ + (m)->alloc.head.cnt = 0,\ + (m)->alloc.head.max = sizeof((m)->alloc.pkts) / sizeof((m)->alloc.pkts[0]) + +int +vpx_codec_pkt_list_add(struct vpx_codec_pkt_list *, + const struct vpx_codec_cx_pkt *); + +const vpx_codec_cx_pkt_t * +vpx_codec_pkt_list_get(struct vpx_codec_pkt_list *list, + vpx_codec_iter_t *iter); + + +#include <stdio.h> +#include <setjmp.h> + +struct vpx_internal_error_info { + vpx_codec_err_t error_code; + int has_detail; + char detail[80]; + int setjmp; + jmp_buf jmp; +}; + +#define CLANG_ANALYZER_NORETURN +#if defined(__has_feature) +#if __has_feature(attribute_analyzer_noreturn) +#undef CLANG_ANALYZER_NORETURN +#define CLANG_ANALYZER_NORETURN __attribute__((analyzer_noreturn)) +#endif +#endif + +void vpx_internal_error(struct vpx_internal_error_info *info, + vpx_codec_err_t error, + const char *fmt, + ...) CLANG_ANALYZER_NORETURN; + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VPX_INTERNAL_VPX_CODEC_INTERNAL_H_ diff --git a/thirdparty/libvpx/vpx/internal/vpx_psnr.h b/thirdparty/libvpx/vpx/internal/vpx_psnr.h new file mode 100644 index 0000000000..07d81bb8d9 --- /dev/null +++ b/thirdparty/libvpx/vpx/internal/vpx_psnr.h @@ -0,0 +1,34 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VPX_INTERNAL_VPX_PSNR_H_ +#define VPX_INTERNAL_VPX_PSNR_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +// TODO(dkovalev) change vpx_sse_to_psnr signature: double -> int64_t + +/*!\brief Converts SSE to PSNR + * + * Converts sum of squared errros (SSE) to peak signal-to-noise ratio (PNSR). + * + * \param[in] samples Number of samples + * \param[in] peak Max sample value + * \param[in] sse Sum of squared errors + */ +double vpx_sse_to_psnr(double samples, double peak, double sse); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VPX_INTERNAL_VPX_PSNR_H_ diff --git a/thirdparty/libvpx/vpx/src/vpx_codec.c b/thirdparty/libvpx/vpx/src/vpx_codec.c new file mode 100644 index 0000000000..5a495ce814 --- /dev/null +++ b/thirdparty/libvpx/vpx/src/vpx_codec.c @@ -0,0 +1,158 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +/*!\file + * \brief Provides the high level interface to wrap decoder algorithms. + * + */ +#include <stdarg.h> +#include <stdlib.h> +#include "vpx/vpx_integer.h" +#include "vpx/internal/vpx_codec_internal.h" +#include "vpx_version.h" + +#define SAVE_STATUS(ctx,var) (ctx?(ctx->err = var):var) + +int vpx_codec_version(void) { + return VERSION_PACKED; +} + + +const char *vpx_codec_version_str(void) { + return VERSION_STRING_NOSP; +} + + +const char *vpx_codec_version_extra_str(void) { + return VERSION_EXTRA; +} + + +const char *vpx_codec_iface_name(vpx_codec_iface_t *iface) { + return iface ? iface->name : "<invalid interface>"; +} + +const char *vpx_codec_err_to_string(vpx_codec_err_t err) { + switch (err) { + case VPX_CODEC_OK: + return "Success"; + case VPX_CODEC_ERROR: + return "Unspecified internal error"; + case VPX_CODEC_MEM_ERROR: + return "Memory allocation error"; + case VPX_CODEC_ABI_MISMATCH: + return "ABI version mismatch"; + case VPX_CODEC_INCAPABLE: + return "Codec does not implement requested capability"; + case VPX_CODEC_UNSUP_BITSTREAM: + return "Bitstream not supported by this decoder"; + case VPX_CODEC_UNSUP_FEATURE: + return "Bitstream required feature not supported by this decoder"; + case VPX_CODEC_CORRUPT_FRAME: + return "Corrupt frame detected"; + case VPX_CODEC_INVALID_PARAM: + return "Invalid parameter"; + case VPX_CODEC_LIST_END: + return "End of iterated list"; + } + + return "Unrecognized error code"; +} + +const char *vpx_codec_error(vpx_codec_ctx_t *ctx) { + return (ctx) ? vpx_codec_err_to_string(ctx->err) + : vpx_codec_err_to_string(VPX_CODEC_INVALID_PARAM); +} + +const char *vpx_codec_error_detail(vpx_codec_ctx_t *ctx) { + if (ctx && ctx->err) + return ctx->priv ? ctx->priv->err_detail : ctx->err_detail; + + return NULL; +} + + +vpx_codec_err_t vpx_codec_destroy(vpx_codec_ctx_t *ctx) { + vpx_codec_err_t res; + + if (!ctx) + res = VPX_CODEC_INVALID_PARAM; + else if (!ctx->iface || !ctx->priv) + res = VPX_CODEC_ERROR; + else { + ctx->iface->destroy((vpx_codec_alg_priv_t *)ctx->priv); + + ctx->iface = NULL; + ctx->name = NULL; + ctx->priv = NULL; + res = VPX_CODEC_OK; + } + + return SAVE_STATUS(ctx, res); +} + + +vpx_codec_caps_t vpx_codec_get_caps(vpx_codec_iface_t *iface) { + return (iface) ? iface->caps : 0; +} + + +vpx_codec_err_t vpx_codec_control_(vpx_codec_ctx_t *ctx, + int ctrl_id, + ...) { + vpx_codec_err_t res; + + if (!ctx || !ctrl_id) + res = VPX_CODEC_INVALID_PARAM; + else if (!ctx->iface || !ctx->priv || !ctx->iface->ctrl_maps) + res = VPX_CODEC_ERROR; + else { + vpx_codec_ctrl_fn_map_t *entry; + + res = VPX_CODEC_ERROR; + + for (entry = ctx->iface->ctrl_maps; entry && entry->fn; entry++) { + if (!entry->ctrl_id || entry->ctrl_id == ctrl_id) { + va_list ap; + + va_start(ap, ctrl_id); + res = entry->fn((vpx_codec_alg_priv_t *)ctx->priv, ap); + va_end(ap); + break; + } + } + } + + return SAVE_STATUS(ctx, res); +} + +void vpx_internal_error(struct vpx_internal_error_info *info, + vpx_codec_err_t error, + const char *fmt, + ...) { + va_list ap; + + info->error_code = error; + info->has_detail = 0; + + if (fmt) { + size_t sz = sizeof(info->detail); + + info->has_detail = 1; + va_start(ap, fmt); + vsnprintf(info->detail, sz - 1, fmt, ap); + va_end(ap); + info->detail[sz - 1] = '\0'; + } + + if (info->setjmp) + longjmp(info->jmp, info->error_code); +} diff --git a/thirdparty/libvpx/vpx/src/vpx_decoder.c b/thirdparty/libvpx/vpx/src/vpx_decoder.c new file mode 100644 index 0000000000..802d8edd8a --- /dev/null +++ b/thirdparty/libvpx/vpx/src/vpx_decoder.c @@ -0,0 +1,197 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +/*!\file + * \brief Provides the high level interface to wrap decoder algorithms. + * + */ +#include <string.h> +#include "vpx/internal/vpx_codec_internal.h" + +#define SAVE_STATUS(ctx,var) (ctx?(ctx->err = var):var) + +static vpx_codec_alg_priv_t *get_alg_priv(vpx_codec_ctx_t *ctx) { + return (vpx_codec_alg_priv_t *)ctx->priv; +} + +vpx_codec_err_t vpx_codec_dec_init_ver(vpx_codec_ctx_t *ctx, + vpx_codec_iface_t *iface, + const vpx_codec_dec_cfg_t *cfg, + vpx_codec_flags_t flags, + int ver) { + vpx_codec_err_t res; + + if (ver != VPX_DECODER_ABI_VERSION) + res = VPX_CODEC_ABI_MISMATCH; + else if (!ctx || !iface) + res = VPX_CODEC_INVALID_PARAM; + else if (iface->abi_version != VPX_CODEC_INTERNAL_ABI_VERSION) + res = VPX_CODEC_ABI_MISMATCH; + else if ((flags & VPX_CODEC_USE_POSTPROC) && !(iface->caps & VPX_CODEC_CAP_POSTPROC)) + res = VPX_CODEC_INCAPABLE; + else if ((flags & VPX_CODEC_USE_ERROR_CONCEALMENT) && + !(iface->caps & VPX_CODEC_CAP_ERROR_CONCEALMENT)) + res = VPX_CODEC_INCAPABLE; + else if ((flags & VPX_CODEC_USE_INPUT_FRAGMENTS) && + !(iface->caps & VPX_CODEC_CAP_INPUT_FRAGMENTS)) + res = VPX_CODEC_INCAPABLE; + else if (!(iface->caps & VPX_CODEC_CAP_DECODER)) + res = VPX_CODEC_INCAPABLE; + else { + memset(ctx, 0, sizeof(*ctx)); + ctx->iface = iface; + ctx->name = iface->name; + ctx->priv = NULL; + ctx->init_flags = flags; + ctx->config.dec = cfg; + + res = ctx->iface->init(ctx, NULL); + if (res) { + ctx->err_detail = ctx->priv ? ctx->priv->err_detail : NULL; + vpx_codec_destroy(ctx); + } + } + + return SAVE_STATUS(ctx, res); +} + + +vpx_codec_err_t vpx_codec_peek_stream_info(vpx_codec_iface_t *iface, + const uint8_t *data, + unsigned int data_sz, + vpx_codec_stream_info_t *si) { + vpx_codec_err_t res; + + if (!iface || !data || !data_sz || !si + || si->sz < sizeof(vpx_codec_stream_info_t)) + res = VPX_CODEC_INVALID_PARAM; + else { + /* Set default/unknown values */ + si->w = 0; + si->h = 0; + + res = iface->dec.peek_si(data, data_sz, si); + } + + return res; +} + + +vpx_codec_err_t vpx_codec_get_stream_info(vpx_codec_ctx_t *ctx, + vpx_codec_stream_info_t *si) { + vpx_codec_err_t res; + + if (!ctx || !si || si->sz < sizeof(vpx_codec_stream_info_t)) + res = VPX_CODEC_INVALID_PARAM; + else if (!ctx->iface || !ctx->priv) + res = VPX_CODEC_ERROR; + else { + /* Set default/unknown values */ + si->w = 0; + si->h = 0; + + res = ctx->iface->dec.get_si(get_alg_priv(ctx), si); + } + + return SAVE_STATUS(ctx, res); +} + + +vpx_codec_err_t vpx_codec_decode(vpx_codec_ctx_t *ctx, + const uint8_t *data, + unsigned int data_sz, + void *user_priv, + long deadline) { + vpx_codec_err_t res; + + /* Sanity checks */ + /* NULL data ptr allowed if data_sz is 0 too */ + if (!ctx || (!data && data_sz) || (data && !data_sz)) + res = VPX_CODEC_INVALID_PARAM; + else if (!ctx->iface || !ctx->priv) + res = VPX_CODEC_ERROR; + else { + res = ctx->iface->dec.decode(get_alg_priv(ctx), data, data_sz, user_priv, + deadline); + } + + return SAVE_STATUS(ctx, res); +} + +vpx_image_t *vpx_codec_get_frame(vpx_codec_ctx_t *ctx, + vpx_codec_iter_t *iter) { + vpx_image_t *img; + + if (!ctx || !iter || !ctx->iface || !ctx->priv) + img = NULL; + else + img = ctx->iface->dec.get_frame(get_alg_priv(ctx), iter); + + return img; +} + + +vpx_codec_err_t vpx_codec_register_put_frame_cb(vpx_codec_ctx_t *ctx, + vpx_codec_put_frame_cb_fn_t cb, + void *user_priv) { + vpx_codec_err_t res; + + if (!ctx || !cb) + res = VPX_CODEC_INVALID_PARAM; + else if (!ctx->iface || !ctx->priv + || !(ctx->iface->caps & VPX_CODEC_CAP_PUT_FRAME)) + res = VPX_CODEC_ERROR; + else { + ctx->priv->dec.put_frame_cb.u.put_frame = cb; + ctx->priv->dec.put_frame_cb.user_priv = user_priv; + res = VPX_CODEC_OK; + } + + return SAVE_STATUS(ctx, res); +} + + +vpx_codec_err_t vpx_codec_register_put_slice_cb(vpx_codec_ctx_t *ctx, + vpx_codec_put_slice_cb_fn_t cb, + void *user_priv) { + vpx_codec_err_t res; + + if (!ctx || !cb) + res = VPX_CODEC_INVALID_PARAM; + else if (!ctx->iface || !ctx->priv + || !(ctx->iface->caps & VPX_CODEC_CAP_PUT_SLICE)) + res = VPX_CODEC_ERROR; + else { + ctx->priv->dec.put_slice_cb.u.put_slice = cb; + ctx->priv->dec.put_slice_cb.user_priv = user_priv; + res = VPX_CODEC_OK; + } + + return SAVE_STATUS(ctx, res); +} + +vpx_codec_err_t vpx_codec_set_frame_buffer_functions( + vpx_codec_ctx_t *ctx, vpx_get_frame_buffer_cb_fn_t cb_get, + vpx_release_frame_buffer_cb_fn_t cb_release, void *cb_priv) { + vpx_codec_err_t res; + + if (!ctx || !cb_get || !cb_release) { + res = VPX_CODEC_INVALID_PARAM; + } else if (!ctx->iface || !ctx->priv || + !(ctx->iface->caps & VPX_CODEC_CAP_EXTERNAL_FRAME_BUFFER)) { + res = VPX_CODEC_ERROR; + } else { + res = ctx->iface->dec.set_fb_fn(get_alg_priv(ctx), cb_get, cb_release, + cb_priv); + } + + return SAVE_STATUS(ctx, res); +} diff --git a/thirdparty/libvpx/vpx/src/vpx_image.c b/thirdparty/libvpx/vpx/src/vpx_image.c new file mode 100644 index 0000000000..9aae12c794 --- /dev/null +++ b/thirdparty/libvpx/vpx/src/vpx_image.c @@ -0,0 +1,285 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <stdlib.h> +#include <string.h> + +#include "vpx/vpx_image.h" +#include "vpx/vpx_integer.h" +#include "vpx_mem/vpx_mem.h" + +static vpx_image_t *img_alloc_helper(vpx_image_t *img, + vpx_img_fmt_t fmt, + unsigned int d_w, + unsigned int d_h, + unsigned int buf_align, + unsigned int stride_align, + unsigned char *img_data) { + unsigned int h, w, s, xcs, ycs, bps; + unsigned int stride_in_bytes; + int align; + + /* Treat align==0 like align==1 */ + if (!buf_align) + buf_align = 1; + + /* Validate alignment (must be power of 2) */ + if (buf_align & (buf_align - 1)) + goto fail; + + /* Treat align==0 like align==1 */ + if (!stride_align) + stride_align = 1; + + /* Validate alignment (must be power of 2) */ + if (stride_align & (stride_align - 1)) + goto fail; + + /* Get sample size for this format */ + switch (fmt) { + case VPX_IMG_FMT_RGB32: + case VPX_IMG_FMT_RGB32_LE: + case VPX_IMG_FMT_ARGB: + case VPX_IMG_FMT_ARGB_LE: + bps = 32; + break; + case VPX_IMG_FMT_RGB24: + case VPX_IMG_FMT_BGR24: + bps = 24; + break; + case VPX_IMG_FMT_RGB565: + case VPX_IMG_FMT_RGB565_LE: + case VPX_IMG_FMT_RGB555: + case VPX_IMG_FMT_RGB555_LE: + case VPX_IMG_FMT_UYVY: + case VPX_IMG_FMT_YUY2: + case VPX_IMG_FMT_YVYU: + bps = 16; + break; + case VPX_IMG_FMT_I420: + case VPX_IMG_FMT_YV12: + case VPX_IMG_FMT_VPXI420: + case VPX_IMG_FMT_VPXYV12: + bps = 12; + break; + case VPX_IMG_FMT_I422: + case VPX_IMG_FMT_I440: + bps = 16; + break; + case VPX_IMG_FMT_I444: + bps = 24; + break; + case VPX_IMG_FMT_I42016: + bps = 24; + break; + case VPX_IMG_FMT_I42216: + case VPX_IMG_FMT_I44016: + bps = 32; + break; + case VPX_IMG_FMT_I44416: + bps = 48; + break; + default: + bps = 16; + break; + } + + /* Get chroma shift values for this format */ + switch (fmt) { + case VPX_IMG_FMT_I420: + case VPX_IMG_FMT_YV12: + case VPX_IMG_FMT_VPXI420: + case VPX_IMG_FMT_VPXYV12: + case VPX_IMG_FMT_I422: + case VPX_IMG_FMT_I42016: + case VPX_IMG_FMT_I42216: + xcs = 1; + break; + default: + xcs = 0; + break; + } + + switch (fmt) { + case VPX_IMG_FMT_I420: + case VPX_IMG_FMT_I440: + case VPX_IMG_FMT_YV12: + case VPX_IMG_FMT_VPXI420: + case VPX_IMG_FMT_VPXYV12: + case VPX_IMG_FMT_I42016: + case VPX_IMG_FMT_I44016: + ycs = 1; + break; + default: + ycs = 0; + break; + } + + /* Calculate storage sizes given the chroma subsampling */ + align = (1 << xcs) - 1; + w = (d_w + align) & ~align; + align = (1 << ycs) - 1; + h = (d_h + align) & ~align; + s = (fmt & VPX_IMG_FMT_PLANAR) ? w : bps * w / 8; + s = (s + stride_align - 1) & ~(stride_align - 1); + stride_in_bytes = (fmt & VPX_IMG_FMT_HIGHBITDEPTH) ? s * 2 : s; + + /* Allocate the new image */ + if (!img) { + img = (vpx_image_t *)calloc(1, sizeof(vpx_image_t)); + + if (!img) + goto fail; + + img->self_allocd = 1; + } else { + memset(img, 0, sizeof(vpx_image_t)); + } + + img->img_data = img_data; + + if (!img_data) { + const uint64_t alloc_size = (fmt & VPX_IMG_FMT_PLANAR) ? + (uint64_t)h * s * bps / 8 : (uint64_t)h * s; + + if (alloc_size != (size_t)alloc_size) + goto fail; + + img->img_data = (uint8_t *)vpx_memalign(buf_align, (size_t)alloc_size); + img->img_data_owner = 1; + } + + if (!img->img_data) + goto fail; + + img->fmt = fmt; + img->bit_depth = (fmt & VPX_IMG_FMT_HIGHBITDEPTH) ? 16 : 8; + img->w = w; + img->h = h; + img->x_chroma_shift = xcs; + img->y_chroma_shift = ycs; + img->bps = bps; + + /* Calculate strides */ + img->stride[VPX_PLANE_Y] = img->stride[VPX_PLANE_ALPHA] = stride_in_bytes; + img->stride[VPX_PLANE_U] = img->stride[VPX_PLANE_V] = stride_in_bytes >> xcs; + + /* Default viewport to entire image */ + if (!vpx_img_set_rect(img, 0, 0, d_w, d_h)) + return img; + +fail: + vpx_img_free(img); + return NULL; +} + +vpx_image_t *vpx_img_alloc(vpx_image_t *img, + vpx_img_fmt_t fmt, + unsigned int d_w, + unsigned int d_h, + unsigned int align) { + return img_alloc_helper(img, fmt, d_w, d_h, align, align, NULL); +} + +vpx_image_t *vpx_img_wrap(vpx_image_t *img, + vpx_img_fmt_t fmt, + unsigned int d_w, + unsigned int d_h, + unsigned int stride_align, + unsigned char *img_data) { + /* By setting buf_align = 1, we don't change buffer alignment in this + * function. */ + return img_alloc_helper(img, fmt, d_w, d_h, 1, stride_align, img_data); +} + +int vpx_img_set_rect(vpx_image_t *img, + unsigned int x, + unsigned int y, + unsigned int w, + unsigned int h) { + unsigned char *data; + + if (x + w <= img->w && y + h <= img->h) { + img->d_w = w; + img->d_h = h; + + /* Calculate plane pointers */ + if (!(img->fmt & VPX_IMG_FMT_PLANAR)) { + img->planes[VPX_PLANE_PACKED] = + img->img_data + x * img->bps / 8 + y * img->stride[VPX_PLANE_PACKED]; + } else { + const int bytes_per_sample = + (img->fmt & VPX_IMG_FMT_HIGHBITDEPTH) ? 2 : 1; + data = img->img_data; + + if (img->fmt & VPX_IMG_FMT_HAS_ALPHA) { + img->planes[VPX_PLANE_ALPHA] = + data + x * bytes_per_sample + y * img->stride[VPX_PLANE_ALPHA]; + data += img->h * img->stride[VPX_PLANE_ALPHA]; + } + + img->planes[VPX_PLANE_Y] = data + x * bytes_per_sample + + y * img->stride[VPX_PLANE_Y]; + data += img->h * img->stride[VPX_PLANE_Y]; + + if (!(img->fmt & VPX_IMG_FMT_UV_FLIP)) { + img->planes[VPX_PLANE_U] = + data + (x >> img->x_chroma_shift) * bytes_per_sample + + (y >> img->y_chroma_shift) * img->stride[VPX_PLANE_U]; + data += (img->h >> img->y_chroma_shift) * img->stride[VPX_PLANE_U]; + img->planes[VPX_PLANE_V] = + data + (x >> img->x_chroma_shift) * bytes_per_sample + + (y >> img->y_chroma_shift) * img->stride[VPX_PLANE_V]; + } else { + img->planes[VPX_PLANE_V] = + data + (x >> img->x_chroma_shift) * bytes_per_sample + + (y >> img->y_chroma_shift) * img->stride[VPX_PLANE_V]; + data += (img->h >> img->y_chroma_shift) * img->stride[VPX_PLANE_V]; + img->planes[VPX_PLANE_U] = + data + (x >> img->x_chroma_shift) * bytes_per_sample + + (y >> img->y_chroma_shift) * img->stride[VPX_PLANE_U]; + } + } + return 0; + } + return -1; +} + +void vpx_img_flip(vpx_image_t *img) { + /* Note: In the calculation pointer adjustment calculation, we want the + * rhs to be promoted to a signed type. Section 6.3.1.8 of the ISO C99 + * standard indicates that if the adjustment parameter is unsigned, the + * stride parameter will be promoted to unsigned, causing errors when + * the lhs is a larger type than the rhs. + */ + img->planes[VPX_PLANE_Y] += (signed)(img->d_h - 1) * img->stride[VPX_PLANE_Y]; + img->stride[VPX_PLANE_Y] = -img->stride[VPX_PLANE_Y]; + + img->planes[VPX_PLANE_U] += (signed)((img->d_h >> img->y_chroma_shift) - 1) + * img->stride[VPX_PLANE_U]; + img->stride[VPX_PLANE_U] = -img->stride[VPX_PLANE_U]; + + img->planes[VPX_PLANE_V] += (signed)((img->d_h >> img->y_chroma_shift) - 1) + * img->stride[VPX_PLANE_V]; + img->stride[VPX_PLANE_V] = -img->stride[VPX_PLANE_V]; + + img->planes[VPX_PLANE_ALPHA] += (signed)(img->d_h - 1) * img->stride[VPX_PLANE_ALPHA]; + img->stride[VPX_PLANE_ALPHA] = -img->stride[VPX_PLANE_ALPHA]; +} + +void vpx_img_free(vpx_image_t *img) { + if (img) { + if (img->img_data && img->img_data_owner) + vpx_free(img->img_data); + + if (img->self_allocd) + free(img); + } +} diff --git a/thirdparty/libvpx/vpx/src/vpx_psnr.c b/thirdparty/libvpx/vpx/src/vpx_psnr.c new file mode 100644 index 0000000000..05843acb61 --- /dev/null +++ b/thirdparty/libvpx/vpx/src/vpx_psnr.c @@ -0,0 +1,24 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <math.h> + +#include "vpx/internal/vpx_psnr.h" + +#define MAX_PSNR 100.0 + +double vpx_sse_to_psnr(double samples, double peak, double sse) { + if (sse > 0.0) { + const double psnr = 10.0 * log10(samples * peak * peak / sse); + return psnr > MAX_PSNR ? MAX_PSNR : psnr; + } else { + return MAX_PSNR; + } +} diff --git a/thirdparty/libvpx/vpx/vp8.h b/thirdparty/libvpx/vpx/vp8.h new file mode 100644 index 0000000000..8a035f9770 --- /dev/null +++ b/thirdparty/libvpx/vpx/vp8.h @@ -0,0 +1,148 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +/*!\defgroup vp8 VP8 + * \ingroup codecs + * VP8 is vpx's newest video compression algorithm that uses motion + * compensated prediction, Discrete Cosine Transform (DCT) coding of the + * prediction error signal and context dependent entropy coding techniques + * based on arithmetic principles. It features: + * - YUV 4:2:0 image format + * - Macro-block based coding (16x16 luma plus two 8x8 chroma) + * - 1/4 (1/8) pixel accuracy motion compensated prediction + * - 4x4 DCT transform + * - 128 level linear quantizer + * - In loop deblocking filter + * - Context-based entropy coding + * + * @{ + */ +/*!\file + * \brief Provides controls common to both the VP8 encoder and decoder. + */ +#ifndef VPX_VP8_H_ +#define VPX_VP8_H_ + +#include "./vpx_codec.h" +#include "./vpx_image.h" + +#ifdef __cplusplus +extern "C" { +#endif + +/*!\brief Control functions + * + * The set of macros define the control functions of VP8 interface + */ +enum vp8_com_control_id { + VP8_SET_REFERENCE = 1, /**< pass in an external frame into decoder to be used as reference frame */ + VP8_COPY_REFERENCE = 2, /**< get a copy of reference frame from the decoder */ + VP8_SET_POSTPROC = 3, /**< set the decoder's post processing settings */ + VP8_SET_DBG_COLOR_REF_FRAME = 4, /**< set the reference frames to color for each macroblock */ + VP8_SET_DBG_COLOR_MB_MODES = 5, /**< set which macro block modes to color */ + VP8_SET_DBG_COLOR_B_MODES = 6, /**< set which blocks modes to color */ + VP8_SET_DBG_DISPLAY_MV = 7, /**< set which motion vector modes to draw */ + + /* TODO(jkoleszar): The encoder incorrectly reuses some of these values (5+) + * for its control ids. These should be migrated to something like the + * VP8_DECODER_CTRL_ID_START range next time we're ready to break the ABI. + */ + VP9_GET_REFERENCE = 128, /**< get a pointer to a reference frame */ + VP8_COMMON_CTRL_ID_MAX, + VP8_DECODER_CTRL_ID_START = 256 +}; + +/*!\brief post process flags + * + * The set of macros define VP8 decoder post processing flags + */ +enum vp8_postproc_level { + VP8_NOFILTERING = 0, + VP8_DEBLOCK = 1 << 0, + VP8_DEMACROBLOCK = 1 << 1, + VP8_ADDNOISE = 1 << 2, + VP8_DEBUG_TXT_FRAME_INFO = 1 << 3, /**< print frame information */ + VP8_DEBUG_TXT_MBLK_MODES = 1 << 4, /**< print macro block modes over each macro block */ + VP8_DEBUG_TXT_DC_DIFF = 1 << 5, /**< print dc diff for each macro block */ + VP8_DEBUG_TXT_RATE_INFO = 1 << 6, /**< print video rate info (encoder only) */ + VP8_MFQE = 1 << 10 +}; + +/*!\brief post process flags + * + * This define a structure that describe the post processing settings. For + * the best objective measure (using the PSNR metric) set post_proc_flag + * to VP8_DEBLOCK and deblocking_level to 1. + */ + +typedef struct vp8_postproc_cfg { + int post_proc_flag; /**< the types of post processing to be done, should be combination of "vp8_postproc_level" */ + int deblocking_level; /**< the strength of deblocking, valid range [0, 16] */ + int noise_level; /**< the strength of additive noise, valid range [0, 16] */ +} vp8_postproc_cfg_t; + +/*!\brief reference frame type + * + * The set of macros define the type of VP8 reference frames + */ +typedef enum vpx_ref_frame_type { + VP8_LAST_FRAME = 1, + VP8_GOLD_FRAME = 2, + VP8_ALTR_FRAME = 4 +} vpx_ref_frame_type_t; + +/*!\brief reference frame data struct + * + * Define the data struct to access vp8 reference frames. + */ +typedef struct vpx_ref_frame { + vpx_ref_frame_type_t frame_type; /**< which reference frame */ + vpx_image_t img; /**< reference frame data in image format */ +} vpx_ref_frame_t; + +/*!\brief VP9 specific reference frame data struct + * + * Define the data struct to access vp9 reference frames. + */ +typedef struct vp9_ref_frame { + int idx; /**< frame index to get (input) */ + vpx_image_t img; /**< img structure to populate (output) */ +} vp9_ref_frame_t; + +/*!\cond */ +/*!\brief vp8 decoder control function parameter type + * + * defines the data type for each of VP8 decoder control function requires + */ +VPX_CTRL_USE_TYPE(VP8_SET_REFERENCE, vpx_ref_frame_t *) +#define VPX_CTRL_VP8_SET_REFERENCE +VPX_CTRL_USE_TYPE(VP8_COPY_REFERENCE, vpx_ref_frame_t *) +#define VPX_CTRL_VP8_COPY_REFERENCE +VPX_CTRL_USE_TYPE(VP8_SET_POSTPROC, vp8_postproc_cfg_t *) +#define VPX_CTRL_VP8_SET_POSTPROC +VPX_CTRL_USE_TYPE(VP8_SET_DBG_COLOR_REF_FRAME, int) +#define VPX_CTRL_VP8_SET_DBG_COLOR_REF_FRAME +VPX_CTRL_USE_TYPE(VP8_SET_DBG_COLOR_MB_MODES, int) +#define VPX_CTRL_VP8_SET_DBG_COLOR_MB_MODES +VPX_CTRL_USE_TYPE(VP8_SET_DBG_COLOR_B_MODES, int) +#define VPX_CTRL_VP8_SET_DBG_COLOR_B_MODES +VPX_CTRL_USE_TYPE(VP8_SET_DBG_DISPLAY_MV, int) +#define VPX_CTRL_VP8_SET_DBG_DISPLAY_MV +VPX_CTRL_USE_TYPE(VP9_GET_REFERENCE, vp9_ref_frame_t *) +#define VPX_CTRL_VP9_GET_REFERENCE + +/*!\endcond */ +/*! @} - end defgroup vp8 */ + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VPX_VP8_H_ diff --git a/thirdparty/libvpx/vpx/vp8dx.h b/thirdparty/libvpx/vpx/vp8dx.h new file mode 100644 index 0000000000..67c97bb6c9 --- /dev/null +++ b/thirdparty/libvpx/vpx/vp8dx.h @@ -0,0 +1,176 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +/*!\defgroup vp8_decoder WebM VP8/VP9 Decoder + * \ingroup vp8 + * + * @{ + */ +/*!\file + * \brief Provides definitions for using VP8 or VP9 within the vpx Decoder + * interface. + */ +#ifndef VPX_VP8DX_H_ +#define VPX_VP8DX_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +/* Include controls common to both the encoder and decoder */ +#include "./vp8.h" + +/*!\name Algorithm interface for VP8 + * + * This interface provides the capability to decode VP8 streams. + * @{ + */ +extern vpx_codec_iface_t vpx_codec_vp8_dx_algo; +extern vpx_codec_iface_t *vpx_codec_vp8_dx(void); +/*!@} - end algorithm interface member group*/ + +/*!\name Algorithm interface for VP9 + * + * This interface provides the capability to decode VP9 streams. + * @{ + */ +extern vpx_codec_iface_t vpx_codec_vp9_dx_algo; +extern vpx_codec_iface_t *vpx_codec_vp9_dx(void); +/*!@} - end algorithm interface member group*/ + +/*!\enum vp8_dec_control_id + * \brief VP8 decoder control functions + * + * This set of macros define the control functions available for the VP8 + * decoder interface. + * + * \sa #vpx_codec_control + */ +enum vp8_dec_control_id { + /** control function to get info on which reference frames were updated + * by the last decode + */ + VP8D_GET_LAST_REF_UPDATES = VP8_DECODER_CTRL_ID_START, + + /** check if the indicated frame is corrupted */ + VP8D_GET_FRAME_CORRUPTED, + + /** control function to get info on which reference frames were used + * by the last decode + */ + VP8D_GET_LAST_REF_USED, + + /** decryption function to decrypt encoded buffer data immediately + * before decoding. Takes a vpx_decrypt_init, which contains + * a callback function and opaque context pointer. + */ + VPXD_SET_DECRYPTOR, + VP8D_SET_DECRYPTOR = VPXD_SET_DECRYPTOR, + + /** control function to get the dimensions that the current frame is decoded + * at. This may be different to the intended display size for the frame as + * specified in the wrapper or frame header (see VP9D_GET_DISPLAY_SIZE). */ + VP9D_GET_FRAME_SIZE, + + /** control function to get the current frame's intended display dimensions + * (as specified in the wrapper or frame header). This may be different to + * the decoded dimensions of this frame (see VP9D_GET_FRAME_SIZE). */ + VP9D_GET_DISPLAY_SIZE, + + /** control function to get the bit depth of the stream. */ + VP9D_GET_BIT_DEPTH, + + /** control function to set the byte alignment of the planes in the reference + * buffers. Valid values are power of 2, from 32 to 1024. A value of 0 sets + * legacy alignment. I.e. Y plane is aligned to 32 bytes, U plane directly + * follows Y plane, and V plane directly follows U plane. Default value is 0. + */ + VP9_SET_BYTE_ALIGNMENT, + + /** control function to invert the decoding order to from right to left. The + * function is used in a test to confirm the decoding independence of tile + * columns. The function may be used in application where this order + * of decoding is desired. + * + * TODO(yaowu): Rework the unit test that uses this control, and in a future + * release, this test-only control shall be removed. + */ + VP9_INVERT_TILE_DECODE_ORDER, + + /** control function to set the skip loop filter flag. Valid values are + * integers. The decoder will skip the loop filter when its value is set to + * nonzero. If the loop filter is skipped the decoder may accumulate decode + * artifacts. The default value is 0. + */ + VP9_SET_SKIP_LOOP_FILTER, + + VP8_DECODER_CTRL_ID_MAX +}; + +/** Decrypt n bytes of data from input -> output, using the decrypt_state + * passed in VPXD_SET_DECRYPTOR. + */ +typedef void (*vpx_decrypt_cb)(void *decrypt_state, const unsigned char *input, + unsigned char *output, int count); + +/*!\brief Structure to hold decryption state + * + * Defines a structure to hold the decryption state and access function. + */ +typedef struct vpx_decrypt_init { + /*! Decrypt callback. */ + vpx_decrypt_cb decrypt_cb; + + /*! Decryption state. */ + void *decrypt_state; +} vpx_decrypt_init; + +/*!\brief A deprecated alias for vpx_decrypt_init. + */ +typedef vpx_decrypt_init vp8_decrypt_init; + + +/*!\cond */ +/*!\brief VP8 decoder control function parameter type + * + * Defines the data types that VP8D control functions take. Note that + * additional common controls are defined in vp8.h + * + */ + + +VPX_CTRL_USE_TYPE(VP8D_GET_LAST_REF_UPDATES, int *) +#define VPX_CTRL_VP8D_GET_LAST_REF_UPDATES +VPX_CTRL_USE_TYPE(VP8D_GET_FRAME_CORRUPTED, int *) +#define VPX_CTRL_VP8D_GET_FRAME_CORRUPTED +VPX_CTRL_USE_TYPE(VP8D_GET_LAST_REF_USED, int *) +#define VPX_CTRL_VP8D_GET_LAST_REF_USED +VPX_CTRL_USE_TYPE(VPXD_SET_DECRYPTOR, vpx_decrypt_init *) +#define VPX_CTRL_VPXD_SET_DECRYPTOR +VPX_CTRL_USE_TYPE(VP8D_SET_DECRYPTOR, vpx_decrypt_init *) +#define VPX_CTRL_VP8D_SET_DECRYPTOR +VPX_CTRL_USE_TYPE(VP9D_GET_DISPLAY_SIZE, int *) +#define VPX_CTRL_VP9D_GET_DISPLAY_SIZE +VPX_CTRL_USE_TYPE(VP9D_GET_BIT_DEPTH, unsigned int *) +#define VPX_CTRL_VP9D_GET_BIT_DEPTH +VPX_CTRL_USE_TYPE(VP9D_GET_FRAME_SIZE, int *) +#define VPX_CTRL_VP9D_GET_FRAME_SIZE +VPX_CTRL_USE_TYPE(VP9_INVERT_TILE_DECODE_ORDER, int) +#define VPX_CTRL_VP9_INVERT_TILE_DECODE_ORDER + +/*!\endcond */ +/*! @} - end defgroup vp8_decoder */ + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VPX_VP8DX_H_ diff --git a/thirdparty/libvpx/vpx/vpx_codec.h b/thirdparty/libvpx/vpx/vpx_codec.h new file mode 100644 index 0000000000..b6037bb4d7 --- /dev/null +++ b/thirdparty/libvpx/vpx/vpx_codec.h @@ -0,0 +1,479 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +/*!\defgroup codec Common Algorithm Interface + * This abstraction allows applications to easily support multiple video + * formats with minimal code duplication. This section describes the interface + * common to all codecs (both encoders and decoders). + * @{ + */ + +/*!\file + * \brief Describes the codec algorithm interface to applications. + * + * This file describes the interface between an application and a + * video codec algorithm. + * + * An application instantiates a specific codec instance by using + * vpx_codec_init() and a pointer to the algorithm's interface structure: + * <pre> + * my_app.c: + * extern vpx_codec_iface_t my_codec; + * { + * vpx_codec_ctx_t algo; + * res = vpx_codec_init(&algo, &my_codec); + * } + * </pre> + * + * Once initialized, the instance is manged using other functions from + * the vpx_codec_* family. + */ +#ifndef VPX_VPX_CODEC_H_ +#define VPX_VPX_CODEC_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +#include "./vpx_integer.h" +#include "./vpx_image.h" + + /*!\brief Decorator indicating a function is deprecated */ +#ifndef DEPRECATED +#if defined(__GNUC__) && __GNUC__ +#define DEPRECATED __attribute__ ((deprecated)) +#elif defined(_MSC_VER) +#define DEPRECATED +#else +#define DEPRECATED +#endif +#endif /* DEPRECATED */ + +#ifndef DECLSPEC_DEPRECATED +#if defined(__GNUC__) && __GNUC__ +#define DECLSPEC_DEPRECATED /**< \copydoc #DEPRECATED */ +#elif defined(_MSC_VER) +#define DECLSPEC_DEPRECATED __declspec(deprecated) /**< \copydoc #DEPRECATED */ +#else +#define DECLSPEC_DEPRECATED /**< \copydoc #DEPRECATED */ +#endif +#endif /* DECLSPEC_DEPRECATED */ + + /*!\brief Decorator indicating a function is potentially unused */ +#ifdef UNUSED +#elif defined(__GNUC__) || defined(__clang__) +#define UNUSED __attribute__ ((unused)) +#else +#define UNUSED +#endif + + /*!\brief Current ABI version number + * + * \internal + * If this file is altered in any way that changes the ABI, this value + * must be bumped. Examples include, but are not limited to, changing + * types, removing or reassigning enums, adding/removing/rearranging + * fields to structures + */ +#define VPX_CODEC_ABI_VERSION (3 + VPX_IMAGE_ABI_VERSION) /**<\hideinitializer*/ + + /*!\brief Algorithm return codes */ + typedef enum { + /*!\brief Operation completed without error */ + VPX_CODEC_OK, + + /*!\brief Unspecified error */ + VPX_CODEC_ERROR, + + /*!\brief Memory operation failed */ + VPX_CODEC_MEM_ERROR, + + /*!\brief ABI version mismatch */ + VPX_CODEC_ABI_MISMATCH, + + /*!\brief Algorithm does not have required capability */ + VPX_CODEC_INCAPABLE, + + /*!\brief The given bitstream is not supported. + * + * The bitstream was unable to be parsed at the highest level. The decoder + * is unable to proceed. This error \ref SHOULD be treated as fatal to the + * stream. */ + VPX_CODEC_UNSUP_BITSTREAM, + + /*!\brief Encoded bitstream uses an unsupported feature + * + * The decoder does not implement a feature required by the encoder. This + * return code should only be used for features that prevent future + * pictures from being properly decoded. This error \ref MAY be treated as + * fatal to the stream or \ref MAY be treated as fatal to the current GOP. + */ + VPX_CODEC_UNSUP_FEATURE, + + /*!\brief The coded data for this stream is corrupt or incomplete + * + * There was a problem decoding the current frame. This return code + * should only be used for failures that prevent future pictures from + * being properly decoded. This error \ref MAY be treated as fatal to the + * stream or \ref MAY be treated as fatal to the current GOP. If decoding + * is continued for the current GOP, artifacts may be present. + */ + VPX_CODEC_CORRUPT_FRAME, + + /*!\brief An application-supplied parameter is not valid. + * + */ + VPX_CODEC_INVALID_PARAM, + + /*!\brief An iterator reached the end of list. + * + */ + VPX_CODEC_LIST_END + + } + vpx_codec_err_t; + + + /*! \brief Codec capabilities bitfield + * + * Each codec advertises the capabilities it supports as part of its + * ::vpx_codec_iface_t interface structure. Capabilities are extra interfaces + * or functionality, and are not required to be supported. + * + * The available flags are specified by VPX_CODEC_CAP_* defines. + */ + typedef long vpx_codec_caps_t; +#define VPX_CODEC_CAP_DECODER 0x1 /**< Is a decoder */ +#define VPX_CODEC_CAP_ENCODER 0x2 /**< Is an encoder */ + + + /*! \brief Initialization-time Feature Enabling + * + * Certain codec features must be known at initialization time, to allow for + * proper memory allocation. + * + * The available flags are specified by VPX_CODEC_USE_* defines. + */ + typedef long vpx_codec_flags_t; + + + /*!\brief Codec interface structure. + * + * Contains function pointers and other data private to the codec + * implementation. This structure is opaque to the application. + */ + typedef const struct vpx_codec_iface vpx_codec_iface_t; + + + /*!\brief Codec private data structure. + * + * Contains data private to the codec implementation. This structure is opaque + * to the application. + */ + typedef struct vpx_codec_priv vpx_codec_priv_t; + + + /*!\brief Iterator + * + * Opaque storage used for iterating over lists. + */ + typedef const void *vpx_codec_iter_t; + + + /*!\brief Codec context structure + * + * All codecs \ref MUST support this context structure fully. In general, + * this data should be considered private to the codec algorithm, and + * not be manipulated or examined by the calling application. Applications + * may reference the 'name' member to get a printable description of the + * algorithm. + */ + typedef struct vpx_codec_ctx { + const char *name; /**< Printable interface name */ + vpx_codec_iface_t *iface; /**< Interface pointers */ + vpx_codec_err_t err; /**< Last returned error */ + const char *err_detail; /**< Detailed info, if available */ + vpx_codec_flags_t init_flags; /**< Flags passed at init time */ + union { + /**< Decoder Configuration Pointer */ + const struct vpx_codec_dec_cfg *dec; + /**< Encoder Configuration Pointer */ + const struct vpx_codec_enc_cfg *enc; + const void *raw; + } config; /**< Configuration pointer aliasing union */ + vpx_codec_priv_t *priv; /**< Algorithm private storage */ + } vpx_codec_ctx_t; + + /*!\brief Bit depth for codec + * * + * This enumeration determines the bit depth of the codec. + */ + typedef enum vpx_bit_depth { + VPX_BITS_8 = 8, /**< 8 bits */ + VPX_BITS_10 = 10, /**< 10 bits */ + VPX_BITS_12 = 12, /**< 12 bits */ + } vpx_bit_depth_t; + + /* + * Library Version Number Interface + * + * For example, see the following sample return values: + * vpx_codec_version() (1<<16 | 2<<8 | 3) + * vpx_codec_version_str() "v1.2.3-rc1-16-gec6a1ba" + * vpx_codec_version_extra_str() "rc1-16-gec6a1ba" + */ + + /*!\brief Return the version information (as an integer) + * + * Returns a packed encoding of the library version number. This will only include + * the major.minor.patch component of the version number. Note that this encoded + * value should be accessed through the macros provided, as the encoding may change + * in the future. + * + */ + int vpx_codec_version(void); +#define VPX_VERSION_MAJOR(v) ((v>>16)&0xff) /**< extract major from packed version */ +#define VPX_VERSION_MINOR(v) ((v>>8)&0xff) /**< extract minor from packed version */ +#define VPX_VERSION_PATCH(v) ((v>>0)&0xff) /**< extract patch from packed version */ + + /*!\brief Return the version major number */ +#define vpx_codec_version_major() ((vpx_codec_version()>>16)&0xff) + + /*!\brief Return the version minor number */ +#define vpx_codec_version_minor() ((vpx_codec_version()>>8)&0xff) + + /*!\brief Return the version patch number */ +#define vpx_codec_version_patch() ((vpx_codec_version()>>0)&0xff) + + + /*!\brief Return the version information (as a string) + * + * Returns a printable string containing the full library version number. This may + * contain additional text following the three digit version number, as to indicate + * release candidates, prerelease versions, etc. + * + */ + const char *vpx_codec_version_str(void); + + + /*!\brief Return the version information (as a string) + * + * Returns a printable "extra string". This is the component of the string returned + * by vpx_codec_version_str() following the three digit version number. + * + */ + const char *vpx_codec_version_extra_str(void); + + + /*!\brief Return the build configuration + * + * Returns a printable string containing an encoded version of the build + * configuration. This may be useful to vpx support. + * + */ + const char *vpx_codec_build_config(void); + + + /*!\brief Return the name for a given interface + * + * Returns a human readable string for name of the given codec interface. + * + * \param[in] iface Interface pointer + * + */ + const char *vpx_codec_iface_name(vpx_codec_iface_t *iface); + + + /*!\brief Convert error number to printable string + * + * Returns a human readable string for the last error returned by the + * algorithm. The returned error will be one line and will not contain + * any newline characters. + * + * + * \param[in] err Error number. + * + */ + const char *vpx_codec_err_to_string(vpx_codec_err_t err); + + + /*!\brief Retrieve error synopsis for codec context + * + * Returns a human readable string for the last error returned by the + * algorithm. The returned error will be one line and will not contain + * any newline characters. + * + * + * \param[in] ctx Pointer to this instance's context. + * + */ + const char *vpx_codec_error(vpx_codec_ctx_t *ctx); + + + /*!\brief Retrieve detailed error information for codec context + * + * Returns a human readable string providing detailed information about + * the last error. + * + * \param[in] ctx Pointer to this instance's context. + * + * \retval NULL + * No detailed information is available. + */ + const char *vpx_codec_error_detail(vpx_codec_ctx_t *ctx); + + + /* REQUIRED FUNCTIONS + * + * The following functions are required to be implemented for all codecs. + * They represent the base case functionality expected of all codecs. + */ + + /*!\brief Destroy a codec instance + * + * Destroys a codec context, freeing any associated memory buffers. + * + * \param[in] ctx Pointer to this instance's context + * + * \retval #VPX_CODEC_OK + * The codec algorithm initialized. + * \retval #VPX_CODEC_MEM_ERROR + * Memory allocation failed. + */ + vpx_codec_err_t vpx_codec_destroy(vpx_codec_ctx_t *ctx); + + + /*!\brief Get the capabilities of an algorithm. + * + * Retrieves the capabilities bitfield from the algorithm's interface. + * + * \param[in] iface Pointer to the algorithm interface + * + */ + vpx_codec_caps_t vpx_codec_get_caps(vpx_codec_iface_t *iface); + + + /*!\brief Control algorithm + * + * This function is used to exchange algorithm specific data with the codec + * instance. This can be used to implement features specific to a particular + * algorithm. + * + * This wrapper function dispatches the request to the helper function + * associated with the given ctrl_id. It tries to call this function + * transparently, but will return #VPX_CODEC_ERROR if the request could not + * be dispatched. + * + * Note that this function should not be used directly. Call the + * #vpx_codec_control wrapper macro instead. + * + * \param[in] ctx Pointer to this instance's context + * \param[in] ctrl_id Algorithm specific control identifier + * + * \retval #VPX_CODEC_OK + * The control request was processed. + * \retval #VPX_CODEC_ERROR + * The control request was not processed. + * \retval #VPX_CODEC_INVALID_PARAM + * The data was not valid. + */ + vpx_codec_err_t vpx_codec_control_(vpx_codec_ctx_t *ctx, + int ctrl_id, + ...); +#if defined(VPX_DISABLE_CTRL_TYPECHECKS) && VPX_DISABLE_CTRL_TYPECHECKS +# define vpx_codec_control(ctx,id,data) vpx_codec_control_(ctx,id,data) +# define VPX_CTRL_USE_TYPE(id, typ) +# define VPX_CTRL_USE_TYPE_DEPRECATED(id, typ) +# define VPX_CTRL_VOID(id, typ) + +#else + /*!\brief vpx_codec_control wrapper macro + * + * This macro allows for type safe conversions across the variadic parameter + * to vpx_codec_control_(). + * + * \internal + * It works by dispatching the call to the control function through a wrapper + * function named with the id parameter. + */ +# define vpx_codec_control(ctx,id,data) vpx_codec_control_##id(ctx,id,data)\ + /**<\hideinitializer*/ + + + /*!\brief vpx_codec_control type definition macro + * + * This macro allows for type safe conversions across the variadic parameter + * to vpx_codec_control_(). It defines the type of the argument for a given + * control identifier. + * + * \internal + * It defines a static function with + * the correctly typed arguments as a wrapper to the type-unsafe internal + * function. + */ +# define VPX_CTRL_USE_TYPE(id, typ) \ + static vpx_codec_err_t \ + vpx_codec_control_##id(vpx_codec_ctx_t*, int, typ) UNUSED;\ + \ + static vpx_codec_err_t \ + vpx_codec_control_##id(vpx_codec_ctx_t *ctx, int ctrl_id, typ data) {\ + return vpx_codec_control_(ctx, ctrl_id, data);\ + } /**<\hideinitializer*/ + + + /*!\brief vpx_codec_control deprecated type definition macro + * + * Like #VPX_CTRL_USE_TYPE, but indicates that the specified control is + * deprecated and should not be used. Consult the documentation for your + * codec for more information. + * + * \internal + * It defines a static function with the correctly typed arguments as a + * wrapper to the type-unsafe internal function. + */ +# define VPX_CTRL_USE_TYPE_DEPRECATED(id, typ) \ + DECLSPEC_DEPRECATED static vpx_codec_err_t \ + vpx_codec_control_##id(vpx_codec_ctx_t*, int, typ) DEPRECATED UNUSED;\ + \ + DECLSPEC_DEPRECATED static vpx_codec_err_t \ + vpx_codec_control_##id(vpx_codec_ctx_t *ctx, int ctrl_id, typ data) {\ + return vpx_codec_control_(ctx, ctrl_id, data);\ + } /**<\hideinitializer*/ + + + /*!\brief vpx_codec_control void type definition macro + * + * This macro allows for type safe conversions across the variadic parameter + * to vpx_codec_control_(). It indicates that a given control identifier takes + * no argument. + * + * \internal + * It defines a static function without a data argument as a wrapper to the + * type-unsafe internal function. + */ +# define VPX_CTRL_VOID(id) \ + static vpx_codec_err_t \ + vpx_codec_control_##id(vpx_codec_ctx_t*, int) UNUSED;\ + \ + static vpx_codec_err_t \ + vpx_codec_control_##id(vpx_codec_ctx_t *ctx, int ctrl_id) {\ + return vpx_codec_control_(ctx, ctrl_id);\ + } /**<\hideinitializer*/ + + +#endif + + /*!@} - end defgroup codec*/ +#ifdef __cplusplus +} +#endif +#endif // VPX_VPX_CODEC_H_ + diff --git a/thirdparty/libvpx/vpx/vpx_decoder.h b/thirdparty/libvpx/vpx/vpx_decoder.h new file mode 100644 index 0000000000..62fd919756 --- /dev/null +++ b/thirdparty/libvpx/vpx/vpx_decoder.h @@ -0,0 +1,378 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ +#ifndef VPX_VPX_DECODER_H_ +#define VPX_VPX_DECODER_H_ + +/*!\defgroup decoder Decoder Algorithm Interface + * \ingroup codec + * This abstraction allows applications using this decoder to easily support + * multiple video formats with minimal code duplication. This section describes + * the interface common to all decoders. + * @{ + */ + +/*!\file + * \brief Describes the decoder algorithm interface to applications. + * + * This file describes the interface between an application and a + * video decoder algorithm. + * + */ +#ifdef __cplusplus +extern "C" { +#endif + +#include "./vpx_codec.h" +#include "./vpx_frame_buffer.h" + + /*!\brief Current ABI version number + * + * \internal + * If this file is altered in any way that changes the ABI, this value + * must be bumped. Examples include, but are not limited to, changing + * types, removing or reassigning enums, adding/removing/rearranging + * fields to structures + */ +#define VPX_DECODER_ABI_VERSION (3 + VPX_CODEC_ABI_VERSION) /**<\hideinitializer*/ + + /*! \brief Decoder capabilities bitfield + * + * Each decoder advertises the capabilities it supports as part of its + * ::vpx_codec_iface_t interface structure. Capabilities are extra interfaces + * or functionality, and are not required to be supported by a decoder. + * + * The available flags are specified by VPX_CODEC_CAP_* defines. + */ +#define VPX_CODEC_CAP_PUT_SLICE 0x10000 /**< Will issue put_slice callbacks */ +#define VPX_CODEC_CAP_PUT_FRAME 0x20000 /**< Will issue put_frame callbacks */ +#define VPX_CODEC_CAP_POSTPROC 0x40000 /**< Can postprocess decoded frame */ +#define VPX_CODEC_CAP_ERROR_CONCEALMENT 0x80000 /**< Can conceal errors due to + packet loss */ +#define VPX_CODEC_CAP_INPUT_FRAGMENTS 0x100000 /**< Can receive encoded frames + one fragment at a time */ + + /*! \brief Initialization-time Feature Enabling + * + * Certain codec features must be known at initialization time, to allow for + * proper memory allocation. + * + * The available flags are specified by VPX_CODEC_USE_* defines. + */ +#define VPX_CODEC_CAP_FRAME_THREADING 0x200000 /**< Can support frame-based + multi-threading */ +#define VPX_CODEC_CAP_EXTERNAL_FRAME_BUFFER 0x400000 /**< Can support external + frame buffers */ + +#define VPX_CODEC_USE_POSTPROC 0x10000 /**< Postprocess decoded frame */ +#define VPX_CODEC_USE_ERROR_CONCEALMENT 0x20000 /**< Conceal errors in decoded + frames */ +#define VPX_CODEC_USE_INPUT_FRAGMENTS 0x40000 /**< The input frame should be + passed to the decoder one + fragment at a time */ +#define VPX_CODEC_USE_FRAME_THREADING 0x80000 /**< Enable frame-based + multi-threading */ + + /*!\brief Stream properties + * + * This structure is used to query or set properties of the decoded + * stream. Algorithms may extend this structure with data specific + * to their bitstream by setting the sz member appropriately. + */ + typedef struct vpx_codec_stream_info { + unsigned int sz; /**< Size of this structure */ + unsigned int w; /**< Width (or 0 for unknown/default) */ + unsigned int h; /**< Height (or 0 for unknown/default) */ + unsigned int is_kf; /**< Current frame is a keyframe */ + } vpx_codec_stream_info_t; + + /* REQUIRED FUNCTIONS + * + * The following functions are required to be implemented for all decoders. + * They represent the base case functionality expected of all decoders. + */ + + + /*!\brief Initialization Configurations + * + * This structure is used to pass init time configuration options to the + * decoder. + */ + typedef struct vpx_codec_dec_cfg { + unsigned int threads; /**< Maximum number of threads to use, default 1 */ + unsigned int w; /**< Width */ + unsigned int h; /**< Height */ + } vpx_codec_dec_cfg_t; /**< alias for struct vpx_codec_dec_cfg */ + + + /*!\brief Initialize a decoder instance + * + * Initializes a decoder context using the given interface. Applications + * should call the vpx_codec_dec_init convenience macro instead of this + * function directly, to ensure that the ABI version number parameter + * is properly initialized. + * + * If the library was configured with --disable-multithread, this call + * is not thread safe and should be guarded with a lock if being used + * in a multithreaded context. + * + * \param[in] ctx Pointer to this instance's context. + * \param[in] iface Pointer to the algorithm interface to use. + * \param[in] cfg Configuration to use, if known. May be NULL. + * \param[in] flags Bitfield of VPX_CODEC_USE_* flags + * \param[in] ver ABI version number. Must be set to + * VPX_DECODER_ABI_VERSION + * \retval #VPX_CODEC_OK + * The decoder algorithm initialized. + * \retval #VPX_CODEC_MEM_ERROR + * Memory allocation failed. + */ + vpx_codec_err_t vpx_codec_dec_init_ver(vpx_codec_ctx_t *ctx, + vpx_codec_iface_t *iface, + const vpx_codec_dec_cfg_t *cfg, + vpx_codec_flags_t flags, + int ver); + + /*!\brief Convenience macro for vpx_codec_dec_init_ver() + * + * Ensures the ABI version parameter is properly set. + */ +#define vpx_codec_dec_init(ctx, iface, cfg, flags) \ + vpx_codec_dec_init_ver(ctx, iface, cfg, flags, VPX_DECODER_ABI_VERSION) + + + /*!\brief Parse stream info from a buffer + * + * Performs high level parsing of the bitstream. Construction of a decoder + * context is not necessary. Can be used to determine if the bitstream is + * of the proper format, and to extract information from the stream. + * + * \param[in] iface Pointer to the algorithm interface + * \param[in] data Pointer to a block of data to parse + * \param[in] data_sz Size of the data buffer + * \param[in,out] si Pointer to stream info to update. The size member + * \ref MUST be properly initialized, but \ref MAY be + * clobbered by the algorithm. This parameter \ref MAY + * be NULL. + * + * \retval #VPX_CODEC_OK + * Bitstream is parsable and stream information updated + */ + vpx_codec_err_t vpx_codec_peek_stream_info(vpx_codec_iface_t *iface, + const uint8_t *data, + unsigned int data_sz, + vpx_codec_stream_info_t *si); + + + /*!\brief Return information about the current stream. + * + * Returns information about the stream that has been parsed during decoding. + * + * \param[in] ctx Pointer to this instance's context + * \param[in,out] si Pointer to stream info to update. The size member + * \ref MUST be properly initialized, but \ref MAY be + * clobbered by the algorithm. This parameter \ref MAY + * be NULL. + * + * \retval #VPX_CODEC_OK + * Bitstream is parsable and stream information updated + */ + vpx_codec_err_t vpx_codec_get_stream_info(vpx_codec_ctx_t *ctx, + vpx_codec_stream_info_t *si); + + + /*!\brief Decode data + * + * Processes a buffer of coded data. If the processing results in a new + * decoded frame becoming available, PUT_SLICE and PUT_FRAME events may be + * generated, as appropriate. Encoded data \ref MUST be passed in DTS (decode + * time stamp) order. Frames produced will always be in PTS (presentation + * time stamp) order. + * If the decoder is configured with VPX_CODEC_USE_INPUT_FRAGMENTS enabled, + * data and data_sz can contain a fragment of the encoded frame. Fragment + * \#n must contain at least partition \#n, but can also contain subsequent + * partitions (\#n+1 - \#n+i), and if so, fragments \#n+1, .., \#n+i must + * be empty. When no more data is available, this function should be called + * with NULL as data and 0 as data_sz. The memory passed to this function + * must be available until the frame has been decoded. + * + * \param[in] ctx Pointer to this instance's context + * \param[in] data Pointer to this block of new coded data. If + * NULL, a VPX_CODEC_CB_PUT_FRAME event is posted + * for the previously decoded frame. + * \param[in] data_sz Size of the coded data, in bytes. + * \param[in] user_priv Application specific data to associate with + * this frame. + * \param[in] deadline Soft deadline the decoder should attempt to meet, + * in us. Set to zero for unlimited. + * + * \return Returns #VPX_CODEC_OK if the coded data was processed completely + * and future pictures can be decoded without error. Otherwise, + * see the descriptions of the other error codes in ::vpx_codec_err_t + * for recoverability capabilities. + */ + vpx_codec_err_t vpx_codec_decode(vpx_codec_ctx_t *ctx, + const uint8_t *data, + unsigned int data_sz, + void *user_priv, + long deadline); + + + /*!\brief Decoded frames iterator + * + * Iterates over a list of the frames available for display. The iterator + * storage should be initialized to NULL to start the iteration. Iteration is + * complete when this function returns NULL. + * + * The list of available frames becomes valid upon completion of the + * vpx_codec_decode call, and remains valid until the next call to vpx_codec_decode. + * + * \param[in] ctx Pointer to this instance's context + * \param[in,out] iter Iterator storage, initialized to NULL + * + * \return Returns a pointer to an image, if one is ready for display. Frames + * produced will always be in PTS (presentation time stamp) order. + */ + vpx_image_t *vpx_codec_get_frame(vpx_codec_ctx_t *ctx, + vpx_codec_iter_t *iter); + + + /*!\defgroup cap_put_frame Frame-Based Decoding Functions + * + * The following functions are required to be implemented for all decoders + * that advertise the VPX_CODEC_CAP_PUT_FRAME capability. Calling these functions + * for codecs that don't advertise this capability will result in an error + * code being returned, usually VPX_CODEC_ERROR + * @{ + */ + + /*!\brief put frame callback prototype + * + * This callback is invoked by the decoder to notify the application of + * the availability of decoded image data. + */ + typedef void (*vpx_codec_put_frame_cb_fn_t)(void *user_priv, + const vpx_image_t *img); + + + /*!\brief Register for notification of frame completion. + * + * Registers a given function to be called when a decoded frame is + * available. + * + * \param[in] ctx Pointer to this instance's context + * \param[in] cb Pointer to the callback function + * \param[in] user_priv User's private data + * + * \retval #VPX_CODEC_OK + * Callback successfully registered. + * \retval #VPX_CODEC_ERROR + * Decoder context not initialized, or algorithm not capable of + * posting slice completion. + */ + vpx_codec_err_t vpx_codec_register_put_frame_cb(vpx_codec_ctx_t *ctx, + vpx_codec_put_frame_cb_fn_t cb, + void *user_priv); + + + /*!@} - end defgroup cap_put_frame */ + + /*!\defgroup cap_put_slice Slice-Based Decoding Functions + * + * The following functions are required to be implemented for all decoders + * that advertise the VPX_CODEC_CAP_PUT_SLICE capability. Calling these functions + * for codecs that don't advertise this capability will result in an error + * code being returned, usually VPX_CODEC_ERROR + * @{ + */ + + /*!\brief put slice callback prototype + * + * This callback is invoked by the decoder to notify the application of + * the availability of partially decoded image data. The + */ + typedef void (*vpx_codec_put_slice_cb_fn_t)(void *user_priv, + const vpx_image_t *img, + const vpx_image_rect_t *valid, + const vpx_image_rect_t *update); + + + /*!\brief Register for notification of slice completion. + * + * Registers a given function to be called when a decoded slice is + * available. + * + * \param[in] ctx Pointer to this instance's context + * \param[in] cb Pointer to the callback function + * \param[in] user_priv User's private data + * + * \retval #VPX_CODEC_OK + * Callback successfully registered. + * \retval #VPX_CODEC_ERROR + * Decoder context not initialized, or algorithm not capable of + * posting slice completion. + */ + vpx_codec_err_t vpx_codec_register_put_slice_cb(vpx_codec_ctx_t *ctx, + vpx_codec_put_slice_cb_fn_t cb, + void *user_priv); + + + /*!@} - end defgroup cap_put_slice*/ + + /*!\defgroup cap_external_frame_buffer External Frame Buffer Functions + * + * The following section is required to be implemented for all decoders + * that advertise the VPX_CODEC_CAP_EXTERNAL_FRAME_BUFFER capability. + * Calling this function for codecs that don't advertise this capability + * will result in an error code being returned, usually VPX_CODEC_ERROR. + * + * \note + * Currently this only works with VP9. + * @{ + */ + + /*!\brief Pass in external frame buffers for the decoder to use. + * + * Registers functions to be called when libvpx needs a frame buffer + * to decode the current frame and a function to be called when libvpx does + * not internally reference the frame buffer. This set function must + * be called before the first call to decode or libvpx will assume the + * default behavior of allocating frame buffers internally. + * + * \param[in] ctx Pointer to this instance's context + * \param[in] cb_get Pointer to the get callback function + * \param[in] cb_release Pointer to the release callback function + * \param[in] cb_priv Callback's private data + * + * \retval #VPX_CODEC_OK + * External frame buffers will be used by libvpx. + * \retval #VPX_CODEC_INVALID_PARAM + * One or more of the callbacks were NULL. + * \retval #VPX_CODEC_ERROR + * Decoder context not initialized, or algorithm not capable of + * using external frame buffers. + * + * \note + * When decoding VP9, the application may be required to pass in at least + * #VP9_MAXIMUM_REF_BUFFERS + #VPX_MAXIMUM_WORK_BUFFERS external frame + * buffers. + */ + vpx_codec_err_t vpx_codec_set_frame_buffer_functions( + vpx_codec_ctx_t *ctx, + vpx_get_frame_buffer_cb_fn_t cb_get, + vpx_release_frame_buffer_cb_fn_t cb_release, void *cb_priv); + + /*!@} - end defgroup cap_external_frame_buffer */ + + /*!@} - end defgroup decoder*/ +#ifdef __cplusplus +} +#endif +#endif // VPX_VPX_DECODER_H_ + diff --git a/thirdparty/libvpx/vpx/vpx_encoder.h b/thirdparty/libvpx/vpx/vpx_encoder.h new file mode 100644 index 0000000000..955e873519 --- /dev/null +++ b/thirdparty/libvpx/vpx/vpx_encoder.h @@ -0,0 +1,1043 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ +#ifndef VPX_VPX_ENCODER_H_ +#define VPX_VPX_ENCODER_H_ + +/*!\defgroup encoder Encoder Algorithm Interface + * \ingroup codec + * This abstraction allows applications using this encoder to easily support + * multiple video formats with minimal code duplication. This section describes + * the interface common to all encoders. + * @{ + */ + +/*!\file + * \brief Describes the encoder algorithm interface to applications. + * + * This file describes the interface between an application and a + * video encoder algorithm. + * + */ +#ifdef __cplusplus +extern "C" { +#endif + +#include "./vpx_codec.h" + + /*! Temporal Scalability: Maximum length of the sequence defining frame + * layer membership + */ +#define VPX_TS_MAX_PERIODICITY 16 + + /*! Temporal Scalability: Maximum number of coding layers */ +#define VPX_TS_MAX_LAYERS 5 + + /*!\deprecated Use #VPX_TS_MAX_PERIODICITY instead. */ +#define MAX_PERIODICITY VPX_TS_MAX_PERIODICITY + +/*! Temporal+Spatial Scalability: Maximum number of coding layers */ +#define VPX_MAX_LAYERS 12 // 3 temporal + 4 spatial layers are allowed. + +/*!\deprecated Use #VPX_MAX_LAYERS instead. */ +#define MAX_LAYERS VPX_MAX_LAYERS // 3 temporal + 4 spatial layers allowed. + +/*! Spatial Scalability: Maximum number of coding layers */ +#define VPX_SS_MAX_LAYERS 5 + +/*! Spatial Scalability: Default number of coding layers */ +#define VPX_SS_DEFAULT_LAYERS 1 + + /*!\brief Current ABI version number + * + * \internal + * If this file is altered in any way that changes the ABI, this value + * must be bumped. Examples include, but are not limited to, changing + * types, removing or reassigning enums, adding/removing/rearranging + * fields to structures + */ +#define VPX_ENCODER_ABI_VERSION (5 + VPX_CODEC_ABI_VERSION) /**<\hideinitializer*/ + + + /*! \brief Encoder capabilities bitfield + * + * Each encoder advertises the capabilities it supports as part of its + * ::vpx_codec_iface_t interface structure. Capabilities are extra + * interfaces or functionality, and are not required to be supported + * by an encoder. + * + * The available flags are specified by VPX_CODEC_CAP_* defines. + */ +#define VPX_CODEC_CAP_PSNR 0x10000 /**< Can issue PSNR packets */ + + /*! Can output one partition at a time. Each partition is returned in its + * own VPX_CODEC_CX_FRAME_PKT, with the FRAME_IS_FRAGMENT flag set for + * every partition but the last. In this mode all frames are always + * returned partition by partition. + */ +#define VPX_CODEC_CAP_OUTPUT_PARTITION 0x20000 + +/*! Can support input images at greater than 8 bitdepth. + */ +#define VPX_CODEC_CAP_HIGHBITDEPTH 0x40000 + + /*! \brief Initialization-time Feature Enabling + * + * Certain codec features must be known at initialization time, to allow + * for proper memory allocation. + * + * The available flags are specified by VPX_CODEC_USE_* defines. + */ +#define VPX_CODEC_USE_PSNR 0x10000 /**< Calculate PSNR on each frame */ +#define VPX_CODEC_USE_OUTPUT_PARTITION 0x20000 /**< Make the encoder output one + partition at a time. */ +#define VPX_CODEC_USE_HIGHBITDEPTH 0x40000 /**< Use high bitdepth */ + + + /*!\brief Generic fixed size buffer structure + * + * This structure is able to hold a reference to any fixed size buffer. + */ + typedef struct vpx_fixed_buf { + void *buf; /**< Pointer to the data */ + size_t sz; /**< Length of the buffer, in chars */ + } vpx_fixed_buf_t; /**< alias for struct vpx_fixed_buf */ + + + /*!\brief Time Stamp Type + * + * An integer, which when multiplied by the stream's time base, provides + * the absolute time of a sample. + */ + typedef int64_t vpx_codec_pts_t; + + + /*!\brief Compressed Frame Flags + * + * This type represents a bitfield containing information about a compressed + * frame that may be useful to an application. The most significant 16 bits + * can be used by an algorithm to provide additional detail, for example to + * support frame types that are codec specific (MPEG-1 D-frames for example) + */ + typedef uint32_t vpx_codec_frame_flags_t; +#define VPX_FRAME_IS_KEY 0x1 /**< frame is the start of a GOP */ +#define VPX_FRAME_IS_DROPPABLE 0x2 /**< frame can be dropped without affecting + the stream (no future frame depends on + this one) */ +#define VPX_FRAME_IS_INVISIBLE 0x4 /**< frame should be decoded but will not + be shown */ +#define VPX_FRAME_IS_FRAGMENT 0x8 /**< this is a fragment of the encoded + frame */ + + /*!\brief Error Resilient flags + * + * These flags define which error resilient features to enable in the + * encoder. The flags are specified through the + * vpx_codec_enc_cfg::g_error_resilient variable. + */ + typedef uint32_t vpx_codec_er_flags_t; +#define VPX_ERROR_RESILIENT_DEFAULT 0x1 /**< Improve resiliency against + losses of whole frames */ +#define VPX_ERROR_RESILIENT_PARTITIONS 0x2 /**< The frame partitions are + independently decodable by the + bool decoder, meaning that + partitions can be decoded even + though earlier partitions have + been lost. Note that intra + prediction is still done over + the partition boundary. */ + + /*!\brief Encoder output packet variants + * + * This enumeration lists the different kinds of data packets that can be + * returned by calls to vpx_codec_get_cx_data(). Algorithms \ref MAY + * extend this list to provide additional functionality. + */ + enum vpx_codec_cx_pkt_kind { + VPX_CODEC_CX_FRAME_PKT, /**< Compressed video frame */ + VPX_CODEC_STATS_PKT, /**< Two-pass statistics for this frame */ + VPX_CODEC_FPMB_STATS_PKT, /**< first pass mb statistics for this frame */ + VPX_CODEC_PSNR_PKT, /**< PSNR statistics for this frame */ + // Spatial SVC is still experimental and may be removed before the next ABI + // bump. +#if VPX_ENCODER_ABI_VERSION > (5 + VPX_CODEC_ABI_VERSION) + VPX_CODEC_SPATIAL_SVC_LAYER_SIZES, /**< Sizes for each layer in this frame*/ + VPX_CODEC_SPATIAL_SVC_LAYER_PSNR, /**< PSNR for each layer in this frame*/ +#endif + VPX_CODEC_CUSTOM_PKT = 256 /**< Algorithm extensions */ + }; + + + /*!\brief Encoder output packet + * + * This structure contains the different kinds of output data the encoder + * may produce while compressing a frame. + */ + typedef struct vpx_codec_cx_pkt { + enum vpx_codec_cx_pkt_kind kind; /**< packet variant */ + union { + struct { + void *buf; /**< compressed data buffer */ + size_t sz; /**< length of compressed data */ + vpx_codec_pts_t pts; /**< time stamp to show frame + (in timebase units) */ + unsigned long duration; /**< duration to show frame + (in timebase units) */ + vpx_codec_frame_flags_t flags; /**< flags for this frame */ + int partition_id; /**< the partition id + defines the decoding order + of the partitions. Only + applicable when "output partition" + mode is enabled. First partition + has id 0.*/ + + } frame; /**< data for compressed frame packet */ + vpx_fixed_buf_t twopass_stats; /**< data for two-pass packet */ + vpx_fixed_buf_t firstpass_mb_stats; /**< first pass mb packet */ + struct vpx_psnr_pkt { + unsigned int samples[4]; /**< Number of samples, total/y/u/v */ + uint64_t sse[4]; /**< sum squared error, total/y/u/v */ + double psnr[4]; /**< PSNR, total/y/u/v */ + } psnr; /**< data for PSNR packet */ + vpx_fixed_buf_t raw; /**< data for arbitrary packets */ + // Spatial SVC is still experimental and may be removed before the next + // ABI bump. +#if VPX_ENCODER_ABI_VERSION > (5 + VPX_CODEC_ABI_VERSION) + size_t layer_sizes[VPX_SS_MAX_LAYERS]; + struct vpx_psnr_pkt layer_psnr[VPX_SS_MAX_LAYERS]; +#endif + + /* This packet size is fixed to allow codecs to extend this + * interface without having to manage storage for raw packets, + * i.e., if it's smaller than 128 bytes, you can store in the + * packet list directly. + */ + char pad[128 - sizeof(enum vpx_codec_cx_pkt_kind)]; /**< fixed sz */ + } data; /**< packet data */ + } vpx_codec_cx_pkt_t; /**< alias for struct vpx_codec_cx_pkt */ + + + /*!\brief Encoder return output buffer callback + * + * This callback function, when registered, returns with packets when each + * spatial layer is encoded. + */ + // putting the definitions here for now. (agrange: find if there + // is a better place for this) + typedef void (* vpx_codec_enc_output_cx_pkt_cb_fn_t)(vpx_codec_cx_pkt_t *pkt, + void *user_data); + + /*!\brief Callback function pointer / user data pair storage */ + typedef struct vpx_codec_enc_output_cx_cb_pair { + vpx_codec_enc_output_cx_pkt_cb_fn_t output_cx_pkt; /**< Callback function */ + void *user_priv; /**< Pointer to private data */ + } vpx_codec_priv_output_cx_pkt_cb_pair_t; + + /*!\brief Rational Number + * + * This structure holds a fractional value. + */ + typedef struct vpx_rational { + int num; /**< fraction numerator */ + int den; /**< fraction denominator */ + } vpx_rational_t; /**< alias for struct vpx_rational */ + + + /*!\brief Multi-pass Encoding Pass */ + enum vpx_enc_pass { + VPX_RC_ONE_PASS, /**< Single pass mode */ + VPX_RC_FIRST_PASS, /**< First pass of multi-pass mode */ + VPX_RC_LAST_PASS /**< Final pass of multi-pass mode */ + }; + + + /*!\brief Rate control mode */ + enum vpx_rc_mode { + VPX_VBR, /**< Variable Bit Rate (VBR) mode */ + VPX_CBR, /**< Constant Bit Rate (CBR) mode */ + VPX_CQ, /**< Constrained Quality (CQ) mode */ + VPX_Q, /**< Constant Quality (Q) mode */ + }; + + + /*!\brief Keyframe placement mode. + * + * This enumeration determines whether keyframes are placed automatically by + * the encoder or whether this behavior is disabled. Older releases of this + * SDK were implemented such that VPX_KF_FIXED meant keyframes were disabled. + * This name is confusing for this behavior, so the new symbols to be used + * are VPX_KF_AUTO and VPX_KF_DISABLED. + */ + enum vpx_kf_mode { + VPX_KF_FIXED, /**< deprecated, implies VPX_KF_DISABLED */ + VPX_KF_AUTO, /**< Encoder determines optimal placement automatically */ + VPX_KF_DISABLED = 0 /**< Encoder does not place keyframes. */ + }; + + + /*!\brief Encoded Frame Flags + * + * This type indicates a bitfield to be passed to vpx_codec_encode(), defining + * per-frame boolean values. By convention, bits common to all codecs will be + * named VPX_EFLAG_*, and bits specific to an algorithm will be named + * /algo/_eflag_*. The lower order 16 bits are reserved for common use. + */ + typedef long vpx_enc_frame_flags_t; +#define VPX_EFLAG_FORCE_KF (1<<0) /**< Force this frame to be a keyframe */ + + + /*!\brief Encoder configuration structure + * + * This structure contains the encoder settings that have common representations + * across all codecs. This doesn't imply that all codecs support all features, + * however. + */ + typedef struct vpx_codec_enc_cfg { + /* + * generic settings (g) + */ + + /*!\brief Algorithm specific "usage" value + * + * Algorithms may define multiple values for usage, which may convey the + * intent of how the application intends to use the stream. If this value + * is non-zero, consult the documentation for the codec to determine its + * meaning. + */ + unsigned int g_usage; + + + /*!\brief Maximum number of threads to use + * + * For multi-threaded implementations, use no more than this number of + * threads. The codec may use fewer threads than allowed. The value + * 0 is equivalent to the value 1. + */ + unsigned int g_threads; + + + /*!\brief Bitstream profile to use + * + * Some codecs support a notion of multiple bitstream profiles. Typically + * this maps to a set of features that are turned on or off. Often the + * profile to use is determined by the features of the intended decoder. + * Consult the documentation for the codec to determine the valid values + * for this parameter, or set to zero for a sane default. + */ + unsigned int g_profile; /**< profile of bitstream to use */ + + + + /*!\brief Width of the frame + * + * This value identifies the presentation resolution of the frame, + * in pixels. Note that the frames passed as input to the encoder must + * have this resolution. Frames will be presented by the decoder in this + * resolution, independent of any spatial resampling the encoder may do. + */ + unsigned int g_w; + + + /*!\brief Height of the frame + * + * This value identifies the presentation resolution of the frame, + * in pixels. Note that the frames passed as input to the encoder must + * have this resolution. Frames will be presented by the decoder in this + * resolution, independent of any spatial resampling the encoder may do. + */ + unsigned int g_h; + + /*!\brief Bit-depth of the codec + * + * This value identifies the bit_depth of the codec, + * Only certain bit-depths are supported as identified in the + * vpx_bit_depth_t enum. + */ + vpx_bit_depth_t g_bit_depth; + + /*!\brief Bit-depth of the input frames + * + * This value identifies the bit_depth of the input frames in bits. + * Note that the frames passed as input to the encoder must have + * this bit-depth. + */ + unsigned int g_input_bit_depth; + + /*!\brief Stream timebase units + * + * Indicates the smallest interval of time, in seconds, used by the stream. + * For fixed frame rate material, or variable frame rate material where + * frames are timed at a multiple of a given clock (ex: video capture), + * the \ref RECOMMENDED method is to set the timebase to the reciprocal + * of the frame rate (ex: 1001/30000 for 29.970 Hz NTSC). This allows the + * pts to correspond to the frame number, which can be handy. For + * re-encoding video from containers with absolute time timestamps, the + * \ref RECOMMENDED method is to set the timebase to that of the parent + * container or multimedia framework (ex: 1/1000 for ms, as in FLV). + */ + struct vpx_rational g_timebase; + + + /*!\brief Enable error resilient modes. + * + * The error resilient bitfield indicates to the encoder which features + * it should enable to take measures for streaming over lossy or noisy + * links. + */ + vpx_codec_er_flags_t g_error_resilient; + + + /*!\brief Multi-pass Encoding Mode + * + * This value should be set to the current phase for multi-pass encoding. + * For single pass, set to #VPX_RC_ONE_PASS. + */ + enum vpx_enc_pass g_pass; + + + /*!\brief Allow lagged encoding + * + * If set, this value allows the encoder to consume a number of input + * frames before producing output frames. This allows the encoder to + * base decisions for the current frame on future frames. This does + * increase the latency of the encoding pipeline, so it is not appropriate + * in all situations (ex: realtime encoding). + * + * Note that this is a maximum value -- the encoder may produce frames + * sooner than the given limit. Set this value to 0 to disable this + * feature. + */ + unsigned int g_lag_in_frames; + + + /* + * rate control settings (rc) + */ + + /*!\brief Temporal resampling configuration, if supported by the codec. + * + * Temporal resampling allows the codec to "drop" frames as a strategy to + * meet its target data rate. This can cause temporal discontinuities in + * the encoded video, which may appear as stuttering during playback. This + * trade-off is often acceptable, but for many applications is not. It can + * be disabled in these cases. + * + * Note that not all codecs support this feature. All vpx VPx codecs do. + * For other codecs, consult the documentation for that algorithm. + * + * This threshold is described as a percentage of the target data buffer. + * When the data buffer falls below this percentage of fullness, a + * dropped frame is indicated. Set the threshold to zero (0) to disable + * this feature. + */ + unsigned int rc_dropframe_thresh; + + + /*!\brief Enable/disable spatial resampling, if supported by the codec. + * + * Spatial resampling allows the codec to compress a lower resolution + * version of the frame, which is then upscaled by the encoder to the + * correct presentation resolution. This increases visual quality at + * low data rates, at the expense of CPU time on the encoder/decoder. + */ + unsigned int rc_resize_allowed; + + /*!\brief Internal coded frame width. + * + * If spatial resampling is enabled this specifies the width of the + * encoded frame. + */ + unsigned int rc_scaled_width; + + /*!\brief Internal coded frame height. + * + * If spatial resampling is enabled this specifies the height of the + * encoded frame. + */ + unsigned int rc_scaled_height; + + /*!\brief Spatial resampling up watermark. + * + * This threshold is described as a percentage of the target data buffer. + * When the data buffer rises above this percentage of fullness, the + * encoder will step up to a higher resolution version of the frame. + */ + unsigned int rc_resize_up_thresh; + + + /*!\brief Spatial resampling down watermark. + * + * This threshold is described as a percentage of the target data buffer. + * When the data buffer falls below this percentage of fullness, the + * encoder will step down to a lower resolution version of the frame. + */ + unsigned int rc_resize_down_thresh; + + + /*!\brief Rate control algorithm to use. + * + * Indicates whether the end usage of this stream is to be streamed over + * a bandwidth constrained link, indicating that Constant Bit Rate (CBR) + * mode should be used, or whether it will be played back on a high + * bandwidth link, as from a local disk, where higher variations in + * bitrate are acceptable. + */ + enum vpx_rc_mode rc_end_usage; + + + /*!\brief Two-pass stats buffer. + * + * A buffer containing all of the stats packets produced in the first + * pass, concatenated. + */ + vpx_fixed_buf_t rc_twopass_stats_in; + + /*!\brief first pass mb stats buffer. + * + * A buffer containing all of the first pass mb stats packets produced + * in the first pass, concatenated. + */ + vpx_fixed_buf_t rc_firstpass_mb_stats_in; + + /*!\brief Target data rate + * + * Target bandwidth to use for this stream, in kilobits per second. + */ + unsigned int rc_target_bitrate; + + + /* + * quantizer settings + */ + + + /*!\brief Minimum (Best Quality) Quantizer + * + * The quantizer is the most direct control over the quality of the + * encoded image. The range of valid values for the quantizer is codec + * specific. Consult the documentation for the codec to determine the + * values to use. To determine the range programmatically, call + * vpx_codec_enc_config_default() with a usage value of 0. + */ + unsigned int rc_min_quantizer; + + + /*!\brief Maximum (Worst Quality) Quantizer + * + * The quantizer is the most direct control over the quality of the + * encoded image. The range of valid values for the quantizer is codec + * specific. Consult the documentation for the codec to determine the + * values to use. To determine the range programmatically, call + * vpx_codec_enc_config_default() with a usage value of 0. + */ + unsigned int rc_max_quantizer; + + + /* + * bitrate tolerance + */ + + + /*!\brief Rate control adaptation undershoot control + * + * This value, expressed as a percentage of the target bitrate, + * controls the maximum allowed adaptation speed of the codec. + * This factor controls the maximum amount of bits that can + * be subtracted from the target bitrate in order to compensate + * for prior overshoot. + * + * Valid values in the range 0-1000. + */ + unsigned int rc_undershoot_pct; + + + /*!\brief Rate control adaptation overshoot control + * + * This value, expressed as a percentage of the target bitrate, + * controls the maximum allowed adaptation speed of the codec. + * This factor controls the maximum amount of bits that can + * be added to the target bitrate in order to compensate for + * prior undershoot. + * + * Valid values in the range 0-1000. + */ + unsigned int rc_overshoot_pct; + + + /* + * decoder buffer model parameters + */ + + + /*!\brief Decoder Buffer Size + * + * This value indicates the amount of data that may be buffered by the + * decoding application. Note that this value is expressed in units of + * time (milliseconds). For example, a value of 5000 indicates that the + * client will buffer (at least) 5000ms worth of encoded data. Use the + * target bitrate (#rc_target_bitrate) to convert to bits/bytes, if + * necessary. + */ + unsigned int rc_buf_sz; + + + /*!\brief Decoder Buffer Initial Size + * + * This value indicates the amount of data that will be buffered by the + * decoding application prior to beginning playback. This value is + * expressed in units of time (milliseconds). Use the target bitrate + * (#rc_target_bitrate) to convert to bits/bytes, if necessary. + */ + unsigned int rc_buf_initial_sz; + + + /*!\brief Decoder Buffer Optimal Size + * + * This value indicates the amount of data that the encoder should try + * to maintain in the decoder's buffer. This value is expressed in units + * of time (milliseconds). Use the target bitrate (#rc_target_bitrate) + * to convert to bits/bytes, if necessary. + */ + unsigned int rc_buf_optimal_sz; + + + /* + * 2 pass rate control parameters + */ + + + /*!\brief Two-pass mode CBR/VBR bias + * + * Bias, expressed on a scale of 0 to 100, for determining target size + * for the current frame. The value 0 indicates the optimal CBR mode + * value should be used. The value 100 indicates the optimal VBR mode + * value should be used. Values in between indicate which way the + * encoder should "lean." + */ + unsigned int rc_2pass_vbr_bias_pct; /**< RC mode bias between CBR and VBR(0-100: 0->CBR, 100->VBR) */ + + + /*!\brief Two-pass mode per-GOP minimum bitrate + * + * This value, expressed as a percentage of the target bitrate, indicates + * the minimum bitrate to be used for a single GOP (aka "section") + */ + unsigned int rc_2pass_vbr_minsection_pct; + + + /*!\brief Two-pass mode per-GOP maximum bitrate + * + * This value, expressed as a percentage of the target bitrate, indicates + * the maximum bitrate to be used for a single GOP (aka "section") + */ + unsigned int rc_2pass_vbr_maxsection_pct; + + + /* + * keyframing settings (kf) + */ + + /*!\brief Keyframe placement mode + * + * This value indicates whether the encoder should place keyframes at a + * fixed interval, or determine the optimal placement automatically + * (as governed by the #kf_min_dist and #kf_max_dist parameters) + */ + enum vpx_kf_mode kf_mode; + + + /*!\brief Keyframe minimum interval + * + * This value, expressed as a number of frames, prevents the encoder from + * placing a keyframe nearer than kf_min_dist to the previous keyframe. At + * least kf_min_dist frames non-keyframes will be coded before the next + * keyframe. Set kf_min_dist equal to kf_max_dist for a fixed interval. + */ + unsigned int kf_min_dist; + + + /*!\brief Keyframe maximum interval + * + * This value, expressed as a number of frames, forces the encoder to code + * a keyframe if one has not been coded in the last kf_max_dist frames. + * A value of 0 implies all frames will be keyframes. Set kf_min_dist + * equal to kf_max_dist for a fixed interval. + */ + unsigned int kf_max_dist; + + /* + * Spatial scalability settings (ss) + */ + + /*!\brief Number of spatial coding layers. + * + * This value specifies the number of spatial coding layers to be used. + */ + unsigned int ss_number_layers; + + /*!\brief Enable auto alt reference flags for each spatial layer. + * + * These values specify if auto alt reference frame is enabled for each + * spatial layer. + */ + int ss_enable_auto_alt_ref[VPX_SS_MAX_LAYERS]; + + /*!\brief Target bitrate for each spatial layer. + * + * These values specify the target coding bitrate to be used for each + * spatial layer. + */ + unsigned int ss_target_bitrate[VPX_SS_MAX_LAYERS]; + + /*!\brief Number of temporal coding layers. + * + * This value specifies the number of temporal layers to be used. + */ + unsigned int ts_number_layers; + + /*!\brief Target bitrate for each temporal layer. + * + * These values specify the target coding bitrate to be used for each + * temporal layer. + */ + unsigned int ts_target_bitrate[VPX_TS_MAX_LAYERS]; + + /*!\brief Frame rate decimation factor for each temporal layer. + * + * These values specify the frame rate decimation factors to apply + * to each temporal layer. + */ + unsigned int ts_rate_decimator[VPX_TS_MAX_LAYERS]; + + /*!\brief Length of the sequence defining frame temporal layer membership. + * + * This value specifies the length of the sequence that defines the + * membership of frames to temporal layers. For example, if the + * ts_periodicity = 8, then the frames are assigned to coding layers with a + * repeated sequence of length 8. + */ + unsigned int ts_periodicity; + + /*!\brief Template defining the membership of frames to temporal layers. + * + * This array defines the membership of frames to temporal coding layers. + * For a 2-layer encoding that assigns even numbered frames to one temporal + * layer (0) and odd numbered frames to a second temporal layer (1) with + * ts_periodicity=8, then ts_layer_id = (0,1,0,1,0,1,0,1). + */ + unsigned int ts_layer_id[VPX_TS_MAX_PERIODICITY]; + + /*!\brief Target bitrate for each spatial/temporal layer. + * + * These values specify the target coding bitrate to be used for each + * spatial/temporal layer. + * + */ + unsigned int layer_target_bitrate[VPX_MAX_LAYERS]; + + /*!\brief Temporal layering mode indicating which temporal layering scheme to use. + * + * The value (refer to VP9E_TEMPORAL_LAYERING_MODE) specifies the + * temporal layering mode to use. + * + */ + int temporal_layering_mode; + } vpx_codec_enc_cfg_t; /**< alias for struct vpx_codec_enc_cfg */ + + /*!\brief vp9 svc extra configure parameters + * + * This defines max/min quantizers and scale factors for each layer + * + */ + typedef struct vpx_svc_parameters { + int max_quantizers[VPX_MAX_LAYERS]; /**< Max Q for each layer */ + int min_quantizers[VPX_MAX_LAYERS]; /**< Min Q for each layer */ + int scaling_factor_num[VPX_MAX_LAYERS]; /**< Scaling factor-numerator */ + int scaling_factor_den[VPX_MAX_LAYERS]; /**< Scaling factor-denominator */ + int temporal_layering_mode; /**< Temporal layering mode */ + } vpx_svc_extra_cfg_t; + + + /*!\brief Initialize an encoder instance + * + * Initializes a encoder context using the given interface. Applications + * should call the vpx_codec_enc_init convenience macro instead of this + * function directly, to ensure that the ABI version number parameter + * is properly initialized. + * + * If the library was configured with --disable-multithread, this call + * is not thread safe and should be guarded with a lock if being used + * in a multithreaded context. + * + * \param[in] ctx Pointer to this instance's context. + * \param[in] iface Pointer to the algorithm interface to use. + * \param[in] cfg Configuration to use, if known. May be NULL. + * \param[in] flags Bitfield of VPX_CODEC_USE_* flags + * \param[in] ver ABI version number. Must be set to + * VPX_ENCODER_ABI_VERSION + * \retval #VPX_CODEC_OK + * The decoder algorithm initialized. + * \retval #VPX_CODEC_MEM_ERROR + * Memory allocation failed. + */ + vpx_codec_err_t vpx_codec_enc_init_ver(vpx_codec_ctx_t *ctx, + vpx_codec_iface_t *iface, + const vpx_codec_enc_cfg_t *cfg, + vpx_codec_flags_t flags, + int ver); + + + /*!\brief Convenience macro for vpx_codec_enc_init_ver() + * + * Ensures the ABI version parameter is properly set. + */ +#define vpx_codec_enc_init(ctx, iface, cfg, flags) \ + vpx_codec_enc_init_ver(ctx, iface, cfg, flags, VPX_ENCODER_ABI_VERSION) + + + /*!\brief Initialize multi-encoder instance + * + * Initializes multi-encoder context using the given interface. + * Applications should call the vpx_codec_enc_init_multi convenience macro + * instead of this function directly, to ensure that the ABI version number + * parameter is properly initialized. + * + * \param[in] ctx Pointer to this instance's context. + * \param[in] iface Pointer to the algorithm interface to use. + * \param[in] cfg Configuration to use, if known. May be NULL. + * \param[in] num_enc Total number of encoders. + * \param[in] flags Bitfield of VPX_CODEC_USE_* flags + * \param[in] dsf Pointer to down-sampling factors. + * \param[in] ver ABI version number. Must be set to + * VPX_ENCODER_ABI_VERSION + * \retval #VPX_CODEC_OK + * The decoder algorithm initialized. + * \retval #VPX_CODEC_MEM_ERROR + * Memory allocation failed. + */ + vpx_codec_err_t vpx_codec_enc_init_multi_ver(vpx_codec_ctx_t *ctx, + vpx_codec_iface_t *iface, + vpx_codec_enc_cfg_t *cfg, + int num_enc, + vpx_codec_flags_t flags, + vpx_rational_t *dsf, + int ver); + + + /*!\brief Convenience macro for vpx_codec_enc_init_multi_ver() + * + * Ensures the ABI version parameter is properly set. + */ +#define vpx_codec_enc_init_multi(ctx, iface, cfg, num_enc, flags, dsf) \ + vpx_codec_enc_init_multi_ver(ctx, iface, cfg, num_enc, flags, dsf, \ + VPX_ENCODER_ABI_VERSION) + + + /*!\brief Get a default configuration + * + * Initializes a encoder configuration structure with default values. Supports + * the notion of "usages" so that an algorithm may offer different default + * settings depending on the user's intended goal. This function \ref SHOULD + * be called by all applications to initialize the configuration structure + * before specializing the configuration with application specific values. + * + * \param[in] iface Pointer to the algorithm interface to use. + * \param[out] cfg Configuration buffer to populate. + * \param[in] reserved Must set to 0 for VP8 and VP9. + * + * \retval #VPX_CODEC_OK + * The configuration was populated. + * \retval #VPX_CODEC_INCAPABLE + * Interface is not an encoder interface. + * \retval #VPX_CODEC_INVALID_PARAM + * A parameter was NULL, or the usage value was not recognized. + */ + vpx_codec_err_t vpx_codec_enc_config_default(vpx_codec_iface_t *iface, + vpx_codec_enc_cfg_t *cfg, + unsigned int reserved); + + + /*!\brief Set or change configuration + * + * Reconfigures an encoder instance according to the given configuration. + * + * \param[in] ctx Pointer to this instance's context + * \param[in] cfg Configuration buffer to use + * + * \retval #VPX_CODEC_OK + * The configuration was populated. + * \retval #VPX_CODEC_INCAPABLE + * Interface is not an encoder interface. + * \retval #VPX_CODEC_INVALID_PARAM + * A parameter was NULL, or the usage value was not recognized. + */ + vpx_codec_err_t vpx_codec_enc_config_set(vpx_codec_ctx_t *ctx, + const vpx_codec_enc_cfg_t *cfg); + + + /*!\brief Get global stream headers + * + * Retrieves a stream level global header packet, if supported by the codec. + * + * \param[in] ctx Pointer to this instance's context + * + * \retval NULL + * Encoder does not support global header + * \retval Non-NULL + * Pointer to buffer containing global header packet + */ + vpx_fixed_buf_t *vpx_codec_get_global_headers(vpx_codec_ctx_t *ctx); + + +#define VPX_DL_REALTIME (1) /**< deadline parameter analogous to + * VPx REALTIME mode. */ +#define VPX_DL_GOOD_QUALITY (1000000) /**< deadline parameter analogous to + * VPx GOOD QUALITY mode. */ +#define VPX_DL_BEST_QUALITY (0) /**< deadline parameter analogous to + * VPx BEST QUALITY mode. */ + /*!\brief Encode a frame + * + * Encodes a video frame at the given "presentation time." The presentation + * time stamp (PTS) \ref MUST be strictly increasing. + * + * The encoder supports the notion of a soft real-time deadline. Given a + * non-zero value to the deadline parameter, the encoder will make a "best + * effort" guarantee to return before the given time slice expires. It is + * implicit that limiting the available time to encode will degrade the + * output quality. The encoder can be given an unlimited time to produce the + * best possible frame by specifying a deadline of '0'. This deadline + * supercedes the VPx notion of "best quality, good quality, realtime". + * Applications that wish to map these former settings to the new deadline + * based system can use the symbols #VPX_DL_REALTIME, #VPX_DL_GOOD_QUALITY, + * and #VPX_DL_BEST_QUALITY. + * + * When the last frame has been passed to the encoder, this function should + * continue to be called, with the img parameter set to NULL. This will + * signal the end-of-stream condition to the encoder and allow it to encode + * any held buffers. Encoding is complete when vpx_codec_encode() is called + * and vpx_codec_get_cx_data() returns no data. + * + * \param[in] ctx Pointer to this instance's context + * \param[in] img Image data to encode, NULL to flush. + * \param[in] pts Presentation time stamp, in timebase units. + * \param[in] duration Duration to show frame, in timebase units. + * \param[in] flags Flags to use for encoding this frame. + * \param[in] deadline Time to spend encoding, in microseconds. (0=infinite) + * + * \retval #VPX_CODEC_OK + * The configuration was populated. + * \retval #VPX_CODEC_INCAPABLE + * Interface is not an encoder interface. + * \retval #VPX_CODEC_INVALID_PARAM + * A parameter was NULL, the image format is unsupported, etc. + */ + vpx_codec_err_t vpx_codec_encode(vpx_codec_ctx_t *ctx, + const vpx_image_t *img, + vpx_codec_pts_t pts, + unsigned long duration, + vpx_enc_frame_flags_t flags, + unsigned long deadline); + + /*!\brief Set compressed data output buffer + * + * Sets the buffer that the codec should output the compressed data + * into. This call effectively sets the buffer pointer returned in the + * next VPX_CODEC_CX_FRAME_PKT packet. Subsequent packets will be + * appended into this buffer. The buffer is preserved across frames, + * so applications must periodically call this function after flushing + * the accumulated compressed data to disk or to the network to reset + * the pointer to the buffer's head. + * + * `pad_before` bytes will be skipped before writing the compressed + * data, and `pad_after` bytes will be appended to the packet. The size + * of the packet will be the sum of the size of the actual compressed + * data, pad_before, and pad_after. The padding bytes will be preserved + * (not overwritten). + * + * Note that calling this function does not guarantee that the returned + * compressed data will be placed into the specified buffer. In the + * event that the encoded data will not fit into the buffer provided, + * the returned packet \ref MAY point to an internal buffer, as it would + * if this call were never used. In this event, the output packet will + * NOT have any padding, and the application must free space and copy it + * to the proper place. This is of particular note in configurations + * that may output multiple packets for a single encoded frame (e.g., lagged + * encoding) or if the application does not reset the buffer periodically. + * + * Applications may restore the default behavior of the codec providing + * the compressed data buffer by calling this function with a NULL + * buffer. + * + * Applications \ref MUSTNOT call this function during iteration of + * vpx_codec_get_cx_data(). + * + * \param[in] ctx Pointer to this instance's context + * \param[in] buf Buffer to store compressed data into + * \param[in] pad_before Bytes to skip before writing compressed data + * \param[in] pad_after Bytes to skip after writing compressed data + * + * \retval #VPX_CODEC_OK + * The buffer was set successfully. + * \retval #VPX_CODEC_INVALID_PARAM + * A parameter was NULL, the image format is unsupported, etc. + */ + vpx_codec_err_t vpx_codec_set_cx_data_buf(vpx_codec_ctx_t *ctx, + const vpx_fixed_buf_t *buf, + unsigned int pad_before, + unsigned int pad_after); + + + /*!\brief Encoded data iterator + * + * Iterates over a list of data packets to be passed from the encoder to the + * application. The different kinds of packets available are enumerated in + * #vpx_codec_cx_pkt_kind. + * + * #VPX_CODEC_CX_FRAME_PKT packets should be passed to the application's + * muxer. Multiple compressed frames may be in the list. + * #VPX_CODEC_STATS_PKT packets should be appended to a global buffer. + * + * The application \ref MUST silently ignore any packet kinds that it does + * not recognize or support. + * + * The data buffers returned from this function are only guaranteed to be + * valid until the application makes another call to any vpx_codec_* function. + * + * \param[in] ctx Pointer to this instance's context + * \param[in,out] iter Iterator storage, initialized to NULL + * + * \return Returns a pointer to an output data packet (compressed frame data, + * two-pass statistics, etc.) or NULL to signal end-of-list. + * + */ + const vpx_codec_cx_pkt_t *vpx_codec_get_cx_data(vpx_codec_ctx_t *ctx, + vpx_codec_iter_t *iter); + + + /*!\brief Get Preview Frame + * + * Returns an image that can be used as a preview. Shows the image as it would + * exist at the decompressor. The application \ref MUST NOT write into this + * image buffer. + * + * \param[in] ctx Pointer to this instance's context + * + * \return Returns a pointer to a preview image, or NULL if no image is + * available. + * + */ + const vpx_image_t *vpx_codec_get_preview_frame(vpx_codec_ctx_t *ctx); + + + /*!@} - end defgroup encoder*/ +#ifdef __cplusplus +} +#endif +#endif // VPX_VPX_ENCODER_H_ + diff --git a/thirdparty/libvpx/vpx/vpx_frame_buffer.h b/thirdparty/libvpx/vpx/vpx_frame_buffer.h new file mode 100644 index 0000000000..9036459af0 --- /dev/null +++ b/thirdparty/libvpx/vpx/vpx_frame_buffer.h @@ -0,0 +1,83 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VPX_VPX_FRAME_BUFFER_H_ +#define VPX_VPX_FRAME_BUFFER_H_ + +/*!\file + * \brief Describes the decoder external frame buffer interface. + */ + +#ifdef __cplusplus +extern "C" { +#endif + +#include "./vpx_integer.h" + +/*!\brief The maximum number of work buffers used by libvpx. + * Support maximum 4 threads to decode video in parallel. + * Each thread will use one work buffer. + * TODO(hkuang): Add support to set number of worker threads dynamically. + */ +#define VPX_MAXIMUM_WORK_BUFFERS 8 + +/*!\brief The maximum number of reference buffers that a VP9 encoder may use. + */ +#define VP9_MAXIMUM_REF_BUFFERS 8 + +/*!\brief External frame buffer + * + * This structure holds allocated frame buffers used by the decoder. + */ +typedef struct vpx_codec_frame_buffer { + uint8_t *data; /**< Pointer to the data buffer */ + size_t size; /**< Size of data in bytes */ + void *priv; /**< Frame's private data */ +} vpx_codec_frame_buffer_t; + +/*!\brief get frame buffer callback prototype + * + * This callback is invoked by the decoder to retrieve data for the frame + * buffer in order for the decode call to complete. The callback must + * allocate at least min_size in bytes and assign it to fb->data. The callback + * must zero out all the data allocated. Then the callback must set fb->size + * to the allocated size. The application does not need to align the allocated + * data. The callback is triggered when the decoder needs a frame buffer to + * decode a compressed image into. This function may be called more than once + * for every call to vpx_codec_decode. The application may set fb->priv to + * some data which will be passed back in the ximage and the release function + * call. |fb| is guaranteed to not be NULL. On success the callback must + * return 0. Any failure the callback must return a value less than 0. + * + * \param[in] priv Callback's private data + * \param[in] new_size Size in bytes needed by the buffer + * \param[in,out] fb Pointer to vpx_codec_frame_buffer_t + */ +typedef int (*vpx_get_frame_buffer_cb_fn_t)( + void *priv, size_t min_size, vpx_codec_frame_buffer_t *fb); + +/*!\brief release frame buffer callback prototype + * + * This callback is invoked by the decoder when the frame buffer is not + * referenced by any other buffers. |fb| is guaranteed to not be NULL. On + * success the callback must return 0. Any failure the callback must return + * a value less than 0. + * + * \param[in] priv Callback's private data + * \param[in] fb Pointer to vpx_codec_frame_buffer_t + */ +typedef int (*vpx_release_frame_buffer_cb_fn_t)( + void *priv, vpx_codec_frame_buffer_t *fb); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VPX_VPX_FRAME_BUFFER_H_ diff --git a/thirdparty/libvpx/vpx/vpx_image.h b/thirdparty/libvpx/vpx/vpx_image.h new file mode 100644 index 0000000000..7958c69806 --- /dev/null +++ b/thirdparty/libvpx/vpx/vpx_image.h @@ -0,0 +1,235 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +/*!\file + * \brief Describes the vpx image descriptor and associated operations + * + */ +#ifndef VPX_VPX_IMAGE_H_ +#define VPX_VPX_IMAGE_H_ + +#ifdef __cplusplus +extern "C" { +#endif + + /*!\brief Current ABI version number + * + * \internal + * If this file is altered in any way that changes the ABI, this value + * must be bumped. Examples include, but are not limited to, changing + * types, removing or reassigning enums, adding/removing/rearranging + * fields to structures + */ +#define VPX_IMAGE_ABI_VERSION (4) /**<\hideinitializer*/ + + +#define VPX_IMG_FMT_PLANAR 0x100 /**< Image is a planar format. */ +#define VPX_IMG_FMT_UV_FLIP 0x200 /**< V plane precedes U in memory. */ +#define VPX_IMG_FMT_HAS_ALPHA 0x400 /**< Image has an alpha channel. */ +#define VPX_IMG_FMT_HIGHBITDEPTH 0x800 /**< Image uses 16bit framebuffer. */ + + /*!\brief List of supported image formats */ + typedef enum vpx_img_fmt { + VPX_IMG_FMT_NONE, + VPX_IMG_FMT_RGB24, /**< 24 bit per pixel packed RGB */ + VPX_IMG_FMT_RGB32, /**< 32 bit per pixel packed 0RGB */ + VPX_IMG_FMT_RGB565, /**< 16 bit per pixel, 565 */ + VPX_IMG_FMT_RGB555, /**< 16 bit per pixel, 555 */ + VPX_IMG_FMT_UYVY, /**< UYVY packed YUV */ + VPX_IMG_FMT_YUY2, /**< YUYV packed YUV */ + VPX_IMG_FMT_YVYU, /**< YVYU packed YUV */ + VPX_IMG_FMT_BGR24, /**< 24 bit per pixel packed BGR */ + VPX_IMG_FMT_RGB32_LE, /**< 32 bit packed BGR0 */ + VPX_IMG_FMT_ARGB, /**< 32 bit packed ARGB, alpha=255 */ + VPX_IMG_FMT_ARGB_LE, /**< 32 bit packed BGRA, alpha=255 */ + VPX_IMG_FMT_RGB565_LE, /**< 16 bit per pixel, gggbbbbb rrrrrggg */ + VPX_IMG_FMT_RGB555_LE, /**< 16 bit per pixel, gggbbbbb 0rrrrrgg */ + VPX_IMG_FMT_YV12 = VPX_IMG_FMT_PLANAR | VPX_IMG_FMT_UV_FLIP | 1, /**< planar YVU */ + VPX_IMG_FMT_I420 = VPX_IMG_FMT_PLANAR | 2, + VPX_IMG_FMT_VPXYV12 = VPX_IMG_FMT_PLANAR | VPX_IMG_FMT_UV_FLIP | 3, /** < planar 4:2:0 format with vpx color space */ + VPX_IMG_FMT_VPXI420 = VPX_IMG_FMT_PLANAR | 4, + VPX_IMG_FMT_I422 = VPX_IMG_FMT_PLANAR | 5, + VPX_IMG_FMT_I444 = VPX_IMG_FMT_PLANAR | 6, + VPX_IMG_FMT_I440 = VPX_IMG_FMT_PLANAR | 7, + VPX_IMG_FMT_444A = VPX_IMG_FMT_PLANAR | VPX_IMG_FMT_HAS_ALPHA | 6, + VPX_IMG_FMT_I42016 = VPX_IMG_FMT_I420 | VPX_IMG_FMT_HIGHBITDEPTH, + VPX_IMG_FMT_I42216 = VPX_IMG_FMT_I422 | VPX_IMG_FMT_HIGHBITDEPTH, + VPX_IMG_FMT_I44416 = VPX_IMG_FMT_I444 | VPX_IMG_FMT_HIGHBITDEPTH, + VPX_IMG_FMT_I44016 = VPX_IMG_FMT_I440 | VPX_IMG_FMT_HIGHBITDEPTH + } vpx_img_fmt_t; /**< alias for enum vpx_img_fmt */ + + /*!\brief List of supported color spaces */ + typedef enum vpx_color_space { + VPX_CS_UNKNOWN = 0, /**< Unknown */ + VPX_CS_BT_601 = 1, /**< BT.601 */ + VPX_CS_BT_709 = 2, /**< BT.709 */ + VPX_CS_SMPTE_170 = 3, /**< SMPTE.170 */ + VPX_CS_SMPTE_240 = 4, /**< SMPTE.240 */ + VPX_CS_BT_2020 = 5, /**< BT.2020 */ + VPX_CS_RESERVED = 6, /**< Reserved */ + VPX_CS_SRGB = 7 /**< sRGB */ + } vpx_color_space_t; /**< alias for enum vpx_color_space */ + + /*!\brief List of supported color range */ + typedef enum vpx_color_range { + VPX_CR_STUDIO_RANGE = 0, /**< Y [16..235], UV [16..240] */ + VPX_CR_FULL_RANGE = 1 /**< YUV/RGB [0..255] */ + } vpx_color_range_t; /**< alias for enum vpx_color_range */ + + /**\brief Image Descriptor */ + typedef struct vpx_image { + vpx_img_fmt_t fmt; /**< Image Format */ + vpx_color_space_t cs; /**< Color Space */ + vpx_color_range_t range; /**< Color Range */ + + /* Image storage dimensions */ + unsigned int w; /**< Stored image width */ + unsigned int h; /**< Stored image height */ + unsigned int bit_depth; /**< Stored image bit-depth */ + + /* Image display dimensions */ + unsigned int d_w; /**< Displayed image width */ + unsigned int d_h; /**< Displayed image height */ + + /* Image intended rendering dimensions */ + unsigned int r_w; /**< Intended rendering image width */ + unsigned int r_h; /**< Intended rendering image height */ + + /* Chroma subsampling info */ + unsigned int x_chroma_shift; /**< subsampling order, X */ + unsigned int y_chroma_shift; /**< subsampling order, Y */ + + /* Image data pointers. */ +#define VPX_PLANE_PACKED 0 /**< To be used for all packed formats */ +#define VPX_PLANE_Y 0 /**< Y (Luminance) plane */ +#define VPX_PLANE_U 1 /**< U (Chroma) plane */ +#define VPX_PLANE_V 2 /**< V (Chroma) plane */ +#define VPX_PLANE_ALPHA 3 /**< A (Transparency) plane */ + unsigned char *planes[4]; /**< pointer to the top left pixel for each plane */ + int stride[4]; /**< stride between rows for each plane */ + + int bps; /**< bits per sample (for packed formats) */ + + /* The following member may be set by the application to associate data + * with this image. + */ + void *user_priv; /**< may be set by the application to associate data + * with this image. */ + + /* The following members should be treated as private. */ + unsigned char *img_data; /**< private */ + int img_data_owner; /**< private */ + int self_allocd; /**< private */ + + void *fb_priv; /**< Frame buffer data associated with the image. */ + } vpx_image_t; /**< alias for struct vpx_image */ + + /**\brief Representation of a rectangle on a surface */ + typedef struct vpx_image_rect { + unsigned int x; /**< leftmost column */ + unsigned int y; /**< topmost row */ + unsigned int w; /**< width */ + unsigned int h; /**< height */ + } vpx_image_rect_t; /**< alias for struct vpx_image_rect */ + + /*!\brief Open a descriptor, allocating storage for the underlying image + * + * Returns a descriptor for storing an image of the given format. The + * storage for the descriptor is allocated on the heap. + * + * \param[in] img Pointer to storage for descriptor. If this parameter + * is NULL, the storage for the descriptor will be + * allocated on the heap. + * \param[in] fmt Format for the image + * \param[in] d_w Width of the image + * \param[in] d_h Height of the image + * \param[in] align Alignment, in bytes, of the image buffer and + * each row in the image(stride). + * + * \return Returns a pointer to the initialized image descriptor. If the img + * parameter is non-null, the value of the img parameter will be + * returned. + */ + vpx_image_t *vpx_img_alloc(vpx_image_t *img, + vpx_img_fmt_t fmt, + unsigned int d_w, + unsigned int d_h, + unsigned int align); + + /*!\brief Open a descriptor, using existing storage for the underlying image + * + * Returns a descriptor for storing an image of the given format. The + * storage for descriptor has been allocated elsewhere, and a descriptor is + * desired to "wrap" that storage. + * + * \param[in] img Pointer to storage for descriptor. If this parameter + * is NULL, the storage for the descriptor will be + * allocated on the heap. + * \param[in] fmt Format for the image + * \param[in] d_w Width of the image + * \param[in] d_h Height of the image + * \param[in] align Alignment, in bytes, of each row in the image. + * \param[in] img_data Storage to use for the image + * + * \return Returns a pointer to the initialized image descriptor. If the img + * parameter is non-null, the value of the img parameter will be + * returned. + */ + vpx_image_t *vpx_img_wrap(vpx_image_t *img, + vpx_img_fmt_t fmt, + unsigned int d_w, + unsigned int d_h, + unsigned int align, + unsigned char *img_data); + + + /*!\brief Set the rectangle identifying the displayed portion of the image + * + * Updates the displayed rectangle (aka viewport) on the image surface to + * match the specified coordinates and size. + * + * \param[in] img Image descriptor + * \param[in] x leftmost column + * \param[in] y topmost row + * \param[in] w width + * \param[in] h height + * + * \return 0 if the requested rectangle is valid, nonzero otherwise. + */ + int vpx_img_set_rect(vpx_image_t *img, + unsigned int x, + unsigned int y, + unsigned int w, + unsigned int h); + + + /*!\brief Flip the image vertically (top for bottom) + * + * Adjusts the image descriptor's pointers and strides to make the image + * be referenced upside-down. + * + * \param[in] img Image descriptor + */ + void vpx_img_flip(vpx_image_t *img); + + /*!\brief Close an image descriptor + * + * Frees all allocated storage associated with an image descriptor. + * + * \param[in] img Image descriptor + */ + void vpx_img_free(vpx_image_t *img); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VPX_VPX_IMAGE_H_ diff --git a/thirdparty/libvpx/vpx/vpx_integer.h b/thirdparty/libvpx/vpx/vpx_integer.h new file mode 100644 index 0000000000..829c9d132c --- /dev/null +++ b/thirdparty/libvpx/vpx/vpx_integer.h @@ -0,0 +1,74 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VPX_VPX_INTEGER_H_ +#define VPX_VPX_INTEGER_H_ + +/* get ptrdiff_t, size_t, wchar_t, NULL */ +#include <stddef.h> + +#if defined(_MSC_VER) +#define VPX_FORCE_INLINE __forceinline +#define VPX_INLINE __inline +#else +#define VPX_FORCE_INLINE __inline__ __attribute__(always_inline) +// TODO(jbb): Allow a way to force inline off for older compilers. +#define VPX_INLINE inline +#endif + +#if (defined(_MSC_VER) && (_MSC_VER < 1600)) || defined(VPX_EMULATE_INTTYPES) +typedef signed char int8_t; +typedef signed short int16_t; +typedef signed int int32_t; + +typedef unsigned char uint8_t; +typedef unsigned short uint16_t; +typedef unsigned int uint32_t; + +#if (defined(_MSC_VER) && (_MSC_VER < 1600)) +typedef signed __int64 int64_t; +typedef unsigned __int64 uint64_t; +#define INT64_MAX _I64_MAX +#define INT32_MAX _I32_MAX +#define INT32_MIN _I32_MIN +#define INT16_MAX _I16_MAX +#define INT16_MIN _I16_MIN +#endif + +#ifndef _UINTPTR_T_DEFINED +typedef size_t uintptr_t; +#endif + +#else + +/* Most platforms have the C99 standard integer types. */ + +#if defined(__cplusplus) +# if !defined(__STDC_FORMAT_MACROS) +# define __STDC_FORMAT_MACROS +# endif +# if !defined(__STDC_LIMIT_MACROS) +# define __STDC_LIMIT_MACROS +# endif +#endif // __cplusplus + +#include <stdint.h> + +#endif + +/* VS2010 defines stdint.h, but not inttypes.h */ +#if defined(_MSC_VER) && _MSC_VER < 1800 +#define PRId64 "I64d" +#else +#include <inttypes.h> +#endif + +#endif // VPX_VPX_INTEGER_H_ diff --git a/thirdparty/libvpx/vpx_dsp/arm/idct16x16_1_add_neon.c b/thirdparty/libvpx/vpx_dsp/arm/idct16x16_1_add_neon.c new file mode 100644 index 0000000000..f734e48027 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/arm/idct16x16_1_add_neon.c @@ -0,0 +1,61 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +#include "vpx_dsp/inv_txfm.h" +#include "vpx_ports/mem.h" + +void vpx_idct16x16_1_add_neon( + int16_t *input, + uint8_t *dest, + int dest_stride) { + uint8x8_t d2u8, d3u8, d30u8, d31u8; + uint64x1_t d2u64, d3u64, d4u64, d5u64; + uint16x8_t q0u16, q9u16, q10u16, q11u16, q12u16; + int16x8_t q0s16; + uint8_t *d1, *d2; + int16_t i, j, a1, cospi_16_64 = 11585; + int16_t out = dct_const_round_shift(input[0] * cospi_16_64); + out = dct_const_round_shift(out * cospi_16_64); + a1 = ROUND_POWER_OF_TWO(out, 6); + + q0s16 = vdupq_n_s16(a1); + q0u16 = vreinterpretq_u16_s16(q0s16); + + for (d1 = d2 = dest, i = 0; i < 4; i++) { + for (j = 0; j < 2; j++) { + d2u64 = vld1_u64((const uint64_t *)d1); + d3u64 = vld1_u64((const uint64_t *)(d1 + 8)); + d1 += dest_stride; + d4u64 = vld1_u64((const uint64_t *)d1); + d5u64 = vld1_u64((const uint64_t *)(d1 + 8)); + d1 += dest_stride; + + q9u16 = vaddw_u8(q0u16, vreinterpret_u8_u64(d2u64)); + q10u16 = vaddw_u8(q0u16, vreinterpret_u8_u64(d3u64)); + q11u16 = vaddw_u8(q0u16, vreinterpret_u8_u64(d4u64)); + q12u16 = vaddw_u8(q0u16, vreinterpret_u8_u64(d5u64)); + + d2u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16)); + d3u8 = vqmovun_s16(vreinterpretq_s16_u16(q10u16)); + d30u8 = vqmovun_s16(vreinterpretq_s16_u16(q11u16)); + d31u8 = vqmovun_s16(vreinterpretq_s16_u16(q12u16)); + + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d2u8)); + vst1_u64((uint64_t *)(d2 + 8), vreinterpret_u64_u8(d3u8)); + d2 += dest_stride; + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d30u8)); + vst1_u64((uint64_t *)(d2 + 8), vreinterpret_u64_u8(d31u8)); + d2 += dest_stride; + } + } + return; +} diff --git a/thirdparty/libvpx/vpx_dsp/arm/idct16x16_add_neon.c b/thirdparty/libvpx/vpx_dsp/arm/idct16x16_add_neon.c new file mode 100644 index 0000000000..651ebb21f9 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/arm/idct16x16_add_neon.c @@ -0,0 +1,1317 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +#include "./vpx_config.h" +#include "vpx_dsp/txfm_common.h" + +static INLINE void TRANSPOSE8X8( + int16x8_t *q8s16, + int16x8_t *q9s16, + int16x8_t *q10s16, + int16x8_t *q11s16, + int16x8_t *q12s16, + int16x8_t *q13s16, + int16x8_t *q14s16, + int16x8_t *q15s16) { + int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16; + int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16; + int32x4x2_t q0x2s32, q1x2s32, q2x2s32, q3x2s32; + int16x8x2_t q0x2s16, q1x2s16, q2x2s16, q3x2s16; + + d16s16 = vget_low_s16(*q8s16); + d17s16 = vget_high_s16(*q8s16); + d18s16 = vget_low_s16(*q9s16); + d19s16 = vget_high_s16(*q9s16); + d20s16 = vget_low_s16(*q10s16); + d21s16 = vget_high_s16(*q10s16); + d22s16 = vget_low_s16(*q11s16); + d23s16 = vget_high_s16(*q11s16); + d24s16 = vget_low_s16(*q12s16); + d25s16 = vget_high_s16(*q12s16); + d26s16 = vget_low_s16(*q13s16); + d27s16 = vget_high_s16(*q13s16); + d28s16 = vget_low_s16(*q14s16); + d29s16 = vget_high_s16(*q14s16); + d30s16 = vget_low_s16(*q15s16); + d31s16 = vget_high_s16(*q15s16); + + *q8s16 = vcombine_s16(d16s16, d24s16); // vswp d17, d24 + *q9s16 = vcombine_s16(d18s16, d26s16); // vswp d19, d26 + *q10s16 = vcombine_s16(d20s16, d28s16); // vswp d21, d28 + *q11s16 = vcombine_s16(d22s16, d30s16); // vswp d23, d30 + *q12s16 = vcombine_s16(d17s16, d25s16); + *q13s16 = vcombine_s16(d19s16, d27s16); + *q14s16 = vcombine_s16(d21s16, d29s16); + *q15s16 = vcombine_s16(d23s16, d31s16); + + q0x2s32 = vtrnq_s32(vreinterpretq_s32_s16(*q8s16), + vreinterpretq_s32_s16(*q10s16)); + q1x2s32 = vtrnq_s32(vreinterpretq_s32_s16(*q9s16), + vreinterpretq_s32_s16(*q11s16)); + q2x2s32 = vtrnq_s32(vreinterpretq_s32_s16(*q12s16), + vreinterpretq_s32_s16(*q14s16)); + q3x2s32 = vtrnq_s32(vreinterpretq_s32_s16(*q13s16), + vreinterpretq_s32_s16(*q15s16)); + + q0x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[0]), // q8 + vreinterpretq_s16_s32(q1x2s32.val[0])); // q9 + q1x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[1]), // q10 + vreinterpretq_s16_s32(q1x2s32.val[1])); // q11 + q2x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[0]), // q12 + vreinterpretq_s16_s32(q3x2s32.val[0])); // q13 + q3x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[1]), // q14 + vreinterpretq_s16_s32(q3x2s32.val[1])); // q15 + + *q8s16 = q0x2s16.val[0]; + *q9s16 = q0x2s16.val[1]; + *q10s16 = q1x2s16.val[0]; + *q11s16 = q1x2s16.val[1]; + *q12s16 = q2x2s16.val[0]; + *q13s16 = q2x2s16.val[1]; + *q14s16 = q3x2s16.val[0]; + *q15s16 = q3x2s16.val[1]; + return; +} + +void vpx_idct16x16_256_add_neon_pass1( + int16_t *in, + int16_t *out, + int output_stride) { + int16x4_t d0s16, d1s16, d2s16, d3s16; + int16x4_t d8s16, d9s16, d10s16, d11s16, d12s16, d13s16, d14s16, d15s16; + int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16; + int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16; + uint64x1_t d16u64, d17u64, d18u64, d19u64, d20u64, d21u64, d22u64, d23u64; + uint64x1_t d24u64, d25u64, d26u64, d27u64, d28u64, d29u64, d30u64, d31u64; + int16x8_t q0s16, q1s16, q2s16, q3s16, q4s16, q5s16, q6s16, q7s16; + int16x8_t q8s16, q9s16, q10s16, q11s16, q12s16, q13s16, q14s16, q15s16; + int32x4_t q0s32, q1s32, q2s32, q3s32, q5s32, q6s32, q9s32; + int32x4_t q10s32, q11s32, q12s32, q13s32, q15s32; + int16x8x2_t q0x2s16; + + q0x2s16 = vld2q_s16(in); + q8s16 = q0x2s16.val[0]; + in += 16; + q0x2s16 = vld2q_s16(in); + q9s16 = q0x2s16.val[0]; + in += 16; + q0x2s16 = vld2q_s16(in); + q10s16 = q0x2s16.val[0]; + in += 16; + q0x2s16 = vld2q_s16(in); + q11s16 = q0x2s16.val[0]; + in += 16; + q0x2s16 = vld2q_s16(in); + q12s16 = q0x2s16.val[0]; + in += 16; + q0x2s16 = vld2q_s16(in); + q13s16 = q0x2s16.val[0]; + in += 16; + q0x2s16 = vld2q_s16(in); + q14s16 = q0x2s16.val[0]; + in += 16; + q0x2s16 = vld2q_s16(in); + q15s16 = q0x2s16.val[0]; + + TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, + &q12s16, &q13s16, &q14s16, &q15s16); + + d16s16 = vget_low_s16(q8s16); + d17s16 = vget_high_s16(q8s16); + d18s16 = vget_low_s16(q9s16); + d19s16 = vget_high_s16(q9s16); + d20s16 = vget_low_s16(q10s16); + d21s16 = vget_high_s16(q10s16); + d22s16 = vget_low_s16(q11s16); + d23s16 = vget_high_s16(q11s16); + d24s16 = vget_low_s16(q12s16); + d25s16 = vget_high_s16(q12s16); + d26s16 = vget_low_s16(q13s16); + d27s16 = vget_high_s16(q13s16); + d28s16 = vget_low_s16(q14s16); + d29s16 = vget_high_s16(q14s16); + d30s16 = vget_low_s16(q15s16); + d31s16 = vget_high_s16(q15s16); + + // stage 3 + d0s16 = vdup_n_s16(cospi_28_64); + d1s16 = vdup_n_s16(cospi_4_64); + + q2s32 = vmull_s16(d18s16, d0s16); + q3s32 = vmull_s16(d19s16, d0s16); + q5s32 = vmull_s16(d18s16, d1s16); + q6s32 = vmull_s16(d19s16, d1s16); + + q2s32 = vmlsl_s16(q2s32, d30s16, d1s16); + q3s32 = vmlsl_s16(q3s32, d31s16, d1s16); + q5s32 = vmlal_s16(q5s32, d30s16, d0s16); + q6s32 = vmlal_s16(q6s32, d31s16, d0s16); + + d2s16 = vdup_n_s16(cospi_12_64); + d3s16 = vdup_n_s16(cospi_20_64); + + d8s16 = vqrshrn_n_s32(q2s32, 14); + d9s16 = vqrshrn_n_s32(q3s32, 14); + d14s16 = vqrshrn_n_s32(q5s32, 14); + d15s16 = vqrshrn_n_s32(q6s32, 14); + q4s16 = vcombine_s16(d8s16, d9s16); + q7s16 = vcombine_s16(d14s16, d15s16); + + q2s32 = vmull_s16(d26s16, d2s16); + q3s32 = vmull_s16(d27s16, d2s16); + q9s32 = vmull_s16(d26s16, d3s16); + q15s32 = vmull_s16(d27s16, d3s16); + + q2s32 = vmlsl_s16(q2s32, d22s16, d3s16); + q3s32 = vmlsl_s16(q3s32, d23s16, d3s16); + q9s32 = vmlal_s16(q9s32, d22s16, d2s16); + q15s32 = vmlal_s16(q15s32, d23s16, d2s16); + + d10s16 = vqrshrn_n_s32(q2s32, 14); + d11s16 = vqrshrn_n_s32(q3s32, 14); + d12s16 = vqrshrn_n_s32(q9s32, 14); + d13s16 = vqrshrn_n_s32(q15s32, 14); + q5s16 = vcombine_s16(d10s16, d11s16); + q6s16 = vcombine_s16(d12s16, d13s16); + + // stage 4 + d30s16 = vdup_n_s16(cospi_16_64); + + q2s32 = vmull_s16(d16s16, d30s16); + q11s32 = vmull_s16(d17s16, d30s16); + q0s32 = vmull_s16(d24s16, d30s16); + q1s32 = vmull_s16(d25s16, d30s16); + + d30s16 = vdup_n_s16(cospi_24_64); + d31s16 = vdup_n_s16(cospi_8_64); + + q3s32 = vaddq_s32(q2s32, q0s32); + q12s32 = vaddq_s32(q11s32, q1s32); + q13s32 = vsubq_s32(q2s32, q0s32); + q1s32 = vsubq_s32(q11s32, q1s32); + + d16s16 = vqrshrn_n_s32(q3s32, 14); + d17s16 = vqrshrn_n_s32(q12s32, 14); + d18s16 = vqrshrn_n_s32(q13s32, 14); + d19s16 = vqrshrn_n_s32(q1s32, 14); + q8s16 = vcombine_s16(d16s16, d17s16); + q9s16 = vcombine_s16(d18s16, d19s16); + + q0s32 = vmull_s16(d20s16, d31s16); + q1s32 = vmull_s16(d21s16, d31s16); + q12s32 = vmull_s16(d20s16, d30s16); + q13s32 = vmull_s16(d21s16, d30s16); + + q0s32 = vmlal_s16(q0s32, d28s16, d30s16); + q1s32 = vmlal_s16(q1s32, d29s16, d30s16); + q12s32 = vmlsl_s16(q12s32, d28s16, d31s16); + q13s32 = vmlsl_s16(q13s32, d29s16, d31s16); + + d22s16 = vqrshrn_n_s32(q0s32, 14); + d23s16 = vqrshrn_n_s32(q1s32, 14); + d20s16 = vqrshrn_n_s32(q12s32, 14); + d21s16 = vqrshrn_n_s32(q13s32, 14); + q10s16 = vcombine_s16(d20s16, d21s16); + q11s16 = vcombine_s16(d22s16, d23s16); + + q13s16 = vsubq_s16(q4s16, q5s16); + q4s16 = vaddq_s16(q4s16, q5s16); + q14s16 = vsubq_s16(q7s16, q6s16); + q15s16 = vaddq_s16(q6s16, q7s16); + d26s16 = vget_low_s16(q13s16); + d27s16 = vget_high_s16(q13s16); + d28s16 = vget_low_s16(q14s16); + d29s16 = vget_high_s16(q14s16); + + // stage 5 + q0s16 = vaddq_s16(q8s16, q11s16); + q1s16 = vaddq_s16(q9s16, q10s16); + q2s16 = vsubq_s16(q9s16, q10s16); + q3s16 = vsubq_s16(q8s16, q11s16); + + d16s16 = vdup_n_s16(cospi_16_64); + + q11s32 = vmull_s16(d26s16, d16s16); + q12s32 = vmull_s16(d27s16, d16s16); + q9s32 = vmull_s16(d28s16, d16s16); + q10s32 = vmull_s16(d29s16, d16s16); + + q6s32 = vsubq_s32(q9s32, q11s32); + q13s32 = vsubq_s32(q10s32, q12s32); + q9s32 = vaddq_s32(q9s32, q11s32); + q10s32 = vaddq_s32(q10s32, q12s32); + + d10s16 = vqrshrn_n_s32(q6s32, 14); + d11s16 = vqrshrn_n_s32(q13s32, 14); + d12s16 = vqrshrn_n_s32(q9s32, 14); + d13s16 = vqrshrn_n_s32(q10s32, 14); + q5s16 = vcombine_s16(d10s16, d11s16); + q6s16 = vcombine_s16(d12s16, d13s16); + + // stage 6 + q8s16 = vaddq_s16(q0s16, q15s16); + q9s16 = vaddq_s16(q1s16, q6s16); + q10s16 = vaddq_s16(q2s16, q5s16); + q11s16 = vaddq_s16(q3s16, q4s16); + q12s16 = vsubq_s16(q3s16, q4s16); + q13s16 = vsubq_s16(q2s16, q5s16); + q14s16 = vsubq_s16(q1s16, q6s16); + q15s16 = vsubq_s16(q0s16, q15s16); + + d16u64 = vreinterpret_u64_s16(vget_low_s16(q8s16)); + d17u64 = vreinterpret_u64_s16(vget_high_s16(q8s16)); + d18u64 = vreinterpret_u64_s16(vget_low_s16(q9s16)); + d19u64 = vreinterpret_u64_s16(vget_high_s16(q9s16)); + d20u64 = vreinterpret_u64_s16(vget_low_s16(q10s16)); + d21u64 = vreinterpret_u64_s16(vget_high_s16(q10s16)); + d22u64 = vreinterpret_u64_s16(vget_low_s16(q11s16)); + d23u64 = vreinterpret_u64_s16(vget_high_s16(q11s16)); + d24u64 = vreinterpret_u64_s16(vget_low_s16(q12s16)); + d25u64 = vreinterpret_u64_s16(vget_high_s16(q12s16)); + d26u64 = vreinterpret_u64_s16(vget_low_s16(q13s16)); + d27u64 = vreinterpret_u64_s16(vget_high_s16(q13s16)); + d28u64 = vreinterpret_u64_s16(vget_low_s16(q14s16)); + d29u64 = vreinterpret_u64_s16(vget_high_s16(q14s16)); + d30u64 = vreinterpret_u64_s16(vget_low_s16(q15s16)); + d31u64 = vreinterpret_u64_s16(vget_high_s16(q15s16)); + + // store the data + output_stride >>= 1; // output_stride / 2, out is int16_t + vst1_u64((uint64_t *)out, d16u64); + out += output_stride; + vst1_u64((uint64_t *)out, d17u64); + out += output_stride; + vst1_u64((uint64_t *)out, d18u64); + out += output_stride; + vst1_u64((uint64_t *)out, d19u64); + out += output_stride; + vst1_u64((uint64_t *)out, d20u64); + out += output_stride; + vst1_u64((uint64_t *)out, d21u64); + out += output_stride; + vst1_u64((uint64_t *)out, d22u64); + out += output_stride; + vst1_u64((uint64_t *)out, d23u64); + out += output_stride; + vst1_u64((uint64_t *)out, d24u64); + out += output_stride; + vst1_u64((uint64_t *)out, d25u64); + out += output_stride; + vst1_u64((uint64_t *)out, d26u64); + out += output_stride; + vst1_u64((uint64_t *)out, d27u64); + out += output_stride; + vst1_u64((uint64_t *)out, d28u64); + out += output_stride; + vst1_u64((uint64_t *)out, d29u64); + out += output_stride; + vst1_u64((uint64_t *)out, d30u64); + out += output_stride; + vst1_u64((uint64_t *)out, d31u64); + return; +} + +void vpx_idct16x16_256_add_neon_pass2( + int16_t *src, + int16_t *out, + int16_t *pass1Output, + int16_t skip_adding, + uint8_t *dest, + int dest_stride) { + uint8_t *d; + uint8x8_t d12u8, d13u8; + int16x4_t d0s16, d1s16, d2s16, d3s16, d4s16, d5s16, d6s16, d7s16; + int16x4_t d8s16, d9s16, d10s16, d11s16, d12s16, d13s16, d14s16, d15s16; + int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16; + int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16; + uint64x1_t d24u64, d25u64, d26u64, d27u64; + int64x1_t d12s64, d13s64; + uint16x8_t q2u16, q3u16, q4u16, q5u16, q8u16; + uint16x8_t q9u16, q12u16, q13u16, q14u16, q15u16; + int16x8_t q0s16, q1s16, q2s16, q3s16, q4s16, q5s16, q6s16, q7s16; + int16x8_t q8s16, q9s16, q10s16, q11s16, q12s16, q13s16, q14s16, q15s16; + int32x4_t q0s32, q1s32, q2s32, q3s32, q4s32, q5s32, q6s32, q8s32, q9s32; + int32x4_t q10s32, q11s32, q12s32, q13s32; + int16x8x2_t q0x2s16; + + q0x2s16 = vld2q_s16(src); + q8s16 = q0x2s16.val[0]; + src += 16; + q0x2s16 = vld2q_s16(src); + q9s16 = q0x2s16.val[0]; + src += 16; + q0x2s16 = vld2q_s16(src); + q10s16 = q0x2s16.val[0]; + src += 16; + q0x2s16 = vld2q_s16(src); + q11s16 = q0x2s16.val[0]; + src += 16; + q0x2s16 = vld2q_s16(src); + q12s16 = q0x2s16.val[0]; + src += 16; + q0x2s16 = vld2q_s16(src); + q13s16 = q0x2s16.val[0]; + src += 16; + q0x2s16 = vld2q_s16(src); + q14s16 = q0x2s16.val[0]; + src += 16; + q0x2s16 = vld2q_s16(src); + q15s16 = q0x2s16.val[0]; + + TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, + &q12s16, &q13s16, &q14s16, &q15s16); + + d16s16 = vget_low_s16(q8s16); + d17s16 = vget_high_s16(q8s16); + d18s16 = vget_low_s16(q9s16); + d19s16 = vget_high_s16(q9s16); + d20s16 = vget_low_s16(q10s16); + d21s16 = vget_high_s16(q10s16); + d22s16 = vget_low_s16(q11s16); + d23s16 = vget_high_s16(q11s16); + d24s16 = vget_low_s16(q12s16); + d25s16 = vget_high_s16(q12s16); + d26s16 = vget_low_s16(q13s16); + d27s16 = vget_high_s16(q13s16); + d28s16 = vget_low_s16(q14s16); + d29s16 = vget_high_s16(q14s16); + d30s16 = vget_low_s16(q15s16); + d31s16 = vget_high_s16(q15s16); + + // stage 3 + d12s16 = vdup_n_s16(cospi_30_64); + d13s16 = vdup_n_s16(cospi_2_64); + + q2s32 = vmull_s16(d16s16, d12s16); + q3s32 = vmull_s16(d17s16, d12s16); + q1s32 = vmull_s16(d16s16, d13s16); + q4s32 = vmull_s16(d17s16, d13s16); + + q2s32 = vmlsl_s16(q2s32, d30s16, d13s16); + q3s32 = vmlsl_s16(q3s32, d31s16, d13s16); + q1s32 = vmlal_s16(q1s32, d30s16, d12s16); + q4s32 = vmlal_s16(q4s32, d31s16, d12s16); + + d0s16 = vqrshrn_n_s32(q2s32, 14); + d1s16 = vqrshrn_n_s32(q3s32, 14); + d14s16 = vqrshrn_n_s32(q1s32, 14); + d15s16 = vqrshrn_n_s32(q4s32, 14); + q0s16 = vcombine_s16(d0s16, d1s16); + q7s16 = vcombine_s16(d14s16, d15s16); + + d30s16 = vdup_n_s16(cospi_14_64); + d31s16 = vdup_n_s16(cospi_18_64); + + q2s32 = vmull_s16(d24s16, d30s16); + q3s32 = vmull_s16(d25s16, d30s16); + q4s32 = vmull_s16(d24s16, d31s16); + q5s32 = vmull_s16(d25s16, d31s16); + + q2s32 = vmlsl_s16(q2s32, d22s16, d31s16); + q3s32 = vmlsl_s16(q3s32, d23s16, d31s16); + q4s32 = vmlal_s16(q4s32, d22s16, d30s16); + q5s32 = vmlal_s16(q5s32, d23s16, d30s16); + + d2s16 = vqrshrn_n_s32(q2s32, 14); + d3s16 = vqrshrn_n_s32(q3s32, 14); + d12s16 = vqrshrn_n_s32(q4s32, 14); + d13s16 = vqrshrn_n_s32(q5s32, 14); + q1s16 = vcombine_s16(d2s16, d3s16); + q6s16 = vcombine_s16(d12s16, d13s16); + + d30s16 = vdup_n_s16(cospi_22_64); + d31s16 = vdup_n_s16(cospi_10_64); + + q11s32 = vmull_s16(d20s16, d30s16); + q12s32 = vmull_s16(d21s16, d30s16); + q4s32 = vmull_s16(d20s16, d31s16); + q5s32 = vmull_s16(d21s16, d31s16); + + q11s32 = vmlsl_s16(q11s32, d26s16, d31s16); + q12s32 = vmlsl_s16(q12s32, d27s16, d31s16); + q4s32 = vmlal_s16(q4s32, d26s16, d30s16); + q5s32 = vmlal_s16(q5s32, d27s16, d30s16); + + d4s16 = vqrshrn_n_s32(q11s32, 14); + d5s16 = vqrshrn_n_s32(q12s32, 14); + d11s16 = vqrshrn_n_s32(q5s32, 14); + d10s16 = vqrshrn_n_s32(q4s32, 14); + q2s16 = vcombine_s16(d4s16, d5s16); + q5s16 = vcombine_s16(d10s16, d11s16); + + d30s16 = vdup_n_s16(cospi_6_64); + d31s16 = vdup_n_s16(cospi_26_64); + + q10s32 = vmull_s16(d28s16, d30s16); + q11s32 = vmull_s16(d29s16, d30s16); + q12s32 = vmull_s16(d28s16, d31s16); + q13s32 = vmull_s16(d29s16, d31s16); + + q10s32 = vmlsl_s16(q10s32, d18s16, d31s16); + q11s32 = vmlsl_s16(q11s32, d19s16, d31s16); + q12s32 = vmlal_s16(q12s32, d18s16, d30s16); + q13s32 = vmlal_s16(q13s32, d19s16, d30s16); + + d6s16 = vqrshrn_n_s32(q10s32, 14); + d7s16 = vqrshrn_n_s32(q11s32, 14); + d8s16 = vqrshrn_n_s32(q12s32, 14); + d9s16 = vqrshrn_n_s32(q13s32, 14); + q3s16 = vcombine_s16(d6s16, d7s16); + q4s16 = vcombine_s16(d8s16, d9s16); + + // stage 3 + q9s16 = vsubq_s16(q0s16, q1s16); + q0s16 = vaddq_s16(q0s16, q1s16); + q10s16 = vsubq_s16(q3s16, q2s16); + q11s16 = vaddq_s16(q2s16, q3s16); + q12s16 = vaddq_s16(q4s16, q5s16); + q13s16 = vsubq_s16(q4s16, q5s16); + q14s16 = vsubq_s16(q7s16, q6s16); + q7s16 = vaddq_s16(q6s16, q7s16); + + // stage 4 + d18s16 = vget_low_s16(q9s16); + d19s16 = vget_high_s16(q9s16); + d20s16 = vget_low_s16(q10s16); + d21s16 = vget_high_s16(q10s16); + d26s16 = vget_low_s16(q13s16); + d27s16 = vget_high_s16(q13s16); + d28s16 = vget_low_s16(q14s16); + d29s16 = vget_high_s16(q14s16); + + d30s16 = vdup_n_s16(cospi_8_64); + d31s16 = vdup_n_s16(cospi_24_64); + + q2s32 = vmull_s16(d18s16, d31s16); + q3s32 = vmull_s16(d19s16, d31s16); + q4s32 = vmull_s16(d28s16, d31s16); + q5s32 = vmull_s16(d29s16, d31s16); + + q2s32 = vmlal_s16(q2s32, d28s16, d30s16); + q3s32 = vmlal_s16(q3s32, d29s16, d30s16); + q4s32 = vmlsl_s16(q4s32, d18s16, d30s16); + q5s32 = vmlsl_s16(q5s32, d19s16, d30s16); + + d12s16 = vqrshrn_n_s32(q2s32, 14); + d13s16 = vqrshrn_n_s32(q3s32, 14); + d2s16 = vqrshrn_n_s32(q4s32, 14); + d3s16 = vqrshrn_n_s32(q5s32, 14); + q1s16 = vcombine_s16(d2s16, d3s16); + q6s16 = vcombine_s16(d12s16, d13s16); + + q3s16 = q11s16; + q4s16 = q12s16; + + d30s16 = vdup_n_s16(-cospi_8_64); + q11s32 = vmull_s16(d26s16, d30s16); + q12s32 = vmull_s16(d27s16, d30s16); + q8s32 = vmull_s16(d20s16, d30s16); + q9s32 = vmull_s16(d21s16, d30s16); + + q11s32 = vmlsl_s16(q11s32, d20s16, d31s16); + q12s32 = vmlsl_s16(q12s32, d21s16, d31s16); + q8s32 = vmlal_s16(q8s32, d26s16, d31s16); + q9s32 = vmlal_s16(q9s32, d27s16, d31s16); + + d4s16 = vqrshrn_n_s32(q11s32, 14); + d5s16 = vqrshrn_n_s32(q12s32, 14); + d10s16 = vqrshrn_n_s32(q8s32, 14); + d11s16 = vqrshrn_n_s32(q9s32, 14); + q2s16 = vcombine_s16(d4s16, d5s16); + q5s16 = vcombine_s16(d10s16, d11s16); + + // stage 5 + q8s16 = vaddq_s16(q0s16, q3s16); + q9s16 = vaddq_s16(q1s16, q2s16); + q10s16 = vsubq_s16(q1s16, q2s16); + q11s16 = vsubq_s16(q0s16, q3s16); + q12s16 = vsubq_s16(q7s16, q4s16); + q13s16 = vsubq_s16(q6s16, q5s16); + q14s16 = vaddq_s16(q6s16, q5s16); + q15s16 = vaddq_s16(q7s16, q4s16); + + // stage 6 + d20s16 = vget_low_s16(q10s16); + d21s16 = vget_high_s16(q10s16); + d22s16 = vget_low_s16(q11s16); + d23s16 = vget_high_s16(q11s16); + d24s16 = vget_low_s16(q12s16); + d25s16 = vget_high_s16(q12s16); + d26s16 = vget_low_s16(q13s16); + d27s16 = vget_high_s16(q13s16); + + d14s16 = vdup_n_s16(cospi_16_64); + + q3s32 = vmull_s16(d26s16, d14s16); + q4s32 = vmull_s16(d27s16, d14s16); + q0s32 = vmull_s16(d20s16, d14s16); + q1s32 = vmull_s16(d21s16, d14s16); + + q5s32 = vsubq_s32(q3s32, q0s32); + q6s32 = vsubq_s32(q4s32, q1s32); + q10s32 = vaddq_s32(q3s32, q0s32); + q4s32 = vaddq_s32(q4s32, q1s32); + + d4s16 = vqrshrn_n_s32(q5s32, 14); + d5s16 = vqrshrn_n_s32(q6s32, 14); + d10s16 = vqrshrn_n_s32(q10s32, 14); + d11s16 = vqrshrn_n_s32(q4s32, 14); + q2s16 = vcombine_s16(d4s16, d5s16); + q5s16 = vcombine_s16(d10s16, d11s16); + + q0s32 = vmull_s16(d22s16, d14s16); + q1s32 = vmull_s16(d23s16, d14s16); + q13s32 = vmull_s16(d24s16, d14s16); + q6s32 = vmull_s16(d25s16, d14s16); + + q10s32 = vsubq_s32(q13s32, q0s32); + q4s32 = vsubq_s32(q6s32, q1s32); + q13s32 = vaddq_s32(q13s32, q0s32); + q6s32 = vaddq_s32(q6s32, q1s32); + + d6s16 = vqrshrn_n_s32(q10s32, 14); + d7s16 = vqrshrn_n_s32(q4s32, 14); + d8s16 = vqrshrn_n_s32(q13s32, 14); + d9s16 = vqrshrn_n_s32(q6s32, 14); + q3s16 = vcombine_s16(d6s16, d7s16); + q4s16 = vcombine_s16(d8s16, d9s16); + + // stage 7 + if (skip_adding != 0) { + d = dest; + // load the data in pass1 + q0s16 = vld1q_s16(pass1Output); + pass1Output += 8; + q1s16 = vld1q_s16(pass1Output); + pass1Output += 8; + d12s64 = vld1_s64((int64_t *)dest); + dest += dest_stride; + d13s64 = vld1_s64((int64_t *)dest); + dest += dest_stride; + + q12s16 = vaddq_s16(q0s16, q15s16); + q13s16 = vaddq_s16(q1s16, q14s16); + q12s16 = vrshrq_n_s16(q12s16, 6); + q13s16 = vrshrq_n_s16(q13s16, 6); + q12u16 = vaddw_u8(vreinterpretq_u16_s16(q12s16), + vreinterpret_u8_s64(d12s64)); + q13u16 = vaddw_u8(vreinterpretq_u16_s16(q13s16), + vreinterpret_u8_s64(d13s64)); + d12u8 = vqmovun_s16(vreinterpretq_s16_u16(q12u16)); + d13u8 = vqmovun_s16(vreinterpretq_s16_u16(q13u16)); + vst1_u64((uint64_t *)d, vreinterpret_u64_u8(d12u8)); + d += dest_stride; + vst1_u64((uint64_t *)d, vreinterpret_u64_u8(d13u8)); + d += dest_stride; + q14s16 = vsubq_s16(q1s16, q14s16); + q15s16 = vsubq_s16(q0s16, q15s16); + + q10s16 = vld1q_s16(pass1Output); + pass1Output += 8; + q11s16 = vld1q_s16(pass1Output); + pass1Output += 8; + d12s64 = vld1_s64((int64_t *)dest); + dest += dest_stride; + d13s64 = vld1_s64((int64_t *)dest); + dest += dest_stride; + q12s16 = vaddq_s16(q10s16, q5s16); + q13s16 = vaddq_s16(q11s16, q4s16); + q12s16 = vrshrq_n_s16(q12s16, 6); + q13s16 = vrshrq_n_s16(q13s16, 6); + q12u16 = vaddw_u8(vreinterpretq_u16_s16(q12s16), + vreinterpret_u8_s64(d12s64)); + q13u16 = vaddw_u8(vreinterpretq_u16_s16(q13s16), + vreinterpret_u8_s64(d13s64)); + d12u8 = vqmovun_s16(vreinterpretq_s16_u16(q12u16)); + d13u8 = vqmovun_s16(vreinterpretq_s16_u16(q13u16)); + vst1_u64((uint64_t *)d, vreinterpret_u64_u8(d12u8)); + d += dest_stride; + vst1_u64((uint64_t *)d, vreinterpret_u64_u8(d13u8)); + d += dest_stride; + q4s16 = vsubq_s16(q11s16, q4s16); + q5s16 = vsubq_s16(q10s16, q5s16); + + q0s16 = vld1q_s16(pass1Output); + pass1Output += 8; + q1s16 = vld1q_s16(pass1Output); + pass1Output += 8; + d12s64 = vld1_s64((int64_t *)dest); + dest += dest_stride; + d13s64 = vld1_s64((int64_t *)dest); + dest += dest_stride; + q12s16 = vaddq_s16(q0s16, q3s16); + q13s16 = vaddq_s16(q1s16, q2s16); + q12s16 = vrshrq_n_s16(q12s16, 6); + q13s16 = vrshrq_n_s16(q13s16, 6); + q12u16 = vaddw_u8(vreinterpretq_u16_s16(q12s16), + vreinterpret_u8_s64(d12s64)); + q13u16 = vaddw_u8(vreinterpretq_u16_s16(q13s16), + vreinterpret_u8_s64(d13s64)); + d12u8 = vqmovun_s16(vreinterpretq_s16_u16(q12u16)); + d13u8 = vqmovun_s16(vreinterpretq_s16_u16(q13u16)); + vst1_u64((uint64_t *)d, vreinterpret_u64_u8(d12u8)); + d += dest_stride; + vst1_u64((uint64_t *)d, vreinterpret_u64_u8(d13u8)); + d += dest_stride; + q2s16 = vsubq_s16(q1s16, q2s16); + q3s16 = vsubq_s16(q0s16, q3s16); + + q10s16 = vld1q_s16(pass1Output); + pass1Output += 8; + q11s16 = vld1q_s16(pass1Output); + d12s64 = vld1_s64((int64_t *)dest); + dest += dest_stride; + d13s64 = vld1_s64((int64_t *)dest); + dest += dest_stride; + q12s16 = vaddq_s16(q10s16, q9s16); + q13s16 = vaddq_s16(q11s16, q8s16); + q12s16 = vrshrq_n_s16(q12s16, 6); + q13s16 = vrshrq_n_s16(q13s16, 6); + q12u16 = vaddw_u8(vreinterpretq_u16_s16(q12s16), + vreinterpret_u8_s64(d12s64)); + q13u16 = vaddw_u8(vreinterpretq_u16_s16(q13s16), + vreinterpret_u8_s64(d13s64)); + d12u8 = vqmovun_s16(vreinterpretq_s16_u16(q12u16)); + d13u8 = vqmovun_s16(vreinterpretq_s16_u16(q13u16)); + vst1_u64((uint64_t *)d, vreinterpret_u64_u8(d12u8)); + d += dest_stride; + vst1_u64((uint64_t *)d, vreinterpret_u64_u8(d13u8)); + d += dest_stride; + q8s16 = vsubq_s16(q11s16, q8s16); + q9s16 = vsubq_s16(q10s16, q9s16); + + // store the data out 8,9,10,11,12,13,14,15 + d12s64 = vld1_s64((int64_t *)dest); + dest += dest_stride; + q8s16 = vrshrq_n_s16(q8s16, 6); + q8u16 = vaddw_u8(vreinterpretq_u16_s16(q8s16), + vreinterpret_u8_s64(d12s64)); + d12u8 = vqmovun_s16(vreinterpretq_s16_u16(q8u16)); + vst1_u64((uint64_t *)d, vreinterpret_u64_u8(d12u8)); + d += dest_stride; + + d12s64 = vld1_s64((int64_t *)dest); + dest += dest_stride; + q9s16 = vrshrq_n_s16(q9s16, 6); + q9u16 = vaddw_u8(vreinterpretq_u16_s16(q9s16), + vreinterpret_u8_s64(d12s64)); + d12u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16)); + vst1_u64((uint64_t *)d, vreinterpret_u64_u8(d12u8)); + d += dest_stride; + + d12s64 = vld1_s64((int64_t *)dest); + dest += dest_stride; + q2s16 = vrshrq_n_s16(q2s16, 6); + q2u16 = vaddw_u8(vreinterpretq_u16_s16(q2s16), + vreinterpret_u8_s64(d12s64)); + d12u8 = vqmovun_s16(vreinterpretq_s16_u16(q2u16)); + vst1_u64((uint64_t *)d, vreinterpret_u64_u8(d12u8)); + d += dest_stride; + + d12s64 = vld1_s64((int64_t *)dest); + dest += dest_stride; + q3s16 = vrshrq_n_s16(q3s16, 6); + q3u16 = vaddw_u8(vreinterpretq_u16_s16(q3s16), + vreinterpret_u8_s64(d12s64)); + d12u8 = vqmovun_s16(vreinterpretq_s16_u16(q3u16)); + vst1_u64((uint64_t *)d, vreinterpret_u64_u8(d12u8)); + d += dest_stride; + + d12s64 = vld1_s64((int64_t *)dest); + dest += dest_stride; + q4s16 = vrshrq_n_s16(q4s16, 6); + q4u16 = vaddw_u8(vreinterpretq_u16_s16(q4s16), + vreinterpret_u8_s64(d12s64)); + d12u8 = vqmovun_s16(vreinterpretq_s16_u16(q4u16)); + vst1_u64((uint64_t *)d, vreinterpret_u64_u8(d12u8)); + d += dest_stride; + + d12s64 = vld1_s64((int64_t *)dest); + dest += dest_stride; + q5s16 = vrshrq_n_s16(q5s16, 6); + q5u16 = vaddw_u8(vreinterpretq_u16_s16(q5s16), + vreinterpret_u8_s64(d12s64)); + d12u8 = vqmovun_s16(vreinterpretq_s16_u16(q5u16)); + vst1_u64((uint64_t *)d, vreinterpret_u64_u8(d12u8)); + d += dest_stride; + + d12s64 = vld1_s64((int64_t *)dest); + dest += dest_stride; + q14s16 = vrshrq_n_s16(q14s16, 6); + q14u16 = vaddw_u8(vreinterpretq_u16_s16(q14s16), + vreinterpret_u8_s64(d12s64)); + d12u8 = vqmovun_s16(vreinterpretq_s16_u16(q14u16)); + vst1_u64((uint64_t *)d, vreinterpret_u64_u8(d12u8)); + d += dest_stride; + + d12s64 = vld1_s64((int64_t *)dest); + q15s16 = vrshrq_n_s16(q15s16, 6); + q15u16 = vaddw_u8(vreinterpretq_u16_s16(q15s16), + vreinterpret_u8_s64(d12s64)); + d12u8 = vqmovun_s16(vreinterpretq_s16_u16(q15u16)); + vst1_u64((uint64_t *)d, vreinterpret_u64_u8(d12u8)); + } else { // skip_adding_dest + q0s16 = vld1q_s16(pass1Output); + pass1Output += 8; + q1s16 = vld1q_s16(pass1Output); + pass1Output += 8; + q12s16 = vaddq_s16(q0s16, q15s16); + q13s16 = vaddq_s16(q1s16, q14s16); + d24u64 = vreinterpret_u64_s16(vget_low_s16(q12s16)); + d25u64 = vreinterpret_u64_s16(vget_high_s16(q12s16)); + d26u64 = vreinterpret_u64_s16(vget_low_s16(q13s16)); + d27u64 = vreinterpret_u64_s16(vget_high_s16(q13s16)); + vst1_u64((uint64_t *)out, d24u64); + out += 4; + vst1_u64((uint64_t *)out, d25u64); + out += 12; + vst1_u64((uint64_t *)out, d26u64); + out += 4; + vst1_u64((uint64_t *)out, d27u64); + out += 12; + q14s16 = vsubq_s16(q1s16, q14s16); + q15s16 = vsubq_s16(q0s16, q15s16); + + q10s16 = vld1q_s16(pass1Output); + pass1Output += 8; + q11s16 = vld1q_s16(pass1Output); + pass1Output += 8; + q12s16 = vaddq_s16(q10s16, q5s16); + q13s16 = vaddq_s16(q11s16, q4s16); + d24u64 = vreinterpret_u64_s16(vget_low_s16(q12s16)); + d25u64 = vreinterpret_u64_s16(vget_high_s16(q12s16)); + d26u64 = vreinterpret_u64_s16(vget_low_s16(q13s16)); + d27u64 = vreinterpret_u64_s16(vget_high_s16(q13s16)); + vst1_u64((uint64_t *)out, d24u64); + out += 4; + vst1_u64((uint64_t *)out, d25u64); + out += 12; + vst1_u64((uint64_t *)out, d26u64); + out += 4; + vst1_u64((uint64_t *)out, d27u64); + out += 12; + q4s16 = vsubq_s16(q11s16, q4s16); + q5s16 = vsubq_s16(q10s16, q5s16); + + q0s16 = vld1q_s16(pass1Output); + pass1Output += 8; + q1s16 = vld1q_s16(pass1Output); + pass1Output += 8; + q12s16 = vaddq_s16(q0s16, q3s16); + q13s16 = vaddq_s16(q1s16, q2s16); + d24u64 = vreinterpret_u64_s16(vget_low_s16(q12s16)); + d25u64 = vreinterpret_u64_s16(vget_high_s16(q12s16)); + d26u64 = vreinterpret_u64_s16(vget_low_s16(q13s16)); + d27u64 = vreinterpret_u64_s16(vget_high_s16(q13s16)); + vst1_u64((uint64_t *)out, d24u64); + out += 4; + vst1_u64((uint64_t *)out, d25u64); + out += 12; + vst1_u64((uint64_t *)out, d26u64); + out += 4; + vst1_u64((uint64_t *)out, d27u64); + out += 12; + q2s16 = vsubq_s16(q1s16, q2s16); + q3s16 = vsubq_s16(q0s16, q3s16); + + q10s16 = vld1q_s16(pass1Output); + pass1Output += 8; + q11s16 = vld1q_s16(pass1Output); + pass1Output += 8; + q12s16 = vaddq_s16(q10s16, q9s16); + q13s16 = vaddq_s16(q11s16, q8s16); + d24u64 = vreinterpret_u64_s16(vget_low_s16(q12s16)); + d25u64 = vreinterpret_u64_s16(vget_high_s16(q12s16)); + d26u64 = vreinterpret_u64_s16(vget_low_s16(q13s16)); + d27u64 = vreinterpret_u64_s16(vget_high_s16(q13s16)); + vst1_u64((uint64_t *)out, d24u64); + out += 4; + vst1_u64((uint64_t *)out, d25u64); + out += 12; + vst1_u64((uint64_t *)out, d26u64); + out += 4; + vst1_u64((uint64_t *)out, d27u64); + out += 12; + q8s16 = vsubq_s16(q11s16, q8s16); + q9s16 = vsubq_s16(q10s16, q9s16); + + vst1_u64((uint64_t *)out, vreinterpret_u64_s16(vget_low_s16(q8s16))); + out += 4; + vst1_u64((uint64_t *)out, vreinterpret_u64_s16(vget_high_s16(q8s16))); + out += 12; + vst1_u64((uint64_t *)out, vreinterpret_u64_s16(vget_low_s16(q9s16))); + out += 4; + vst1_u64((uint64_t *)out, vreinterpret_u64_s16(vget_high_s16(q9s16))); + out += 12; + vst1_u64((uint64_t *)out, vreinterpret_u64_s16(vget_low_s16(q2s16))); + out += 4; + vst1_u64((uint64_t *)out, vreinterpret_u64_s16(vget_high_s16(q2s16))); + out += 12; + vst1_u64((uint64_t *)out, vreinterpret_u64_s16(vget_low_s16(q3s16))); + out += 4; + vst1_u64((uint64_t *)out, vreinterpret_u64_s16(vget_high_s16(q3s16))); + out += 12; + vst1_u64((uint64_t *)out, vreinterpret_u64_s16(vget_low_s16(q4s16))); + out += 4; + vst1_u64((uint64_t *)out, vreinterpret_u64_s16(vget_high_s16(q4s16))); + out += 12; + vst1_u64((uint64_t *)out, vreinterpret_u64_s16(vget_low_s16(q5s16))); + out += 4; + vst1_u64((uint64_t *)out, vreinterpret_u64_s16(vget_high_s16(q5s16))); + out += 12; + vst1_u64((uint64_t *)out, vreinterpret_u64_s16(vget_low_s16(q14s16))); + out += 4; + vst1_u64((uint64_t *)out, vreinterpret_u64_s16(vget_high_s16(q14s16))); + out += 12; + vst1_u64((uint64_t *)out, vreinterpret_u64_s16(vget_low_s16(q15s16))); + out += 4; + vst1_u64((uint64_t *)out, vreinterpret_u64_s16(vget_high_s16(q15s16))); + } + return; +} + +void vpx_idct16x16_10_add_neon_pass1( + int16_t *in, + int16_t *out, + int output_stride) { + int16x4_t d4s16; + int16x4_t d8s16, d9s16, d10s16, d11s16, d12s16, d13s16, d14s16, d15s16; + uint64x1_t d4u64, d5u64, d18u64, d19u64, d20u64, d21u64, d22u64, d23u64; + uint64x1_t d24u64, d25u64, d26u64, d27u64, d28u64, d29u64, d30u64, d31u64; + int16x8_t q0s16, q1s16, q2s16, q4s16, q5s16, q6s16, q7s16; + int16x8_t q8s16, q9s16, q10s16, q11s16, q12s16, q13s16, q14s16, q15s16; + int32x4_t q6s32, q9s32; + int32x4_t q10s32, q11s32, q12s32, q15s32; + int16x8x2_t q0x2s16; + + q0x2s16 = vld2q_s16(in); + q8s16 = q0x2s16.val[0]; + in += 16; + q0x2s16 = vld2q_s16(in); + q9s16 = q0x2s16.val[0]; + in += 16; + q0x2s16 = vld2q_s16(in); + q10s16 = q0x2s16.val[0]; + in += 16; + q0x2s16 = vld2q_s16(in); + q11s16 = q0x2s16.val[0]; + in += 16; + q0x2s16 = vld2q_s16(in); + q12s16 = q0x2s16.val[0]; + in += 16; + q0x2s16 = vld2q_s16(in); + q13s16 = q0x2s16.val[0]; + in += 16; + q0x2s16 = vld2q_s16(in); + q14s16 = q0x2s16.val[0]; + in += 16; + q0x2s16 = vld2q_s16(in); + q15s16 = q0x2s16.val[0]; + + TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, + &q12s16, &q13s16, &q14s16, &q15s16); + + // stage 3 + q0s16 = vdupq_n_s16(cospi_28_64 * 2); + q1s16 = vdupq_n_s16(cospi_4_64 * 2); + + q4s16 = vqrdmulhq_s16(q9s16, q0s16); + q7s16 = vqrdmulhq_s16(q9s16, q1s16); + + // stage 4 + q1s16 = vdupq_n_s16(cospi_16_64 * 2); + d4s16 = vdup_n_s16(cospi_16_64); + + q8s16 = vqrdmulhq_s16(q8s16, q1s16); + + d8s16 = vget_low_s16(q4s16); + d9s16 = vget_high_s16(q4s16); + d14s16 = vget_low_s16(q7s16); + d15s16 = vget_high_s16(q7s16); + q9s32 = vmull_s16(d14s16, d4s16); + q10s32 = vmull_s16(d15s16, d4s16); + q12s32 = vmull_s16(d9s16, d4s16); + q11s32 = vmull_s16(d8s16, d4s16); + + q15s32 = vsubq_s32(q10s32, q12s32); + q6s32 = vsubq_s32(q9s32, q11s32); + q9s32 = vaddq_s32(q9s32, q11s32); + q10s32 = vaddq_s32(q10s32, q12s32); + + d11s16 = vqrshrn_n_s32(q15s32, 14); + d10s16 = vqrshrn_n_s32(q6s32, 14); + d12s16 = vqrshrn_n_s32(q9s32, 14); + d13s16 = vqrshrn_n_s32(q10s32, 14); + q5s16 = vcombine_s16(d10s16, d11s16); + q6s16 = vcombine_s16(d12s16, d13s16); + + // stage 6 + q2s16 = vaddq_s16(q8s16, q7s16); + q9s16 = vaddq_s16(q8s16, q6s16); + q10s16 = vaddq_s16(q8s16, q5s16); + q11s16 = vaddq_s16(q8s16, q4s16); + q12s16 = vsubq_s16(q8s16, q4s16); + q13s16 = vsubq_s16(q8s16, q5s16); + q14s16 = vsubq_s16(q8s16, q6s16); + q15s16 = vsubq_s16(q8s16, q7s16); + + d4u64 = vreinterpret_u64_s16(vget_low_s16(q2s16)); + d5u64 = vreinterpret_u64_s16(vget_high_s16(q2s16)); + d18u64 = vreinterpret_u64_s16(vget_low_s16(q9s16)); + d19u64 = vreinterpret_u64_s16(vget_high_s16(q9s16)); + d20u64 = vreinterpret_u64_s16(vget_low_s16(q10s16)); + d21u64 = vreinterpret_u64_s16(vget_high_s16(q10s16)); + d22u64 = vreinterpret_u64_s16(vget_low_s16(q11s16)); + d23u64 = vreinterpret_u64_s16(vget_high_s16(q11s16)); + d24u64 = vreinterpret_u64_s16(vget_low_s16(q12s16)); + d25u64 = vreinterpret_u64_s16(vget_high_s16(q12s16)); + d26u64 = vreinterpret_u64_s16(vget_low_s16(q13s16)); + d27u64 = vreinterpret_u64_s16(vget_high_s16(q13s16)); + d28u64 = vreinterpret_u64_s16(vget_low_s16(q14s16)); + d29u64 = vreinterpret_u64_s16(vget_high_s16(q14s16)); + d30u64 = vreinterpret_u64_s16(vget_low_s16(q15s16)); + d31u64 = vreinterpret_u64_s16(vget_high_s16(q15s16)); + + // store the data + output_stride >>= 1; // output_stride / 2, out is int16_t + vst1_u64((uint64_t *)out, d4u64); + out += output_stride; + vst1_u64((uint64_t *)out, d5u64); + out += output_stride; + vst1_u64((uint64_t *)out, d18u64); + out += output_stride; + vst1_u64((uint64_t *)out, d19u64); + out += output_stride; + vst1_u64((uint64_t *)out, d20u64); + out += output_stride; + vst1_u64((uint64_t *)out, d21u64); + out += output_stride; + vst1_u64((uint64_t *)out, d22u64); + out += output_stride; + vst1_u64((uint64_t *)out, d23u64); + out += output_stride; + vst1_u64((uint64_t *)out, d24u64); + out += output_stride; + vst1_u64((uint64_t *)out, d25u64); + out += output_stride; + vst1_u64((uint64_t *)out, d26u64); + out += output_stride; + vst1_u64((uint64_t *)out, d27u64); + out += output_stride; + vst1_u64((uint64_t *)out, d28u64); + out += output_stride; + vst1_u64((uint64_t *)out, d29u64); + out += output_stride; + vst1_u64((uint64_t *)out, d30u64); + out += output_stride; + vst1_u64((uint64_t *)out, d31u64); + return; +} + +void vpx_idct16x16_10_add_neon_pass2( + int16_t *src, + int16_t *out, + int16_t *pass1Output, + int16_t skip_adding, + uint8_t *dest, + int dest_stride) { + int16x4_t d0s16, d1s16, d2s16, d3s16, d4s16, d5s16, d6s16, d7s16; + int16x4_t d8s16, d9s16, d10s16, d11s16, d12s16, d13s16, d14s16, d15s16; + int16x4_t d20s16, d21s16, d22s16, d23s16; + int16x4_t d24s16, d25s16, d26s16, d27s16, d30s16, d31s16; + uint64x1_t d4u64, d5u64, d6u64, d7u64, d8u64, d9u64, d10u64, d11u64; + uint64x1_t d16u64, d17u64, d18u64, d19u64; + uint64x1_t d24u64, d25u64, d26u64, d27u64, d28u64, d29u64, d30u64, d31u64; + int16x8_t q0s16, q1s16, q2s16, q3s16, q4s16, q5s16, q6s16, q7s16; + int16x8_t q8s16, q9s16, q10s16, q11s16, q12s16, q13s16, q14s16, q15s16; + int32x4_t q0s32, q1s32, q2s32, q3s32, q4s32, q5s32, q6s32, q8s32, q9s32; + int32x4_t q10s32, q11s32, q12s32, q13s32; + int16x8x2_t q0x2s16; + (void)skip_adding; + (void)dest; + (void)dest_stride; + + q0x2s16 = vld2q_s16(src); + q8s16 = q0x2s16.val[0]; + src += 16; + q0x2s16 = vld2q_s16(src); + q9s16 = q0x2s16.val[0]; + src += 16; + q0x2s16 = vld2q_s16(src); + q10s16 = q0x2s16.val[0]; + src += 16; + q0x2s16 = vld2q_s16(src); + q11s16 = q0x2s16.val[0]; + src += 16; + q0x2s16 = vld2q_s16(src); + q12s16 = q0x2s16.val[0]; + src += 16; + q0x2s16 = vld2q_s16(src); + q13s16 = q0x2s16.val[0]; + src += 16; + q0x2s16 = vld2q_s16(src); + q14s16 = q0x2s16.val[0]; + src += 16; + q0x2s16 = vld2q_s16(src); + q15s16 = q0x2s16.val[0]; + + TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, + &q12s16, &q13s16, &q14s16, &q15s16); + + // stage 3 + q6s16 = vdupq_n_s16(cospi_30_64 * 2); + q0s16 = vqrdmulhq_s16(q8s16, q6s16); + q6s16 = vdupq_n_s16(cospi_2_64 * 2); + q7s16 = vqrdmulhq_s16(q8s16, q6s16); + + q15s16 = vdupq_n_s16(-cospi_26_64 * 2); + q14s16 = vdupq_n_s16(cospi_6_64 * 2); + q3s16 = vqrdmulhq_s16(q9s16, q15s16); + q4s16 = vqrdmulhq_s16(q9s16, q14s16); + + // stage 4 + d0s16 = vget_low_s16(q0s16); + d1s16 = vget_high_s16(q0s16); + d6s16 = vget_low_s16(q3s16); + d7s16 = vget_high_s16(q3s16); + d8s16 = vget_low_s16(q4s16); + d9s16 = vget_high_s16(q4s16); + d14s16 = vget_low_s16(q7s16); + d15s16 = vget_high_s16(q7s16); + + d30s16 = vdup_n_s16(cospi_8_64); + d31s16 = vdup_n_s16(cospi_24_64); + + q12s32 = vmull_s16(d14s16, d31s16); + q5s32 = vmull_s16(d15s16, d31s16); + q2s32 = vmull_s16(d0s16, d31s16); + q11s32 = vmull_s16(d1s16, d31s16); + + q12s32 = vmlsl_s16(q12s32, d0s16, d30s16); + q5s32 = vmlsl_s16(q5s32, d1s16, d30s16); + q2s32 = vmlal_s16(q2s32, d14s16, d30s16); + q11s32 = vmlal_s16(q11s32, d15s16, d30s16); + + d2s16 = vqrshrn_n_s32(q12s32, 14); + d3s16 = vqrshrn_n_s32(q5s32, 14); + d12s16 = vqrshrn_n_s32(q2s32, 14); + d13s16 = vqrshrn_n_s32(q11s32, 14); + q1s16 = vcombine_s16(d2s16, d3s16); + q6s16 = vcombine_s16(d12s16, d13s16); + + d30s16 = vdup_n_s16(-cospi_8_64); + q10s32 = vmull_s16(d8s16, d30s16); + q13s32 = vmull_s16(d9s16, d30s16); + q8s32 = vmull_s16(d6s16, d30s16); + q9s32 = vmull_s16(d7s16, d30s16); + + q10s32 = vmlsl_s16(q10s32, d6s16, d31s16); + q13s32 = vmlsl_s16(q13s32, d7s16, d31s16); + q8s32 = vmlal_s16(q8s32, d8s16, d31s16); + q9s32 = vmlal_s16(q9s32, d9s16, d31s16); + + d4s16 = vqrshrn_n_s32(q10s32, 14); + d5s16 = vqrshrn_n_s32(q13s32, 14); + d10s16 = vqrshrn_n_s32(q8s32, 14); + d11s16 = vqrshrn_n_s32(q9s32, 14); + q2s16 = vcombine_s16(d4s16, d5s16); + q5s16 = vcombine_s16(d10s16, d11s16); + + // stage 5 + q8s16 = vaddq_s16(q0s16, q3s16); + q9s16 = vaddq_s16(q1s16, q2s16); + q10s16 = vsubq_s16(q1s16, q2s16); + q11s16 = vsubq_s16(q0s16, q3s16); + q12s16 = vsubq_s16(q7s16, q4s16); + q13s16 = vsubq_s16(q6s16, q5s16); + q14s16 = vaddq_s16(q6s16, q5s16); + q15s16 = vaddq_s16(q7s16, q4s16); + + // stage 6 + d20s16 = vget_low_s16(q10s16); + d21s16 = vget_high_s16(q10s16); + d22s16 = vget_low_s16(q11s16); + d23s16 = vget_high_s16(q11s16); + d24s16 = vget_low_s16(q12s16); + d25s16 = vget_high_s16(q12s16); + d26s16 = vget_low_s16(q13s16); + d27s16 = vget_high_s16(q13s16); + + d14s16 = vdup_n_s16(cospi_16_64); + q3s32 = vmull_s16(d26s16, d14s16); + q4s32 = vmull_s16(d27s16, d14s16); + q0s32 = vmull_s16(d20s16, d14s16); + q1s32 = vmull_s16(d21s16, d14s16); + + q5s32 = vsubq_s32(q3s32, q0s32); + q6s32 = vsubq_s32(q4s32, q1s32); + q0s32 = vaddq_s32(q3s32, q0s32); + q4s32 = vaddq_s32(q4s32, q1s32); + + d4s16 = vqrshrn_n_s32(q5s32, 14); + d5s16 = vqrshrn_n_s32(q6s32, 14); + d10s16 = vqrshrn_n_s32(q0s32, 14); + d11s16 = vqrshrn_n_s32(q4s32, 14); + q2s16 = vcombine_s16(d4s16, d5s16); + q5s16 = vcombine_s16(d10s16, d11s16); + + q0s32 = vmull_s16(d22s16, d14s16); + q1s32 = vmull_s16(d23s16, d14s16); + q13s32 = vmull_s16(d24s16, d14s16); + q6s32 = vmull_s16(d25s16, d14s16); + + q10s32 = vsubq_s32(q13s32, q0s32); + q4s32 = vsubq_s32(q6s32, q1s32); + q13s32 = vaddq_s32(q13s32, q0s32); + q6s32 = vaddq_s32(q6s32, q1s32); + + d6s16 = vqrshrn_n_s32(q10s32, 14); + d7s16 = vqrshrn_n_s32(q4s32, 14); + d8s16 = vqrshrn_n_s32(q13s32, 14); + d9s16 = vqrshrn_n_s32(q6s32, 14); + q3s16 = vcombine_s16(d6s16, d7s16); + q4s16 = vcombine_s16(d8s16, d9s16); + + // stage 7 + q0s16 = vld1q_s16(pass1Output); + pass1Output += 8; + q1s16 = vld1q_s16(pass1Output); + pass1Output += 8; + q12s16 = vaddq_s16(q0s16, q15s16); + q13s16 = vaddq_s16(q1s16, q14s16); + d24u64 = vreinterpret_u64_s16(vget_low_s16(q12s16)); + d25u64 = vreinterpret_u64_s16(vget_high_s16(q12s16)); + d26u64 = vreinterpret_u64_s16(vget_low_s16(q13s16)); + d27u64 = vreinterpret_u64_s16(vget_high_s16(q13s16)); + vst1_u64((uint64_t *)out, d24u64); + out += 4; + vst1_u64((uint64_t *)out, d25u64); + out += 12; + vst1_u64((uint64_t *)out, d26u64); + out += 4; + vst1_u64((uint64_t *)out, d27u64); + out += 12; + q14s16 = vsubq_s16(q1s16, q14s16); + q15s16 = vsubq_s16(q0s16, q15s16); + + q10s16 = vld1q_s16(pass1Output); + pass1Output += 8; + q11s16 = vld1q_s16(pass1Output); + pass1Output += 8; + q12s16 = vaddq_s16(q10s16, q5s16); + q13s16 = vaddq_s16(q11s16, q4s16); + d24u64 = vreinterpret_u64_s16(vget_low_s16(q12s16)); + d25u64 = vreinterpret_u64_s16(vget_high_s16(q12s16)); + d26u64 = vreinterpret_u64_s16(vget_low_s16(q13s16)); + d27u64 = vreinterpret_u64_s16(vget_high_s16(q13s16)); + vst1_u64((uint64_t *)out, d24u64); + out += 4; + vst1_u64((uint64_t *)out, d25u64); + out += 12; + vst1_u64((uint64_t *)out, d26u64); + out += 4; + vst1_u64((uint64_t *)out, d27u64); + out += 12; + q4s16 = vsubq_s16(q11s16, q4s16); + q5s16 = vsubq_s16(q10s16, q5s16); + + q0s16 = vld1q_s16(pass1Output); + pass1Output += 8; + q1s16 = vld1q_s16(pass1Output); + pass1Output += 8; + q12s16 = vaddq_s16(q0s16, q3s16); + q13s16 = vaddq_s16(q1s16, q2s16); + d24u64 = vreinterpret_u64_s16(vget_low_s16(q12s16)); + d25u64 = vreinterpret_u64_s16(vget_high_s16(q12s16)); + d26u64 = vreinterpret_u64_s16(vget_low_s16(q13s16)); + d27u64 = vreinterpret_u64_s16(vget_high_s16(q13s16)); + vst1_u64((uint64_t *)out, d24u64); + out += 4; + vst1_u64((uint64_t *)out, d25u64); + out += 12; + vst1_u64((uint64_t *)out, d26u64); + out += 4; + vst1_u64((uint64_t *)out, d27u64); + out += 12; + q2s16 = vsubq_s16(q1s16, q2s16); + q3s16 = vsubq_s16(q0s16, q3s16); + + q10s16 = vld1q_s16(pass1Output); + pass1Output += 8; + q11s16 = vld1q_s16(pass1Output); + q12s16 = vaddq_s16(q10s16, q9s16); + q13s16 = vaddq_s16(q11s16, q8s16); + d24u64 = vreinterpret_u64_s16(vget_low_s16(q12s16)); + d25u64 = vreinterpret_u64_s16(vget_high_s16(q12s16)); + d26u64 = vreinterpret_u64_s16(vget_low_s16(q13s16)); + d27u64 = vreinterpret_u64_s16(vget_high_s16(q13s16)); + vst1_u64((uint64_t *)out, d24u64); + out += 4; + vst1_u64((uint64_t *)out, d25u64); + out += 12; + vst1_u64((uint64_t *)out, d26u64); + out += 4; + vst1_u64((uint64_t *)out, d27u64); + out += 12; + q8s16 = vsubq_s16(q11s16, q8s16); + q9s16 = vsubq_s16(q10s16, q9s16); + + d4u64 = vreinterpret_u64_s16(vget_low_s16(q2s16)); + d5u64 = vreinterpret_u64_s16(vget_high_s16(q2s16)); + d6u64 = vreinterpret_u64_s16(vget_low_s16(q3s16)); + d7u64 = vreinterpret_u64_s16(vget_high_s16(q3s16)); + d8u64 = vreinterpret_u64_s16(vget_low_s16(q4s16)); + d9u64 = vreinterpret_u64_s16(vget_high_s16(q4s16)); + d10u64 = vreinterpret_u64_s16(vget_low_s16(q5s16)); + d11u64 = vreinterpret_u64_s16(vget_high_s16(q5s16)); + d16u64 = vreinterpret_u64_s16(vget_low_s16(q8s16)); + d17u64 = vreinterpret_u64_s16(vget_high_s16(q8s16)); + d18u64 = vreinterpret_u64_s16(vget_low_s16(q9s16)); + d19u64 = vreinterpret_u64_s16(vget_high_s16(q9s16)); + d28u64 = vreinterpret_u64_s16(vget_low_s16(q14s16)); + d29u64 = vreinterpret_u64_s16(vget_high_s16(q14s16)); + d30u64 = vreinterpret_u64_s16(vget_low_s16(q15s16)); + d31u64 = vreinterpret_u64_s16(vget_high_s16(q15s16)); + + vst1_u64((uint64_t *)out, d16u64); + out += 4; + vst1_u64((uint64_t *)out, d17u64); + out += 12; + vst1_u64((uint64_t *)out, d18u64); + out += 4; + vst1_u64((uint64_t *)out, d19u64); + out += 12; + vst1_u64((uint64_t *)out, d4u64); + out += 4; + vst1_u64((uint64_t *)out, d5u64); + out += 12; + vst1_u64((uint64_t *)out, d6u64); + out += 4; + vst1_u64((uint64_t *)out, d7u64); + out += 12; + vst1_u64((uint64_t *)out, d8u64); + out += 4; + vst1_u64((uint64_t *)out, d9u64); + out += 12; + vst1_u64((uint64_t *)out, d10u64); + out += 4; + vst1_u64((uint64_t *)out, d11u64); + out += 12; + vst1_u64((uint64_t *)out, d28u64); + out += 4; + vst1_u64((uint64_t *)out, d29u64); + out += 12; + vst1_u64((uint64_t *)out, d30u64); + out += 4; + vst1_u64((uint64_t *)out, d31u64); + return; +} diff --git a/thirdparty/libvpx/vpx_dsp/arm/idct16x16_neon.c b/thirdparty/libvpx/vpx_dsp/arm/idct16x16_neon.c new file mode 100644 index 0000000000..352979aa16 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/arm/idct16x16_neon.c @@ -0,0 +1,185 @@ +/* + * Copyright (c) 2013 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "vpx_dsp/vpx_dsp_common.h" + +void vpx_idct16x16_256_add_neon_pass1(const int16_t *input, + int16_t *output, + int output_stride); +void vpx_idct16x16_256_add_neon_pass2(const int16_t *src, + int16_t *output, + int16_t *pass1Output, + int16_t skip_adding, + uint8_t *dest, + int dest_stride); +void vpx_idct16x16_10_add_neon_pass1(const int16_t *input, + int16_t *output, + int output_stride); +void vpx_idct16x16_10_add_neon_pass2(const int16_t *src, + int16_t *output, + int16_t *pass1Output, + int16_t skip_adding, + uint8_t *dest, + int dest_stride); + +#if HAVE_NEON_ASM +/* For ARM NEON, d8-d15 are callee-saved registers, and need to be saved. */ +extern void vpx_push_neon(int64_t *store); +extern void vpx_pop_neon(int64_t *store); +#endif // HAVE_NEON_ASM + +void vpx_idct16x16_256_add_neon(const int16_t *input, + uint8_t *dest, int dest_stride) { +#if HAVE_NEON_ASM + int64_t store_reg[8]; +#endif + int16_t pass1_output[16*16] = {0}; + int16_t row_idct_output[16*16] = {0}; + +#if HAVE_NEON_ASM + // save d8-d15 register values. + vpx_push_neon(store_reg); +#endif + + /* Parallel idct on the upper 8 rows */ + // First pass processes even elements 0, 2, 4, 6, 8, 10, 12, 14 and save the + // stage 6 result in pass1_output. + vpx_idct16x16_256_add_neon_pass1(input, pass1_output, 8); + + // Second pass processes odd elements 1, 3, 5, 7, 9, 11, 13, 15 and combines + // with result in pass1(pass1_output) to calculate final result in stage 7 + // which will be saved into row_idct_output. + vpx_idct16x16_256_add_neon_pass2(input+1, + row_idct_output, + pass1_output, + 0, + dest, + dest_stride); + + /* Parallel idct on the lower 8 rows */ + // First pass processes even elements 0, 2, 4, 6, 8, 10, 12, 14 and save the + // stage 6 result in pass1_output. + vpx_idct16x16_256_add_neon_pass1(input+8*16, pass1_output, 8); + + // Second pass processes odd elements 1, 3, 5, 7, 9, 11, 13, 15 and combines + // with result in pass1(pass1_output) to calculate final result in stage 7 + // which will be saved into row_idct_output. + vpx_idct16x16_256_add_neon_pass2(input+8*16+1, + row_idct_output+8, + pass1_output, + 0, + dest, + dest_stride); + + /* Parallel idct on the left 8 columns */ + // First pass processes even elements 0, 2, 4, 6, 8, 10, 12, 14 and save the + // stage 6 result in pass1_output. + vpx_idct16x16_256_add_neon_pass1(row_idct_output, pass1_output, 8); + + // Second pass processes odd elements 1, 3, 5, 7, 9, 11, 13, 15 and combines + // with result in pass1(pass1_output) to calculate final result in stage 7. + // Then add the result to the destination data. + vpx_idct16x16_256_add_neon_pass2(row_idct_output+1, + row_idct_output, + pass1_output, + 1, + dest, + dest_stride); + + /* Parallel idct on the right 8 columns */ + // First pass processes even elements 0, 2, 4, 6, 8, 10, 12, 14 and save the + // stage 6 result in pass1_output. + vpx_idct16x16_256_add_neon_pass1(row_idct_output+8*16, pass1_output, 8); + + // Second pass processes odd elements 1, 3, 5, 7, 9, 11, 13, 15 and combines + // with result in pass1(pass1_output) to calculate final result in stage 7. + // Then add the result to the destination data. + vpx_idct16x16_256_add_neon_pass2(row_idct_output+8*16+1, + row_idct_output+8, + pass1_output, + 1, + dest+8, + dest_stride); + +#if HAVE_NEON_ASM + // restore d8-d15 register values. + vpx_pop_neon(store_reg); +#endif + + return; +} + +void vpx_idct16x16_10_add_neon(const int16_t *input, + uint8_t *dest, int dest_stride) { +#if HAVE_NEON_ASM + int64_t store_reg[8]; +#endif + int16_t pass1_output[16*16] = {0}; + int16_t row_idct_output[16*16] = {0}; + +#if HAVE_NEON_ASM + // save d8-d15 register values. + vpx_push_neon(store_reg); +#endif + + /* Parallel idct on the upper 8 rows */ + // First pass processes even elements 0, 2, 4, 6, 8, 10, 12, 14 and save the + // stage 6 result in pass1_output. + vpx_idct16x16_10_add_neon_pass1(input, pass1_output, 8); + + // Second pass processes odd elements 1, 3, 5, 7, 9, 11, 13, 15 and combines + // with result in pass1(pass1_output) to calculate final result in stage 7 + // which will be saved into row_idct_output. + vpx_idct16x16_10_add_neon_pass2(input+1, + row_idct_output, + pass1_output, + 0, + dest, + dest_stride); + + /* Skip Parallel idct on the lower 8 rows as they are all 0s */ + + /* Parallel idct on the left 8 columns */ + // First pass processes even elements 0, 2, 4, 6, 8, 10, 12, 14 and save the + // stage 6 result in pass1_output. + vpx_idct16x16_256_add_neon_pass1(row_idct_output, pass1_output, 8); + + // Second pass processes odd elements 1, 3, 5, 7, 9, 11, 13, 15 and combines + // with result in pass1(pass1_output) to calculate final result in stage 7. + // Then add the result to the destination data. + vpx_idct16x16_256_add_neon_pass2(row_idct_output+1, + row_idct_output, + pass1_output, + 1, + dest, + dest_stride); + + /* Parallel idct on the right 8 columns */ + // First pass processes even elements 0, 2, 4, 6, 8, 10, 12, 14 and save the + // stage 6 result in pass1_output. + vpx_idct16x16_256_add_neon_pass1(row_idct_output+8*16, pass1_output, 8); + + // Second pass processes odd elements 1, 3, 5, 7, 9, 11, 13, 15 and combines + // with result in pass1(pass1_output) to calculate final result in stage 7. + // Then add the result to the destination data. + vpx_idct16x16_256_add_neon_pass2(row_idct_output+8*16+1, + row_idct_output+8, + pass1_output, + 1, + dest+8, + dest_stride); + +#if HAVE_NEON_ASM + // restore d8-d15 register values. + vpx_pop_neon(store_reg); +#endif + + return; +} diff --git a/thirdparty/libvpx/vpx_dsp/arm/idct32x32_1_add_neon.c b/thirdparty/libvpx/vpx_dsp/arm/idct32x32_1_add_neon.c new file mode 100644 index 0000000000..c25c0c4a5c --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/arm/idct32x32_1_add_neon.c @@ -0,0 +1,165 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +#include "./vpx_config.h" + +#include "vpx_dsp/inv_txfm.h" +#include "vpx_ports/mem.h" + +static INLINE void LD_16x8( + uint8_t *d, + int d_stride, + uint8x16_t *q8u8, + uint8x16_t *q9u8, + uint8x16_t *q10u8, + uint8x16_t *q11u8, + uint8x16_t *q12u8, + uint8x16_t *q13u8, + uint8x16_t *q14u8, + uint8x16_t *q15u8) { + *q8u8 = vld1q_u8(d); + d += d_stride; + *q9u8 = vld1q_u8(d); + d += d_stride; + *q10u8 = vld1q_u8(d); + d += d_stride; + *q11u8 = vld1q_u8(d); + d += d_stride; + *q12u8 = vld1q_u8(d); + d += d_stride; + *q13u8 = vld1q_u8(d); + d += d_stride; + *q14u8 = vld1q_u8(d); + d += d_stride; + *q15u8 = vld1q_u8(d); + return; +} + +static INLINE void ADD_DIFF_16x8( + uint8x16_t qdiffu8, + uint8x16_t *q8u8, + uint8x16_t *q9u8, + uint8x16_t *q10u8, + uint8x16_t *q11u8, + uint8x16_t *q12u8, + uint8x16_t *q13u8, + uint8x16_t *q14u8, + uint8x16_t *q15u8) { + *q8u8 = vqaddq_u8(*q8u8, qdiffu8); + *q9u8 = vqaddq_u8(*q9u8, qdiffu8); + *q10u8 = vqaddq_u8(*q10u8, qdiffu8); + *q11u8 = vqaddq_u8(*q11u8, qdiffu8); + *q12u8 = vqaddq_u8(*q12u8, qdiffu8); + *q13u8 = vqaddq_u8(*q13u8, qdiffu8); + *q14u8 = vqaddq_u8(*q14u8, qdiffu8); + *q15u8 = vqaddq_u8(*q15u8, qdiffu8); + return; +} + +static INLINE void SUB_DIFF_16x8( + uint8x16_t qdiffu8, + uint8x16_t *q8u8, + uint8x16_t *q9u8, + uint8x16_t *q10u8, + uint8x16_t *q11u8, + uint8x16_t *q12u8, + uint8x16_t *q13u8, + uint8x16_t *q14u8, + uint8x16_t *q15u8) { + *q8u8 = vqsubq_u8(*q8u8, qdiffu8); + *q9u8 = vqsubq_u8(*q9u8, qdiffu8); + *q10u8 = vqsubq_u8(*q10u8, qdiffu8); + *q11u8 = vqsubq_u8(*q11u8, qdiffu8); + *q12u8 = vqsubq_u8(*q12u8, qdiffu8); + *q13u8 = vqsubq_u8(*q13u8, qdiffu8); + *q14u8 = vqsubq_u8(*q14u8, qdiffu8); + *q15u8 = vqsubq_u8(*q15u8, qdiffu8); + return; +} + +static INLINE void ST_16x8( + uint8_t *d, + int d_stride, + uint8x16_t *q8u8, + uint8x16_t *q9u8, + uint8x16_t *q10u8, + uint8x16_t *q11u8, + uint8x16_t *q12u8, + uint8x16_t *q13u8, + uint8x16_t *q14u8, + uint8x16_t *q15u8) { + vst1q_u8(d, *q8u8); + d += d_stride; + vst1q_u8(d, *q9u8); + d += d_stride; + vst1q_u8(d, *q10u8); + d += d_stride; + vst1q_u8(d, *q11u8); + d += d_stride; + vst1q_u8(d, *q12u8); + d += d_stride; + vst1q_u8(d, *q13u8); + d += d_stride; + vst1q_u8(d, *q14u8); + d += d_stride; + vst1q_u8(d, *q15u8); + return; +} + +void vpx_idct32x32_1_add_neon( + int16_t *input, + uint8_t *dest, + int dest_stride) { + uint8x16_t q0u8, q8u8, q9u8, q10u8, q11u8, q12u8, q13u8, q14u8, q15u8; + int i, j, dest_stride8; + uint8_t *d; + int16_t a1, cospi_16_64 = 11585; + int16_t out = dct_const_round_shift(input[0] * cospi_16_64); + + out = dct_const_round_shift(out * cospi_16_64); + a1 = ROUND_POWER_OF_TWO(out, 6); + + dest_stride8 = dest_stride * 8; + if (a1 >= 0) { // diff_positive_32_32 + a1 = a1 < 0 ? 0 : a1 > 255 ? 255 : a1; + q0u8 = vdupq_n_u8(a1); + for (i = 0; i < 2; i++, dest += 16) { // diff_positive_32_32_loop + d = dest; + for (j = 0; j < 4; j++) { + LD_16x8(d, dest_stride, &q8u8, &q9u8, &q10u8, &q11u8, + &q12u8, &q13u8, &q14u8, &q15u8); + ADD_DIFF_16x8(q0u8, &q8u8, &q9u8, &q10u8, &q11u8, + &q12u8, &q13u8, &q14u8, &q15u8); + ST_16x8(d, dest_stride, &q8u8, &q9u8, &q10u8, &q11u8, + &q12u8, &q13u8, &q14u8, &q15u8); + d += dest_stride8; + } + } + } else { // diff_negative_32_32 + a1 = -a1; + a1 = a1 < 0 ? 0 : a1 > 255 ? 255 : a1; + q0u8 = vdupq_n_u8(a1); + for (i = 0; i < 2; i++, dest += 16) { // diff_negative_32_32_loop + d = dest; + for (j = 0; j < 4; j++) { + LD_16x8(d, dest_stride, &q8u8, &q9u8, &q10u8, &q11u8, + &q12u8, &q13u8, &q14u8, &q15u8); + SUB_DIFF_16x8(q0u8, &q8u8, &q9u8, &q10u8, &q11u8, + &q12u8, &q13u8, &q14u8, &q15u8); + ST_16x8(d, dest_stride, &q8u8, &q9u8, &q10u8, &q11u8, + &q12u8, &q13u8, &q14u8, &q15u8); + d += dest_stride8; + } + } + } + return; +} diff --git a/thirdparty/libvpx/vpx_dsp/arm/idct32x32_add_neon.c b/thirdparty/libvpx/vpx_dsp/arm/idct32x32_add_neon.c new file mode 100644 index 0000000000..025437eb96 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/arm/idct32x32_add_neon.c @@ -0,0 +1,719 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +#include "./vpx_config.h" +#include "vpx_dsp/txfm_common.h" + +#define LOAD_FROM_TRANSPOSED(prev, first, second) \ + q14s16 = vld1q_s16(trans_buf + first * 8); \ + q13s16 = vld1q_s16(trans_buf + second * 8); + +#define LOAD_FROM_OUTPUT(prev, first, second, qA, qB) \ + qA = vld1q_s16(out + first * 32); \ + qB = vld1q_s16(out + second * 32); + +#define STORE_IN_OUTPUT(prev, first, second, qA, qB) \ + vst1q_s16(out + first * 32, qA); \ + vst1q_s16(out + second * 32, qB); + +#define STORE_COMBINE_CENTER_RESULTS(r10, r9) \ + __STORE_COMBINE_CENTER_RESULTS(r10, r9, stride, \ + q6s16, q7s16, q8s16, q9s16); +static INLINE void __STORE_COMBINE_CENTER_RESULTS( + uint8_t *p1, + uint8_t *p2, + int stride, + int16x8_t q6s16, + int16x8_t q7s16, + int16x8_t q8s16, + int16x8_t q9s16) { + int16x4_t d8s16, d9s16, d10s16, d11s16; + + d8s16 = vld1_s16((int16_t *)p1); + p1 += stride; + d11s16 = vld1_s16((int16_t *)p2); + p2 -= stride; + d9s16 = vld1_s16((int16_t *)p1); + d10s16 = vld1_s16((int16_t *)p2); + + q7s16 = vrshrq_n_s16(q7s16, 6); + q8s16 = vrshrq_n_s16(q8s16, 6); + q9s16 = vrshrq_n_s16(q9s16, 6); + q6s16 = vrshrq_n_s16(q6s16, 6); + + q7s16 = vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(q7s16), + vreinterpret_u8_s16(d9s16))); + q8s16 = vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(q8s16), + vreinterpret_u8_s16(d10s16))); + q9s16 = vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(q9s16), + vreinterpret_u8_s16(d11s16))); + q6s16 = vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(q6s16), + vreinterpret_u8_s16(d8s16))); + + d9s16 = vreinterpret_s16_u8(vqmovun_s16(q7s16)); + d10s16 = vreinterpret_s16_u8(vqmovun_s16(q8s16)); + d11s16 = vreinterpret_s16_u8(vqmovun_s16(q9s16)); + d8s16 = vreinterpret_s16_u8(vqmovun_s16(q6s16)); + + vst1_s16((int16_t *)p1, d9s16); + p1 -= stride; + vst1_s16((int16_t *)p2, d10s16); + p2 += stride; + vst1_s16((int16_t *)p1, d8s16); + vst1_s16((int16_t *)p2, d11s16); + return; +} + +#define STORE_COMBINE_EXTREME_RESULTS(r7, r6); \ + __STORE_COMBINE_EXTREME_RESULTS(r7, r6, stride, \ + q4s16, q5s16, q6s16, q7s16); +static INLINE void __STORE_COMBINE_EXTREME_RESULTS( + uint8_t *p1, + uint8_t *p2, + int stride, + int16x8_t q4s16, + int16x8_t q5s16, + int16x8_t q6s16, + int16x8_t q7s16) { + int16x4_t d4s16, d5s16, d6s16, d7s16; + + d4s16 = vld1_s16((int16_t *)p1); + p1 += stride; + d7s16 = vld1_s16((int16_t *)p2); + p2 -= stride; + d5s16 = vld1_s16((int16_t *)p1); + d6s16 = vld1_s16((int16_t *)p2); + + q5s16 = vrshrq_n_s16(q5s16, 6); + q6s16 = vrshrq_n_s16(q6s16, 6); + q7s16 = vrshrq_n_s16(q7s16, 6); + q4s16 = vrshrq_n_s16(q4s16, 6); + + q5s16 = vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(q5s16), + vreinterpret_u8_s16(d5s16))); + q6s16 = vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(q6s16), + vreinterpret_u8_s16(d6s16))); + q7s16 = vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(q7s16), + vreinterpret_u8_s16(d7s16))); + q4s16 = vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(q4s16), + vreinterpret_u8_s16(d4s16))); + + d5s16 = vreinterpret_s16_u8(vqmovun_s16(q5s16)); + d6s16 = vreinterpret_s16_u8(vqmovun_s16(q6s16)); + d7s16 = vreinterpret_s16_u8(vqmovun_s16(q7s16)); + d4s16 = vreinterpret_s16_u8(vqmovun_s16(q4s16)); + + vst1_s16((int16_t *)p1, d5s16); + p1 -= stride; + vst1_s16((int16_t *)p2, d6s16); + p2 += stride; + vst1_s16((int16_t *)p2, d7s16); + vst1_s16((int16_t *)p1, d4s16); + return; +} + +#define DO_BUTTERFLY_STD(const_1, const_2, qA, qB) \ + DO_BUTTERFLY(q14s16, q13s16, const_1, const_2, qA, qB); +static INLINE void DO_BUTTERFLY( + int16x8_t q14s16, + int16x8_t q13s16, + int16_t first_const, + int16_t second_const, + int16x8_t *qAs16, + int16x8_t *qBs16) { + int16x4_t d30s16, d31s16; + int32x4_t q8s32, q9s32, q10s32, q11s32, q12s32, q15s32; + int16x4_t dCs16, dDs16, dAs16, dBs16; + + dCs16 = vget_low_s16(q14s16); + dDs16 = vget_high_s16(q14s16); + dAs16 = vget_low_s16(q13s16); + dBs16 = vget_high_s16(q13s16); + + d30s16 = vdup_n_s16(first_const); + d31s16 = vdup_n_s16(second_const); + + q8s32 = vmull_s16(dCs16, d30s16); + q10s32 = vmull_s16(dAs16, d31s16); + q9s32 = vmull_s16(dDs16, d30s16); + q11s32 = vmull_s16(dBs16, d31s16); + q12s32 = vmull_s16(dCs16, d31s16); + + q8s32 = vsubq_s32(q8s32, q10s32); + q9s32 = vsubq_s32(q9s32, q11s32); + + q10s32 = vmull_s16(dDs16, d31s16); + q11s32 = vmull_s16(dAs16, d30s16); + q15s32 = vmull_s16(dBs16, d30s16); + + q11s32 = vaddq_s32(q12s32, q11s32); + q10s32 = vaddq_s32(q10s32, q15s32); + + *qAs16 = vcombine_s16(vqrshrn_n_s32(q8s32, 14), + vqrshrn_n_s32(q9s32, 14)); + *qBs16 = vcombine_s16(vqrshrn_n_s32(q11s32, 14), + vqrshrn_n_s32(q10s32, 14)); + return; +} + +static INLINE void idct32_transpose_pair( + int16_t *input, + int16_t *t_buf) { + int16_t *in; + int i; + const int stride = 32; + int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16; + int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16; + int16x8_t q8s16, q9s16, q10s16, q11s16, q12s16, q13s16, q14s16, q15s16; + int32x4x2_t q0x2s32, q1x2s32, q2x2s32, q3x2s32; + int16x8x2_t q0x2s16, q1x2s16, q2x2s16, q3x2s16; + + for (i = 0; i < 4; i++, input += 8) { + in = input; + q8s16 = vld1q_s16(in); + in += stride; + q9s16 = vld1q_s16(in); + in += stride; + q10s16 = vld1q_s16(in); + in += stride; + q11s16 = vld1q_s16(in); + in += stride; + q12s16 = vld1q_s16(in); + in += stride; + q13s16 = vld1q_s16(in); + in += stride; + q14s16 = vld1q_s16(in); + in += stride; + q15s16 = vld1q_s16(in); + + d16s16 = vget_low_s16(q8s16); + d17s16 = vget_high_s16(q8s16); + d18s16 = vget_low_s16(q9s16); + d19s16 = vget_high_s16(q9s16); + d20s16 = vget_low_s16(q10s16); + d21s16 = vget_high_s16(q10s16); + d22s16 = vget_low_s16(q11s16); + d23s16 = vget_high_s16(q11s16); + d24s16 = vget_low_s16(q12s16); + d25s16 = vget_high_s16(q12s16); + d26s16 = vget_low_s16(q13s16); + d27s16 = vget_high_s16(q13s16); + d28s16 = vget_low_s16(q14s16); + d29s16 = vget_high_s16(q14s16); + d30s16 = vget_low_s16(q15s16); + d31s16 = vget_high_s16(q15s16); + + q8s16 = vcombine_s16(d16s16, d24s16); // vswp d17, d24 + q9s16 = vcombine_s16(d18s16, d26s16); // vswp d19, d26 + q10s16 = vcombine_s16(d20s16, d28s16); // vswp d21, d28 + q11s16 = vcombine_s16(d22s16, d30s16); // vswp d23, d30 + q12s16 = vcombine_s16(d17s16, d25s16); + q13s16 = vcombine_s16(d19s16, d27s16); + q14s16 = vcombine_s16(d21s16, d29s16); + q15s16 = vcombine_s16(d23s16, d31s16); + + q0x2s32 = vtrnq_s32(vreinterpretq_s32_s16(q8s16), + vreinterpretq_s32_s16(q10s16)); + q1x2s32 = vtrnq_s32(vreinterpretq_s32_s16(q9s16), + vreinterpretq_s32_s16(q11s16)); + q2x2s32 = vtrnq_s32(vreinterpretq_s32_s16(q12s16), + vreinterpretq_s32_s16(q14s16)); + q3x2s32 = vtrnq_s32(vreinterpretq_s32_s16(q13s16), + vreinterpretq_s32_s16(q15s16)); + + q0x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[0]), // q8 + vreinterpretq_s16_s32(q1x2s32.val[0])); // q9 + q1x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[1]), // q10 + vreinterpretq_s16_s32(q1x2s32.val[1])); // q11 + q2x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[0]), // q12 + vreinterpretq_s16_s32(q3x2s32.val[0])); // q13 + q3x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[1]), // q14 + vreinterpretq_s16_s32(q3x2s32.val[1])); // q15 + + vst1q_s16(t_buf, q0x2s16.val[0]); + t_buf += 8; + vst1q_s16(t_buf, q0x2s16.val[1]); + t_buf += 8; + vst1q_s16(t_buf, q1x2s16.val[0]); + t_buf += 8; + vst1q_s16(t_buf, q1x2s16.val[1]); + t_buf += 8; + vst1q_s16(t_buf, q2x2s16.val[0]); + t_buf += 8; + vst1q_s16(t_buf, q2x2s16.val[1]); + t_buf += 8; + vst1q_s16(t_buf, q3x2s16.val[0]); + t_buf += 8; + vst1q_s16(t_buf, q3x2s16.val[1]); + t_buf += 8; + } + return; +} + +static INLINE void idct32_bands_end_1st_pass( + int16_t *out, + int16x8_t q2s16, + int16x8_t q3s16, + int16x8_t q6s16, + int16x8_t q7s16, + int16x8_t q8s16, + int16x8_t q9s16, + int16x8_t q10s16, + int16x8_t q11s16, + int16x8_t q12s16, + int16x8_t q13s16, + int16x8_t q14s16, + int16x8_t q15s16) { + int16x8_t q0s16, q1s16, q4s16, q5s16; + + STORE_IN_OUTPUT(17, 16, 17, q6s16, q7s16); + STORE_IN_OUTPUT(17, 14, 15, q8s16, q9s16); + + LOAD_FROM_OUTPUT(15, 30, 31, q0s16, q1s16); + q4s16 = vaddq_s16(q2s16, q1s16); + q5s16 = vaddq_s16(q3s16, q0s16); + q6s16 = vsubq_s16(q3s16, q0s16); + q7s16 = vsubq_s16(q2s16, q1s16); + STORE_IN_OUTPUT(31, 30, 31, q6s16, q7s16); + STORE_IN_OUTPUT(31, 0, 1, q4s16, q5s16); + + LOAD_FROM_OUTPUT(1, 12, 13, q0s16, q1s16); + q2s16 = vaddq_s16(q10s16, q1s16); + q3s16 = vaddq_s16(q11s16, q0s16); + q4s16 = vsubq_s16(q11s16, q0s16); + q5s16 = vsubq_s16(q10s16, q1s16); + + LOAD_FROM_OUTPUT(13, 18, 19, q0s16, q1s16); + q8s16 = vaddq_s16(q4s16, q1s16); + q9s16 = vaddq_s16(q5s16, q0s16); + q6s16 = vsubq_s16(q5s16, q0s16); + q7s16 = vsubq_s16(q4s16, q1s16); + STORE_IN_OUTPUT(19, 18, 19, q6s16, q7s16); + STORE_IN_OUTPUT(19, 12, 13, q8s16, q9s16); + + LOAD_FROM_OUTPUT(13, 28, 29, q0s16, q1s16); + q4s16 = vaddq_s16(q2s16, q1s16); + q5s16 = vaddq_s16(q3s16, q0s16); + q6s16 = vsubq_s16(q3s16, q0s16); + q7s16 = vsubq_s16(q2s16, q1s16); + STORE_IN_OUTPUT(29, 28, 29, q6s16, q7s16); + STORE_IN_OUTPUT(29, 2, 3, q4s16, q5s16); + + LOAD_FROM_OUTPUT(3, 10, 11, q0s16, q1s16); + q2s16 = vaddq_s16(q12s16, q1s16); + q3s16 = vaddq_s16(q13s16, q0s16); + q4s16 = vsubq_s16(q13s16, q0s16); + q5s16 = vsubq_s16(q12s16, q1s16); + + LOAD_FROM_OUTPUT(11, 20, 21, q0s16, q1s16); + q8s16 = vaddq_s16(q4s16, q1s16); + q9s16 = vaddq_s16(q5s16, q0s16); + q6s16 = vsubq_s16(q5s16, q0s16); + q7s16 = vsubq_s16(q4s16, q1s16); + STORE_IN_OUTPUT(21, 20, 21, q6s16, q7s16); + STORE_IN_OUTPUT(21, 10, 11, q8s16, q9s16); + + LOAD_FROM_OUTPUT(11, 26, 27, q0s16, q1s16); + q4s16 = vaddq_s16(q2s16, q1s16); + q5s16 = vaddq_s16(q3s16, q0s16); + q6s16 = vsubq_s16(q3s16, q0s16); + q7s16 = vsubq_s16(q2s16, q1s16); + STORE_IN_OUTPUT(27, 26, 27, q6s16, q7s16); + STORE_IN_OUTPUT(27, 4, 5, q4s16, q5s16); + + LOAD_FROM_OUTPUT(5, 8, 9, q0s16, q1s16); + q2s16 = vaddq_s16(q14s16, q1s16); + q3s16 = vaddq_s16(q15s16, q0s16); + q4s16 = vsubq_s16(q15s16, q0s16); + q5s16 = vsubq_s16(q14s16, q1s16); + + LOAD_FROM_OUTPUT(9, 22, 23, q0s16, q1s16); + q8s16 = vaddq_s16(q4s16, q1s16); + q9s16 = vaddq_s16(q5s16, q0s16); + q6s16 = vsubq_s16(q5s16, q0s16); + q7s16 = vsubq_s16(q4s16, q1s16); + STORE_IN_OUTPUT(23, 22, 23, q6s16, q7s16); + STORE_IN_OUTPUT(23, 8, 9, q8s16, q9s16); + + LOAD_FROM_OUTPUT(9, 24, 25, q0s16, q1s16); + q4s16 = vaddq_s16(q2s16, q1s16); + q5s16 = vaddq_s16(q3s16, q0s16); + q6s16 = vsubq_s16(q3s16, q0s16); + q7s16 = vsubq_s16(q2s16, q1s16); + STORE_IN_OUTPUT(25, 24, 25, q6s16, q7s16); + STORE_IN_OUTPUT(25, 6, 7, q4s16, q5s16); + return; +} + +static INLINE void idct32_bands_end_2nd_pass( + int16_t *out, + uint8_t *dest, + int stride, + int16x8_t q2s16, + int16x8_t q3s16, + int16x8_t q6s16, + int16x8_t q7s16, + int16x8_t q8s16, + int16x8_t q9s16, + int16x8_t q10s16, + int16x8_t q11s16, + int16x8_t q12s16, + int16x8_t q13s16, + int16x8_t q14s16, + int16x8_t q15s16) { + uint8_t *r6 = dest + 31 * stride; + uint8_t *r7 = dest/* + 0 * stride*/; + uint8_t *r9 = dest + 15 * stride; + uint8_t *r10 = dest + 16 * stride; + int str2 = stride << 1; + int16x8_t q0s16, q1s16, q4s16, q5s16; + + STORE_COMBINE_CENTER_RESULTS(r10, r9); + r10 += str2; r9 -= str2; + + LOAD_FROM_OUTPUT(17, 30, 31, q0s16, q1s16) + q4s16 = vaddq_s16(q2s16, q1s16); + q5s16 = vaddq_s16(q3s16, q0s16); + q6s16 = vsubq_s16(q3s16, q0s16); + q7s16 = vsubq_s16(q2s16, q1s16); + STORE_COMBINE_EXTREME_RESULTS(r7, r6); + r7 += str2; r6 -= str2; + + LOAD_FROM_OUTPUT(31, 12, 13, q0s16, q1s16) + q2s16 = vaddq_s16(q10s16, q1s16); + q3s16 = vaddq_s16(q11s16, q0s16); + q4s16 = vsubq_s16(q11s16, q0s16); + q5s16 = vsubq_s16(q10s16, q1s16); + + LOAD_FROM_OUTPUT(13, 18, 19, q0s16, q1s16) + q8s16 = vaddq_s16(q4s16, q1s16); + q9s16 = vaddq_s16(q5s16, q0s16); + q6s16 = vsubq_s16(q5s16, q0s16); + q7s16 = vsubq_s16(q4s16, q1s16); + STORE_COMBINE_CENTER_RESULTS(r10, r9); + r10 += str2; r9 -= str2; + + LOAD_FROM_OUTPUT(19, 28, 29, q0s16, q1s16) + q4s16 = vaddq_s16(q2s16, q1s16); + q5s16 = vaddq_s16(q3s16, q0s16); + q6s16 = vsubq_s16(q3s16, q0s16); + q7s16 = vsubq_s16(q2s16, q1s16); + STORE_COMBINE_EXTREME_RESULTS(r7, r6); + r7 += str2; r6 -= str2; + + LOAD_FROM_OUTPUT(29, 10, 11, q0s16, q1s16) + q2s16 = vaddq_s16(q12s16, q1s16); + q3s16 = vaddq_s16(q13s16, q0s16); + q4s16 = vsubq_s16(q13s16, q0s16); + q5s16 = vsubq_s16(q12s16, q1s16); + + LOAD_FROM_OUTPUT(11, 20, 21, q0s16, q1s16) + q8s16 = vaddq_s16(q4s16, q1s16); + q9s16 = vaddq_s16(q5s16, q0s16); + q6s16 = vsubq_s16(q5s16, q0s16); + q7s16 = vsubq_s16(q4s16, q1s16); + STORE_COMBINE_CENTER_RESULTS(r10, r9); + r10 += str2; r9 -= str2; + + LOAD_FROM_OUTPUT(21, 26, 27, q0s16, q1s16) + q4s16 = vaddq_s16(q2s16, q1s16); + q5s16 = vaddq_s16(q3s16, q0s16); + q6s16 = vsubq_s16(q3s16, q0s16); + q7s16 = vsubq_s16(q2s16, q1s16); + STORE_COMBINE_EXTREME_RESULTS(r7, r6); + r7 += str2; r6 -= str2; + + LOAD_FROM_OUTPUT(27, 8, 9, q0s16, q1s16) + q2s16 = vaddq_s16(q14s16, q1s16); + q3s16 = vaddq_s16(q15s16, q0s16); + q4s16 = vsubq_s16(q15s16, q0s16); + q5s16 = vsubq_s16(q14s16, q1s16); + + LOAD_FROM_OUTPUT(9, 22, 23, q0s16, q1s16) + q8s16 = vaddq_s16(q4s16, q1s16); + q9s16 = vaddq_s16(q5s16, q0s16); + q6s16 = vsubq_s16(q5s16, q0s16); + q7s16 = vsubq_s16(q4s16, q1s16); + STORE_COMBINE_CENTER_RESULTS(r10, r9); + + LOAD_FROM_OUTPUT(23, 24, 25, q0s16, q1s16) + q4s16 = vaddq_s16(q2s16, q1s16); + q5s16 = vaddq_s16(q3s16, q0s16); + q6s16 = vsubq_s16(q3s16, q0s16); + q7s16 = vsubq_s16(q2s16, q1s16); + STORE_COMBINE_EXTREME_RESULTS(r7, r6); + return; +} + +void vpx_idct32x32_1024_add_neon( + int16_t *input, + uint8_t *dest, + int stride) { + int i, idct32_pass_loop; + int16_t trans_buf[32 * 8]; + int16_t pass1[32 * 32]; + int16_t pass2[32 * 32]; + int16_t *out; + int16x8_t q0s16, q1s16, q2s16, q3s16, q4s16, q5s16, q6s16, q7s16; + int16x8_t q8s16, q9s16, q10s16, q11s16, q12s16, q13s16, q14s16, q15s16; + + for (idct32_pass_loop = 0, out = pass1; + idct32_pass_loop < 2; + idct32_pass_loop++, + input = pass1, // the input of pass2 is the result of pass1 + out = pass2) { + for (i = 0; + i < 4; i++, + input += 32 * 8, out += 8) { // idct32_bands_loop + idct32_transpose_pair(input, trans_buf); + + // ----------------------------------------- + // BLOCK A: 16-19,28-31 + // ----------------------------------------- + // generate 16,17,30,31 + // part of stage 1 + LOAD_FROM_TRANSPOSED(0, 1, 31) + DO_BUTTERFLY_STD(cospi_31_64, cospi_1_64, &q0s16, &q2s16) + LOAD_FROM_TRANSPOSED(31, 17, 15) + DO_BUTTERFLY_STD(cospi_15_64, cospi_17_64, &q1s16, &q3s16) + // part of stage 2 + q4s16 = vaddq_s16(q0s16, q1s16); + q13s16 = vsubq_s16(q0s16, q1s16); + q6s16 = vaddq_s16(q2s16, q3s16); + q14s16 = vsubq_s16(q2s16, q3s16); + // part of stage 3 + DO_BUTTERFLY_STD(cospi_28_64, cospi_4_64, &q5s16, &q7s16) + + // generate 18,19,28,29 + // part of stage 1 + LOAD_FROM_TRANSPOSED(15, 9, 23) + DO_BUTTERFLY_STD(cospi_23_64, cospi_9_64, &q0s16, &q2s16) + LOAD_FROM_TRANSPOSED(23, 25, 7) + DO_BUTTERFLY_STD(cospi_7_64, cospi_25_64, &q1s16, &q3s16) + // part of stage 2 + q13s16 = vsubq_s16(q3s16, q2s16); + q3s16 = vaddq_s16(q3s16, q2s16); + q14s16 = vsubq_s16(q1s16, q0s16); + q2s16 = vaddq_s16(q1s16, q0s16); + // part of stage 3 + DO_BUTTERFLY_STD(-cospi_4_64, -cospi_28_64, &q1s16, &q0s16) + // part of stage 4 + q8s16 = vaddq_s16(q4s16, q2s16); + q9s16 = vaddq_s16(q5s16, q0s16); + q10s16 = vaddq_s16(q7s16, q1s16); + q15s16 = vaddq_s16(q6s16, q3s16); + q13s16 = vsubq_s16(q5s16, q0s16); + q14s16 = vsubq_s16(q7s16, q1s16); + STORE_IN_OUTPUT(0, 16, 31, q8s16, q15s16) + STORE_IN_OUTPUT(31, 17, 30, q9s16, q10s16) + // part of stage 5 + DO_BUTTERFLY_STD(cospi_24_64, cospi_8_64, &q0s16, &q1s16) + STORE_IN_OUTPUT(30, 29, 18, q1s16, q0s16) + // part of stage 4 + q13s16 = vsubq_s16(q4s16, q2s16); + q14s16 = vsubq_s16(q6s16, q3s16); + // part of stage 5 + DO_BUTTERFLY_STD(cospi_24_64, cospi_8_64, &q4s16, &q6s16) + STORE_IN_OUTPUT(18, 19, 28, q4s16, q6s16) + + // ----------------------------------------- + // BLOCK B: 20-23,24-27 + // ----------------------------------------- + // generate 20,21,26,27 + // part of stage 1 + LOAD_FROM_TRANSPOSED(7, 5, 27) + DO_BUTTERFLY_STD(cospi_27_64, cospi_5_64, &q0s16, &q2s16) + LOAD_FROM_TRANSPOSED(27, 21, 11) + DO_BUTTERFLY_STD(cospi_11_64, cospi_21_64, &q1s16, &q3s16) + // part of stage 2 + q13s16 = vsubq_s16(q0s16, q1s16); + q0s16 = vaddq_s16(q0s16, q1s16); + q14s16 = vsubq_s16(q2s16, q3s16); + q2s16 = vaddq_s16(q2s16, q3s16); + // part of stage 3 + DO_BUTTERFLY_STD(cospi_12_64, cospi_20_64, &q1s16, &q3s16) + + // generate 22,23,24,25 + // part of stage 1 + LOAD_FROM_TRANSPOSED(11, 13, 19) + DO_BUTTERFLY_STD(cospi_19_64, cospi_13_64, &q5s16, &q7s16) + LOAD_FROM_TRANSPOSED(19, 29, 3) + DO_BUTTERFLY_STD(cospi_3_64, cospi_29_64, &q4s16, &q6s16) + // part of stage 2 + q14s16 = vsubq_s16(q4s16, q5s16); + q5s16 = vaddq_s16(q4s16, q5s16); + q13s16 = vsubq_s16(q6s16, q7s16); + q6s16 = vaddq_s16(q6s16, q7s16); + // part of stage 3 + DO_BUTTERFLY_STD(-cospi_20_64, -cospi_12_64, &q4s16, &q7s16) + // part of stage 4 + q10s16 = vaddq_s16(q7s16, q1s16); + q11s16 = vaddq_s16(q5s16, q0s16); + q12s16 = vaddq_s16(q6s16, q2s16); + q15s16 = vaddq_s16(q4s16, q3s16); + // part of stage 6 + LOAD_FROM_OUTPUT(28, 16, 17, q14s16, q13s16) + q8s16 = vaddq_s16(q14s16, q11s16); + q9s16 = vaddq_s16(q13s16, q10s16); + q13s16 = vsubq_s16(q13s16, q10s16); + q11s16 = vsubq_s16(q14s16, q11s16); + STORE_IN_OUTPUT(17, 17, 16, q9s16, q8s16) + LOAD_FROM_OUTPUT(16, 30, 31, q14s16, q9s16) + q8s16 = vsubq_s16(q9s16, q12s16); + q10s16 = vaddq_s16(q14s16, q15s16); + q14s16 = vsubq_s16(q14s16, q15s16); + q12s16 = vaddq_s16(q9s16, q12s16); + STORE_IN_OUTPUT(31, 30, 31, q10s16, q12s16) + // part of stage 7 + DO_BUTTERFLY_STD(cospi_16_64, cospi_16_64, &q13s16, &q14s16) + STORE_IN_OUTPUT(31, 25, 22, q14s16, q13s16) + q13s16 = q11s16; + q14s16 = q8s16; + DO_BUTTERFLY_STD(cospi_16_64, cospi_16_64, &q13s16, &q14s16) + STORE_IN_OUTPUT(22, 24, 23, q14s16, q13s16) + // part of stage 4 + q14s16 = vsubq_s16(q5s16, q0s16); + q13s16 = vsubq_s16(q6s16, q2s16); + DO_BUTTERFLY_STD(-cospi_8_64, -cospi_24_64, &q5s16, &q6s16); + q14s16 = vsubq_s16(q7s16, q1s16); + q13s16 = vsubq_s16(q4s16, q3s16); + DO_BUTTERFLY_STD(-cospi_8_64, -cospi_24_64, &q0s16, &q1s16); + // part of stage 6 + LOAD_FROM_OUTPUT(23, 18, 19, q14s16, q13s16) + q8s16 = vaddq_s16(q14s16, q1s16); + q9s16 = vaddq_s16(q13s16, q6s16); + q13s16 = vsubq_s16(q13s16, q6s16); + q1s16 = vsubq_s16(q14s16, q1s16); + STORE_IN_OUTPUT(19, 18, 19, q8s16, q9s16) + LOAD_FROM_OUTPUT(19, 28, 29, q8s16, q9s16) + q14s16 = vsubq_s16(q8s16, q5s16); + q10s16 = vaddq_s16(q8s16, q5s16); + q11s16 = vaddq_s16(q9s16, q0s16); + q0s16 = vsubq_s16(q9s16, q0s16); + STORE_IN_OUTPUT(29, 28, 29, q10s16, q11s16) + // part of stage 7 + DO_BUTTERFLY_STD(cospi_16_64, cospi_16_64, &q13s16, &q14s16) + STORE_IN_OUTPUT(29, 20, 27, q13s16, q14s16) + DO_BUTTERFLY(q0s16, q1s16, cospi_16_64, cospi_16_64, + &q1s16, &q0s16); + STORE_IN_OUTPUT(27, 21, 26, q1s16, q0s16) + + // ----------------------------------------- + // BLOCK C: 8-10,11-15 + // ----------------------------------------- + // generate 8,9,14,15 + // part of stage 2 + LOAD_FROM_TRANSPOSED(3, 2, 30) + DO_BUTTERFLY_STD(cospi_30_64, cospi_2_64, &q0s16, &q2s16) + LOAD_FROM_TRANSPOSED(30, 18, 14) + DO_BUTTERFLY_STD(cospi_14_64, cospi_18_64, &q1s16, &q3s16) + // part of stage 3 + q13s16 = vsubq_s16(q0s16, q1s16); + q0s16 = vaddq_s16(q0s16, q1s16); + q14s16 = vsubq_s16(q2s16, q3s16); + q2s16 = vaddq_s16(q2s16, q3s16); + // part of stage 4 + DO_BUTTERFLY_STD(cospi_24_64, cospi_8_64, &q1s16, &q3s16) + + // generate 10,11,12,13 + // part of stage 2 + LOAD_FROM_TRANSPOSED(14, 10, 22) + DO_BUTTERFLY_STD(cospi_22_64, cospi_10_64, &q5s16, &q7s16) + LOAD_FROM_TRANSPOSED(22, 26, 6) + DO_BUTTERFLY_STD(cospi_6_64, cospi_26_64, &q4s16, &q6s16) + // part of stage 3 + q14s16 = vsubq_s16(q4s16, q5s16); + q5s16 = vaddq_s16(q4s16, q5s16); + q13s16 = vsubq_s16(q6s16, q7s16); + q6s16 = vaddq_s16(q6s16, q7s16); + // part of stage 4 + DO_BUTTERFLY_STD(-cospi_8_64, -cospi_24_64, &q4s16, &q7s16) + // part of stage 5 + q8s16 = vaddq_s16(q0s16, q5s16); + q9s16 = vaddq_s16(q1s16, q7s16); + q13s16 = vsubq_s16(q1s16, q7s16); + q14s16 = vsubq_s16(q3s16, q4s16); + q10s16 = vaddq_s16(q3s16, q4s16); + q15s16 = vaddq_s16(q2s16, q6s16); + STORE_IN_OUTPUT(26, 8, 15, q8s16, q15s16) + STORE_IN_OUTPUT(15, 9, 14, q9s16, q10s16) + // part of stage 6 + DO_BUTTERFLY_STD(cospi_16_64, cospi_16_64, &q1s16, &q3s16) + STORE_IN_OUTPUT(14, 13, 10, q3s16, q1s16) + q13s16 = vsubq_s16(q0s16, q5s16); + q14s16 = vsubq_s16(q2s16, q6s16); + DO_BUTTERFLY_STD(cospi_16_64, cospi_16_64, &q1s16, &q3s16) + STORE_IN_OUTPUT(10, 11, 12, q1s16, q3s16) + + // ----------------------------------------- + // BLOCK D: 0-3,4-7 + // ----------------------------------------- + // generate 4,5,6,7 + // part of stage 3 + LOAD_FROM_TRANSPOSED(6, 4, 28) + DO_BUTTERFLY_STD(cospi_28_64, cospi_4_64, &q0s16, &q2s16) + LOAD_FROM_TRANSPOSED(28, 20, 12) + DO_BUTTERFLY_STD(cospi_12_64, cospi_20_64, &q1s16, &q3s16) + // part of stage 4 + q13s16 = vsubq_s16(q0s16, q1s16); + q0s16 = vaddq_s16(q0s16, q1s16); + q14s16 = vsubq_s16(q2s16, q3s16); + q2s16 = vaddq_s16(q2s16, q3s16); + // part of stage 5 + DO_BUTTERFLY_STD(cospi_16_64, cospi_16_64, &q1s16, &q3s16) + + // generate 0,1,2,3 + // part of stage 4 + LOAD_FROM_TRANSPOSED(12, 0, 16) + DO_BUTTERFLY_STD(cospi_16_64, cospi_16_64, &q5s16, &q7s16) + LOAD_FROM_TRANSPOSED(16, 8, 24) + DO_BUTTERFLY_STD(cospi_24_64, cospi_8_64, &q14s16, &q6s16) + // part of stage 5 + q4s16 = vaddq_s16(q7s16, q6s16); + q7s16 = vsubq_s16(q7s16, q6s16); + q6s16 = vsubq_s16(q5s16, q14s16); + q5s16 = vaddq_s16(q5s16, q14s16); + // part of stage 6 + q8s16 = vaddq_s16(q4s16, q2s16); + q9s16 = vaddq_s16(q5s16, q3s16); + q10s16 = vaddq_s16(q6s16, q1s16); + q11s16 = vaddq_s16(q7s16, q0s16); + q12s16 = vsubq_s16(q7s16, q0s16); + q13s16 = vsubq_s16(q6s16, q1s16); + q14s16 = vsubq_s16(q5s16, q3s16); + q15s16 = vsubq_s16(q4s16, q2s16); + // part of stage 7 + LOAD_FROM_OUTPUT(12, 14, 15, q0s16, q1s16) + q2s16 = vaddq_s16(q8s16, q1s16); + q3s16 = vaddq_s16(q9s16, q0s16); + q4s16 = vsubq_s16(q9s16, q0s16); + q5s16 = vsubq_s16(q8s16, q1s16); + LOAD_FROM_OUTPUT(15, 16, 17, q0s16, q1s16) + q8s16 = vaddq_s16(q4s16, q1s16); + q9s16 = vaddq_s16(q5s16, q0s16); + q6s16 = vsubq_s16(q5s16, q0s16); + q7s16 = vsubq_s16(q4s16, q1s16); + + if (idct32_pass_loop == 0) { + idct32_bands_end_1st_pass(out, + q2s16, q3s16, q6s16, q7s16, q8s16, q9s16, + q10s16, q11s16, q12s16, q13s16, q14s16, q15s16); + } else { + idct32_bands_end_2nd_pass(out, dest, stride, + q2s16, q3s16, q6s16, q7s16, q8s16, q9s16, + q10s16, q11s16, q12s16, q13s16, q14s16, q15s16); + dest += 8; + } + } + } + return; +} diff --git a/thirdparty/libvpx/vpx_dsp/arm/idct4x4_1_add_neon.c b/thirdparty/libvpx/vpx_dsp/arm/idct4x4_1_add_neon.c new file mode 100644 index 0000000000..ea618700c9 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/arm/idct4x4_1_add_neon.c @@ -0,0 +1,50 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +#include "vpx_dsp/inv_txfm.h" +#include "vpx_ports/mem.h" + +void vpx_idct4x4_1_add_neon( + int16_t *input, + uint8_t *dest, + int dest_stride) { + uint8x8_t d6u8; + uint32x2_t d2u32 = vdup_n_u32(0); + uint16x8_t q8u16; + int16x8_t q0s16; + uint8_t *d1, *d2; + int16_t i, a1, cospi_16_64 = 11585; + int16_t out = dct_const_round_shift(input[0] * cospi_16_64); + out = dct_const_round_shift(out * cospi_16_64); + a1 = ROUND_POWER_OF_TWO(out, 4); + + q0s16 = vdupq_n_s16(a1); + + // dc_only_idct_add + d1 = d2 = dest; + for (i = 0; i < 2; i++) { + d2u32 = vld1_lane_u32((const uint32_t *)d1, d2u32, 0); + d1 += dest_stride; + d2u32 = vld1_lane_u32((const uint32_t *)d1, d2u32, 1); + d1 += dest_stride; + + q8u16 = vaddw_u8(vreinterpretq_u16_s16(q0s16), + vreinterpret_u8_u32(d2u32)); + d6u8 = vqmovun_s16(vreinterpretq_s16_u16(q8u16)); + + vst1_lane_u32((uint32_t *)d2, vreinterpret_u32_u8(d6u8), 0); + d2 += dest_stride; + vst1_lane_u32((uint32_t *)d2, vreinterpret_u32_u8(d6u8), 1); + d2 += dest_stride; + } + return; +} diff --git a/thirdparty/libvpx/vpx_dsp/arm/idct4x4_add_neon.c b/thirdparty/libvpx/vpx_dsp/arm/idct4x4_add_neon.c new file mode 100644 index 0000000000..3c975c99b7 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/arm/idct4x4_add_neon.c @@ -0,0 +1,151 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +void vpx_idct4x4_16_add_neon( + int16_t *input, + uint8_t *dest, + int dest_stride) { + uint8x8_t d26u8, d27u8; + uint32x2_t d26u32, d27u32; + uint16x8_t q8u16, q9u16; + int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16; + int16x4_t d22s16, d23s16, d24s16, d26s16, d27s16, d28s16, d29s16; + int16x8_t q8s16, q9s16, q13s16, q14s16; + int32x4_t q1s32, q13s32, q14s32, q15s32; + int16x4x2_t d0x2s16, d1x2s16; + int32x4x2_t q0x2s32; + uint8_t *d; + int16_t cospi_8_64 = 15137; + int16_t cospi_16_64 = 11585; + int16_t cospi_24_64 = 6270; + + d26u32 = d27u32 = vdup_n_u32(0); + + q8s16 = vld1q_s16(input); + q9s16 = vld1q_s16(input + 8); + + d16s16 = vget_low_s16(q8s16); + d17s16 = vget_high_s16(q8s16); + d18s16 = vget_low_s16(q9s16); + d19s16 = vget_high_s16(q9s16); + + d0x2s16 = vtrn_s16(d16s16, d17s16); + d1x2s16 = vtrn_s16(d18s16, d19s16); + q8s16 = vcombine_s16(d0x2s16.val[0], d0x2s16.val[1]); + q9s16 = vcombine_s16(d1x2s16.val[0], d1x2s16.val[1]); + + d20s16 = vdup_n_s16(cospi_8_64); + d21s16 = vdup_n_s16(cospi_16_64); + + q0x2s32 = vtrnq_s32(vreinterpretq_s32_s16(q8s16), + vreinterpretq_s32_s16(q9s16)); + d16s16 = vget_low_s16(vreinterpretq_s16_s32(q0x2s32.val[0])); + d17s16 = vget_high_s16(vreinterpretq_s16_s32(q0x2s32.val[0])); + d18s16 = vget_low_s16(vreinterpretq_s16_s32(q0x2s32.val[1])); + d19s16 = vget_high_s16(vreinterpretq_s16_s32(q0x2s32.val[1])); + + d22s16 = vdup_n_s16(cospi_24_64); + + // stage 1 + d23s16 = vadd_s16(d16s16, d18s16); + d24s16 = vsub_s16(d16s16, d18s16); + + q15s32 = vmull_s16(d17s16, d22s16); + q1s32 = vmull_s16(d17s16, d20s16); + q13s32 = vmull_s16(d23s16, d21s16); + q14s32 = vmull_s16(d24s16, d21s16); + + q15s32 = vmlsl_s16(q15s32, d19s16, d20s16); + q1s32 = vmlal_s16(q1s32, d19s16, d22s16); + + d26s16 = vqrshrn_n_s32(q13s32, 14); + d27s16 = vqrshrn_n_s32(q14s32, 14); + d29s16 = vqrshrn_n_s32(q15s32, 14); + d28s16 = vqrshrn_n_s32(q1s32, 14); + q13s16 = vcombine_s16(d26s16, d27s16); + q14s16 = vcombine_s16(d28s16, d29s16); + + // stage 2 + q8s16 = vaddq_s16(q13s16, q14s16); + q9s16 = vsubq_s16(q13s16, q14s16); + + d16s16 = vget_low_s16(q8s16); + d17s16 = vget_high_s16(q8s16); + d18s16 = vget_high_s16(q9s16); // vswp d18 d19 + d19s16 = vget_low_s16(q9s16); + + d0x2s16 = vtrn_s16(d16s16, d17s16); + d1x2s16 = vtrn_s16(d18s16, d19s16); + q8s16 = vcombine_s16(d0x2s16.val[0], d0x2s16.val[1]); + q9s16 = vcombine_s16(d1x2s16.val[0], d1x2s16.val[1]); + + q0x2s32 = vtrnq_s32(vreinterpretq_s32_s16(q8s16), + vreinterpretq_s32_s16(q9s16)); + d16s16 = vget_low_s16(vreinterpretq_s16_s32(q0x2s32.val[0])); + d17s16 = vget_high_s16(vreinterpretq_s16_s32(q0x2s32.val[0])); + d18s16 = vget_low_s16(vreinterpretq_s16_s32(q0x2s32.val[1])); + d19s16 = vget_high_s16(vreinterpretq_s16_s32(q0x2s32.val[1])); + + // do the transform on columns + // stage 1 + d23s16 = vadd_s16(d16s16, d18s16); + d24s16 = vsub_s16(d16s16, d18s16); + + q15s32 = vmull_s16(d17s16, d22s16); + q1s32 = vmull_s16(d17s16, d20s16); + q13s32 = vmull_s16(d23s16, d21s16); + q14s32 = vmull_s16(d24s16, d21s16); + + q15s32 = vmlsl_s16(q15s32, d19s16, d20s16); + q1s32 = vmlal_s16(q1s32, d19s16, d22s16); + + d26s16 = vqrshrn_n_s32(q13s32, 14); + d27s16 = vqrshrn_n_s32(q14s32, 14); + d29s16 = vqrshrn_n_s32(q15s32, 14); + d28s16 = vqrshrn_n_s32(q1s32, 14); + q13s16 = vcombine_s16(d26s16, d27s16); + q14s16 = vcombine_s16(d28s16, d29s16); + + // stage 2 + q8s16 = vaddq_s16(q13s16, q14s16); + q9s16 = vsubq_s16(q13s16, q14s16); + + q8s16 = vrshrq_n_s16(q8s16, 4); + q9s16 = vrshrq_n_s16(q9s16, 4); + + d = dest; + d26u32 = vld1_lane_u32((const uint32_t *)d, d26u32, 0); + d += dest_stride; + d26u32 = vld1_lane_u32((const uint32_t *)d, d26u32, 1); + d += dest_stride; + d27u32 = vld1_lane_u32((const uint32_t *)d, d27u32, 1); + d += dest_stride; + d27u32 = vld1_lane_u32((const uint32_t *)d, d27u32, 0); + + q8u16 = vaddw_u8(vreinterpretq_u16_s16(q8s16), + vreinterpret_u8_u32(d26u32)); + q9u16 = vaddw_u8(vreinterpretq_u16_s16(q9s16), + vreinterpret_u8_u32(d27u32)); + + d26u8 = vqmovun_s16(vreinterpretq_s16_u16(q8u16)); + d27u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16)); + + d = dest; + vst1_lane_u32((uint32_t *)d, vreinterpret_u32_u8(d26u8), 0); + d += dest_stride; + vst1_lane_u32((uint32_t *)d, vreinterpret_u32_u8(d26u8), 1); + d += dest_stride; + vst1_lane_u32((uint32_t *)d, vreinterpret_u32_u8(d27u8), 1); + d += dest_stride; + vst1_lane_u32((uint32_t *)d, vreinterpret_u32_u8(d27u8), 0); + return; +} diff --git a/thirdparty/libvpx/vpx_dsp/arm/idct8x8_1_add_neon.c b/thirdparty/libvpx/vpx_dsp/arm/idct8x8_1_add_neon.c new file mode 100644 index 0000000000..c1b801fad5 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/arm/idct8x8_1_add_neon.c @@ -0,0 +1,64 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +#include "vpx_dsp/inv_txfm.h" +#include "vpx_ports/mem.h" + +void vpx_idct8x8_1_add_neon( + int16_t *input, + uint8_t *dest, + int dest_stride) { + uint8x8_t d2u8, d3u8, d30u8, d31u8; + uint64x1_t d2u64, d3u64, d4u64, d5u64; + uint16x8_t q0u16, q9u16, q10u16, q11u16, q12u16; + int16x8_t q0s16; + uint8_t *d1, *d2; + int16_t i, a1, cospi_16_64 = 11585; + int16_t out = dct_const_round_shift(input[0] * cospi_16_64); + out = dct_const_round_shift(out * cospi_16_64); + a1 = ROUND_POWER_OF_TWO(out, 5); + + q0s16 = vdupq_n_s16(a1); + q0u16 = vreinterpretq_u16_s16(q0s16); + + d1 = d2 = dest; + for (i = 0; i < 2; i++) { + d2u64 = vld1_u64((const uint64_t *)d1); + d1 += dest_stride; + d3u64 = vld1_u64((const uint64_t *)d1); + d1 += dest_stride; + d4u64 = vld1_u64((const uint64_t *)d1); + d1 += dest_stride; + d5u64 = vld1_u64((const uint64_t *)d1); + d1 += dest_stride; + + q9u16 = vaddw_u8(q0u16, vreinterpret_u8_u64(d2u64)); + q10u16 = vaddw_u8(q0u16, vreinterpret_u8_u64(d3u64)); + q11u16 = vaddw_u8(q0u16, vreinterpret_u8_u64(d4u64)); + q12u16 = vaddw_u8(q0u16, vreinterpret_u8_u64(d5u64)); + + d2u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16)); + d3u8 = vqmovun_s16(vreinterpretq_s16_u16(q10u16)); + d30u8 = vqmovun_s16(vreinterpretq_s16_u16(q11u16)); + d31u8 = vqmovun_s16(vreinterpretq_s16_u16(q12u16)); + + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d2u8)); + d2 += dest_stride; + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d3u8)); + d2 += dest_stride; + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d30u8)); + d2 += dest_stride; + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d31u8)); + d2 += dest_stride; + } + return; +} diff --git a/thirdparty/libvpx/vpx_dsp/arm/idct8x8_add_neon.c b/thirdparty/libvpx/vpx_dsp/arm/idct8x8_add_neon.c new file mode 100644 index 0000000000..4b2c2a6f83 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/arm/idct8x8_add_neon.c @@ -0,0 +1,540 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +#include "./vpx_config.h" +#include "vpx_dsp/txfm_common.h" + +static INLINE void TRANSPOSE8X8( + int16x8_t *q8s16, + int16x8_t *q9s16, + int16x8_t *q10s16, + int16x8_t *q11s16, + int16x8_t *q12s16, + int16x8_t *q13s16, + int16x8_t *q14s16, + int16x8_t *q15s16) { + int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16; + int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16; + int32x4x2_t q0x2s32, q1x2s32, q2x2s32, q3x2s32; + int16x8x2_t q0x2s16, q1x2s16, q2x2s16, q3x2s16; + + d16s16 = vget_low_s16(*q8s16); + d17s16 = vget_high_s16(*q8s16); + d18s16 = vget_low_s16(*q9s16); + d19s16 = vget_high_s16(*q9s16); + d20s16 = vget_low_s16(*q10s16); + d21s16 = vget_high_s16(*q10s16); + d22s16 = vget_low_s16(*q11s16); + d23s16 = vget_high_s16(*q11s16); + d24s16 = vget_low_s16(*q12s16); + d25s16 = vget_high_s16(*q12s16); + d26s16 = vget_low_s16(*q13s16); + d27s16 = vget_high_s16(*q13s16); + d28s16 = vget_low_s16(*q14s16); + d29s16 = vget_high_s16(*q14s16); + d30s16 = vget_low_s16(*q15s16); + d31s16 = vget_high_s16(*q15s16); + + *q8s16 = vcombine_s16(d16s16, d24s16); // vswp d17, d24 + *q9s16 = vcombine_s16(d18s16, d26s16); // vswp d19, d26 + *q10s16 = vcombine_s16(d20s16, d28s16); // vswp d21, d28 + *q11s16 = vcombine_s16(d22s16, d30s16); // vswp d23, d30 + *q12s16 = vcombine_s16(d17s16, d25s16); + *q13s16 = vcombine_s16(d19s16, d27s16); + *q14s16 = vcombine_s16(d21s16, d29s16); + *q15s16 = vcombine_s16(d23s16, d31s16); + + q0x2s32 = vtrnq_s32(vreinterpretq_s32_s16(*q8s16), + vreinterpretq_s32_s16(*q10s16)); + q1x2s32 = vtrnq_s32(vreinterpretq_s32_s16(*q9s16), + vreinterpretq_s32_s16(*q11s16)); + q2x2s32 = vtrnq_s32(vreinterpretq_s32_s16(*q12s16), + vreinterpretq_s32_s16(*q14s16)); + q3x2s32 = vtrnq_s32(vreinterpretq_s32_s16(*q13s16), + vreinterpretq_s32_s16(*q15s16)); + + q0x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[0]), // q8 + vreinterpretq_s16_s32(q1x2s32.val[0])); // q9 + q1x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[1]), // q10 + vreinterpretq_s16_s32(q1x2s32.val[1])); // q11 + q2x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[0]), // q12 + vreinterpretq_s16_s32(q3x2s32.val[0])); // q13 + q3x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[1]), // q14 + vreinterpretq_s16_s32(q3x2s32.val[1])); // q15 + + *q8s16 = q0x2s16.val[0]; + *q9s16 = q0x2s16.val[1]; + *q10s16 = q1x2s16.val[0]; + *q11s16 = q1x2s16.val[1]; + *q12s16 = q2x2s16.val[0]; + *q13s16 = q2x2s16.val[1]; + *q14s16 = q3x2s16.val[0]; + *q15s16 = q3x2s16.val[1]; + return; +} + +static INLINE void IDCT8x8_1D( + int16x8_t *q8s16, + int16x8_t *q9s16, + int16x8_t *q10s16, + int16x8_t *q11s16, + int16x8_t *q12s16, + int16x8_t *q13s16, + int16x8_t *q14s16, + int16x8_t *q15s16) { + int16x4_t d0s16, d1s16, d2s16, d3s16; + int16x4_t d8s16, d9s16, d10s16, d11s16, d12s16, d13s16, d14s16, d15s16; + int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16; + int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16; + int16x8_t q0s16, q1s16, q2s16, q3s16, q4s16, q5s16, q6s16, q7s16; + int32x4_t q2s32, q3s32, q5s32, q6s32, q8s32, q9s32; + int32x4_t q10s32, q11s32, q12s32, q13s32, q15s32; + + d0s16 = vdup_n_s16(cospi_28_64); + d1s16 = vdup_n_s16(cospi_4_64); + d2s16 = vdup_n_s16(cospi_12_64); + d3s16 = vdup_n_s16(cospi_20_64); + + d16s16 = vget_low_s16(*q8s16); + d17s16 = vget_high_s16(*q8s16); + d18s16 = vget_low_s16(*q9s16); + d19s16 = vget_high_s16(*q9s16); + d20s16 = vget_low_s16(*q10s16); + d21s16 = vget_high_s16(*q10s16); + d22s16 = vget_low_s16(*q11s16); + d23s16 = vget_high_s16(*q11s16); + d24s16 = vget_low_s16(*q12s16); + d25s16 = vget_high_s16(*q12s16); + d26s16 = vget_low_s16(*q13s16); + d27s16 = vget_high_s16(*q13s16); + d28s16 = vget_low_s16(*q14s16); + d29s16 = vget_high_s16(*q14s16); + d30s16 = vget_low_s16(*q15s16); + d31s16 = vget_high_s16(*q15s16); + + q2s32 = vmull_s16(d18s16, d0s16); + q3s32 = vmull_s16(d19s16, d0s16); + q5s32 = vmull_s16(d26s16, d2s16); + q6s32 = vmull_s16(d27s16, d2s16); + + q2s32 = vmlsl_s16(q2s32, d30s16, d1s16); + q3s32 = vmlsl_s16(q3s32, d31s16, d1s16); + q5s32 = vmlsl_s16(q5s32, d22s16, d3s16); + q6s32 = vmlsl_s16(q6s32, d23s16, d3s16); + + d8s16 = vqrshrn_n_s32(q2s32, 14); + d9s16 = vqrshrn_n_s32(q3s32, 14); + d10s16 = vqrshrn_n_s32(q5s32, 14); + d11s16 = vqrshrn_n_s32(q6s32, 14); + q4s16 = vcombine_s16(d8s16, d9s16); + q5s16 = vcombine_s16(d10s16, d11s16); + + q2s32 = vmull_s16(d18s16, d1s16); + q3s32 = vmull_s16(d19s16, d1s16); + q9s32 = vmull_s16(d26s16, d3s16); + q13s32 = vmull_s16(d27s16, d3s16); + + q2s32 = vmlal_s16(q2s32, d30s16, d0s16); + q3s32 = vmlal_s16(q3s32, d31s16, d0s16); + q9s32 = vmlal_s16(q9s32, d22s16, d2s16); + q13s32 = vmlal_s16(q13s32, d23s16, d2s16); + + d14s16 = vqrshrn_n_s32(q2s32, 14); + d15s16 = vqrshrn_n_s32(q3s32, 14); + d12s16 = vqrshrn_n_s32(q9s32, 14); + d13s16 = vqrshrn_n_s32(q13s32, 14); + q6s16 = vcombine_s16(d12s16, d13s16); + q7s16 = vcombine_s16(d14s16, d15s16); + + d0s16 = vdup_n_s16(cospi_16_64); + + q2s32 = vmull_s16(d16s16, d0s16); + q3s32 = vmull_s16(d17s16, d0s16); + q13s32 = vmull_s16(d16s16, d0s16); + q15s32 = vmull_s16(d17s16, d0s16); + + q2s32 = vmlal_s16(q2s32, d24s16, d0s16); + q3s32 = vmlal_s16(q3s32, d25s16, d0s16); + q13s32 = vmlsl_s16(q13s32, d24s16, d0s16); + q15s32 = vmlsl_s16(q15s32, d25s16, d0s16); + + d0s16 = vdup_n_s16(cospi_24_64); + d1s16 = vdup_n_s16(cospi_8_64); + + d18s16 = vqrshrn_n_s32(q2s32, 14); + d19s16 = vqrshrn_n_s32(q3s32, 14); + d22s16 = vqrshrn_n_s32(q13s32, 14); + d23s16 = vqrshrn_n_s32(q15s32, 14); + *q9s16 = vcombine_s16(d18s16, d19s16); + *q11s16 = vcombine_s16(d22s16, d23s16); + + q2s32 = vmull_s16(d20s16, d0s16); + q3s32 = vmull_s16(d21s16, d0s16); + q8s32 = vmull_s16(d20s16, d1s16); + q12s32 = vmull_s16(d21s16, d1s16); + + q2s32 = vmlsl_s16(q2s32, d28s16, d1s16); + q3s32 = vmlsl_s16(q3s32, d29s16, d1s16); + q8s32 = vmlal_s16(q8s32, d28s16, d0s16); + q12s32 = vmlal_s16(q12s32, d29s16, d0s16); + + d26s16 = vqrshrn_n_s32(q2s32, 14); + d27s16 = vqrshrn_n_s32(q3s32, 14); + d30s16 = vqrshrn_n_s32(q8s32, 14); + d31s16 = vqrshrn_n_s32(q12s32, 14); + *q13s16 = vcombine_s16(d26s16, d27s16); + *q15s16 = vcombine_s16(d30s16, d31s16); + + q0s16 = vaddq_s16(*q9s16, *q15s16); + q1s16 = vaddq_s16(*q11s16, *q13s16); + q2s16 = vsubq_s16(*q11s16, *q13s16); + q3s16 = vsubq_s16(*q9s16, *q15s16); + + *q13s16 = vsubq_s16(q4s16, q5s16); + q4s16 = vaddq_s16(q4s16, q5s16); + *q14s16 = vsubq_s16(q7s16, q6s16); + q7s16 = vaddq_s16(q7s16, q6s16); + d26s16 = vget_low_s16(*q13s16); + d27s16 = vget_high_s16(*q13s16); + d28s16 = vget_low_s16(*q14s16); + d29s16 = vget_high_s16(*q14s16); + + d16s16 = vdup_n_s16(cospi_16_64); + + q9s32 = vmull_s16(d28s16, d16s16); + q10s32 = vmull_s16(d29s16, d16s16); + q11s32 = vmull_s16(d28s16, d16s16); + q12s32 = vmull_s16(d29s16, d16s16); + + q9s32 = vmlsl_s16(q9s32, d26s16, d16s16); + q10s32 = vmlsl_s16(q10s32, d27s16, d16s16); + q11s32 = vmlal_s16(q11s32, d26s16, d16s16); + q12s32 = vmlal_s16(q12s32, d27s16, d16s16); + + d10s16 = vqrshrn_n_s32(q9s32, 14); + d11s16 = vqrshrn_n_s32(q10s32, 14); + d12s16 = vqrshrn_n_s32(q11s32, 14); + d13s16 = vqrshrn_n_s32(q12s32, 14); + q5s16 = vcombine_s16(d10s16, d11s16); + q6s16 = vcombine_s16(d12s16, d13s16); + + *q8s16 = vaddq_s16(q0s16, q7s16); + *q9s16 = vaddq_s16(q1s16, q6s16); + *q10s16 = vaddq_s16(q2s16, q5s16); + *q11s16 = vaddq_s16(q3s16, q4s16); + *q12s16 = vsubq_s16(q3s16, q4s16); + *q13s16 = vsubq_s16(q2s16, q5s16); + *q14s16 = vsubq_s16(q1s16, q6s16); + *q15s16 = vsubq_s16(q0s16, q7s16); + return; +} + +void vpx_idct8x8_64_add_neon( + int16_t *input, + uint8_t *dest, + int dest_stride) { + uint8_t *d1, *d2; + uint8x8_t d0u8, d1u8, d2u8, d3u8; + uint64x1_t d0u64, d1u64, d2u64, d3u64; + int16x8_t q8s16, q9s16, q10s16, q11s16, q12s16, q13s16, q14s16, q15s16; + uint16x8_t q8u16, q9u16, q10u16, q11u16; + + q8s16 = vld1q_s16(input); + q9s16 = vld1q_s16(input + 8); + q10s16 = vld1q_s16(input + 16); + q11s16 = vld1q_s16(input + 24); + q12s16 = vld1q_s16(input + 32); + q13s16 = vld1q_s16(input + 40); + q14s16 = vld1q_s16(input + 48); + q15s16 = vld1q_s16(input + 56); + + TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, + &q12s16, &q13s16, &q14s16, &q15s16); + + IDCT8x8_1D(&q8s16, &q9s16, &q10s16, &q11s16, + &q12s16, &q13s16, &q14s16, &q15s16); + + TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, + &q12s16, &q13s16, &q14s16, &q15s16); + + IDCT8x8_1D(&q8s16, &q9s16, &q10s16, &q11s16, + &q12s16, &q13s16, &q14s16, &q15s16); + + q8s16 = vrshrq_n_s16(q8s16, 5); + q9s16 = vrshrq_n_s16(q9s16, 5); + q10s16 = vrshrq_n_s16(q10s16, 5); + q11s16 = vrshrq_n_s16(q11s16, 5); + q12s16 = vrshrq_n_s16(q12s16, 5); + q13s16 = vrshrq_n_s16(q13s16, 5); + q14s16 = vrshrq_n_s16(q14s16, 5); + q15s16 = vrshrq_n_s16(q15s16, 5); + + d1 = d2 = dest; + + d0u64 = vld1_u64((uint64_t *)d1); + d1 += dest_stride; + d1u64 = vld1_u64((uint64_t *)d1); + d1 += dest_stride; + d2u64 = vld1_u64((uint64_t *)d1); + d1 += dest_stride; + d3u64 = vld1_u64((uint64_t *)d1); + d1 += dest_stride; + + q8u16 = vaddw_u8(vreinterpretq_u16_s16(q8s16), + vreinterpret_u8_u64(d0u64)); + q9u16 = vaddw_u8(vreinterpretq_u16_s16(q9s16), + vreinterpret_u8_u64(d1u64)); + q10u16 = vaddw_u8(vreinterpretq_u16_s16(q10s16), + vreinterpret_u8_u64(d2u64)); + q11u16 = vaddw_u8(vreinterpretq_u16_s16(q11s16), + vreinterpret_u8_u64(d3u64)); + + d0u8 = vqmovun_s16(vreinterpretq_s16_u16(q8u16)); + d1u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16)); + d2u8 = vqmovun_s16(vreinterpretq_s16_u16(q10u16)); + d3u8 = vqmovun_s16(vreinterpretq_s16_u16(q11u16)); + + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d0u8)); + d2 += dest_stride; + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d1u8)); + d2 += dest_stride; + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d2u8)); + d2 += dest_stride; + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d3u8)); + d2 += dest_stride; + + q8s16 = q12s16; + q9s16 = q13s16; + q10s16 = q14s16; + q11s16 = q15s16; + + d0u64 = vld1_u64((uint64_t *)d1); + d1 += dest_stride; + d1u64 = vld1_u64((uint64_t *)d1); + d1 += dest_stride; + d2u64 = vld1_u64((uint64_t *)d1); + d1 += dest_stride; + d3u64 = vld1_u64((uint64_t *)d1); + d1 += dest_stride; + + q8u16 = vaddw_u8(vreinterpretq_u16_s16(q8s16), + vreinterpret_u8_u64(d0u64)); + q9u16 = vaddw_u8(vreinterpretq_u16_s16(q9s16), + vreinterpret_u8_u64(d1u64)); + q10u16 = vaddw_u8(vreinterpretq_u16_s16(q10s16), + vreinterpret_u8_u64(d2u64)); + q11u16 = vaddw_u8(vreinterpretq_u16_s16(q11s16), + vreinterpret_u8_u64(d3u64)); + + d0u8 = vqmovun_s16(vreinterpretq_s16_u16(q8u16)); + d1u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16)); + d2u8 = vqmovun_s16(vreinterpretq_s16_u16(q10u16)); + d3u8 = vqmovun_s16(vreinterpretq_s16_u16(q11u16)); + + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d0u8)); + d2 += dest_stride; + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d1u8)); + d2 += dest_stride; + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d2u8)); + d2 += dest_stride; + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d3u8)); + d2 += dest_stride; + return; +} + +void vpx_idct8x8_12_add_neon( + int16_t *input, + uint8_t *dest, + int dest_stride) { + uint8_t *d1, *d2; + uint8x8_t d0u8, d1u8, d2u8, d3u8; + int16x4_t d10s16, d11s16, d12s16, d13s16, d16s16; + int16x4_t d26s16, d27s16, d28s16, d29s16; + uint64x1_t d0u64, d1u64, d2u64, d3u64; + int16x8_t q0s16, q1s16, q2s16, q3s16, q4s16, q5s16, q6s16, q7s16; + int16x8_t q8s16, q9s16, q10s16, q11s16, q12s16, q13s16, q14s16, q15s16; + uint16x8_t q8u16, q9u16, q10u16, q11u16; + int32x4_t q9s32, q10s32, q11s32, q12s32; + + q8s16 = vld1q_s16(input); + q9s16 = vld1q_s16(input + 8); + q10s16 = vld1q_s16(input + 16); + q11s16 = vld1q_s16(input + 24); + q12s16 = vld1q_s16(input + 32); + q13s16 = vld1q_s16(input + 40); + q14s16 = vld1q_s16(input + 48); + q15s16 = vld1q_s16(input + 56); + + TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, + &q12s16, &q13s16, &q14s16, &q15s16); + + // First transform rows + // stage 1 + q0s16 = vdupq_n_s16(cospi_28_64 * 2); + q1s16 = vdupq_n_s16(cospi_4_64 * 2); + + q4s16 = vqrdmulhq_s16(q9s16, q0s16); + + q0s16 = vdupq_n_s16(-cospi_20_64 * 2); + + q7s16 = vqrdmulhq_s16(q9s16, q1s16); + + q1s16 = vdupq_n_s16(cospi_12_64 * 2); + + q5s16 = vqrdmulhq_s16(q11s16, q0s16); + + q0s16 = vdupq_n_s16(cospi_16_64 * 2); + + q6s16 = vqrdmulhq_s16(q11s16, q1s16); + + // stage 2 & stage 3 - even half + q1s16 = vdupq_n_s16(cospi_24_64 * 2); + + q9s16 = vqrdmulhq_s16(q8s16, q0s16); + + q0s16 = vdupq_n_s16(cospi_8_64 * 2); + + q13s16 = vqrdmulhq_s16(q10s16, q1s16); + + q15s16 = vqrdmulhq_s16(q10s16, q0s16); + + // stage 3 -odd half + q0s16 = vaddq_s16(q9s16, q15s16); + q1s16 = vaddq_s16(q9s16, q13s16); + q2s16 = vsubq_s16(q9s16, q13s16); + q3s16 = vsubq_s16(q9s16, q15s16); + + // stage 2 - odd half + q13s16 = vsubq_s16(q4s16, q5s16); + q4s16 = vaddq_s16(q4s16, q5s16); + q14s16 = vsubq_s16(q7s16, q6s16); + q7s16 = vaddq_s16(q7s16, q6s16); + d26s16 = vget_low_s16(q13s16); + d27s16 = vget_high_s16(q13s16); + d28s16 = vget_low_s16(q14s16); + d29s16 = vget_high_s16(q14s16); + + d16s16 = vdup_n_s16(cospi_16_64); + q9s32 = vmull_s16(d28s16, d16s16); + q10s32 = vmull_s16(d29s16, d16s16); + q11s32 = vmull_s16(d28s16, d16s16); + q12s32 = vmull_s16(d29s16, d16s16); + + q9s32 = vmlsl_s16(q9s32, d26s16, d16s16); + q10s32 = vmlsl_s16(q10s32, d27s16, d16s16); + q11s32 = vmlal_s16(q11s32, d26s16, d16s16); + q12s32 = vmlal_s16(q12s32, d27s16, d16s16); + + d10s16 = vqrshrn_n_s32(q9s32, 14); + d11s16 = vqrshrn_n_s32(q10s32, 14); + d12s16 = vqrshrn_n_s32(q11s32, 14); + d13s16 = vqrshrn_n_s32(q12s32, 14); + q5s16 = vcombine_s16(d10s16, d11s16); + q6s16 = vcombine_s16(d12s16, d13s16); + + // stage 4 + q8s16 = vaddq_s16(q0s16, q7s16); + q9s16 = vaddq_s16(q1s16, q6s16); + q10s16 = vaddq_s16(q2s16, q5s16); + q11s16 = vaddq_s16(q3s16, q4s16); + q12s16 = vsubq_s16(q3s16, q4s16); + q13s16 = vsubq_s16(q2s16, q5s16); + q14s16 = vsubq_s16(q1s16, q6s16); + q15s16 = vsubq_s16(q0s16, q7s16); + + TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, + &q12s16, &q13s16, &q14s16, &q15s16); + + IDCT8x8_1D(&q8s16, &q9s16, &q10s16, &q11s16, + &q12s16, &q13s16, &q14s16, &q15s16); + + q8s16 = vrshrq_n_s16(q8s16, 5); + q9s16 = vrshrq_n_s16(q9s16, 5); + q10s16 = vrshrq_n_s16(q10s16, 5); + q11s16 = vrshrq_n_s16(q11s16, 5); + q12s16 = vrshrq_n_s16(q12s16, 5); + q13s16 = vrshrq_n_s16(q13s16, 5); + q14s16 = vrshrq_n_s16(q14s16, 5); + q15s16 = vrshrq_n_s16(q15s16, 5); + + d1 = d2 = dest; + + d0u64 = vld1_u64((uint64_t *)d1); + d1 += dest_stride; + d1u64 = vld1_u64((uint64_t *)d1); + d1 += dest_stride; + d2u64 = vld1_u64((uint64_t *)d1); + d1 += dest_stride; + d3u64 = vld1_u64((uint64_t *)d1); + d1 += dest_stride; + + q8u16 = vaddw_u8(vreinterpretq_u16_s16(q8s16), + vreinterpret_u8_u64(d0u64)); + q9u16 = vaddw_u8(vreinterpretq_u16_s16(q9s16), + vreinterpret_u8_u64(d1u64)); + q10u16 = vaddw_u8(vreinterpretq_u16_s16(q10s16), + vreinterpret_u8_u64(d2u64)); + q11u16 = vaddw_u8(vreinterpretq_u16_s16(q11s16), + vreinterpret_u8_u64(d3u64)); + + d0u8 = vqmovun_s16(vreinterpretq_s16_u16(q8u16)); + d1u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16)); + d2u8 = vqmovun_s16(vreinterpretq_s16_u16(q10u16)); + d3u8 = vqmovun_s16(vreinterpretq_s16_u16(q11u16)); + + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d0u8)); + d2 += dest_stride; + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d1u8)); + d2 += dest_stride; + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d2u8)); + d2 += dest_stride; + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d3u8)); + d2 += dest_stride; + + q8s16 = q12s16; + q9s16 = q13s16; + q10s16 = q14s16; + q11s16 = q15s16; + + d0u64 = vld1_u64((uint64_t *)d1); + d1 += dest_stride; + d1u64 = vld1_u64((uint64_t *)d1); + d1 += dest_stride; + d2u64 = vld1_u64((uint64_t *)d1); + d1 += dest_stride; + d3u64 = vld1_u64((uint64_t *)d1); + d1 += dest_stride; + + q8u16 = vaddw_u8(vreinterpretq_u16_s16(q8s16), + vreinterpret_u8_u64(d0u64)); + q9u16 = vaddw_u8(vreinterpretq_u16_s16(q9s16), + vreinterpret_u8_u64(d1u64)); + q10u16 = vaddw_u8(vreinterpretq_u16_s16(q10s16), + vreinterpret_u8_u64(d2u64)); + q11u16 = vaddw_u8(vreinterpretq_u16_s16(q11s16), + vreinterpret_u8_u64(d3u64)); + + d0u8 = vqmovun_s16(vreinterpretq_s16_u16(q8u16)); + d1u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16)); + d2u8 = vqmovun_s16(vreinterpretq_s16_u16(q10u16)); + d3u8 = vqmovun_s16(vreinterpretq_s16_u16(q11u16)); + + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d0u8)); + d2 += dest_stride; + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d1u8)); + d2 += dest_stride; + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d2u8)); + d2 += dest_stride; + vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d3u8)); + d2 += dest_stride; + return; +} diff --git a/thirdparty/libvpx/vpx_dsp/arm/intrapred_neon.c b/thirdparty/libvpx/vpx_dsp/arm/intrapred_neon.c new file mode 100644 index 0000000000..0a376104d2 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/arm/intrapred_neon.c @@ -0,0 +1,822 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +#include "./vpx_config.h" +#include "./vpx_dsp_rtcd.h" +#include "vpx/vpx_integer.h" + +//------------------------------------------------------------------------------ +// DC 4x4 + +// 'do_above' and 'do_left' facilitate branch removal when inlined. +static INLINE void dc_4x4(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left, + int do_above, int do_left) { + uint16x8_t sum_top; + uint16x8_t sum_left; + uint8x8_t dc0; + + if (do_above) { + const uint8x8_t A = vld1_u8(above); // top row + const uint16x4_t p0 = vpaddl_u8(A); // cascading summation of the top + const uint16x4_t p1 = vpadd_u16(p0, p0); + sum_top = vcombine_u16(p1, p1); + } + + if (do_left) { + const uint8x8_t L = vld1_u8(left); // left border + const uint16x4_t p0 = vpaddl_u8(L); // cascading summation of the left + const uint16x4_t p1 = vpadd_u16(p0, p0); + sum_left = vcombine_u16(p1, p1); + } + + if (do_above && do_left) { + const uint16x8_t sum = vaddq_u16(sum_left, sum_top); + dc0 = vrshrn_n_u16(sum, 3); + } else if (do_above) { + dc0 = vrshrn_n_u16(sum_top, 2); + } else if (do_left) { + dc0 = vrshrn_n_u16(sum_left, 2); + } else { + dc0 = vdup_n_u8(0x80); + } + + { + const uint8x8_t dc = vdup_lane_u8(dc0, 0); + int i; + for (i = 0; i < 4; ++i) { + vst1_lane_u32((uint32_t*)(dst + i * stride), vreinterpret_u32_u8(dc), 0); + } + } +} + +void vpx_dc_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + dc_4x4(dst, stride, above, left, 1, 1); +} + +void vpx_dc_left_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + (void)above; + dc_4x4(dst, stride, NULL, left, 0, 1); +} + +void vpx_dc_top_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + (void)left; + dc_4x4(dst, stride, above, NULL, 1, 0); +} + +void vpx_dc_128_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + (void)above; + (void)left; + dc_4x4(dst, stride, NULL, NULL, 0, 0); +} + +//------------------------------------------------------------------------------ +// DC 8x8 + +// 'do_above' and 'do_left' facilitate branch removal when inlined. +static INLINE void dc_8x8(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left, + int do_above, int do_left) { + uint16x8_t sum_top; + uint16x8_t sum_left; + uint8x8_t dc0; + + if (do_above) { + const uint8x8_t A = vld1_u8(above); // top row + const uint16x4_t p0 = vpaddl_u8(A); // cascading summation of the top + const uint16x4_t p1 = vpadd_u16(p0, p0); + const uint16x4_t p2 = vpadd_u16(p1, p1); + sum_top = vcombine_u16(p2, p2); + } + + if (do_left) { + const uint8x8_t L = vld1_u8(left); // left border + const uint16x4_t p0 = vpaddl_u8(L); // cascading summation of the left + const uint16x4_t p1 = vpadd_u16(p0, p0); + const uint16x4_t p2 = vpadd_u16(p1, p1); + sum_left = vcombine_u16(p2, p2); + } + + if (do_above && do_left) { + const uint16x8_t sum = vaddq_u16(sum_left, sum_top); + dc0 = vrshrn_n_u16(sum, 4); + } else if (do_above) { + dc0 = vrshrn_n_u16(sum_top, 3); + } else if (do_left) { + dc0 = vrshrn_n_u16(sum_left, 3); + } else { + dc0 = vdup_n_u8(0x80); + } + + { + const uint8x8_t dc = vdup_lane_u8(dc0, 0); + int i; + for (i = 0; i < 8; ++i) { + vst1_u32((uint32_t*)(dst + i * stride), vreinterpret_u32_u8(dc)); + } + } +} + +void vpx_dc_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + dc_8x8(dst, stride, above, left, 1, 1); +} + +void vpx_dc_left_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + (void)above; + dc_8x8(dst, stride, NULL, left, 0, 1); +} + +void vpx_dc_top_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + (void)left; + dc_8x8(dst, stride, above, NULL, 1, 0); +} + +void vpx_dc_128_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + (void)above; + (void)left; + dc_8x8(dst, stride, NULL, NULL, 0, 0); +} + +//------------------------------------------------------------------------------ +// DC 16x16 + +// 'do_above' and 'do_left' facilitate branch removal when inlined. +static INLINE void dc_16x16(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left, + int do_above, int do_left) { + uint16x8_t sum_top; + uint16x8_t sum_left; + uint8x8_t dc0; + + if (do_above) { + const uint8x16_t A = vld1q_u8(above); // top row + const uint16x8_t p0 = vpaddlq_u8(A); // cascading summation of the top + const uint16x4_t p1 = vadd_u16(vget_low_u16(p0), vget_high_u16(p0)); + const uint16x4_t p2 = vpadd_u16(p1, p1); + const uint16x4_t p3 = vpadd_u16(p2, p2); + sum_top = vcombine_u16(p3, p3); + } + + if (do_left) { + const uint8x16_t L = vld1q_u8(left); // left row + const uint16x8_t p0 = vpaddlq_u8(L); // cascading summation of the left + const uint16x4_t p1 = vadd_u16(vget_low_u16(p0), vget_high_u16(p0)); + const uint16x4_t p2 = vpadd_u16(p1, p1); + const uint16x4_t p3 = vpadd_u16(p2, p2); + sum_left = vcombine_u16(p3, p3); + } + + if (do_above && do_left) { + const uint16x8_t sum = vaddq_u16(sum_left, sum_top); + dc0 = vrshrn_n_u16(sum, 5); + } else if (do_above) { + dc0 = vrshrn_n_u16(sum_top, 4); + } else if (do_left) { + dc0 = vrshrn_n_u16(sum_left, 4); + } else { + dc0 = vdup_n_u8(0x80); + } + + { + const uint8x16_t dc = vdupq_lane_u8(dc0, 0); + int i; + for (i = 0; i < 16; ++i) { + vst1q_u8(dst + i * stride, dc); + } + } +} + +void vpx_dc_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + dc_16x16(dst, stride, above, left, 1, 1); +} + +void vpx_dc_left_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, + const uint8_t *left) { + (void)above; + dc_16x16(dst, stride, NULL, left, 0, 1); +} + +void vpx_dc_top_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, + const uint8_t *left) { + (void)left; + dc_16x16(dst, stride, above, NULL, 1, 0); +} + +void vpx_dc_128_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, + const uint8_t *left) { + (void)above; + (void)left; + dc_16x16(dst, stride, NULL, NULL, 0, 0); +} + +//------------------------------------------------------------------------------ +// DC 32x32 + +// 'do_above' and 'do_left' facilitate branch removal when inlined. +static INLINE void dc_32x32(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left, + int do_above, int do_left) { + uint16x8_t sum_top; + uint16x8_t sum_left; + uint8x8_t dc0; + + if (do_above) { + const uint8x16_t A0 = vld1q_u8(above); // top row + const uint8x16_t A1 = vld1q_u8(above + 16); + const uint16x8_t p0 = vpaddlq_u8(A0); // cascading summation of the top + const uint16x8_t p1 = vpaddlq_u8(A1); + const uint16x8_t p2 = vaddq_u16(p0, p1); + const uint16x4_t p3 = vadd_u16(vget_low_u16(p2), vget_high_u16(p2)); + const uint16x4_t p4 = vpadd_u16(p3, p3); + const uint16x4_t p5 = vpadd_u16(p4, p4); + sum_top = vcombine_u16(p5, p5); + } + + if (do_left) { + const uint8x16_t L0 = vld1q_u8(left); // left row + const uint8x16_t L1 = vld1q_u8(left + 16); + const uint16x8_t p0 = vpaddlq_u8(L0); // cascading summation of the left + const uint16x8_t p1 = vpaddlq_u8(L1); + const uint16x8_t p2 = vaddq_u16(p0, p1); + const uint16x4_t p3 = vadd_u16(vget_low_u16(p2), vget_high_u16(p2)); + const uint16x4_t p4 = vpadd_u16(p3, p3); + const uint16x4_t p5 = vpadd_u16(p4, p4); + sum_left = vcombine_u16(p5, p5); + } + + if (do_above && do_left) { + const uint16x8_t sum = vaddq_u16(sum_left, sum_top); + dc0 = vrshrn_n_u16(sum, 6); + } else if (do_above) { + dc0 = vrshrn_n_u16(sum_top, 5); + } else if (do_left) { + dc0 = vrshrn_n_u16(sum_left, 5); + } else { + dc0 = vdup_n_u8(0x80); + } + + { + const uint8x16_t dc = vdupq_lane_u8(dc0, 0); + int i; + for (i = 0; i < 32; ++i) { + vst1q_u8(dst + i * stride, dc); + vst1q_u8(dst + i * stride + 16, dc); + } + } +} + +void vpx_dc_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + dc_32x32(dst, stride, above, left, 1, 1); +} + +void vpx_dc_left_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, + const uint8_t *left) { + (void)above; + dc_32x32(dst, stride, NULL, left, 0, 1); +} + +void vpx_dc_top_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, + const uint8_t *left) { + (void)left; + dc_32x32(dst, stride, above, NULL, 1, 0); +} + +void vpx_dc_128_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, + const uint8_t *left) { + (void)above; + (void)left; + dc_32x32(dst, stride, NULL, NULL, 0, 0); +} + +// ----------------------------------------------------------------------------- + +void vpx_d45_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + const uint64x1_t A0 = vreinterpret_u64_u8(vld1_u8(above)); // top row + const uint64x1_t A1 = vshr_n_u64(A0, 8); + const uint64x1_t A2 = vshr_n_u64(A0, 16); + const uint8x8_t ABCDEFGH = vreinterpret_u8_u64(A0); + const uint8x8_t BCDEFGH0 = vreinterpret_u8_u64(A1); + const uint8x8_t CDEFGH00 = vreinterpret_u8_u64(A2); + const uint8x8_t avg1 = vhadd_u8(ABCDEFGH, CDEFGH00); + const uint8x8_t avg2 = vrhadd_u8(avg1, BCDEFGH0); + const uint64x1_t avg2_u64 = vreinterpret_u64_u8(avg2); + const uint32x2_t r0 = vreinterpret_u32_u8(avg2); + const uint32x2_t r1 = vreinterpret_u32_u64(vshr_n_u64(avg2_u64, 8)); + const uint32x2_t r2 = vreinterpret_u32_u64(vshr_n_u64(avg2_u64, 16)); + const uint32x2_t r3 = vreinterpret_u32_u64(vshr_n_u64(avg2_u64, 24)); + (void)left; + vst1_lane_u32((uint32_t *)(dst + 0 * stride), r0, 0); + vst1_lane_u32((uint32_t *)(dst + 1 * stride), r1, 0); + vst1_lane_u32((uint32_t *)(dst + 2 * stride), r2, 0); + vst1_lane_u32((uint32_t *)(dst + 3 * stride), r3, 0); + dst[3 * stride + 3] = above[7]; +} + +void vpx_d45_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + static const uint8_t shuffle1[8] = { 1, 2, 3, 4, 5, 6, 7, 7 }; + static const uint8_t shuffle2[8] = { 2, 3, 4, 5, 6, 7, 7, 7 }; + const uint8x8_t sh_12345677 = vld1_u8(shuffle1); + const uint8x8_t sh_23456777 = vld1_u8(shuffle2); + const uint8x8_t A0 = vld1_u8(above); // top row + const uint8x8_t A1 = vtbl1_u8(A0, sh_12345677); + const uint8x8_t A2 = vtbl1_u8(A0, sh_23456777); + const uint8x8_t avg1 = vhadd_u8(A0, A2); + uint8x8_t row = vrhadd_u8(avg1, A1); + int i; + (void)left; + for (i = 0; i < 7; ++i) { + vst1_u8(dst + i * stride, row); + row = vtbl1_u8(row, sh_12345677); + } + vst1_u8(dst + i * stride, row); +} + +void vpx_d45_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + const uint8x16_t A0 = vld1q_u8(above); // top row + const uint8x16_t above_right = vld1q_dup_u8(above + 15); + const uint8x16_t A1 = vextq_u8(A0, above_right, 1); + const uint8x16_t A2 = vextq_u8(A0, above_right, 2); + const uint8x16_t avg1 = vhaddq_u8(A0, A2); + uint8x16_t row = vrhaddq_u8(avg1, A1); + int i; + (void)left; + for (i = 0; i < 15; ++i) { + vst1q_u8(dst + i * stride, row); + row = vextq_u8(row, above_right, 1); + } + vst1q_u8(dst + i * stride, row); +} + +// ----------------------------------------------------------------------------- + +void vpx_d135_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + const uint8x8_t XABCD_u8 = vld1_u8(above - 1); + const uint64x1_t XABCD = vreinterpret_u64_u8(XABCD_u8); + const uint64x1_t ____XABC = vshl_n_u64(XABCD, 32); + const uint32x2_t zero = vdup_n_u32(0); + const uint32x2_t IJKL = vld1_lane_u32((const uint32_t *)left, zero, 0); + const uint8x8_t IJKL_u8 = vreinterpret_u8_u32(IJKL); + const uint64x1_t LKJI____ = vreinterpret_u64_u8(vrev32_u8(IJKL_u8)); + const uint64x1_t LKJIXABC = vorr_u64(LKJI____, ____XABC); + const uint8x8_t KJIXABC_ = vreinterpret_u8_u64(vshr_n_u64(LKJIXABC, 8)); + const uint8x8_t JIXABC__ = vreinterpret_u8_u64(vshr_n_u64(LKJIXABC, 16)); + const uint8_t D = vget_lane_u8(XABCD_u8, 4); + const uint8x8_t JIXABCD_ = vset_lane_u8(D, JIXABC__, 6); + const uint8x8_t LKJIXABC_u8 = vreinterpret_u8_u64(LKJIXABC); + const uint8x8_t avg1 = vhadd_u8(JIXABCD_, LKJIXABC_u8); + const uint8x8_t avg2 = vrhadd_u8(avg1, KJIXABC_); + const uint64x1_t avg2_u64 = vreinterpret_u64_u8(avg2); + const uint32x2_t r3 = vreinterpret_u32_u8(avg2); + const uint32x2_t r2 = vreinterpret_u32_u64(vshr_n_u64(avg2_u64, 8)); + const uint32x2_t r1 = vreinterpret_u32_u64(vshr_n_u64(avg2_u64, 16)); + const uint32x2_t r0 = vreinterpret_u32_u64(vshr_n_u64(avg2_u64, 24)); + vst1_lane_u32((uint32_t *)(dst + 0 * stride), r0, 0); + vst1_lane_u32((uint32_t *)(dst + 1 * stride), r1, 0); + vst1_lane_u32((uint32_t *)(dst + 2 * stride), r2, 0); + vst1_lane_u32((uint32_t *)(dst + 3 * stride), r3, 0); +} + +#if !HAVE_NEON_ASM + +void vpx_v_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + int i; + uint32x2_t d0u32 = vdup_n_u32(0); + (void)left; + + d0u32 = vld1_lane_u32((const uint32_t *)above, d0u32, 0); + for (i = 0; i < 4; i++, dst += stride) + vst1_lane_u32((uint32_t *)dst, d0u32, 0); +} + +void vpx_v_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + int i; + uint8x8_t d0u8 = vdup_n_u8(0); + (void)left; + + d0u8 = vld1_u8(above); + for (i = 0; i < 8; i++, dst += stride) + vst1_u8(dst, d0u8); +} + +void vpx_v_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + int i; + uint8x16_t q0u8 = vdupq_n_u8(0); + (void)left; + + q0u8 = vld1q_u8(above); + for (i = 0; i < 16; i++, dst += stride) + vst1q_u8(dst, q0u8); +} + +void vpx_v_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + int i; + uint8x16_t q0u8 = vdupq_n_u8(0); + uint8x16_t q1u8 = vdupq_n_u8(0); + (void)left; + + q0u8 = vld1q_u8(above); + q1u8 = vld1q_u8(above + 16); + for (i = 0; i < 32; i++, dst += stride) { + vst1q_u8(dst, q0u8); + vst1q_u8(dst + 16, q1u8); + } +} + +void vpx_h_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + uint8x8_t d0u8 = vdup_n_u8(0); + uint32x2_t d1u32 = vdup_n_u32(0); + (void)above; + + d1u32 = vld1_lane_u32((const uint32_t *)left, d1u32, 0); + + d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 0); + vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0); + dst += stride; + d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 1); + vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0); + dst += stride; + d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 2); + vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0); + dst += stride; + d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 3); + vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0); +} + +void vpx_h_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + uint8x8_t d0u8 = vdup_n_u8(0); + uint64x1_t d1u64 = vdup_n_u64(0); + (void)above; + + d1u64 = vld1_u64((const uint64_t *)left); + + d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 0); + vst1_u8(dst, d0u8); + dst += stride; + d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 1); + vst1_u8(dst, d0u8); + dst += stride; + d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 2); + vst1_u8(dst, d0u8); + dst += stride; + d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 3); + vst1_u8(dst, d0u8); + dst += stride; + d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 4); + vst1_u8(dst, d0u8); + dst += stride; + d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 5); + vst1_u8(dst, d0u8); + dst += stride; + d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 6); + vst1_u8(dst, d0u8); + dst += stride; + d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 7); + vst1_u8(dst, d0u8); +} + +void vpx_h_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + int j; + uint8x8_t d2u8 = vdup_n_u8(0); + uint8x16_t q0u8 = vdupq_n_u8(0); + uint8x16_t q1u8 = vdupq_n_u8(0); + (void)above; + + q1u8 = vld1q_u8(left); + d2u8 = vget_low_u8(q1u8); + for (j = 0; j < 2; j++, d2u8 = vget_high_u8(q1u8)) { + q0u8 = vdupq_lane_u8(d2u8, 0); + vst1q_u8(dst, q0u8); + dst += stride; + q0u8 = vdupq_lane_u8(d2u8, 1); + vst1q_u8(dst, q0u8); + dst += stride; + q0u8 = vdupq_lane_u8(d2u8, 2); + vst1q_u8(dst, q0u8); + dst += stride; + q0u8 = vdupq_lane_u8(d2u8, 3); + vst1q_u8(dst, q0u8); + dst += stride; + q0u8 = vdupq_lane_u8(d2u8, 4); + vst1q_u8(dst, q0u8); + dst += stride; + q0u8 = vdupq_lane_u8(d2u8, 5); + vst1q_u8(dst, q0u8); + dst += stride; + q0u8 = vdupq_lane_u8(d2u8, 6); + vst1q_u8(dst, q0u8); + dst += stride; + q0u8 = vdupq_lane_u8(d2u8, 7); + vst1q_u8(dst, q0u8); + dst += stride; + } +} + +void vpx_h_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + int j, k; + uint8x8_t d2u8 = vdup_n_u8(0); + uint8x16_t q0u8 = vdupq_n_u8(0); + uint8x16_t q1u8 = vdupq_n_u8(0); + (void)above; + + for (k = 0; k < 2; k++, left += 16) { + q1u8 = vld1q_u8(left); + d2u8 = vget_low_u8(q1u8); + for (j = 0; j < 2; j++, d2u8 = vget_high_u8(q1u8)) { + q0u8 = vdupq_lane_u8(d2u8, 0); + vst1q_u8(dst, q0u8); + vst1q_u8(dst + 16, q0u8); + dst += stride; + q0u8 = vdupq_lane_u8(d2u8, 1); + vst1q_u8(dst, q0u8); + vst1q_u8(dst + 16, q0u8); + dst += stride; + q0u8 = vdupq_lane_u8(d2u8, 2); + vst1q_u8(dst, q0u8); + vst1q_u8(dst + 16, q0u8); + dst += stride; + q0u8 = vdupq_lane_u8(d2u8, 3); + vst1q_u8(dst, q0u8); + vst1q_u8(dst + 16, q0u8); + dst += stride; + q0u8 = vdupq_lane_u8(d2u8, 4); + vst1q_u8(dst, q0u8); + vst1q_u8(dst + 16, q0u8); + dst += stride; + q0u8 = vdupq_lane_u8(d2u8, 5); + vst1q_u8(dst, q0u8); + vst1q_u8(dst + 16, q0u8); + dst += stride; + q0u8 = vdupq_lane_u8(d2u8, 6); + vst1q_u8(dst, q0u8); + vst1q_u8(dst + 16, q0u8); + dst += stride; + q0u8 = vdupq_lane_u8(d2u8, 7); + vst1q_u8(dst, q0u8); + vst1q_u8(dst + 16, q0u8); + dst += stride; + } + } +} + +void vpx_tm_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + int i; + uint16x8_t q1u16, q3u16; + int16x8_t q1s16; + uint8x8_t d0u8 = vdup_n_u8(0); + uint32x2_t d2u32 = vdup_n_u32(0); + + d0u8 = vld1_dup_u8(above - 1); + d2u32 = vld1_lane_u32((const uint32_t *)above, d2u32, 0); + q3u16 = vsubl_u8(vreinterpret_u8_u32(d2u32), d0u8); + for (i = 0; i < 4; i++, dst += stride) { + q1u16 = vdupq_n_u16((uint16_t)left[i]); + q1s16 = vaddq_s16(vreinterpretq_s16_u16(q1u16), + vreinterpretq_s16_u16(q3u16)); + d0u8 = vqmovun_s16(q1s16); + vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0); + } +} + +void vpx_tm_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + int j; + uint16x8_t q0u16, q3u16, q10u16; + int16x8_t q0s16; + uint16x4_t d20u16; + uint8x8_t d0u8, d2u8, d30u8; + + d0u8 = vld1_dup_u8(above - 1); + d30u8 = vld1_u8(left); + d2u8 = vld1_u8(above); + q10u16 = vmovl_u8(d30u8); + q3u16 = vsubl_u8(d2u8, d0u8); + d20u16 = vget_low_u16(q10u16); + for (j = 0; j < 2; j++, d20u16 = vget_high_u16(q10u16)) { + q0u16 = vdupq_lane_u16(d20u16, 0); + q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16), + vreinterpretq_s16_u16(q0u16)); + d0u8 = vqmovun_s16(q0s16); + vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8)); + dst += stride; + q0u16 = vdupq_lane_u16(d20u16, 1); + q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16), + vreinterpretq_s16_u16(q0u16)); + d0u8 = vqmovun_s16(q0s16); + vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8)); + dst += stride; + q0u16 = vdupq_lane_u16(d20u16, 2); + q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16), + vreinterpretq_s16_u16(q0u16)); + d0u8 = vqmovun_s16(q0s16); + vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8)); + dst += stride; + q0u16 = vdupq_lane_u16(d20u16, 3); + q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16), + vreinterpretq_s16_u16(q0u16)); + d0u8 = vqmovun_s16(q0s16); + vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8)); + dst += stride; + } +} + +void vpx_tm_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + int j, k; + uint16x8_t q0u16, q2u16, q3u16, q8u16, q10u16; + uint8x16_t q0u8, q1u8; + int16x8_t q0s16, q1s16, q8s16, q11s16; + uint16x4_t d20u16; + uint8x8_t d2u8, d3u8, d18u8, d22u8, d23u8; + + q0u8 = vld1q_dup_u8(above - 1); + q1u8 = vld1q_u8(above); + q2u16 = vsubl_u8(vget_low_u8(q1u8), vget_low_u8(q0u8)); + q3u16 = vsubl_u8(vget_high_u8(q1u8), vget_high_u8(q0u8)); + for (k = 0; k < 2; k++, left += 8) { + d18u8 = vld1_u8(left); + q10u16 = vmovl_u8(d18u8); + d20u16 = vget_low_u16(q10u16); + for (j = 0; j < 2; j++, d20u16 = vget_high_u16(q10u16)) { + q0u16 = vdupq_lane_u16(d20u16, 0); + q8u16 = vdupq_lane_u16(d20u16, 1); + q1s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), + vreinterpretq_s16_u16(q2u16)); + q0s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), + vreinterpretq_s16_u16(q3u16)); + q11s16 = vaddq_s16(vreinterpretq_s16_u16(q8u16), + vreinterpretq_s16_u16(q2u16)); + q8s16 = vaddq_s16(vreinterpretq_s16_u16(q8u16), + vreinterpretq_s16_u16(q3u16)); + d2u8 = vqmovun_s16(q1s16); + d3u8 = vqmovun_s16(q0s16); + d22u8 = vqmovun_s16(q11s16); + d23u8 = vqmovun_s16(q8s16); + vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d2u8)); + vst1_u64((uint64_t *)(dst + 8), vreinterpret_u64_u8(d3u8)); + dst += stride; + vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d22u8)); + vst1_u64((uint64_t *)(dst + 8), vreinterpret_u64_u8(d23u8)); + dst += stride; + + q0u16 = vdupq_lane_u16(d20u16, 2); + q8u16 = vdupq_lane_u16(d20u16, 3); + q1s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), + vreinterpretq_s16_u16(q2u16)); + q0s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), + vreinterpretq_s16_u16(q3u16)); + q11s16 = vaddq_s16(vreinterpretq_s16_u16(q8u16), + vreinterpretq_s16_u16(q2u16)); + q8s16 = vaddq_s16(vreinterpretq_s16_u16(q8u16), + vreinterpretq_s16_u16(q3u16)); + d2u8 = vqmovun_s16(q1s16); + d3u8 = vqmovun_s16(q0s16); + d22u8 = vqmovun_s16(q11s16); + d23u8 = vqmovun_s16(q8s16); + vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d2u8)); + vst1_u64((uint64_t *)(dst + 8), vreinterpret_u64_u8(d3u8)); + dst += stride; + vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d22u8)); + vst1_u64((uint64_t *)(dst + 8), vreinterpret_u64_u8(d23u8)); + dst += stride; + } + } +} + +void vpx_tm_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + int j, k; + uint16x8_t q0u16, q3u16, q8u16, q9u16, q10u16, q11u16; + uint8x16_t q0u8, q1u8, q2u8; + int16x8_t q12s16, q13s16, q14s16, q15s16; + uint16x4_t d6u16; + uint8x8_t d0u8, d1u8, d2u8, d3u8, d26u8; + + q0u8 = vld1q_dup_u8(above - 1); + q1u8 = vld1q_u8(above); + q2u8 = vld1q_u8(above + 16); + q8u16 = vsubl_u8(vget_low_u8(q1u8), vget_low_u8(q0u8)); + q9u16 = vsubl_u8(vget_high_u8(q1u8), vget_high_u8(q0u8)); + q10u16 = vsubl_u8(vget_low_u8(q2u8), vget_low_u8(q0u8)); + q11u16 = vsubl_u8(vget_high_u8(q2u8), vget_high_u8(q0u8)); + for (k = 0; k < 4; k++, left += 8) { + d26u8 = vld1_u8(left); + q3u16 = vmovl_u8(d26u8); + d6u16 = vget_low_u16(q3u16); + for (j = 0; j < 2; j++, d6u16 = vget_high_u16(q3u16)) { + q0u16 = vdupq_lane_u16(d6u16, 0); + q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), + vreinterpretq_s16_u16(q8u16)); + q13s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), + vreinterpretq_s16_u16(q9u16)); + q14s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), + vreinterpretq_s16_u16(q10u16)); + q15s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), + vreinterpretq_s16_u16(q11u16)); + d0u8 = vqmovun_s16(q12s16); + d1u8 = vqmovun_s16(q13s16); + d2u8 = vqmovun_s16(q14s16); + d3u8 = vqmovun_s16(q15s16); + q0u8 = vcombine_u8(d0u8, d1u8); + q1u8 = vcombine_u8(d2u8, d3u8); + vst1q_u64((uint64_t *)dst, vreinterpretq_u64_u8(q0u8)); + vst1q_u64((uint64_t *)(dst + 16), vreinterpretq_u64_u8(q1u8)); + dst += stride; + + q0u16 = vdupq_lane_u16(d6u16, 1); + q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), + vreinterpretq_s16_u16(q8u16)); + q13s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), + vreinterpretq_s16_u16(q9u16)); + q14s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), + vreinterpretq_s16_u16(q10u16)); + q15s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), + vreinterpretq_s16_u16(q11u16)); + d0u8 = vqmovun_s16(q12s16); + d1u8 = vqmovun_s16(q13s16); + d2u8 = vqmovun_s16(q14s16); + d3u8 = vqmovun_s16(q15s16); + q0u8 = vcombine_u8(d0u8, d1u8); + q1u8 = vcombine_u8(d2u8, d3u8); + vst1q_u64((uint64_t *)dst, vreinterpretq_u64_u8(q0u8)); + vst1q_u64((uint64_t *)(dst + 16), vreinterpretq_u64_u8(q1u8)); + dst += stride; + + q0u16 = vdupq_lane_u16(d6u16, 2); + q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), + vreinterpretq_s16_u16(q8u16)); + q13s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), + vreinterpretq_s16_u16(q9u16)); + q14s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), + vreinterpretq_s16_u16(q10u16)); + q15s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), + vreinterpretq_s16_u16(q11u16)); + d0u8 = vqmovun_s16(q12s16); + d1u8 = vqmovun_s16(q13s16); + d2u8 = vqmovun_s16(q14s16); + d3u8 = vqmovun_s16(q15s16); + q0u8 = vcombine_u8(d0u8, d1u8); + q1u8 = vcombine_u8(d2u8, d3u8); + vst1q_u64((uint64_t *)dst, vreinterpretq_u64_u8(q0u8)); + vst1q_u64((uint64_t *)(dst + 16), vreinterpretq_u64_u8(q1u8)); + dst += stride; + + q0u16 = vdupq_lane_u16(d6u16, 3); + q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), + vreinterpretq_s16_u16(q8u16)); + q13s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), + vreinterpretq_s16_u16(q9u16)); + q14s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), + vreinterpretq_s16_u16(q10u16)); + q15s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), + vreinterpretq_s16_u16(q11u16)); + d0u8 = vqmovun_s16(q12s16); + d1u8 = vqmovun_s16(q13s16); + d2u8 = vqmovun_s16(q14s16); + d3u8 = vqmovun_s16(q15s16); + q0u8 = vcombine_u8(d0u8, d1u8); + q1u8 = vcombine_u8(d2u8, d3u8); + vst1q_u64((uint64_t *)dst, vreinterpretq_u64_u8(q0u8)); + vst1q_u64((uint64_t *)(dst + 16), vreinterpretq_u64_u8(q1u8)); + dst += stride; + } + } +} +#endif // !HAVE_NEON_ASM diff --git a/thirdparty/libvpx/vpx_dsp/arm/intrapred_neon_asm.asm b/thirdparty/libvpx/vpx_dsp/arm/intrapred_neon_asm.asm new file mode 100644 index 0000000000..115790d480 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/arm/intrapred_neon_asm.asm @@ -0,0 +1,630 @@ +; +; Copyright (c) 2014 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + + EXPORT |vpx_v_predictor_4x4_neon| + EXPORT |vpx_v_predictor_8x8_neon| + EXPORT |vpx_v_predictor_16x16_neon| + EXPORT |vpx_v_predictor_32x32_neon| + EXPORT |vpx_h_predictor_4x4_neon| + EXPORT |vpx_h_predictor_8x8_neon| + EXPORT |vpx_h_predictor_16x16_neon| + EXPORT |vpx_h_predictor_32x32_neon| + EXPORT |vpx_tm_predictor_4x4_neon| + EXPORT |vpx_tm_predictor_8x8_neon| + EXPORT |vpx_tm_predictor_16x16_neon| + EXPORT |vpx_tm_predictor_32x32_neon| + ARM + REQUIRE8 + PRESERVE8 + + AREA ||.text||, CODE, READONLY, ALIGN=2 + +;void vpx_v_predictor_4x4_neon(uint8_t *dst, ptrdiff_t y_stride, +; const uint8_t *above, +; const uint8_t *left) +; r0 uint8_t *dst +; r1 ptrdiff_t y_stride +; r2 const uint8_t *above +; r3 const uint8_t *left + +|vpx_v_predictor_4x4_neon| PROC + vld1.32 {d0[0]}, [r2] + vst1.32 {d0[0]}, [r0], r1 + vst1.32 {d0[0]}, [r0], r1 + vst1.32 {d0[0]}, [r0], r1 + vst1.32 {d0[0]}, [r0], r1 + bx lr + ENDP ; |vpx_v_predictor_4x4_neon| + +;void vpx_v_predictor_8x8_neon(uint8_t *dst, ptrdiff_t y_stride, +; const uint8_t *above, +; const uint8_t *left) +; r0 uint8_t *dst +; r1 ptrdiff_t y_stride +; r2 const uint8_t *above +; r3 const uint8_t *left + +|vpx_v_predictor_8x8_neon| PROC + vld1.8 {d0}, [r2] + vst1.8 {d0}, [r0], r1 + vst1.8 {d0}, [r0], r1 + vst1.8 {d0}, [r0], r1 + vst1.8 {d0}, [r0], r1 + vst1.8 {d0}, [r0], r1 + vst1.8 {d0}, [r0], r1 + vst1.8 {d0}, [r0], r1 + vst1.8 {d0}, [r0], r1 + bx lr + ENDP ; |vpx_v_predictor_8x8_neon| + +;void vpx_v_predictor_16x16_neon(uint8_t *dst, ptrdiff_t y_stride, +; const uint8_t *above, +; const uint8_t *left) +; r0 uint8_t *dst +; r1 ptrdiff_t y_stride +; r2 const uint8_t *above +; r3 const uint8_t *left + +|vpx_v_predictor_16x16_neon| PROC + vld1.8 {q0}, [r2] + vst1.8 {q0}, [r0], r1 + vst1.8 {q0}, [r0], r1 + vst1.8 {q0}, [r0], r1 + vst1.8 {q0}, [r0], r1 + vst1.8 {q0}, [r0], r1 + vst1.8 {q0}, [r0], r1 + vst1.8 {q0}, [r0], r1 + vst1.8 {q0}, [r0], r1 + vst1.8 {q0}, [r0], r1 + vst1.8 {q0}, [r0], r1 + vst1.8 {q0}, [r0], r1 + vst1.8 {q0}, [r0], r1 + vst1.8 {q0}, [r0], r1 + vst1.8 {q0}, [r0], r1 + vst1.8 {q0}, [r0], r1 + vst1.8 {q0}, [r0], r1 + bx lr + ENDP ; |vpx_v_predictor_16x16_neon| + +;void vpx_v_predictor_32x32_neon(uint8_t *dst, ptrdiff_t y_stride, +; const uint8_t *above, +; const uint8_t *left) +; r0 uint8_t *dst +; r1 ptrdiff_t y_stride +; r2 const uint8_t *above +; r3 const uint8_t *left + +|vpx_v_predictor_32x32_neon| PROC + vld1.8 {q0, q1}, [r2] + mov r2, #2 +loop_v + vst1.8 {q0, q1}, [r0], r1 + vst1.8 {q0, q1}, [r0], r1 + vst1.8 {q0, q1}, [r0], r1 + vst1.8 {q0, q1}, [r0], r1 + vst1.8 {q0, q1}, [r0], r1 + vst1.8 {q0, q1}, [r0], r1 + vst1.8 {q0, q1}, [r0], r1 + vst1.8 {q0, q1}, [r0], r1 + vst1.8 {q0, q1}, [r0], r1 + vst1.8 {q0, q1}, [r0], r1 + vst1.8 {q0, q1}, [r0], r1 + vst1.8 {q0, q1}, [r0], r1 + vst1.8 {q0, q1}, [r0], r1 + vst1.8 {q0, q1}, [r0], r1 + vst1.8 {q0, q1}, [r0], r1 + vst1.8 {q0, q1}, [r0], r1 + subs r2, r2, #1 + bgt loop_v + bx lr + ENDP ; |vpx_v_predictor_32x32_neon| + +;void vpx_h_predictor_4x4_neon(uint8_t *dst, ptrdiff_t y_stride, +; const uint8_t *above, +; const uint8_t *left) +; r0 uint8_t *dst +; r1 ptrdiff_t y_stride +; r2 const uint8_t *above +; r3 const uint8_t *left + +|vpx_h_predictor_4x4_neon| PROC + vld1.32 {d1[0]}, [r3] + vdup.8 d0, d1[0] + vst1.32 {d0[0]}, [r0], r1 + vdup.8 d0, d1[1] + vst1.32 {d0[0]}, [r0], r1 + vdup.8 d0, d1[2] + vst1.32 {d0[0]}, [r0], r1 + vdup.8 d0, d1[3] + vst1.32 {d0[0]}, [r0], r1 + bx lr + ENDP ; |vpx_h_predictor_4x4_neon| + +;void vpx_h_predictor_8x8_neon(uint8_t *dst, ptrdiff_t y_stride, +; const uint8_t *above, +; const uint8_t *left) +; r0 uint8_t *dst +; r1 ptrdiff_t y_stride +; r2 const uint8_t *above +; r3 const uint8_t *left + +|vpx_h_predictor_8x8_neon| PROC + vld1.64 {d1}, [r3] + vdup.8 d0, d1[0] + vst1.64 {d0}, [r0], r1 + vdup.8 d0, d1[1] + vst1.64 {d0}, [r0], r1 + vdup.8 d0, d1[2] + vst1.64 {d0}, [r0], r1 + vdup.8 d0, d1[3] + vst1.64 {d0}, [r0], r1 + vdup.8 d0, d1[4] + vst1.64 {d0}, [r0], r1 + vdup.8 d0, d1[5] + vst1.64 {d0}, [r0], r1 + vdup.8 d0, d1[6] + vst1.64 {d0}, [r0], r1 + vdup.8 d0, d1[7] + vst1.64 {d0}, [r0], r1 + bx lr + ENDP ; |vpx_h_predictor_8x8_neon| + +;void vpx_h_predictor_16x16_neon(uint8_t *dst, ptrdiff_t y_stride, +; const uint8_t *above, +; const uint8_t *left) +; r0 uint8_t *dst +; r1 ptrdiff_t y_stride +; r2 const uint8_t *above +; r3 const uint8_t *left + +|vpx_h_predictor_16x16_neon| PROC + vld1.8 {q1}, [r3] + vdup.8 q0, d2[0] + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d2[1] + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d2[2] + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d2[3] + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d2[4] + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d2[5] + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d2[6] + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d2[7] + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d3[0] + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d3[1] + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d3[2] + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d3[3] + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d3[4] + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d3[5] + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d3[6] + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d3[7] + vst1.8 {q0}, [r0], r1 + bx lr + ENDP ; |vpx_h_predictor_16x16_neon| + +;void vpx_h_predictor_32x32_neon(uint8_t *dst, ptrdiff_t y_stride, +; const uint8_t *above, +; const uint8_t *left) +; r0 uint8_t *dst +; r1 ptrdiff_t y_stride +; r2 const uint8_t *above +; r3 const uint8_t *left + +|vpx_h_predictor_32x32_neon| PROC + sub r1, r1, #16 + mov r2, #2 +loop_h + vld1.8 {q1}, [r3]! + vdup.8 q0, d2[0] + vst1.8 {q0}, [r0]! + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d2[1] + vst1.8 {q0}, [r0]! + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d2[2] + vst1.8 {q0}, [r0]! + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d2[3] + vst1.8 {q0}, [r0]! + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d2[4] + vst1.8 {q0}, [r0]! + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d2[5] + vst1.8 {q0}, [r0]! + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d2[6] + vst1.8 {q0}, [r0]! + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d2[7] + vst1.8 {q0}, [r0]! + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d3[0] + vst1.8 {q0}, [r0]! + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d3[1] + vst1.8 {q0}, [r0]! + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d3[2] + vst1.8 {q0}, [r0]! + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d3[3] + vst1.8 {q0}, [r0]! + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d3[4] + vst1.8 {q0}, [r0]! + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d3[5] + vst1.8 {q0}, [r0]! + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d3[6] + vst1.8 {q0}, [r0]! + vst1.8 {q0}, [r0], r1 + vdup.8 q0, d3[7] + vst1.8 {q0}, [r0]! + vst1.8 {q0}, [r0], r1 + subs r2, r2, #1 + bgt loop_h + bx lr + ENDP ; |vpx_h_predictor_32x32_neon| + +;void vpx_tm_predictor_4x4_neon (uint8_t *dst, ptrdiff_t y_stride, +; const uint8_t *above, +; const uint8_t *left) +; r0 uint8_t *dst +; r1 ptrdiff_t y_stride +; r2 const uint8_t *above +; r3 const uint8_t *left + +|vpx_tm_predictor_4x4_neon| PROC + ; Load ytop_left = above[-1]; + sub r12, r2, #1 + vld1.u8 {d0[]}, [r12] + + ; Load above 4 pixels + vld1.32 {d2[0]}, [r2] + + ; Compute above - ytop_left + vsubl.u8 q3, d2, d0 + + ; Load left row by row and compute left + (above - ytop_left) + ; 1st row and 2nd row + vld1.u8 {d2[]}, [r3]! + vld1.u8 {d4[]}, [r3]! + vmovl.u8 q1, d2 + vmovl.u8 q2, d4 + vadd.s16 q1, q1, q3 + vadd.s16 q2, q2, q3 + vqmovun.s16 d0, q1 + vqmovun.s16 d1, q2 + vst1.32 {d0[0]}, [r0], r1 + vst1.32 {d1[0]}, [r0], r1 + + ; 3rd row and 4th row + vld1.u8 {d2[]}, [r3]! + vld1.u8 {d4[]}, [r3] + vmovl.u8 q1, d2 + vmovl.u8 q2, d4 + vadd.s16 q1, q1, q3 + vadd.s16 q2, q2, q3 + vqmovun.s16 d0, q1 + vqmovun.s16 d1, q2 + vst1.32 {d0[0]}, [r0], r1 + vst1.32 {d1[0]}, [r0], r1 + bx lr + ENDP ; |vpx_tm_predictor_4x4_neon| + +;void vpx_tm_predictor_8x8_neon (uint8_t *dst, ptrdiff_t y_stride, +; const uint8_t *above, +; const uint8_t *left) +; r0 uint8_t *dst +; r1 ptrdiff_t y_stride +; r2 const uint8_t *above +; r3 const uint8_t *left + +|vpx_tm_predictor_8x8_neon| PROC + ; Load ytop_left = above[-1]; + sub r12, r2, #1 + vld1.8 {d0[]}, [r12] + + ; preload 8 left + vld1.8 {d30}, [r3] + + ; Load above 8 pixels + vld1.64 {d2}, [r2] + + vmovl.u8 q10, d30 + + ; Compute above - ytop_left + vsubl.u8 q3, d2, d0 + + ; Load left row by row and compute left + (above - ytop_left) + ; 1st row and 2nd row + vdup.16 q0, d20[0] + vdup.16 q1, d20[1] + vadd.s16 q0, q3, q0 + vadd.s16 q1, q3, q1 + + ; 3rd row and 4th row + vdup.16 q8, d20[2] + vdup.16 q9, d20[3] + vadd.s16 q8, q3, q8 + vadd.s16 q9, q3, q9 + + vqmovun.s16 d0, q0 + vqmovun.s16 d1, q1 + vqmovun.s16 d2, q8 + vqmovun.s16 d3, q9 + + vst1.64 {d0}, [r0], r1 + vst1.64 {d1}, [r0], r1 + vst1.64 {d2}, [r0], r1 + vst1.64 {d3}, [r0], r1 + + ; 5th row and 6th row + vdup.16 q0, d21[0] + vdup.16 q1, d21[1] + vadd.s16 q0, q3, q0 + vadd.s16 q1, q3, q1 + + ; 7th row and 8th row + vdup.16 q8, d21[2] + vdup.16 q9, d21[3] + vadd.s16 q8, q3, q8 + vadd.s16 q9, q3, q9 + + vqmovun.s16 d0, q0 + vqmovun.s16 d1, q1 + vqmovun.s16 d2, q8 + vqmovun.s16 d3, q9 + + vst1.64 {d0}, [r0], r1 + vst1.64 {d1}, [r0], r1 + vst1.64 {d2}, [r0], r1 + vst1.64 {d3}, [r0], r1 + + bx lr + ENDP ; |vpx_tm_predictor_8x8_neon| + +;void vpx_tm_predictor_16x16_neon (uint8_t *dst, ptrdiff_t y_stride, +; const uint8_t *above, +; const uint8_t *left) +; r0 uint8_t *dst +; r1 ptrdiff_t y_stride +; r2 const uint8_t *above +; r3 const uint8_t *left + +|vpx_tm_predictor_16x16_neon| PROC + ; Load ytop_left = above[-1]; + sub r12, r2, #1 + vld1.8 {d0[]}, [r12] + + ; Load above 8 pixels + vld1.8 {q1}, [r2] + + ; preload 8 left into r12 + vld1.8 {d18}, [r3]! + + ; Compute above - ytop_left + vsubl.u8 q2, d2, d0 + vsubl.u8 q3, d3, d0 + + vmovl.u8 q10, d18 + + ; Load left row by row and compute left + (above - ytop_left) + ; Process 8 rows in each single loop and loop 2 times to process 16 rows. + mov r2, #2 + +loop_16x16_neon + ; Process two rows. + vdup.16 q0, d20[0] + vdup.16 q8, d20[1] + vadd.s16 q1, q0, q2 + vadd.s16 q0, q0, q3 + vadd.s16 q11, q8, q2 + vadd.s16 q8, q8, q3 + vqmovun.s16 d2, q1 + vqmovun.s16 d3, q0 + vqmovun.s16 d22, q11 + vqmovun.s16 d23, q8 + vdup.16 q0, d20[2] ; proload next 2 rows data + vdup.16 q8, d20[3] + vst1.64 {d2,d3}, [r0], r1 + vst1.64 {d22,d23}, [r0], r1 + + ; Process two rows. + vadd.s16 q1, q0, q2 + vadd.s16 q0, q0, q3 + vadd.s16 q11, q8, q2 + vadd.s16 q8, q8, q3 + vqmovun.s16 d2, q1 + vqmovun.s16 d3, q0 + vqmovun.s16 d22, q11 + vqmovun.s16 d23, q8 + vdup.16 q0, d21[0] ; proload next 2 rows data + vdup.16 q8, d21[1] + vst1.64 {d2,d3}, [r0], r1 + vst1.64 {d22,d23}, [r0], r1 + + vadd.s16 q1, q0, q2 + vadd.s16 q0, q0, q3 + vadd.s16 q11, q8, q2 + vadd.s16 q8, q8, q3 + vqmovun.s16 d2, q1 + vqmovun.s16 d3, q0 + vqmovun.s16 d22, q11 + vqmovun.s16 d23, q8 + vdup.16 q0, d21[2] ; proload next 2 rows data + vdup.16 q8, d21[3] + vst1.64 {d2,d3}, [r0], r1 + vst1.64 {d22,d23}, [r0], r1 + + + vadd.s16 q1, q0, q2 + vadd.s16 q0, q0, q3 + vadd.s16 q11, q8, q2 + vadd.s16 q8, q8, q3 + vqmovun.s16 d2, q1 + vqmovun.s16 d3, q0 + vqmovun.s16 d22, q11 + vqmovun.s16 d23, q8 + vld1.8 {d18}, [r3]! ; preload 8 left into r12 + vmovl.u8 q10, d18 + vst1.64 {d2,d3}, [r0], r1 + vst1.64 {d22,d23}, [r0], r1 + + subs r2, r2, #1 + bgt loop_16x16_neon + + bx lr + ENDP ; |vpx_tm_predictor_16x16_neon| + +;void vpx_tm_predictor_32x32_neon (uint8_t *dst, ptrdiff_t y_stride, +; const uint8_t *above, +; const uint8_t *left) +; r0 uint8_t *dst +; r1 ptrdiff_t y_stride +; r2 const uint8_t *above +; r3 const uint8_t *left + +|vpx_tm_predictor_32x32_neon| PROC + ; Load ytop_left = above[-1]; + sub r12, r2, #1 + vld1.8 {d0[]}, [r12] + + ; Load above 32 pixels + vld1.8 {q1}, [r2]! + vld1.8 {q2}, [r2] + + ; preload 8 left pixels + vld1.8 {d26}, [r3]! + + ; Compute above - ytop_left + vsubl.u8 q8, d2, d0 + vsubl.u8 q9, d3, d0 + vsubl.u8 q10, d4, d0 + vsubl.u8 q11, d5, d0 + + vmovl.u8 q3, d26 + + ; Load left row by row and compute left + (above - ytop_left) + ; Process 8 rows in each single loop and loop 4 times to process 32 rows. + mov r2, #4 + +loop_32x32_neon + ; Process two rows. + vdup.16 q0, d6[0] + vdup.16 q2, d6[1] + vadd.s16 q12, q0, q8 + vadd.s16 q13, q0, q9 + vadd.s16 q14, q0, q10 + vadd.s16 q15, q0, q11 + vqmovun.s16 d0, q12 + vqmovun.s16 d1, q13 + vadd.s16 q12, q2, q8 + vadd.s16 q13, q2, q9 + vqmovun.s16 d2, q14 + vqmovun.s16 d3, q15 + vadd.s16 q14, q2, q10 + vadd.s16 q15, q2, q11 + vst1.64 {d0-d3}, [r0], r1 + vqmovun.s16 d24, q12 + vqmovun.s16 d25, q13 + vqmovun.s16 d26, q14 + vqmovun.s16 d27, q15 + vdup.16 q1, d6[2] + vdup.16 q2, d6[3] + vst1.64 {d24-d27}, [r0], r1 + + ; Process two rows. + vadd.s16 q12, q1, q8 + vadd.s16 q13, q1, q9 + vadd.s16 q14, q1, q10 + vadd.s16 q15, q1, q11 + vqmovun.s16 d0, q12 + vqmovun.s16 d1, q13 + vadd.s16 q12, q2, q8 + vadd.s16 q13, q2, q9 + vqmovun.s16 d2, q14 + vqmovun.s16 d3, q15 + vadd.s16 q14, q2, q10 + vadd.s16 q15, q2, q11 + vst1.64 {d0-d3}, [r0], r1 + vqmovun.s16 d24, q12 + vqmovun.s16 d25, q13 + vqmovun.s16 d26, q14 + vqmovun.s16 d27, q15 + vdup.16 q0, d7[0] + vdup.16 q2, d7[1] + vst1.64 {d24-d27}, [r0], r1 + + ; Process two rows. + vadd.s16 q12, q0, q8 + vadd.s16 q13, q0, q9 + vadd.s16 q14, q0, q10 + vadd.s16 q15, q0, q11 + vqmovun.s16 d0, q12 + vqmovun.s16 d1, q13 + vadd.s16 q12, q2, q8 + vadd.s16 q13, q2, q9 + vqmovun.s16 d2, q14 + vqmovun.s16 d3, q15 + vadd.s16 q14, q2, q10 + vadd.s16 q15, q2, q11 + vst1.64 {d0-d3}, [r0], r1 + vqmovun.s16 d24, q12 + vqmovun.s16 d25, q13 + vqmovun.s16 d26, q14 + vqmovun.s16 d27, q15 + vdup.16 q0, d7[2] + vdup.16 q2, d7[3] + vst1.64 {d24-d27}, [r0], r1 + + ; Process two rows. + vadd.s16 q12, q0, q8 + vadd.s16 q13, q0, q9 + vadd.s16 q14, q0, q10 + vadd.s16 q15, q0, q11 + vqmovun.s16 d0, q12 + vqmovun.s16 d1, q13 + vadd.s16 q12, q2, q8 + vadd.s16 q13, q2, q9 + vqmovun.s16 d2, q14 + vqmovun.s16 d3, q15 + vadd.s16 q14, q2, q10 + vadd.s16 q15, q2, q11 + vst1.64 {d0-d3}, [r0], r1 + vqmovun.s16 d24, q12 + vqmovun.s16 d25, q13 + vld1.8 {d0}, [r3]! ; preload 8 left pixels + vqmovun.s16 d26, q14 + vqmovun.s16 d27, q15 + vmovl.u8 q3, d0 + vst1.64 {d24-d27}, [r0], r1 + + subs r2, r2, #1 + bgt loop_32x32_neon + + bx lr + ENDP ; |vpx_tm_predictor_32x32_neon| + + END diff --git a/thirdparty/libvpx/vpx_dsp/arm/loopfilter_16_neon.c b/thirdparty/libvpx/vpx_dsp/arm/loopfilter_16_neon.c new file mode 100644 index 0000000000..d24e6adc8a --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/arm/loopfilter_16_neon.c @@ -0,0 +1,179 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +#include "./vpx_dsp_rtcd.h" +#include "./vpx_config.h" +#include "vpx/vpx_integer.h" + +static INLINE void loop_filter_neon_16( + uint8x16_t qblimit, // blimit + uint8x16_t qlimit, // limit + uint8x16_t qthresh, // thresh + uint8x16_t q3, // p3 + uint8x16_t q4, // p2 + uint8x16_t q5, // p1 + uint8x16_t q6, // p0 + uint8x16_t q7, // q0 + uint8x16_t q8, // q1 + uint8x16_t q9, // q2 + uint8x16_t q10, // q3 + uint8x16_t *q5r, // p1 + uint8x16_t *q6r, // p0 + uint8x16_t *q7r, // q0 + uint8x16_t *q8r) { // q1 + uint8x16_t q1u8, q2u8, q11u8, q12u8, q13u8, q14u8, q15u8; + int16x8_t q2s16, q11s16; + uint16x8_t q4u16; + int8x16_t q0s8, q1s8, q2s8, q11s8, q12s8, q13s8; + int8x8_t d2s8, d3s8; + + q11u8 = vabdq_u8(q3, q4); + q12u8 = vabdq_u8(q4, q5); + q13u8 = vabdq_u8(q5, q6); + q14u8 = vabdq_u8(q8, q7); + q3 = vabdq_u8(q9, q8); + q4 = vabdq_u8(q10, q9); + + q11u8 = vmaxq_u8(q11u8, q12u8); + q12u8 = vmaxq_u8(q13u8, q14u8); + q3 = vmaxq_u8(q3, q4); + q15u8 = vmaxq_u8(q11u8, q12u8); + + q9 = vabdq_u8(q6, q7); + + // vp8_hevmask + q13u8 = vcgtq_u8(q13u8, qthresh); + q14u8 = vcgtq_u8(q14u8, qthresh); + q15u8 = vmaxq_u8(q15u8, q3); + + q2u8 = vabdq_u8(q5, q8); + q9 = vqaddq_u8(q9, q9); + + q15u8 = vcgeq_u8(qlimit, q15u8); + + // vp8_filter() function + // convert to signed + q10 = vdupq_n_u8(0x80); + q8 = veorq_u8(q8, q10); + q7 = veorq_u8(q7, q10); + q6 = veorq_u8(q6, q10); + q5 = veorq_u8(q5, q10); + + q2u8 = vshrq_n_u8(q2u8, 1); + q9 = vqaddq_u8(q9, q2u8); + + q2s16 = vsubl_s8(vget_low_s8(vreinterpretq_s8_u8(q7)), + vget_low_s8(vreinterpretq_s8_u8(q6))); + q11s16 = vsubl_s8(vget_high_s8(vreinterpretq_s8_u8(q7)), + vget_high_s8(vreinterpretq_s8_u8(q6))); + + q9 = vcgeq_u8(qblimit, q9); + + q1s8 = vqsubq_s8(vreinterpretq_s8_u8(q5), + vreinterpretq_s8_u8(q8)); + + q14u8 = vorrq_u8(q13u8, q14u8); + + q4u16 = vdupq_n_u16(3); + q2s16 = vmulq_s16(q2s16, vreinterpretq_s16_u16(q4u16)); + q11s16 = vmulq_s16(q11s16, vreinterpretq_s16_u16(q4u16)); + + q1u8 = vandq_u8(vreinterpretq_u8_s8(q1s8), q14u8); + q15u8 = vandq_u8(q15u8, q9); + + q1s8 = vreinterpretq_s8_u8(q1u8); + q2s16 = vaddw_s8(q2s16, vget_low_s8(q1s8)); + q11s16 = vaddw_s8(q11s16, vget_high_s8(q1s8)); + + q4 = vdupq_n_u8(3); + q9 = vdupq_n_u8(4); + // vp8_filter = clamp(vp8_filter + 3 * ( qs0 - ps0)) + d2s8 = vqmovn_s16(q2s16); + d3s8 = vqmovn_s16(q11s16); + q1s8 = vcombine_s8(d2s8, d3s8); + q1u8 = vandq_u8(vreinterpretq_u8_s8(q1s8), q15u8); + q1s8 = vreinterpretq_s8_u8(q1u8); + + q2s8 = vqaddq_s8(q1s8, vreinterpretq_s8_u8(q4)); + q1s8 = vqaddq_s8(q1s8, vreinterpretq_s8_u8(q9)); + q2s8 = vshrq_n_s8(q2s8, 3); + q1s8 = vshrq_n_s8(q1s8, 3); + + q11s8 = vqaddq_s8(vreinterpretq_s8_u8(q6), q2s8); + q0s8 = vqsubq_s8(vreinterpretq_s8_u8(q7), q1s8); + + q1s8 = vrshrq_n_s8(q1s8, 1); + q1s8 = vbicq_s8(q1s8, vreinterpretq_s8_u8(q14u8)); + + q13s8 = vqaddq_s8(vreinterpretq_s8_u8(q5), q1s8); + q12s8 = vqsubq_s8(vreinterpretq_s8_u8(q8), q1s8); + + *q8r = veorq_u8(vreinterpretq_u8_s8(q12s8), q10); + *q7r = veorq_u8(vreinterpretq_u8_s8(q0s8), q10); + *q6r = veorq_u8(vreinterpretq_u8_s8(q11s8), q10); + *q5r = veorq_u8(vreinterpretq_u8_s8(q13s8), q10); + return; +} + +void vpx_lpf_horizontal_4_dual_neon(uint8_t *s, int p /* pitch */, + const uint8_t *blimit0, + const uint8_t *limit0, + const uint8_t *thresh0, + const uint8_t *blimit1, + const uint8_t *limit1, + const uint8_t *thresh1) { + uint8x8_t dblimit0, dlimit0, dthresh0, dblimit1, dlimit1, dthresh1; + uint8x16_t qblimit, qlimit, qthresh; + uint8x16_t q3u8, q4u8, q5u8, q6u8, q7u8, q8u8, q9u8, q10u8; + + dblimit0 = vld1_u8(blimit0); + dlimit0 = vld1_u8(limit0); + dthresh0 = vld1_u8(thresh0); + dblimit1 = vld1_u8(blimit1); + dlimit1 = vld1_u8(limit1); + dthresh1 = vld1_u8(thresh1); + qblimit = vcombine_u8(dblimit0, dblimit1); + qlimit = vcombine_u8(dlimit0, dlimit1); + qthresh = vcombine_u8(dthresh0, dthresh1); + + s -= (p << 2); + + q3u8 = vld1q_u8(s); + s += p; + q4u8 = vld1q_u8(s); + s += p; + q5u8 = vld1q_u8(s); + s += p; + q6u8 = vld1q_u8(s); + s += p; + q7u8 = vld1q_u8(s); + s += p; + q8u8 = vld1q_u8(s); + s += p; + q9u8 = vld1q_u8(s); + s += p; + q10u8 = vld1q_u8(s); + + loop_filter_neon_16(qblimit, qlimit, qthresh, + q3u8, q4u8, q5u8, q6u8, q7u8, q8u8, q9u8, q10u8, + &q5u8, &q6u8, &q7u8, &q8u8); + + s -= (p * 5); + vst1q_u8(s, q5u8); + s += p; + vst1q_u8(s, q6u8); + s += p; + vst1q_u8(s, q7u8); + s += p; + vst1q_u8(s, q8u8); + return; +} diff --git a/thirdparty/libvpx/vpx_dsp/arm/loopfilter_4_neon.c b/thirdparty/libvpx/vpx_dsp/arm/loopfilter_4_neon.c new file mode 100644 index 0000000000..7f3ee70b94 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/arm/loopfilter_4_neon.c @@ -0,0 +1,266 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +#include "./vpx_dsp_rtcd.h" + +static INLINE void loop_filter_neon( + uint8x8_t dblimit, // flimit + uint8x8_t dlimit, // limit + uint8x8_t dthresh, // thresh + uint8x8_t d3u8, // p3 + uint8x8_t d4u8, // p2 + uint8x8_t d5u8, // p1 + uint8x8_t d6u8, // p0 + uint8x8_t d7u8, // q0 + uint8x8_t d16u8, // q1 + uint8x8_t d17u8, // q2 + uint8x8_t d18u8, // q3 + uint8x8_t *d4ru8, // p1 + uint8x8_t *d5ru8, // p0 + uint8x8_t *d6ru8, // q0 + uint8x8_t *d7ru8) { // q1 + uint8x8_t d19u8, d20u8, d21u8, d22u8, d23u8, d27u8, d28u8; + int16x8_t q12s16; + int8x8_t d19s8, d20s8, d21s8, d26s8, d27s8, d28s8; + + d19u8 = vabd_u8(d3u8, d4u8); + d20u8 = vabd_u8(d4u8, d5u8); + d21u8 = vabd_u8(d5u8, d6u8); + d22u8 = vabd_u8(d16u8, d7u8); + d3u8 = vabd_u8(d17u8, d16u8); + d4u8 = vabd_u8(d18u8, d17u8); + + d19u8 = vmax_u8(d19u8, d20u8); + d20u8 = vmax_u8(d21u8, d22u8); + d3u8 = vmax_u8(d3u8, d4u8); + d23u8 = vmax_u8(d19u8, d20u8); + + d17u8 = vabd_u8(d6u8, d7u8); + + d21u8 = vcgt_u8(d21u8, dthresh); + d22u8 = vcgt_u8(d22u8, dthresh); + d23u8 = vmax_u8(d23u8, d3u8); + + d28u8 = vabd_u8(d5u8, d16u8); + d17u8 = vqadd_u8(d17u8, d17u8); + + d23u8 = vcge_u8(dlimit, d23u8); + + d18u8 = vdup_n_u8(0x80); + d5u8 = veor_u8(d5u8, d18u8); + d6u8 = veor_u8(d6u8, d18u8); + d7u8 = veor_u8(d7u8, d18u8); + d16u8 = veor_u8(d16u8, d18u8); + + d28u8 = vshr_n_u8(d28u8, 1); + d17u8 = vqadd_u8(d17u8, d28u8); + + d19u8 = vdup_n_u8(3); + + d28s8 = vsub_s8(vreinterpret_s8_u8(d7u8), + vreinterpret_s8_u8(d6u8)); + + d17u8 = vcge_u8(dblimit, d17u8); + + d27s8 = vqsub_s8(vreinterpret_s8_u8(d5u8), + vreinterpret_s8_u8(d16u8)); + + d22u8 = vorr_u8(d21u8, d22u8); + + q12s16 = vmull_s8(d28s8, vreinterpret_s8_u8(d19u8)); + + d27u8 = vand_u8(vreinterpret_u8_s8(d27s8), d22u8); + d23u8 = vand_u8(d23u8, d17u8); + + q12s16 = vaddw_s8(q12s16, vreinterpret_s8_u8(d27u8)); + + d17u8 = vdup_n_u8(4); + + d27s8 = vqmovn_s16(q12s16); + d27u8 = vand_u8(vreinterpret_u8_s8(d27s8), d23u8); + d27s8 = vreinterpret_s8_u8(d27u8); + + d28s8 = vqadd_s8(d27s8, vreinterpret_s8_u8(d19u8)); + d27s8 = vqadd_s8(d27s8, vreinterpret_s8_u8(d17u8)); + d28s8 = vshr_n_s8(d28s8, 3); + d27s8 = vshr_n_s8(d27s8, 3); + + d19s8 = vqadd_s8(vreinterpret_s8_u8(d6u8), d28s8); + d26s8 = vqsub_s8(vreinterpret_s8_u8(d7u8), d27s8); + + d27s8 = vrshr_n_s8(d27s8, 1); + d27s8 = vbic_s8(d27s8, vreinterpret_s8_u8(d22u8)); + + d21s8 = vqadd_s8(vreinterpret_s8_u8(d5u8), d27s8); + d20s8 = vqsub_s8(vreinterpret_s8_u8(d16u8), d27s8); + + *d4ru8 = veor_u8(vreinterpret_u8_s8(d21s8), d18u8); + *d5ru8 = veor_u8(vreinterpret_u8_s8(d19s8), d18u8); + *d6ru8 = veor_u8(vreinterpret_u8_s8(d26s8), d18u8); + *d7ru8 = veor_u8(vreinterpret_u8_s8(d20s8), d18u8); + return; +} + +void vpx_lpf_horizontal_4_neon( + uint8_t *src, + int pitch, + const uint8_t *blimit, + const uint8_t *limit, + const uint8_t *thresh) { + int i; + uint8_t *s, *psrc; + uint8x8_t dblimit, dlimit, dthresh; + uint8x8_t d3u8, d4u8, d5u8, d6u8, d7u8, d16u8, d17u8, d18u8; + + dblimit = vld1_u8(blimit); + dlimit = vld1_u8(limit); + dthresh = vld1_u8(thresh); + + psrc = src - (pitch << 2); + for (i = 0; i < 1; i++) { + s = psrc + i * 8; + + d3u8 = vld1_u8(s); + s += pitch; + d4u8 = vld1_u8(s); + s += pitch; + d5u8 = vld1_u8(s); + s += pitch; + d6u8 = vld1_u8(s); + s += pitch; + d7u8 = vld1_u8(s); + s += pitch; + d16u8 = vld1_u8(s); + s += pitch; + d17u8 = vld1_u8(s); + s += pitch; + d18u8 = vld1_u8(s); + + loop_filter_neon(dblimit, dlimit, dthresh, + d3u8, d4u8, d5u8, d6u8, d7u8, d16u8, d17u8, d18u8, + &d4u8, &d5u8, &d6u8, &d7u8); + + s -= (pitch * 5); + vst1_u8(s, d4u8); + s += pitch; + vst1_u8(s, d5u8); + s += pitch; + vst1_u8(s, d6u8); + s += pitch; + vst1_u8(s, d7u8); + } + return; +} + +void vpx_lpf_vertical_4_neon( + uint8_t *src, + int pitch, + const uint8_t *blimit, + const uint8_t *limit, + const uint8_t *thresh) { + int i, pitch8; + uint8_t *s; + uint8x8_t dblimit, dlimit, dthresh; + uint8x8_t d3u8, d4u8, d5u8, d6u8, d7u8, d16u8, d17u8, d18u8; + uint32x2x2_t d2tmp0, d2tmp1, d2tmp2, d2tmp3; + uint16x4x2_t d2tmp4, d2tmp5, d2tmp6, d2tmp7; + uint8x8x2_t d2tmp8, d2tmp9, d2tmp10, d2tmp11; + uint8x8x4_t d4Result; + + dblimit = vld1_u8(blimit); + dlimit = vld1_u8(limit); + dthresh = vld1_u8(thresh); + + pitch8 = pitch * 8; + for (i = 0; i < 1; i++, src += pitch8) { + s = src - (i + 1) * 4; + + d3u8 = vld1_u8(s); + s += pitch; + d4u8 = vld1_u8(s); + s += pitch; + d5u8 = vld1_u8(s); + s += pitch; + d6u8 = vld1_u8(s); + s += pitch; + d7u8 = vld1_u8(s); + s += pitch; + d16u8 = vld1_u8(s); + s += pitch; + d17u8 = vld1_u8(s); + s += pitch; + d18u8 = vld1_u8(s); + + d2tmp0 = vtrn_u32(vreinterpret_u32_u8(d3u8), + vreinterpret_u32_u8(d7u8)); + d2tmp1 = vtrn_u32(vreinterpret_u32_u8(d4u8), + vreinterpret_u32_u8(d16u8)); + d2tmp2 = vtrn_u32(vreinterpret_u32_u8(d5u8), + vreinterpret_u32_u8(d17u8)); + d2tmp3 = vtrn_u32(vreinterpret_u32_u8(d6u8), + vreinterpret_u32_u8(d18u8)); + + d2tmp4 = vtrn_u16(vreinterpret_u16_u32(d2tmp0.val[0]), + vreinterpret_u16_u32(d2tmp2.val[0])); + d2tmp5 = vtrn_u16(vreinterpret_u16_u32(d2tmp1.val[0]), + vreinterpret_u16_u32(d2tmp3.val[0])); + d2tmp6 = vtrn_u16(vreinterpret_u16_u32(d2tmp0.val[1]), + vreinterpret_u16_u32(d2tmp2.val[1])); + d2tmp7 = vtrn_u16(vreinterpret_u16_u32(d2tmp1.val[1]), + vreinterpret_u16_u32(d2tmp3.val[1])); + + d2tmp8 = vtrn_u8(vreinterpret_u8_u16(d2tmp4.val[0]), + vreinterpret_u8_u16(d2tmp5.val[0])); + d2tmp9 = vtrn_u8(vreinterpret_u8_u16(d2tmp4.val[1]), + vreinterpret_u8_u16(d2tmp5.val[1])); + d2tmp10 = vtrn_u8(vreinterpret_u8_u16(d2tmp6.val[0]), + vreinterpret_u8_u16(d2tmp7.val[0])); + d2tmp11 = vtrn_u8(vreinterpret_u8_u16(d2tmp6.val[1]), + vreinterpret_u8_u16(d2tmp7.val[1])); + + d3u8 = d2tmp8.val[0]; + d4u8 = d2tmp8.val[1]; + d5u8 = d2tmp9.val[0]; + d6u8 = d2tmp9.val[1]; + d7u8 = d2tmp10.val[0]; + d16u8 = d2tmp10.val[1]; + d17u8 = d2tmp11.val[0]; + d18u8 = d2tmp11.val[1]; + + loop_filter_neon(dblimit, dlimit, dthresh, + d3u8, d4u8, d5u8, d6u8, d7u8, d16u8, d17u8, d18u8, + &d4u8, &d5u8, &d6u8, &d7u8); + + d4Result.val[0] = d4u8; + d4Result.val[1] = d5u8; + d4Result.val[2] = d6u8; + d4Result.val[3] = d7u8; + + src -= 2; + vst4_lane_u8(src, d4Result, 0); + src += pitch; + vst4_lane_u8(src, d4Result, 1); + src += pitch; + vst4_lane_u8(src, d4Result, 2); + src += pitch; + vst4_lane_u8(src, d4Result, 3); + src += pitch; + vst4_lane_u8(src, d4Result, 4); + src += pitch; + vst4_lane_u8(src, d4Result, 5); + src += pitch; + vst4_lane_u8(src, d4Result, 6); + src += pitch; + vst4_lane_u8(src, d4Result, 7); + } + return; +} diff --git a/thirdparty/libvpx/vpx_dsp/arm/loopfilter_8_neon.c b/thirdparty/libvpx/vpx_dsp/arm/loopfilter_8_neon.c new file mode 100644 index 0000000000..ec3757380d --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/arm/loopfilter_8_neon.c @@ -0,0 +1,445 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +#include "./vpx_dsp_rtcd.h" + +static INLINE void mbloop_filter_neon( + uint8x8_t dblimit, // mblimit + uint8x8_t dlimit, // limit + uint8x8_t dthresh, // thresh + uint8x8_t d3u8, // p2 + uint8x8_t d4u8, // p2 + uint8x8_t d5u8, // p1 + uint8x8_t d6u8, // p0 + uint8x8_t d7u8, // q0 + uint8x8_t d16u8, // q1 + uint8x8_t d17u8, // q2 + uint8x8_t d18u8, // q3 + uint8x8_t *d0ru8, // p1 + uint8x8_t *d1ru8, // p1 + uint8x8_t *d2ru8, // p0 + uint8x8_t *d3ru8, // q0 + uint8x8_t *d4ru8, // q1 + uint8x8_t *d5ru8) { // q1 + uint32_t flat; + uint8x8_t d0u8, d1u8, d2u8, d19u8, d20u8, d21u8, d22u8, d23u8, d24u8; + uint8x8_t d25u8, d26u8, d27u8, d28u8, d29u8, d30u8, d31u8; + int16x8_t q15s16; + uint16x8_t q10u16, q14u16; + int8x8_t d21s8, d24s8, d25s8, d26s8, d28s8, d29s8, d30s8; + + d19u8 = vabd_u8(d3u8, d4u8); + d20u8 = vabd_u8(d4u8, d5u8); + d21u8 = vabd_u8(d5u8, d6u8); + d22u8 = vabd_u8(d16u8, d7u8); + d23u8 = vabd_u8(d17u8, d16u8); + d24u8 = vabd_u8(d18u8, d17u8); + + d19u8 = vmax_u8(d19u8, d20u8); + d20u8 = vmax_u8(d21u8, d22u8); + + d25u8 = vabd_u8(d6u8, d4u8); + + d23u8 = vmax_u8(d23u8, d24u8); + + d26u8 = vabd_u8(d7u8, d17u8); + + d19u8 = vmax_u8(d19u8, d20u8); + + d24u8 = vabd_u8(d6u8, d7u8); + d27u8 = vabd_u8(d3u8, d6u8); + d28u8 = vabd_u8(d18u8, d7u8); + + d19u8 = vmax_u8(d19u8, d23u8); + + d23u8 = vabd_u8(d5u8, d16u8); + d24u8 = vqadd_u8(d24u8, d24u8); + + + d19u8 = vcge_u8(dlimit, d19u8); + + + d25u8 = vmax_u8(d25u8, d26u8); + d26u8 = vmax_u8(d27u8, d28u8); + + d23u8 = vshr_n_u8(d23u8, 1); + + d25u8 = vmax_u8(d25u8, d26u8); + + d24u8 = vqadd_u8(d24u8, d23u8); + + d20u8 = vmax_u8(d20u8, d25u8); + + d23u8 = vdup_n_u8(1); + d24u8 = vcge_u8(dblimit, d24u8); + + d21u8 = vcgt_u8(d21u8, dthresh); + + d20u8 = vcge_u8(d23u8, d20u8); + + d19u8 = vand_u8(d19u8, d24u8); + + d23u8 = vcgt_u8(d22u8, dthresh); + + d20u8 = vand_u8(d20u8, d19u8); + + d22u8 = vdup_n_u8(0x80); + + d23u8 = vorr_u8(d21u8, d23u8); + + q10u16 = vcombine_u16(vreinterpret_u16_u8(d20u8), + vreinterpret_u16_u8(d21u8)); + + d30u8 = vshrn_n_u16(q10u16, 4); + flat = vget_lane_u32(vreinterpret_u32_u8(d30u8), 0); + + if (flat == 0xffffffff) { // Check for all 1's, power_branch_only + d27u8 = vdup_n_u8(3); + d21u8 = vdup_n_u8(2); + q14u16 = vaddl_u8(d6u8, d7u8); + q14u16 = vmlal_u8(q14u16, d3u8, d27u8); + q14u16 = vmlal_u8(q14u16, d4u8, d21u8); + q14u16 = vaddw_u8(q14u16, d5u8); + *d0ru8 = vqrshrn_n_u16(q14u16, 3); + + q14u16 = vsubw_u8(q14u16, d3u8); + q14u16 = vsubw_u8(q14u16, d4u8); + q14u16 = vaddw_u8(q14u16, d5u8); + q14u16 = vaddw_u8(q14u16, d16u8); + *d1ru8 = vqrshrn_n_u16(q14u16, 3); + + q14u16 = vsubw_u8(q14u16, d3u8); + q14u16 = vsubw_u8(q14u16, d5u8); + q14u16 = vaddw_u8(q14u16, d6u8); + q14u16 = vaddw_u8(q14u16, d17u8); + *d2ru8 = vqrshrn_n_u16(q14u16, 3); + + q14u16 = vsubw_u8(q14u16, d3u8); + q14u16 = vsubw_u8(q14u16, d6u8); + q14u16 = vaddw_u8(q14u16, d7u8); + q14u16 = vaddw_u8(q14u16, d18u8); + *d3ru8 = vqrshrn_n_u16(q14u16, 3); + + q14u16 = vsubw_u8(q14u16, d4u8); + q14u16 = vsubw_u8(q14u16, d7u8); + q14u16 = vaddw_u8(q14u16, d16u8); + q14u16 = vaddw_u8(q14u16, d18u8); + *d4ru8 = vqrshrn_n_u16(q14u16, 3); + + q14u16 = vsubw_u8(q14u16, d5u8); + q14u16 = vsubw_u8(q14u16, d16u8); + q14u16 = vaddw_u8(q14u16, d17u8); + q14u16 = vaddw_u8(q14u16, d18u8); + *d5ru8 = vqrshrn_n_u16(q14u16, 3); + } else { + d21u8 = veor_u8(d7u8, d22u8); + d24u8 = veor_u8(d6u8, d22u8); + d25u8 = veor_u8(d5u8, d22u8); + d26u8 = veor_u8(d16u8, d22u8); + + d27u8 = vdup_n_u8(3); + + d28s8 = vsub_s8(vreinterpret_s8_u8(d21u8), vreinterpret_s8_u8(d24u8)); + d29s8 = vqsub_s8(vreinterpret_s8_u8(d25u8), vreinterpret_s8_u8(d26u8)); + + q15s16 = vmull_s8(d28s8, vreinterpret_s8_u8(d27u8)); + + d29s8 = vand_s8(d29s8, vreinterpret_s8_u8(d23u8)); + + q15s16 = vaddw_s8(q15s16, d29s8); + + d29u8 = vdup_n_u8(4); + + d28s8 = vqmovn_s16(q15s16); + + d28s8 = vand_s8(d28s8, vreinterpret_s8_u8(d19u8)); + + d30s8 = vqadd_s8(d28s8, vreinterpret_s8_u8(d27u8)); + d29s8 = vqadd_s8(d28s8, vreinterpret_s8_u8(d29u8)); + d30s8 = vshr_n_s8(d30s8, 3); + d29s8 = vshr_n_s8(d29s8, 3); + + d24s8 = vqadd_s8(vreinterpret_s8_u8(d24u8), d30s8); + d21s8 = vqsub_s8(vreinterpret_s8_u8(d21u8), d29s8); + + d29s8 = vrshr_n_s8(d29s8, 1); + d29s8 = vbic_s8(d29s8, vreinterpret_s8_u8(d23u8)); + + d25s8 = vqadd_s8(vreinterpret_s8_u8(d25u8), d29s8); + d26s8 = vqsub_s8(vreinterpret_s8_u8(d26u8), d29s8); + + if (flat == 0) { // filter_branch_only + *d0ru8 = d4u8; + *d1ru8 = veor_u8(vreinterpret_u8_s8(d25s8), d22u8); + *d2ru8 = veor_u8(vreinterpret_u8_s8(d24s8), d22u8); + *d3ru8 = veor_u8(vreinterpret_u8_s8(d21s8), d22u8); + *d4ru8 = veor_u8(vreinterpret_u8_s8(d26s8), d22u8); + *d5ru8 = d17u8; + return; + } + + d21u8 = veor_u8(vreinterpret_u8_s8(d21s8), d22u8); + d24u8 = veor_u8(vreinterpret_u8_s8(d24s8), d22u8); + d25u8 = veor_u8(vreinterpret_u8_s8(d25s8), d22u8); + d26u8 = veor_u8(vreinterpret_u8_s8(d26s8), d22u8); + + d23u8 = vdup_n_u8(2); + q14u16 = vaddl_u8(d6u8, d7u8); + q14u16 = vmlal_u8(q14u16, d3u8, d27u8); + q14u16 = vmlal_u8(q14u16, d4u8, d23u8); + + d0u8 = vbsl_u8(d20u8, dblimit, d4u8); + + q14u16 = vaddw_u8(q14u16, d5u8); + + d1u8 = vbsl_u8(d20u8, dlimit, d25u8); + + d30u8 = vqrshrn_n_u16(q14u16, 3); + + q14u16 = vsubw_u8(q14u16, d3u8); + q14u16 = vsubw_u8(q14u16, d4u8); + q14u16 = vaddw_u8(q14u16, d5u8); + q14u16 = vaddw_u8(q14u16, d16u8); + + d2u8 = vbsl_u8(d20u8, dthresh, d24u8); + + d31u8 = vqrshrn_n_u16(q14u16, 3); + + q14u16 = vsubw_u8(q14u16, d3u8); + q14u16 = vsubw_u8(q14u16, d5u8); + q14u16 = vaddw_u8(q14u16, d6u8); + q14u16 = vaddw_u8(q14u16, d17u8); + + *d0ru8 = vbsl_u8(d20u8, d30u8, d0u8); + + d23u8 = vqrshrn_n_u16(q14u16, 3); + + q14u16 = vsubw_u8(q14u16, d3u8); + q14u16 = vsubw_u8(q14u16, d6u8); + q14u16 = vaddw_u8(q14u16, d7u8); + + *d1ru8 = vbsl_u8(d20u8, d31u8, d1u8); + + q14u16 = vaddw_u8(q14u16, d18u8); + + *d2ru8 = vbsl_u8(d20u8, d23u8, d2u8); + + d22u8 = vqrshrn_n_u16(q14u16, 3); + + q14u16 = vsubw_u8(q14u16, d4u8); + q14u16 = vsubw_u8(q14u16, d7u8); + q14u16 = vaddw_u8(q14u16, d16u8); + + d3u8 = vbsl_u8(d20u8, d3u8, d21u8); + + q14u16 = vaddw_u8(q14u16, d18u8); + + d4u8 = vbsl_u8(d20u8, d4u8, d26u8); + + d6u8 = vqrshrn_n_u16(q14u16, 3); + + q14u16 = vsubw_u8(q14u16, d5u8); + q14u16 = vsubw_u8(q14u16, d16u8); + q14u16 = vaddw_u8(q14u16, d17u8); + q14u16 = vaddw_u8(q14u16, d18u8); + + d5u8 = vbsl_u8(d20u8, d5u8, d17u8); + + d7u8 = vqrshrn_n_u16(q14u16, 3); + + *d3ru8 = vbsl_u8(d20u8, d22u8, d3u8); + *d4ru8 = vbsl_u8(d20u8, d6u8, d4u8); + *d5ru8 = vbsl_u8(d20u8, d7u8, d5u8); + } + return; +} + +void vpx_lpf_horizontal_8_neon( + uint8_t *src, + int pitch, + const uint8_t *blimit, + const uint8_t *limit, + const uint8_t *thresh) { + int i; + uint8_t *s, *psrc; + uint8x8_t dblimit, dlimit, dthresh; + uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8, d7u8; + uint8x8_t d16u8, d17u8, d18u8; + + dblimit = vld1_u8(blimit); + dlimit = vld1_u8(limit); + dthresh = vld1_u8(thresh); + + psrc = src - (pitch << 2); + for (i = 0; i < 1; i++) { + s = psrc + i * 8; + + d3u8 = vld1_u8(s); + s += pitch; + d4u8 = vld1_u8(s); + s += pitch; + d5u8 = vld1_u8(s); + s += pitch; + d6u8 = vld1_u8(s); + s += pitch; + d7u8 = vld1_u8(s); + s += pitch; + d16u8 = vld1_u8(s); + s += pitch; + d17u8 = vld1_u8(s); + s += pitch; + d18u8 = vld1_u8(s); + + mbloop_filter_neon(dblimit, dlimit, dthresh, + d3u8, d4u8, d5u8, d6u8, d7u8, d16u8, d17u8, d18u8, + &d0u8, &d1u8, &d2u8, &d3u8, &d4u8, &d5u8); + + s -= (pitch * 6); + vst1_u8(s, d0u8); + s += pitch; + vst1_u8(s, d1u8); + s += pitch; + vst1_u8(s, d2u8); + s += pitch; + vst1_u8(s, d3u8); + s += pitch; + vst1_u8(s, d4u8); + s += pitch; + vst1_u8(s, d5u8); + } + return; +} + +void vpx_lpf_vertical_8_neon( + uint8_t *src, + int pitch, + const uint8_t *blimit, + const uint8_t *limit, + const uint8_t *thresh) { + int i; + uint8_t *s; + uint8x8_t dblimit, dlimit, dthresh; + uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8, d7u8; + uint8x8_t d16u8, d17u8, d18u8; + uint32x2x2_t d2tmp0, d2tmp1, d2tmp2, d2tmp3; + uint16x4x2_t d2tmp4, d2tmp5, d2tmp6, d2tmp7; + uint8x8x2_t d2tmp8, d2tmp9, d2tmp10, d2tmp11; + uint8x8x4_t d4Result; + uint8x8x2_t d2Result; + + dblimit = vld1_u8(blimit); + dlimit = vld1_u8(limit); + dthresh = vld1_u8(thresh); + + for (i = 0; i < 1; i++) { + s = src + (i * (pitch << 3)) - 4; + + d3u8 = vld1_u8(s); + s += pitch; + d4u8 = vld1_u8(s); + s += pitch; + d5u8 = vld1_u8(s); + s += pitch; + d6u8 = vld1_u8(s); + s += pitch; + d7u8 = vld1_u8(s); + s += pitch; + d16u8 = vld1_u8(s); + s += pitch; + d17u8 = vld1_u8(s); + s += pitch; + d18u8 = vld1_u8(s); + + d2tmp0 = vtrn_u32(vreinterpret_u32_u8(d3u8), + vreinterpret_u32_u8(d7u8)); + d2tmp1 = vtrn_u32(vreinterpret_u32_u8(d4u8), + vreinterpret_u32_u8(d16u8)); + d2tmp2 = vtrn_u32(vreinterpret_u32_u8(d5u8), + vreinterpret_u32_u8(d17u8)); + d2tmp3 = vtrn_u32(vreinterpret_u32_u8(d6u8), + vreinterpret_u32_u8(d18u8)); + + d2tmp4 = vtrn_u16(vreinterpret_u16_u32(d2tmp0.val[0]), + vreinterpret_u16_u32(d2tmp2.val[0])); + d2tmp5 = vtrn_u16(vreinterpret_u16_u32(d2tmp1.val[0]), + vreinterpret_u16_u32(d2tmp3.val[0])); + d2tmp6 = vtrn_u16(vreinterpret_u16_u32(d2tmp0.val[1]), + vreinterpret_u16_u32(d2tmp2.val[1])); + d2tmp7 = vtrn_u16(vreinterpret_u16_u32(d2tmp1.val[1]), + vreinterpret_u16_u32(d2tmp3.val[1])); + + d2tmp8 = vtrn_u8(vreinterpret_u8_u16(d2tmp4.val[0]), + vreinterpret_u8_u16(d2tmp5.val[0])); + d2tmp9 = vtrn_u8(vreinterpret_u8_u16(d2tmp4.val[1]), + vreinterpret_u8_u16(d2tmp5.val[1])); + d2tmp10 = vtrn_u8(vreinterpret_u8_u16(d2tmp6.val[0]), + vreinterpret_u8_u16(d2tmp7.val[0])); + d2tmp11 = vtrn_u8(vreinterpret_u8_u16(d2tmp6.val[1]), + vreinterpret_u8_u16(d2tmp7.val[1])); + + d3u8 = d2tmp8.val[0]; + d4u8 = d2tmp8.val[1]; + d5u8 = d2tmp9.val[0]; + d6u8 = d2tmp9.val[1]; + d7u8 = d2tmp10.val[0]; + d16u8 = d2tmp10.val[1]; + d17u8 = d2tmp11.val[0]; + d18u8 = d2tmp11.val[1]; + + mbloop_filter_neon(dblimit, dlimit, dthresh, + d3u8, d4u8, d5u8, d6u8, d7u8, d16u8, d17u8, d18u8, + &d0u8, &d1u8, &d2u8, &d3u8, &d4u8, &d5u8); + + d4Result.val[0] = d0u8; + d4Result.val[1] = d1u8; + d4Result.val[2] = d2u8; + d4Result.val[3] = d3u8; + + d2Result.val[0] = d4u8; + d2Result.val[1] = d5u8; + + s = src - 3; + vst4_lane_u8(s, d4Result, 0); + s += pitch; + vst4_lane_u8(s, d4Result, 1); + s += pitch; + vst4_lane_u8(s, d4Result, 2); + s += pitch; + vst4_lane_u8(s, d4Result, 3); + s += pitch; + vst4_lane_u8(s, d4Result, 4); + s += pitch; + vst4_lane_u8(s, d4Result, 5); + s += pitch; + vst4_lane_u8(s, d4Result, 6); + s += pitch; + vst4_lane_u8(s, d4Result, 7); + + s = src + 1; + vst2_lane_u8(s, d2Result, 0); + s += pitch; + vst2_lane_u8(s, d2Result, 1); + s += pitch; + vst2_lane_u8(s, d2Result, 2); + s += pitch; + vst2_lane_u8(s, d2Result, 3); + s += pitch; + vst2_lane_u8(s, d2Result, 4); + s += pitch; + vst2_lane_u8(s, d2Result, 5); + s += pitch; + vst2_lane_u8(s, d2Result, 6); + s += pitch; + vst2_lane_u8(s, d2Result, 7); + } + return; +} diff --git a/thirdparty/libvpx/vpx_dsp/arm/loopfilter_mb_neon.asm b/thirdparty/libvpx/vpx_dsp/arm/loopfilter_mb_neon.asm new file mode 100644 index 0000000000..d5da7a8409 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/arm/loopfilter_mb_neon.asm @@ -0,0 +1,635 @@ +; +; Copyright (c) 2013 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + + EXPORT |vpx_lpf_horizontal_edge_8_neon| + EXPORT |vpx_lpf_horizontal_edge_16_neon| + EXPORT |vpx_lpf_vertical_16_neon| + ARM + + AREA ||.text||, CODE, READONLY, ALIGN=2 + +; void mb_lpf_horizontal_edge(uint8_t *s, int p, +; const uint8_t *blimit, +; const uint8_t *limit, +; const uint8_t *thresh, +; int count) +; r0 uint8_t *s, +; r1 int p, /* pitch */ +; r2 const uint8_t *blimit, +; r3 const uint8_t *limit, +; sp const uint8_t *thresh, +; r12 int count +|mb_lpf_horizontal_edge| PROC + push {r4-r8, lr} + vpush {d8-d15} + ldr r4, [sp, #88] ; load thresh + +h_count + vld1.8 {d16[]}, [r2] ; load *blimit + vld1.8 {d17[]}, [r3] ; load *limit + vld1.8 {d18[]}, [r4] ; load *thresh + + sub r8, r0, r1, lsl #3 ; move src pointer down by 8 lines + + vld1.u8 {d0}, [r8@64], r1 ; p7 + vld1.u8 {d1}, [r8@64], r1 ; p6 + vld1.u8 {d2}, [r8@64], r1 ; p5 + vld1.u8 {d3}, [r8@64], r1 ; p4 + vld1.u8 {d4}, [r8@64], r1 ; p3 + vld1.u8 {d5}, [r8@64], r1 ; p2 + vld1.u8 {d6}, [r8@64], r1 ; p1 + vld1.u8 {d7}, [r8@64], r1 ; p0 + vld1.u8 {d8}, [r8@64], r1 ; q0 + vld1.u8 {d9}, [r8@64], r1 ; q1 + vld1.u8 {d10}, [r8@64], r1 ; q2 + vld1.u8 {d11}, [r8@64], r1 ; q3 + vld1.u8 {d12}, [r8@64], r1 ; q4 + vld1.u8 {d13}, [r8@64], r1 ; q5 + vld1.u8 {d14}, [r8@64], r1 ; q6 + vld1.u8 {d15}, [r8@64], r1 ; q7 + + bl vpx_wide_mbfilter_neon + + tst r7, #1 + beq h_mbfilter + + ; flat && mask were not set for any of the channels. Just store the values + ; from filter. + sub r8, r0, r1, lsl #1 + + vst1.u8 {d25}, [r8@64], r1 ; store op1 + vst1.u8 {d24}, [r8@64], r1 ; store op0 + vst1.u8 {d23}, [r8@64], r1 ; store oq0 + vst1.u8 {d26}, [r8@64], r1 ; store oq1 + + b h_next + +h_mbfilter + tst r7, #2 + beq h_wide_mbfilter + + ; flat2 was not set for any of the channels. Just store the values from + ; mbfilter. + sub r8, r0, r1, lsl #1 + sub r8, r8, r1 + + vst1.u8 {d18}, [r8@64], r1 ; store op2 + vst1.u8 {d19}, [r8@64], r1 ; store op1 + vst1.u8 {d20}, [r8@64], r1 ; store op0 + vst1.u8 {d21}, [r8@64], r1 ; store oq0 + vst1.u8 {d22}, [r8@64], r1 ; store oq1 + vst1.u8 {d23}, [r8@64], r1 ; store oq2 + + b h_next + +h_wide_mbfilter + sub r8, r0, r1, lsl #3 + add r8, r8, r1 + + vst1.u8 {d16}, [r8@64], r1 ; store op6 + vst1.u8 {d24}, [r8@64], r1 ; store op5 + vst1.u8 {d25}, [r8@64], r1 ; store op4 + vst1.u8 {d26}, [r8@64], r1 ; store op3 + vst1.u8 {d27}, [r8@64], r1 ; store op2 + vst1.u8 {d18}, [r8@64], r1 ; store op1 + vst1.u8 {d19}, [r8@64], r1 ; store op0 + vst1.u8 {d20}, [r8@64], r1 ; store oq0 + vst1.u8 {d21}, [r8@64], r1 ; store oq1 + vst1.u8 {d22}, [r8@64], r1 ; store oq2 + vst1.u8 {d23}, [r8@64], r1 ; store oq3 + vst1.u8 {d1}, [r8@64], r1 ; store oq4 + vst1.u8 {d2}, [r8@64], r1 ; store oq5 + vst1.u8 {d3}, [r8@64], r1 ; store oq6 + +h_next + add r0, r0, #8 + subs r12, r12, #1 + bne h_count + + vpop {d8-d15} + pop {r4-r8, pc} + + ENDP ; |mb_lpf_horizontal_edge| + +; void vpx_lpf_horizontal_edge_8_neon(uint8_t *s, int pitch, +; const uint8_t *blimit, +; const uint8_t *limit, +; const uint8_t *thresh) +; r0 uint8_t *s, +; r1 int pitch, +; r2 const uint8_t *blimit, +; r3 const uint8_t *limit, +; sp const uint8_t *thresh +|vpx_lpf_horizontal_edge_8_neon| PROC + mov r12, #1 + b mb_lpf_horizontal_edge + ENDP ; |vpx_lpf_horizontal_edge_8_neon| + +; void vpx_lpf_horizontal_edge_16_neon(uint8_t *s, int pitch, +; const uint8_t *blimit, +; const uint8_t *limit, +; const uint8_t *thresh) +; r0 uint8_t *s, +; r1 int pitch, +; r2 const uint8_t *blimit, +; r3 const uint8_t *limit, +; sp const uint8_t *thresh +|vpx_lpf_horizontal_edge_16_neon| PROC + mov r12, #2 + b mb_lpf_horizontal_edge + ENDP ; |vpx_lpf_horizontal_edge_16_neon| + +; void vpx_lpf_vertical_16_neon(uint8_t *s, int p, +; const uint8_t *blimit, +; const uint8_t *limit, +; const uint8_t *thresh) +; r0 uint8_t *s, +; r1 int p, /* pitch */ +; r2 const uint8_t *blimit, +; r3 const uint8_t *limit, +; sp const uint8_t *thresh, +|vpx_lpf_vertical_16_neon| PROC + push {r4-r8, lr} + vpush {d8-d15} + ldr r4, [sp, #88] ; load thresh + + vld1.8 {d16[]}, [r2] ; load *blimit + vld1.8 {d17[]}, [r3] ; load *limit + vld1.8 {d18[]}, [r4] ; load *thresh + + sub r8, r0, #8 + + vld1.8 {d0}, [r8@64], r1 + vld1.8 {d8}, [r0@64], r1 + vld1.8 {d1}, [r8@64], r1 + vld1.8 {d9}, [r0@64], r1 + vld1.8 {d2}, [r8@64], r1 + vld1.8 {d10}, [r0@64], r1 + vld1.8 {d3}, [r8@64], r1 + vld1.8 {d11}, [r0@64], r1 + vld1.8 {d4}, [r8@64], r1 + vld1.8 {d12}, [r0@64], r1 + vld1.8 {d5}, [r8@64], r1 + vld1.8 {d13}, [r0@64], r1 + vld1.8 {d6}, [r8@64], r1 + vld1.8 {d14}, [r0@64], r1 + vld1.8 {d7}, [r8@64], r1 + vld1.8 {d15}, [r0@64], r1 + + sub r0, r0, r1, lsl #3 + + vtrn.32 q0, q2 + vtrn.32 q1, q3 + vtrn.32 q4, q6 + vtrn.32 q5, q7 + + vtrn.16 q0, q1 + vtrn.16 q2, q3 + vtrn.16 q4, q5 + vtrn.16 q6, q7 + + vtrn.8 d0, d1 + vtrn.8 d2, d3 + vtrn.8 d4, d5 + vtrn.8 d6, d7 + + vtrn.8 d8, d9 + vtrn.8 d10, d11 + vtrn.8 d12, d13 + vtrn.8 d14, d15 + + bl vpx_wide_mbfilter_neon + + tst r7, #1 + beq v_mbfilter + + ; flat && mask were not set for any of the channels. Just store the values + ; from filter. + sub r8, r0, #2 + + vswp d23, d25 + + vst4.8 {d23[0], d24[0], d25[0], d26[0]}, [r8], r1 + vst4.8 {d23[1], d24[1], d25[1], d26[1]}, [r8], r1 + vst4.8 {d23[2], d24[2], d25[2], d26[2]}, [r8], r1 + vst4.8 {d23[3], d24[3], d25[3], d26[3]}, [r8], r1 + vst4.8 {d23[4], d24[4], d25[4], d26[4]}, [r8], r1 + vst4.8 {d23[5], d24[5], d25[5], d26[5]}, [r8], r1 + vst4.8 {d23[6], d24[6], d25[6], d26[6]}, [r8], r1 + vst4.8 {d23[7], d24[7], d25[7], d26[7]}, [r8], r1 + + b v_end + +v_mbfilter + tst r7, #2 + beq v_wide_mbfilter + + ; flat2 was not set for any of the channels. Just store the values from + ; mbfilter. + sub r8, r0, #3 + + vst3.8 {d18[0], d19[0], d20[0]}, [r8], r1 + vst3.8 {d21[0], d22[0], d23[0]}, [r0], r1 + vst3.8 {d18[1], d19[1], d20[1]}, [r8], r1 + vst3.8 {d21[1], d22[1], d23[1]}, [r0], r1 + vst3.8 {d18[2], d19[2], d20[2]}, [r8], r1 + vst3.8 {d21[2], d22[2], d23[2]}, [r0], r1 + vst3.8 {d18[3], d19[3], d20[3]}, [r8], r1 + vst3.8 {d21[3], d22[3], d23[3]}, [r0], r1 + vst3.8 {d18[4], d19[4], d20[4]}, [r8], r1 + vst3.8 {d21[4], d22[4], d23[4]}, [r0], r1 + vst3.8 {d18[5], d19[5], d20[5]}, [r8], r1 + vst3.8 {d21[5], d22[5], d23[5]}, [r0], r1 + vst3.8 {d18[6], d19[6], d20[6]}, [r8], r1 + vst3.8 {d21[6], d22[6], d23[6]}, [r0], r1 + vst3.8 {d18[7], d19[7], d20[7]}, [r8], r1 + vst3.8 {d21[7], d22[7], d23[7]}, [r0], r1 + + b v_end + +v_wide_mbfilter + sub r8, r0, #8 + + vtrn.32 d0, d26 + vtrn.32 d16, d27 + vtrn.32 d24, d18 + vtrn.32 d25, d19 + + vtrn.16 d0, d24 + vtrn.16 d16, d25 + vtrn.16 d26, d18 + vtrn.16 d27, d19 + + vtrn.8 d0, d16 + vtrn.8 d24, d25 + vtrn.8 d26, d27 + vtrn.8 d18, d19 + + vtrn.32 d20, d1 + vtrn.32 d21, d2 + vtrn.32 d22, d3 + vtrn.32 d23, d15 + + vtrn.16 d20, d22 + vtrn.16 d21, d23 + vtrn.16 d1, d3 + vtrn.16 d2, d15 + + vtrn.8 d20, d21 + vtrn.8 d22, d23 + vtrn.8 d1, d2 + vtrn.8 d3, d15 + + vst1.8 {d0}, [r8@64], r1 + vst1.8 {d20}, [r0@64], r1 + vst1.8 {d16}, [r8@64], r1 + vst1.8 {d21}, [r0@64], r1 + vst1.8 {d24}, [r8@64], r1 + vst1.8 {d22}, [r0@64], r1 + vst1.8 {d25}, [r8@64], r1 + vst1.8 {d23}, [r0@64], r1 + vst1.8 {d26}, [r8@64], r1 + vst1.8 {d1}, [r0@64], r1 + vst1.8 {d27}, [r8@64], r1 + vst1.8 {d2}, [r0@64], r1 + vst1.8 {d18}, [r8@64], r1 + vst1.8 {d3}, [r0@64], r1 + vst1.8 {d19}, [r8@64], r1 + vst1.8 {d15}, [r0@64], r1 + +v_end + vpop {d8-d15} + pop {r4-r8, pc} + + ENDP ; |vpx_lpf_vertical_16_neon| + +; void vpx_wide_mbfilter_neon(); +; This is a helper function for the loopfilters. The invidual functions do the +; necessary load, transpose (if necessary) and store. +; +; r0-r3 PRESERVE +; d16 blimit +; d17 limit +; d18 thresh +; d0 p7 +; d1 p6 +; d2 p5 +; d3 p4 +; d4 p3 +; d5 p2 +; d6 p1 +; d7 p0 +; d8 q0 +; d9 q1 +; d10 q2 +; d11 q3 +; d12 q4 +; d13 q5 +; d14 q6 +; d15 q7 +|vpx_wide_mbfilter_neon| PROC + mov r7, #0 + + ; filter_mask + vabd.u8 d19, d4, d5 ; abs(p3 - p2) + vabd.u8 d20, d5, d6 ; abs(p2 - p1) + vabd.u8 d21, d6, d7 ; abs(p1 - p0) + vabd.u8 d22, d9, d8 ; abs(q1 - q0) + vabd.u8 d23, d10, d9 ; abs(q2 - q1) + vabd.u8 d24, d11, d10 ; abs(q3 - q2) + + ; only compare the largest value to limit + vmax.u8 d19, d19, d20 ; max(abs(p3 - p2), abs(p2 - p1)) + vmax.u8 d20, d21, d22 ; max(abs(p1 - p0), abs(q1 - q0)) + vmax.u8 d23, d23, d24 ; max(abs(q2 - q1), abs(q3 - q2)) + vmax.u8 d19, d19, d20 + + vabd.u8 d24, d7, d8 ; abs(p0 - q0) + + vmax.u8 d19, d19, d23 + + vabd.u8 d23, d6, d9 ; a = abs(p1 - q1) + vqadd.u8 d24, d24, d24 ; b = abs(p0 - q0) * 2 + + ; abs () > limit + vcge.u8 d19, d17, d19 + + ; flatmask4 + vabd.u8 d25, d7, d5 ; abs(p0 - p2) + vabd.u8 d26, d8, d10 ; abs(q0 - q2) + vabd.u8 d27, d4, d7 ; abs(p3 - p0) + vabd.u8 d28, d11, d8 ; abs(q3 - q0) + + ; only compare the largest value to thresh + vmax.u8 d25, d25, d26 ; max(abs(p0 - p2), abs(q0 - q2)) + vmax.u8 d26, d27, d28 ; max(abs(p3 - p0), abs(q3 - q0)) + vmax.u8 d25, d25, d26 + vmax.u8 d20, d20, d25 + + vshr.u8 d23, d23, #1 ; a = a / 2 + vqadd.u8 d24, d24, d23 ; a = b + a + + vmov.u8 d30, #1 + vcge.u8 d24, d16, d24 ; (a > blimit * 2 + limit) * -1 + + vcge.u8 d20, d30, d20 ; flat + + vand d19, d19, d24 ; mask + + ; hevmask + vcgt.u8 d21, d21, d18 ; (abs(p1 - p0) > thresh)*-1 + vcgt.u8 d22, d22, d18 ; (abs(q1 - q0) > thresh)*-1 + vorr d21, d21, d22 ; hev + + vand d16, d20, d19 ; flat && mask + vmov r5, r6, d16 + + ; flatmask5(1, p7, p6, p5, p4, p0, q0, q4, q5, q6, q7) + vabd.u8 d22, d3, d7 ; abs(p4 - p0) + vabd.u8 d23, d12, d8 ; abs(q4 - q0) + vabd.u8 d24, d7, d2 ; abs(p0 - p5) + vabd.u8 d25, d8, d13 ; abs(q0 - q5) + vabd.u8 d26, d1, d7 ; abs(p6 - p0) + vabd.u8 d27, d14, d8 ; abs(q6 - q0) + vabd.u8 d28, d0, d7 ; abs(p7 - p0) + vabd.u8 d29, d15, d8 ; abs(q7 - q0) + + ; only compare the largest value to thresh + vmax.u8 d22, d22, d23 ; max(abs(p4 - p0), abs(q4 - q0)) + vmax.u8 d23, d24, d25 ; max(abs(p0 - p5), abs(q0 - q5)) + vmax.u8 d24, d26, d27 ; max(abs(p6 - p0), abs(q6 - q0)) + vmax.u8 d25, d28, d29 ; max(abs(p7 - p0), abs(q7 - q0)) + + vmax.u8 d26, d22, d23 + vmax.u8 d27, d24, d25 + vmax.u8 d23, d26, d27 + + vcge.u8 d18, d30, d23 ; flat2 + + vmov.u8 d22, #0x80 + + orrs r5, r5, r6 ; Check for 0 + orreq r7, r7, #1 ; Only do filter branch + + vand d17, d18, d16 ; flat2 && flat && mask + vmov r5, r6, d17 + + ; mbfilter() function + + ; filter() function + ; convert to signed + veor d23, d8, d22 ; qs0 + veor d24, d7, d22 ; ps0 + veor d25, d6, d22 ; ps1 + veor d26, d9, d22 ; qs1 + + vmov.u8 d27, #3 + + vsub.s8 d28, d23, d24 ; ( qs0 - ps0) + vqsub.s8 d29, d25, d26 ; filter = clamp(ps1-qs1) + vmull.s8 q15, d28, d27 ; 3 * ( qs0 - ps0) + vand d29, d29, d21 ; filter &= hev + vaddw.s8 q15, q15, d29 ; filter + 3 * (qs0 - ps0) + vmov.u8 d29, #4 + + ; filter = clamp(filter + 3 * ( qs0 - ps0)) + vqmovn.s16 d28, q15 + + vand d28, d28, d19 ; filter &= mask + + vqadd.s8 d30, d28, d27 ; filter2 = clamp(filter+3) + vqadd.s8 d29, d28, d29 ; filter1 = clamp(filter+4) + vshr.s8 d30, d30, #3 ; filter2 >>= 3 + vshr.s8 d29, d29, #3 ; filter1 >>= 3 + + + vqadd.s8 d24, d24, d30 ; op0 = clamp(ps0 + filter2) + vqsub.s8 d23, d23, d29 ; oq0 = clamp(qs0 - filter1) + + ; outer tap adjustments: ++filter1 >> 1 + vrshr.s8 d29, d29, #1 + vbic d29, d29, d21 ; filter &= ~hev + + vqadd.s8 d25, d25, d29 ; op1 = clamp(ps1 + filter) + vqsub.s8 d26, d26, d29 ; oq1 = clamp(qs1 - filter) + + veor d24, d24, d22 ; *f_op0 = u^0x80 + veor d23, d23, d22 ; *f_oq0 = u^0x80 + veor d25, d25, d22 ; *f_op1 = u^0x80 + veor d26, d26, d22 ; *f_oq1 = u^0x80 + + tst r7, #1 + bxne lr + + orrs r5, r5, r6 ; Check for 0 + orreq r7, r7, #2 ; Only do mbfilter branch + + ; mbfilter flat && mask branch + ; TODO(fgalligan): Can I decrease the cycles shifting to consective d's + ; and using vibt on the q's? + vmov.u8 d29, #2 + vaddl.u8 q15, d7, d8 ; op2 = p0 + q0 + vmlal.u8 q15, d4, d27 ; op2 = p0 + q0 + p3 * 3 + vmlal.u8 q15, d5, d29 ; op2 = p0 + q0 + p3 * 3 + p2 * 2 + vaddl.u8 q10, d4, d5 + vaddw.u8 q15, d6 ; op2=p1 + p0 + q0 + p3 * 3 + p2 *2 + vaddl.u8 q14, d6, d9 + vqrshrn.u16 d18, q15, #3 ; r_op2 + + vsub.i16 q15, q10 + vaddl.u8 q10, d4, d6 + vadd.i16 q15, q14 + vaddl.u8 q14, d7, d10 + vqrshrn.u16 d19, q15, #3 ; r_op1 + + vsub.i16 q15, q10 + vadd.i16 q15, q14 + vaddl.u8 q14, d8, d11 + vqrshrn.u16 d20, q15, #3 ; r_op0 + + vsubw.u8 q15, d4 ; oq0 = op0 - p3 + vsubw.u8 q15, d7 ; oq0 -= p0 + vadd.i16 q15, q14 + vaddl.u8 q14, d9, d11 + vqrshrn.u16 d21, q15, #3 ; r_oq0 + + vsubw.u8 q15, d5 ; oq1 = oq0 - p2 + vsubw.u8 q15, d8 ; oq1 -= q0 + vadd.i16 q15, q14 + vaddl.u8 q14, d10, d11 + vqrshrn.u16 d22, q15, #3 ; r_oq1 + + vsubw.u8 q15, d6 ; oq2 = oq0 - p1 + vsubw.u8 q15, d9 ; oq2 -= q1 + vadd.i16 q15, q14 + vqrshrn.u16 d27, q15, #3 ; r_oq2 + + ; Filter does not set op2 or oq2, so use p2 and q2. + vbif d18, d5, d16 ; t_op2 |= p2 & ~(flat & mask) + vbif d19, d25, d16 ; t_op1 |= f_op1 & ~(flat & mask) + vbif d20, d24, d16 ; t_op0 |= f_op0 & ~(flat & mask) + vbif d21, d23, d16 ; t_oq0 |= f_oq0 & ~(flat & mask) + vbif d22, d26, d16 ; t_oq1 |= f_oq1 & ~(flat & mask) + + vbit d23, d27, d16 ; t_oq2 |= r_oq2 & (flat & mask) + vbif d23, d10, d16 ; t_oq2 |= q2 & ~(flat & mask) + + tst r7, #2 + bxne lr + + ; wide_mbfilter flat2 && flat && mask branch + vmov.u8 d16, #7 + vaddl.u8 q15, d7, d8 ; op6 = p0 + q0 + vaddl.u8 q12, d2, d3 + vaddl.u8 q13, d4, d5 + vaddl.u8 q14, d1, d6 + vmlal.u8 q15, d0, d16 ; op6 += p7 * 3 + vadd.i16 q12, q13 + vadd.i16 q15, q14 + vaddl.u8 q14, d2, d9 + vadd.i16 q15, q12 + vaddl.u8 q12, d0, d1 + vaddw.u8 q15, d1 + vaddl.u8 q13, d0, d2 + vadd.i16 q14, q15, q14 + vqrshrn.u16 d16, q15, #4 ; w_op6 + + vsub.i16 q15, q14, q12 + vaddl.u8 q14, d3, d10 + vqrshrn.u16 d24, q15, #4 ; w_op5 + + vsub.i16 q15, q13 + vaddl.u8 q13, d0, d3 + vadd.i16 q15, q14 + vaddl.u8 q14, d4, d11 + vqrshrn.u16 d25, q15, #4 ; w_op4 + + vadd.i16 q15, q14 + vaddl.u8 q14, d0, d4 + vsub.i16 q15, q13 + vsub.i16 q14, q15, q14 + vqrshrn.u16 d26, q15, #4 ; w_op3 + + vaddw.u8 q15, q14, d5 ; op2 += p2 + vaddl.u8 q14, d0, d5 + vaddw.u8 q15, d12 ; op2 += q4 + vbif d26, d4, d17 ; op3 |= p3 & ~(f2 & f & m) + vqrshrn.u16 d27, q15, #4 ; w_op2 + + vsub.i16 q15, q14 + vaddl.u8 q14, d0, d6 + vaddw.u8 q15, d6 ; op1 += p1 + vaddw.u8 q15, d13 ; op1 += q5 + vbif d27, d18, d17 ; op2 |= t_op2 & ~(f2 & f & m) + vqrshrn.u16 d18, q15, #4 ; w_op1 + + vsub.i16 q15, q14 + vaddl.u8 q14, d0, d7 + vaddw.u8 q15, d7 ; op0 += p0 + vaddw.u8 q15, d14 ; op0 += q6 + vbif d18, d19, d17 ; op1 |= t_op1 & ~(f2 & f & m) + vqrshrn.u16 d19, q15, #4 ; w_op0 + + vsub.i16 q15, q14 + vaddl.u8 q14, d1, d8 + vaddw.u8 q15, d8 ; oq0 += q0 + vaddw.u8 q15, d15 ; oq0 += q7 + vbif d19, d20, d17 ; op0 |= t_op0 & ~(f2 & f & m) + vqrshrn.u16 d20, q15, #4 ; w_oq0 + + vsub.i16 q15, q14 + vaddl.u8 q14, d2, d9 + vaddw.u8 q15, d9 ; oq1 += q1 + vaddl.u8 q4, d10, d15 + vaddw.u8 q15, d15 ; oq1 += q7 + vbif d20, d21, d17 ; oq0 |= t_oq0 & ~(f2 & f & m) + vqrshrn.u16 d21, q15, #4 ; w_oq1 + + vsub.i16 q15, q14 + vaddl.u8 q14, d3, d10 + vadd.i16 q15, q4 + vaddl.u8 q4, d11, d15 + vbif d21, d22, d17 ; oq1 |= t_oq1 & ~(f2 & f & m) + vqrshrn.u16 d22, q15, #4 ; w_oq2 + + vsub.i16 q15, q14 + vaddl.u8 q14, d4, d11 + vadd.i16 q15, q4 + vaddl.u8 q4, d12, d15 + vbif d22, d23, d17 ; oq2 |= t_oq2 & ~(f2 & f & m) + vqrshrn.u16 d23, q15, #4 ; w_oq3 + + vsub.i16 q15, q14 + vaddl.u8 q14, d5, d12 + vadd.i16 q15, q4 + vaddl.u8 q4, d13, d15 + vbif d16, d1, d17 ; op6 |= p6 & ~(f2 & f & m) + vqrshrn.u16 d1, q15, #4 ; w_oq4 + + vsub.i16 q15, q14 + vaddl.u8 q14, d6, d13 + vadd.i16 q15, q4 + vaddl.u8 q4, d14, d15 + vbif d24, d2, d17 ; op5 |= p5 & ~(f2 & f & m) + vqrshrn.u16 d2, q15, #4 ; w_oq5 + + vsub.i16 q15, q14 + vbif d25, d3, d17 ; op4 |= p4 & ~(f2 & f & m) + vadd.i16 q15, q4 + vbif d23, d11, d17 ; oq3 |= q3 & ~(f2 & f & m) + vqrshrn.u16 d3, q15, #4 ; w_oq6 + vbif d1, d12, d17 ; oq4 |= q4 & ~(f2 & f & m) + vbif d2, d13, d17 ; oq5 |= q5 & ~(f2 & f & m) + vbif d3, d14, d17 ; oq6 |= q6 & ~(f2 & f & m) + + bx lr + ENDP ; |vpx_wide_mbfilter_neon| + + END diff --git a/thirdparty/libvpx/vpx_dsp/arm/loopfilter_neon.c b/thirdparty/libvpx/vpx_dsp/arm/loopfilter_neon.c new file mode 100644 index 0000000000..aa31f29358 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/arm/loopfilter_neon.c @@ -0,0 +1,58 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +#include "./vpx_dsp_rtcd.h" +#include "./vpx_config.h" +#include "vpx/vpx_integer.h" + +void vpx_lpf_vertical_4_dual_neon(uint8_t *s, int p, + const uint8_t *blimit0, + const uint8_t *limit0, + const uint8_t *thresh0, + const uint8_t *blimit1, + const uint8_t *limit1, + const uint8_t *thresh1) { + vpx_lpf_vertical_4_neon(s, p, blimit0, limit0, thresh0); + vpx_lpf_vertical_4_neon(s + 8 * p, p, blimit1, limit1, thresh1); +} + +#if HAVE_NEON_ASM +void vpx_lpf_horizontal_8_dual_neon(uint8_t *s, int p /* pitch */, + const uint8_t *blimit0, + const uint8_t *limit0, + const uint8_t *thresh0, + const uint8_t *blimit1, + const uint8_t *limit1, + const uint8_t *thresh1) { + vpx_lpf_horizontal_8_neon(s, p, blimit0, limit0, thresh0); + vpx_lpf_horizontal_8_neon(s + 8, p, blimit1, limit1, thresh1); +} + +void vpx_lpf_vertical_8_dual_neon(uint8_t *s, int p, + const uint8_t *blimit0, + const uint8_t *limit0, + const uint8_t *thresh0, + const uint8_t *blimit1, + const uint8_t *limit1, + const uint8_t *thresh1) { + vpx_lpf_vertical_8_neon(s, p, blimit0, limit0, thresh0); + vpx_lpf_vertical_8_neon(s + 8 * p, p, blimit1, limit1, thresh1); +} + +void vpx_lpf_vertical_16_dual_neon(uint8_t *s, int p, + const uint8_t *blimit, + const uint8_t *limit, + const uint8_t *thresh) { + vpx_lpf_vertical_16_neon(s, p, blimit, limit, thresh); + vpx_lpf_vertical_16_neon(s + 8 * p, p, blimit, limit, thresh); +} +#endif // HAVE_NEON_ASM diff --git a/thirdparty/libvpx/vpx_dsp/arm/save_reg_neon.asm b/thirdparty/libvpx/vpx_dsp/arm/save_reg_neon.asm new file mode 100644 index 0000000000..c9ca10801d --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/arm/save_reg_neon.asm @@ -0,0 +1,36 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + + + EXPORT |vpx_push_neon| + EXPORT |vpx_pop_neon| + + ARM + REQUIRE8 + PRESERVE8 + + AREA ||.text||, CODE, READONLY, ALIGN=2 + +|vpx_push_neon| PROC + vst1.i64 {d8, d9, d10, d11}, [r0]! + vst1.i64 {d12, d13, d14, d15}, [r0]! + bx lr + + ENDP + +|vpx_pop_neon| PROC + vld1.i64 {d8, d9, d10, d11}, [r0]! + vld1.i64 {d12, d13, d14, d15}, [r0]! + bx lr + + ENDP + + END + diff --git a/thirdparty/libvpx/vpx_dsp/arm/vpx_convolve8_avg_neon.c b/thirdparty/libvpx/vpx_dsp/arm/vpx_convolve8_avg_neon.c new file mode 100644 index 0000000000..8632250138 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/arm/vpx_convolve8_avg_neon.c @@ -0,0 +1,373 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> +#include <assert.h> + +#include "./vpx_config.h" +#include "./vpx_dsp_rtcd.h" +#include "vpx/vpx_integer.h" +#include "vpx_ports/mem.h" + +static INLINE int32x4_t MULTIPLY_BY_Q0( + int16x4_t dsrc0, + int16x4_t dsrc1, + int16x4_t dsrc2, + int16x4_t dsrc3, + int16x4_t dsrc4, + int16x4_t dsrc5, + int16x4_t dsrc6, + int16x4_t dsrc7, + int16x8_t q0s16) { + int32x4_t qdst; + int16x4_t d0s16, d1s16; + + d0s16 = vget_low_s16(q0s16); + d1s16 = vget_high_s16(q0s16); + + qdst = vmull_lane_s16(dsrc0, d0s16, 0); + qdst = vmlal_lane_s16(qdst, dsrc1, d0s16, 1); + qdst = vmlal_lane_s16(qdst, dsrc2, d0s16, 2); + qdst = vmlal_lane_s16(qdst, dsrc3, d0s16, 3); + qdst = vmlal_lane_s16(qdst, dsrc4, d1s16, 0); + qdst = vmlal_lane_s16(qdst, dsrc5, d1s16, 1); + qdst = vmlal_lane_s16(qdst, dsrc6, d1s16, 2); + qdst = vmlal_lane_s16(qdst, dsrc7, d1s16, 3); + return qdst; +} + +void vpx_convolve8_avg_horiz_neon( + const uint8_t *src, + ptrdiff_t src_stride, + uint8_t *dst, + ptrdiff_t dst_stride, + const int16_t *filter_x, + int x_step_q4, + const int16_t *filter_y, // unused + int y_step_q4, // unused + int w, + int h) { + int width; + const uint8_t *s; + uint8_t *d; + uint8x8_t d2u8, d3u8, d24u8, d25u8, d26u8, d27u8, d28u8, d29u8; + uint32x2_t d2u32, d3u32, d6u32, d7u32, d28u32, d29u32, d30u32, d31u32; + uint8x16_t q1u8, q3u8, q12u8, q13u8, q14u8, q15u8; + int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d22s16, d23s16; + int16x4_t d24s16, d25s16, d26s16, d27s16; + uint16x4_t d2u16, d3u16, d4u16, d5u16, d16u16, d17u16, d18u16, d19u16; + int16x8_t q0s16; + uint16x8_t q1u16, q2u16, q8u16, q9u16, q10u16, q11u16, q12u16, q13u16; + int32x4_t q1s32, q2s32, q14s32, q15s32; + uint16x8x2_t q0x2u16; + uint8x8x2_t d0x2u8, d1x2u8; + uint32x2x2_t d0x2u32; + uint16x4x2_t d0x2u16, d1x2u16; + uint32x4x2_t q0x2u32; + + assert(x_step_q4 == 16); + + q0s16 = vld1q_s16(filter_x); + + src -= 3; // adjust for taps + for (; h > 0; h -= 4) { // loop_horiz_v + s = src; + d24u8 = vld1_u8(s); + s += src_stride; + d25u8 = vld1_u8(s); + s += src_stride; + d26u8 = vld1_u8(s); + s += src_stride; + d27u8 = vld1_u8(s); + + q12u8 = vcombine_u8(d24u8, d25u8); + q13u8 = vcombine_u8(d26u8, d27u8); + + q0x2u16 = vtrnq_u16(vreinterpretq_u16_u8(q12u8), + vreinterpretq_u16_u8(q13u8)); + d24u8 = vreinterpret_u8_u16(vget_low_u16(q0x2u16.val[0])); + d25u8 = vreinterpret_u8_u16(vget_high_u16(q0x2u16.val[0])); + d26u8 = vreinterpret_u8_u16(vget_low_u16(q0x2u16.val[1])); + d27u8 = vreinterpret_u8_u16(vget_high_u16(q0x2u16.val[1])); + d0x2u8 = vtrn_u8(d24u8, d25u8); + d1x2u8 = vtrn_u8(d26u8, d27u8); + + __builtin_prefetch(src + src_stride * 4); + __builtin_prefetch(src + src_stride * 5); + + q8u16 = vmovl_u8(d0x2u8.val[0]); + q9u16 = vmovl_u8(d0x2u8.val[1]); + q10u16 = vmovl_u8(d1x2u8.val[0]); + q11u16 = vmovl_u8(d1x2u8.val[1]); + + src += 7; + d16u16 = vget_low_u16(q8u16); + d17u16 = vget_high_u16(q8u16); + d18u16 = vget_low_u16(q9u16); + d19u16 = vget_high_u16(q9u16); + q8u16 = vcombine_u16(d16u16, d18u16); // vswp 17 18 + q9u16 = vcombine_u16(d17u16, d19u16); + + d20s16 = vreinterpret_s16_u16(vget_low_u16(q10u16)); + d23s16 = vreinterpret_s16_u16(vget_high_u16(q10u16)); // vmov 23 21 + for (width = w; + width > 0; + width -= 4, src += 4, dst += 4) { // loop_horiz + s = src; + d28u32 = vld1_dup_u32((const uint32_t *)s); + s += src_stride; + d29u32 = vld1_dup_u32((const uint32_t *)s); + s += src_stride; + d31u32 = vld1_dup_u32((const uint32_t *)s); + s += src_stride; + d30u32 = vld1_dup_u32((const uint32_t *)s); + + __builtin_prefetch(src + 64); + + d0x2u16 = vtrn_u16(vreinterpret_u16_u32(d28u32), + vreinterpret_u16_u32(d31u32)); + d1x2u16 = vtrn_u16(vreinterpret_u16_u32(d29u32), + vreinterpret_u16_u32(d30u32)); + d0x2u8 = vtrn_u8(vreinterpret_u8_u16(d0x2u16.val[0]), // d28 + vreinterpret_u8_u16(d1x2u16.val[0])); // d29 + d1x2u8 = vtrn_u8(vreinterpret_u8_u16(d0x2u16.val[1]), // d31 + vreinterpret_u8_u16(d1x2u16.val[1])); // d30 + + __builtin_prefetch(src + 64 + src_stride); + + q14u8 = vcombine_u8(d0x2u8.val[0], d0x2u8.val[1]); + q15u8 = vcombine_u8(d1x2u8.val[1], d1x2u8.val[0]); + q0x2u32 = vtrnq_u32(vreinterpretq_u32_u8(q14u8), + vreinterpretq_u32_u8(q15u8)); + + d28u8 = vreinterpret_u8_u32(vget_low_u32(q0x2u32.val[0])); + d29u8 = vreinterpret_u8_u32(vget_high_u32(q0x2u32.val[0])); + q12u16 = vmovl_u8(d28u8); + q13u16 = vmovl_u8(d29u8); + + __builtin_prefetch(src + 64 + src_stride * 2); + + d = dst; + d6u32 = vld1_lane_u32((const uint32_t *)d, d6u32, 0); + d += dst_stride; + d7u32 = vld1_lane_u32((const uint32_t *)d, d7u32, 0); + d += dst_stride; + d6u32 = vld1_lane_u32((const uint32_t *)d, d6u32, 1); + d += dst_stride; + d7u32 = vld1_lane_u32((const uint32_t *)d, d7u32, 1); + + d16s16 = vreinterpret_s16_u16(vget_low_u16(q8u16)); + d17s16 = vreinterpret_s16_u16(vget_high_u16(q8u16)); + d18s16 = vreinterpret_s16_u16(vget_low_u16(q9u16)); + d19s16 = vreinterpret_s16_u16(vget_high_u16(q9u16)); + d22s16 = vreinterpret_s16_u16(vget_low_u16(q11u16)); + d24s16 = vreinterpret_s16_u16(vget_low_u16(q12u16)); + d25s16 = vreinterpret_s16_u16(vget_high_u16(q12u16)); + d26s16 = vreinterpret_s16_u16(vget_low_u16(q13u16)); + d27s16 = vreinterpret_s16_u16(vget_high_u16(q13u16)); + + q1s32 = MULTIPLY_BY_Q0(d16s16, d17s16, d20s16, d22s16, + d18s16, d19s16, d23s16, d24s16, q0s16); + q2s32 = MULTIPLY_BY_Q0(d17s16, d20s16, d22s16, d18s16, + d19s16, d23s16, d24s16, d26s16, q0s16); + q14s32 = MULTIPLY_BY_Q0(d20s16, d22s16, d18s16, d19s16, + d23s16, d24s16, d26s16, d27s16, q0s16); + q15s32 = MULTIPLY_BY_Q0(d22s16, d18s16, d19s16, d23s16, + d24s16, d26s16, d27s16, d25s16, q0s16); + + __builtin_prefetch(src + 64 + src_stride * 3); + + d2u16 = vqrshrun_n_s32(q1s32, 7); + d3u16 = vqrshrun_n_s32(q2s32, 7); + d4u16 = vqrshrun_n_s32(q14s32, 7); + d5u16 = vqrshrun_n_s32(q15s32, 7); + + q1u16 = vcombine_u16(d2u16, d3u16); + q2u16 = vcombine_u16(d4u16, d5u16); + + d2u8 = vqmovn_u16(q1u16); + d3u8 = vqmovn_u16(q2u16); + + d0x2u16 = vtrn_u16(vreinterpret_u16_u8(d2u8), + vreinterpret_u16_u8(d3u8)); + d0x2u32 = vtrn_u32(vreinterpret_u32_u16(d0x2u16.val[0]), + vreinterpret_u32_u16(d0x2u16.val[1])); + d0x2u8 = vtrn_u8(vreinterpret_u8_u32(d0x2u32.val[0]), + vreinterpret_u8_u32(d0x2u32.val[1])); + + q1u8 = vcombine_u8(d0x2u8.val[0], d0x2u8.val[1]); + q3u8 = vreinterpretq_u8_u32(vcombine_u32(d6u32, d7u32)); + + q1u8 = vrhaddq_u8(q1u8, q3u8); + + d2u32 = vreinterpret_u32_u8(vget_low_u8(q1u8)); + d3u32 = vreinterpret_u32_u8(vget_high_u8(q1u8)); + + d = dst; + vst1_lane_u32((uint32_t *)d, d2u32, 0); + d += dst_stride; + vst1_lane_u32((uint32_t *)d, d3u32, 0); + d += dst_stride; + vst1_lane_u32((uint32_t *)d, d2u32, 1); + d += dst_stride; + vst1_lane_u32((uint32_t *)d, d3u32, 1); + + q8u16 = q9u16; + d20s16 = d23s16; + q11u16 = q12u16; + q9u16 = q13u16; + d23s16 = vreinterpret_s16_u16(vget_high_u16(q11u16)); + } + src += src_stride * 4 - w - 7; + dst += dst_stride * 4 - w; + } + return; +} + +void vpx_convolve8_avg_vert_neon( + const uint8_t *src, + ptrdiff_t src_stride, + uint8_t *dst, + ptrdiff_t dst_stride, + const int16_t *filter_x, // unused + int x_step_q4, // unused + const int16_t *filter_y, + int y_step_q4, + int w, + int h) { + int height; + const uint8_t *s; + uint8_t *d; + uint8x8_t d2u8, d3u8; + uint32x2_t d2u32, d3u32, d6u32, d7u32; + uint32x2_t d16u32, d18u32, d20u32, d22u32, d24u32, d26u32; + uint8x16_t q1u8, q3u8; + int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16; + int16x4_t d24s16, d25s16, d26s16, d27s16; + uint16x4_t d2u16, d3u16, d4u16, d5u16; + int16x8_t q0s16; + uint16x8_t q1u16, q2u16, q8u16, q9u16, q10u16, q11u16, q12u16, q13u16; + int32x4_t q1s32, q2s32, q14s32, q15s32; + + assert(y_step_q4 == 16); + + src -= src_stride * 3; + q0s16 = vld1q_s16(filter_y); + for (; w > 0; w -= 4, src += 4, dst += 4) { // loop_vert_h + s = src; + d16u32 = vld1_lane_u32((const uint32_t *)s, d16u32, 0); + s += src_stride; + d16u32 = vld1_lane_u32((const uint32_t *)s, d16u32, 1); + s += src_stride; + d18u32 = vld1_lane_u32((const uint32_t *)s, d18u32, 0); + s += src_stride; + d18u32 = vld1_lane_u32((const uint32_t *)s, d18u32, 1); + s += src_stride; + d20u32 = vld1_lane_u32((const uint32_t *)s, d20u32, 0); + s += src_stride; + d20u32 = vld1_lane_u32((const uint32_t *)s, d20u32, 1); + s += src_stride; + d22u32 = vld1_lane_u32((const uint32_t *)s, d22u32, 0); + s += src_stride; + + q8u16 = vmovl_u8(vreinterpret_u8_u32(d16u32)); + q9u16 = vmovl_u8(vreinterpret_u8_u32(d18u32)); + q10u16 = vmovl_u8(vreinterpret_u8_u32(d20u32)); + q11u16 = vmovl_u8(vreinterpret_u8_u32(d22u32)); + + d18s16 = vreinterpret_s16_u16(vget_low_u16(q9u16)); + d19s16 = vreinterpret_s16_u16(vget_high_u16(q9u16)); + d22s16 = vreinterpret_s16_u16(vget_low_u16(q11u16)); + d = dst; + for (height = h; height > 0; height -= 4) { // loop_vert + d24u32 = vld1_lane_u32((const uint32_t *)s, d24u32, 0); + s += src_stride; + d26u32 = vld1_lane_u32((const uint32_t *)s, d26u32, 0); + s += src_stride; + d26u32 = vld1_lane_u32((const uint32_t *)s, d26u32, 1); + s += src_stride; + d24u32 = vld1_lane_u32((const uint32_t *)s, d24u32, 1); + s += src_stride; + + q12u16 = vmovl_u8(vreinterpret_u8_u32(d24u32)); + q13u16 = vmovl_u8(vreinterpret_u8_u32(d26u32)); + + d6u32 = vld1_lane_u32((const uint32_t *)d, d6u32, 0); + d += dst_stride; + d6u32 = vld1_lane_u32((const uint32_t *)d, d6u32, 1); + d += dst_stride; + d7u32 = vld1_lane_u32((const uint32_t *)d, d7u32, 0); + d += dst_stride; + d7u32 = vld1_lane_u32((const uint32_t *)d, d7u32, 1); + d -= dst_stride * 3; + + d16s16 = vreinterpret_s16_u16(vget_low_u16(q8u16)); + d17s16 = vreinterpret_s16_u16(vget_high_u16(q8u16)); + d20s16 = vreinterpret_s16_u16(vget_low_u16(q10u16)); + d21s16 = vreinterpret_s16_u16(vget_high_u16(q10u16)); + d24s16 = vreinterpret_s16_u16(vget_low_u16(q12u16)); + d25s16 = vreinterpret_s16_u16(vget_high_u16(q12u16)); + d26s16 = vreinterpret_s16_u16(vget_low_u16(q13u16)); + d27s16 = vreinterpret_s16_u16(vget_high_u16(q13u16)); + + __builtin_prefetch(s); + __builtin_prefetch(s + src_stride); + q1s32 = MULTIPLY_BY_Q0(d16s16, d17s16, d18s16, d19s16, + d20s16, d21s16, d22s16, d24s16, q0s16); + __builtin_prefetch(s + src_stride * 2); + __builtin_prefetch(s + src_stride * 3); + q2s32 = MULTIPLY_BY_Q0(d17s16, d18s16, d19s16, d20s16, + d21s16, d22s16, d24s16, d26s16, q0s16); + __builtin_prefetch(d); + __builtin_prefetch(d + dst_stride); + q14s32 = MULTIPLY_BY_Q0(d18s16, d19s16, d20s16, d21s16, + d22s16, d24s16, d26s16, d27s16, q0s16); + __builtin_prefetch(d + dst_stride * 2); + __builtin_prefetch(d + dst_stride * 3); + q15s32 = MULTIPLY_BY_Q0(d19s16, d20s16, d21s16, d22s16, + d24s16, d26s16, d27s16, d25s16, q0s16); + + d2u16 = vqrshrun_n_s32(q1s32, 7); + d3u16 = vqrshrun_n_s32(q2s32, 7); + d4u16 = vqrshrun_n_s32(q14s32, 7); + d5u16 = vqrshrun_n_s32(q15s32, 7); + + q1u16 = vcombine_u16(d2u16, d3u16); + q2u16 = vcombine_u16(d4u16, d5u16); + + d2u8 = vqmovn_u16(q1u16); + d3u8 = vqmovn_u16(q2u16); + + q1u8 = vcombine_u8(d2u8, d3u8); + q3u8 = vreinterpretq_u8_u32(vcombine_u32(d6u32, d7u32)); + + q1u8 = vrhaddq_u8(q1u8, q3u8); + + d2u32 = vreinterpret_u32_u8(vget_low_u8(q1u8)); + d3u32 = vreinterpret_u32_u8(vget_high_u8(q1u8)); + + vst1_lane_u32((uint32_t *)d, d2u32, 0); + d += dst_stride; + vst1_lane_u32((uint32_t *)d, d2u32, 1); + d += dst_stride; + vst1_lane_u32((uint32_t *)d, d3u32, 0); + d += dst_stride; + vst1_lane_u32((uint32_t *)d, d3u32, 1); + d += dst_stride; + + q8u16 = q10u16; + d18s16 = d22s16; + d19s16 = d24s16; + q10u16 = q13u16; + d22s16 = d25s16; + } + } + return; +} diff --git a/thirdparty/libvpx/vpx_dsp/arm/vpx_convolve8_neon.c b/thirdparty/libvpx/vpx_dsp/arm/vpx_convolve8_neon.c new file mode 100644 index 0000000000..9bd715e2c6 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/arm/vpx_convolve8_neon.c @@ -0,0 +1,340 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> +#include <assert.h> + +#include "./vpx_config.h" +#include "./vpx_dsp_rtcd.h" +#include "vpx/vpx_integer.h" +#include "vpx_ports/mem.h" + +static INLINE int32x4_t MULTIPLY_BY_Q0( + int16x4_t dsrc0, + int16x4_t dsrc1, + int16x4_t dsrc2, + int16x4_t dsrc3, + int16x4_t dsrc4, + int16x4_t dsrc5, + int16x4_t dsrc6, + int16x4_t dsrc7, + int16x8_t q0s16) { + int32x4_t qdst; + int16x4_t d0s16, d1s16; + + d0s16 = vget_low_s16(q0s16); + d1s16 = vget_high_s16(q0s16); + + qdst = vmull_lane_s16(dsrc0, d0s16, 0); + qdst = vmlal_lane_s16(qdst, dsrc1, d0s16, 1); + qdst = vmlal_lane_s16(qdst, dsrc2, d0s16, 2); + qdst = vmlal_lane_s16(qdst, dsrc3, d0s16, 3); + qdst = vmlal_lane_s16(qdst, dsrc4, d1s16, 0); + qdst = vmlal_lane_s16(qdst, dsrc5, d1s16, 1); + qdst = vmlal_lane_s16(qdst, dsrc6, d1s16, 2); + qdst = vmlal_lane_s16(qdst, dsrc7, d1s16, 3); + return qdst; +} + +void vpx_convolve8_horiz_neon( + const uint8_t *src, + ptrdiff_t src_stride, + uint8_t *dst, + ptrdiff_t dst_stride, + const int16_t *filter_x, + int x_step_q4, + const int16_t *filter_y, // unused + int y_step_q4, // unused + int w, + int h) { + int width; + const uint8_t *s, *psrc; + uint8_t *d, *pdst; + uint8x8_t d2u8, d3u8, d24u8, d25u8, d26u8, d27u8, d28u8, d29u8; + uint32x2_t d2u32, d3u32, d28u32, d29u32, d30u32, d31u32; + uint8x16_t q12u8, q13u8, q14u8, q15u8; + int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d22s16, d23s16; + int16x4_t d24s16, d25s16, d26s16, d27s16; + uint16x4_t d2u16, d3u16, d4u16, d5u16, d16u16, d17u16, d18u16, d19u16; + int16x8_t q0s16; + uint16x8_t q1u16, q2u16, q8u16, q9u16, q10u16, q11u16, q12u16, q13u16; + int32x4_t q1s32, q2s32, q14s32, q15s32; + uint16x8x2_t q0x2u16; + uint8x8x2_t d0x2u8, d1x2u8; + uint32x2x2_t d0x2u32; + uint16x4x2_t d0x2u16, d1x2u16; + uint32x4x2_t q0x2u32; + + assert(x_step_q4 == 16); + + q0s16 = vld1q_s16(filter_x); + + src -= 3; // adjust for taps + for (; h > 0; h -= 4, + src += src_stride * 4, + dst += dst_stride * 4) { // loop_horiz_v + s = src; + d24u8 = vld1_u8(s); + s += src_stride; + d25u8 = vld1_u8(s); + s += src_stride; + d26u8 = vld1_u8(s); + s += src_stride; + d27u8 = vld1_u8(s); + + q12u8 = vcombine_u8(d24u8, d25u8); + q13u8 = vcombine_u8(d26u8, d27u8); + + q0x2u16 = vtrnq_u16(vreinterpretq_u16_u8(q12u8), + vreinterpretq_u16_u8(q13u8)); + d24u8 = vreinterpret_u8_u16(vget_low_u16(q0x2u16.val[0])); + d25u8 = vreinterpret_u8_u16(vget_high_u16(q0x2u16.val[0])); + d26u8 = vreinterpret_u8_u16(vget_low_u16(q0x2u16.val[1])); + d27u8 = vreinterpret_u8_u16(vget_high_u16(q0x2u16.val[1])); + d0x2u8 = vtrn_u8(d24u8, d25u8); + d1x2u8 = vtrn_u8(d26u8, d27u8); + + __builtin_prefetch(src + src_stride * 4); + __builtin_prefetch(src + src_stride * 5); + __builtin_prefetch(src + src_stride * 6); + + q8u16 = vmovl_u8(d0x2u8.val[0]); + q9u16 = vmovl_u8(d0x2u8.val[1]); + q10u16 = vmovl_u8(d1x2u8.val[0]); + q11u16 = vmovl_u8(d1x2u8.val[1]); + + d16u16 = vget_low_u16(q8u16); + d17u16 = vget_high_u16(q8u16); + d18u16 = vget_low_u16(q9u16); + d19u16 = vget_high_u16(q9u16); + q8u16 = vcombine_u16(d16u16, d18u16); // vswp 17 18 + q9u16 = vcombine_u16(d17u16, d19u16); + + d20s16 = vreinterpret_s16_u16(vget_low_u16(q10u16)); + d23s16 = vreinterpret_s16_u16(vget_high_u16(q10u16)); // vmov 23 21 + for (width = w, psrc = src + 7, pdst = dst; + width > 0; + width -= 4, psrc += 4, pdst += 4) { // loop_horiz + s = psrc; + d28u32 = vld1_dup_u32((const uint32_t *)s); + s += src_stride; + d29u32 = vld1_dup_u32((const uint32_t *)s); + s += src_stride; + d31u32 = vld1_dup_u32((const uint32_t *)s); + s += src_stride; + d30u32 = vld1_dup_u32((const uint32_t *)s); + + __builtin_prefetch(psrc + 64); + + d0x2u16 = vtrn_u16(vreinterpret_u16_u32(d28u32), + vreinterpret_u16_u32(d31u32)); + d1x2u16 = vtrn_u16(vreinterpret_u16_u32(d29u32), + vreinterpret_u16_u32(d30u32)); + d0x2u8 = vtrn_u8(vreinterpret_u8_u16(d0x2u16.val[0]), // d28 + vreinterpret_u8_u16(d1x2u16.val[0])); // d29 + d1x2u8 = vtrn_u8(vreinterpret_u8_u16(d0x2u16.val[1]), // d31 + vreinterpret_u8_u16(d1x2u16.val[1])); // d30 + + __builtin_prefetch(psrc + 64 + src_stride); + + q14u8 = vcombine_u8(d0x2u8.val[0], d0x2u8.val[1]); + q15u8 = vcombine_u8(d1x2u8.val[1], d1x2u8.val[0]); + q0x2u32 = vtrnq_u32(vreinterpretq_u32_u8(q14u8), + vreinterpretq_u32_u8(q15u8)); + + d28u8 = vreinterpret_u8_u32(vget_low_u32(q0x2u32.val[0])); + d29u8 = vreinterpret_u8_u32(vget_high_u32(q0x2u32.val[0])); + q12u16 = vmovl_u8(d28u8); + q13u16 = vmovl_u8(d29u8); + + __builtin_prefetch(psrc + 64 + src_stride * 2); + + d16s16 = vreinterpret_s16_u16(vget_low_u16(q8u16)); + d17s16 = vreinterpret_s16_u16(vget_high_u16(q8u16)); + d18s16 = vreinterpret_s16_u16(vget_low_u16(q9u16)); + d19s16 = vreinterpret_s16_u16(vget_high_u16(q9u16)); + d22s16 = vreinterpret_s16_u16(vget_low_u16(q11u16)); + d24s16 = vreinterpret_s16_u16(vget_low_u16(q12u16)); + d25s16 = vreinterpret_s16_u16(vget_high_u16(q12u16)); + d26s16 = vreinterpret_s16_u16(vget_low_u16(q13u16)); + d27s16 = vreinterpret_s16_u16(vget_high_u16(q13u16)); + + q1s32 = MULTIPLY_BY_Q0(d16s16, d17s16, d20s16, d22s16, + d18s16, d19s16, d23s16, d24s16, q0s16); + q2s32 = MULTIPLY_BY_Q0(d17s16, d20s16, d22s16, d18s16, + d19s16, d23s16, d24s16, d26s16, q0s16); + q14s32 = MULTIPLY_BY_Q0(d20s16, d22s16, d18s16, d19s16, + d23s16, d24s16, d26s16, d27s16, q0s16); + q15s32 = MULTIPLY_BY_Q0(d22s16, d18s16, d19s16, d23s16, + d24s16, d26s16, d27s16, d25s16, q0s16); + + __builtin_prefetch(psrc + 60 + src_stride * 3); + + d2u16 = vqrshrun_n_s32(q1s32, 7); + d3u16 = vqrshrun_n_s32(q2s32, 7); + d4u16 = vqrshrun_n_s32(q14s32, 7); + d5u16 = vqrshrun_n_s32(q15s32, 7); + + q1u16 = vcombine_u16(d2u16, d3u16); + q2u16 = vcombine_u16(d4u16, d5u16); + + d2u8 = vqmovn_u16(q1u16); + d3u8 = vqmovn_u16(q2u16); + + d0x2u16 = vtrn_u16(vreinterpret_u16_u8(d2u8), + vreinterpret_u16_u8(d3u8)); + d0x2u32 = vtrn_u32(vreinterpret_u32_u16(d0x2u16.val[0]), + vreinterpret_u32_u16(d0x2u16.val[1])); + d0x2u8 = vtrn_u8(vreinterpret_u8_u32(d0x2u32.val[0]), + vreinterpret_u8_u32(d0x2u32.val[1])); + + d2u32 = vreinterpret_u32_u8(d0x2u8.val[0]); + d3u32 = vreinterpret_u32_u8(d0x2u8.val[1]); + + d = pdst; + vst1_lane_u32((uint32_t *)d, d2u32, 0); + d += dst_stride; + vst1_lane_u32((uint32_t *)d, d3u32, 0); + d += dst_stride; + vst1_lane_u32((uint32_t *)d, d2u32, 1); + d += dst_stride; + vst1_lane_u32((uint32_t *)d, d3u32, 1); + + q8u16 = q9u16; + d20s16 = d23s16; + q11u16 = q12u16; + q9u16 = q13u16; + d23s16 = vreinterpret_s16_u16(vget_high_u16(q11u16)); + } + } + return; +} + +void vpx_convolve8_vert_neon( + const uint8_t *src, + ptrdiff_t src_stride, + uint8_t *dst, + ptrdiff_t dst_stride, + const int16_t *filter_x, // unused + int x_step_q4, // unused + const int16_t *filter_y, + int y_step_q4, + int w, + int h) { + int height; + const uint8_t *s; + uint8_t *d; + uint32x2_t d2u32, d3u32; + uint32x2_t d16u32, d18u32, d20u32, d22u32, d24u32, d26u32; + int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16; + int16x4_t d24s16, d25s16, d26s16, d27s16; + uint16x4_t d2u16, d3u16, d4u16, d5u16; + int16x8_t q0s16; + uint16x8_t q1u16, q2u16, q8u16, q9u16, q10u16, q11u16, q12u16, q13u16; + int32x4_t q1s32, q2s32, q14s32, q15s32; + + assert(y_step_q4 == 16); + + src -= src_stride * 3; + q0s16 = vld1q_s16(filter_y); + for (; w > 0; w -= 4, src += 4, dst += 4) { // loop_vert_h + s = src; + d16u32 = vld1_lane_u32((const uint32_t *)s, d16u32, 0); + s += src_stride; + d16u32 = vld1_lane_u32((const uint32_t *)s, d16u32, 1); + s += src_stride; + d18u32 = vld1_lane_u32((const uint32_t *)s, d18u32, 0); + s += src_stride; + d18u32 = vld1_lane_u32((const uint32_t *)s, d18u32, 1); + s += src_stride; + d20u32 = vld1_lane_u32((const uint32_t *)s, d20u32, 0); + s += src_stride; + d20u32 = vld1_lane_u32((const uint32_t *)s, d20u32, 1); + s += src_stride; + d22u32 = vld1_lane_u32((const uint32_t *)s, d22u32, 0); + s += src_stride; + + q8u16 = vmovl_u8(vreinterpret_u8_u32(d16u32)); + q9u16 = vmovl_u8(vreinterpret_u8_u32(d18u32)); + q10u16 = vmovl_u8(vreinterpret_u8_u32(d20u32)); + q11u16 = vmovl_u8(vreinterpret_u8_u32(d22u32)); + + d18s16 = vreinterpret_s16_u16(vget_low_u16(q9u16)); + d19s16 = vreinterpret_s16_u16(vget_high_u16(q9u16)); + d22s16 = vreinterpret_s16_u16(vget_low_u16(q11u16)); + d = dst; + for (height = h; height > 0; height -= 4) { // loop_vert + d24u32 = vld1_lane_u32((const uint32_t *)s, d24u32, 0); + s += src_stride; + d26u32 = vld1_lane_u32((const uint32_t *)s, d26u32, 0); + s += src_stride; + d26u32 = vld1_lane_u32((const uint32_t *)s, d26u32, 1); + s += src_stride; + d24u32 = vld1_lane_u32((const uint32_t *)s, d24u32, 1); + s += src_stride; + + q12u16 = vmovl_u8(vreinterpret_u8_u32(d24u32)); + q13u16 = vmovl_u8(vreinterpret_u8_u32(d26u32)); + + d16s16 = vreinterpret_s16_u16(vget_low_u16(q8u16)); + d17s16 = vreinterpret_s16_u16(vget_high_u16(q8u16)); + d20s16 = vreinterpret_s16_u16(vget_low_u16(q10u16)); + d21s16 = vreinterpret_s16_u16(vget_high_u16(q10u16)); + d24s16 = vreinterpret_s16_u16(vget_low_u16(q12u16)); + d25s16 = vreinterpret_s16_u16(vget_high_u16(q12u16)); + d26s16 = vreinterpret_s16_u16(vget_low_u16(q13u16)); + d27s16 = vreinterpret_s16_u16(vget_high_u16(q13u16)); + + __builtin_prefetch(d); + __builtin_prefetch(d + dst_stride); + q1s32 = MULTIPLY_BY_Q0(d16s16, d17s16, d18s16, d19s16, + d20s16, d21s16, d22s16, d24s16, q0s16); + __builtin_prefetch(d + dst_stride * 2); + __builtin_prefetch(d + dst_stride * 3); + q2s32 = MULTIPLY_BY_Q0(d17s16, d18s16, d19s16, d20s16, + d21s16, d22s16, d24s16, d26s16, q0s16); + __builtin_prefetch(s); + __builtin_prefetch(s + src_stride); + q14s32 = MULTIPLY_BY_Q0(d18s16, d19s16, d20s16, d21s16, + d22s16, d24s16, d26s16, d27s16, q0s16); + __builtin_prefetch(s + src_stride * 2); + __builtin_prefetch(s + src_stride * 3); + q15s32 = MULTIPLY_BY_Q0(d19s16, d20s16, d21s16, d22s16, + d24s16, d26s16, d27s16, d25s16, q0s16); + + d2u16 = vqrshrun_n_s32(q1s32, 7); + d3u16 = vqrshrun_n_s32(q2s32, 7); + d4u16 = vqrshrun_n_s32(q14s32, 7); + d5u16 = vqrshrun_n_s32(q15s32, 7); + + q1u16 = vcombine_u16(d2u16, d3u16); + q2u16 = vcombine_u16(d4u16, d5u16); + + d2u32 = vreinterpret_u32_u8(vqmovn_u16(q1u16)); + d3u32 = vreinterpret_u32_u8(vqmovn_u16(q2u16)); + + vst1_lane_u32((uint32_t *)d, d2u32, 0); + d += dst_stride; + vst1_lane_u32((uint32_t *)d, d2u32, 1); + d += dst_stride; + vst1_lane_u32((uint32_t *)d, d3u32, 0); + d += dst_stride; + vst1_lane_u32((uint32_t *)d, d3u32, 1); + d += dst_stride; + + q8u16 = q10u16; + d18s16 = d22s16; + d19s16 = d24s16; + q10u16 = q13u16; + d22s16 = d25s16; + } + } + return; +} diff --git a/thirdparty/libvpx/vpx_dsp/arm/vpx_convolve_avg_neon.c b/thirdparty/libvpx/vpx_dsp/arm/vpx_convolve_avg_neon.c new file mode 100644 index 0000000000..dc58a332f8 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/arm/vpx_convolve_avg_neon.c @@ -0,0 +1,147 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +#include "./vpx_dsp_rtcd.h" +#include "vpx/vpx_integer.h" + +void vpx_convolve_avg_neon( + const uint8_t *src, // r0 + ptrdiff_t src_stride, // r1 + uint8_t *dst, // r2 + ptrdiff_t dst_stride, // r3 + const int16_t *filter_x, + int filter_x_stride, + const int16_t *filter_y, + int filter_y_stride, + int w, + int h) { + uint8_t *d; + uint8x8_t d0u8, d1u8, d2u8, d3u8; + uint32x2_t d0u32, d2u32; + uint8x16_t q0u8, q1u8, q2u8, q3u8, q8u8, q9u8, q10u8, q11u8; + (void)filter_x; (void)filter_x_stride; + (void)filter_y; (void)filter_y_stride; + + d = dst; + if (w > 32) { // avg64 + for (; h > 0; h -= 1) { + q0u8 = vld1q_u8(src); + q1u8 = vld1q_u8(src + 16); + q2u8 = vld1q_u8(src + 32); + q3u8 = vld1q_u8(src + 48); + src += src_stride; + q8u8 = vld1q_u8(d); + q9u8 = vld1q_u8(d + 16); + q10u8 = vld1q_u8(d + 32); + q11u8 = vld1q_u8(d + 48); + d += dst_stride; + + q0u8 = vrhaddq_u8(q0u8, q8u8); + q1u8 = vrhaddq_u8(q1u8, q9u8); + q2u8 = vrhaddq_u8(q2u8, q10u8); + q3u8 = vrhaddq_u8(q3u8, q11u8); + + vst1q_u8(dst, q0u8); + vst1q_u8(dst + 16, q1u8); + vst1q_u8(dst + 32, q2u8); + vst1q_u8(dst + 48, q3u8); + dst += dst_stride; + } + } else if (w == 32) { // avg32 + for (; h > 0; h -= 2) { + q0u8 = vld1q_u8(src); + q1u8 = vld1q_u8(src + 16); + src += src_stride; + q2u8 = vld1q_u8(src); + q3u8 = vld1q_u8(src + 16); + src += src_stride; + q8u8 = vld1q_u8(d); + q9u8 = vld1q_u8(d + 16); + d += dst_stride; + q10u8 = vld1q_u8(d); + q11u8 = vld1q_u8(d + 16); + d += dst_stride; + + q0u8 = vrhaddq_u8(q0u8, q8u8); + q1u8 = vrhaddq_u8(q1u8, q9u8); + q2u8 = vrhaddq_u8(q2u8, q10u8); + q3u8 = vrhaddq_u8(q3u8, q11u8); + + vst1q_u8(dst, q0u8); + vst1q_u8(dst + 16, q1u8); + dst += dst_stride; + vst1q_u8(dst, q2u8); + vst1q_u8(dst + 16, q3u8); + dst += dst_stride; + } + } else if (w > 8) { // avg16 + for (; h > 0; h -= 2) { + q0u8 = vld1q_u8(src); + src += src_stride; + q1u8 = vld1q_u8(src); + src += src_stride; + q2u8 = vld1q_u8(d); + d += dst_stride; + q3u8 = vld1q_u8(d); + d += dst_stride; + + q0u8 = vrhaddq_u8(q0u8, q2u8); + q1u8 = vrhaddq_u8(q1u8, q3u8); + + vst1q_u8(dst, q0u8); + dst += dst_stride; + vst1q_u8(dst, q1u8); + dst += dst_stride; + } + } else if (w == 8) { // avg8 + for (; h > 0; h -= 2) { + d0u8 = vld1_u8(src); + src += src_stride; + d1u8 = vld1_u8(src); + src += src_stride; + d2u8 = vld1_u8(d); + d += dst_stride; + d3u8 = vld1_u8(d); + d += dst_stride; + + q0u8 = vcombine_u8(d0u8, d1u8); + q1u8 = vcombine_u8(d2u8, d3u8); + q0u8 = vrhaddq_u8(q0u8, q1u8); + + vst1_u8(dst, vget_low_u8(q0u8)); + dst += dst_stride; + vst1_u8(dst, vget_high_u8(q0u8)); + dst += dst_stride; + } + } else { // avg4 + for (; h > 0; h -= 2) { + d0u32 = vld1_lane_u32((const uint32_t *)src, d0u32, 0); + src += src_stride; + d0u32 = vld1_lane_u32((const uint32_t *)src, d0u32, 1); + src += src_stride; + d2u32 = vld1_lane_u32((const uint32_t *)d, d2u32, 0); + d += dst_stride; + d2u32 = vld1_lane_u32((const uint32_t *)d, d2u32, 1); + d += dst_stride; + + d0u8 = vrhadd_u8(vreinterpret_u8_u32(d0u32), + vreinterpret_u8_u32(d2u32)); + + d0u32 = vreinterpret_u32_u8(d0u8); + vst1_lane_u32((uint32_t *)dst, d0u32, 0); + dst += dst_stride; + vst1_lane_u32((uint32_t *)dst, d0u32, 1); + dst += dst_stride; + } + } + return; +} diff --git a/thirdparty/libvpx/vpx_dsp/arm/vpx_convolve_copy_neon.c b/thirdparty/libvpx/vpx_dsp/arm/vpx_convolve_copy_neon.c new file mode 100644 index 0000000000..d8fb97a861 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/arm/vpx_convolve_copy_neon.c @@ -0,0 +1,94 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> + +#include "./vpx_dsp_rtcd.h" +#include "vpx/vpx_integer.h" + +void vpx_convolve_copy_neon( + const uint8_t *src, // r0 + ptrdiff_t src_stride, // r1 + uint8_t *dst, // r2 + ptrdiff_t dst_stride, // r3 + const int16_t *filter_x, + int filter_x_stride, + const int16_t *filter_y, + int filter_y_stride, + int w, + int h) { + uint8x8_t d0u8, d2u8; + uint8x16_t q0u8, q1u8, q2u8, q3u8; + (void)filter_x; (void)filter_x_stride; + (void)filter_y; (void)filter_y_stride; + + if (w > 32) { // copy64 + for (; h > 0; h--) { + q0u8 = vld1q_u8(src); + q1u8 = vld1q_u8(src + 16); + q2u8 = vld1q_u8(src + 32); + q3u8 = vld1q_u8(src + 48); + src += src_stride; + + vst1q_u8(dst, q0u8); + vst1q_u8(dst + 16, q1u8); + vst1q_u8(dst + 32, q2u8); + vst1q_u8(dst + 48, q3u8); + dst += dst_stride; + } + } else if (w == 32) { // copy32 + for (; h > 0; h -= 2) { + q0u8 = vld1q_u8(src); + q1u8 = vld1q_u8(src + 16); + src += src_stride; + q2u8 = vld1q_u8(src); + q3u8 = vld1q_u8(src + 16); + src += src_stride; + + vst1q_u8(dst, q0u8); + vst1q_u8(dst + 16, q1u8); + dst += dst_stride; + vst1q_u8(dst, q2u8); + vst1q_u8(dst + 16, q3u8); + dst += dst_stride; + } + } else if (w > 8) { // copy16 + for (; h > 0; h -= 2) { + q0u8 = vld1q_u8(src); + src += src_stride; + q1u8 = vld1q_u8(src); + src += src_stride; + + vst1q_u8(dst, q0u8); + dst += dst_stride; + vst1q_u8(dst, q1u8); + dst += dst_stride; + } + } else if (w == 8) { // copy8 + for (; h > 0; h -= 2) { + d0u8 = vld1_u8(src); + src += src_stride; + d2u8 = vld1_u8(src); + src += src_stride; + + vst1_u8(dst, d0u8); + dst += dst_stride; + vst1_u8(dst, d2u8); + dst += dst_stride; + } + } else { // copy4 + for (; h > 0; h--) { + *(uint32_t *)dst = *(const uint32_t *)src; + src += src_stride; + dst += dst_stride; + } + } + return; +} diff --git a/thirdparty/libvpx/vpx_dsp/arm/vpx_convolve_neon.c b/thirdparty/libvpx/vpx_dsp/arm/vpx_convolve_neon.c new file mode 100644 index 0000000000..1506ce6203 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/arm/vpx_convolve_neon.c @@ -0,0 +1,72 @@ +/* + * Copyright (c) 2013 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <assert.h> + +#include "./vpx_dsp_rtcd.h" +#include "vpx_dsp/vpx_dsp_common.h" +#include "vpx_ports/mem.h" + +void vpx_convolve8_neon(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h) { + /* Given our constraints: w <= 64, h <= 64, taps == 8 we can reduce the + * maximum buffer size to 64 * 64 + 7 (+ 1 to make it divisible by 4). + */ + DECLARE_ALIGNED(8, uint8_t, temp[64 * 72]); + + // Account for the vertical phase needing 3 lines prior and 4 lines post + int intermediate_height = h + 7; + + assert(y_step_q4 == 16); + assert(x_step_q4 == 16); + + /* Filter starting 3 lines back. The neon implementation will ignore the + * given height and filter a multiple of 4 lines. Since this goes in to + * the temp buffer which has lots of extra room and is subsequently discarded + * this is safe if somewhat less than ideal. + */ + vpx_convolve8_horiz_neon(src - src_stride * 3, src_stride, + temp, 64, + filter_x, x_step_q4, filter_y, y_step_q4, + w, intermediate_height); + + /* Step into the temp buffer 3 lines to get the actual frame data */ + vpx_convolve8_vert_neon(temp + 64 * 3, 64, + dst, dst_stride, + filter_x, x_step_q4, filter_y, y_step_q4, + w, h); +} + +void vpx_convolve8_avg_neon(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h) { + DECLARE_ALIGNED(8, uint8_t, temp[64 * 72]); + int intermediate_height = h + 7; + + assert(y_step_q4 == 16); + assert(x_step_q4 == 16); + + /* This implementation has the same issues as above. In addition, we only want + * to average the values after both passes. + */ + vpx_convolve8_horiz_neon(src - src_stride * 3, src_stride, + temp, 64, + filter_x, x_step_q4, filter_y, y_step_q4, + w, intermediate_height); + vpx_convolve8_avg_vert_neon(temp + 64 * 3, + 64, dst, dst_stride, + filter_x, x_step_q4, filter_y, y_step_q4, + w, h); +} diff --git a/thirdparty/libvpx/vpx_dsp/bitreader.c b/thirdparty/libvpx/vpx_dsp/bitreader.c new file mode 100644 index 0000000000..8140e78e70 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/bitreader.c @@ -0,0 +1,103 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ +#include <stdlib.h> + +#include "./vpx_config.h" + +#include "vpx_dsp/bitreader.h" +#include "vpx_dsp/prob.h" +#include "vpx_dsp/vpx_dsp_common.h" +#include "vpx_ports/mem.h" +#include "vpx_mem/vpx_mem.h" +#include "vpx_util/endian_inl.h" + +int vpx_reader_init(vpx_reader *r, + const uint8_t *buffer, + size_t size, + vpx_decrypt_cb decrypt_cb, + void *decrypt_state) { + if (size && !buffer) { + return 1; + } else { + r->buffer_end = buffer + size; + r->buffer = buffer; + r->value = 0; + r->count = -8; + r->range = 255; + r->decrypt_cb = decrypt_cb; + r->decrypt_state = decrypt_state; + vpx_reader_fill(r); + return vpx_read_bit(r) != 0; // marker bit + } +} + +void vpx_reader_fill(vpx_reader *r) { + const uint8_t *const buffer_end = r->buffer_end; + const uint8_t *buffer = r->buffer; + const uint8_t *buffer_start = buffer; + BD_VALUE value = r->value; + int count = r->count; + const size_t bytes_left = buffer_end - buffer; + const size_t bits_left = bytes_left * CHAR_BIT; + int shift = BD_VALUE_SIZE - CHAR_BIT - (count + CHAR_BIT); + + if (r->decrypt_cb) { + size_t n = VPXMIN(sizeof(r->clear_buffer), bytes_left); + r->decrypt_cb(r->decrypt_state, buffer, r->clear_buffer, (int)n); + buffer = r->clear_buffer; + buffer_start = r->clear_buffer; + } + if (bits_left > BD_VALUE_SIZE) { + const int bits = (shift & 0xfffffff8) + CHAR_BIT; + BD_VALUE nv; + BD_VALUE big_endian_values; + memcpy(&big_endian_values, buffer, sizeof(BD_VALUE)); +#if SIZE_MAX == 0xffffffffffffffffULL + big_endian_values = HToBE64(big_endian_values); +#else + big_endian_values = HToBE32(big_endian_values); +#endif + nv = big_endian_values >> (BD_VALUE_SIZE - bits); + count += bits; + buffer += (bits >> 3); + value = r->value | (nv << (shift & 0x7)); + } else { + const int bits_over = (int)(shift + CHAR_BIT - (int)bits_left); + int loop_end = 0; + if (bits_over >= 0) { + count += LOTS_OF_BITS; + loop_end = bits_over; + } + + if (bits_over < 0 || bits_left) { + while (shift >= loop_end) { + count += CHAR_BIT; + value |= (BD_VALUE)*buffer++ << shift; + shift -= CHAR_BIT; + } + } + } + + // NOTE: Variable 'buffer' may not relate to 'r->buffer' after decryption, + // so we increase 'r->buffer' by the amount that 'buffer' moved, rather than + // assign 'buffer' to 'r->buffer'. + r->buffer += buffer - buffer_start; + r->value = value; + r->count = count; +} + +const uint8_t *vpx_reader_find_end(vpx_reader *r) { + // Find the end of the coded buffer + while (r->count > CHAR_BIT && r->count < BD_VALUE_SIZE) { + r->count -= CHAR_BIT; + r->buffer--; + } + return r->buffer; +} diff --git a/thirdparty/libvpx/vpx_dsp/bitreader.h b/thirdparty/libvpx/vpx_dsp/bitreader.h new file mode 100644 index 0000000000..9a441b4107 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/bitreader.h @@ -0,0 +1,140 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VPX_DSP_BITREADER_H_ +#define VPX_DSP_BITREADER_H_ + +#include <stddef.h> +#include <limits.h> + +#include "./vpx_config.h" +#include "vpx_ports/mem.h" +#include "vpx/vp8dx.h" +#include "vpx/vpx_integer.h" +#include "vpx_dsp/prob.h" + +#ifdef __cplusplus +extern "C" { +#endif + +typedef size_t BD_VALUE; + +#define BD_VALUE_SIZE ((int)sizeof(BD_VALUE) * CHAR_BIT) + +// This is meant to be a large, positive constant that can still be efficiently +// loaded as an immediate (on platforms like ARM, for example). +// Even relatively modest values like 100 would work fine. +#define LOTS_OF_BITS 0x40000000 + +typedef struct { + // Be careful when reordering this struct, it may impact the cache negatively. + BD_VALUE value; + unsigned int range; + int count; + const uint8_t *buffer_end; + const uint8_t *buffer; + vpx_decrypt_cb decrypt_cb; + void *decrypt_state; + uint8_t clear_buffer[sizeof(BD_VALUE) + 1]; +} vpx_reader; + +int vpx_reader_init(vpx_reader *r, + const uint8_t *buffer, + size_t size, + vpx_decrypt_cb decrypt_cb, + void *decrypt_state); + +void vpx_reader_fill(vpx_reader *r); + +const uint8_t *vpx_reader_find_end(vpx_reader *r); + +static INLINE int vpx_reader_has_error(vpx_reader *r) { + // Check if we have reached the end of the buffer. + // + // Variable 'count' stores the number of bits in the 'value' buffer, minus + // 8. The top byte is part of the algorithm, and the remainder is buffered + // to be shifted into it. So if count == 8, the top 16 bits of 'value' are + // occupied, 8 for the algorithm and 8 in the buffer. + // + // When reading a byte from the user's buffer, count is filled with 8 and + // one byte is filled into the value buffer. When we reach the end of the + // data, count is additionally filled with LOTS_OF_BITS. So when + // count == LOTS_OF_BITS - 1, the user's data has been exhausted. + // + // 1 if we have tried to decode bits after the end of stream was encountered. + // 0 No error. + return r->count > BD_VALUE_SIZE && r->count < LOTS_OF_BITS; +} + +static INLINE int vpx_read(vpx_reader *r, int prob) { + unsigned int bit = 0; + BD_VALUE value; + BD_VALUE bigsplit; + int count; + unsigned int range; + unsigned int split = (r->range * prob + (256 - prob)) >> CHAR_BIT; + + if (r->count < 0) + vpx_reader_fill(r); + + value = r->value; + count = r->count; + + bigsplit = (BD_VALUE)split << (BD_VALUE_SIZE - CHAR_BIT); + + range = split; + + if (value >= bigsplit) { + range = r->range - split; + value = value - bigsplit; + bit = 1; + } + + { + register int shift = vpx_norm[range]; + range <<= shift; + value <<= shift; + count -= shift; + } + r->value = value; + r->count = count; + r->range = range; + + return bit; +} + +static INLINE int vpx_read_bit(vpx_reader *r) { + return vpx_read(r, 128); // vpx_prob_half +} + +static INLINE int vpx_read_literal(vpx_reader *r, int bits) { + int literal = 0, bit; + + for (bit = bits - 1; bit >= 0; bit--) + literal |= vpx_read_bit(r) << bit; + + return literal; +} + +static INLINE int vpx_read_tree(vpx_reader *r, const vpx_tree_index *tree, + const vpx_prob *probs) { + vpx_tree_index i = 0; + + while ((i = tree[i + vpx_read(r, probs[i >> 1])]) > 0) + continue; + + return -i; +} + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VPX_DSP_BITREADER_H_ diff --git a/thirdparty/libvpx/vpx_dsp/bitreader_buffer.c b/thirdparty/libvpx/vpx_dsp/bitreader_buffer.c new file mode 100644 index 0000000000..d7b55cf9f4 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/bitreader_buffer.c @@ -0,0 +1,53 @@ +/* + * Copyright (c) 2013 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ +#include "./vpx_config.h" +#include "./bitreader_buffer.h" + +size_t vpx_rb_bytes_read(struct vpx_read_bit_buffer *rb) { + return (rb->bit_offset + 7) >> 3; +} + +int vpx_rb_read_bit(struct vpx_read_bit_buffer *rb) { + const size_t off = rb->bit_offset; + const size_t p = off >> 3; + const int q = 7 - (int)(off & 0x7); + if (rb->bit_buffer + p < rb->bit_buffer_end) { + const int bit = (rb->bit_buffer[p] >> q) & 1; + rb->bit_offset = off + 1; + return bit; + } else { + rb->error_handler(rb->error_handler_data); + return 0; + } +} + +int vpx_rb_read_literal(struct vpx_read_bit_buffer *rb, int bits) { + int value = 0, bit; + for (bit = bits - 1; bit >= 0; bit--) + value |= vpx_rb_read_bit(rb) << bit; + return value; +} + +int vpx_rb_read_signed_literal(struct vpx_read_bit_buffer *rb, + int bits) { + const int value = vpx_rb_read_literal(rb, bits); + return vpx_rb_read_bit(rb) ? -value : value; +} + +int vpx_rb_read_inv_signed_literal(struct vpx_read_bit_buffer *rb, + int bits) { +#if CONFIG_MISC_FIXES + const int nbits = sizeof(unsigned) * 8 - bits - 1; + const unsigned value = (unsigned)vpx_rb_read_literal(rb, bits + 1) << nbits; + return ((int) value) >> nbits; +#else + return vpx_rb_read_signed_literal(rb, bits); +#endif +} diff --git a/thirdparty/libvpx/vpx_dsp/bitreader_buffer.h b/thirdparty/libvpx/vpx_dsp/bitreader_buffer.h new file mode 100644 index 0000000000..8a48a95ed1 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/bitreader_buffer.h @@ -0,0 +1,47 @@ +/* + * Copyright (c) 2013 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VPX_DSP_BITREADER_BUFFER_H_ +#define VPX_DSP_BITREADER_BUFFER_H_ + +#include <limits.h> + +#include "vpx/vpx_integer.h" + +#ifdef __cplusplus +extern "C" { +#endif + +typedef void (*vpx_rb_error_handler)(void *data); + +struct vpx_read_bit_buffer { + const uint8_t *bit_buffer; + const uint8_t *bit_buffer_end; + size_t bit_offset; + + void *error_handler_data; + vpx_rb_error_handler error_handler; +}; + +size_t vpx_rb_bytes_read(struct vpx_read_bit_buffer *rb); + +int vpx_rb_read_bit(struct vpx_read_bit_buffer *rb); + +int vpx_rb_read_literal(struct vpx_read_bit_buffer *rb, int bits); + +int vpx_rb_read_signed_literal(struct vpx_read_bit_buffer *rb, int bits); + +int vpx_rb_read_inv_signed_literal(struct vpx_read_bit_buffer *rb, int bits); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VPX_DSP_BITREADER_BUFFER_H_ diff --git a/thirdparty/libvpx/vpx_dsp/intrapred.c b/thirdparty/libvpx/vpx_dsp/intrapred.c new file mode 100644 index 0000000000..cc4a74bd26 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/intrapred.c @@ -0,0 +1,870 @@ +/* + * Copyright (c) 2015 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "./vpx_config.h" +#include "./vpx_dsp_rtcd.h" + +#include "vpx_dsp/vpx_dsp_common.h" +#include "vpx_mem/vpx_mem.h" + +#define DST(x, y) dst[(x) + (y) * stride] +#define AVG3(a, b, c) (((a) + 2 * (b) + (c) + 2) >> 2) +#define AVG2(a, b) (((a) + (b) + 1) >> 1) + +static INLINE void d207_predictor(uint8_t *dst, ptrdiff_t stride, int bs, + const uint8_t *above, const uint8_t *left) { + int r, c; + (void) above; + // first column + for (r = 0; r < bs - 1; ++r) + dst[r * stride] = AVG2(left[r], left[r + 1]); + dst[(bs - 1) * stride] = left[bs - 1]; + dst++; + + // second column + for (r = 0; r < bs - 2; ++r) + dst[r * stride] = AVG3(left[r], left[r + 1], left[r + 2]); + dst[(bs - 2) * stride] = AVG3(left[bs - 2], left[bs - 1], left[bs - 1]); + dst[(bs - 1) * stride] = left[bs - 1]; + dst++; + + // rest of last row + for (c = 0; c < bs - 2; ++c) + dst[(bs - 1) * stride + c] = left[bs - 1]; + + for (r = bs - 2; r >= 0; --r) + for (c = 0; c < bs - 2; ++c) + dst[r * stride + c] = dst[(r + 1) * stride + c - 2]; +} + +#if CONFIG_MISC_FIXES +static INLINE void d207e_predictor(uint8_t *dst, ptrdiff_t stride, int bs, + const uint8_t *above, const uint8_t *left) { + int r, c; + (void) above; + + for (r = 0; r < bs; ++r) { + for (c = 0; c < bs; ++c) { + dst[c] = c & 1 ? AVG3(left[(c >> 1) + r], left[(c >> 1) + r + 1], + left[(c >> 1) + r + 2]) + : AVG2(left[(c >> 1) + r], left[(c >> 1) + r + 1]); + } + dst += stride; + } +} +#endif // CONFIG_MISC_FIXES + +static INLINE void d63_predictor(uint8_t *dst, ptrdiff_t stride, int bs, + const uint8_t *above, const uint8_t *left) { + int r, c; + int size; + (void)left; + for (c = 0; c < bs; ++c) { + dst[c] = AVG2(above[c], above[c + 1]); + dst[stride + c] = AVG3(above[c], above[c + 1], above[c + 2]); + } + for (r = 2, size = bs - 2; r < bs; r += 2, --size) { + memcpy(dst + (r + 0) * stride, dst + (r >> 1), size); + memset(dst + (r + 0) * stride + size, above[bs - 1], bs - size); + memcpy(dst + (r + 1) * stride, dst + stride + (r >> 1), size); + memset(dst + (r + 1) * stride + size, above[bs - 1], bs - size); + } +} + +#if CONFIG_MISC_FIXES +static INLINE void d63e_predictor(uint8_t *dst, ptrdiff_t stride, int bs, + const uint8_t *above, const uint8_t *left) { + int r, c; + (void) left; + for (r = 0; r < bs; ++r) { + for (c = 0; c < bs; ++c) { + dst[c] = r & 1 ? AVG3(above[(r >> 1) + c], above[(r >> 1) + c + 1], + above[(r >> 1) + c + 2]) + : AVG2(above[(r >> 1) + c], above[(r >> 1) + c + 1]); + } + dst += stride; + } +} +#endif // CONFIG_MISC_FIXES + +static INLINE void d45_predictor(uint8_t *dst, ptrdiff_t stride, int bs, + const uint8_t *above, const uint8_t *left) { + const uint8_t above_right = above[bs - 1]; + const uint8_t *const dst_row0 = dst; + int x, size; + (void)left; + + for (x = 0; x < bs - 1; ++x) { + dst[x] = AVG3(above[x], above[x + 1], above[x + 2]); + } + dst[bs - 1] = above_right; + dst += stride; + for (x = 1, size = bs - 2; x < bs; ++x, --size) { + memcpy(dst, dst_row0 + x, size); + memset(dst + size, above_right, x + 1); + dst += stride; + } +} + +#if CONFIG_MISC_FIXES +static INLINE void d45e_predictor(uint8_t *dst, ptrdiff_t stride, int bs, + const uint8_t *above, const uint8_t *left) { + int r, c; + (void) left; + for (r = 0; r < bs; ++r) { + for (c = 0; c < bs; ++c) { + dst[c] = AVG3(above[r + c], above[r + c + 1], + above[r + c + 1 + (r + c + 2 < bs * 2)]); + } + dst += stride; + } +} +#endif // CONFIG_MISC_FIXES + +static INLINE void d117_predictor(uint8_t *dst, ptrdiff_t stride, int bs, + const uint8_t *above, const uint8_t *left) { + int r, c; + + // first row + for (c = 0; c < bs; c++) + dst[c] = AVG2(above[c - 1], above[c]); + dst += stride; + + // second row + dst[0] = AVG3(left[0], above[-1], above[0]); + for (c = 1; c < bs; c++) + dst[c] = AVG3(above[c - 2], above[c - 1], above[c]); + dst += stride; + + // the rest of first col + dst[0] = AVG3(above[-1], left[0], left[1]); + for (r = 3; r < bs; ++r) + dst[(r - 2) * stride] = AVG3(left[r - 3], left[r - 2], left[r - 1]); + + // the rest of the block + for (r = 2; r < bs; ++r) { + for (c = 1; c < bs; c++) + dst[c] = dst[-2 * stride + c - 1]; + dst += stride; + } +} + +static INLINE void d135_predictor(uint8_t *dst, ptrdiff_t stride, int bs, + const uint8_t *above, const uint8_t *left) { + int i; +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ > 7 + // silence a spurious -Warray-bounds warning, possibly related to: + // https://gcc.gnu.org/bugzilla/show_bug.cgi?id=56273 + uint8_t border[69]; +#else + uint8_t border[32 + 32 - 1]; // outer border from bottom-left to top-right +#endif + + // dst(bs, bs - 2)[0], i.e., border starting at bottom-left + for (i = 0; i < bs - 2; ++i) { + border[i] = AVG3(left[bs - 3 - i], left[bs - 2 - i], left[bs - 1 - i]); + } + border[bs - 2] = AVG3(above[-1], left[0], left[1]); + border[bs - 1] = AVG3(left[0], above[-1], above[0]); + border[bs - 0] = AVG3(above[-1], above[0], above[1]); + // dst[0][2, size), i.e., remaining top border ascending + for (i = 0; i < bs - 2; ++i) { + border[bs + 1 + i] = AVG3(above[i], above[i + 1], above[i + 2]); + } + + for (i = 0; i < bs; ++i) { + memcpy(dst + i * stride, border + bs - 1 - i, bs); + } +} + +static INLINE void d153_predictor(uint8_t *dst, ptrdiff_t stride, int bs, + const uint8_t *above, const uint8_t *left) { + int r, c; + dst[0] = AVG2(above[-1], left[0]); + for (r = 1; r < bs; r++) + dst[r * stride] = AVG2(left[r - 1], left[r]); + dst++; + + dst[0] = AVG3(left[0], above[-1], above[0]); + dst[stride] = AVG3(above[-1], left[0], left[1]); + for (r = 2; r < bs; r++) + dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]); + dst++; + + for (c = 0; c < bs - 2; c++) + dst[c] = AVG3(above[c - 1], above[c], above[c + 1]); + dst += stride; + + for (r = 1; r < bs; ++r) { + for (c = 0; c < bs - 2; c++) + dst[c] = dst[-stride + c - 2]; + dst += stride; + } +} + +static INLINE void v_predictor(uint8_t *dst, ptrdiff_t stride, int bs, + const uint8_t *above, const uint8_t *left) { + int r; + (void) left; + + for (r = 0; r < bs; r++) { + memcpy(dst, above, bs); + dst += stride; + } +} + +static INLINE void h_predictor(uint8_t *dst, ptrdiff_t stride, int bs, + const uint8_t *above, const uint8_t *left) { + int r; + (void) above; + + for (r = 0; r < bs; r++) { + memset(dst, left[r], bs); + dst += stride; + } +} + +static INLINE void tm_predictor(uint8_t *dst, ptrdiff_t stride, int bs, + const uint8_t *above, const uint8_t *left) { + int r, c; + int ytop_left = above[-1]; + + for (r = 0; r < bs; r++) { + for (c = 0; c < bs; c++) + dst[c] = clip_pixel(left[r] + above[c] - ytop_left); + dst += stride; + } +} + +static INLINE void dc_128_predictor(uint8_t *dst, ptrdiff_t stride, int bs, + const uint8_t *above, const uint8_t *left) { + int r; + (void) above; + (void) left; + + for (r = 0; r < bs; r++) { + memset(dst, 128, bs); + dst += stride; + } +} + +static INLINE void dc_left_predictor(uint8_t *dst, ptrdiff_t stride, int bs, + const uint8_t *above, + const uint8_t *left) { + int i, r, expected_dc, sum = 0; + (void) above; + + for (i = 0; i < bs; i++) + sum += left[i]; + expected_dc = (sum + (bs >> 1)) / bs; + + for (r = 0; r < bs; r++) { + memset(dst, expected_dc, bs); + dst += stride; + } +} + +static INLINE void dc_top_predictor(uint8_t *dst, ptrdiff_t stride, int bs, + const uint8_t *above, const uint8_t *left) { + int i, r, expected_dc, sum = 0; + (void) left; + + for (i = 0; i < bs; i++) + sum += above[i]; + expected_dc = (sum + (bs >> 1)) / bs; + + for (r = 0; r < bs; r++) { + memset(dst, expected_dc, bs); + dst += stride; + } +} + +static INLINE void dc_predictor(uint8_t *dst, ptrdiff_t stride, int bs, + const uint8_t *above, const uint8_t *left) { + int i, r, expected_dc, sum = 0; + const int count = 2 * bs; + + for (i = 0; i < bs; i++) { + sum += above[i]; + sum += left[i]; + } + + expected_dc = (sum + (count >> 1)) / count; + + for (r = 0; r < bs; r++) { + memset(dst, expected_dc, bs); + dst += stride; + } +} + +void vpx_he_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + const int H = above[-1]; + const int I = left[0]; + const int J = left[1]; + const int K = left[2]; + const int L = left[3]; + + memset(dst + stride * 0, AVG3(H, I, J), 4); + memset(dst + stride * 1, AVG3(I, J, K), 4); + memset(dst + stride * 2, AVG3(J, K, L), 4); + memset(dst + stride * 3, AVG3(K, L, L), 4); +} + +void vpx_ve_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + const int H = above[-1]; + const int I = above[0]; + const int J = above[1]; + const int K = above[2]; + const int L = above[3]; + const int M = above[4]; + (void)left; + + dst[0] = AVG3(H, I, J); + dst[1] = AVG3(I, J, K); + dst[2] = AVG3(J, K, L); + dst[3] = AVG3(K, L, M); + memcpy(dst + stride * 1, dst, 4); + memcpy(dst + stride * 2, dst, 4); + memcpy(dst + stride * 3, dst, 4); +} + +void vpx_d207_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + const int I = left[0]; + const int J = left[1]; + const int K = left[2]; + const int L = left[3]; + (void)above; + DST(0, 0) = AVG2(I, J); + DST(2, 0) = DST(0, 1) = AVG2(J, K); + DST(2, 1) = DST(0, 2) = AVG2(K, L); + DST(1, 0) = AVG3(I, J, K); + DST(3, 0) = DST(1, 1) = AVG3(J, K, L); + DST(3, 1) = DST(1, 2) = AVG3(K, L, L); + DST(3, 2) = DST(2, 2) = + DST(0, 3) = DST(1, 3) = DST(2, 3) = DST(3, 3) = L; +} + +void vpx_d63_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + const int A = above[0]; + const int B = above[1]; + const int C = above[2]; + const int D = above[3]; + const int E = above[4]; + const int F = above[5]; + const int G = above[6]; + (void)left; + DST(0, 0) = AVG2(A, B); + DST(1, 0) = DST(0, 2) = AVG2(B, C); + DST(2, 0) = DST(1, 2) = AVG2(C, D); + DST(3, 0) = DST(2, 2) = AVG2(D, E); + DST(3, 2) = AVG2(E, F); // differs from vp8 + + DST(0, 1) = AVG3(A, B, C); + DST(1, 1) = DST(0, 3) = AVG3(B, C, D); + DST(2, 1) = DST(1, 3) = AVG3(C, D, E); + DST(3, 1) = DST(2, 3) = AVG3(D, E, F); + DST(3, 3) = AVG3(E, F, G); // differs from vp8 +} + +void vpx_d63f_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + const int A = above[0]; + const int B = above[1]; + const int C = above[2]; + const int D = above[3]; + const int E = above[4]; + const int F = above[5]; + const int G = above[6]; + const int H = above[7]; + (void)left; + DST(0, 0) = AVG2(A, B); + DST(1, 0) = DST(0, 2) = AVG2(B, C); + DST(2, 0) = DST(1, 2) = AVG2(C, D); + DST(3, 0) = DST(2, 2) = AVG2(D, E); + DST(3, 2) = AVG3(E, F, G); + + DST(0, 1) = AVG3(A, B, C); + DST(1, 1) = DST(0, 3) = AVG3(B, C, D); + DST(2, 1) = DST(1, 3) = AVG3(C, D, E); + DST(3, 1) = DST(2, 3) = AVG3(D, E, F); + DST(3, 3) = AVG3(F, G, H); +} + +void vpx_d45_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + const int A = above[0]; + const int B = above[1]; + const int C = above[2]; + const int D = above[3]; + const int E = above[4]; + const int F = above[5]; + const int G = above[6]; + const int H = above[7]; + (void)stride; + (void)left; + DST(0, 0) = AVG3(A, B, C); + DST(1, 0) = DST(0, 1) = AVG3(B, C, D); + DST(2, 0) = DST(1, 1) = DST(0, 2) = AVG3(C, D, E); + DST(3, 0) = DST(2, 1) = DST(1, 2) = DST(0, 3) = AVG3(D, E, F); + DST(3, 1) = DST(2, 2) = DST(1, 3) = AVG3(E, F, G); + DST(3, 2) = DST(2, 3) = AVG3(F, G, H); + DST(3, 3) = H; // differs from vp8 +} + +void vpx_d45e_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + const int A = above[0]; + const int B = above[1]; + const int C = above[2]; + const int D = above[3]; + const int E = above[4]; + const int F = above[5]; + const int G = above[6]; + const int H = above[7]; + (void)stride; + (void)left; + DST(0, 0) = AVG3(A, B, C); + DST(1, 0) = DST(0, 1) = AVG3(B, C, D); + DST(2, 0) = DST(1, 1) = DST(0, 2) = AVG3(C, D, E); + DST(3, 0) = DST(2, 1) = DST(1, 2) = DST(0, 3) = AVG3(D, E, F); + DST(3, 1) = DST(2, 2) = DST(1, 3) = AVG3(E, F, G); + DST(3, 2) = DST(2, 3) = AVG3(F, G, H); + DST(3, 3) = AVG3(G, H, H); +} + +void vpx_d117_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + const int I = left[0]; + const int J = left[1]; + const int K = left[2]; + const int X = above[-1]; + const int A = above[0]; + const int B = above[1]; + const int C = above[2]; + const int D = above[3]; + DST(0, 0) = DST(1, 2) = AVG2(X, A); + DST(1, 0) = DST(2, 2) = AVG2(A, B); + DST(2, 0) = DST(3, 2) = AVG2(B, C); + DST(3, 0) = AVG2(C, D); + + DST(0, 3) = AVG3(K, J, I); + DST(0, 2) = AVG3(J, I, X); + DST(0, 1) = DST(1, 3) = AVG3(I, X, A); + DST(1, 1) = DST(2, 3) = AVG3(X, A, B); + DST(2, 1) = DST(3, 3) = AVG3(A, B, C); + DST(3, 1) = AVG3(B, C, D); +} + +void vpx_d135_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + const int I = left[0]; + const int J = left[1]; + const int K = left[2]; + const int L = left[3]; + const int X = above[-1]; + const int A = above[0]; + const int B = above[1]; + const int C = above[2]; + const int D = above[3]; + (void)stride; + DST(0, 3) = AVG3(J, K, L); + DST(1, 3) = DST(0, 2) = AVG3(I, J, K); + DST(2, 3) = DST(1, 2) = DST(0, 1) = AVG3(X, I, J); + DST(3, 3) = DST(2, 2) = DST(1, 1) = DST(0, 0) = AVG3(A, X, I); + DST(3, 2) = DST(2, 1) = DST(1, 0) = AVG3(B, A, X); + DST(3, 1) = DST(2, 0) = AVG3(C, B, A); + DST(3, 0) = AVG3(D, C, B); +} + +void vpx_d153_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride, + const uint8_t *above, const uint8_t *left) { + const int I = left[0]; + const int J = left[1]; + const int K = left[2]; + const int L = left[3]; + const int X = above[-1]; + const int A = above[0]; + const int B = above[1]; + const int C = above[2]; + + DST(0, 0) = DST(2, 1) = AVG2(I, X); + DST(0, 1) = DST(2, 2) = AVG2(J, I); + DST(0, 2) = DST(2, 3) = AVG2(K, J); + DST(0, 3) = AVG2(L, K); + + DST(3, 0) = AVG3(A, B, C); + DST(2, 0) = AVG3(X, A, B); + DST(1, 0) = DST(3, 1) = AVG3(I, X, A); + DST(1, 1) = DST(3, 2) = AVG3(J, I, X); + DST(1, 2) = DST(3, 3) = AVG3(K, J, I); + DST(1, 3) = AVG3(L, K, J); +} + +#if CONFIG_VP9_HIGHBITDEPTH +static INLINE void highbd_d207_predictor(uint16_t *dst, ptrdiff_t stride, + int bs, const uint16_t *above, + const uint16_t *left, int bd) { + int r, c; + (void) above; + (void) bd; + + // First column. + for (r = 0; r < bs - 1; ++r) { + dst[r * stride] = AVG2(left[r], left[r + 1]); + } + dst[(bs - 1) * stride] = left[bs - 1]; + dst++; + + // Second column. + for (r = 0; r < bs - 2; ++r) { + dst[r * stride] = AVG3(left[r], left[r + 1], left[r + 2]); + } + dst[(bs - 2) * stride] = AVG3(left[bs - 2], left[bs - 1], left[bs - 1]); + dst[(bs - 1) * stride] = left[bs - 1]; + dst++; + + // Rest of last row. + for (c = 0; c < bs - 2; ++c) + dst[(bs - 1) * stride + c] = left[bs - 1]; + + for (r = bs - 2; r >= 0; --r) { + for (c = 0; c < bs - 2; ++c) + dst[r * stride + c] = dst[(r + 1) * stride + c - 2]; + } +} + +#if CONFIG_MISC_FIXES +static INLINE void highbd_d207e_predictor(uint16_t *dst, ptrdiff_t stride, + int bs, const uint16_t *above, + const uint16_t *left, int bd) { + int r, c; + (void) above; + (void) bd; + + for (r = 0; r < bs; ++r) { + for (c = 0; c < bs; ++c) { + dst[c] = c & 1 ? AVG3(left[(c >> 1) + r], left[(c >> 1) + r + 1], + left[(c >> 1) + r + 2]) + : AVG2(left[(c >> 1) + r], left[(c >> 1) + r + 1]); + } + dst += stride; + } +} +#endif // CONFIG_MISC_FIXES + +static INLINE void highbd_d63_predictor(uint16_t *dst, ptrdiff_t stride, + int bs, const uint16_t *above, + const uint16_t *left, int bd) { + int r, c; + (void) left; + (void) bd; + for (r = 0; r < bs; ++r) { + for (c = 0; c < bs; ++c) { + dst[c] = r & 1 ? AVG3(above[(r >> 1) + c], above[(r >> 1) + c + 1], + above[(r >> 1) + c + 2]) + : AVG2(above[(r >> 1) + c], above[(r >> 1) + c + 1]); + } + dst += stride; + } +} + +#define highbd_d63e_predictor highbd_d63_predictor + +static INLINE void highbd_d45_predictor(uint16_t *dst, ptrdiff_t stride, int bs, + const uint16_t *above, + const uint16_t *left, int bd) { + int r, c; + (void) left; + (void) bd; + for (r = 0; r < bs; ++r) { + for (c = 0; c < bs; ++c) { + dst[c] = r + c + 2 < bs * 2 ? AVG3(above[r + c], above[r + c + 1], + above[r + c + 2]) + : above[bs * 2 - 1]; + } + dst += stride; + } +} + +#if CONFIG_MISC_FIXES +static INLINE void highbd_d45e_predictor(uint16_t *dst, ptrdiff_t stride, + int bs, const uint16_t *above, + const uint16_t *left, int bd) { + int r, c; + (void) left; + (void) bd; + for (r = 0; r < bs; ++r) { + for (c = 0; c < bs; ++c) { + dst[c] = AVG3(above[r + c], above[r + c + 1], + above[r + c + 1 + (r + c + 2 < bs * 2)]); + } + dst += stride; + } +} +#endif // CONFIG_MISC_FIXES + +static INLINE void highbd_d117_predictor(uint16_t *dst, ptrdiff_t stride, + int bs, const uint16_t *above, + const uint16_t *left, int bd) { + int r, c; + (void) bd; + + // first row + for (c = 0; c < bs; c++) + dst[c] = AVG2(above[c - 1], above[c]); + dst += stride; + + // second row + dst[0] = AVG3(left[0], above[-1], above[0]); + for (c = 1; c < bs; c++) + dst[c] = AVG3(above[c - 2], above[c - 1], above[c]); + dst += stride; + + // the rest of first col + dst[0] = AVG3(above[-1], left[0], left[1]); + for (r = 3; r < bs; ++r) + dst[(r - 2) * stride] = AVG3(left[r - 3], left[r - 2], left[r - 1]); + + // the rest of the block + for (r = 2; r < bs; ++r) { + for (c = 1; c < bs; c++) + dst[c] = dst[-2 * stride + c - 1]; + dst += stride; + } +} + +static INLINE void highbd_d135_predictor(uint16_t *dst, ptrdiff_t stride, + int bs, const uint16_t *above, + const uint16_t *left, int bd) { + int r, c; + (void) bd; + dst[0] = AVG3(left[0], above[-1], above[0]); + for (c = 1; c < bs; c++) + dst[c] = AVG3(above[c - 2], above[c - 1], above[c]); + + dst[stride] = AVG3(above[-1], left[0], left[1]); + for (r = 2; r < bs; ++r) + dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]); + + dst += stride; + for (r = 1; r < bs; ++r) { + for (c = 1; c < bs; c++) + dst[c] = dst[-stride + c - 1]; + dst += stride; + } +} + +static INLINE void highbd_d153_predictor(uint16_t *dst, ptrdiff_t stride, + int bs, const uint16_t *above, + const uint16_t *left, int bd) { + int r, c; + (void) bd; + dst[0] = AVG2(above[-1], left[0]); + for (r = 1; r < bs; r++) + dst[r * stride] = AVG2(left[r - 1], left[r]); + dst++; + + dst[0] = AVG3(left[0], above[-1], above[0]); + dst[stride] = AVG3(above[-1], left[0], left[1]); + for (r = 2; r < bs; r++) + dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]); + dst++; + + for (c = 0; c < bs - 2; c++) + dst[c] = AVG3(above[c - 1], above[c], above[c + 1]); + dst += stride; + + for (r = 1; r < bs; ++r) { + for (c = 0; c < bs - 2; c++) + dst[c] = dst[-stride + c - 2]; + dst += stride; + } +} + +static INLINE void highbd_v_predictor(uint16_t *dst, ptrdiff_t stride, + int bs, const uint16_t *above, + const uint16_t *left, int bd) { + int r; + (void) left; + (void) bd; + for (r = 0; r < bs; r++) { + memcpy(dst, above, bs * sizeof(uint16_t)); + dst += stride; + } +} + +static INLINE void highbd_h_predictor(uint16_t *dst, ptrdiff_t stride, + int bs, const uint16_t *above, + const uint16_t *left, int bd) { + int r; + (void) above; + (void) bd; + for (r = 0; r < bs; r++) { + vpx_memset16(dst, left[r], bs); + dst += stride; + } +} + +static INLINE void highbd_tm_predictor(uint16_t *dst, ptrdiff_t stride, + int bs, const uint16_t *above, + const uint16_t *left, int bd) { + int r, c; + int ytop_left = above[-1]; + (void) bd; + + for (r = 0; r < bs; r++) { + for (c = 0; c < bs; c++) + dst[c] = clip_pixel_highbd(left[r] + above[c] - ytop_left, bd); + dst += stride; + } +} + +static INLINE void highbd_dc_128_predictor(uint16_t *dst, ptrdiff_t stride, + int bs, const uint16_t *above, + const uint16_t *left, int bd) { + int r; + (void) above; + (void) left; + + for (r = 0; r < bs; r++) { + vpx_memset16(dst, 128 << (bd - 8), bs); + dst += stride; + } +} + +static INLINE void highbd_dc_left_predictor(uint16_t *dst, ptrdiff_t stride, + int bs, const uint16_t *above, + const uint16_t *left, int bd) { + int i, r, expected_dc, sum = 0; + (void) above; + (void) bd; + + for (i = 0; i < bs; i++) + sum += left[i]; + expected_dc = (sum + (bs >> 1)) / bs; + + for (r = 0; r < bs; r++) { + vpx_memset16(dst, expected_dc, bs); + dst += stride; + } +} + +static INLINE void highbd_dc_top_predictor(uint16_t *dst, ptrdiff_t stride, + int bs, const uint16_t *above, + const uint16_t *left, int bd) { + int i, r, expected_dc, sum = 0; + (void) left; + (void) bd; + + for (i = 0; i < bs; i++) + sum += above[i]; + expected_dc = (sum + (bs >> 1)) / bs; + + for (r = 0; r < bs; r++) { + vpx_memset16(dst, expected_dc, bs); + dst += stride; + } +} + +static INLINE void highbd_dc_predictor(uint16_t *dst, ptrdiff_t stride, + int bs, const uint16_t *above, + const uint16_t *left, int bd) { + int i, r, expected_dc, sum = 0; + const int count = 2 * bs; + (void) bd; + + for (i = 0; i < bs; i++) { + sum += above[i]; + sum += left[i]; + } + + expected_dc = (sum + (count >> 1)) / count; + + for (r = 0; r < bs; r++) { + vpx_memset16(dst, expected_dc, bs); + dst += stride; + } +} +#endif // CONFIG_VP9_HIGHBITDEPTH + +// This serves as a wrapper function, so that all the prediction functions +// can be unified and accessed as a pointer array. Note that the boundary +// above and left are not necessarily used all the time. +#define intra_pred_sized(type, size) \ + void vpx_##type##_predictor_##size##x##size##_c(uint8_t *dst, \ + ptrdiff_t stride, \ + const uint8_t *above, \ + const uint8_t *left) { \ + type##_predictor(dst, stride, size, above, left); \ + } + +#if CONFIG_VP9_HIGHBITDEPTH +#define intra_pred_highbd_sized(type, size) \ + void vpx_highbd_##type##_predictor_##size##x##size##_c( \ + uint16_t *dst, ptrdiff_t stride, const uint16_t *above, \ + const uint16_t *left, int bd) { \ + highbd_##type##_predictor(dst, stride, size, above, left, bd); \ + } + +#define intra_pred_allsizes(type) \ + intra_pred_sized(type, 4) \ + intra_pred_sized(type, 8) \ + intra_pred_sized(type, 16) \ + intra_pred_sized(type, 32) \ + intra_pred_highbd_sized(type, 4) \ + intra_pred_highbd_sized(type, 8) \ + intra_pred_highbd_sized(type, 16) \ + intra_pred_highbd_sized(type, 32) + +#define intra_pred_no_4x4(type) \ + intra_pred_sized(type, 8) \ + intra_pred_sized(type, 16) \ + intra_pred_sized(type, 32) \ + intra_pred_highbd_sized(type, 4) \ + intra_pred_highbd_sized(type, 8) \ + intra_pred_highbd_sized(type, 16) \ + intra_pred_highbd_sized(type, 32) + +#else +#define intra_pred_allsizes(type) \ + intra_pred_sized(type, 4) \ + intra_pred_sized(type, 8) \ + intra_pred_sized(type, 16) \ + intra_pred_sized(type, 32) + +#define intra_pred_no_4x4(type) \ + intra_pred_sized(type, 8) \ + intra_pred_sized(type, 16) \ + intra_pred_sized(type, 32) +#endif // CONFIG_VP9_HIGHBITDEPTH + +intra_pred_no_4x4(d207) +intra_pred_no_4x4(d63) +intra_pred_no_4x4(d45) +#if CONFIG_MISC_FIXES +intra_pred_allsizes(d207e) +intra_pred_allsizes(d63e) +intra_pred_no_4x4(d45e) +#endif +intra_pred_no_4x4(d117) +intra_pred_no_4x4(d135) +intra_pred_no_4x4(d153) +intra_pred_allsizes(v) +intra_pred_allsizes(h) +intra_pred_allsizes(tm) +intra_pred_allsizes(dc_128) +intra_pred_allsizes(dc_left) +intra_pred_allsizes(dc_top) +intra_pred_allsizes(dc) +#undef intra_pred_allsizes diff --git a/thirdparty/libvpx/vpx_dsp/inv_txfm.c b/thirdparty/libvpx/vpx_dsp/inv_txfm.c new file mode 100644 index 0000000000..e18d31d7aa --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/inv_txfm.c @@ -0,0 +1,2518 @@ +/* + * Copyright (c) 2015 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <math.h> +#include <string.h> + +#include "./vpx_dsp_rtcd.h" +#include "vpx_dsp/inv_txfm.h" + +void vpx_iwht4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int stride) { +/* 4-point reversible, orthonormal inverse Walsh-Hadamard in 3.5 adds, + 0.5 shifts per pixel. */ + int i; + tran_low_t output[16]; + tran_high_t a1, b1, c1, d1, e1; + const tran_low_t *ip = input; + tran_low_t *op = output; + + for (i = 0; i < 4; i++) { + a1 = ip[0] >> UNIT_QUANT_SHIFT; + c1 = ip[1] >> UNIT_QUANT_SHIFT; + d1 = ip[2] >> UNIT_QUANT_SHIFT; + b1 = ip[3] >> UNIT_QUANT_SHIFT; + a1 += c1; + d1 -= b1; + e1 = (a1 - d1) >> 1; + b1 = e1 - b1; + c1 = e1 - c1; + a1 -= b1; + d1 += c1; + op[0] = WRAPLOW(a1); + op[1] = WRAPLOW(b1); + op[2] = WRAPLOW(c1); + op[3] = WRAPLOW(d1); + ip += 4; + op += 4; + } + + ip = output; + for (i = 0; i < 4; i++) { + a1 = ip[4 * 0]; + c1 = ip[4 * 1]; + d1 = ip[4 * 2]; + b1 = ip[4 * 3]; + a1 += c1; + d1 -= b1; + e1 = (a1 - d1) >> 1; + b1 = e1 - b1; + c1 = e1 - c1; + a1 -= b1; + d1 += c1; + dest[stride * 0] = clip_pixel_add(dest[stride * 0], WRAPLOW(a1)); + dest[stride * 1] = clip_pixel_add(dest[stride * 1], WRAPLOW(b1)); + dest[stride * 2] = clip_pixel_add(dest[stride * 2], WRAPLOW(c1)); + dest[stride * 3] = clip_pixel_add(dest[stride * 3], WRAPLOW(d1)); + + ip++; + dest++; + } +} + +void vpx_iwht4x4_1_add_c(const tran_low_t *in, uint8_t *dest, int dest_stride) { + int i; + tran_high_t a1, e1; + tran_low_t tmp[4]; + const tran_low_t *ip = in; + tran_low_t *op = tmp; + + a1 = ip[0] >> UNIT_QUANT_SHIFT; + e1 = a1 >> 1; + a1 -= e1; + op[0] = WRAPLOW(a1); + op[1] = op[2] = op[3] = WRAPLOW(e1); + + ip = tmp; + for (i = 0; i < 4; i++) { + e1 = ip[0] >> 1; + a1 = ip[0] - e1; + dest[dest_stride * 0] = clip_pixel_add(dest[dest_stride * 0], a1); + dest[dest_stride * 1] = clip_pixel_add(dest[dest_stride * 1], e1); + dest[dest_stride * 2] = clip_pixel_add(dest[dest_stride * 2], e1); + dest[dest_stride * 3] = clip_pixel_add(dest[dest_stride * 3], e1); + ip++; + dest++; + } +} + +void idct4_c(const tran_low_t *input, tran_low_t *output) { + tran_low_t step[4]; + tran_high_t temp1, temp2; + // stage 1 + temp1 = (input[0] + input[2]) * cospi_16_64; + temp2 = (input[0] - input[2]) * cospi_16_64; + step[0] = WRAPLOW(dct_const_round_shift(temp1)); + step[1] = WRAPLOW(dct_const_round_shift(temp2)); + temp1 = input[1] * cospi_24_64 - input[3] * cospi_8_64; + temp2 = input[1] * cospi_8_64 + input[3] * cospi_24_64; + step[2] = WRAPLOW(dct_const_round_shift(temp1)); + step[3] = WRAPLOW(dct_const_round_shift(temp2)); + + // stage 2 + output[0] = WRAPLOW(step[0] + step[3]); + output[1] = WRAPLOW(step[1] + step[2]); + output[2] = WRAPLOW(step[1] - step[2]); + output[3] = WRAPLOW(step[0] - step[3]); +} + +void vpx_idct4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int stride) { + tran_low_t out[4 * 4]; + tran_low_t *outptr = out; + int i, j; + tran_low_t temp_in[4], temp_out[4]; + + // Rows + for (i = 0; i < 4; ++i) { + idct4_c(input, outptr); + input += 4; + outptr += 4; + } + + // Columns + for (i = 0; i < 4; ++i) { + for (j = 0; j < 4; ++j) + temp_in[j] = out[j * 4 + i]; + idct4_c(temp_in, temp_out); + for (j = 0; j < 4; ++j) { + dest[j * stride + i] = clip_pixel_add(dest[j * stride + i], + ROUND_POWER_OF_TWO(temp_out[j], 4)); + } + } +} + +void vpx_idct4x4_1_add_c(const tran_low_t *input, uint8_t *dest, + int dest_stride) { + int i; + tran_high_t a1; + tran_low_t out = WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64)); + out = WRAPLOW(dct_const_round_shift(out * cospi_16_64)); + a1 = ROUND_POWER_OF_TWO(out, 4); + + for (i = 0; i < 4; i++) { + dest[0] = clip_pixel_add(dest[0], a1); + dest[1] = clip_pixel_add(dest[1], a1); + dest[2] = clip_pixel_add(dest[2], a1); + dest[3] = clip_pixel_add(dest[3], a1); + dest += dest_stride; + } +} + +void idct8_c(const tran_low_t *input, tran_low_t *output) { + tran_low_t step1[8], step2[8]; + tran_high_t temp1, temp2; + // stage 1 + step1[0] = input[0]; + step1[2] = input[4]; + step1[1] = input[2]; + step1[3] = input[6]; + temp1 = input[1] * cospi_28_64 - input[7] * cospi_4_64; + temp2 = input[1] * cospi_4_64 + input[7] * cospi_28_64; + step1[4] = WRAPLOW(dct_const_round_shift(temp1)); + step1[7] = WRAPLOW(dct_const_round_shift(temp2)); + temp1 = input[5] * cospi_12_64 - input[3] * cospi_20_64; + temp2 = input[5] * cospi_20_64 + input[3] * cospi_12_64; + step1[5] = WRAPLOW(dct_const_round_shift(temp1)); + step1[6] = WRAPLOW(dct_const_round_shift(temp2)); + + // stage 2 + temp1 = (step1[0] + step1[2]) * cospi_16_64; + temp2 = (step1[0] - step1[2]) * cospi_16_64; + step2[0] = WRAPLOW(dct_const_round_shift(temp1)); + step2[1] = WRAPLOW(dct_const_round_shift(temp2)); + temp1 = step1[1] * cospi_24_64 - step1[3] * cospi_8_64; + temp2 = step1[1] * cospi_8_64 + step1[3] * cospi_24_64; + step2[2] = WRAPLOW(dct_const_round_shift(temp1)); + step2[3] = WRAPLOW(dct_const_round_shift(temp2)); + step2[4] = WRAPLOW(step1[4] + step1[5]); + step2[5] = WRAPLOW(step1[4] - step1[5]); + step2[6] = WRAPLOW(-step1[6] + step1[7]); + step2[7] = WRAPLOW(step1[6] + step1[7]); + + // stage 3 + step1[0] = WRAPLOW(step2[0] + step2[3]); + step1[1] = WRAPLOW(step2[1] + step2[2]); + step1[2] = WRAPLOW(step2[1] - step2[2]); + step1[3] = WRAPLOW(step2[0] - step2[3]); + step1[4] = step2[4]; + temp1 = (step2[6] - step2[5]) * cospi_16_64; + temp2 = (step2[5] + step2[6]) * cospi_16_64; + step1[5] = WRAPLOW(dct_const_round_shift(temp1)); + step1[6] = WRAPLOW(dct_const_round_shift(temp2)); + step1[7] = step2[7]; + + // stage 4 + output[0] = WRAPLOW(step1[0] + step1[7]); + output[1] = WRAPLOW(step1[1] + step1[6]); + output[2] = WRAPLOW(step1[2] + step1[5]); + output[3] = WRAPLOW(step1[3] + step1[4]); + output[4] = WRAPLOW(step1[3] - step1[4]); + output[5] = WRAPLOW(step1[2] - step1[5]); + output[6] = WRAPLOW(step1[1] - step1[6]); + output[7] = WRAPLOW(step1[0] - step1[7]); +} + +void vpx_idct8x8_64_add_c(const tran_low_t *input, uint8_t *dest, int stride) { + tran_low_t out[8 * 8]; + tran_low_t *outptr = out; + int i, j; + tran_low_t temp_in[8], temp_out[8]; + + // First transform rows + for (i = 0; i < 8; ++i) { + idct8_c(input, outptr); + input += 8; + outptr += 8; + } + + // Then transform columns + for (i = 0; i < 8; ++i) { + for (j = 0; j < 8; ++j) + temp_in[j] = out[j * 8 + i]; + idct8_c(temp_in, temp_out); + for (j = 0; j < 8; ++j) { + dest[j * stride + i] = clip_pixel_add(dest[j * stride + i], + ROUND_POWER_OF_TWO(temp_out[j], 5)); + } + } +} + +void vpx_idct8x8_1_add_c(const tran_low_t *input, uint8_t *dest, int stride) { + int i, j; + tran_high_t a1; + tran_low_t out = WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64)); + out = WRAPLOW(dct_const_round_shift(out * cospi_16_64)); + a1 = ROUND_POWER_OF_TWO(out, 5); + for (j = 0; j < 8; ++j) { + for (i = 0; i < 8; ++i) + dest[i] = clip_pixel_add(dest[i], a1); + dest += stride; + } +} + +void iadst4_c(const tran_low_t *input, tran_low_t *output) { + tran_high_t s0, s1, s2, s3, s4, s5, s6, s7; + + tran_low_t x0 = input[0]; + tran_low_t x1 = input[1]; + tran_low_t x2 = input[2]; + tran_low_t x3 = input[3]; + + if (!(x0 | x1 | x2 | x3)) { + output[0] = output[1] = output[2] = output[3] = 0; + return; + } + + s0 = sinpi_1_9 * x0; + s1 = sinpi_2_9 * x0; + s2 = sinpi_3_9 * x1; + s3 = sinpi_4_9 * x2; + s4 = sinpi_1_9 * x2; + s5 = sinpi_2_9 * x3; + s6 = sinpi_4_9 * x3; + s7 = WRAPLOW(x0 - x2 + x3); + + s0 = s0 + s3 + s5; + s1 = s1 - s4 - s6; + s3 = s2; + s2 = sinpi_3_9 * s7; + + // 1-D transform scaling factor is sqrt(2). + // The overall dynamic range is 14b (input) + 14b (multiplication scaling) + // + 1b (addition) = 29b. + // Hence the output bit depth is 15b. + output[0] = WRAPLOW(dct_const_round_shift(s0 + s3)); + output[1] = WRAPLOW(dct_const_round_shift(s1 + s3)); + output[2] = WRAPLOW(dct_const_round_shift(s2)); + output[3] = WRAPLOW(dct_const_round_shift(s0 + s1 - s3)); +} + +void iadst8_c(const tran_low_t *input, tran_low_t *output) { + int s0, s1, s2, s3, s4, s5, s6, s7; + + tran_high_t x0 = input[7]; + tran_high_t x1 = input[0]; + tran_high_t x2 = input[5]; + tran_high_t x3 = input[2]; + tran_high_t x4 = input[3]; + tran_high_t x5 = input[4]; + tran_high_t x6 = input[1]; + tran_high_t x7 = input[6]; + + if (!(x0 | x1 | x2 | x3 | x4 | x5 | x6 | x7)) { + output[0] = output[1] = output[2] = output[3] = output[4] + = output[5] = output[6] = output[7] = 0; + return; + } + + // stage 1 + s0 = (int)(cospi_2_64 * x0 + cospi_30_64 * x1); + s1 = (int)(cospi_30_64 * x0 - cospi_2_64 * x1); + s2 = (int)(cospi_10_64 * x2 + cospi_22_64 * x3); + s3 = (int)(cospi_22_64 * x2 - cospi_10_64 * x3); + s4 = (int)(cospi_18_64 * x4 + cospi_14_64 * x5); + s5 = (int)(cospi_14_64 * x4 - cospi_18_64 * x5); + s6 = (int)(cospi_26_64 * x6 + cospi_6_64 * x7); + s7 = (int)(cospi_6_64 * x6 - cospi_26_64 * x7); + + x0 = WRAPLOW(dct_const_round_shift(s0 + s4)); + x1 = WRAPLOW(dct_const_round_shift(s1 + s5)); + x2 = WRAPLOW(dct_const_round_shift(s2 + s6)); + x3 = WRAPLOW(dct_const_round_shift(s3 + s7)); + x4 = WRAPLOW(dct_const_round_shift(s0 - s4)); + x5 = WRAPLOW(dct_const_round_shift(s1 - s5)); + x6 = WRAPLOW(dct_const_round_shift(s2 - s6)); + x7 = WRAPLOW(dct_const_round_shift(s3 - s7)); + + // stage 2 + s0 = (int)x0; + s1 = (int)x1; + s2 = (int)x2; + s3 = (int)x3; + s4 = (int)(cospi_8_64 * x4 + cospi_24_64 * x5); + s5 = (int)(cospi_24_64 * x4 - cospi_8_64 * x5); + s6 = (int)(-cospi_24_64 * x6 + cospi_8_64 * x7); + s7 = (int)(cospi_8_64 * x6 + cospi_24_64 * x7); + + x0 = WRAPLOW(s0 + s2); + x1 = WRAPLOW(s1 + s3); + x2 = WRAPLOW(s0 - s2); + x3 = WRAPLOW(s1 - s3); + x4 = WRAPLOW(dct_const_round_shift(s4 + s6)); + x5 = WRAPLOW(dct_const_round_shift(s5 + s7)); + x6 = WRAPLOW(dct_const_round_shift(s4 - s6)); + x7 = WRAPLOW(dct_const_round_shift(s5 - s7)); + + // stage 3 + s2 = (int)(cospi_16_64 * (x2 + x3)); + s3 = (int)(cospi_16_64 * (x2 - x3)); + s6 = (int)(cospi_16_64 * (x6 + x7)); + s7 = (int)(cospi_16_64 * (x6 - x7)); + + x2 = WRAPLOW(dct_const_round_shift(s2)); + x3 = WRAPLOW(dct_const_round_shift(s3)); + x6 = WRAPLOW(dct_const_round_shift(s6)); + x7 = WRAPLOW(dct_const_round_shift(s7)); + + output[0] = WRAPLOW(x0); + output[1] = WRAPLOW(-x4); + output[2] = WRAPLOW(x6); + output[3] = WRAPLOW(-x2); + output[4] = WRAPLOW(x3); + output[5] = WRAPLOW(-x7); + output[6] = WRAPLOW(x5); + output[7] = WRAPLOW(-x1); +} + +void vpx_idct8x8_12_add_c(const tran_low_t *input, uint8_t *dest, int stride) { + tran_low_t out[8 * 8] = { 0 }; + tran_low_t *outptr = out; + int i, j; + tran_low_t temp_in[8], temp_out[8]; + + // First transform rows + // only first 4 row has non-zero coefs + for (i = 0; i < 4; ++i) { + idct8_c(input, outptr); + input += 8; + outptr += 8; + } + + // Then transform columns + for (i = 0; i < 8; ++i) { + for (j = 0; j < 8; ++j) + temp_in[j] = out[j * 8 + i]; + idct8_c(temp_in, temp_out); + for (j = 0; j < 8; ++j) { + dest[j * stride + i] = clip_pixel_add(dest[j * stride + i], + ROUND_POWER_OF_TWO(temp_out[j], 5)); + } + } +} + +void idct16_c(const tran_low_t *input, tran_low_t *output) { + tran_low_t step1[16], step2[16]; + tran_high_t temp1, temp2; + + // stage 1 + step1[0] = input[0/2]; + step1[1] = input[16/2]; + step1[2] = input[8/2]; + step1[3] = input[24/2]; + step1[4] = input[4/2]; + step1[5] = input[20/2]; + step1[6] = input[12/2]; + step1[7] = input[28/2]; + step1[8] = input[2/2]; + step1[9] = input[18/2]; + step1[10] = input[10/2]; + step1[11] = input[26/2]; + step1[12] = input[6/2]; + step1[13] = input[22/2]; + step1[14] = input[14/2]; + step1[15] = input[30/2]; + + // stage 2 + step2[0] = step1[0]; + step2[1] = step1[1]; + step2[2] = step1[2]; + step2[3] = step1[3]; + step2[4] = step1[4]; + step2[5] = step1[5]; + step2[6] = step1[6]; + step2[7] = step1[7]; + + temp1 = step1[8] * cospi_30_64 - step1[15] * cospi_2_64; + temp2 = step1[8] * cospi_2_64 + step1[15] * cospi_30_64; + step2[8] = WRAPLOW(dct_const_round_shift(temp1)); + step2[15] = WRAPLOW(dct_const_round_shift(temp2)); + + temp1 = step1[9] * cospi_14_64 - step1[14] * cospi_18_64; + temp2 = step1[9] * cospi_18_64 + step1[14] * cospi_14_64; + step2[9] = WRAPLOW(dct_const_round_shift(temp1)); + step2[14] = WRAPLOW(dct_const_round_shift(temp2)); + + temp1 = step1[10] * cospi_22_64 - step1[13] * cospi_10_64; + temp2 = step1[10] * cospi_10_64 + step1[13] * cospi_22_64; + step2[10] = WRAPLOW(dct_const_round_shift(temp1)); + step2[13] = WRAPLOW(dct_const_round_shift(temp2)); + + temp1 = step1[11] * cospi_6_64 - step1[12] * cospi_26_64; + temp2 = step1[11] * cospi_26_64 + step1[12] * cospi_6_64; + step2[11] = WRAPLOW(dct_const_round_shift(temp1)); + step2[12] = WRAPLOW(dct_const_round_shift(temp2)); + + // stage 3 + step1[0] = step2[0]; + step1[1] = step2[1]; + step1[2] = step2[2]; + step1[3] = step2[3]; + + temp1 = step2[4] * cospi_28_64 - step2[7] * cospi_4_64; + temp2 = step2[4] * cospi_4_64 + step2[7] * cospi_28_64; + step1[4] = WRAPLOW(dct_const_round_shift(temp1)); + step1[7] = WRAPLOW(dct_const_round_shift(temp2)); + temp1 = step2[5] * cospi_12_64 - step2[6] * cospi_20_64; + temp2 = step2[5] * cospi_20_64 + step2[6] * cospi_12_64; + step1[5] = WRAPLOW(dct_const_round_shift(temp1)); + step1[6] = WRAPLOW(dct_const_round_shift(temp2)); + + step1[8] = WRAPLOW(step2[8] + step2[9]); + step1[9] = WRAPLOW(step2[8] - step2[9]); + step1[10] = WRAPLOW(-step2[10] + step2[11]); + step1[11] = WRAPLOW(step2[10] + step2[11]); + step1[12] = WRAPLOW(step2[12] + step2[13]); + step1[13] = WRAPLOW(step2[12] - step2[13]); + step1[14] = WRAPLOW(-step2[14] + step2[15]); + step1[15] = WRAPLOW(step2[14] + step2[15]); + + // stage 4 + temp1 = (step1[0] + step1[1]) * cospi_16_64; + temp2 = (step1[0] - step1[1]) * cospi_16_64; + step2[0] = WRAPLOW(dct_const_round_shift(temp1)); + step2[1] = WRAPLOW(dct_const_round_shift(temp2)); + temp1 = step1[2] * cospi_24_64 - step1[3] * cospi_8_64; + temp2 = step1[2] * cospi_8_64 + step1[3] * cospi_24_64; + step2[2] = WRAPLOW(dct_const_round_shift(temp1)); + step2[3] = WRAPLOW(dct_const_round_shift(temp2)); + step2[4] = WRAPLOW(step1[4] + step1[5]); + step2[5] = WRAPLOW(step1[4] - step1[5]); + step2[6] = WRAPLOW(-step1[6] + step1[7]); + step2[7] = WRAPLOW(step1[6] + step1[7]); + + step2[8] = step1[8]; + step2[15] = step1[15]; + temp1 = -step1[9] * cospi_8_64 + step1[14] * cospi_24_64; + temp2 = step1[9] * cospi_24_64 + step1[14] * cospi_8_64; + step2[9] = WRAPLOW(dct_const_round_shift(temp1)); + step2[14] = WRAPLOW(dct_const_round_shift(temp2)); + temp1 = -step1[10] * cospi_24_64 - step1[13] * cospi_8_64; + temp2 = -step1[10] * cospi_8_64 + step1[13] * cospi_24_64; + step2[10] = WRAPLOW(dct_const_round_shift(temp1)); + step2[13] = WRAPLOW(dct_const_round_shift(temp2)); + step2[11] = step1[11]; + step2[12] = step1[12]; + + // stage 5 + step1[0] = WRAPLOW(step2[0] + step2[3]); + step1[1] = WRAPLOW(step2[1] + step2[2]); + step1[2] = WRAPLOW(step2[1] - step2[2]); + step1[3] = WRAPLOW(step2[0] - step2[3]); + step1[4] = step2[4]; + temp1 = (step2[6] - step2[5]) * cospi_16_64; + temp2 = (step2[5] + step2[6]) * cospi_16_64; + step1[5] = WRAPLOW(dct_const_round_shift(temp1)); + step1[6] = WRAPLOW(dct_const_round_shift(temp2)); + step1[7] = step2[7]; + + step1[8] = WRAPLOW(step2[8] + step2[11]); + step1[9] = WRAPLOW(step2[9] + step2[10]); + step1[10] = WRAPLOW(step2[9] - step2[10]); + step1[11] = WRAPLOW(step2[8] - step2[11]); + step1[12] = WRAPLOW(-step2[12] + step2[15]); + step1[13] = WRAPLOW(-step2[13] + step2[14]); + step1[14] = WRAPLOW(step2[13] + step2[14]); + step1[15] = WRAPLOW(step2[12] + step2[15]); + + // stage 6 + step2[0] = WRAPLOW(step1[0] + step1[7]); + step2[1] = WRAPLOW(step1[1] + step1[6]); + step2[2] = WRAPLOW(step1[2] + step1[5]); + step2[3] = WRAPLOW(step1[3] + step1[4]); + step2[4] = WRAPLOW(step1[3] - step1[4]); + step2[5] = WRAPLOW(step1[2] - step1[5]); + step2[6] = WRAPLOW(step1[1] - step1[6]); + step2[7] = WRAPLOW(step1[0] - step1[7]); + step2[8] = step1[8]; + step2[9] = step1[9]; + temp1 = (-step1[10] + step1[13]) * cospi_16_64; + temp2 = (step1[10] + step1[13]) * cospi_16_64; + step2[10] = WRAPLOW(dct_const_round_shift(temp1)); + step2[13] = WRAPLOW(dct_const_round_shift(temp2)); + temp1 = (-step1[11] + step1[12]) * cospi_16_64; + temp2 = (step1[11] + step1[12]) * cospi_16_64; + step2[11] = WRAPLOW(dct_const_round_shift(temp1)); + step2[12] = WRAPLOW(dct_const_round_shift(temp2)); + step2[14] = step1[14]; + step2[15] = step1[15]; + + // stage 7 + output[0] = WRAPLOW(step2[0] + step2[15]); + output[1] = WRAPLOW(step2[1] + step2[14]); + output[2] = WRAPLOW(step2[2] + step2[13]); + output[3] = WRAPLOW(step2[3] + step2[12]); + output[4] = WRAPLOW(step2[4] + step2[11]); + output[5] = WRAPLOW(step2[5] + step2[10]); + output[6] = WRAPLOW(step2[6] + step2[9]); + output[7] = WRAPLOW(step2[7] + step2[8]); + output[8] = WRAPLOW(step2[7] - step2[8]); + output[9] = WRAPLOW(step2[6] - step2[9]); + output[10] = WRAPLOW(step2[5] - step2[10]); + output[11] = WRAPLOW(step2[4] - step2[11]); + output[12] = WRAPLOW(step2[3] - step2[12]); + output[13] = WRAPLOW(step2[2] - step2[13]); + output[14] = WRAPLOW(step2[1] - step2[14]); + output[15] = WRAPLOW(step2[0] - step2[15]); +} + +void vpx_idct16x16_256_add_c(const tran_low_t *input, uint8_t *dest, + int stride) { + tran_low_t out[16 * 16]; + tran_low_t *outptr = out; + int i, j; + tran_low_t temp_in[16], temp_out[16]; + + // First transform rows + for (i = 0; i < 16; ++i) { + idct16_c(input, outptr); + input += 16; + outptr += 16; + } + + // Then transform columns + for (i = 0; i < 16; ++i) { + for (j = 0; j < 16; ++j) + temp_in[j] = out[j * 16 + i]; + idct16_c(temp_in, temp_out); + for (j = 0; j < 16; ++j) { + dest[j * stride + i] = clip_pixel_add(dest[j * stride + i], + ROUND_POWER_OF_TWO(temp_out[j], 6)); + } + } +} + +void iadst16_c(const tran_low_t *input, tran_low_t *output) { + tran_high_t s0, s1, s2, s3, s4, s5, s6, s7, s8; + tran_high_t s9, s10, s11, s12, s13, s14, s15; + + tran_high_t x0 = input[15]; + tran_high_t x1 = input[0]; + tran_high_t x2 = input[13]; + tran_high_t x3 = input[2]; + tran_high_t x4 = input[11]; + tran_high_t x5 = input[4]; + tran_high_t x6 = input[9]; + tran_high_t x7 = input[6]; + tran_high_t x8 = input[7]; + tran_high_t x9 = input[8]; + tran_high_t x10 = input[5]; + tran_high_t x11 = input[10]; + tran_high_t x12 = input[3]; + tran_high_t x13 = input[12]; + tran_high_t x14 = input[1]; + tran_high_t x15 = input[14]; + + if (!(x0 | x1 | x2 | x3 | x4 | x5 | x6 | x7 | x8 + | x9 | x10 | x11 | x12 | x13 | x14 | x15)) { + output[0] = output[1] = output[2] = output[3] = output[4] + = output[5] = output[6] = output[7] = output[8] + = output[9] = output[10] = output[11] = output[12] + = output[13] = output[14] = output[15] = 0; + return; + } + + // stage 1 + s0 = x0 * cospi_1_64 + x1 * cospi_31_64; + s1 = x0 * cospi_31_64 - x1 * cospi_1_64; + s2 = x2 * cospi_5_64 + x3 * cospi_27_64; + s3 = x2 * cospi_27_64 - x3 * cospi_5_64; + s4 = x4 * cospi_9_64 + x5 * cospi_23_64; + s5 = x4 * cospi_23_64 - x5 * cospi_9_64; + s6 = x6 * cospi_13_64 + x7 * cospi_19_64; + s7 = x6 * cospi_19_64 - x7 * cospi_13_64; + s8 = x8 * cospi_17_64 + x9 * cospi_15_64; + s9 = x8 * cospi_15_64 - x9 * cospi_17_64; + s10 = x10 * cospi_21_64 + x11 * cospi_11_64; + s11 = x10 * cospi_11_64 - x11 * cospi_21_64; + s12 = x12 * cospi_25_64 + x13 * cospi_7_64; + s13 = x12 * cospi_7_64 - x13 * cospi_25_64; + s14 = x14 * cospi_29_64 + x15 * cospi_3_64; + s15 = x14 * cospi_3_64 - x15 * cospi_29_64; + + x0 = WRAPLOW(dct_const_round_shift(s0 + s8)); + x1 = WRAPLOW(dct_const_round_shift(s1 + s9)); + x2 = WRAPLOW(dct_const_round_shift(s2 + s10)); + x3 = WRAPLOW(dct_const_round_shift(s3 + s11)); + x4 = WRAPLOW(dct_const_round_shift(s4 + s12)); + x5 = WRAPLOW(dct_const_round_shift(s5 + s13)); + x6 = WRAPLOW(dct_const_round_shift(s6 + s14)); + x7 = WRAPLOW(dct_const_round_shift(s7 + s15)); + x8 = WRAPLOW(dct_const_round_shift(s0 - s8)); + x9 = WRAPLOW(dct_const_round_shift(s1 - s9)); + x10 = WRAPLOW(dct_const_round_shift(s2 - s10)); + x11 = WRAPLOW(dct_const_round_shift(s3 - s11)); + x12 = WRAPLOW(dct_const_round_shift(s4 - s12)); + x13 = WRAPLOW(dct_const_round_shift(s5 - s13)); + x14 = WRAPLOW(dct_const_round_shift(s6 - s14)); + x15 = WRAPLOW(dct_const_round_shift(s7 - s15)); + + // stage 2 + s0 = x0; + s1 = x1; + s2 = x2; + s3 = x3; + s4 = x4; + s5 = x5; + s6 = x6; + s7 = x7; + s8 = x8 * cospi_4_64 + x9 * cospi_28_64; + s9 = x8 * cospi_28_64 - x9 * cospi_4_64; + s10 = x10 * cospi_20_64 + x11 * cospi_12_64; + s11 = x10 * cospi_12_64 - x11 * cospi_20_64; + s12 = - x12 * cospi_28_64 + x13 * cospi_4_64; + s13 = x12 * cospi_4_64 + x13 * cospi_28_64; + s14 = - x14 * cospi_12_64 + x15 * cospi_20_64; + s15 = x14 * cospi_20_64 + x15 * cospi_12_64; + + x0 = WRAPLOW(s0 + s4); + x1 = WRAPLOW(s1 + s5); + x2 = WRAPLOW(s2 + s6); + x3 = WRAPLOW(s3 + s7); + x4 = WRAPLOW(s0 - s4); + x5 = WRAPLOW(s1 - s5); + x6 = WRAPLOW(s2 - s6); + x7 = WRAPLOW(s3 - s7); + x8 = WRAPLOW(dct_const_round_shift(s8 + s12)); + x9 = WRAPLOW(dct_const_round_shift(s9 + s13)); + x10 = WRAPLOW(dct_const_round_shift(s10 + s14)); + x11 = WRAPLOW(dct_const_round_shift(s11 + s15)); + x12 = WRAPLOW(dct_const_round_shift(s8 - s12)); + x13 = WRAPLOW(dct_const_round_shift(s9 - s13)); + x14 = WRAPLOW(dct_const_round_shift(s10 - s14)); + x15 = WRAPLOW(dct_const_round_shift(s11 - s15)); + + // stage 3 + s0 = x0; + s1 = x1; + s2 = x2; + s3 = x3; + s4 = x4 * cospi_8_64 + x5 * cospi_24_64; + s5 = x4 * cospi_24_64 - x5 * cospi_8_64; + s6 = - x6 * cospi_24_64 + x7 * cospi_8_64; + s7 = x6 * cospi_8_64 + x7 * cospi_24_64; + s8 = x8; + s9 = x9; + s10 = x10; + s11 = x11; + s12 = x12 * cospi_8_64 + x13 * cospi_24_64; + s13 = x12 * cospi_24_64 - x13 * cospi_8_64; + s14 = - x14 * cospi_24_64 + x15 * cospi_8_64; + s15 = x14 * cospi_8_64 + x15 * cospi_24_64; + + x0 = WRAPLOW(s0 + s2); + x1 = WRAPLOW(s1 + s3); + x2 = WRAPLOW(s0 - s2); + x3 = WRAPLOW(s1 - s3); + x4 = WRAPLOW(dct_const_round_shift(s4 + s6)); + x5 = WRAPLOW(dct_const_round_shift(s5 + s7)); + x6 = WRAPLOW(dct_const_round_shift(s4 - s6)); + x7 = WRAPLOW(dct_const_round_shift(s5 - s7)); + x8 = WRAPLOW(s8 + s10); + x9 = WRAPLOW(s9 + s11); + x10 = WRAPLOW(s8 - s10); + x11 = WRAPLOW(s9 - s11); + x12 = WRAPLOW(dct_const_round_shift(s12 + s14)); + x13 = WRAPLOW(dct_const_round_shift(s13 + s15)); + x14 = WRAPLOW(dct_const_round_shift(s12 - s14)); + x15 = WRAPLOW(dct_const_round_shift(s13 - s15)); + + // stage 4 + s2 = (- cospi_16_64) * (x2 + x3); + s3 = cospi_16_64 * (x2 - x3); + s6 = cospi_16_64 * (x6 + x7); + s7 = cospi_16_64 * (- x6 + x7); + s10 = cospi_16_64 * (x10 + x11); + s11 = cospi_16_64 * (- x10 + x11); + s14 = (- cospi_16_64) * (x14 + x15); + s15 = cospi_16_64 * (x14 - x15); + + x2 = WRAPLOW(dct_const_round_shift(s2)); + x3 = WRAPLOW(dct_const_round_shift(s3)); + x6 = WRAPLOW(dct_const_round_shift(s6)); + x7 = WRAPLOW(dct_const_round_shift(s7)); + x10 = WRAPLOW(dct_const_round_shift(s10)); + x11 = WRAPLOW(dct_const_round_shift(s11)); + x14 = WRAPLOW(dct_const_round_shift(s14)); + x15 = WRAPLOW(dct_const_round_shift(s15)); + + output[0] = WRAPLOW(x0); + output[1] = WRAPLOW(-x8); + output[2] = WRAPLOW(x12); + output[3] = WRAPLOW(-x4); + output[4] = WRAPLOW(x6); + output[5] = WRAPLOW(x14); + output[6] = WRAPLOW(x10); + output[7] = WRAPLOW(x2); + output[8] = WRAPLOW(x3); + output[9] = WRAPLOW(x11); + output[10] = WRAPLOW(x15); + output[11] = WRAPLOW(x7); + output[12] = WRAPLOW(x5); + output[13] = WRAPLOW(-x13); + output[14] = WRAPLOW(x9); + output[15] = WRAPLOW(-x1); +} + +void vpx_idct16x16_10_add_c(const tran_low_t *input, uint8_t *dest, + int stride) { + tran_low_t out[16 * 16] = { 0 }; + tran_low_t *outptr = out; + int i, j; + tran_low_t temp_in[16], temp_out[16]; + + // First transform rows. Since all non-zero dct coefficients are in + // upper-left 4x4 area, we only need to calculate first 4 rows here. + for (i = 0; i < 4; ++i) { + idct16_c(input, outptr); + input += 16; + outptr += 16; + } + + // Then transform columns + for (i = 0; i < 16; ++i) { + for (j = 0; j < 16; ++j) + temp_in[j] = out[j*16 + i]; + idct16_c(temp_in, temp_out); + for (j = 0; j < 16; ++j) { + dest[j * stride + i] = clip_pixel_add(dest[j * stride + i], + ROUND_POWER_OF_TWO(temp_out[j], 6)); + } + } +} + +void vpx_idct16x16_1_add_c(const tran_low_t *input, uint8_t *dest, int stride) { + int i, j; + tran_high_t a1; + tran_low_t out = WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64)); + out = WRAPLOW(dct_const_round_shift(out * cospi_16_64)); + a1 = ROUND_POWER_OF_TWO(out, 6); + for (j = 0; j < 16; ++j) { + for (i = 0; i < 16; ++i) + dest[i] = clip_pixel_add(dest[i], a1); + dest += stride; + } +} + +void idct32_c(const tran_low_t *input, tran_low_t *output) { + tran_low_t step1[32], step2[32]; + tran_high_t temp1, temp2; + + // stage 1 + step1[0] = input[0]; + step1[1] = input[16]; + step1[2] = input[8]; + step1[3] = input[24]; + step1[4] = input[4]; + step1[5] = input[20]; + step1[6] = input[12]; + step1[7] = input[28]; + step1[8] = input[2]; + step1[9] = input[18]; + step1[10] = input[10]; + step1[11] = input[26]; + step1[12] = input[6]; + step1[13] = input[22]; + step1[14] = input[14]; + step1[15] = input[30]; + + temp1 = input[1] * cospi_31_64 - input[31] * cospi_1_64; + temp2 = input[1] * cospi_1_64 + input[31] * cospi_31_64; + step1[16] = WRAPLOW(dct_const_round_shift(temp1)); + step1[31] = WRAPLOW(dct_const_round_shift(temp2)); + + temp1 = input[17] * cospi_15_64 - input[15] * cospi_17_64; + temp2 = input[17] * cospi_17_64 + input[15] * cospi_15_64; + step1[17] = WRAPLOW(dct_const_round_shift(temp1)); + step1[30] = WRAPLOW(dct_const_round_shift(temp2)); + + temp1 = input[9] * cospi_23_64 - input[23] * cospi_9_64; + temp2 = input[9] * cospi_9_64 + input[23] * cospi_23_64; + step1[18] = WRAPLOW(dct_const_round_shift(temp1)); + step1[29] = WRAPLOW(dct_const_round_shift(temp2)); + + temp1 = input[25] * cospi_7_64 - input[7] * cospi_25_64; + temp2 = input[25] * cospi_25_64 + input[7] * cospi_7_64; + step1[19] = WRAPLOW(dct_const_round_shift(temp1)); + step1[28] = WRAPLOW(dct_const_round_shift(temp2)); + + temp1 = input[5] * cospi_27_64 - input[27] * cospi_5_64; + temp2 = input[5] * cospi_5_64 + input[27] * cospi_27_64; + step1[20] = WRAPLOW(dct_const_round_shift(temp1)); + step1[27] = WRAPLOW(dct_const_round_shift(temp2)); + + temp1 = input[21] * cospi_11_64 - input[11] * cospi_21_64; + temp2 = input[21] * cospi_21_64 + input[11] * cospi_11_64; + step1[21] = WRAPLOW(dct_const_round_shift(temp1)); + step1[26] = WRAPLOW(dct_const_round_shift(temp2)); + + temp1 = input[13] * cospi_19_64 - input[19] * cospi_13_64; + temp2 = input[13] * cospi_13_64 + input[19] * cospi_19_64; + step1[22] = WRAPLOW(dct_const_round_shift(temp1)); + step1[25] = WRAPLOW(dct_const_round_shift(temp2)); + + temp1 = input[29] * cospi_3_64 - input[3] * cospi_29_64; + temp2 = input[29] * cospi_29_64 + input[3] * cospi_3_64; + step1[23] = WRAPLOW(dct_const_round_shift(temp1)); + step1[24] = WRAPLOW(dct_const_round_shift(temp2)); + + // stage 2 + step2[0] = step1[0]; + step2[1] = step1[1]; + step2[2] = step1[2]; + step2[3] = step1[3]; + step2[4] = step1[4]; + step2[5] = step1[5]; + step2[6] = step1[6]; + step2[7] = step1[7]; + + temp1 = step1[8] * cospi_30_64 - step1[15] * cospi_2_64; + temp2 = step1[8] * cospi_2_64 + step1[15] * cospi_30_64; + step2[8] = WRAPLOW(dct_const_round_shift(temp1)); + step2[15] = WRAPLOW(dct_const_round_shift(temp2)); + + temp1 = step1[9] * cospi_14_64 - step1[14] * cospi_18_64; + temp2 = step1[9] * cospi_18_64 + step1[14] * cospi_14_64; + step2[9] = WRAPLOW(dct_const_round_shift(temp1)); + step2[14] = WRAPLOW(dct_const_round_shift(temp2)); + + temp1 = step1[10] * cospi_22_64 - step1[13] * cospi_10_64; + temp2 = step1[10] * cospi_10_64 + step1[13] * cospi_22_64; + step2[10] = WRAPLOW(dct_const_round_shift(temp1)); + step2[13] = WRAPLOW(dct_const_round_shift(temp2)); + + temp1 = step1[11] * cospi_6_64 - step1[12] * cospi_26_64; + temp2 = step1[11] * cospi_26_64 + step1[12] * cospi_6_64; + step2[11] = WRAPLOW(dct_const_round_shift(temp1)); + step2[12] = WRAPLOW(dct_const_round_shift(temp2)); + + step2[16] = WRAPLOW(step1[16] + step1[17]); + step2[17] = WRAPLOW(step1[16] - step1[17]); + step2[18] = WRAPLOW(-step1[18] + step1[19]); + step2[19] = WRAPLOW(step1[18] + step1[19]); + step2[20] = WRAPLOW(step1[20] + step1[21]); + step2[21] = WRAPLOW(step1[20] - step1[21]); + step2[22] = WRAPLOW(-step1[22] + step1[23]); + step2[23] = WRAPLOW(step1[22] + step1[23]); + step2[24] = WRAPLOW(step1[24] + step1[25]); + step2[25] = WRAPLOW(step1[24] - step1[25]); + step2[26] = WRAPLOW(-step1[26] + step1[27]); + step2[27] = WRAPLOW(step1[26] + step1[27]); + step2[28] = WRAPLOW(step1[28] + step1[29]); + step2[29] = WRAPLOW(step1[28] - step1[29]); + step2[30] = WRAPLOW(-step1[30] + step1[31]); + step2[31] = WRAPLOW(step1[30] + step1[31]); + + // stage 3 + step1[0] = step2[0]; + step1[1] = step2[1]; + step1[2] = step2[2]; + step1[3] = step2[3]; + + temp1 = step2[4] * cospi_28_64 - step2[7] * cospi_4_64; + temp2 = step2[4] * cospi_4_64 + step2[7] * cospi_28_64; + step1[4] = WRAPLOW(dct_const_round_shift(temp1)); + step1[7] = WRAPLOW(dct_const_round_shift(temp2)); + temp1 = step2[5] * cospi_12_64 - step2[6] * cospi_20_64; + temp2 = step2[5] * cospi_20_64 + step2[6] * cospi_12_64; + step1[5] = WRAPLOW(dct_const_round_shift(temp1)); + step1[6] = WRAPLOW(dct_const_round_shift(temp2)); + + step1[8] = WRAPLOW(step2[8] + step2[9]); + step1[9] = WRAPLOW(step2[8] - step2[9]); + step1[10] = WRAPLOW(-step2[10] + step2[11]); + step1[11] = WRAPLOW(step2[10] + step2[11]); + step1[12] = WRAPLOW(step2[12] + step2[13]); + step1[13] = WRAPLOW(step2[12] - step2[13]); + step1[14] = WRAPLOW(-step2[14] + step2[15]); + step1[15] = WRAPLOW(step2[14] + step2[15]); + + step1[16] = step2[16]; + step1[31] = step2[31]; + temp1 = -step2[17] * cospi_4_64 + step2[30] * cospi_28_64; + temp2 = step2[17] * cospi_28_64 + step2[30] * cospi_4_64; + step1[17] = WRAPLOW(dct_const_round_shift(temp1)); + step1[30] = WRAPLOW(dct_const_round_shift(temp2)); + temp1 = -step2[18] * cospi_28_64 - step2[29] * cospi_4_64; + temp2 = -step2[18] * cospi_4_64 + step2[29] * cospi_28_64; + step1[18] = WRAPLOW(dct_const_round_shift(temp1)); + step1[29] = WRAPLOW(dct_const_round_shift(temp2)); + step1[19] = step2[19]; + step1[20] = step2[20]; + temp1 = -step2[21] * cospi_20_64 + step2[26] * cospi_12_64; + temp2 = step2[21] * cospi_12_64 + step2[26] * cospi_20_64; + step1[21] = WRAPLOW(dct_const_round_shift(temp1)); + step1[26] = WRAPLOW(dct_const_round_shift(temp2)); + temp1 = -step2[22] * cospi_12_64 - step2[25] * cospi_20_64; + temp2 = -step2[22] * cospi_20_64 + step2[25] * cospi_12_64; + step1[22] = WRAPLOW(dct_const_round_shift(temp1)); + step1[25] = WRAPLOW(dct_const_round_shift(temp2)); + step1[23] = step2[23]; + step1[24] = step2[24]; + step1[27] = step2[27]; + step1[28] = step2[28]; + + // stage 4 + temp1 = (step1[0] + step1[1]) * cospi_16_64; + temp2 = (step1[0] - step1[1]) * cospi_16_64; + step2[0] = WRAPLOW(dct_const_round_shift(temp1)); + step2[1] = WRAPLOW(dct_const_round_shift(temp2)); + temp1 = step1[2] * cospi_24_64 - step1[3] * cospi_8_64; + temp2 = step1[2] * cospi_8_64 + step1[3] * cospi_24_64; + step2[2] = WRAPLOW(dct_const_round_shift(temp1)); + step2[3] = WRAPLOW(dct_const_round_shift(temp2)); + step2[4] = WRAPLOW(step1[4] + step1[5]); + step2[5] = WRAPLOW(step1[4] - step1[5]); + step2[6] = WRAPLOW(-step1[6] + step1[7]); + step2[7] = WRAPLOW(step1[6] + step1[7]); + + step2[8] = step1[8]; + step2[15] = step1[15]; + temp1 = -step1[9] * cospi_8_64 + step1[14] * cospi_24_64; + temp2 = step1[9] * cospi_24_64 + step1[14] * cospi_8_64; + step2[9] = WRAPLOW(dct_const_round_shift(temp1)); + step2[14] = WRAPLOW(dct_const_round_shift(temp2)); + temp1 = -step1[10] * cospi_24_64 - step1[13] * cospi_8_64; + temp2 = -step1[10] * cospi_8_64 + step1[13] * cospi_24_64; + step2[10] = WRAPLOW(dct_const_round_shift(temp1)); + step2[13] = WRAPLOW(dct_const_round_shift(temp2)); + step2[11] = step1[11]; + step2[12] = step1[12]; + + step2[16] = WRAPLOW(step1[16] + step1[19]); + step2[17] = WRAPLOW(step1[17] + step1[18]); + step2[18] = WRAPLOW(step1[17] - step1[18]); + step2[19] = WRAPLOW(step1[16] - step1[19]); + step2[20] = WRAPLOW(-step1[20] + step1[23]); + step2[21] = WRAPLOW(-step1[21] + step1[22]); + step2[22] = WRAPLOW(step1[21] + step1[22]); + step2[23] = WRAPLOW(step1[20] + step1[23]); + + step2[24] = WRAPLOW(step1[24] + step1[27]); + step2[25] = WRAPLOW(step1[25] + step1[26]); + step2[26] = WRAPLOW(step1[25] - step1[26]); + step2[27] = WRAPLOW(step1[24] - step1[27]); + step2[28] = WRAPLOW(-step1[28] + step1[31]); + step2[29] = WRAPLOW(-step1[29] + step1[30]); + step2[30] = WRAPLOW(step1[29] + step1[30]); + step2[31] = WRAPLOW(step1[28] + step1[31]); + + // stage 5 + step1[0] = WRAPLOW(step2[0] + step2[3]); + step1[1] = WRAPLOW(step2[1] + step2[2]); + step1[2] = WRAPLOW(step2[1] - step2[2]); + step1[3] = WRAPLOW(step2[0] - step2[3]); + step1[4] = step2[4]; + temp1 = (step2[6] - step2[5]) * cospi_16_64; + temp2 = (step2[5] + step2[6]) * cospi_16_64; + step1[5] = WRAPLOW(dct_const_round_shift(temp1)); + step1[6] = WRAPLOW(dct_const_round_shift(temp2)); + step1[7] = step2[7]; + + step1[8] = WRAPLOW(step2[8] + step2[11]); + step1[9] = WRAPLOW(step2[9] + step2[10]); + step1[10] = WRAPLOW(step2[9] - step2[10]); + step1[11] = WRAPLOW(step2[8] - step2[11]); + step1[12] = WRAPLOW(-step2[12] + step2[15]); + step1[13] = WRAPLOW(-step2[13] + step2[14]); + step1[14] = WRAPLOW(step2[13] + step2[14]); + step1[15] = WRAPLOW(step2[12] + step2[15]); + + step1[16] = step2[16]; + step1[17] = step2[17]; + temp1 = -step2[18] * cospi_8_64 + step2[29] * cospi_24_64; + temp2 = step2[18] * cospi_24_64 + step2[29] * cospi_8_64; + step1[18] = WRAPLOW(dct_const_round_shift(temp1)); + step1[29] = WRAPLOW(dct_const_round_shift(temp2)); + temp1 = -step2[19] * cospi_8_64 + step2[28] * cospi_24_64; + temp2 = step2[19] * cospi_24_64 + step2[28] * cospi_8_64; + step1[19] = WRAPLOW(dct_const_round_shift(temp1)); + step1[28] = WRAPLOW(dct_const_round_shift(temp2)); + temp1 = -step2[20] * cospi_24_64 - step2[27] * cospi_8_64; + temp2 = -step2[20] * cospi_8_64 + step2[27] * cospi_24_64; + step1[20] = WRAPLOW(dct_const_round_shift(temp1)); + step1[27] = WRAPLOW(dct_const_round_shift(temp2)); + temp1 = -step2[21] * cospi_24_64 - step2[26] * cospi_8_64; + temp2 = -step2[21] * cospi_8_64 + step2[26] * cospi_24_64; + step1[21] = WRAPLOW(dct_const_round_shift(temp1)); + step1[26] = WRAPLOW(dct_const_round_shift(temp2)); + step1[22] = step2[22]; + step1[23] = step2[23]; + step1[24] = step2[24]; + step1[25] = step2[25]; + step1[30] = step2[30]; + step1[31] = step2[31]; + + // stage 6 + step2[0] = WRAPLOW(step1[0] + step1[7]); + step2[1] = WRAPLOW(step1[1] + step1[6]); + step2[2] = WRAPLOW(step1[2] + step1[5]); + step2[3] = WRAPLOW(step1[3] + step1[4]); + step2[4] = WRAPLOW(step1[3] - step1[4]); + step2[5] = WRAPLOW(step1[2] - step1[5]); + step2[6] = WRAPLOW(step1[1] - step1[6]); + step2[7] = WRAPLOW(step1[0] - step1[7]); + step2[8] = step1[8]; + step2[9] = step1[9]; + temp1 = (-step1[10] + step1[13]) * cospi_16_64; + temp2 = (step1[10] + step1[13]) * cospi_16_64; + step2[10] = WRAPLOW(dct_const_round_shift(temp1)); + step2[13] = WRAPLOW(dct_const_round_shift(temp2)); + temp1 = (-step1[11] + step1[12]) * cospi_16_64; + temp2 = (step1[11] + step1[12]) * cospi_16_64; + step2[11] = WRAPLOW(dct_const_round_shift(temp1)); + step2[12] = WRAPLOW(dct_const_round_shift(temp2)); + step2[14] = step1[14]; + step2[15] = step1[15]; + + step2[16] = WRAPLOW(step1[16] + step1[23]); + step2[17] = WRAPLOW(step1[17] + step1[22]); + step2[18] = WRAPLOW(step1[18] + step1[21]); + step2[19] = WRAPLOW(step1[19] + step1[20]); + step2[20] = WRAPLOW(step1[19] - step1[20]); + step2[21] = WRAPLOW(step1[18] - step1[21]); + step2[22] = WRAPLOW(step1[17] - step1[22]); + step2[23] = WRAPLOW(step1[16] - step1[23]); + + step2[24] = WRAPLOW(-step1[24] + step1[31]); + step2[25] = WRAPLOW(-step1[25] + step1[30]); + step2[26] = WRAPLOW(-step1[26] + step1[29]); + step2[27] = WRAPLOW(-step1[27] + step1[28]); + step2[28] = WRAPLOW(step1[27] + step1[28]); + step2[29] = WRAPLOW(step1[26] + step1[29]); + step2[30] = WRAPLOW(step1[25] + step1[30]); + step2[31] = WRAPLOW(step1[24] + step1[31]); + + // stage 7 + step1[0] = WRAPLOW(step2[0] + step2[15]); + step1[1] = WRAPLOW(step2[1] + step2[14]); + step1[2] = WRAPLOW(step2[2] + step2[13]); + step1[3] = WRAPLOW(step2[3] + step2[12]); + step1[4] = WRAPLOW(step2[4] + step2[11]); + step1[5] = WRAPLOW(step2[5] + step2[10]); + step1[6] = WRAPLOW(step2[6] + step2[9]); + step1[7] = WRAPLOW(step2[7] + step2[8]); + step1[8] = WRAPLOW(step2[7] - step2[8]); + step1[9] = WRAPLOW(step2[6] - step2[9]); + step1[10] = WRAPLOW(step2[5] - step2[10]); + step1[11] = WRAPLOW(step2[4] - step2[11]); + step1[12] = WRAPLOW(step2[3] - step2[12]); + step1[13] = WRAPLOW(step2[2] - step2[13]); + step1[14] = WRAPLOW(step2[1] - step2[14]); + step1[15] = WRAPLOW(step2[0] - step2[15]); + + step1[16] = step2[16]; + step1[17] = step2[17]; + step1[18] = step2[18]; + step1[19] = step2[19]; + temp1 = (-step2[20] + step2[27]) * cospi_16_64; + temp2 = (step2[20] + step2[27]) * cospi_16_64; + step1[20] = WRAPLOW(dct_const_round_shift(temp1)); + step1[27] = WRAPLOW(dct_const_round_shift(temp2)); + temp1 = (-step2[21] + step2[26]) * cospi_16_64; + temp2 = (step2[21] + step2[26]) * cospi_16_64; + step1[21] = WRAPLOW(dct_const_round_shift(temp1)); + step1[26] = WRAPLOW(dct_const_round_shift(temp2)); + temp1 = (-step2[22] + step2[25]) * cospi_16_64; + temp2 = (step2[22] + step2[25]) * cospi_16_64; + step1[22] = WRAPLOW(dct_const_round_shift(temp1)); + step1[25] = WRAPLOW(dct_const_round_shift(temp2)); + temp1 = (-step2[23] + step2[24]) * cospi_16_64; + temp2 = (step2[23] + step2[24]) * cospi_16_64; + step1[23] = WRAPLOW(dct_const_round_shift(temp1)); + step1[24] = WRAPLOW(dct_const_round_shift(temp2)); + step1[28] = step2[28]; + step1[29] = step2[29]; + step1[30] = step2[30]; + step1[31] = step2[31]; + + // final stage + output[0] = WRAPLOW(step1[0] + step1[31]); + output[1] = WRAPLOW(step1[1] + step1[30]); + output[2] = WRAPLOW(step1[2] + step1[29]); + output[3] = WRAPLOW(step1[3] + step1[28]); + output[4] = WRAPLOW(step1[4] + step1[27]); + output[5] = WRAPLOW(step1[5] + step1[26]); + output[6] = WRAPLOW(step1[6] + step1[25]); + output[7] = WRAPLOW(step1[7] + step1[24]); + output[8] = WRAPLOW(step1[8] + step1[23]); + output[9] = WRAPLOW(step1[9] + step1[22]); + output[10] = WRAPLOW(step1[10] + step1[21]); + output[11] = WRAPLOW(step1[11] + step1[20]); + output[12] = WRAPLOW(step1[12] + step1[19]); + output[13] = WRAPLOW(step1[13] + step1[18]); + output[14] = WRAPLOW(step1[14] + step1[17]); + output[15] = WRAPLOW(step1[15] + step1[16]); + output[16] = WRAPLOW(step1[15] - step1[16]); + output[17] = WRAPLOW(step1[14] - step1[17]); + output[18] = WRAPLOW(step1[13] - step1[18]); + output[19] = WRAPLOW(step1[12] - step1[19]); + output[20] = WRAPLOW(step1[11] - step1[20]); + output[21] = WRAPLOW(step1[10] - step1[21]); + output[22] = WRAPLOW(step1[9] - step1[22]); + output[23] = WRAPLOW(step1[8] - step1[23]); + output[24] = WRAPLOW(step1[7] - step1[24]); + output[25] = WRAPLOW(step1[6] - step1[25]); + output[26] = WRAPLOW(step1[5] - step1[26]); + output[27] = WRAPLOW(step1[4] - step1[27]); + output[28] = WRAPLOW(step1[3] - step1[28]); + output[29] = WRAPLOW(step1[2] - step1[29]); + output[30] = WRAPLOW(step1[1] - step1[30]); + output[31] = WRAPLOW(step1[0] - step1[31]); +} + +void vpx_idct32x32_1024_add_c(const tran_low_t *input, uint8_t *dest, + int stride) { + tran_low_t out[32 * 32]; + tran_low_t *outptr = out; + int i, j; + tran_low_t temp_in[32], temp_out[32]; + + // Rows + for (i = 0; i < 32; ++i) { + int16_t zero_coeff[16]; + for (j = 0; j < 16; ++j) + zero_coeff[j] = input[2 * j] | input[2 * j + 1]; + for (j = 0; j < 8; ++j) + zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1]; + for (j = 0; j < 4; ++j) + zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1]; + for (j = 0; j < 2; ++j) + zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1]; + + if (zero_coeff[0] | zero_coeff[1]) + idct32_c(input, outptr); + else + memset(outptr, 0, sizeof(tran_low_t) * 32); + input += 32; + outptr += 32; + } + + // Columns + for (i = 0; i < 32; ++i) { + for (j = 0; j < 32; ++j) + temp_in[j] = out[j * 32 + i]; + idct32_c(temp_in, temp_out); + for (j = 0; j < 32; ++j) { + dest[j * stride + i] = clip_pixel_add(dest[j * stride + i], + ROUND_POWER_OF_TWO(temp_out[j], 6)); + } + } +} + +void vpx_idct32x32_135_add_c(const tran_low_t *input, uint8_t *dest, + int stride) { + tran_low_t out[32 * 32] = {0}; + tran_low_t *outptr = out; + int i, j; + tran_low_t temp_in[32], temp_out[32]; + + // Rows + // only upper-left 16x16 has non-zero coeff + for (i = 0; i < 16; ++i) { + idct32_c(input, outptr); + input += 32; + outptr += 32; + } + + // Columns + for (i = 0; i < 32; ++i) { + for (j = 0; j < 32; ++j) + temp_in[j] = out[j * 32 + i]; + idct32_c(temp_in, temp_out); + for (j = 0; j < 32; ++j) { + dest[j * stride + i] = clip_pixel_add(dest[j * stride + i], + ROUND_POWER_OF_TWO(temp_out[j], 6)); + } + } +} + +void vpx_idct32x32_34_add_c(const tran_low_t *input, uint8_t *dest, + int stride) { + tran_low_t out[32 * 32] = {0}; + tran_low_t *outptr = out; + int i, j; + tran_low_t temp_in[32], temp_out[32]; + + // Rows + // only upper-left 8x8 has non-zero coeff + for (i = 0; i < 8; ++i) { + idct32_c(input, outptr); + input += 32; + outptr += 32; + } + + // Columns + for (i = 0; i < 32; ++i) { + for (j = 0; j < 32; ++j) + temp_in[j] = out[j * 32 + i]; + idct32_c(temp_in, temp_out); + for (j = 0; j < 32; ++j) { + dest[j * stride + i] = clip_pixel_add(dest[j * stride + i], + ROUND_POWER_OF_TWO(temp_out[j], 6)); + } + } +} + +void vpx_idct32x32_1_add_c(const tran_low_t *input, uint8_t *dest, int stride) { + int i, j; + tran_high_t a1; + + tran_low_t out = WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64)); + out = WRAPLOW(dct_const_round_shift(out * cospi_16_64)); + a1 = ROUND_POWER_OF_TWO(out, 6); + + for (j = 0; j < 32; ++j) { + for (i = 0; i < 32; ++i) + dest[i] = clip_pixel_add(dest[i], a1); + dest += stride; + } +} + +#if CONFIG_VP9_HIGHBITDEPTH +void vpx_highbd_iwht4x4_16_add_c(const tran_low_t *input, uint8_t *dest8, + int stride, int bd) { + /* 4-point reversible, orthonormal inverse Walsh-Hadamard in 3.5 adds, + 0.5 shifts per pixel. */ + int i; + tran_low_t output[16]; + tran_high_t a1, b1, c1, d1, e1; + const tran_low_t *ip = input; + tran_low_t *op = output; + uint16_t *dest = CONVERT_TO_SHORTPTR(dest8); + + for (i = 0; i < 4; i++) { + a1 = ip[0] >> UNIT_QUANT_SHIFT; + c1 = ip[1] >> UNIT_QUANT_SHIFT; + d1 = ip[2] >> UNIT_QUANT_SHIFT; + b1 = ip[3] >> UNIT_QUANT_SHIFT; + a1 += c1; + d1 -= b1; + e1 = (a1 - d1) >> 1; + b1 = e1 - b1; + c1 = e1 - c1; + a1 -= b1; + d1 += c1; + op[0] = HIGHBD_WRAPLOW(a1, bd); + op[1] = HIGHBD_WRAPLOW(b1, bd); + op[2] = HIGHBD_WRAPLOW(c1, bd); + op[3] = HIGHBD_WRAPLOW(d1, bd); + ip += 4; + op += 4; + } + + ip = output; + for (i = 0; i < 4; i++) { + a1 = ip[4 * 0]; + c1 = ip[4 * 1]; + d1 = ip[4 * 2]; + b1 = ip[4 * 3]; + a1 += c1; + d1 -= b1; + e1 = (a1 - d1) >> 1; + b1 = e1 - b1; + c1 = e1 - c1; + a1 -= b1; + d1 += c1; + dest[stride * 0] = highbd_clip_pixel_add(dest[stride * 0], + HIGHBD_WRAPLOW(a1, bd), bd); + dest[stride * 1] = highbd_clip_pixel_add(dest[stride * 1], + HIGHBD_WRAPLOW(b1, bd), bd); + dest[stride * 2] = highbd_clip_pixel_add(dest[stride * 2], + HIGHBD_WRAPLOW(c1, bd), bd); + dest[stride * 3] = highbd_clip_pixel_add(dest[stride * 3], + HIGHBD_WRAPLOW(d1, bd), bd); + + ip++; + dest++; + } +} + +void vpx_highbd_iwht4x4_1_add_c(const tran_low_t *in, uint8_t *dest8, + int dest_stride, int bd) { + int i; + tran_high_t a1, e1; + tran_low_t tmp[4]; + const tran_low_t *ip = in; + tran_low_t *op = tmp; + uint16_t *dest = CONVERT_TO_SHORTPTR(dest8); + (void) bd; + + a1 = ip[0] >> UNIT_QUANT_SHIFT; + e1 = a1 >> 1; + a1 -= e1; + op[0] = HIGHBD_WRAPLOW(a1, bd); + op[1] = op[2] = op[3] = HIGHBD_WRAPLOW(e1, bd); + + ip = tmp; + for (i = 0; i < 4; i++) { + e1 = ip[0] >> 1; + a1 = ip[0] - e1; + dest[dest_stride * 0] = highbd_clip_pixel_add( + dest[dest_stride * 0], a1, bd); + dest[dest_stride * 1] = highbd_clip_pixel_add( + dest[dest_stride * 1], e1, bd); + dest[dest_stride * 2] = highbd_clip_pixel_add( + dest[dest_stride * 2], e1, bd); + dest[dest_stride * 3] = highbd_clip_pixel_add( + dest[dest_stride * 3], e1, bd); + ip++; + dest++; + } +} + +void vpx_highbd_idct4_c(const tran_low_t *input, tran_low_t *output, int bd) { + tran_low_t step[4]; + tran_high_t temp1, temp2; + (void) bd; + // stage 1 + temp1 = (input[0] + input[2]) * cospi_16_64; + temp2 = (input[0] - input[2]) * cospi_16_64; + step[0] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step[1] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + temp1 = input[1] * cospi_24_64 - input[3] * cospi_8_64; + temp2 = input[1] * cospi_8_64 + input[3] * cospi_24_64; + step[2] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step[3] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + + // stage 2 + output[0] = HIGHBD_WRAPLOW(step[0] + step[3], bd); + output[1] = HIGHBD_WRAPLOW(step[1] + step[2], bd); + output[2] = HIGHBD_WRAPLOW(step[1] - step[2], bd); + output[3] = HIGHBD_WRAPLOW(step[0] - step[3], bd); +} + +void vpx_highbd_idct4x4_16_add_c(const tran_low_t *input, uint8_t *dest8, + int stride, int bd) { + tran_low_t out[4 * 4]; + tran_low_t *outptr = out; + int i, j; + tran_low_t temp_in[4], temp_out[4]; + uint16_t *dest = CONVERT_TO_SHORTPTR(dest8); + + // Rows + for (i = 0; i < 4; ++i) { + vpx_highbd_idct4_c(input, outptr, bd); + input += 4; + outptr += 4; + } + + // Columns + for (i = 0; i < 4; ++i) { + for (j = 0; j < 4; ++j) + temp_in[j] = out[j * 4 + i]; + vpx_highbd_idct4_c(temp_in, temp_out, bd); + for (j = 0; j < 4; ++j) { + dest[j * stride + i] = highbd_clip_pixel_add( + dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 4), bd); + } + } +} + +void vpx_highbd_idct4x4_1_add_c(const tran_low_t *input, uint8_t *dest8, + int dest_stride, int bd) { + int i; + tran_high_t a1; + tran_low_t out = HIGHBD_WRAPLOW( + highbd_dct_const_round_shift(input[0] * cospi_16_64), bd); + uint16_t *dest = CONVERT_TO_SHORTPTR(dest8); + + out = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(out * cospi_16_64), bd); + a1 = ROUND_POWER_OF_TWO(out, 4); + + for (i = 0; i < 4; i++) { + dest[0] = highbd_clip_pixel_add(dest[0], a1, bd); + dest[1] = highbd_clip_pixel_add(dest[1], a1, bd); + dest[2] = highbd_clip_pixel_add(dest[2], a1, bd); + dest[3] = highbd_clip_pixel_add(dest[3], a1, bd); + dest += dest_stride; + } +} + +void vpx_highbd_idct8_c(const tran_low_t *input, tran_low_t *output, int bd) { + tran_low_t step1[8], step2[8]; + tran_high_t temp1, temp2; + // stage 1 + step1[0] = input[0]; + step1[2] = input[4]; + step1[1] = input[2]; + step1[3] = input[6]; + temp1 = input[1] * cospi_28_64 - input[7] * cospi_4_64; + temp2 = input[1] * cospi_4_64 + input[7] * cospi_28_64; + step1[4] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[7] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + temp1 = input[5] * cospi_12_64 - input[3] * cospi_20_64; + temp2 = input[5] * cospi_20_64 + input[3] * cospi_12_64; + step1[5] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[6] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + + // stage 2 & stage 3 - even half + vpx_highbd_idct4_c(step1, step1, bd); + + // stage 2 - odd half + step2[4] = HIGHBD_WRAPLOW(step1[4] + step1[5], bd); + step2[5] = HIGHBD_WRAPLOW(step1[4] - step1[5], bd); + step2[6] = HIGHBD_WRAPLOW(-step1[6] + step1[7], bd); + step2[7] = HIGHBD_WRAPLOW(step1[6] + step1[7], bd); + + // stage 3 - odd half + step1[4] = step2[4]; + temp1 = (step2[6] - step2[5]) * cospi_16_64; + temp2 = (step2[5] + step2[6]) * cospi_16_64; + step1[5] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[6] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + step1[7] = step2[7]; + + // stage 4 + output[0] = HIGHBD_WRAPLOW(step1[0] + step1[7], bd); + output[1] = HIGHBD_WRAPLOW(step1[1] + step1[6], bd); + output[2] = HIGHBD_WRAPLOW(step1[2] + step1[5], bd); + output[3] = HIGHBD_WRAPLOW(step1[3] + step1[4], bd); + output[4] = HIGHBD_WRAPLOW(step1[3] - step1[4], bd); + output[5] = HIGHBD_WRAPLOW(step1[2] - step1[5], bd); + output[6] = HIGHBD_WRAPLOW(step1[1] - step1[6], bd); + output[7] = HIGHBD_WRAPLOW(step1[0] - step1[7], bd); +} + +void vpx_highbd_idct8x8_64_add_c(const tran_low_t *input, uint8_t *dest8, + int stride, int bd) { + tran_low_t out[8 * 8]; + tran_low_t *outptr = out; + int i, j; + tran_low_t temp_in[8], temp_out[8]; + uint16_t *dest = CONVERT_TO_SHORTPTR(dest8); + + // First transform rows. + for (i = 0; i < 8; ++i) { + vpx_highbd_idct8_c(input, outptr, bd); + input += 8; + outptr += 8; + } + + // Then transform columns. + for (i = 0; i < 8; ++i) { + for (j = 0; j < 8; ++j) + temp_in[j] = out[j * 8 + i]; + vpx_highbd_idct8_c(temp_in, temp_out, bd); + for (j = 0; j < 8; ++j) { + dest[j * stride + i] = highbd_clip_pixel_add( + dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 5), bd); + } + } +} + +void vpx_highbd_idct8x8_1_add_c(const tran_low_t *input, uint8_t *dest8, + int stride, int bd) { + int i, j; + tran_high_t a1; + tran_low_t out = HIGHBD_WRAPLOW( + highbd_dct_const_round_shift(input[0] * cospi_16_64), bd); + uint16_t *dest = CONVERT_TO_SHORTPTR(dest8); + out = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(out * cospi_16_64), bd); + a1 = ROUND_POWER_OF_TWO(out, 5); + for (j = 0; j < 8; ++j) { + for (i = 0; i < 8; ++i) + dest[i] = highbd_clip_pixel_add(dest[i], a1, bd); + dest += stride; + } +} + +void vpx_highbd_iadst4_c(const tran_low_t *input, tran_low_t *output, int bd) { + tran_high_t s0, s1, s2, s3, s4, s5, s6, s7; + + tran_low_t x0 = input[0]; + tran_low_t x1 = input[1]; + tran_low_t x2 = input[2]; + tran_low_t x3 = input[3]; + (void) bd; + + if (!(x0 | x1 | x2 | x3)) { + memset(output, 0, 4 * sizeof(*output)); + return; + } + + s0 = sinpi_1_9 * x0; + s1 = sinpi_2_9 * x0; + s2 = sinpi_3_9 * x1; + s3 = sinpi_4_9 * x2; + s4 = sinpi_1_9 * x2; + s5 = sinpi_2_9 * x3; + s6 = sinpi_4_9 * x3; + s7 = (tran_high_t)HIGHBD_WRAPLOW(x0 - x2 + x3, bd); + + s0 = s0 + s3 + s5; + s1 = s1 - s4 - s6; + s3 = s2; + s2 = sinpi_3_9 * s7; + + // 1-D transform scaling factor is sqrt(2). + // The overall dynamic range is 14b (input) + 14b (multiplication scaling) + // + 1b (addition) = 29b. + // Hence the output bit depth is 15b. + output[0] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s0 + s3), bd); + output[1] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s1 + s3), bd); + output[2] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s2), bd); + output[3] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s0 + s1 - s3), bd); +} + +void vpx_highbd_iadst8_c(const tran_low_t *input, tran_low_t *output, int bd) { + tran_high_t s0, s1, s2, s3, s4, s5, s6, s7; + + tran_low_t x0 = input[7]; + tran_low_t x1 = input[0]; + tran_low_t x2 = input[5]; + tran_low_t x3 = input[2]; + tran_low_t x4 = input[3]; + tran_low_t x5 = input[4]; + tran_low_t x6 = input[1]; + tran_low_t x7 = input[6]; + (void) bd; + + if (!(x0 | x1 | x2 | x3 | x4 | x5 | x6 | x7)) { + memset(output, 0, 8 * sizeof(*output)); + return; + } + + // stage 1 + s0 = cospi_2_64 * x0 + cospi_30_64 * x1; + s1 = cospi_30_64 * x0 - cospi_2_64 * x1; + s2 = cospi_10_64 * x2 + cospi_22_64 * x3; + s3 = cospi_22_64 * x2 - cospi_10_64 * x3; + s4 = cospi_18_64 * x4 + cospi_14_64 * x5; + s5 = cospi_14_64 * x4 - cospi_18_64 * x5; + s6 = cospi_26_64 * x6 + cospi_6_64 * x7; + s7 = cospi_6_64 * x6 - cospi_26_64 * x7; + + x0 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s0 + s4), bd); + x1 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s1 + s5), bd); + x2 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s2 + s6), bd); + x3 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s3 + s7), bd); + x4 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s0 - s4), bd); + x5 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s1 - s5), bd); + x6 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s2 - s6), bd); + x7 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s3 - s7), bd); + + // stage 2 + s0 = x0; + s1 = x1; + s2 = x2; + s3 = x3; + s4 = cospi_8_64 * x4 + cospi_24_64 * x5; + s5 = cospi_24_64 * x4 - cospi_8_64 * x5; + s6 = -cospi_24_64 * x6 + cospi_8_64 * x7; + s7 = cospi_8_64 * x6 + cospi_24_64 * x7; + + x0 = HIGHBD_WRAPLOW(s0 + s2, bd); + x1 = HIGHBD_WRAPLOW(s1 + s3, bd); + x2 = HIGHBD_WRAPLOW(s0 - s2, bd); + x3 = HIGHBD_WRAPLOW(s1 - s3, bd); + x4 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s4 + s6), bd); + x5 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s5 + s7), bd); + x6 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s4 - s6), bd); + x7 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s5 - s7), bd); + + // stage 3 + s2 = cospi_16_64 * (x2 + x3); + s3 = cospi_16_64 * (x2 - x3); + s6 = cospi_16_64 * (x6 + x7); + s7 = cospi_16_64 * (x6 - x7); + + x2 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s2), bd); + x3 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s3), bd); + x6 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s6), bd); + x7 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s7), bd); + + output[0] = HIGHBD_WRAPLOW(x0, bd); + output[1] = HIGHBD_WRAPLOW(-x4, bd); + output[2] = HIGHBD_WRAPLOW(x6, bd); + output[3] = HIGHBD_WRAPLOW(-x2, bd); + output[4] = HIGHBD_WRAPLOW(x3, bd); + output[5] = HIGHBD_WRAPLOW(-x7, bd); + output[6] = HIGHBD_WRAPLOW(x5, bd); + output[7] = HIGHBD_WRAPLOW(-x1, bd); +} + +void vpx_highbd_idct8x8_10_add_c(const tran_low_t *input, uint8_t *dest8, + int stride, int bd) { + tran_low_t out[8 * 8] = { 0 }; + tran_low_t *outptr = out; + int i, j; + tran_low_t temp_in[8], temp_out[8]; + uint16_t *dest = CONVERT_TO_SHORTPTR(dest8); + + // First transform rows. + // Only first 4 row has non-zero coefs. + for (i = 0; i < 4; ++i) { + vpx_highbd_idct8_c(input, outptr, bd); + input += 8; + outptr += 8; + } + // Then transform columns. + for (i = 0; i < 8; ++i) { + for (j = 0; j < 8; ++j) + temp_in[j] = out[j * 8 + i]; + vpx_highbd_idct8_c(temp_in, temp_out, bd); + for (j = 0; j < 8; ++j) { + dest[j * stride + i] = highbd_clip_pixel_add( + dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 5), bd); + } + } +} + +void vpx_highbd_idct16_c(const tran_low_t *input, tran_low_t *output, int bd) { + tran_low_t step1[16], step2[16]; + tran_high_t temp1, temp2; + (void) bd; + + // stage 1 + step1[0] = input[0/2]; + step1[1] = input[16/2]; + step1[2] = input[8/2]; + step1[3] = input[24/2]; + step1[4] = input[4/2]; + step1[5] = input[20/2]; + step1[6] = input[12/2]; + step1[7] = input[28/2]; + step1[8] = input[2/2]; + step1[9] = input[18/2]; + step1[10] = input[10/2]; + step1[11] = input[26/2]; + step1[12] = input[6/2]; + step1[13] = input[22/2]; + step1[14] = input[14/2]; + step1[15] = input[30/2]; + + // stage 2 + step2[0] = step1[0]; + step2[1] = step1[1]; + step2[2] = step1[2]; + step2[3] = step1[3]; + step2[4] = step1[4]; + step2[5] = step1[5]; + step2[6] = step1[6]; + step2[7] = step1[7]; + + temp1 = step1[8] * cospi_30_64 - step1[15] * cospi_2_64; + temp2 = step1[8] * cospi_2_64 + step1[15] * cospi_30_64; + step2[8] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step2[15] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + + temp1 = step1[9] * cospi_14_64 - step1[14] * cospi_18_64; + temp2 = step1[9] * cospi_18_64 + step1[14] * cospi_14_64; + step2[9] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step2[14] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + + temp1 = step1[10] * cospi_22_64 - step1[13] * cospi_10_64; + temp2 = step1[10] * cospi_10_64 + step1[13] * cospi_22_64; + step2[10] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step2[13] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + + temp1 = step1[11] * cospi_6_64 - step1[12] * cospi_26_64; + temp2 = step1[11] * cospi_26_64 + step1[12] * cospi_6_64; + step2[11] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step2[12] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + + // stage 3 + step1[0] = step2[0]; + step1[1] = step2[1]; + step1[2] = step2[2]; + step1[3] = step2[3]; + + temp1 = step2[4] * cospi_28_64 - step2[7] * cospi_4_64; + temp2 = step2[4] * cospi_4_64 + step2[7] * cospi_28_64; + step1[4] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[7] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + temp1 = step2[5] * cospi_12_64 - step2[6] * cospi_20_64; + temp2 = step2[5] * cospi_20_64 + step2[6] * cospi_12_64; + step1[5] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[6] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + + step1[8] = HIGHBD_WRAPLOW(step2[8] + step2[9], bd); + step1[9] = HIGHBD_WRAPLOW(step2[8] - step2[9], bd); + step1[10] = HIGHBD_WRAPLOW(-step2[10] + step2[11], bd); + step1[11] = HIGHBD_WRAPLOW(step2[10] + step2[11], bd); + step1[12] = HIGHBD_WRAPLOW(step2[12] + step2[13], bd); + step1[13] = HIGHBD_WRAPLOW(step2[12] - step2[13], bd); + step1[14] = HIGHBD_WRAPLOW(-step2[14] + step2[15], bd); + step1[15] = HIGHBD_WRAPLOW(step2[14] + step2[15], bd); + + // stage 4 + temp1 = (step1[0] + step1[1]) * cospi_16_64; + temp2 = (step1[0] - step1[1]) * cospi_16_64; + step2[0] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step2[1] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + temp1 = step1[2] * cospi_24_64 - step1[3] * cospi_8_64; + temp2 = step1[2] * cospi_8_64 + step1[3] * cospi_24_64; + step2[2] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step2[3] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + step2[4] = HIGHBD_WRAPLOW(step1[4] + step1[5], bd); + step2[5] = HIGHBD_WRAPLOW(step1[4] - step1[5], bd); + step2[6] = HIGHBD_WRAPLOW(-step1[6] + step1[7], bd); + step2[7] = HIGHBD_WRAPLOW(step1[6] + step1[7], bd); + + step2[8] = step1[8]; + step2[15] = step1[15]; + temp1 = -step1[9] * cospi_8_64 + step1[14] * cospi_24_64; + temp2 = step1[9] * cospi_24_64 + step1[14] * cospi_8_64; + step2[9] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step2[14] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + temp1 = -step1[10] * cospi_24_64 - step1[13] * cospi_8_64; + temp2 = -step1[10] * cospi_8_64 + step1[13] * cospi_24_64; + step2[10] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step2[13] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + step2[11] = step1[11]; + step2[12] = step1[12]; + + // stage 5 + step1[0] = HIGHBD_WRAPLOW(step2[0] + step2[3], bd); + step1[1] = HIGHBD_WRAPLOW(step2[1] + step2[2], bd); + step1[2] = HIGHBD_WRAPLOW(step2[1] - step2[2], bd); + step1[3] = HIGHBD_WRAPLOW(step2[0] - step2[3], bd); + step1[4] = step2[4]; + temp1 = (step2[6] - step2[5]) * cospi_16_64; + temp2 = (step2[5] + step2[6]) * cospi_16_64; + step1[5] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[6] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + step1[7] = step2[7]; + + step1[8] = HIGHBD_WRAPLOW(step2[8] + step2[11], bd); + step1[9] = HIGHBD_WRAPLOW(step2[9] + step2[10], bd); + step1[10] = HIGHBD_WRAPLOW(step2[9] - step2[10], bd); + step1[11] = HIGHBD_WRAPLOW(step2[8] - step2[11], bd); + step1[12] = HIGHBD_WRAPLOW(-step2[12] + step2[15], bd); + step1[13] = HIGHBD_WRAPLOW(-step2[13] + step2[14], bd); + step1[14] = HIGHBD_WRAPLOW(step2[13] + step2[14], bd); + step1[15] = HIGHBD_WRAPLOW(step2[12] + step2[15], bd); + + // stage 6 + step2[0] = HIGHBD_WRAPLOW(step1[0] + step1[7], bd); + step2[1] = HIGHBD_WRAPLOW(step1[1] + step1[6], bd); + step2[2] = HIGHBD_WRAPLOW(step1[2] + step1[5], bd); + step2[3] = HIGHBD_WRAPLOW(step1[3] + step1[4], bd); + step2[4] = HIGHBD_WRAPLOW(step1[3] - step1[4], bd); + step2[5] = HIGHBD_WRAPLOW(step1[2] - step1[5], bd); + step2[6] = HIGHBD_WRAPLOW(step1[1] - step1[6], bd); + step2[7] = HIGHBD_WRAPLOW(step1[0] - step1[7], bd); + step2[8] = step1[8]; + step2[9] = step1[9]; + temp1 = (-step1[10] + step1[13]) * cospi_16_64; + temp2 = (step1[10] + step1[13]) * cospi_16_64; + step2[10] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step2[13] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + temp1 = (-step1[11] + step1[12]) * cospi_16_64; + temp2 = (step1[11] + step1[12]) * cospi_16_64; + step2[11] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step2[12] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + step2[14] = step1[14]; + step2[15] = step1[15]; + + // stage 7 + output[0] = HIGHBD_WRAPLOW(step2[0] + step2[15], bd); + output[1] = HIGHBD_WRAPLOW(step2[1] + step2[14], bd); + output[2] = HIGHBD_WRAPLOW(step2[2] + step2[13], bd); + output[3] = HIGHBD_WRAPLOW(step2[3] + step2[12], bd); + output[4] = HIGHBD_WRAPLOW(step2[4] + step2[11], bd); + output[5] = HIGHBD_WRAPLOW(step2[5] + step2[10], bd); + output[6] = HIGHBD_WRAPLOW(step2[6] + step2[9], bd); + output[7] = HIGHBD_WRAPLOW(step2[7] + step2[8], bd); + output[8] = HIGHBD_WRAPLOW(step2[7] - step2[8], bd); + output[9] = HIGHBD_WRAPLOW(step2[6] - step2[9], bd); + output[10] = HIGHBD_WRAPLOW(step2[5] - step2[10], bd); + output[11] = HIGHBD_WRAPLOW(step2[4] - step2[11], bd); + output[12] = HIGHBD_WRAPLOW(step2[3] - step2[12], bd); + output[13] = HIGHBD_WRAPLOW(step2[2] - step2[13], bd); + output[14] = HIGHBD_WRAPLOW(step2[1] - step2[14], bd); + output[15] = HIGHBD_WRAPLOW(step2[0] - step2[15], bd); +} + +void vpx_highbd_idct16x16_256_add_c(const tran_low_t *input, uint8_t *dest8, + int stride, int bd) { + tran_low_t out[16 * 16]; + tran_low_t *outptr = out; + int i, j; + tran_low_t temp_in[16], temp_out[16]; + uint16_t *dest = CONVERT_TO_SHORTPTR(dest8); + + // First transform rows. + for (i = 0; i < 16; ++i) { + vpx_highbd_idct16_c(input, outptr, bd); + input += 16; + outptr += 16; + } + + // Then transform columns. + for (i = 0; i < 16; ++i) { + for (j = 0; j < 16; ++j) + temp_in[j] = out[j * 16 + i]; + vpx_highbd_idct16_c(temp_in, temp_out, bd); + for (j = 0; j < 16; ++j) { + dest[j * stride + i] = highbd_clip_pixel_add( + dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 6), bd); + } + } +} + +void vpx_highbd_iadst16_c(const tran_low_t *input, tran_low_t *output, int bd) { + tran_high_t s0, s1, s2, s3, s4, s5, s6, s7, s8; + tran_high_t s9, s10, s11, s12, s13, s14, s15; + + tran_low_t x0 = input[15]; + tran_low_t x1 = input[0]; + tran_low_t x2 = input[13]; + tran_low_t x3 = input[2]; + tran_low_t x4 = input[11]; + tran_low_t x5 = input[4]; + tran_low_t x6 = input[9]; + tran_low_t x7 = input[6]; + tran_low_t x8 = input[7]; + tran_low_t x9 = input[8]; + tran_low_t x10 = input[5]; + tran_low_t x11 = input[10]; + tran_low_t x12 = input[3]; + tran_low_t x13 = input[12]; + tran_low_t x14 = input[1]; + tran_low_t x15 = input[14]; + (void) bd; + + if (!(x0 | x1 | x2 | x3 | x4 | x5 | x6 | x7 | x8 + | x9 | x10 | x11 | x12 | x13 | x14 | x15)) { + memset(output, 0, 16 * sizeof(*output)); + return; + } + + // stage 1 + s0 = x0 * cospi_1_64 + x1 * cospi_31_64; + s1 = x0 * cospi_31_64 - x1 * cospi_1_64; + s2 = x2 * cospi_5_64 + x3 * cospi_27_64; + s3 = x2 * cospi_27_64 - x3 * cospi_5_64; + s4 = x4 * cospi_9_64 + x5 * cospi_23_64; + s5 = x4 * cospi_23_64 - x5 * cospi_9_64; + s6 = x6 * cospi_13_64 + x7 * cospi_19_64; + s7 = x6 * cospi_19_64 - x7 * cospi_13_64; + s8 = x8 * cospi_17_64 + x9 * cospi_15_64; + s9 = x8 * cospi_15_64 - x9 * cospi_17_64; + s10 = x10 * cospi_21_64 + x11 * cospi_11_64; + s11 = x10 * cospi_11_64 - x11 * cospi_21_64; + s12 = x12 * cospi_25_64 + x13 * cospi_7_64; + s13 = x12 * cospi_7_64 - x13 * cospi_25_64; + s14 = x14 * cospi_29_64 + x15 * cospi_3_64; + s15 = x14 * cospi_3_64 - x15 * cospi_29_64; + + x0 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s0 + s8), bd); + x1 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s1 + s9), bd); + x2 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s2 + s10), bd); + x3 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s3 + s11), bd); + x4 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s4 + s12), bd); + x5 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s5 + s13), bd); + x6 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s6 + s14), bd); + x7 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s7 + s15), bd); + x8 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s0 - s8), bd); + x9 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s1 - s9), bd); + x10 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s2 - s10), bd); + x11 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s3 - s11), bd); + x12 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s4 - s12), bd); + x13 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s5 - s13), bd); + x14 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s6 - s14), bd); + x15 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s7 - s15), bd); + + // stage 2 + s0 = x0; + s1 = x1; + s2 = x2; + s3 = x3; + s4 = x4; + s5 = x5; + s6 = x6; + s7 = x7; + s8 = x8 * cospi_4_64 + x9 * cospi_28_64; + s9 = x8 * cospi_28_64 - x9 * cospi_4_64; + s10 = x10 * cospi_20_64 + x11 * cospi_12_64; + s11 = x10 * cospi_12_64 - x11 * cospi_20_64; + s12 = -x12 * cospi_28_64 + x13 * cospi_4_64; + s13 = x12 * cospi_4_64 + x13 * cospi_28_64; + s14 = -x14 * cospi_12_64 + x15 * cospi_20_64; + s15 = x14 * cospi_20_64 + x15 * cospi_12_64; + + x0 = HIGHBD_WRAPLOW(s0 + s4, bd); + x1 = HIGHBD_WRAPLOW(s1 + s5, bd); + x2 = HIGHBD_WRAPLOW(s2 + s6, bd); + x3 = HIGHBD_WRAPLOW(s3 + s7, bd); + x4 = HIGHBD_WRAPLOW(s0 - s4, bd); + x5 = HIGHBD_WRAPLOW(s1 - s5, bd); + x6 = HIGHBD_WRAPLOW(s2 - s6, bd); + x7 = HIGHBD_WRAPLOW(s3 - s7, bd); + x8 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s8 + s12), bd); + x9 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s9 + s13), bd); + x10 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s10 + s14), bd); + x11 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s11 + s15), bd); + x12 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s8 - s12), bd); + x13 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s9 - s13), bd); + x14 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s10 - s14), bd); + x15 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s11 - s15), bd); + + // stage 3 + s0 = x0; + s1 = x1; + s2 = x2; + s3 = x3; + s4 = x4 * cospi_8_64 + x5 * cospi_24_64; + s5 = x4 * cospi_24_64 - x5 * cospi_8_64; + s6 = -x6 * cospi_24_64 + x7 * cospi_8_64; + s7 = x6 * cospi_8_64 + x7 * cospi_24_64; + s8 = x8; + s9 = x9; + s10 = x10; + s11 = x11; + s12 = x12 * cospi_8_64 + x13 * cospi_24_64; + s13 = x12 * cospi_24_64 - x13 * cospi_8_64; + s14 = -x14 * cospi_24_64 + x15 * cospi_8_64; + s15 = x14 * cospi_8_64 + x15 * cospi_24_64; + + x0 = HIGHBD_WRAPLOW(s0 + s2, bd); + x1 = HIGHBD_WRAPLOW(s1 + s3, bd); + x2 = HIGHBD_WRAPLOW(s0 - s2, bd); + x3 = HIGHBD_WRAPLOW(s1 - s3, bd); + x4 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s4 + s6), bd); + x5 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s5 + s7), bd); + x6 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s4 - s6), bd); + x7 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s5 - s7), bd); + x8 = HIGHBD_WRAPLOW(s8 + s10, bd); + x9 = HIGHBD_WRAPLOW(s9 + s11, bd); + x10 = HIGHBD_WRAPLOW(s8 - s10, bd); + x11 = HIGHBD_WRAPLOW(s9 - s11, bd); + x12 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s12 + s14), bd); + x13 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s13 + s15), bd); + x14 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s12 - s14), bd); + x15 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s13 - s15), bd); + + // stage 4 + s2 = (- cospi_16_64) * (x2 + x3); + s3 = cospi_16_64 * (x2 - x3); + s6 = cospi_16_64 * (x6 + x7); + s7 = cospi_16_64 * (-x6 + x7); + s10 = cospi_16_64 * (x10 + x11); + s11 = cospi_16_64 * (-x10 + x11); + s14 = (- cospi_16_64) * (x14 + x15); + s15 = cospi_16_64 * (x14 - x15); + + x2 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s2), bd); + x3 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s3), bd); + x6 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s6), bd); + x7 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s7), bd); + x10 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s10), bd); + x11 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s11), bd); + x14 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s14), bd); + x15 = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(s15), bd); + + output[0] = HIGHBD_WRAPLOW(x0, bd); + output[1] = HIGHBD_WRAPLOW(-x8, bd); + output[2] = HIGHBD_WRAPLOW(x12, bd); + output[3] = HIGHBD_WRAPLOW(-x4, bd); + output[4] = HIGHBD_WRAPLOW(x6, bd); + output[5] = HIGHBD_WRAPLOW(x14, bd); + output[6] = HIGHBD_WRAPLOW(x10, bd); + output[7] = HIGHBD_WRAPLOW(x2, bd); + output[8] = HIGHBD_WRAPLOW(x3, bd); + output[9] = HIGHBD_WRAPLOW(x11, bd); + output[10] = HIGHBD_WRAPLOW(x15, bd); + output[11] = HIGHBD_WRAPLOW(x7, bd); + output[12] = HIGHBD_WRAPLOW(x5, bd); + output[13] = HIGHBD_WRAPLOW(-x13, bd); + output[14] = HIGHBD_WRAPLOW(x9, bd); + output[15] = HIGHBD_WRAPLOW(-x1, bd); +} + +void vpx_highbd_idct16x16_10_add_c(const tran_low_t *input, uint8_t *dest8, + int stride, int bd) { + tran_low_t out[16 * 16] = { 0 }; + tran_low_t *outptr = out; + int i, j; + tran_low_t temp_in[16], temp_out[16]; + uint16_t *dest = CONVERT_TO_SHORTPTR(dest8); + + // First transform rows. Since all non-zero dct coefficients are in + // upper-left 4x4 area, we only need to calculate first 4 rows here. + for (i = 0; i < 4; ++i) { + vpx_highbd_idct16_c(input, outptr, bd); + input += 16; + outptr += 16; + } + + // Then transform columns. + for (i = 0; i < 16; ++i) { + for (j = 0; j < 16; ++j) + temp_in[j] = out[j*16 + i]; + vpx_highbd_idct16_c(temp_in, temp_out, bd); + for (j = 0; j < 16; ++j) { + dest[j * stride + i] = highbd_clip_pixel_add( + dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 6), bd); + } + } +} + +void vpx_highbd_idct16x16_1_add_c(const tran_low_t *input, uint8_t *dest8, + int stride, int bd) { + int i, j; + tran_high_t a1; + tran_low_t out = HIGHBD_WRAPLOW( + highbd_dct_const_round_shift(input[0] * cospi_16_64), bd); + uint16_t *dest = CONVERT_TO_SHORTPTR(dest8); + + out = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(out * cospi_16_64), bd); + a1 = ROUND_POWER_OF_TWO(out, 6); + for (j = 0; j < 16; ++j) { + for (i = 0; i < 16; ++i) + dest[i] = highbd_clip_pixel_add(dest[i], a1, bd); + dest += stride; + } +} + +static void highbd_idct32_c(const tran_low_t *input, + tran_low_t *output, int bd) { + tran_low_t step1[32], step2[32]; + tran_high_t temp1, temp2; + (void) bd; + + // stage 1 + step1[0] = input[0]; + step1[1] = input[16]; + step1[2] = input[8]; + step1[3] = input[24]; + step1[4] = input[4]; + step1[5] = input[20]; + step1[6] = input[12]; + step1[7] = input[28]; + step1[8] = input[2]; + step1[9] = input[18]; + step1[10] = input[10]; + step1[11] = input[26]; + step1[12] = input[6]; + step1[13] = input[22]; + step1[14] = input[14]; + step1[15] = input[30]; + + temp1 = input[1] * cospi_31_64 - input[31] * cospi_1_64; + temp2 = input[1] * cospi_1_64 + input[31] * cospi_31_64; + step1[16] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[31] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + + temp1 = input[17] * cospi_15_64 - input[15] * cospi_17_64; + temp2 = input[17] * cospi_17_64 + input[15] * cospi_15_64; + step1[17] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[30] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + + temp1 = input[9] * cospi_23_64 - input[23] * cospi_9_64; + temp2 = input[9] * cospi_9_64 + input[23] * cospi_23_64; + step1[18] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[29] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + + temp1 = input[25] * cospi_7_64 - input[7] * cospi_25_64; + temp2 = input[25] * cospi_25_64 + input[7] * cospi_7_64; + step1[19] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[28] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + + temp1 = input[5] * cospi_27_64 - input[27] * cospi_5_64; + temp2 = input[5] * cospi_5_64 + input[27] * cospi_27_64; + step1[20] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[27] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + + temp1 = input[21] * cospi_11_64 - input[11] * cospi_21_64; + temp2 = input[21] * cospi_21_64 + input[11] * cospi_11_64; + step1[21] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[26] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + + temp1 = input[13] * cospi_19_64 - input[19] * cospi_13_64; + temp2 = input[13] * cospi_13_64 + input[19] * cospi_19_64; + step1[22] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[25] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + + temp1 = input[29] * cospi_3_64 - input[3] * cospi_29_64; + temp2 = input[29] * cospi_29_64 + input[3] * cospi_3_64; + step1[23] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[24] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + + // stage 2 + step2[0] = step1[0]; + step2[1] = step1[1]; + step2[2] = step1[2]; + step2[3] = step1[3]; + step2[4] = step1[4]; + step2[5] = step1[5]; + step2[6] = step1[6]; + step2[7] = step1[7]; + + temp1 = step1[8] * cospi_30_64 - step1[15] * cospi_2_64; + temp2 = step1[8] * cospi_2_64 + step1[15] * cospi_30_64; + step2[8] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step2[15] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + + temp1 = step1[9] * cospi_14_64 - step1[14] * cospi_18_64; + temp2 = step1[9] * cospi_18_64 + step1[14] * cospi_14_64; + step2[9] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step2[14] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + + temp1 = step1[10] * cospi_22_64 - step1[13] * cospi_10_64; + temp2 = step1[10] * cospi_10_64 + step1[13] * cospi_22_64; + step2[10] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step2[13] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + + temp1 = step1[11] * cospi_6_64 - step1[12] * cospi_26_64; + temp2 = step1[11] * cospi_26_64 + step1[12] * cospi_6_64; + step2[11] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step2[12] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + + step2[16] = HIGHBD_WRAPLOW(step1[16] + step1[17], bd); + step2[17] = HIGHBD_WRAPLOW(step1[16] - step1[17], bd); + step2[18] = HIGHBD_WRAPLOW(-step1[18] + step1[19], bd); + step2[19] = HIGHBD_WRAPLOW(step1[18] + step1[19], bd); + step2[20] = HIGHBD_WRAPLOW(step1[20] + step1[21], bd); + step2[21] = HIGHBD_WRAPLOW(step1[20] - step1[21], bd); + step2[22] = HIGHBD_WRAPLOW(-step1[22] + step1[23], bd); + step2[23] = HIGHBD_WRAPLOW(step1[22] + step1[23], bd); + step2[24] = HIGHBD_WRAPLOW(step1[24] + step1[25], bd); + step2[25] = HIGHBD_WRAPLOW(step1[24] - step1[25], bd); + step2[26] = HIGHBD_WRAPLOW(-step1[26] + step1[27], bd); + step2[27] = HIGHBD_WRAPLOW(step1[26] + step1[27], bd); + step2[28] = HIGHBD_WRAPLOW(step1[28] + step1[29], bd); + step2[29] = HIGHBD_WRAPLOW(step1[28] - step1[29], bd); + step2[30] = HIGHBD_WRAPLOW(-step1[30] + step1[31], bd); + step2[31] = HIGHBD_WRAPLOW(step1[30] + step1[31], bd); + + // stage 3 + step1[0] = step2[0]; + step1[1] = step2[1]; + step1[2] = step2[2]; + step1[3] = step2[3]; + + temp1 = step2[4] * cospi_28_64 - step2[7] * cospi_4_64; + temp2 = step2[4] * cospi_4_64 + step2[7] * cospi_28_64; + step1[4] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[7] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + temp1 = step2[5] * cospi_12_64 - step2[6] * cospi_20_64; + temp2 = step2[5] * cospi_20_64 + step2[6] * cospi_12_64; + step1[5] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[6] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + + step1[8] = HIGHBD_WRAPLOW(step2[8] + step2[9], bd); + step1[9] = HIGHBD_WRAPLOW(step2[8] - step2[9], bd); + step1[10] = HIGHBD_WRAPLOW(-step2[10] + step2[11], bd); + step1[11] = HIGHBD_WRAPLOW(step2[10] + step2[11], bd); + step1[12] = HIGHBD_WRAPLOW(step2[12] + step2[13], bd); + step1[13] = HIGHBD_WRAPLOW(step2[12] - step2[13], bd); + step1[14] = HIGHBD_WRAPLOW(-step2[14] + step2[15], bd); + step1[15] = HIGHBD_WRAPLOW(step2[14] + step2[15], bd); + + step1[16] = step2[16]; + step1[31] = step2[31]; + temp1 = -step2[17] * cospi_4_64 + step2[30] * cospi_28_64; + temp2 = step2[17] * cospi_28_64 + step2[30] * cospi_4_64; + step1[17] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[30] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + temp1 = -step2[18] * cospi_28_64 - step2[29] * cospi_4_64; + temp2 = -step2[18] * cospi_4_64 + step2[29] * cospi_28_64; + step1[18] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[29] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + step1[19] = step2[19]; + step1[20] = step2[20]; + temp1 = -step2[21] * cospi_20_64 + step2[26] * cospi_12_64; + temp2 = step2[21] * cospi_12_64 + step2[26] * cospi_20_64; + step1[21] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[26] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + temp1 = -step2[22] * cospi_12_64 - step2[25] * cospi_20_64; + temp2 = -step2[22] * cospi_20_64 + step2[25] * cospi_12_64; + step1[22] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[25] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + step1[23] = step2[23]; + step1[24] = step2[24]; + step1[27] = step2[27]; + step1[28] = step2[28]; + + // stage 4 + temp1 = (step1[0] + step1[1]) * cospi_16_64; + temp2 = (step1[0] - step1[1]) * cospi_16_64; + step2[0] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step2[1] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + temp1 = step1[2] * cospi_24_64 - step1[3] * cospi_8_64; + temp2 = step1[2] * cospi_8_64 + step1[3] * cospi_24_64; + step2[2] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step2[3] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + step2[4] = HIGHBD_WRAPLOW(step1[4] + step1[5], bd); + step2[5] = HIGHBD_WRAPLOW(step1[4] - step1[5], bd); + step2[6] = HIGHBD_WRAPLOW(-step1[6] + step1[7], bd); + step2[7] = HIGHBD_WRAPLOW(step1[6] + step1[7], bd); + + step2[8] = step1[8]; + step2[15] = step1[15]; + temp1 = -step1[9] * cospi_8_64 + step1[14] * cospi_24_64; + temp2 = step1[9] * cospi_24_64 + step1[14] * cospi_8_64; + step2[9] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step2[14] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + temp1 = -step1[10] * cospi_24_64 - step1[13] * cospi_8_64; + temp2 = -step1[10] * cospi_8_64 + step1[13] * cospi_24_64; + step2[10] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step2[13] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + step2[11] = step1[11]; + step2[12] = step1[12]; + + step2[16] = HIGHBD_WRAPLOW(step1[16] + step1[19], bd); + step2[17] = HIGHBD_WRAPLOW(step1[17] + step1[18], bd); + step2[18] = HIGHBD_WRAPLOW(step1[17] - step1[18], bd); + step2[19] = HIGHBD_WRAPLOW(step1[16] - step1[19], bd); + step2[20] = HIGHBD_WRAPLOW(-step1[20] + step1[23], bd); + step2[21] = HIGHBD_WRAPLOW(-step1[21] + step1[22], bd); + step2[22] = HIGHBD_WRAPLOW(step1[21] + step1[22], bd); + step2[23] = HIGHBD_WRAPLOW(step1[20] + step1[23], bd); + + step2[24] = HIGHBD_WRAPLOW(step1[24] + step1[27], bd); + step2[25] = HIGHBD_WRAPLOW(step1[25] + step1[26], bd); + step2[26] = HIGHBD_WRAPLOW(step1[25] - step1[26], bd); + step2[27] = HIGHBD_WRAPLOW(step1[24] - step1[27], bd); + step2[28] = HIGHBD_WRAPLOW(-step1[28] + step1[31], bd); + step2[29] = HIGHBD_WRAPLOW(-step1[29] + step1[30], bd); + step2[30] = HIGHBD_WRAPLOW(step1[29] + step1[30], bd); + step2[31] = HIGHBD_WRAPLOW(step1[28] + step1[31], bd); + + // stage 5 + step1[0] = HIGHBD_WRAPLOW(step2[0] + step2[3], bd); + step1[1] = HIGHBD_WRAPLOW(step2[1] + step2[2], bd); + step1[2] = HIGHBD_WRAPLOW(step2[1] - step2[2], bd); + step1[3] = HIGHBD_WRAPLOW(step2[0] - step2[3], bd); + step1[4] = step2[4]; + temp1 = (step2[6] - step2[5]) * cospi_16_64; + temp2 = (step2[5] + step2[6]) * cospi_16_64; + step1[5] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[6] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + step1[7] = step2[7]; + + step1[8] = HIGHBD_WRAPLOW(step2[8] + step2[11], bd); + step1[9] = HIGHBD_WRAPLOW(step2[9] + step2[10], bd); + step1[10] = HIGHBD_WRAPLOW(step2[9] - step2[10], bd); + step1[11] = HIGHBD_WRAPLOW(step2[8] - step2[11], bd); + step1[12] = HIGHBD_WRAPLOW(-step2[12] + step2[15], bd); + step1[13] = HIGHBD_WRAPLOW(-step2[13] + step2[14], bd); + step1[14] = HIGHBD_WRAPLOW(step2[13] + step2[14], bd); + step1[15] = HIGHBD_WRAPLOW(step2[12] + step2[15], bd); + + step1[16] = step2[16]; + step1[17] = step2[17]; + temp1 = -step2[18] * cospi_8_64 + step2[29] * cospi_24_64; + temp2 = step2[18] * cospi_24_64 + step2[29] * cospi_8_64; + step1[18] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[29] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + temp1 = -step2[19] * cospi_8_64 + step2[28] * cospi_24_64; + temp2 = step2[19] * cospi_24_64 + step2[28] * cospi_8_64; + step1[19] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[28] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + temp1 = -step2[20] * cospi_24_64 - step2[27] * cospi_8_64; + temp2 = -step2[20] * cospi_8_64 + step2[27] * cospi_24_64; + step1[20] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[27] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + temp1 = -step2[21] * cospi_24_64 - step2[26] * cospi_8_64; + temp2 = -step2[21] * cospi_8_64 + step2[26] * cospi_24_64; + step1[21] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[26] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + step1[22] = step2[22]; + step1[23] = step2[23]; + step1[24] = step2[24]; + step1[25] = step2[25]; + step1[30] = step2[30]; + step1[31] = step2[31]; + + // stage 6 + step2[0] = HIGHBD_WRAPLOW(step1[0] + step1[7], bd); + step2[1] = HIGHBD_WRAPLOW(step1[1] + step1[6], bd); + step2[2] = HIGHBD_WRAPLOW(step1[2] + step1[5], bd); + step2[3] = HIGHBD_WRAPLOW(step1[3] + step1[4], bd); + step2[4] = HIGHBD_WRAPLOW(step1[3] - step1[4], bd); + step2[5] = HIGHBD_WRAPLOW(step1[2] - step1[5], bd); + step2[6] = HIGHBD_WRAPLOW(step1[1] - step1[6], bd); + step2[7] = HIGHBD_WRAPLOW(step1[0] - step1[7], bd); + step2[8] = step1[8]; + step2[9] = step1[9]; + temp1 = (-step1[10] + step1[13]) * cospi_16_64; + temp2 = (step1[10] + step1[13]) * cospi_16_64; + step2[10] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step2[13] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + temp1 = (-step1[11] + step1[12]) * cospi_16_64; + temp2 = (step1[11] + step1[12]) * cospi_16_64; + step2[11] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step2[12] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + step2[14] = step1[14]; + step2[15] = step1[15]; + + step2[16] = HIGHBD_WRAPLOW(step1[16] + step1[23], bd); + step2[17] = HIGHBD_WRAPLOW(step1[17] + step1[22], bd); + step2[18] = HIGHBD_WRAPLOW(step1[18] + step1[21], bd); + step2[19] = HIGHBD_WRAPLOW(step1[19] + step1[20], bd); + step2[20] = HIGHBD_WRAPLOW(step1[19] - step1[20], bd); + step2[21] = HIGHBD_WRAPLOW(step1[18] - step1[21], bd); + step2[22] = HIGHBD_WRAPLOW(step1[17] - step1[22], bd); + step2[23] = HIGHBD_WRAPLOW(step1[16] - step1[23], bd); + + step2[24] = HIGHBD_WRAPLOW(-step1[24] + step1[31], bd); + step2[25] = HIGHBD_WRAPLOW(-step1[25] + step1[30], bd); + step2[26] = HIGHBD_WRAPLOW(-step1[26] + step1[29], bd); + step2[27] = HIGHBD_WRAPLOW(-step1[27] + step1[28], bd); + step2[28] = HIGHBD_WRAPLOW(step1[27] + step1[28], bd); + step2[29] = HIGHBD_WRAPLOW(step1[26] + step1[29], bd); + step2[30] = HIGHBD_WRAPLOW(step1[25] + step1[30], bd); + step2[31] = HIGHBD_WRAPLOW(step1[24] + step1[31], bd); + + // stage 7 + step1[0] = HIGHBD_WRAPLOW(step2[0] + step2[15], bd); + step1[1] = HIGHBD_WRAPLOW(step2[1] + step2[14], bd); + step1[2] = HIGHBD_WRAPLOW(step2[2] + step2[13], bd); + step1[3] = HIGHBD_WRAPLOW(step2[3] + step2[12], bd); + step1[4] = HIGHBD_WRAPLOW(step2[4] + step2[11], bd); + step1[5] = HIGHBD_WRAPLOW(step2[5] + step2[10], bd); + step1[6] = HIGHBD_WRAPLOW(step2[6] + step2[9], bd); + step1[7] = HIGHBD_WRAPLOW(step2[7] + step2[8], bd); + step1[8] = HIGHBD_WRAPLOW(step2[7] - step2[8], bd); + step1[9] = HIGHBD_WRAPLOW(step2[6] - step2[9], bd); + step1[10] = HIGHBD_WRAPLOW(step2[5] - step2[10], bd); + step1[11] = HIGHBD_WRAPLOW(step2[4] - step2[11], bd); + step1[12] = HIGHBD_WRAPLOW(step2[3] - step2[12], bd); + step1[13] = HIGHBD_WRAPLOW(step2[2] - step2[13], bd); + step1[14] = HIGHBD_WRAPLOW(step2[1] - step2[14], bd); + step1[15] = HIGHBD_WRAPLOW(step2[0] - step2[15], bd); + + step1[16] = step2[16]; + step1[17] = step2[17]; + step1[18] = step2[18]; + step1[19] = step2[19]; + temp1 = (-step2[20] + step2[27]) * cospi_16_64; + temp2 = (step2[20] + step2[27]) * cospi_16_64; + step1[20] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[27] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + temp1 = (-step2[21] + step2[26]) * cospi_16_64; + temp2 = (step2[21] + step2[26]) * cospi_16_64; + step1[21] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[26] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + temp1 = (-step2[22] + step2[25]) * cospi_16_64; + temp2 = (step2[22] + step2[25]) * cospi_16_64; + step1[22] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[25] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + temp1 = (-step2[23] + step2[24]) * cospi_16_64; + temp2 = (step2[23] + step2[24]) * cospi_16_64; + step1[23] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp1), bd); + step1[24] = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(temp2), bd); + step1[28] = step2[28]; + step1[29] = step2[29]; + step1[30] = step2[30]; + step1[31] = step2[31]; + + // final stage + output[0] = HIGHBD_WRAPLOW(step1[0] + step1[31], bd); + output[1] = HIGHBD_WRAPLOW(step1[1] + step1[30], bd); + output[2] = HIGHBD_WRAPLOW(step1[2] + step1[29], bd); + output[3] = HIGHBD_WRAPLOW(step1[3] + step1[28], bd); + output[4] = HIGHBD_WRAPLOW(step1[4] + step1[27], bd); + output[5] = HIGHBD_WRAPLOW(step1[5] + step1[26], bd); + output[6] = HIGHBD_WRAPLOW(step1[6] + step1[25], bd); + output[7] = HIGHBD_WRAPLOW(step1[7] + step1[24], bd); + output[8] = HIGHBD_WRAPLOW(step1[8] + step1[23], bd); + output[9] = HIGHBD_WRAPLOW(step1[9] + step1[22], bd); + output[10] = HIGHBD_WRAPLOW(step1[10] + step1[21], bd); + output[11] = HIGHBD_WRAPLOW(step1[11] + step1[20], bd); + output[12] = HIGHBD_WRAPLOW(step1[12] + step1[19], bd); + output[13] = HIGHBD_WRAPLOW(step1[13] + step1[18], bd); + output[14] = HIGHBD_WRAPLOW(step1[14] + step1[17], bd); + output[15] = HIGHBD_WRAPLOW(step1[15] + step1[16], bd); + output[16] = HIGHBD_WRAPLOW(step1[15] - step1[16], bd); + output[17] = HIGHBD_WRAPLOW(step1[14] - step1[17], bd); + output[18] = HIGHBD_WRAPLOW(step1[13] - step1[18], bd); + output[19] = HIGHBD_WRAPLOW(step1[12] - step1[19], bd); + output[20] = HIGHBD_WRAPLOW(step1[11] - step1[20], bd); + output[21] = HIGHBD_WRAPLOW(step1[10] - step1[21], bd); + output[22] = HIGHBD_WRAPLOW(step1[9] - step1[22], bd); + output[23] = HIGHBD_WRAPLOW(step1[8] - step1[23], bd); + output[24] = HIGHBD_WRAPLOW(step1[7] - step1[24], bd); + output[25] = HIGHBD_WRAPLOW(step1[6] - step1[25], bd); + output[26] = HIGHBD_WRAPLOW(step1[5] - step1[26], bd); + output[27] = HIGHBD_WRAPLOW(step1[4] - step1[27], bd); + output[28] = HIGHBD_WRAPLOW(step1[3] - step1[28], bd); + output[29] = HIGHBD_WRAPLOW(step1[2] - step1[29], bd); + output[30] = HIGHBD_WRAPLOW(step1[1] - step1[30], bd); + output[31] = HIGHBD_WRAPLOW(step1[0] - step1[31], bd); +} + +void vpx_highbd_idct32x32_1024_add_c(const tran_low_t *input, uint8_t *dest8, + int stride, int bd) { + tran_low_t out[32 * 32]; + tran_low_t *outptr = out; + int i, j; + tran_low_t temp_in[32], temp_out[32]; + uint16_t *dest = CONVERT_TO_SHORTPTR(dest8); + + // Rows + for (i = 0; i < 32; ++i) { + tran_low_t zero_coeff[16]; + for (j = 0; j < 16; ++j) + zero_coeff[j] = input[2 * j] | input[2 * j + 1]; + for (j = 0; j < 8; ++j) + zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1]; + for (j = 0; j < 4; ++j) + zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1]; + for (j = 0; j < 2; ++j) + zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1]; + + if (zero_coeff[0] | zero_coeff[1]) + highbd_idct32_c(input, outptr, bd); + else + memset(outptr, 0, sizeof(tran_low_t) * 32); + input += 32; + outptr += 32; + } + + // Columns + for (i = 0; i < 32; ++i) { + for (j = 0; j < 32; ++j) + temp_in[j] = out[j * 32 + i]; + highbd_idct32_c(temp_in, temp_out, bd); + for (j = 0; j < 32; ++j) { + dest[j * stride + i] = highbd_clip_pixel_add( + dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 6), bd); + } + } +} + +void vpx_highbd_idct32x32_34_add_c(const tran_low_t *input, uint8_t *dest8, + int stride, int bd) { + tran_low_t out[32 * 32] = {0}; + tran_low_t *outptr = out; + int i, j; + tran_low_t temp_in[32], temp_out[32]; + uint16_t *dest = CONVERT_TO_SHORTPTR(dest8); + + // Rows + // Only upper-left 8x8 has non-zero coeff. + for (i = 0; i < 8; ++i) { + highbd_idct32_c(input, outptr, bd); + input += 32; + outptr += 32; + } + // Columns + for (i = 0; i < 32; ++i) { + for (j = 0; j < 32; ++j) + temp_in[j] = out[j * 32 + i]; + highbd_idct32_c(temp_in, temp_out, bd); + for (j = 0; j < 32; ++j) { + dest[j * stride + i] = highbd_clip_pixel_add( + dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 6), bd); + } + } +} + +void vpx_highbd_idct32x32_1_add_c(const tran_low_t *input, uint8_t *dest8, + int stride, int bd) { + int i, j; + int a1; + uint16_t *dest = CONVERT_TO_SHORTPTR(dest8); + + tran_low_t out = HIGHBD_WRAPLOW( + highbd_dct_const_round_shift(input[0] * cospi_16_64), bd); + out = HIGHBD_WRAPLOW(highbd_dct_const_round_shift(out * cospi_16_64), bd); + a1 = ROUND_POWER_OF_TWO(out, 6); + + for (j = 0; j < 32; ++j) { + for (i = 0; i < 32; ++i) + dest[i] = highbd_clip_pixel_add(dest[i], a1, bd); + dest += stride; + } +} +#endif // CONFIG_VP9_HIGHBITDEPTH diff --git a/thirdparty/libvpx/vpx_dsp/inv_txfm.h b/thirdparty/libvpx/vpx_dsp/inv_txfm.h new file mode 100644 index 0000000000..9cfe1be3a7 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/inv_txfm.h @@ -0,0 +1,133 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VPX_DSP_INV_TXFM_H_ +#define VPX_DSP_INV_TXFM_H_ + +#include <assert.h> + +#include "./vpx_config.h" +#include "vpx_dsp/txfm_common.h" +#include "vpx_ports/mem.h" + +#ifdef __cplusplus +extern "C" { +#endif + +static INLINE tran_high_t check_range(tran_high_t input) { +#if CONFIG_COEFFICIENT_RANGE_CHECKING + // For valid VP9 input streams, intermediate stage coefficients should always + // stay within the range of a signed 16 bit integer. Coefficients can go out + // of this range for invalid/corrupt VP9 streams. However, strictly checking + // this range for every intermediate coefficient can burdensome for a decoder, + // therefore the following assertion is only enabled when configured with + // --enable-coefficient-range-checking. + assert(INT16_MIN <= input); + assert(input <= INT16_MAX); +#endif // CONFIG_COEFFICIENT_RANGE_CHECKING + return input; +} + +static INLINE tran_high_t dct_const_round_shift(tran_high_t input) { + tran_high_t rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS); + return (tran_high_t)rv; +} + +#if CONFIG_VP9_HIGHBITDEPTH +static INLINE tran_high_t highbd_check_range(tran_high_t input, + int bd) { +#if CONFIG_COEFFICIENT_RANGE_CHECKING + // For valid highbitdepth VP9 streams, intermediate stage coefficients will + // stay within the ranges: + // - 8 bit: signed 16 bit integer + // - 10 bit: signed 18 bit integer + // - 12 bit: signed 20 bit integer + const int32_t int_max = (1 << (7 + bd)) - 1; + const int32_t int_min = -int_max - 1; + assert(int_min <= input); + assert(input <= int_max); + (void) int_min; +#endif // CONFIG_COEFFICIENT_RANGE_CHECKING + (void) bd; + return input; +} + +static INLINE tran_high_t highbd_dct_const_round_shift(tran_high_t input) { + tran_high_t rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS); + return (tran_high_t)rv; +} +#endif // CONFIG_VP9_HIGHBITDEPTH + +#if CONFIG_EMULATE_HARDWARE +// When CONFIG_EMULATE_HARDWARE is 1 the transform performs a +// non-normative method to handle overflows. A stream that causes +// overflows in the inverse transform is considered invalid in VP9, +// and a hardware implementer is free to choose any reasonable +// method to handle overflows. However to aid in hardware +// verification they can use a specific implementation of the +// WRAPLOW() macro below that is identical to their intended +// hardware implementation (and also use configure options to trigger +// the C-implementation of the transform). +// +// The particular WRAPLOW implementation below performs strict +// overflow wrapping to match common hardware implementations. +// bd of 8 uses trans_low with 16bits, need to remove 16bits +// bd of 10 uses trans_low with 18bits, need to remove 14bits +// bd of 12 uses trans_low with 20bits, need to remove 12bits +// bd of x uses trans_low with 8+x bits, need to remove 24-x bits + +#define WRAPLOW(x) ((((int32_t)check_range(x)) << 16) >> 16) +#if CONFIG_VP9_HIGHBITDEPTH +#define HIGHBD_WRAPLOW(x, bd) \ + ((((int32_t)highbd_check_range((x), bd)) << (24 - bd)) >> (24 - bd)) +#endif // CONFIG_VP9_HIGHBITDEPTH + +#else // CONFIG_EMULATE_HARDWARE + +#define WRAPLOW(x) ((int32_t)check_range(x)) +#if CONFIG_VP9_HIGHBITDEPTH +#define HIGHBD_WRAPLOW(x, bd) \ + ((int32_t)highbd_check_range((x), bd)) +#endif // CONFIG_VP9_HIGHBITDEPTH +#endif // CONFIG_EMULATE_HARDWARE + +void idct4_c(const tran_low_t *input, tran_low_t *output); +void idct8_c(const tran_low_t *input, tran_low_t *output); +void idct16_c(const tran_low_t *input, tran_low_t *output); +void idct32_c(const tran_low_t *input, tran_low_t *output); +void iadst4_c(const tran_low_t *input, tran_low_t *output); +void iadst8_c(const tran_low_t *input, tran_low_t *output); +void iadst16_c(const tran_low_t *input, tran_low_t *output); + +#if CONFIG_VP9_HIGHBITDEPTH +void vpx_highbd_idct4_c(const tran_low_t *input, tran_low_t *output, int bd); +void vpx_highbd_idct8_c(const tran_low_t *input, tran_low_t *output, int bd); +void vpx_highbd_idct16_c(const tran_low_t *input, tran_low_t *output, int bd); + +void vpx_highbd_iadst4_c(const tran_low_t *input, tran_low_t *output, int bd); +void vpx_highbd_iadst8_c(const tran_low_t *input, tran_low_t *output, int bd); +void vpx_highbd_iadst16_c(const tran_low_t *input, tran_low_t *output, int bd); + +static INLINE uint16_t highbd_clip_pixel_add(uint16_t dest, tran_high_t trans, + int bd) { + trans = HIGHBD_WRAPLOW(trans, bd); + return clip_pixel_highbd(dest + (int)trans, bd); +} +#endif + +static INLINE uint8_t clip_pixel_add(uint8_t dest, tran_high_t trans) { + trans = WRAPLOW(trans); + return clip_pixel(dest + (int)trans); +} +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VPX_DSP_INV_TXFM_H_ diff --git a/thirdparty/libvpx/vpx_dsp/loopfilter.c b/thirdparty/libvpx/vpx_dsp/loopfilter.c new file mode 100644 index 0000000000..645a1ab95e --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/loopfilter.c @@ -0,0 +1,767 @@ +/* + * Copyright (c) 2015 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <stdlib.h> + +#include "./vpx_config.h" +#include "./vpx_dsp_rtcd.h" +#include "vpx_dsp/vpx_dsp_common.h" +#include "vpx_ports/mem.h" + +static INLINE int8_t signed_char_clamp(int t) { + return (int8_t)clamp(t, -128, 127); +} + +#if CONFIG_VP9_HIGHBITDEPTH +static INLINE int16_t signed_char_clamp_high(int t, int bd) { + switch (bd) { + case 10: + return (int16_t)clamp(t, -128*4, 128*4-1); + case 12: + return (int16_t)clamp(t, -128*16, 128*16-1); + case 8: + default: + return (int16_t)clamp(t, -128, 128-1); + } +} +#endif + +// should we apply any filter at all: 11111111 yes, 00000000 no +static INLINE int8_t filter_mask(uint8_t limit, uint8_t blimit, + uint8_t p3, uint8_t p2, + uint8_t p1, uint8_t p0, + uint8_t q0, uint8_t q1, + uint8_t q2, uint8_t q3) { + int8_t mask = 0; + mask |= (abs(p3 - p2) > limit) * -1; + mask |= (abs(p2 - p1) > limit) * -1; + mask |= (abs(p1 - p0) > limit) * -1; + mask |= (abs(q1 - q0) > limit) * -1; + mask |= (abs(q2 - q1) > limit) * -1; + mask |= (abs(q3 - q2) > limit) * -1; + mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2 > blimit) * -1; + return ~mask; +} + +static INLINE int8_t flat_mask4(uint8_t thresh, + uint8_t p3, uint8_t p2, + uint8_t p1, uint8_t p0, + uint8_t q0, uint8_t q1, + uint8_t q2, uint8_t q3) { + int8_t mask = 0; + mask |= (abs(p1 - p0) > thresh) * -1; + mask |= (abs(q1 - q0) > thresh) * -1; + mask |= (abs(p2 - p0) > thresh) * -1; + mask |= (abs(q2 - q0) > thresh) * -1; + mask |= (abs(p3 - p0) > thresh) * -1; + mask |= (abs(q3 - q0) > thresh) * -1; + return ~mask; +} + +static INLINE int8_t flat_mask5(uint8_t thresh, + uint8_t p4, uint8_t p3, + uint8_t p2, uint8_t p1, + uint8_t p0, uint8_t q0, + uint8_t q1, uint8_t q2, + uint8_t q3, uint8_t q4) { + int8_t mask = ~flat_mask4(thresh, p3, p2, p1, p0, q0, q1, q2, q3); + mask |= (abs(p4 - p0) > thresh) * -1; + mask |= (abs(q4 - q0) > thresh) * -1; + return ~mask; +} + +// is there high edge variance internal edge: 11111111 yes, 00000000 no +static INLINE int8_t hev_mask(uint8_t thresh, uint8_t p1, uint8_t p0, + uint8_t q0, uint8_t q1) { + int8_t hev = 0; + hev |= (abs(p1 - p0) > thresh) * -1; + hev |= (abs(q1 - q0) > thresh) * -1; + return hev; +} + +static INLINE void filter4(int8_t mask, uint8_t thresh, uint8_t *op1, + uint8_t *op0, uint8_t *oq0, uint8_t *oq1) { + int8_t filter1, filter2; + + const int8_t ps1 = (int8_t) *op1 ^ 0x80; + const int8_t ps0 = (int8_t) *op0 ^ 0x80; + const int8_t qs0 = (int8_t) *oq0 ^ 0x80; + const int8_t qs1 = (int8_t) *oq1 ^ 0x80; + const uint8_t hev = hev_mask(thresh, *op1, *op0, *oq0, *oq1); + + // add outer taps if we have high edge variance + int8_t filter = signed_char_clamp(ps1 - qs1) & hev; + + // inner taps + filter = signed_char_clamp(filter + 3 * (qs0 - ps0)) & mask; + + // save bottom 3 bits so that we round one side +4 and the other +3 + // if it equals 4 we'll set to adjust by -1 to account for the fact + // we'd round 3 the other way + filter1 = signed_char_clamp(filter + 4) >> 3; + filter2 = signed_char_clamp(filter + 3) >> 3; + + *oq0 = signed_char_clamp(qs0 - filter1) ^ 0x80; + *op0 = signed_char_clamp(ps0 + filter2) ^ 0x80; + + // outer tap adjustments + filter = ROUND_POWER_OF_TWO(filter1, 1) & ~hev; + + *oq1 = signed_char_clamp(qs1 - filter) ^ 0x80; + *op1 = signed_char_clamp(ps1 + filter) ^ 0x80; +} + +void vpx_lpf_horizontal_4_c(uint8_t *s, int p /* pitch */, + const uint8_t *blimit, const uint8_t *limit, + const uint8_t *thresh) { + int i; + + // loop filter designed to work using chars so that we can make maximum use + // of 8 bit simd instructions. + for (i = 0; i < 8; ++i) { + const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p]; + const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p]; + const int8_t mask = filter_mask(*limit, *blimit, + p3, p2, p1, p0, q0, q1, q2, q3); + filter4(mask, *thresh, s - 2 * p, s - 1 * p, s, s + 1 * p); + ++s; + } +} + +void vpx_lpf_horizontal_4_dual_c(uint8_t *s, int p, const uint8_t *blimit0, + const uint8_t *limit0, const uint8_t *thresh0, + const uint8_t *blimit1, const uint8_t *limit1, + const uint8_t *thresh1) { + vpx_lpf_horizontal_4_c(s, p, blimit0, limit0, thresh0); + vpx_lpf_horizontal_4_c(s + 8, p, blimit1, limit1, thresh1); +} + +void vpx_lpf_vertical_4_c(uint8_t *s, int pitch, const uint8_t *blimit, + const uint8_t *limit, const uint8_t *thresh) { + int i; + + // loop filter designed to work using chars so that we can make maximum use + // of 8 bit simd instructions. + for (i = 0; i < 8; ++i) { + const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1]; + const uint8_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3]; + const int8_t mask = filter_mask(*limit, *blimit, + p3, p2, p1, p0, q0, q1, q2, q3); + filter4(mask, *thresh, s - 2, s - 1, s, s + 1); + s += pitch; + } +} + +void vpx_lpf_vertical_4_dual_c(uint8_t *s, int pitch, const uint8_t *blimit0, + const uint8_t *limit0, const uint8_t *thresh0, + const uint8_t *blimit1, const uint8_t *limit1, + const uint8_t *thresh1) { + vpx_lpf_vertical_4_c(s, pitch, blimit0, limit0, thresh0); + vpx_lpf_vertical_4_c(s + 8 * pitch, pitch, blimit1, limit1, thresh1); +} + +static INLINE void filter8(int8_t mask, uint8_t thresh, uint8_t flat, + uint8_t *op3, uint8_t *op2, + uint8_t *op1, uint8_t *op0, + uint8_t *oq0, uint8_t *oq1, + uint8_t *oq2, uint8_t *oq3) { + if (flat && mask) { + const uint8_t p3 = *op3, p2 = *op2, p1 = *op1, p0 = *op0; + const uint8_t q0 = *oq0, q1 = *oq1, q2 = *oq2, q3 = *oq3; + + // 7-tap filter [1, 1, 1, 2, 1, 1, 1] + *op2 = ROUND_POWER_OF_TWO(p3 + p3 + p3 + 2 * p2 + p1 + p0 + q0, 3); + *op1 = ROUND_POWER_OF_TWO(p3 + p3 + p2 + 2 * p1 + p0 + q0 + q1, 3); + *op0 = ROUND_POWER_OF_TWO(p3 + p2 + p1 + 2 * p0 + q0 + q1 + q2, 3); + *oq0 = ROUND_POWER_OF_TWO(p2 + p1 + p0 + 2 * q0 + q1 + q2 + q3, 3); + *oq1 = ROUND_POWER_OF_TWO(p1 + p0 + q0 + 2 * q1 + q2 + q3 + q3, 3); + *oq2 = ROUND_POWER_OF_TWO(p0 + q0 + q1 + 2 * q2 + q3 + q3 + q3, 3); + } else { + filter4(mask, thresh, op1, op0, oq0, oq1); + } +} + +void vpx_lpf_horizontal_8_c(uint8_t *s, int p, const uint8_t *blimit, + const uint8_t *limit, const uint8_t *thresh) { + int i; + + // loop filter designed to work using chars so that we can make maximum use + // of 8 bit simd instructions. + for (i = 0; i < 8; ++i) { + const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p]; + const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p]; + + const int8_t mask = filter_mask(*limit, *blimit, + p3, p2, p1, p0, q0, q1, q2, q3); + const int8_t flat = flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3); + filter8(mask, *thresh, flat, s - 4 * p, s - 3 * p, s - 2 * p, s - 1 * p, + s, s + 1 * p, s + 2 * p, s + 3 * p); + ++s; + } +} + +void vpx_lpf_horizontal_8_dual_c(uint8_t *s, int p, const uint8_t *blimit0, + const uint8_t *limit0, const uint8_t *thresh0, + const uint8_t *blimit1, const uint8_t *limit1, + const uint8_t *thresh1) { + vpx_lpf_horizontal_8_c(s, p, blimit0, limit0, thresh0); + vpx_lpf_horizontal_8_c(s + 8, p, blimit1, limit1, thresh1); +} + +void vpx_lpf_vertical_8_c(uint8_t *s, int pitch, const uint8_t *blimit, + const uint8_t *limit, const uint8_t *thresh) { + int i; + + for (i = 0; i < 8; ++i) { + const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1]; + const uint8_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3]; + const int8_t mask = filter_mask(*limit, *blimit, + p3, p2, p1, p0, q0, q1, q2, q3); + const int8_t flat = flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3); + filter8(mask, *thresh, flat, s - 4, s - 3, s - 2, s - 1, + s, s + 1, s + 2, s + 3); + s += pitch; + } +} + +void vpx_lpf_vertical_8_dual_c(uint8_t *s, int pitch, const uint8_t *blimit0, + const uint8_t *limit0, const uint8_t *thresh0, + const uint8_t *blimit1, const uint8_t *limit1, + const uint8_t *thresh1) { + vpx_lpf_vertical_8_c(s, pitch, blimit0, limit0, thresh0); + vpx_lpf_vertical_8_c(s + 8 * pitch, pitch, blimit1, limit1, thresh1); +} + +static INLINE void filter16(int8_t mask, uint8_t thresh, + uint8_t flat, uint8_t flat2, + uint8_t *op7, uint8_t *op6, + uint8_t *op5, uint8_t *op4, + uint8_t *op3, uint8_t *op2, + uint8_t *op1, uint8_t *op0, + uint8_t *oq0, uint8_t *oq1, + uint8_t *oq2, uint8_t *oq3, + uint8_t *oq4, uint8_t *oq5, + uint8_t *oq6, uint8_t *oq7) { + if (flat2 && flat && mask) { + const uint8_t p7 = *op7, p6 = *op6, p5 = *op5, p4 = *op4, + p3 = *op3, p2 = *op2, p1 = *op1, p0 = *op0; + + const uint8_t q0 = *oq0, q1 = *oq1, q2 = *oq2, q3 = *oq3, + q4 = *oq4, q5 = *oq5, q6 = *oq6, q7 = *oq7; + + // 15-tap filter [1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1] + *op6 = ROUND_POWER_OF_TWO(p7 * 7 + p6 * 2 + p5 + p4 + p3 + p2 + p1 + p0 + + q0, 4); + *op5 = ROUND_POWER_OF_TWO(p7 * 6 + p6 + p5 * 2 + p4 + p3 + p2 + p1 + p0 + + q0 + q1, 4); + *op4 = ROUND_POWER_OF_TWO(p7 * 5 + p6 + p5 + p4 * 2 + p3 + p2 + p1 + p0 + + q0 + q1 + q2, 4); + *op3 = ROUND_POWER_OF_TWO(p7 * 4 + p6 + p5 + p4 + p3 * 2 + p2 + p1 + p0 + + q0 + q1 + q2 + q3, 4); + *op2 = ROUND_POWER_OF_TWO(p7 * 3 + p6 + p5 + p4 + p3 + p2 * 2 + p1 + p0 + + q0 + q1 + q2 + q3 + q4, 4); + *op1 = ROUND_POWER_OF_TWO(p7 * 2 + p6 + p5 + p4 + p3 + p2 + p1 * 2 + p0 + + q0 + q1 + q2 + q3 + q4 + q5, 4); + *op0 = ROUND_POWER_OF_TWO(p7 + p6 + p5 + p4 + p3 + p2 + p1 + p0 * 2 + + q0 + q1 + q2 + q3 + q4 + q5 + q6, 4); + *oq0 = ROUND_POWER_OF_TWO(p6 + p5 + p4 + p3 + p2 + p1 + p0 + + q0 * 2 + q1 + q2 + q3 + q4 + q5 + q6 + q7, 4); + *oq1 = ROUND_POWER_OF_TWO(p5 + p4 + p3 + p2 + p1 + p0 + + q0 + q1 * 2 + q2 + q3 + q4 + q5 + q6 + q7 * 2, 4); + *oq2 = ROUND_POWER_OF_TWO(p4 + p3 + p2 + p1 + p0 + + q0 + q1 + q2 * 2 + q3 + q4 + q5 + q6 + q7 * 3, 4); + *oq3 = ROUND_POWER_OF_TWO(p3 + p2 + p1 + p0 + + q0 + q1 + q2 + q3 * 2 + q4 + q5 + q6 + q7 * 4, 4); + *oq4 = ROUND_POWER_OF_TWO(p2 + p1 + p0 + + q0 + q1 + q2 + q3 + q4 * 2 + q5 + q6 + q7 * 5, 4); + *oq5 = ROUND_POWER_OF_TWO(p1 + p0 + + q0 + q1 + q2 + q3 + q4 + q5 * 2 + q6 + q7 * 6, 4); + *oq6 = ROUND_POWER_OF_TWO(p0 + + q0 + q1 + q2 + q3 + q4 + q5 + q6 * 2 + q7 * 7, 4); + } else { + filter8(mask, thresh, flat, op3, op2, op1, op0, oq0, oq1, oq2, oq3); + } +} + +static void mb_lpf_horizontal_edge_w(uint8_t *s, int p, const uint8_t *blimit, + const uint8_t *limit, + const uint8_t *thresh, int count) { + int i; + + // loop filter designed to work using chars so that we can make maximum use + // of 8 bit simd instructions. + for (i = 0; i < 8 * count; ++i) { + const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p]; + const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p]; + const int8_t mask = filter_mask(*limit, *blimit, + p3, p2, p1, p0, q0, q1, q2, q3); + const int8_t flat = flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3); + const int8_t flat2 = flat_mask5(1, + s[-8 * p], s[-7 * p], s[-6 * p], s[-5 * p], p0, + q0, s[4 * p], s[5 * p], s[6 * p], s[7 * p]); + + filter16(mask, *thresh, flat, flat2, + s - 8 * p, s - 7 * p, s - 6 * p, s - 5 * p, + s - 4 * p, s - 3 * p, s - 2 * p, s - 1 * p, + s, s + 1 * p, s + 2 * p, s + 3 * p, + s + 4 * p, s + 5 * p, s + 6 * p, s + 7 * p); + ++s; + } +} + +void vpx_lpf_horizontal_edge_8_c(uint8_t *s, int p, const uint8_t *blimit, + const uint8_t *limit, const uint8_t *thresh) { + mb_lpf_horizontal_edge_w(s, p, blimit, limit, thresh, 1); +} + +void vpx_lpf_horizontal_edge_16_c(uint8_t *s, int p, const uint8_t *blimit, + const uint8_t *limit, const uint8_t *thresh) { + mb_lpf_horizontal_edge_w(s, p, blimit, limit, thresh, 2); +} + +static void mb_lpf_vertical_edge_w(uint8_t *s, int p, + const uint8_t *blimit, + const uint8_t *limit, + const uint8_t *thresh, + int count) { + int i; + + for (i = 0; i < count; ++i) { + const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1]; + const uint8_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3]; + const int8_t mask = filter_mask(*limit, *blimit, + p3, p2, p1, p0, q0, q1, q2, q3); + const int8_t flat = flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3); + const int8_t flat2 = flat_mask5(1, s[-8], s[-7], s[-6], s[-5], p0, + q0, s[4], s[5], s[6], s[7]); + + filter16(mask, *thresh, flat, flat2, + s - 8, s - 7, s - 6, s - 5, s - 4, s - 3, s - 2, s - 1, + s, s + 1, s + 2, s + 3, s + 4, s + 5, s + 6, s + 7); + s += p; + } +} + +void vpx_lpf_vertical_16_c(uint8_t *s, int p, const uint8_t *blimit, + const uint8_t *limit, const uint8_t *thresh) { + mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 8); +} + +void vpx_lpf_vertical_16_dual_c(uint8_t *s, int p, const uint8_t *blimit, + const uint8_t *limit, const uint8_t *thresh) { + mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 16); +} + +#if CONFIG_VP9_HIGHBITDEPTH +// Should we apply any filter at all: 11111111 yes, 00000000 no ? +static INLINE int8_t highbd_filter_mask(uint8_t limit, uint8_t blimit, + uint16_t p3, uint16_t p2, + uint16_t p1, uint16_t p0, + uint16_t q0, uint16_t q1, + uint16_t q2, uint16_t q3, int bd) { + int8_t mask = 0; + int16_t limit16 = (uint16_t)limit << (bd - 8); + int16_t blimit16 = (uint16_t)blimit << (bd - 8); + mask |= (abs(p3 - p2) > limit16) * -1; + mask |= (abs(p2 - p1) > limit16) * -1; + mask |= (abs(p1 - p0) > limit16) * -1; + mask |= (abs(q1 - q0) > limit16) * -1; + mask |= (abs(q2 - q1) > limit16) * -1; + mask |= (abs(q3 - q2) > limit16) * -1; + mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2 > blimit16) * -1; + return ~mask; +} + +static INLINE int8_t highbd_flat_mask4(uint8_t thresh, + uint16_t p3, uint16_t p2, + uint16_t p1, uint16_t p0, + uint16_t q0, uint16_t q1, + uint16_t q2, uint16_t q3, int bd) { + int8_t mask = 0; + int16_t thresh16 = (uint16_t)thresh << (bd - 8); + mask |= (abs(p1 - p0) > thresh16) * -1; + mask |= (abs(q1 - q0) > thresh16) * -1; + mask |= (abs(p2 - p0) > thresh16) * -1; + mask |= (abs(q2 - q0) > thresh16) * -1; + mask |= (abs(p3 - p0) > thresh16) * -1; + mask |= (abs(q3 - q0) > thresh16) * -1; + return ~mask; +} + +static INLINE int8_t highbd_flat_mask5(uint8_t thresh, + uint16_t p4, uint16_t p3, + uint16_t p2, uint16_t p1, + uint16_t p0, uint16_t q0, + uint16_t q1, uint16_t q2, + uint16_t q3, uint16_t q4, int bd) { + int8_t mask = ~highbd_flat_mask4(thresh, p3, p2, p1, p0, q0, q1, q2, q3, bd); + int16_t thresh16 = (uint16_t)thresh << (bd - 8); + mask |= (abs(p4 - p0) > thresh16) * -1; + mask |= (abs(q4 - q0) > thresh16) * -1; + return ~mask; +} + +// Is there high edge variance internal edge: +// 11111111_11111111 yes, 00000000_00000000 no ? +static INLINE int16_t highbd_hev_mask(uint8_t thresh, uint16_t p1, uint16_t p0, + uint16_t q0, uint16_t q1, int bd) { + int16_t hev = 0; + int16_t thresh16 = (uint16_t)thresh << (bd - 8); + hev |= (abs(p1 - p0) > thresh16) * -1; + hev |= (abs(q1 - q0) > thresh16) * -1; + return hev; +} + +static INLINE void highbd_filter4(int8_t mask, uint8_t thresh, uint16_t *op1, + uint16_t *op0, uint16_t *oq0, uint16_t *oq1, + int bd) { + int16_t filter1, filter2; + // ^0x80 equivalent to subtracting 0x80 from the values to turn them + // into -128 to +127 instead of 0 to 255. + int shift = bd - 8; + const int16_t ps1 = (int16_t)*op1 - (0x80 << shift); + const int16_t ps0 = (int16_t)*op0 - (0x80 << shift); + const int16_t qs0 = (int16_t)*oq0 - (0x80 << shift); + const int16_t qs1 = (int16_t)*oq1 - (0x80 << shift); + const uint16_t hev = highbd_hev_mask(thresh, *op1, *op0, *oq0, *oq1, bd); + + // Add outer taps if we have high edge variance. + int16_t filter = signed_char_clamp_high(ps1 - qs1, bd) & hev; + + // Inner taps. + filter = signed_char_clamp_high(filter + 3 * (qs0 - ps0), bd) & mask; + + // Save bottom 3 bits so that we round one side +4 and the other +3 + // if it equals 4 we'll set to adjust by -1 to account for the fact + // we'd round 3 the other way. + filter1 = signed_char_clamp_high(filter + 4, bd) >> 3; + filter2 = signed_char_clamp_high(filter + 3, bd) >> 3; + + *oq0 = signed_char_clamp_high(qs0 - filter1, bd) + (0x80 << shift); + *op0 = signed_char_clamp_high(ps0 + filter2, bd) + (0x80 << shift); + + // Outer tap adjustments. + filter = ROUND_POWER_OF_TWO(filter1, 1) & ~hev; + + *oq1 = signed_char_clamp_high(qs1 - filter, bd) + (0x80 << shift); + *op1 = signed_char_clamp_high(ps1 + filter, bd) + (0x80 << shift); +} + +void vpx_highbd_lpf_horizontal_4_c(uint16_t *s, int p /* pitch */, + const uint8_t *blimit, const uint8_t *limit, + const uint8_t *thresh, int bd) { + int i; + + // loop filter designed to work using chars so that we can make maximum use + // of 8 bit simd instructions. + for (i = 0; i < 8; ++i) { + const uint16_t p3 = s[-4 * p]; + const uint16_t p2 = s[-3 * p]; + const uint16_t p1 = s[-2 * p]; + const uint16_t p0 = s[-p]; + const uint16_t q0 = s[0 * p]; + const uint16_t q1 = s[1 * p]; + const uint16_t q2 = s[2 * p]; + const uint16_t q3 = s[3 * p]; + const int8_t mask = highbd_filter_mask(*limit, *blimit, + p3, p2, p1, p0, q0, q1, q2, q3, bd); + highbd_filter4(mask, *thresh, s - 2 * p, s - 1 * p, s, s + 1 * p, bd); + ++s; + } +} + +void vpx_highbd_lpf_horizontal_4_dual_c(uint16_t *s, int p, + const uint8_t *blimit0, + const uint8_t *limit0, + const uint8_t *thresh0, + const uint8_t *blimit1, + const uint8_t *limit1, + const uint8_t *thresh1, + int bd) { + vpx_highbd_lpf_horizontal_4_c(s, p, blimit0, limit0, thresh0, bd); + vpx_highbd_lpf_horizontal_4_c(s + 8, p, blimit1, limit1, thresh1, bd); +} + +void vpx_highbd_lpf_vertical_4_c(uint16_t *s, int pitch, const uint8_t *blimit, + const uint8_t *limit, const uint8_t *thresh, + int bd) { + int i; + + // loop filter designed to work using chars so that we can make maximum use + // of 8 bit simd instructions. + for (i = 0; i < 8; ++i) { + const uint16_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1]; + const uint16_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3]; + const int8_t mask = highbd_filter_mask(*limit, *blimit, + p3, p2, p1, p0, q0, q1, q2, q3, bd); + highbd_filter4(mask, *thresh, s - 2, s - 1, s, s + 1, bd); + s += pitch; + } +} + +void vpx_highbd_lpf_vertical_4_dual_c(uint16_t *s, int pitch, + const uint8_t *blimit0, + const uint8_t *limit0, + const uint8_t *thresh0, + const uint8_t *blimit1, + const uint8_t *limit1, + const uint8_t *thresh1, + int bd) { + vpx_highbd_lpf_vertical_4_c(s, pitch, blimit0, limit0, thresh0, bd); + vpx_highbd_lpf_vertical_4_c(s + 8 * pitch, pitch, blimit1, limit1, + thresh1, bd); +} + +static INLINE void highbd_filter8(int8_t mask, uint8_t thresh, uint8_t flat, + uint16_t *op3, uint16_t *op2, + uint16_t *op1, uint16_t *op0, + uint16_t *oq0, uint16_t *oq1, + uint16_t *oq2, uint16_t *oq3, int bd) { + if (flat && mask) { + const uint16_t p3 = *op3, p2 = *op2, p1 = *op1, p0 = *op0; + const uint16_t q0 = *oq0, q1 = *oq1, q2 = *oq2, q3 = *oq3; + + // 7-tap filter [1, 1, 1, 2, 1, 1, 1] + *op2 = ROUND_POWER_OF_TWO(p3 + p3 + p3 + 2 * p2 + p1 + p0 + q0, 3); + *op1 = ROUND_POWER_OF_TWO(p3 + p3 + p2 + 2 * p1 + p0 + q0 + q1, 3); + *op0 = ROUND_POWER_OF_TWO(p3 + p2 + p1 + 2 * p0 + q0 + q1 + q2, 3); + *oq0 = ROUND_POWER_OF_TWO(p2 + p1 + p0 + 2 * q0 + q1 + q2 + q3, 3); + *oq1 = ROUND_POWER_OF_TWO(p1 + p0 + q0 + 2 * q1 + q2 + q3 + q3, 3); + *oq2 = ROUND_POWER_OF_TWO(p0 + q0 + q1 + 2 * q2 + q3 + q3 + q3, 3); + } else { + highbd_filter4(mask, thresh, op1, op0, oq0, oq1, bd); + } +} + +void vpx_highbd_lpf_horizontal_8_c(uint16_t *s, int p, const uint8_t *blimit, + const uint8_t *limit, const uint8_t *thresh, + int bd) { + int i; + + // loop filter designed to work using chars so that we can make maximum use + // of 8 bit simd instructions. + for (i = 0; i < 8; ++i) { + const uint16_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p]; + const uint16_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p]; + + const int8_t mask = highbd_filter_mask(*limit, *blimit, + p3, p2, p1, p0, q0, q1, q2, q3, bd); + const int8_t flat = highbd_flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3, + bd); + highbd_filter8(mask, *thresh, flat, + s - 4 * p, s - 3 * p, s - 2 * p, s - 1 * p, + s, s + 1 * p, s + 2 * p, s + 3 * p, bd); + ++s; + } +} + +void vpx_highbd_lpf_horizontal_8_dual_c(uint16_t *s, int p, + const uint8_t *blimit0, + const uint8_t *limit0, + const uint8_t *thresh0, + const uint8_t *blimit1, + const uint8_t *limit1, + const uint8_t *thresh1, + int bd) { + vpx_highbd_lpf_horizontal_8_c(s, p, blimit0, limit0, thresh0, bd); + vpx_highbd_lpf_horizontal_8_c(s + 8, p, blimit1, limit1, thresh1, bd); +} + +void vpx_highbd_lpf_vertical_8_c(uint16_t *s, int pitch, const uint8_t *blimit, + const uint8_t *limit, const uint8_t *thresh, + int bd) { + int i; + + for (i = 0; i < 8; ++i) { + const uint16_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1]; + const uint16_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3]; + const int8_t mask = highbd_filter_mask(*limit, *blimit, + p3, p2, p1, p0, q0, q1, q2, q3, bd); + const int8_t flat = highbd_flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3, + bd); + highbd_filter8(mask, *thresh, flat, + s - 4, s - 3, s - 2, s - 1, + s, s + 1, s + 2, s + 3, + bd); + s += pitch; + } +} + +void vpx_highbd_lpf_vertical_8_dual_c(uint16_t *s, int pitch, + const uint8_t *blimit0, + const uint8_t *limit0, + const uint8_t *thresh0, + const uint8_t *blimit1, + const uint8_t *limit1, + const uint8_t *thresh1, + int bd) { + vpx_highbd_lpf_vertical_8_c(s, pitch, blimit0, limit0, thresh0, bd); + vpx_highbd_lpf_vertical_8_c(s + 8 * pitch, pitch, blimit1, limit1, + thresh1, bd); +} + +static INLINE void highbd_filter16(int8_t mask, uint8_t thresh, + uint8_t flat, uint8_t flat2, + uint16_t *op7, uint16_t *op6, + uint16_t *op5, uint16_t *op4, + uint16_t *op3, uint16_t *op2, + uint16_t *op1, uint16_t *op0, + uint16_t *oq0, uint16_t *oq1, + uint16_t *oq2, uint16_t *oq3, + uint16_t *oq4, uint16_t *oq5, + uint16_t *oq6, uint16_t *oq7, int bd) { + if (flat2 && flat && mask) { + const uint16_t p7 = *op7; + const uint16_t p6 = *op6; + const uint16_t p5 = *op5; + const uint16_t p4 = *op4; + const uint16_t p3 = *op3; + const uint16_t p2 = *op2; + const uint16_t p1 = *op1; + const uint16_t p0 = *op0; + const uint16_t q0 = *oq0; + const uint16_t q1 = *oq1; + const uint16_t q2 = *oq2; + const uint16_t q3 = *oq3; + const uint16_t q4 = *oq4; + const uint16_t q5 = *oq5; + const uint16_t q6 = *oq6; + const uint16_t q7 = *oq7; + + // 15-tap filter [1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1] + *op6 = ROUND_POWER_OF_TWO(p7 * 7 + p6 * 2 + p5 + p4 + p3 + p2 + p1 + p0 + + q0, 4); + *op5 = ROUND_POWER_OF_TWO(p7 * 6 + p6 + p5 * 2 + p4 + p3 + p2 + p1 + p0 + + q0 + q1, 4); + *op4 = ROUND_POWER_OF_TWO(p7 * 5 + p6 + p5 + p4 * 2 + p3 + p2 + p1 + p0 + + q0 + q1 + q2, 4); + *op3 = ROUND_POWER_OF_TWO(p7 * 4 + p6 + p5 + p4 + p3 * 2 + p2 + p1 + p0 + + q0 + q1 + q2 + q3, 4); + *op2 = ROUND_POWER_OF_TWO(p7 * 3 + p6 + p5 + p4 + p3 + p2 * 2 + p1 + p0 + + q0 + q1 + q2 + q3 + q4, 4); + *op1 = ROUND_POWER_OF_TWO(p7 * 2 + p6 + p5 + p4 + p3 + p2 + p1 * 2 + p0 + + q0 + q1 + q2 + q3 + q4 + q5, 4); + *op0 = ROUND_POWER_OF_TWO(p7 + p6 + p5 + p4 + p3 + p2 + p1 + p0 * 2 + + q0 + q1 + q2 + q3 + q4 + q5 + q6, 4); + *oq0 = ROUND_POWER_OF_TWO(p6 + p5 + p4 + p3 + p2 + p1 + p0 + + q0 * 2 + q1 + q2 + q3 + q4 + q5 + q6 + q7, 4); + *oq1 = ROUND_POWER_OF_TWO(p5 + p4 + p3 + p2 + p1 + p0 + + q0 + q1 * 2 + q2 + q3 + q4 + q5 + q6 + q7 * 2, 4); + *oq2 = ROUND_POWER_OF_TWO(p4 + p3 + p2 + p1 + p0 + + q0 + q1 + q2 * 2 + q3 + q4 + q5 + q6 + q7 * 3, 4); + *oq3 = ROUND_POWER_OF_TWO(p3 + p2 + p1 + p0 + + q0 + q1 + q2 + q3 * 2 + q4 + q5 + q6 + q7 * 4, 4); + *oq4 = ROUND_POWER_OF_TWO(p2 + p1 + p0 + + q0 + q1 + q2 + q3 + q4 * 2 + q5 + q6 + q7 * 5, 4); + *oq5 = ROUND_POWER_OF_TWO(p1 + p0 + + q0 + q1 + q2 + q3 + q4 + q5 * 2 + q6 + q7 * 6, 4); + *oq6 = ROUND_POWER_OF_TWO(p0 + + q0 + q1 + q2 + q3 + q4 + q5 + q6 * 2 + q7 * 7, 4); + } else { + highbd_filter8(mask, thresh, flat, op3, op2, op1, op0, oq0, oq1, oq2, oq3, + bd); + } +} + +static void highbd_mb_lpf_horizontal_edge_w(uint16_t *s, int p, + const uint8_t *blimit, + const uint8_t *limit, + const uint8_t *thresh, + int count, int bd) { + int i; + + // loop filter designed to work using chars so that we can make maximum use + // of 8 bit simd instructions. + for (i = 0; i < 8 * count; ++i) { + const uint16_t p3 = s[-4 * p]; + const uint16_t p2 = s[-3 * p]; + const uint16_t p1 = s[-2 * p]; + const uint16_t p0 = s[-p]; + const uint16_t q0 = s[0 * p]; + const uint16_t q1 = s[1 * p]; + const uint16_t q2 = s[2 * p]; + const uint16_t q3 = s[3 * p]; + const int8_t mask = highbd_filter_mask(*limit, *blimit, + p3, p2, p1, p0, q0, q1, q2, q3, bd); + const int8_t flat = highbd_flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3, + bd); + const int8_t flat2 = highbd_flat_mask5( + 1, s[-8 * p], s[-7 * p], s[-6 * p], s[-5 * p], p0, + q0, s[4 * p], s[5 * p], s[6 * p], s[7 * p], bd); + + highbd_filter16(mask, *thresh, flat, flat2, + s - 8 * p, s - 7 * p, s - 6 * p, s - 5 * p, + s - 4 * p, s - 3 * p, s - 2 * p, s - 1 * p, + s, s + 1 * p, s + 2 * p, s + 3 * p, + s + 4 * p, s + 5 * p, s + 6 * p, s + 7 * p, + bd); + ++s; + } +} + +void vpx_highbd_lpf_horizontal_edge_8_c(uint16_t *s, int p, + const uint8_t *blimit, + const uint8_t *limit, + const uint8_t *thresh, int bd) { + highbd_mb_lpf_horizontal_edge_w(s, p, blimit, limit, thresh, 1, bd); +} + +void vpx_highbd_lpf_horizontal_edge_16_c(uint16_t *s, int p, + const uint8_t *blimit, + const uint8_t *limit, + const uint8_t *thresh, int bd) { + highbd_mb_lpf_horizontal_edge_w(s, p, blimit, limit, thresh, 2, bd); +} + +static void highbd_mb_lpf_vertical_edge_w(uint16_t *s, int p, + const uint8_t *blimit, + const uint8_t *limit, + const uint8_t *thresh, + int count, int bd) { + int i; + + for (i = 0; i < count; ++i) { + const uint16_t p3 = s[-4]; + const uint16_t p2 = s[-3]; + const uint16_t p1 = s[-2]; + const uint16_t p0 = s[-1]; + const uint16_t q0 = s[0]; + const uint16_t q1 = s[1]; + const uint16_t q2 = s[2]; + const uint16_t q3 = s[3]; + const int8_t mask = highbd_filter_mask(*limit, *blimit, + p3, p2, p1, p0, q0, q1, q2, q3, bd); + const int8_t flat = highbd_flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3, + bd); + const int8_t flat2 = highbd_flat_mask5(1, s[-8], s[-7], s[-6], s[-5], p0, + q0, s[4], s[5], s[6], s[7], bd); + + highbd_filter16(mask, *thresh, flat, flat2, + s - 8, s - 7, s - 6, s - 5, s - 4, s - 3, s - 2, s - 1, + s, s + 1, s + 2, s + 3, s + 4, s + 5, s + 6, s + 7, + bd); + s += p; + } +} + +void vpx_highbd_lpf_vertical_16_c(uint16_t *s, int p, const uint8_t *blimit, + const uint8_t *limit, const uint8_t *thresh, + int bd) { + highbd_mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 8, bd); +} + +void vpx_highbd_lpf_vertical_16_dual_c(uint16_t *s, int p, + const uint8_t *blimit, + const uint8_t *limit, + const uint8_t *thresh, + int bd) { + highbd_mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 16, bd); +} +#endif // CONFIG_VP9_HIGHBITDEPTH diff --git a/thirdparty/libvpx/vpx_dsp/prob.c b/thirdparty/libvpx/vpx_dsp/prob.c new file mode 100644 index 0000000000..639d24dd2f --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/prob.c @@ -0,0 +1,53 @@ +/* + * Copyright (c) 2013 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "./prob.h" + +const uint8_t vpx_norm[256] = { + 0, 7, 6, 6, 5, 5, 5, 5, 4, 4, 4, 4, 4, 4, 4, 4, + 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, + 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, + 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 +}; + +static unsigned int tree_merge_probs_impl(unsigned int i, + const vpx_tree_index *tree, + const vpx_prob *pre_probs, + const unsigned int *counts, + vpx_prob *probs) { + const int l = tree[i]; + const unsigned int left_count = (l <= 0) + ? counts[-l] + : tree_merge_probs_impl(l, tree, pre_probs, counts, probs); + const int r = tree[i + 1]; + const unsigned int right_count = (r <= 0) + ? counts[-r] + : tree_merge_probs_impl(r, tree, pre_probs, counts, probs); + const unsigned int ct[2] = { left_count, right_count }; + probs[i >> 1] = mode_mv_merge_probs(pre_probs[i >> 1], ct); + return left_count + right_count; +} + +void vpx_tree_merge_probs(const vpx_tree_index *tree, const vpx_prob *pre_probs, + const unsigned int *counts, vpx_prob *probs) { + tree_merge_probs_impl(0, tree, pre_probs, counts, probs); +} diff --git a/thirdparty/libvpx/vpx_dsp/prob.h b/thirdparty/libvpx/vpx_dsp/prob.h new file mode 100644 index 0000000000..c3cb103ffb --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/prob.h @@ -0,0 +1,103 @@ +/* + * Copyright (c) 2013 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VPX_DSP_PROB_H_ +#define VPX_DSP_PROB_H_ + +#include "./vpx_config.h" +#include "./vpx_dsp_common.h" + +#include "vpx_ports/mem.h" + +#ifdef __cplusplus +extern "C" { +#endif + +typedef uint8_t vpx_prob; + +#define MAX_PROB 255 + +#define vpx_prob_half ((vpx_prob) 128) + +typedef int8_t vpx_tree_index; + +#define TREE_SIZE(leaf_count) (2 * (leaf_count) - 2) + +#define vpx_complement(x) (255 - x) + +#define MODE_MV_COUNT_SAT 20 + +/* We build coding trees compactly in arrays. + Each node of the tree is a pair of vpx_tree_indices. + Array index often references a corresponding probability table. + Index <= 0 means done encoding/decoding and value = -Index, + Index > 0 means need another bit, specification at index. + Nonnegative indices are always even; processing begins at node 0. */ + +typedef const vpx_tree_index vpx_tree[]; + +static INLINE vpx_prob clip_prob(int p) { + return (p > 255) ? 255 : (p < 1) ? 1 : p; +} + +static INLINE vpx_prob get_prob(int num, int den) { + return (den == 0) ? 128u : clip_prob(((int64_t)num * 256 + (den >> 1)) / den); +} + +static INLINE vpx_prob get_binary_prob(int n0, int n1) { + return get_prob(n0, n0 + n1); +} + +/* This function assumes prob1 and prob2 are already within [1,255] range. */ +static INLINE vpx_prob weighted_prob(int prob1, int prob2, int factor) { + return ROUND_POWER_OF_TWO(prob1 * (256 - factor) + prob2 * factor, 8); +} + +static INLINE vpx_prob merge_probs(vpx_prob pre_prob, + const unsigned int ct[2], + unsigned int count_sat, + unsigned int max_update_factor) { + const vpx_prob prob = get_binary_prob(ct[0], ct[1]); + const unsigned int count = VPXMIN(ct[0] + ct[1], count_sat); + const unsigned int factor = max_update_factor * count / count_sat; + return weighted_prob(pre_prob, prob, factor); +} + +// MODE_MV_MAX_UPDATE_FACTOR (128) * count / MODE_MV_COUNT_SAT; +static const int count_to_update_factor[MODE_MV_COUNT_SAT + 1] = { + 0, 6, 12, 19, 25, 32, 38, 44, 51, 57, 64, + 70, 76, 83, 89, 96, 102, 108, 115, 121, 128 +}; + +static INLINE vpx_prob mode_mv_merge_probs(vpx_prob pre_prob, + const unsigned int ct[2]) { + const unsigned int den = ct[0] + ct[1]; + if (den == 0) { + return pre_prob; + } else { + const unsigned int count = VPXMIN(den, MODE_MV_COUNT_SAT); + const unsigned int factor = count_to_update_factor[count]; + const vpx_prob prob = + clip_prob(((int64_t)(ct[0]) * 256 + (den >> 1)) / den); + return weighted_prob(pre_prob, prob, factor); + } +} + +void vpx_tree_merge_probs(const vpx_tree_index *tree, const vpx_prob *pre_probs, + const unsigned int *counts, vpx_prob *probs); + + +DECLARE_ALIGNED(16, extern const uint8_t, vpx_norm[256]); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VPX_DSP_PROB_H_ diff --git a/thirdparty/libvpx/vpx_dsp/txfm_common.h b/thirdparty/libvpx/vpx_dsp/txfm_common.h new file mode 100644 index 0000000000..442e6a57b5 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/txfm_common.h @@ -0,0 +1,66 @@ +/* + * Copyright (c) 2015 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VPX_DSP_TXFM_COMMON_H_ +#define VPX_DSP_TXFM_COMMON_H_ + +#include "vpx_dsp/vpx_dsp_common.h" + +// Constants and Macros used by all idct/dct functions +#define DCT_CONST_BITS 14 +#define DCT_CONST_ROUNDING (1 << (DCT_CONST_BITS - 1)) + +#define UNIT_QUANT_SHIFT 2 +#define UNIT_QUANT_FACTOR (1 << UNIT_QUANT_SHIFT) + +// Constants: +// for (int i = 1; i< 32; ++i) +// printf("static const int cospi_%d_64 = %.0f;\n", i, +// round(16384 * cos(i*M_PI/64))); +// Note: sin(k*Pi/64) = cos((32-k)*Pi/64) +static const tran_high_t cospi_1_64 = 16364; +static const tran_high_t cospi_2_64 = 16305; +static const tran_high_t cospi_3_64 = 16207; +static const tran_high_t cospi_4_64 = 16069; +static const tran_high_t cospi_5_64 = 15893; +static const tran_high_t cospi_6_64 = 15679; +static const tran_high_t cospi_7_64 = 15426; +static const tran_high_t cospi_8_64 = 15137; +static const tran_high_t cospi_9_64 = 14811; +static const tran_high_t cospi_10_64 = 14449; +static const tran_high_t cospi_11_64 = 14053; +static const tran_high_t cospi_12_64 = 13623; +static const tran_high_t cospi_13_64 = 13160; +static const tran_high_t cospi_14_64 = 12665; +static const tran_high_t cospi_15_64 = 12140; +static const tran_high_t cospi_16_64 = 11585; +static const tran_high_t cospi_17_64 = 11003; +static const tran_high_t cospi_18_64 = 10394; +static const tran_high_t cospi_19_64 = 9760; +static const tran_high_t cospi_20_64 = 9102; +static const tran_high_t cospi_21_64 = 8423; +static const tran_high_t cospi_22_64 = 7723; +static const tran_high_t cospi_23_64 = 7005; +static const tran_high_t cospi_24_64 = 6270; +static const tran_high_t cospi_25_64 = 5520; +static const tran_high_t cospi_26_64 = 4756; +static const tran_high_t cospi_27_64 = 3981; +static const tran_high_t cospi_28_64 = 3196; +static const tran_high_t cospi_29_64 = 2404; +static const tran_high_t cospi_30_64 = 1606; +static const tran_high_t cospi_31_64 = 804; + +// 16384 * sqrt(2) * sin(kPi/9) * 2 / 3 +static const tran_high_t sinpi_1_9 = 5283; +static const tran_high_t sinpi_2_9 = 9929; +static const tran_high_t sinpi_3_9 = 13377; +static const tran_high_t sinpi_4_9 = 15212; + +#endif // VPX_DSP_TXFM_COMMON_H_ diff --git a/thirdparty/libvpx/vpx_dsp/vpx_convolve.c b/thirdparty/libvpx/vpx_dsp/vpx_convolve.c new file mode 100644 index 0000000000..2d1c927cbe --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/vpx_convolve.c @@ -0,0 +1,612 @@ +/* + * Copyright (c) 2013 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <assert.h> +#include <string.h> + +#include "./vpx_config.h" +#include "./vpx_dsp_rtcd.h" +#include "vpx/vpx_integer.h" +#include "vpx_dsp/vpx_convolve.h" +#include "vpx_dsp/vpx_dsp_common.h" +#include "vpx_dsp/vpx_filter.h" +#include "vpx_ports/mem.h" + +static void convolve_horiz(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const InterpKernel *x_filters, + int x0_q4, int x_step_q4, int w, int h) { + int x, y; + src -= SUBPEL_TAPS / 2 - 1; + for (y = 0; y < h; ++y) { + int x_q4 = x0_q4; + for (x = 0; x < w; ++x) { + const uint8_t *const src_x = &src[x_q4 >> SUBPEL_BITS]; + const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK]; + int k, sum = 0; + for (k = 0; k < SUBPEL_TAPS; ++k) + sum += src_x[k] * x_filter[k]; + dst[x] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)); + x_q4 += x_step_q4; + } + src += src_stride; + dst += dst_stride; + } +} + +static void convolve_avg_horiz(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const InterpKernel *x_filters, + int x0_q4, int x_step_q4, int w, int h) { + int x, y; + src -= SUBPEL_TAPS / 2 - 1; + for (y = 0; y < h; ++y) { + int x_q4 = x0_q4; + for (x = 0; x < w; ++x) { + const uint8_t *const src_x = &src[x_q4 >> SUBPEL_BITS]; + const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK]; + int k, sum = 0; + for (k = 0; k < SUBPEL_TAPS; ++k) + sum += src_x[k] * x_filter[k]; + dst[x] = ROUND_POWER_OF_TWO(dst[x] + + clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)), 1); + x_q4 += x_step_q4; + } + src += src_stride; + dst += dst_stride; + } +} + +static void convolve_vert(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const InterpKernel *y_filters, + int y0_q4, int y_step_q4, int w, int h) { + int x, y; + src -= src_stride * (SUBPEL_TAPS / 2 - 1); + + for (x = 0; x < w; ++x) { + int y_q4 = y0_q4; + for (y = 0; y < h; ++y) { + const unsigned char *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride]; + const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK]; + int k, sum = 0; + for (k = 0; k < SUBPEL_TAPS; ++k) + sum += src_y[k * src_stride] * y_filter[k]; + dst[y * dst_stride] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)); + y_q4 += y_step_q4; + } + ++src; + ++dst; + } +} + +static void convolve_avg_vert(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const InterpKernel *y_filters, + int y0_q4, int y_step_q4, int w, int h) { + int x, y; + src -= src_stride * (SUBPEL_TAPS / 2 - 1); + + for (x = 0; x < w; ++x) { + int y_q4 = y0_q4; + for (y = 0; y < h; ++y) { + const unsigned char *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride]; + const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK]; + int k, sum = 0; + for (k = 0; k < SUBPEL_TAPS; ++k) + sum += src_y[k * src_stride] * y_filter[k]; + dst[y * dst_stride] = ROUND_POWER_OF_TWO(dst[y * dst_stride] + + clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)), 1); + y_q4 += y_step_q4; + } + ++src; + ++dst; + } +} + +static void convolve(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const InterpKernel *const x_filters, + int x0_q4, int x_step_q4, + const InterpKernel *const y_filters, + int y0_q4, int y_step_q4, + int w, int h) { + // Note: Fixed size intermediate buffer, temp, places limits on parameters. + // 2d filtering proceeds in 2 steps: + // (1) Interpolate horizontally into an intermediate buffer, temp. + // (2) Interpolate temp vertically to derive the sub-pixel result. + // Deriving the maximum number of rows in the temp buffer (135): + // --Smallest scaling factor is x1/2 ==> y_step_q4 = 32 (Normative). + // --Largest block size is 64x64 pixels. + // --64 rows in the downscaled frame span a distance of (64 - 1) * 32 in the + // original frame (in 1/16th pixel units). + // --Must round-up because block may be located at sub-pixel position. + // --Require an additional SUBPEL_TAPS rows for the 8-tap filter tails. + // --((64 - 1) * 32 + 15) >> 4 + 8 = 135. + uint8_t temp[135 * 64]; + int intermediate_height = + (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS; + + assert(w <= 64); + assert(h <= 64); + assert(y_step_q4 <= 32); + assert(x_step_q4 <= 32); + + convolve_horiz(src - src_stride * (SUBPEL_TAPS / 2 - 1), src_stride, temp, 64, + x_filters, x0_q4, x_step_q4, w, intermediate_height); + convolve_vert(temp + 64 * (SUBPEL_TAPS / 2 - 1), 64, dst, dst_stride, + y_filters, y0_q4, y_step_q4, w, h); +} + +static const InterpKernel *get_filter_base(const int16_t *filter) { + // NOTE: This assumes that the filter table is 256-byte aligned. + // TODO(agrange) Modify to make independent of table alignment. + return (const InterpKernel *)(((intptr_t)filter) & ~((intptr_t)0xFF)); +} + +static int get_filter_offset(const int16_t *f, const InterpKernel *base) { + return (int)((const InterpKernel *)(intptr_t)f - base); +} + +void vpx_convolve8_horiz_c(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h) { + const InterpKernel *const filters_x = get_filter_base(filter_x); + const int x0_q4 = get_filter_offset(filter_x, filters_x); + + (void)filter_y; + (void)y_step_q4; + + convolve_horiz(src, src_stride, dst, dst_stride, filters_x, + x0_q4, x_step_q4, w, h); +} + +void vpx_convolve8_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h) { + const InterpKernel *const filters_x = get_filter_base(filter_x); + const int x0_q4 = get_filter_offset(filter_x, filters_x); + + (void)filter_y; + (void)y_step_q4; + + convolve_avg_horiz(src, src_stride, dst, dst_stride, filters_x, + x0_q4, x_step_q4, w, h); +} + +void vpx_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h) { + const InterpKernel *const filters_y = get_filter_base(filter_y); + const int y0_q4 = get_filter_offset(filter_y, filters_y); + + (void)filter_x; + (void)x_step_q4; + + convolve_vert(src, src_stride, dst, dst_stride, filters_y, + y0_q4, y_step_q4, w, h); +} + +void vpx_convolve8_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h) { + const InterpKernel *const filters_y = get_filter_base(filter_y); + const int y0_q4 = get_filter_offset(filter_y, filters_y); + + (void)filter_x; + (void)x_step_q4; + + convolve_avg_vert(src, src_stride, dst, dst_stride, filters_y, + y0_q4, y_step_q4, w, h); +} + +void vpx_convolve8_c(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h) { + const InterpKernel *const filters_x = get_filter_base(filter_x); + const int x0_q4 = get_filter_offset(filter_x, filters_x); + + const InterpKernel *const filters_y = get_filter_base(filter_y); + const int y0_q4 = get_filter_offset(filter_y, filters_y); + + convolve(src, src_stride, dst, dst_stride, + filters_x, x0_q4, x_step_q4, + filters_y, y0_q4, y_step_q4, w, h); +} + +void vpx_convolve8_avg_c(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h) { + /* Fixed size intermediate buffer places limits on parameters. */ + DECLARE_ALIGNED(16, uint8_t, temp[64 * 64]); + assert(w <= 64); + assert(h <= 64); + + vpx_convolve8_c(src, src_stride, temp, 64, + filter_x, x_step_q4, filter_y, y_step_q4, w, h); + vpx_convolve_avg_c(temp, 64, dst, dst_stride, NULL, 0, NULL, 0, w, h); +} + +void vpx_convolve_copy_c(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int filter_x_stride, + const int16_t *filter_y, int filter_y_stride, + int w, int h) { + int r; + + (void)filter_x; (void)filter_x_stride; + (void)filter_y; (void)filter_y_stride; + + for (r = h; r > 0; --r) { + memcpy(dst, src, w); + src += src_stride; + dst += dst_stride; + } +} + +void vpx_convolve_avg_c(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int filter_x_stride, + const int16_t *filter_y, int filter_y_stride, + int w, int h) { + int x, y; + + (void)filter_x; (void)filter_x_stride; + (void)filter_y; (void)filter_y_stride; + + for (y = 0; y < h; ++y) { + for (x = 0; x < w; ++x) + dst[x] = ROUND_POWER_OF_TWO(dst[x] + src[x], 1); + + src += src_stride; + dst += dst_stride; + } +} + +void vpx_scaled_horiz_c(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h) { + vpx_convolve8_horiz_c(src, src_stride, dst, dst_stride, filter_x, x_step_q4, + filter_y, y_step_q4, w, h); +} + +void vpx_scaled_vert_c(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h) { + vpx_convolve8_vert_c(src, src_stride, dst, dst_stride, filter_x, x_step_q4, + filter_y, y_step_q4, w, h); +} + +void vpx_scaled_2d_c(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h) { + vpx_convolve8_c(src, src_stride, dst, dst_stride, filter_x, x_step_q4, + filter_y, y_step_q4, w, h); +} + +void vpx_scaled_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h) { + vpx_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride, filter_x, + x_step_q4, filter_y, y_step_q4, w, h); +} + +void vpx_scaled_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h) { + vpx_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, filter_x, + x_step_q4, filter_y, y_step_q4, w, h); +} + +void vpx_scaled_avg_2d_c(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h) { + vpx_convolve8_avg_c(src, src_stride, dst, dst_stride, filter_x, x_step_q4, + filter_y, y_step_q4, w, h); +} + +#if CONFIG_VP9_HIGHBITDEPTH +static void highbd_convolve_horiz(const uint8_t *src8, ptrdiff_t src_stride, + uint8_t *dst8, ptrdiff_t dst_stride, + const InterpKernel *x_filters, + int x0_q4, int x_step_q4, + int w, int h, int bd) { + int x, y; + uint16_t *src = CONVERT_TO_SHORTPTR(src8); + uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); + src -= SUBPEL_TAPS / 2 - 1; + for (y = 0; y < h; ++y) { + int x_q4 = x0_q4; + for (x = 0; x < w; ++x) { + const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS]; + const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK]; + int k, sum = 0; + for (k = 0; k < SUBPEL_TAPS; ++k) + sum += src_x[k] * x_filter[k]; + dst[x] = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd); + x_q4 += x_step_q4; + } + src += src_stride; + dst += dst_stride; + } +} + +static void highbd_convolve_avg_horiz(const uint8_t *src8, ptrdiff_t src_stride, + uint8_t *dst8, ptrdiff_t dst_stride, + const InterpKernel *x_filters, + int x0_q4, int x_step_q4, + int w, int h, int bd) { + int x, y; + uint16_t *src = CONVERT_TO_SHORTPTR(src8); + uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); + src -= SUBPEL_TAPS / 2 - 1; + for (y = 0; y < h; ++y) { + int x_q4 = x0_q4; + for (x = 0; x < w; ++x) { + const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS]; + const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK]; + int k, sum = 0; + for (k = 0; k < SUBPEL_TAPS; ++k) + sum += src_x[k] * x_filter[k]; + dst[x] = ROUND_POWER_OF_TWO(dst[x] + + clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd), 1); + x_q4 += x_step_q4; + } + src += src_stride; + dst += dst_stride; + } +} + +static void highbd_convolve_vert(const uint8_t *src8, ptrdiff_t src_stride, + uint8_t *dst8, ptrdiff_t dst_stride, + const InterpKernel *y_filters, + int y0_q4, int y_step_q4, int w, int h, + int bd) { + int x, y; + uint16_t *src = CONVERT_TO_SHORTPTR(src8); + uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); + src -= src_stride * (SUBPEL_TAPS / 2 - 1); + for (x = 0; x < w; ++x) { + int y_q4 = y0_q4; + for (y = 0; y < h; ++y) { + const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride]; + const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK]; + int k, sum = 0; + for (k = 0; k < SUBPEL_TAPS; ++k) + sum += src_y[k * src_stride] * y_filter[k]; + dst[y * dst_stride] = clip_pixel_highbd( + ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd); + y_q4 += y_step_q4; + } + ++src; + ++dst; + } +} + +static void highbd_convolve_avg_vert(const uint8_t *src8, ptrdiff_t src_stride, + uint8_t *dst8, ptrdiff_t dst_stride, + const InterpKernel *y_filters, + int y0_q4, int y_step_q4, int w, int h, + int bd) { + int x, y; + uint16_t *src = CONVERT_TO_SHORTPTR(src8); + uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); + src -= src_stride * (SUBPEL_TAPS / 2 - 1); + for (x = 0; x < w; ++x) { + int y_q4 = y0_q4; + for (y = 0; y < h; ++y) { + const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride]; + const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK]; + int k, sum = 0; + for (k = 0; k < SUBPEL_TAPS; ++k) + sum += src_y[k * src_stride] * y_filter[k]; + dst[y * dst_stride] = ROUND_POWER_OF_TWO(dst[y * dst_stride] + + clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd), 1); + y_q4 += y_step_q4; + } + ++src; + ++dst; + } +} + +static void highbd_convolve(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const InterpKernel *const x_filters, + int x0_q4, int x_step_q4, + const InterpKernel *const y_filters, + int y0_q4, int y_step_q4, + int w, int h, int bd) { + // Note: Fixed size intermediate buffer, temp, places limits on parameters. + // 2d filtering proceeds in 2 steps: + // (1) Interpolate horizontally into an intermediate buffer, temp. + // (2) Interpolate temp vertically to derive the sub-pixel result. + // Deriving the maximum number of rows in the temp buffer (135): + // --Smallest scaling factor is x1/2 ==> y_step_q4 = 32 (Normative). + // --Largest block size is 64x64 pixels. + // --64 rows in the downscaled frame span a distance of (64 - 1) * 32 in the + // original frame (in 1/16th pixel units). + // --Must round-up because block may be located at sub-pixel position. + // --Require an additional SUBPEL_TAPS rows for the 8-tap filter tails. + // --((64 - 1) * 32 + 15) >> 4 + 8 = 135. + uint16_t temp[64 * 135]; + int intermediate_height = + (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS; + + assert(w <= 64); + assert(h <= 64); + assert(y_step_q4 <= 32); + assert(x_step_q4 <= 32); + + highbd_convolve_horiz(src - src_stride * (SUBPEL_TAPS / 2 - 1), + src_stride, CONVERT_TO_BYTEPTR(temp), 64, + x_filters, x0_q4, x_step_q4, w, + intermediate_height, bd); + highbd_convolve_vert(CONVERT_TO_BYTEPTR(temp) + 64 * (SUBPEL_TAPS / 2 - 1), + 64, dst, dst_stride, y_filters, y0_q4, y_step_q4, + w, h, bd); +} + + +void vpx_highbd_convolve8_horiz_c(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h, int bd) { + const InterpKernel *const filters_x = get_filter_base(filter_x); + const int x0_q4 = get_filter_offset(filter_x, filters_x); + (void)filter_y; + (void)y_step_q4; + + highbd_convolve_horiz(src, src_stride, dst, dst_stride, filters_x, + x0_q4, x_step_q4, w, h, bd); +} + +void vpx_highbd_convolve8_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h, int bd) { + const InterpKernel *const filters_x = get_filter_base(filter_x); + const int x0_q4 = get_filter_offset(filter_x, filters_x); + (void)filter_y; + (void)y_step_q4; + + highbd_convolve_avg_horiz(src, src_stride, dst, dst_stride, filters_x, + x0_q4, x_step_q4, w, h, bd); +} + +void vpx_highbd_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h, int bd) { + const InterpKernel *const filters_y = get_filter_base(filter_y); + const int y0_q4 = get_filter_offset(filter_y, filters_y); + (void)filter_x; + (void)x_step_q4; + + highbd_convolve_vert(src, src_stride, dst, dst_stride, filters_y, + y0_q4, y_step_q4, w, h, bd); +} + +void vpx_highbd_convolve8_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h, int bd) { + const InterpKernel *const filters_y = get_filter_base(filter_y); + const int y0_q4 = get_filter_offset(filter_y, filters_y); + (void)filter_x; + (void)x_step_q4; + + highbd_convolve_avg_vert(src, src_stride, dst, dst_stride, filters_y, + y0_q4, y_step_q4, w, h, bd); +} + +void vpx_highbd_convolve8_c(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h, int bd) { + const InterpKernel *const filters_x = get_filter_base(filter_x); + const int x0_q4 = get_filter_offset(filter_x, filters_x); + + const InterpKernel *const filters_y = get_filter_base(filter_y); + const int y0_q4 = get_filter_offset(filter_y, filters_y); + + highbd_convolve(src, src_stride, dst, dst_stride, + filters_x, x0_q4, x_step_q4, + filters_y, y0_q4, y_step_q4, w, h, bd); +} + +void vpx_highbd_convolve8_avg_c(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h, int bd) { + // Fixed size intermediate buffer places limits on parameters. + DECLARE_ALIGNED(16, uint16_t, temp[64 * 64]); + assert(w <= 64); + assert(h <= 64); + + vpx_highbd_convolve8_c(src, src_stride, CONVERT_TO_BYTEPTR(temp), 64, + filter_x, x_step_q4, filter_y, y_step_q4, w, h, bd); + vpx_highbd_convolve_avg_c(CONVERT_TO_BYTEPTR(temp), 64, dst, dst_stride, + NULL, 0, NULL, 0, w, h, bd); +} + +void vpx_highbd_convolve_copy_c(const uint8_t *src8, ptrdiff_t src_stride, + uint8_t *dst8, ptrdiff_t dst_stride, + const int16_t *filter_x, int filter_x_stride, + const int16_t *filter_y, int filter_y_stride, + int w, int h, int bd) { + int r; + uint16_t *src = CONVERT_TO_SHORTPTR(src8); + uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); + (void)filter_x; + (void)filter_y; + (void)filter_x_stride; + (void)filter_y_stride; + (void)bd; + + for (r = h; r > 0; --r) { + memcpy(dst, src, w * sizeof(uint16_t)); + src += src_stride; + dst += dst_stride; + } +} + +void vpx_highbd_convolve_avg_c(const uint8_t *src8, ptrdiff_t src_stride, + uint8_t *dst8, ptrdiff_t dst_stride, + const int16_t *filter_x, int filter_x_stride, + const int16_t *filter_y, int filter_y_stride, + int w, int h, int bd) { + int x, y; + uint16_t *src = CONVERT_TO_SHORTPTR(src8); + uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); + (void)filter_x; + (void)filter_y; + (void)filter_x_stride; + (void)filter_y_stride; + (void)bd; + + for (y = 0; y < h; ++y) { + for (x = 0; x < w; ++x) { + dst[x] = ROUND_POWER_OF_TWO(dst[x] + src[x], 1); + } + src += src_stride; + dst += dst_stride; + } +} +#endif diff --git a/thirdparty/libvpx/vpx_dsp/vpx_convolve.h b/thirdparty/libvpx/vpx_dsp/vpx_convolve.h new file mode 100644 index 0000000000..9ed3f1750f --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/vpx_convolve.h @@ -0,0 +1,38 @@ +/* + * Copyright (c) 2013 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ +#ifndef VPX_DSP_VPX_CONVOLVE_H_ +#define VPX_DSP_VPX_CONVOLVE_H_ + +#include "./vpx_config.h" +#include "vpx/vpx_integer.h" + +#ifdef __cplusplus +extern "C" { +#endif + +typedef void (*convolve_fn_t)(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h); + +#if CONFIG_VP9_HIGHBITDEPTH +typedef void (*highbd_convolve_fn_t)(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h, int bd); +#endif + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VPX_DSP_VPX_CONVOLVE_H_ diff --git a/thirdparty/libvpx/vpx_dsp/vpx_dsp_common.h b/thirdparty/libvpx/vpx_dsp/vpx_dsp_common.h new file mode 100644 index 0000000000..a1d0a51ef5 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/vpx_dsp_common.h @@ -0,0 +1,69 @@ +/* + * Copyright (c) 2015 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VPX_DSP_VPX_DSP_COMMON_H_ +#define VPX_DSP_VPX_DSP_COMMON_H_ + +#include "./vpx_config.h" +#include "vpx/vpx_integer.h" +#include "vpx_ports/mem.h" + +#ifdef __cplusplus +extern "C" { +#endif + +#define VPXMIN(x, y) (((x) < (y)) ? (x) : (y)) +#define VPXMAX(x, y) (((x) > (y)) ? (x) : (y)) + +#if CONFIG_VP9_HIGHBITDEPTH +// Note: +// tran_low_t is the datatype used for final transform coefficients. +// tran_high_t is the datatype used for intermediate transform stages. +typedef int64_t tran_high_t; +typedef int32_t tran_low_t; +#else +// Note: +// tran_low_t is the datatype used for final transform coefficients. +// tran_high_t is the datatype used for intermediate transform stages. +typedef int32_t tran_high_t; +typedef int16_t tran_low_t; +#endif // CONFIG_VP9_HIGHBITDEPTH + +static INLINE uint8_t clip_pixel(int val) { + return (val > 255) ? 255 : (val < 0) ? 0 : val; +} + +static INLINE int clamp(int value, int low, int high) { + return value < low ? low : (value > high ? high : value); +} + +static INLINE double fclamp(double value, double low, double high) { + return value < low ? low : (value > high ? high : value); +} + +#if CONFIG_VP9_HIGHBITDEPTH +static INLINE uint16_t clip_pixel_highbd(int val, int bd) { + switch (bd) { + case 8: + default: + return (uint16_t)clamp(val, 0, 255); + case 10: + return (uint16_t)clamp(val, 0, 1023); + case 12: + return (uint16_t)clamp(val, 0, 4095); + } +} +#endif // CONFIG_VP9_HIGHBITDEPTH + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VPX_DSP_VPX_DSP_COMMON_H_ diff --git a/thirdparty/libvpx/vpx_dsp/vpx_dsp_rtcd.c b/thirdparty/libvpx/vpx_dsp/vpx_dsp_rtcd.c new file mode 100644 index 0000000000..5fe27b614b --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/vpx_dsp_rtcd.c @@ -0,0 +1,17 @@ +/* + * Copyright (c) 2015 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ +#include "./vpx_config.h" +#define RTCD_C +#include "./vpx_dsp_rtcd.h" +#include "vpx_ports/vpx_once.h" + +void vpx_dsp_rtcd() { + once(setup_rtcd_internal); +} diff --git a/thirdparty/libvpx/vpx_dsp/vpx_filter.h b/thirdparty/libvpx/vpx_dsp/vpx_filter.h new file mode 100644 index 0000000000..2617febf3b --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/vpx_filter.h @@ -0,0 +1,34 @@ +/* + * Copyright (c) 2015 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VPX_DSP_VPX_FILTER_H_ +#define VPX_DSP_VPX_FILTER_H_ + +#include "vpx/vpx_integer.h" + + +#ifdef __cplusplus +extern "C" { +#endif + +#define FILTER_BITS 7 + +#define SUBPEL_BITS 4 +#define SUBPEL_MASK ((1 << SUBPEL_BITS) - 1) +#define SUBPEL_SHIFTS (1 << SUBPEL_BITS) +#define SUBPEL_TAPS 8 + +typedef int16_t InterpKernel[SUBPEL_TAPS]; + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VPX_DSP_VPX_FILTER_H_ diff --git a/thirdparty/libvpx/vpx_dsp/x86/convolve.h b/thirdparty/libvpx/vpx_dsp/x86/convolve.h new file mode 100644 index 0000000000..7e43eb7c72 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/x86/convolve.h @@ -0,0 +1,274 @@ +/* + * Copyright (c) 2015 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ +#ifndef VPX_DSP_X86_CONVOLVE_H_ +#define VPX_DSP_X86_CONVOLVE_H_ + +#include <assert.h> + +#include "./vpx_config.h" +#include "vpx/vpx_integer.h" +#include "vpx_ports/mem.h" + +typedef void filter8_1dfunction ( + const uint8_t *src_ptr, + ptrdiff_t src_pitch, + uint8_t *output_ptr, + ptrdiff_t out_pitch, + uint32_t output_height, + const int16_t *filter +); + +#define FUN_CONV_1D(name, step_q4, filter, dir, src_start, avg, opt) \ + void vpx_convolve8_##name##_##opt(const uint8_t *src, ptrdiff_t src_stride, \ + uint8_t *dst, ptrdiff_t dst_stride, \ + const int16_t *filter_x, int x_step_q4, \ + const int16_t *filter_y, int y_step_q4, \ + int w, int h) { \ + assert(filter[3] != 128); \ + assert(step_q4 == 16); \ + if (filter[0] | filter[1] | filter[2]) { \ + while (w >= 16) { \ + vpx_filter_block1d16_##dir##8_##avg##opt(src_start, \ + src_stride, \ + dst, \ + dst_stride, \ + h, \ + filter); \ + src += 16; \ + dst += 16; \ + w -= 16; \ + } \ + if (w == 8) { \ + vpx_filter_block1d8_##dir##8_##avg##opt(src_start, \ + src_stride, \ + dst, \ + dst_stride, \ + h, \ + filter); \ + } else if (w == 4) { \ + vpx_filter_block1d4_##dir##8_##avg##opt(src_start, \ + src_stride, \ + dst, \ + dst_stride, \ + h, \ + filter); \ + } \ + } else { \ + while (w >= 16) { \ + vpx_filter_block1d16_##dir##2_##avg##opt(src, \ + src_stride, \ + dst, \ + dst_stride, \ + h, \ + filter); \ + src += 16; \ + dst += 16; \ + w -= 16; \ + } \ + if (w == 8) { \ + vpx_filter_block1d8_##dir##2_##avg##opt(src, \ + src_stride, \ + dst, \ + dst_stride, \ + h, \ + filter); \ + } else if (w == 4) { \ + vpx_filter_block1d4_##dir##2_##avg##opt(src, \ + src_stride, \ + dst, \ + dst_stride, \ + h, \ + filter); \ + } \ + } \ +} + +#define FUN_CONV_2D(avg, opt) \ +void vpx_convolve8_##avg##opt(const uint8_t *src, ptrdiff_t src_stride, \ + uint8_t *dst, ptrdiff_t dst_stride, \ + const int16_t *filter_x, int x_step_q4, \ + const int16_t *filter_y, int y_step_q4, \ + int w, int h) { \ + assert(filter_x[3] != 128); \ + assert(filter_y[3] != 128); \ + assert(w <= 64); \ + assert(h <= 64); \ + assert(x_step_q4 == 16); \ + assert(y_step_q4 == 16); \ + if (filter_x[0] | filter_x[1] | filter_x[2]) { \ + DECLARE_ALIGNED(16, uint8_t, fdata2[64 * 71]); \ + vpx_convolve8_horiz_##opt(src - 3 * src_stride, src_stride, fdata2, 64, \ + filter_x, x_step_q4, filter_y, y_step_q4, \ + w, h + 7); \ + vpx_convolve8_##avg##vert_##opt(fdata2 + 3 * 64, 64, dst, dst_stride, \ + filter_x, x_step_q4, filter_y, \ + y_step_q4, w, h); \ + } else { \ + DECLARE_ALIGNED(16, uint8_t, fdata2[64 * 65]); \ + vpx_convolve8_horiz_##opt(src, src_stride, fdata2, 64, \ + filter_x, x_step_q4, filter_y, y_step_q4, \ + w, h + 1); \ + vpx_convolve8_##avg##vert_##opt(fdata2, 64, dst, dst_stride, \ + filter_x, x_step_q4, filter_y, \ + y_step_q4, w, h); \ + } \ +} + +#if CONFIG_VP9_HIGHBITDEPTH + +typedef void highbd_filter8_1dfunction ( + const uint16_t *src_ptr, + const ptrdiff_t src_pitch, + uint16_t *output_ptr, + ptrdiff_t out_pitch, + unsigned int output_height, + const int16_t *filter, + int bd +); + +#define HIGH_FUN_CONV_1D(name, step_q4, filter, dir, src_start, avg, opt) \ + void vpx_highbd_convolve8_##name##_##opt(const uint8_t *src8, \ + ptrdiff_t src_stride, \ + uint8_t *dst8, \ + ptrdiff_t dst_stride, \ + const int16_t *filter_x, \ + int x_step_q4, \ + const int16_t *filter_y, \ + int y_step_q4, \ + int w, int h, int bd) { \ + if (step_q4 == 16 && filter[3] != 128) { \ + uint16_t *src = CONVERT_TO_SHORTPTR(src8); \ + uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); \ + if (filter[0] | filter[1] | filter[2]) { \ + while (w >= 16) { \ + vpx_highbd_filter_block1d16_##dir##8_##avg##opt(src_start, \ + src_stride, \ + dst, \ + dst_stride, \ + h, \ + filter, \ + bd); \ + src += 16; \ + dst += 16; \ + w -= 16; \ + } \ + while (w >= 8) { \ + vpx_highbd_filter_block1d8_##dir##8_##avg##opt(src_start, \ + src_stride, \ + dst, \ + dst_stride, \ + h, \ + filter, \ + bd); \ + src += 8; \ + dst += 8; \ + w -= 8; \ + } \ + while (w >= 4) { \ + vpx_highbd_filter_block1d4_##dir##8_##avg##opt(src_start, \ + src_stride, \ + dst, \ + dst_stride, \ + h, \ + filter, \ + bd); \ + src += 4; \ + dst += 4; \ + w -= 4; \ + } \ + } else { \ + while (w >= 16) { \ + vpx_highbd_filter_block1d16_##dir##2_##avg##opt(src, \ + src_stride, \ + dst, \ + dst_stride, \ + h, \ + filter, \ + bd); \ + src += 16; \ + dst += 16; \ + w -= 16; \ + } \ + while (w >= 8) { \ + vpx_highbd_filter_block1d8_##dir##2_##avg##opt(src, \ + src_stride, \ + dst, \ + dst_stride, \ + h, \ + filter, \ + bd); \ + src += 8; \ + dst += 8; \ + w -= 8; \ + } \ + while (w >= 4) { \ + vpx_highbd_filter_block1d4_##dir##2_##avg##opt(src, \ + src_stride, \ + dst, \ + dst_stride, \ + h, \ + filter, \ + bd); \ + src += 4; \ + dst += 4; \ + w -= 4; \ + } \ + } \ + } \ + if (w) { \ + vpx_highbd_convolve8_##name##_c(src8, src_stride, dst8, dst_stride, \ + filter_x, x_step_q4, filter_y, y_step_q4, \ + w, h, bd); \ + } \ +} + +#define HIGH_FUN_CONV_2D(avg, opt) \ +void vpx_highbd_convolve8_##avg##opt(const uint8_t *src, ptrdiff_t src_stride, \ + uint8_t *dst, ptrdiff_t dst_stride, \ + const int16_t *filter_x, int x_step_q4, \ + const int16_t *filter_y, int y_step_q4, \ + int w, int h, int bd) { \ + assert(w <= 64); \ + assert(h <= 64); \ + if (x_step_q4 == 16 && y_step_q4 == 16) { \ + if ((filter_x[0] | filter_x[1] | filter_x[2]) || filter_x[3] == 128) { \ + DECLARE_ALIGNED(16, uint16_t, fdata2[64 * 71]); \ + vpx_highbd_convolve8_horiz_##opt(src - 3 * src_stride, src_stride, \ + CONVERT_TO_BYTEPTR(fdata2), 64, \ + filter_x, x_step_q4, \ + filter_y, y_step_q4, \ + w, h + 7, bd); \ + vpx_highbd_convolve8_##avg##vert_##opt(CONVERT_TO_BYTEPTR(fdata2) + 192, \ + 64, dst, dst_stride, \ + filter_x, x_step_q4, \ + filter_y, y_step_q4, \ + w, h, bd); \ + } else { \ + DECLARE_ALIGNED(16, uint16_t, fdata2[64 * 65]); \ + vpx_highbd_convolve8_horiz_##opt(src, src_stride, \ + CONVERT_TO_BYTEPTR(fdata2), 64, \ + filter_x, x_step_q4, \ + filter_y, y_step_q4, \ + w, h + 1, bd); \ + vpx_highbd_convolve8_##avg##vert_##opt(CONVERT_TO_BYTEPTR(fdata2), 64, \ + dst, dst_stride, \ + filter_x, x_step_q4, \ + filter_y, y_step_q4, \ + w, h, bd); \ + } \ + } else { \ + vpx_highbd_convolve8_##avg##c(src, src_stride, dst, dst_stride, \ + filter_x, x_step_q4, filter_y, y_step_q4, w, \ + h, bd); \ + } \ +} +#endif // CONFIG_VP9_HIGHBITDEPTH + +#endif // VPX_DSP_X86_CONVOLVE_H_ diff --git a/thirdparty/libvpx/vpx_dsp/x86/intrapred_sse2.asm b/thirdparty/libvpx/vpx_dsp/x86/intrapred_sse2.asm new file mode 100644 index 0000000000..cd6a6ae982 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/x86/intrapred_sse2.asm @@ -0,0 +1,860 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + +%include "third_party/x86inc/x86inc.asm" + +SECTION_RODATA +pb_1: times 16 db 1 +pw_4: times 8 dw 4 +pw_8: times 8 dw 8 +pw_16: times 8 dw 16 +pw_32: times 8 dw 32 +dc_128: times 16 db 128 +pw2_4: times 8 dw 2 +pw2_8: times 8 dw 4 +pw2_16: times 8 dw 8 +pw2_32: times 8 dw 16 + +SECTION .text + +; ------------------------------------------ +; input: x, y, z, result +; +; trick from pascal +; (x+2y+z+2)>>2 can be calculated as: +; result = avg(x,z) +; result -= xor(x,z) & 1 +; result = avg(result,y) +; ------------------------------------------ +%macro X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 4 + pavgb %4, %1, %3 + pxor %3, %1 + pand %3, [GLOBAL(pb_1)] + psubb %4, %3 + pavgb %4, %2 +%endmacro + +INIT_XMM sse2 +cglobal d45_predictor_4x4, 3, 4, 4, dst, stride, above, goffset + GET_GOT goffsetq + + movq m0, [aboveq] + DEFINE_ARGS dst, stride, temp + psrldq m1, m0, 1 + psrldq m2, m0, 2 + X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m1, m2, m3 + + ; store 4 lines + movd [dstq ], m3 + psrlq m3, 8 + movd [dstq+strideq ], m3 + lea dstq, [dstq+strideq*2] + psrlq m3, 8 + movd [dstq ], m3 + psrlq m3, 8 + movd [dstq+strideq ], m3 + psrlq m0, 56 + movd tempq, m0 + mov [dstq+strideq+3], tempb + + RESTORE_GOT + RET + +INIT_XMM sse2 +cglobal d45_predictor_8x8, 3, 4, 4, dst, stride, above, goffset + GET_GOT goffsetq + + movu m1, [aboveq] + pslldq m0, m1, 1 + psrldq m2, m1, 1 + DEFINE_ARGS dst, stride, stride3 + lea stride3q, [strideq*3] + X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m1, m2, m3 + punpckhbw m0, m0 ; 7 7 + punpcklwd m0, m0 ; 7 7 7 7 + punpckldq m0, m0 ; 7 7 7 7 7 7 7 7 + punpcklqdq m3, m0 ; -1 0 1 2 3 4 5 6 7 7 7 7 7 7 7 7 + + ; store 4 lines + psrldq m3, 1 + movq [dstq ], m3 + psrldq m3, 1 + movq [dstq+strideq ], m3 + psrldq m3, 1 + movq [dstq+strideq*2], m3 + psrldq m3, 1 + movq [dstq+stride3q ], m3 + lea dstq, [dstq+strideq*4] + + ; store next 4 lines + psrldq m3, 1 + movq [dstq ], m3 + psrldq m3, 1 + movq [dstq+strideq ], m3 + psrldq m3, 1 + movq [dstq+strideq*2], m3 + psrldq m3, 1 + movq [dstq+stride3q ], m3 + + RESTORE_GOT + RET + +INIT_XMM sse2 +cglobal d207_predictor_4x4, 4, 4, 5, dst, stride, unused, left, goffset + GET_GOT goffsetq + + movd m0, [leftq] ; abcd [byte] + punpcklbw m4, m0, m0 ; aabb ccdd + punpcklwd m4, m4 ; aaaa bbbb cccc dddd + psrldq m4, 12 ; dddd + punpckldq m0, m4 ; abcd dddd + psrldq m1, m0, 1 ; bcdd + psrldq m2, m0, 2 ; cddd + + X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m1, m2, m3 ; a2bc b2cd c3d d + pavgb m1, m0 ; ab, bc, cd, d [byte] + + punpcklbw m1, m3 ; ab, a2bc, bc, b2cd, cd, c3d, d, d + movd [dstq ], m1 + psrlq m1, 16 ; bc, b2cd, cd, c3d, d, d + movd [dstq+strideq], m1 + + lea dstq, [dstq+strideq*2] + psrlq m1, 16 ; cd, c3d, d, d + movd [dstq ], m1 + movd [dstq+strideq], m4 ; d, d, d, d + RESTORE_GOT + RET + +INIT_XMM sse2 +cglobal dc_predictor_4x4, 4, 5, 3, dst, stride, above, left, goffset + GET_GOT goffsetq + + movd m2, [leftq] + movd m0, [aboveq] + pxor m1, m1 + punpckldq m0, m2 + psadbw m0, m1 + paddw m0, [GLOBAL(pw_4)] + psraw m0, 3 + pshuflw m0, m0, 0x0 + packuswb m0, m0 + movd [dstq ], m0 + movd [dstq+strideq], m0 + lea dstq, [dstq+strideq*2] + movd [dstq ], m0 + movd [dstq+strideq], m0 + + RESTORE_GOT + RET + +INIT_XMM sse2 +cglobal dc_left_predictor_4x4, 2, 5, 2, dst, stride, above, left, goffset + movifnidn leftq, leftmp + GET_GOT goffsetq + + pxor m1, m1 + movd m0, [leftq] + psadbw m0, m1 + paddw m0, [GLOBAL(pw2_4)] + psraw m0, 2 + pshuflw m0, m0, 0x0 + packuswb m0, m0 + movd [dstq ], m0 + movd [dstq+strideq], m0 + lea dstq, [dstq+strideq*2] + movd [dstq ], m0 + movd [dstq+strideq], m0 + + RESTORE_GOT + RET + +INIT_XMM sse2 +cglobal dc_top_predictor_4x4, 3, 5, 2, dst, stride, above, left, goffset + GET_GOT goffsetq + + pxor m1, m1 + movd m0, [aboveq] + psadbw m0, m1 + paddw m0, [GLOBAL(pw2_4)] + psraw m0, 2 + pshuflw m0, m0, 0x0 + packuswb m0, m0 + movd [dstq ], m0 + movd [dstq+strideq], m0 + lea dstq, [dstq+strideq*2] + movd [dstq ], m0 + movd [dstq+strideq], m0 + + RESTORE_GOT + RET + +INIT_XMM sse2 +cglobal dc_predictor_8x8, 4, 5, 3, dst, stride, above, left, goffset + GET_GOT goffsetq + + pxor m1, m1 + movq m0, [aboveq] + movq m2, [leftq] + DEFINE_ARGS dst, stride, stride3 + lea stride3q, [strideq*3] + psadbw m0, m1 + psadbw m2, m1 + paddw m0, m2 + paddw m0, [GLOBAL(pw_8)] + psraw m0, 4 + punpcklbw m0, m0 + pshuflw m0, m0, 0x0 + movq [dstq ], m0 + movq [dstq+strideq ], m0 + movq [dstq+strideq*2], m0 + movq [dstq+stride3q ], m0 + lea dstq, [dstq+strideq*4] + movq [dstq ], m0 + movq [dstq+strideq ], m0 + movq [dstq+strideq*2], m0 + movq [dstq+stride3q ], m0 + + RESTORE_GOT + RET + +INIT_XMM sse2 +cglobal dc_top_predictor_8x8, 3, 5, 2, dst, stride, above, left, goffset + GET_GOT goffsetq + + pxor m1, m1 + movq m0, [aboveq] + DEFINE_ARGS dst, stride, stride3 + lea stride3q, [strideq*3] + psadbw m0, m1 + paddw m0, [GLOBAL(pw2_8)] + psraw m0, 3 + punpcklbw m0, m0 + pshuflw m0, m0, 0x0 + movq [dstq ], m0 + movq [dstq+strideq ], m0 + movq [dstq+strideq*2], m0 + movq [dstq+stride3q ], m0 + lea dstq, [dstq+strideq*4] + movq [dstq ], m0 + movq [dstq+strideq ], m0 + movq [dstq+strideq*2], m0 + movq [dstq+stride3q ], m0 + + RESTORE_GOT + RET + +INIT_XMM sse2 +cglobal dc_left_predictor_8x8, 2, 5, 2, dst, stride, above, left, goffset + movifnidn leftq, leftmp + GET_GOT goffsetq + + pxor m1, m1 + movq m0, [leftq] + DEFINE_ARGS dst, stride, stride3 + lea stride3q, [strideq*3] + psadbw m0, m1 + paddw m0, [GLOBAL(pw2_8)] + psraw m0, 3 + punpcklbw m0, m0 + pshuflw m0, m0, 0x0 + movq [dstq ], m0 + movq [dstq+strideq ], m0 + movq [dstq+strideq*2], m0 + movq [dstq+stride3q ], m0 + lea dstq, [dstq+strideq*4] + movq [dstq ], m0 + movq [dstq+strideq ], m0 + movq [dstq+strideq*2], m0 + movq [dstq+stride3q ], m0 + + RESTORE_GOT + RET + +INIT_XMM sse2 +cglobal dc_128_predictor_4x4, 2, 5, 1, dst, stride, above, left, goffset + GET_GOT goffsetq + + DEFINE_ARGS dst, stride, stride3 + lea stride3q, [strideq*3] + movd m0, [GLOBAL(dc_128)] + movd [dstq ], m0 + movd [dstq+strideq ], m0 + movd [dstq+strideq*2], m0 + movd [dstq+stride3q ], m0 + RESTORE_GOT + RET + +INIT_XMM sse2 +cglobal dc_128_predictor_8x8, 2, 5, 1, dst, stride, above, left, goffset + GET_GOT goffsetq + + DEFINE_ARGS dst, stride, stride3 + lea stride3q, [strideq*3] + movq m0, [GLOBAL(dc_128)] + movq [dstq ], m0 + movq [dstq+strideq ], m0 + movq [dstq+strideq*2], m0 + movq [dstq+stride3q ], m0 + lea dstq, [dstq+strideq*4] + movq [dstq ], m0 + movq [dstq+strideq ], m0 + movq [dstq+strideq*2], m0 + movq [dstq+stride3q ], m0 + RESTORE_GOT + RET + +INIT_XMM sse2 +cglobal dc_predictor_16x16, 4, 5, 3, dst, stride, above, left, goffset + GET_GOT goffsetq + + pxor m1, m1 + mova m0, [aboveq] + mova m2, [leftq] + DEFINE_ARGS dst, stride, stride3, lines4 + lea stride3q, [strideq*3] + mov lines4d, 4 + psadbw m0, m1 + psadbw m2, m1 + paddw m0, m2 + movhlps m2, m0 + paddw m0, m2 + paddw m0, [GLOBAL(pw_16)] + psraw m0, 5 + pshuflw m0, m0, 0x0 + punpcklqdq m0, m0 + packuswb m0, m0 +.loop: + mova [dstq ], m0 + mova [dstq+strideq ], m0 + mova [dstq+strideq*2], m0 + mova [dstq+stride3q ], m0 + lea dstq, [dstq+strideq*4] + dec lines4d + jnz .loop + + RESTORE_GOT + REP_RET + + +INIT_XMM sse2 +cglobal dc_top_predictor_16x16, 4, 5, 3, dst, stride, above, left, goffset + GET_GOT goffsetq + + pxor m1, m1 + mova m0, [aboveq] + DEFINE_ARGS dst, stride, stride3, lines4 + lea stride3q, [strideq*3] + mov lines4d, 4 + psadbw m0, m1 + movhlps m2, m0 + paddw m0, m2 + paddw m0, [GLOBAL(pw2_16)] + psraw m0, 4 + pshuflw m0, m0, 0x0 + punpcklqdq m0, m0 + packuswb m0, m0 +.loop: + mova [dstq ], m0 + mova [dstq+strideq ], m0 + mova [dstq+strideq*2], m0 + mova [dstq+stride3q ], m0 + lea dstq, [dstq+strideq*4] + dec lines4d + jnz .loop + + RESTORE_GOT + REP_RET + +INIT_XMM sse2 +cglobal dc_left_predictor_16x16, 4, 5, 3, dst, stride, above, left, goffset + GET_GOT goffsetq + + pxor m1, m1 + mova m0, [leftq] + DEFINE_ARGS dst, stride, stride3, lines4 + lea stride3q, [strideq*3] + mov lines4d, 4 + psadbw m0, m1 + movhlps m2, m0 + paddw m0, m2 + paddw m0, [GLOBAL(pw2_16)] + psraw m0, 4 + pshuflw m0, m0, 0x0 + punpcklqdq m0, m0 + packuswb m0, m0 +.loop: + mova [dstq ], m0 + mova [dstq+strideq ], m0 + mova [dstq+strideq*2], m0 + mova [dstq+stride3q ], m0 + lea dstq, [dstq+strideq*4] + dec lines4d + jnz .loop + + RESTORE_GOT + REP_RET + +INIT_XMM sse2 +cglobal dc_128_predictor_16x16, 4, 5, 3, dst, stride, above, left, goffset + GET_GOT goffsetq + + DEFINE_ARGS dst, stride, stride3, lines4 + lea stride3q, [strideq*3] + mov lines4d, 4 + mova m0, [GLOBAL(dc_128)] +.loop: + mova [dstq ], m0 + mova [dstq+strideq ], m0 + mova [dstq+strideq*2], m0 + mova [dstq+stride3q ], m0 + lea dstq, [dstq+strideq*4] + dec lines4d + jnz .loop + RESTORE_GOT + RET + + +INIT_XMM sse2 +cglobal dc_predictor_32x32, 4, 5, 5, dst, stride, above, left, goffset + GET_GOT goffsetq + + pxor m1, m1 + mova m0, [aboveq] + mova m2, [aboveq+16] + mova m3, [leftq] + mova m4, [leftq+16] + DEFINE_ARGS dst, stride, stride3, lines4 + lea stride3q, [strideq*3] + mov lines4d, 8 + psadbw m0, m1 + psadbw m2, m1 + psadbw m3, m1 + psadbw m4, m1 + paddw m0, m2 + paddw m0, m3 + paddw m0, m4 + movhlps m2, m0 + paddw m0, m2 + paddw m0, [GLOBAL(pw_32)] + psraw m0, 6 + pshuflw m0, m0, 0x0 + punpcklqdq m0, m0 + packuswb m0, m0 +.loop: + mova [dstq ], m0 + mova [dstq +16], m0 + mova [dstq+strideq ], m0 + mova [dstq+strideq +16], m0 + mova [dstq+strideq*2 ], m0 + mova [dstq+strideq*2+16], m0 + mova [dstq+stride3q ], m0 + mova [dstq+stride3q +16], m0 + lea dstq, [dstq+strideq*4] + dec lines4d + jnz .loop + + RESTORE_GOT + REP_RET + +INIT_XMM sse2 +cglobal dc_top_predictor_32x32, 4, 5, 5, dst, stride, above, left, goffset + GET_GOT goffsetq + + pxor m1, m1 + mova m0, [aboveq] + mova m2, [aboveq+16] + DEFINE_ARGS dst, stride, stride3, lines4 + lea stride3q, [strideq*3] + mov lines4d, 8 + psadbw m0, m1 + psadbw m2, m1 + paddw m0, m2 + movhlps m2, m0 + paddw m0, m2 + paddw m0, [GLOBAL(pw2_32)] + psraw m0, 5 + pshuflw m0, m0, 0x0 + punpcklqdq m0, m0 + packuswb m0, m0 +.loop: + mova [dstq ], m0 + mova [dstq +16], m0 + mova [dstq+strideq ], m0 + mova [dstq+strideq +16], m0 + mova [dstq+strideq*2 ], m0 + mova [dstq+strideq*2+16], m0 + mova [dstq+stride3q ], m0 + mova [dstq+stride3q +16], m0 + lea dstq, [dstq+strideq*4] + dec lines4d + jnz .loop + + RESTORE_GOT + REP_RET + +INIT_XMM sse2 +cglobal dc_left_predictor_32x32, 4, 5, 5, dst, stride, above, left, goffset + GET_GOT goffsetq + + pxor m1, m1 + mova m0, [leftq] + mova m2, [leftq+16] + DEFINE_ARGS dst, stride, stride3, lines4 + lea stride3q, [strideq*3] + mov lines4d, 8 + psadbw m0, m1 + psadbw m2, m1 + paddw m0, m2 + movhlps m2, m0 + paddw m0, m2 + paddw m0, [GLOBAL(pw2_32)] + psraw m0, 5 + pshuflw m0, m0, 0x0 + punpcklqdq m0, m0 + packuswb m0, m0 +.loop: + mova [dstq ], m0 + mova [dstq +16], m0 + mova [dstq+strideq ], m0 + mova [dstq+strideq +16], m0 + mova [dstq+strideq*2 ], m0 + mova [dstq+strideq*2+16], m0 + mova [dstq+stride3q ], m0 + mova [dstq+stride3q +16], m0 + lea dstq, [dstq+strideq*4] + dec lines4d + jnz .loop + + RESTORE_GOT + REP_RET + +INIT_XMM sse2 +cglobal dc_128_predictor_32x32, 4, 5, 3, dst, stride, above, left, goffset + GET_GOT goffsetq + + DEFINE_ARGS dst, stride, stride3, lines4 + lea stride3q, [strideq*3] + mov lines4d, 8 + mova m0, [GLOBAL(dc_128)] +.loop: + mova [dstq ], m0 + mova [dstq +16], m0 + mova [dstq+strideq ], m0 + mova [dstq+strideq +16], m0 + mova [dstq+strideq*2 ], m0 + mova [dstq+strideq*2+16], m0 + mova [dstq+stride3q ], m0 + mova [dstq+stride3q +16], m0 + lea dstq, [dstq+strideq*4] + dec lines4d + jnz .loop + RESTORE_GOT + RET + +INIT_XMM sse2 +cglobal v_predictor_4x4, 3, 3, 1, dst, stride, above + movd m0, [aboveq] + movd [dstq ], m0 + movd [dstq+strideq], m0 + lea dstq, [dstq+strideq*2] + movd [dstq ], m0 + movd [dstq+strideq], m0 + RET + +INIT_XMM sse2 +cglobal v_predictor_8x8, 3, 3, 1, dst, stride, above + movq m0, [aboveq] + DEFINE_ARGS dst, stride, stride3 + lea stride3q, [strideq*3] + movq [dstq ], m0 + movq [dstq+strideq ], m0 + movq [dstq+strideq*2], m0 + movq [dstq+stride3q ], m0 + lea dstq, [dstq+strideq*4] + movq [dstq ], m0 + movq [dstq+strideq ], m0 + movq [dstq+strideq*2], m0 + movq [dstq+stride3q ], m0 + RET + +INIT_XMM sse2 +cglobal v_predictor_16x16, 3, 4, 1, dst, stride, above + mova m0, [aboveq] + DEFINE_ARGS dst, stride, stride3, nlines4 + lea stride3q, [strideq*3] + mov nlines4d, 4 +.loop: + mova [dstq ], m0 + mova [dstq+strideq ], m0 + mova [dstq+strideq*2], m0 + mova [dstq+stride3q ], m0 + lea dstq, [dstq+strideq*4] + dec nlines4d + jnz .loop + REP_RET + +INIT_XMM sse2 +cglobal v_predictor_32x32, 3, 4, 2, dst, stride, above + mova m0, [aboveq] + mova m1, [aboveq+16] + DEFINE_ARGS dst, stride, stride3, nlines4 + lea stride3q, [strideq*3] + mov nlines4d, 8 +.loop: + mova [dstq ], m0 + mova [dstq +16], m1 + mova [dstq+strideq ], m0 + mova [dstq+strideq +16], m1 + mova [dstq+strideq*2 ], m0 + mova [dstq+strideq*2+16], m1 + mova [dstq+stride3q ], m0 + mova [dstq+stride3q +16], m1 + lea dstq, [dstq+strideq*4] + dec nlines4d + jnz .loop + REP_RET + +INIT_XMM sse2 +cglobal h_predictor_4x4, 2, 4, 4, dst, stride, line, left + movifnidn leftq, leftmp + movd m0, [leftq] + punpcklbw m0, m0 + punpcklbw m0, m0 + pshufd m1, m0, 0x1 + movd [dstq ], m0 + movd [dstq+strideq], m1 + pshufd m2, m0, 0x2 + lea dstq, [dstq+strideq*2] + pshufd m3, m0, 0x3 + movd [dstq ], m2 + movd [dstq+strideq], m3 + RET + +INIT_XMM sse2 +cglobal h_predictor_8x8, 2, 5, 3, dst, stride, line, left + movifnidn leftq, leftmp + mov lineq, -2 + DEFINE_ARGS dst, stride, line, left, stride3 + lea stride3q, [strideq*3] + movq m0, [leftq ] + punpcklbw m0, m0 ; l1 l1 l2 l2 ... l8 l8 +.loop: + pshuflw m1, m0, 0x0 ; l1 l1 l1 l1 l1 l1 l1 l1 + pshuflw m2, m0, 0x55 ; l2 l2 l2 l2 l2 l2 l2 l2 + movq [dstq ], m1 + movq [dstq+strideq], m2 + pshuflw m1, m0, 0xaa + pshuflw m2, m0, 0xff + movq [dstq+strideq*2], m1 + movq [dstq+stride3q ], m2 + pshufd m0, m0, 0xe ; [63:0] l5 l5 l6 l6 l7 l7 l8 l8 + inc lineq + lea dstq, [dstq+strideq*4] + jnz .loop + REP_RET + +INIT_XMM sse2 +cglobal h_predictor_16x16, 2, 5, 3, dst, stride, line, left + movifnidn leftq, leftmp + mov lineq, -4 + DEFINE_ARGS dst, stride, line, left, stride3 + lea stride3q, [strideq*3] +.loop: + movd m0, [leftq] + punpcklbw m0, m0 + punpcklbw m0, m0 ; l1 to l4 each repeated 4 times + pshufd m1, m0, 0x0 ; l1 repeated 16 times + pshufd m2, m0, 0x55 ; l2 repeated 16 times + mova [dstq ], m1 + mova [dstq+strideq ], m2 + pshufd m1, m0, 0xaa + pshufd m2, m0, 0xff + mova [dstq+strideq*2], m1 + mova [dstq+stride3q ], m2 + inc lineq + lea leftq, [leftq+4 ] + lea dstq, [dstq+strideq*4] + jnz .loop + REP_RET + +INIT_XMM sse2 +cglobal h_predictor_32x32, 2, 5, 3, dst, stride, line, left + movifnidn leftq, leftmp + mov lineq, -8 + DEFINE_ARGS dst, stride, line, left, stride3 + lea stride3q, [strideq*3] +.loop: + movd m0, [leftq] + punpcklbw m0, m0 + punpcklbw m0, m0 ; l1 to l4 each repeated 4 times + pshufd m1, m0, 0x0 ; l1 repeated 16 times + pshufd m2, m0, 0x55 ; l2 repeated 16 times + mova [dstq ], m1 + mova [dstq+16 ], m1 + mova [dstq+strideq ], m2 + mova [dstq+strideq+16 ], m2 + pshufd m1, m0, 0xaa + pshufd m2, m0, 0xff + mova [dstq+strideq*2 ], m1 + mova [dstq+strideq*2+16], m1 + mova [dstq+stride3q ], m2 + mova [dstq+stride3q+16 ], m2 + inc lineq + lea leftq, [leftq+4 ] + lea dstq, [dstq+strideq*4] + jnz .loop + REP_RET + +INIT_XMM sse2 +cglobal tm_predictor_4x4, 4, 4, 5, dst, stride, above, left + pxor m1, m1 + movq m0, [aboveq-1]; [63:0] tl t1 t2 t3 t4 x x x + punpcklbw m0, m1 + pshuflw m2, m0, 0x0 ; [63:0] tl tl tl tl [word] + psrldq m0, 2 + psubw m0, m2 ; [63:0] t1-tl t2-tl t3-tl t4-tl [word] + movd m2, [leftq] + punpcklbw m2, m1 + pshuflw m4, m2, 0x0 ; [63:0] l1 l1 l1 l1 [word] + pshuflw m3, m2, 0x55 ; [63:0] l2 l2 l2 l2 [word] + paddw m4, m0 + paddw m3, m0 + packuswb m4, m4 + packuswb m3, m3 + movd [dstq ], m4 + movd [dstq+strideq], m3 + lea dstq, [dstq+strideq*2] + pshuflw m4, m2, 0xaa + pshuflw m3, m2, 0xff + paddw m4, m0 + paddw m3, m0 + packuswb m4, m4 + packuswb m3, m3 + movd [dstq ], m4 + movd [dstq+strideq], m3 + RET + +INIT_XMM sse2 +cglobal tm_predictor_8x8, 4, 4, 5, dst, stride, above, left + pxor m1, m1 + movd m2, [aboveq-1] + movq m0, [aboveq] + punpcklbw m2, m1 + punpcklbw m0, m1 ; t1 t2 t3 t4 t5 t6 t7 t8 [word] + pshuflw m2, m2, 0x0 ; [63:0] tl tl tl tl [word] + DEFINE_ARGS dst, stride, line, left + mov lineq, -4 + punpcklqdq m2, m2 ; tl tl tl tl tl tl tl tl [word] + psubw m0, m2 ; t1-tl t2-tl ... t8-tl [word] + movq m2, [leftq] + punpcklbw m2, m1 ; l1 l2 l3 l4 l5 l6 l7 l8 [word] +.loop + pshuflw m4, m2, 0x0 ; [63:0] l1 l1 l1 l1 [word] + pshuflw m3, m2, 0x55 ; [63:0] l2 l2 l2 l2 [word] + punpcklqdq m4, m4 ; l1 l1 l1 l1 l1 l1 l1 l1 [word] + punpcklqdq m3, m3 ; l2 l2 l2 l2 l2 l2 l2 l2 [word] + paddw m4, m0 + paddw m3, m0 + packuswb m4, m3 + movq [dstq ], m4 + movhps [dstq+strideq], m4 + lea dstq, [dstq+strideq*2] + psrldq m2, 4 + inc lineq + jnz .loop + REP_RET + +INIT_XMM sse2 +cglobal tm_predictor_16x16, 4, 5, 8, dst, stride, above, left + pxor m1, m1 + mova m2, [aboveq-16]; + mova m0, [aboveq] ; t1 t2 ... t16 [byte] + punpckhbw m2, m1 ; [127:112] tl [word] + punpckhbw m4, m0, m1 + punpcklbw m0, m1 ; m0:m4 t1 t2 ... t16 [word] + DEFINE_ARGS dst, stride, line, left, stride8 + mov lineq, -8 + pshufhw m2, m2, 0xff + mova m3, [leftq] ; l1 l2 ... l16 [byte] + punpckhqdq m2, m2 ; tl repeated 8 times [word] + psubw m0, m2 + psubw m4, m2 ; m0:m4 t1-tl t2-tl ... t16-tl [word] + punpckhbw m5, m3, m1 + punpcklbw m3, m1 ; m3:m5 l1 l2 ... l16 [word] + lea stride8q, [strideq*8] +.loop: + pshuflw m6, m3, 0x0 + pshuflw m7, m5, 0x0 + punpcklqdq m6, m6 ; l1 repeated 8 times [word] + punpcklqdq m7, m7 ; l8 repeated 8 times [word] + paddw m1, m6, m0 + paddw m6, m4 ; m1:m6 ti-tl+l1 [i=1,15] [word] + psrldq m5, 2 + packuswb m1, m6 + mova [dstq ], m1 + paddw m1, m7, m0 + paddw m7, m4 ; m1:m7 ti-tl+l8 [i=1,15] [word] + psrldq m3, 2 + packuswb m1, m7 + mova [dstq+stride8q], m1 + inc lineq + lea dstq, [dstq+strideq] + jnz .loop + REP_RET + +INIT_XMM sse2 +cglobal tm_predictor_32x32, 4, 4, 8, dst, stride, above, left + pxor m1, m1 + movd m2, [aboveq-1] + mova m0, [aboveq] + mova m4, [aboveq+16] + punpcklbw m2, m1 + punpckhbw m3, m0, m1 + punpckhbw m5, m4, m1 + punpcklbw m0, m1 + punpcklbw m4, m1 + pshuflw m2, m2, 0x0 + DEFINE_ARGS dst, stride, line, left + mov lineq, -16 + punpcklqdq m2, m2 + add leftq, 32 + psubw m0, m2 + psubw m3, m2 + psubw m4, m2 + psubw m5, m2 +.loop: + movd m2, [leftq+lineq*2] + pxor m1, m1 + punpcklbw m2, m1 + pshuflw m7, m2, 0x55 + pshuflw m2, m2, 0x0 + punpcklqdq m2, m2 + punpcklqdq m7, m7 + paddw m6, m2, m3 + paddw m1, m2, m0 + packuswb m1, m6 + mova [dstq ], m1 + paddw m6, m2, m5 + paddw m1, m2, m4 + packuswb m1, m6 + mova [dstq+16 ], m1 + paddw m6, m7, m3 + paddw m1, m7, m0 + packuswb m1, m6 + mova [dstq+strideq ], m1 + paddw m6, m7, m5 + paddw m1, m7, m4 + packuswb m1, m6 + mova [dstq+strideq+16], m1 + lea dstq, [dstq+strideq*2] + inc lineq + jnz .loop + REP_RET diff --git a/thirdparty/libvpx/vpx_dsp/x86/intrapred_ssse3.asm b/thirdparty/libvpx/vpx_dsp/x86/intrapred_ssse3.asm new file mode 100644 index 0000000000..5e0139fa8d --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/x86/intrapred_ssse3.asm @@ -0,0 +1,871 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + +%include "third_party/x86inc/x86inc.asm" + +SECTION_RODATA + +pb_1: times 16 db 1 +sh_b12345677: db 1, 2, 3, 4, 5, 6, 7, 7, 0, 0, 0, 0, 0, 0, 0, 0 +sh_b23456777: db 2, 3, 4, 5, 6, 7, 7, 7, 0, 0, 0, 0, 0, 0, 0, 0 +sh_b0123456777777777: db 0, 1, 2, 3, 4, 5, 6, 7, 7, 7, 7, 7, 7, 7, 7, 7 +sh_b1234567777777777: db 1, 2, 3, 4, 5, 6, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7 +sh_b2345677777777777: db 2, 3, 4, 5, 6, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7 +sh_b123456789abcdeff: db 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 15 +sh_b23456789abcdefff: db 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 15, 15 +sh_b32104567: db 3, 2, 1, 0, 4, 5, 6, 7, 0, 0, 0, 0, 0, 0, 0, 0 +sh_b8091a2b345: db 8, 0, 9, 1, 10, 2, 11, 3, 4, 5, 0, 0, 0, 0, 0, 0 +sh_b76543210: db 7, 6, 5, 4, 3, 2, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0 +sh_b65432108: db 6, 5, 4, 3, 2, 1, 0, 8, 0, 0, 0, 0, 0, 0, 0, 0 +sh_b54321089: db 5, 4, 3, 2, 1, 0, 8, 9, 0, 0, 0, 0, 0, 0, 0, 0 +sh_b89abcdef: db 8, 9, 10, 11, 12, 13, 14, 15, 0, 0, 0, 0, 0, 0, 0, 0 +sh_bfedcba9876543210: db 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0 + +SECTION .text + +INIT_XMM ssse3 +cglobal d45_predictor_16x16, 3, 6, 4, dst, stride, above, dst8, line, goffset + GET_GOT goffsetq + + mova m0, [aboveq] + DEFINE_ARGS dst, stride, stride3, dst8, line + lea stride3q, [strideq*3] + lea dst8q, [dstq+strideq*8] + mova m1, [GLOBAL(sh_b123456789abcdeff)] + pshufb m2, m0, [GLOBAL(sh_b23456789abcdefff)] + pavgb m3, m2, m0 + pxor m2, m0 + pshufb m0, m1 + pand m2, [GLOBAL(pb_1)] + psubb m3, m2 + pavgb m0, m3 + + ; first 4 lines and first half of 3rd 4 lines + mov lined, 2 +.loop: + mova [dstq ], m0 + movhps [dst8q ], m0 + pshufb m0, m1 + mova [dstq +strideq ], m0 + movhps [dst8q+strideq ], m0 + pshufb m0, m1 + mova [dstq +strideq*2 ], m0 + movhps [dst8q+strideq*2 ], m0 + pshufb m0, m1 + mova [dstq +stride3q ], m0 + movhps [dst8q+stride3q ], m0 + pshufb m0, m1 + lea dstq, [dstq +strideq*4] + lea dst8q, [dst8q+strideq*4] + dec lined + jnz .loop + + ; bottom-right 8x8 block + movhps [dstq +8], m0 + movhps [dstq+strideq +8], m0 + movhps [dstq+strideq*2+8], m0 + movhps [dstq+stride3q +8], m0 + lea dstq, [dstq+strideq*4] + movhps [dstq +8], m0 + movhps [dstq+strideq +8], m0 + movhps [dstq+strideq*2+8], m0 + movhps [dstq+stride3q +8], m0 + + RESTORE_GOT + RET + +INIT_XMM ssse3 +cglobal d45_predictor_32x32, 3, 6, 7, dst, stride, above, dst16, line, goffset + GET_GOT goffsetq + + mova m0, [aboveq] + mova m4, [aboveq+16] + DEFINE_ARGS dst, stride, stride3, dst16, line + lea stride3q, [strideq*3] + lea dst16q, [dstq +strideq*8] + lea dst16q, [dst16q+strideq*8] + mova m1, [GLOBAL(sh_b123456789abcdeff)] + pshufb m2, m4, [GLOBAL(sh_b23456789abcdefff)] + pavgb m3, m2, m4 + pxor m2, m4 + palignr m5, m4, m0, 1 + palignr m6, m4, m0, 2 + pshufb m4, m1 + pand m2, [GLOBAL(pb_1)] + psubb m3, m2 + pavgb m4, m3 + pavgb m3, m0, m6 + pxor m0, m6 + pand m0, [GLOBAL(pb_1)] + psubb m3, m0 + pavgb m5, m3 + + ; write 4x4 lines (and the first half of the second 4x4 lines) + mov lined, 4 +.loop: + mova [dstq ], m5 + mova [dstq +16], m4 + mova [dst16q ], m4 + palignr m3, m4, m5, 1 + pshufb m4, m1 + mova [dstq +strideq ], m3 + mova [dstq +strideq +16], m4 + mova [dst16q+strideq ], m4 + palignr m5, m4, m3, 1 + pshufb m4, m1 + mova [dstq +strideq*2 ], m5 + mova [dstq +strideq*2+16], m4 + mova [dst16q+strideq*2 ], m4 + palignr m3, m4, m5, 1 + pshufb m4, m1 + mova [dstq +stride3q ], m3 + mova [dstq +stride3q +16], m4 + mova [dst16q+stride3q ], m4 + palignr m5, m4, m3, 1 + pshufb m4, m1 + lea dstq, [dstq +strideq*4] + lea dst16q, [dst16q+strideq*4] + dec lined + jnz .loop + + ; write second half of second 4x4 lines + mova [dstq +16], m4 + mova [dstq +strideq +16], m4 + mova [dstq +strideq*2+16], m4 + mova [dstq +stride3q +16], m4 + lea dstq, [dstq +strideq*4] + mova [dstq +16], m4 + mova [dstq +strideq +16], m4 + mova [dstq +strideq*2+16], m4 + mova [dstq +stride3q +16], m4 + lea dstq, [dstq +strideq*4] + mova [dstq +16], m4 + mova [dstq +strideq +16], m4 + mova [dstq +strideq*2+16], m4 + mova [dstq +stride3q +16], m4 + lea dstq, [dstq +strideq*4] + mova [dstq +16], m4 + mova [dstq +strideq +16], m4 + mova [dstq +strideq*2+16], m4 + mova [dstq +stride3q +16], m4 + + RESTORE_GOT + RET + +; ------------------------------------------ +; input: x, y, z, result +; +; trick from pascal +; (x+2y+z+2)>>2 can be calculated as: +; result = avg(x,z) +; result -= xor(x,z) & 1 +; result = avg(result,y) +; ------------------------------------------ +%macro X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 4 + pavgb %4, %1, %3 + pxor %3, %1 + pand %3, [GLOBAL(pb_1)] + psubb %4, %3 + pavgb %4, %2 +%endmacro + +INIT_XMM ssse3 +cglobal d63_predictor_4x4, 3, 4, 5, dst, stride, above, goffset + GET_GOT goffsetq + + movq m3, [aboveq] + pshufb m1, m3, [GLOBAL(sh_b23456777)] + pshufb m2, m3, [GLOBAL(sh_b12345677)] + + X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m3, m2, m1, m4 + pavgb m3, m2 + + ; store 4 lines + movd [dstq ], m3 + movd [dstq+strideq], m4 + lea dstq, [dstq+strideq*2] + psrldq m3, 1 + psrldq m4, 1 + movd [dstq ], m3 + movd [dstq+strideq], m4 + RESTORE_GOT + RET + +INIT_XMM ssse3 +cglobal d63_predictor_8x8, 3, 4, 5, dst, stride, above, goffset + GET_GOT goffsetq + + movq m3, [aboveq] + DEFINE_ARGS dst, stride, stride3 + lea stride3q, [strideq*3] + pshufb m1, m3, [GLOBAL(sh_b2345677777777777)] + pshufb m0, m3, [GLOBAL(sh_b0123456777777777)] + pshufb m2, m3, [GLOBAL(sh_b1234567777777777)] + pshufb m3, [GLOBAL(sh_b0123456777777777)] + + X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m2, m1, m4 + pavgb m3, m2 + + ; store 4 lines + movq [dstq ], m3 + movq [dstq+strideq], m4 + psrldq m3, 1 + psrldq m4, 1 + movq [dstq+strideq*2], m3 + movq [dstq+stride3q ], m4 + lea dstq, [dstq+strideq*4] + psrldq m3, 1 + psrldq m4, 1 + + ; store 4 lines + movq [dstq ], m3 + movq [dstq+strideq], m4 + psrldq m3, 1 + psrldq m4, 1 + movq [dstq+strideq*2], m3 + movq [dstq+stride3q ], m4 + RESTORE_GOT + RET + +INIT_XMM ssse3 +cglobal d63_predictor_16x16, 3, 5, 5, dst, stride, above, line, goffset + GET_GOT goffsetq + + mova m0, [aboveq] + DEFINE_ARGS dst, stride, stride3, line + lea stride3q, [strideq*3] + mova m1, [GLOBAL(sh_b123456789abcdeff)] + pshufb m2, m0, [GLOBAL(sh_b23456789abcdefff)] + pshufb m3, m0, m1 + + X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m3, m2, m4 + pavgb m0, m3 + + mov lined, 4 +.loop: + mova [dstq ], m0 + mova [dstq+strideq ], m4 + pshufb m0, m1 + pshufb m4, m1 + mova [dstq+strideq*2], m0 + mova [dstq+stride3q ], m4 + pshufb m0, m1 + pshufb m4, m1 + lea dstq, [dstq+strideq*4] + dec lined + jnz .loop + RESTORE_GOT + REP_RET + +INIT_XMM ssse3 +cglobal d63_predictor_32x32, 3, 5, 8, dst, stride, above, line, goffset + GET_GOT goffsetq + + mova m0, [aboveq] + mova m7, [aboveq+16] + DEFINE_ARGS dst, stride, stride3, line + mova m1, [GLOBAL(sh_b123456789abcdeff)] + lea stride3q, [strideq*3] + pshufb m2, m7, [GLOBAL(sh_b23456789abcdefff)] + pshufb m3, m7, m1 + + X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m7, m3, m2, m4 + palignr m6, m7, m0, 1 + palignr m5, m7, m0, 2 + pavgb m7, m3 + + X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m6, m5, m2 + pavgb m0, m6 + + mov lined, 8 +.loop: + mova [dstq ], m0 + mova [dstq +16], m7 + mova [dstq+strideq ], m2 + mova [dstq+strideq +16], m4 + palignr m3, m7, m0, 1 + palignr m5, m4, m2, 1 + pshufb m7, m1 + pshufb m4, m1 + + mova [dstq+strideq*2 ], m3 + mova [dstq+strideq*2+16], m7 + mova [dstq+stride3q ], m5 + mova [dstq+stride3q +16], m4 + palignr m0, m7, m3, 1 + palignr m2, m4, m5, 1 + pshufb m7, m1 + pshufb m4, m1 + lea dstq, [dstq+strideq*4] + dec lined + jnz .loop + RESTORE_GOT + REP_RET + +INIT_XMM ssse3 +cglobal d153_predictor_4x4, 4, 5, 4, dst, stride, above, left, goffset + GET_GOT goffsetq + movd m0, [leftq] ; l1, l2, l3, l4 + movd m1, [aboveq-1] ; tl, t1, t2, t3 + punpckldq m0, m1 ; l1, l2, l3, l4, tl, t1, t2, t3 + pshufb m0, [GLOBAL(sh_b32104567)]; l4, l3, l2, l1, tl, t1, t2, t3 + psrldq m1, m0, 1 ; l3, l2, l1, tl, t1, t2, t3 + psrldq m2, m0, 2 ; l2, l1, tl, t1, t2, t3 + ; comments below are for a predictor like this + ; A1 B1 C1 D1 + ; A2 B2 A1 B1 + ; A3 B3 A2 B2 + ; A4 B4 A3 B3 + X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m1, m2, m3 ; 3-tap avg B4 B3 B2 B1 C1 D1 + pavgb m1, m0 ; 2-tap avg A4 A3 A2 A1 + + punpcklqdq m3, m1 ; B4 B3 B2 B1 C1 D1 x x A4 A3 A2 A1 .. + + DEFINE_ARGS dst, stride, stride3 + lea stride3q, [strideq*3] + pshufb m3, [GLOBAL(sh_b8091a2b345)] ; A4 B4 A3 B3 A2 B2 A1 B1 C1 D1 .. + movd [dstq+stride3q ], m3 + psrldq m3, 2 ; A3 B3 A2 B2 A1 B1 C1 D1 .. + movd [dstq+strideq*2], m3 + psrldq m3, 2 ; A2 B2 A1 B1 C1 D1 .. + movd [dstq+strideq ], m3 + psrldq m3, 2 ; A1 B1 C1 D1 .. + movd [dstq ], m3 + RESTORE_GOT + RET + +INIT_XMM ssse3 +cglobal d153_predictor_8x8, 4, 5, 8, dst, stride, above, left, goffset + GET_GOT goffsetq + movq m0, [leftq] ; [0- 7] l1-8 [byte] + movhps m0, [aboveq-1] ; [8-15] tl, t1-7 [byte] + pshufb m1, m0, [GLOBAL(sh_b76543210)] ; l8-1 [word] + pshufb m2, m0, [GLOBAL(sh_b65432108)] ; l7-1,tl [word] + pshufb m3, m0, [GLOBAL(sh_b54321089)] ; l6-1,tl,t1 [word] + pshufb m0, [GLOBAL(sh_b89abcdef)] ; tl,t1-7 [word] + psrldq m4, m0, 1 ; t1-7 [word] + psrldq m5, m0, 2 ; t2-7 [word] + ; comments below are for a predictor like this + ; A1 B1 C1 D1 E1 F1 G1 H1 + ; A2 B2 A1 B1 C1 D1 E1 F1 + ; A3 B3 A2 B2 A1 B1 C1 D1 + ; A4 B4 A3 B3 A2 B2 A1 B1 + ; A5 B5 A4 B4 A3 B3 A2 B2 + ; A6 B6 A5 B5 A4 B4 A3 B3 + ; A7 B7 A6 B6 A5 B5 A4 B4 + ; A8 B8 A7 B7 A6 B6 A5 B5 + pavgb m6, m1, m2 ; 2-tap avg A8-A1 + + X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m4, m5, m7 ; 3-tap avg C-H1 + + X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m1, m2, m3, m0 ; 3-tap avg B8-1 + + punpcklbw m6, m0 ; A-B8, A-B7 ... A-B2, A-B1 + + DEFINE_ARGS dst, stride, stride3 + lea stride3q, [strideq*3] + + movhps [dstq+stride3q], m6 ; A-B4, A-B3, A-B2, A-B1 + palignr m0, m7, m6, 10 ; A-B3, A-B2, A-B1, C-H1 + movq [dstq+strideq*2], m0 + psrldq m0, 2 ; A-B2, A-B1, C-H1 + movq [dstq+strideq ], m0 + psrldq m0, 2 ; A-H1 + movq [dstq ], m0 + lea dstq, [dstq+strideq*4] + movq [dstq+stride3q ], m6 ; A-B8, A-B7, A-B6, A-B5 + psrldq m6, 2 ; A-B7, A-B6, A-B5, A-B4 + movq [dstq+strideq*2], m6 + psrldq m6, 2 ; A-B6, A-B5, A-B4, A-B3 + movq [dstq+strideq ], m6 + psrldq m6, 2 ; A-B5, A-B4, A-B3, A-B2 + movq [dstq ], m6 + RESTORE_GOT + RET + +INIT_XMM ssse3 +cglobal d153_predictor_16x16, 4, 5, 8, dst, stride, above, left, goffset + GET_GOT goffsetq + mova m0, [leftq] + movu m7, [aboveq-1] + ; comments below are for a predictor like this + ; A1 B1 C1 D1 E1 F1 G1 H1 I1 J1 K1 L1 M1 N1 O1 P1 + ; A2 B2 A1 B1 C1 D1 E1 F1 G1 H1 I1 J1 K1 L1 M1 N1 + ; A3 B3 A2 B2 A1 B1 C1 D1 E1 F1 G1 H1 I1 J1 K1 L1 + ; A4 B4 A3 B3 A2 B2 A1 B1 C1 D1 E1 F1 G1 H1 I1 J1 + ; A5 B5 A4 B4 A3 B3 A2 B2 A1 B1 C1 D1 E1 F1 G1 H1 + ; A6 B6 A5 B5 A4 B4 A3 B3 A2 B2 A1 B1 C1 D1 E1 F1 + ; A7 B7 A6 B6 A5 B5 A4 B4 A3 B3 A2 B2 A1 B1 C1 D1 + ; A8 B8 A7 B7 A6 B6 A5 B5 A4 B4 A3 B3 A2 B2 A1 B1 + ; A9 B9 A8 B8 A7 B7 A6 B6 A5 B5 A4 B4 A3 B3 A2 B2 + ; Aa Ba A9 B9 A8 B8 A7 B7 A6 B6 A5 B5 A4 B4 A3 B3 + ; Ab Bb Aa Ba A9 B9 A8 B8 A7 B7 A6 B6 A5 B5 A4 B4 + ; Ac Bc Ab Bb Aa Ba A9 B9 A8 B8 A7 B7 A6 B6 A5 B5 + ; Ad Bd Ac Bc Ab Bb Aa Ba A9 B9 A8 B8 A7 B7 A6 B6 + ; Ae Be Ad Bd Ac Bc Ab Bb Aa Ba A9 B9 A8 B8 A7 B7 + ; Af Bf Ae Be Ad Bd Ac Bc Ab Bb Aa Ba A9 B9 A8 B8 + ; Ag Bg Af Bf Ae Be Ad Bd Ac Bc Ab Bb Aa Ba A9 B9 + pshufb m6, m7, [GLOBAL(sh_bfedcba9876543210)] + palignr m5, m0, m6, 15 + palignr m3, m0, m6, 14 + + X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m5, m3, m4 ; 3-tap avg B3-Bg + pshufb m1, m0, [GLOBAL(sh_b123456789abcdeff)] + pavgb m5, m0 ; A1 - Ag + + punpcklbw m0, m4, m5 ; A-B8 ... A-B1 + punpckhbw m4, m5 ; A-B9 ... A-Bg + + pshufb m3, m7, [GLOBAL(sh_b123456789abcdeff)] + pshufb m5, m7, [GLOBAL(sh_b23456789abcdefff)] + + X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m7, m3, m5, m1 ; 3-tap avg C1-P1 + + pshufb m6, m0, [GLOBAL(sh_bfedcba9876543210)] + DEFINE_ARGS dst, stride, stride3 + lea stride3q, [strideq*3] + palignr m2, m1, m6, 14 + mova [dstq ], m2 + palignr m2, m1, m6, 12 + mova [dstq+strideq ], m2 + palignr m2, m1, m6, 10 + mova [dstq+strideq*2], m2 + palignr m2, m1, m6, 8 + mova [dstq+stride3q ], m2 + lea dstq, [dstq+strideq*4] + palignr m2, m1, m6, 6 + mova [dstq ], m2 + palignr m2, m1, m6, 4 + mova [dstq+strideq ], m2 + palignr m2, m1, m6, 2 + mova [dstq+strideq*2], m2 + pshufb m4, [GLOBAL(sh_bfedcba9876543210)] + mova [dstq+stride3q ], m6 + lea dstq, [dstq+strideq*4] + + palignr m2, m6, m4, 14 + mova [dstq ], m2 + palignr m2, m6, m4, 12 + mova [dstq+strideq ], m2 + palignr m2, m6, m4, 10 + mova [dstq+strideq*2], m2 + palignr m2, m6, m4, 8 + mova [dstq+stride3q ], m2 + lea dstq, [dstq+strideq*4] + palignr m2, m6, m4, 6 + mova [dstq ], m2 + palignr m2, m6, m4, 4 + mova [dstq+strideq ], m2 + palignr m2, m6, m4, 2 + mova [dstq+strideq*2], m2 + mova [dstq+stride3q ], m4 + RESTORE_GOT + RET + +INIT_XMM ssse3 +cglobal d153_predictor_32x32, 4, 5, 8, dst, stride, above, left, goffset + GET_GOT goffsetq + mova m0, [leftq] + movu m7, [aboveq-1] + movu m1, [aboveq+15] + + pshufb m4, m1, [GLOBAL(sh_b123456789abcdeff)] + pshufb m6, m1, [GLOBAL(sh_b23456789abcdefff)] + + X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m1, m4, m6, m2 ; 3-tap avg above [high] + + palignr m3, m1, m7, 1 + palignr m5, m1, m7, 2 + + X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m7, m3, m5, m1 ; 3-tap avg above [low] + + pshufb m7, [GLOBAL(sh_bfedcba9876543210)] + palignr m5, m0, m7, 15 + palignr m3, m0, m7, 14 + + X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m5, m3, m4 ; 3-tap avg B3-Bg + pavgb m5, m0 ; A1 - Ag + punpcklbw m6, m4, m5 ; A-B8 ... A-B1 + punpckhbw m4, m5 ; A-B9 ... A-Bg + pshufb m6, [GLOBAL(sh_bfedcba9876543210)] + pshufb m4, [GLOBAL(sh_bfedcba9876543210)] + + DEFINE_ARGS dst, stride, stride3, left, line + lea stride3q, [strideq*3] + + palignr m5, m2, m1, 14 + palignr m7, m1, m6, 14 + mova [dstq ], m7 + mova [dstq+16 ], m5 + palignr m5, m2, m1, 12 + palignr m7, m1, m6, 12 + mova [dstq+strideq ], m7 + mova [dstq+strideq+16 ], m5 + palignr m5, m2, m1, 10 + palignr m7, m1, m6, 10 + mova [dstq+strideq*2 ], m7 + mova [dstq+strideq*2+16], m5 + palignr m5, m2, m1, 8 + palignr m7, m1, m6, 8 + mova [dstq+stride3q ], m7 + mova [dstq+stride3q+16 ], m5 + lea dstq, [dstq+strideq*4] + palignr m5, m2, m1, 6 + palignr m7, m1, m6, 6 + mova [dstq ], m7 + mova [dstq+16 ], m5 + palignr m5, m2, m1, 4 + palignr m7, m1, m6, 4 + mova [dstq+strideq ], m7 + mova [dstq+strideq+16 ], m5 + palignr m5, m2, m1, 2 + palignr m7, m1, m6, 2 + mova [dstq+strideq*2 ], m7 + mova [dstq+strideq*2+16], m5 + mova [dstq+stride3q ], m6 + mova [dstq+stride3q+16 ], m1 + lea dstq, [dstq+strideq*4] + + palignr m5, m1, m6, 14 + palignr m3, m6, m4, 14 + mova [dstq ], m3 + mova [dstq+16 ], m5 + palignr m5, m1, m6, 12 + palignr m3, m6, m4, 12 + mova [dstq+strideq ], m3 + mova [dstq+strideq+16 ], m5 + palignr m5, m1, m6, 10 + palignr m3, m6, m4, 10 + mova [dstq+strideq*2 ], m3 + mova [dstq+strideq*2+16], m5 + palignr m5, m1, m6, 8 + palignr m3, m6, m4, 8 + mova [dstq+stride3q ], m3 + mova [dstq+stride3q+16 ], m5 + lea dstq, [dstq+strideq*4] + palignr m5, m1, m6, 6 + palignr m3, m6, m4, 6 + mova [dstq ], m3 + mova [dstq+16 ], m5 + palignr m5, m1, m6, 4 + palignr m3, m6, m4, 4 + mova [dstq+strideq ], m3 + mova [dstq+strideq+16 ], m5 + palignr m5, m1, m6, 2 + palignr m3, m6, m4, 2 + mova [dstq+strideq*2 ], m3 + mova [dstq+strideq*2+16], m5 + mova [dstq+stride3q ], m4 + mova [dstq+stride3q+16 ], m6 + lea dstq, [dstq+strideq*4] + + mova m7, [leftq] + mova m3, [leftq+16] + palignr m5, m3, m7, 15 + palignr m0, m3, m7, 14 + + X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m3, m5, m0, m2 ; 3-tap avg Bh - + pavgb m5, m3 ; Ah - + punpcklbw m3, m2, m5 ; A-B8 ... A-B1 + punpckhbw m2, m5 ; A-B9 ... A-Bg + pshufb m3, [GLOBAL(sh_bfedcba9876543210)] + pshufb m2, [GLOBAL(sh_bfedcba9876543210)] + + palignr m7, m6, m4, 14 + palignr m0, m4, m3, 14 + mova [dstq ], m0 + mova [dstq+16 ], m7 + palignr m7, m6, m4, 12 + palignr m0, m4, m3, 12 + mova [dstq+strideq ], m0 + mova [dstq+strideq+16 ], m7 + palignr m7, m6, m4, 10 + palignr m0, m4, m3, 10 + mova [dstq+strideq*2 ], m0 + mova [dstq+strideq*2+16], m7 + palignr m7, m6, m4, 8 + palignr m0, m4, m3, 8 + mova [dstq+stride3q ], m0 + mova [dstq+stride3q+16 ], m7 + lea dstq, [dstq+strideq*4] + palignr m7, m6, m4, 6 + palignr m0, m4, m3, 6 + mova [dstq ], m0 + mova [dstq+16 ], m7 + palignr m7, m6, m4, 4 + palignr m0, m4, m3, 4 + mova [dstq+strideq ], m0 + mova [dstq+strideq+16 ], m7 + palignr m7, m6, m4, 2 + palignr m0, m4, m3, 2 + mova [dstq+strideq*2 ], m0 + mova [dstq+strideq*2+16], m7 + mova [dstq+stride3q ], m3 + mova [dstq+stride3q+16 ], m4 + lea dstq, [dstq+strideq*4] + + palignr m7, m4, m3, 14 + palignr m0, m3, m2, 14 + mova [dstq ], m0 + mova [dstq+16 ], m7 + palignr m7, m4, m3, 12 + palignr m0, m3, m2, 12 + mova [dstq+strideq ], m0 + mova [dstq+strideq+16 ], m7 + palignr m7, m4, m3, 10 + palignr m0, m3, m2, 10 + mova [dstq+strideq*2 ], m0 + mova [dstq+strideq*2+16], m7 + palignr m7, m4, m3, 8 + palignr m0, m3, m2, 8 + mova [dstq+stride3q ], m0 + mova [dstq+stride3q+16 ], m7 + lea dstq, [dstq+strideq*4] + palignr m7, m4, m3, 6 + palignr m0, m3, m2, 6 + mova [dstq ], m0 + mova [dstq+16 ], m7 + palignr m7, m4, m3, 4 + palignr m0, m3, m2, 4 + mova [dstq+strideq ], m0 + mova [dstq+strideq+16 ], m7 + palignr m7, m4, m3, 2 + palignr m0, m3, m2, 2 + mova [dstq+strideq*2 ], m0 + mova [dstq+strideq*2+16], m7 + mova [dstq+stride3q ], m2 + mova [dstq+stride3q+16 ], m3 + + RESTORE_GOT + RET + +INIT_XMM ssse3 +cglobal d207_predictor_8x8, 4, 5, 4, dst, stride, stride3, left, goffset + GET_GOT goffsetq + movq m3, [leftq] ; abcdefgh [byte] + lea stride3q, [strideq*3] + + pshufb m1, m3, [GLOBAL(sh_b2345677777777777)] + pshufb m0, m3, [GLOBAL(sh_b0123456777777777)] + pshufb m2, m3, [GLOBAL(sh_b1234567777777777)] + + X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m2, m1, m3 + pavgb m0, m2 + punpcklbw m0, m3 ; interleaved output + + movq [dstq ], m0 + psrldq m0, 2 + movq [dstq+strideq ], m0 + psrldq m0, 2 + movq [dstq+strideq*2], m0 + psrldq m0, 2 + movq [dstq+stride3q ], m0 + lea dstq, [dstq+strideq*4] + pshufhw m0, m0, q0000 ; de, d2ef, ef, e2fg, fg, f2gh, gh, g3h, 8xh + psrldq m0, 2 + movq [dstq ], m0 + psrldq m0, 2 + movq [dstq+strideq ], m0 + psrldq m0, 2 + movq [dstq+strideq*2], m0 + psrldq m0, 2 + movq [dstq+stride3q ], m0 + RESTORE_GOT + RET + +INIT_XMM ssse3 +cglobal d207_predictor_16x16, 4, 5, 5, dst, stride, stride3, left, goffset + GET_GOT goffsetq + lea stride3q, [strideq*3] + mova m0, [leftq] ; abcdefghijklmnop [byte] + pshufb m1, m0, [GLOBAL(sh_b123456789abcdeff)] ; bcdefghijklmnopp + pshufb m2, m0, [GLOBAL(sh_b23456789abcdefff)] + + X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m1, m2, m3 + pavgb m1, m0 ; ab, bc, cd .. no, op, pp [byte] + + punpckhbw m4, m1, m3 ; interleaved input + punpcklbw m1, m3 ; interleaved output + mova [dstq ], m1 + palignr m3, m4, m1, 2 + mova [dstq+strideq ], m3 + palignr m3, m4, m1, 4 + mova [dstq+strideq*2], m3 + palignr m3, m4, m1, 6 + mova [dstq+stride3q ], m3 + lea dstq, [dstq+strideq*4] + palignr m3, m4, m1, 8 + mova [dstq ], m3 + palignr m3, m4, m1, 10 + mova [dstq+strideq ], m3 + palignr m3, m4, m1, 12 + mova [dstq+strideq*2], m3 + palignr m3, m4, m1, 14 + mova [dstq+stride3q ], m3 + DEFINE_ARGS dst, stride, stride3, line + mov lined, 2 + mova m0, [GLOBAL(sh_b23456789abcdefff)] +.loop: + lea dstq, [dstq+strideq*4] + mova [dstq ], m4 + pshufb m4, m0 + mova [dstq+strideq ], m4 + pshufb m4, m0 + mova [dstq+strideq*2], m4 + pshufb m4, m0 + mova [dstq+stride3q ], m4 + pshufb m4, m0 + dec lined + jnz .loop + RESTORE_GOT + REP_RET + +INIT_XMM ssse3 +cglobal d207_predictor_32x32, 4, 5, 8, dst, stride, stride3, left, goffset + GET_GOT goffsetq + lea stride3q, [strideq*3] + mova m1, [leftq] ; 0-15 [byte] + mova m2, [leftq+16] ; 16-31 [byte] + pshufb m0, m2, [GLOBAL(sh_b23456789abcdefff)] + pshufb m4, m2, [GLOBAL(sh_b123456789abcdeff)] + + X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m2, m4, m0, m3 + palignr m6, m2, m1, 1 + palignr m5, m2, m1, 2 + pavgb m2, m4 ; high 16px even lines + + X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m1, m6, m5, m0 + pavgb m1, m6 ; low 16px even lines + + punpckhbw m6, m1, m0 ; interleaved output 2 + punpcklbw m1, m0 ; interleaved output 1 + + punpckhbw m7, m2, m3 ; interleaved output 4 + punpcklbw m2, m3 ; interleaved output 3 + + ; output 1st 8 lines (and half of 2nd 8 lines) + DEFINE_ARGS dst, stride, stride3, dst8 + lea dst8q, [dstq+strideq*8] + mova [dstq ], m1 + mova [dstq +16], m6 + mova [dst8q ], m6 + palignr m0, m6, m1, 2 + palignr m4, m2, m6, 2 + mova [dstq +strideq ], m0 + mova [dstq +strideq +16], m4 + mova [dst8q+strideq ], m4 + palignr m0, m6, m1, 4 + palignr m4, m2, m6, 4 + mova [dstq +strideq*2 ], m0 + mova [dstq +strideq*2+16], m4 + mova [dst8q+strideq*2 ], m4 + palignr m0, m6, m1, 6 + palignr m4, m2, m6, 6 + mova [dstq +stride3q ], m0 + mova [dstq +stride3q +16], m4 + mova [dst8q+stride3q ], m4 + lea dstq, [dstq +strideq*4] + lea dst8q, [dst8q+strideq*4] + palignr m0, m6, m1, 8 + palignr m4, m2, m6, 8 + mova [dstq ], m0 + mova [dstq +16], m4 + mova [dst8q ], m4 + palignr m0, m6, m1, 10 + palignr m4, m2, m6, 10 + mova [dstq +strideq ], m0 + mova [dstq +strideq +16], m4 + mova [dst8q+strideq ], m4 + palignr m0, m6, m1, 12 + palignr m4, m2, m6, 12 + mova [dstq +strideq*2 ], m0 + mova [dstq +strideq*2+16], m4 + mova [dst8q+strideq*2 ], m4 + palignr m0, m6, m1, 14 + palignr m4, m2, m6, 14 + mova [dstq +stride3q ], m0 + mova [dstq +stride3q +16], m4 + mova [dst8q+stride3q ], m4 + lea dstq, [dstq+strideq*4] + lea dst8q, [dst8q+strideq*4] + + ; output 2nd half of 2nd 8 lines and half of 3rd 8 lines + mova [dstq +16], m2 + mova [dst8q ], m2 + palignr m4, m7, m2, 2 + mova [dstq +strideq +16], m4 + mova [dst8q+strideq ], m4 + palignr m4, m7, m2, 4 + mova [dstq +strideq*2+16], m4 + mova [dst8q+strideq*2 ], m4 + palignr m4, m7, m2, 6 + mova [dstq +stride3q +16], m4 + mova [dst8q+stride3q ], m4 + lea dstq, [dstq+strideq*4] + lea dst8q, [dst8q+strideq*4] + palignr m4, m7, m2, 8 + mova [dstq +16], m4 + mova [dst8q ], m4 + palignr m4, m7, m2, 10 + mova [dstq +strideq +16], m4 + mova [dst8q+strideq ], m4 + palignr m4, m7, m2, 12 + mova [dstq +strideq*2+16], m4 + mova [dst8q+strideq*2 ], m4 + palignr m4, m7, m2, 14 + mova [dstq +stride3q +16], m4 + mova [dst8q+stride3q ], m4 + lea dstq, [dstq+strideq*4] + lea dst8q, [dst8q+strideq*4] + + ; output 2nd half of 3rd 8 lines and half of 4th 8 lines + mova m0, [GLOBAL(sh_b23456789abcdefff)] + mova [dstq +16], m7 + mova [dst8q ], m7 + pshufb m7, m0 + mova [dstq +strideq +16], m7 + mova [dst8q+strideq ], m7 + pshufb m7, m0 + mova [dstq +strideq*2+16], m7 + mova [dst8q+strideq*2 ], m7 + pshufb m7, m0 + mova [dstq +stride3q +16], m7 + mova [dst8q+stride3q ], m7 + pshufb m7, m0 + lea dstq, [dstq+strideq*4] + lea dst8q, [dst8q+strideq*4] + mova [dstq +16], m7 + mova [dst8q ], m7 + pshufb m7, m0 + mova [dstq +strideq +16], m7 + mova [dst8q+strideq ], m7 + pshufb m7, m0 + mova [dstq +strideq*2+16], m7 + mova [dst8q+strideq*2 ], m7 + pshufb m7, m0 + mova [dstq +stride3q +16], m7 + mova [dst8q+stride3q ], m7 + pshufb m7, m0 + lea dstq, [dstq+strideq*4] + + ; output last half of 4th 8 lines + mova [dstq +16], m7 + mova [dstq +strideq +16], m7 + mova [dstq +strideq*2+16], m7 + mova [dstq +stride3q +16], m7 + lea dstq, [dstq+strideq*4] + mova [dstq +16], m7 + mova [dstq +strideq +16], m7 + mova [dstq +strideq*2+16], m7 + mova [dstq +stride3q +16], m7 + + ; done! + RESTORE_GOT + RET diff --git a/thirdparty/libvpx/vpx_dsp/x86/inv_txfm_sse2.c b/thirdparty/libvpx/vpx_dsp/x86/inv_txfm_sse2.c new file mode 100644 index 0000000000..df5068c624 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/x86/inv_txfm_sse2.c @@ -0,0 +1,4046 @@ +/* + * Copyright (c) 2015 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "./vpx_dsp_rtcd.h" +#include "vpx_dsp/x86/inv_txfm_sse2.h" +#include "vpx_dsp/x86/txfm_common_sse2.h" + +#define RECON_AND_STORE4X4(dest, in_x) \ +{ \ + __m128i d0 = _mm_cvtsi32_si128(*(const int *)(dest)); \ + d0 = _mm_unpacklo_epi8(d0, zero); \ + d0 = _mm_add_epi16(in_x, d0); \ + d0 = _mm_packus_epi16(d0, d0); \ + *(int *)(dest) = _mm_cvtsi128_si32(d0); \ +} + +void vpx_idct4x4_16_add_sse2(const tran_low_t *input, uint8_t *dest, + int stride) { + const __m128i zero = _mm_setzero_si128(); + const __m128i eight = _mm_set1_epi16(8); + const __m128i cst = _mm_setr_epi16( + (int16_t)cospi_16_64, (int16_t)cospi_16_64, (int16_t)cospi_16_64, + (int16_t)-cospi_16_64, (int16_t)cospi_24_64, (int16_t)-cospi_8_64, + (int16_t)cospi_8_64, (int16_t)cospi_24_64); + const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING); + __m128i input0, input1, input2, input3; + + // Rows + input0 = load_input_data(input); + input2 = load_input_data(input + 8); + + // Construct i3, i1, i3, i1, i2, i0, i2, i0 + input0 = _mm_shufflelo_epi16(input0, 0xd8); + input0 = _mm_shufflehi_epi16(input0, 0xd8); + input2 = _mm_shufflelo_epi16(input2, 0xd8); + input2 = _mm_shufflehi_epi16(input2, 0xd8); + + input1 = _mm_unpackhi_epi32(input0, input0); + input0 = _mm_unpacklo_epi32(input0, input0); + input3 = _mm_unpackhi_epi32(input2, input2); + input2 = _mm_unpacklo_epi32(input2, input2); + + // Stage 1 + input0 = _mm_madd_epi16(input0, cst); + input1 = _mm_madd_epi16(input1, cst); + input2 = _mm_madd_epi16(input2, cst); + input3 = _mm_madd_epi16(input3, cst); + + input0 = _mm_add_epi32(input0, rounding); + input1 = _mm_add_epi32(input1, rounding); + input2 = _mm_add_epi32(input2, rounding); + input3 = _mm_add_epi32(input3, rounding); + + input0 = _mm_srai_epi32(input0, DCT_CONST_BITS); + input1 = _mm_srai_epi32(input1, DCT_CONST_BITS); + input2 = _mm_srai_epi32(input2, DCT_CONST_BITS); + input3 = _mm_srai_epi32(input3, DCT_CONST_BITS); + + // Stage 2 + input0 = _mm_packs_epi32(input0, input1); + input1 = _mm_packs_epi32(input2, input3); + + // Transpose + input2 = _mm_unpacklo_epi16(input0, input1); + input3 = _mm_unpackhi_epi16(input0, input1); + input0 = _mm_unpacklo_epi32(input2, input3); + input1 = _mm_unpackhi_epi32(input2, input3); + + // Switch column2, column 3, and then, we got: + // input2: column1, column 0; input3: column2, column 3. + input1 = _mm_shuffle_epi32(input1, 0x4e); + input2 = _mm_add_epi16(input0, input1); + input3 = _mm_sub_epi16(input0, input1); + + // Columns + // Construct i3, i1, i3, i1, i2, i0, i2, i0 + input0 = _mm_unpacklo_epi32(input2, input2); + input1 = _mm_unpackhi_epi32(input2, input2); + input2 = _mm_unpackhi_epi32(input3, input3); + input3 = _mm_unpacklo_epi32(input3, input3); + + // Stage 1 + input0 = _mm_madd_epi16(input0, cst); + input1 = _mm_madd_epi16(input1, cst); + input2 = _mm_madd_epi16(input2, cst); + input3 = _mm_madd_epi16(input3, cst); + + input0 = _mm_add_epi32(input0, rounding); + input1 = _mm_add_epi32(input1, rounding); + input2 = _mm_add_epi32(input2, rounding); + input3 = _mm_add_epi32(input3, rounding); + + input0 = _mm_srai_epi32(input0, DCT_CONST_BITS); + input1 = _mm_srai_epi32(input1, DCT_CONST_BITS); + input2 = _mm_srai_epi32(input2, DCT_CONST_BITS); + input3 = _mm_srai_epi32(input3, DCT_CONST_BITS); + + // Stage 2 + input0 = _mm_packs_epi32(input0, input2); + input1 = _mm_packs_epi32(input1, input3); + + // Transpose + input2 = _mm_unpacklo_epi16(input0, input1); + input3 = _mm_unpackhi_epi16(input0, input1); + input0 = _mm_unpacklo_epi32(input2, input3); + input1 = _mm_unpackhi_epi32(input2, input3); + + // Switch column2, column 3, and then, we got: + // input2: column1, column 0; input3: column2, column 3. + input1 = _mm_shuffle_epi32(input1, 0x4e); + input2 = _mm_add_epi16(input0, input1); + input3 = _mm_sub_epi16(input0, input1); + + // Final round and shift + input2 = _mm_add_epi16(input2, eight); + input3 = _mm_add_epi16(input3, eight); + + input2 = _mm_srai_epi16(input2, 4); + input3 = _mm_srai_epi16(input3, 4); + + // Reconstruction and Store + { + __m128i d0 = _mm_cvtsi32_si128(*(const int *)(dest)); + __m128i d2 = _mm_cvtsi32_si128(*(const int *)(dest + stride * 2)); + d0 = _mm_unpacklo_epi32(d0, + _mm_cvtsi32_si128(*(const int *)(dest + stride))); + d2 = _mm_unpacklo_epi32( + _mm_cvtsi32_si128(*(const int *)(dest + stride * 3)), d2); + d0 = _mm_unpacklo_epi8(d0, zero); + d2 = _mm_unpacklo_epi8(d2, zero); + d0 = _mm_add_epi16(d0, input2); + d2 = _mm_add_epi16(d2, input3); + d0 = _mm_packus_epi16(d0, d2); + // store input0 + *(int *)dest = _mm_cvtsi128_si32(d0); + // store input1 + d0 = _mm_srli_si128(d0, 4); + *(int *)(dest + stride) = _mm_cvtsi128_si32(d0); + // store input2 + d0 = _mm_srli_si128(d0, 4); + *(int *)(dest + stride * 3) = _mm_cvtsi128_si32(d0); + // store input3 + d0 = _mm_srli_si128(d0, 4); + *(int *)(dest + stride * 2) = _mm_cvtsi128_si32(d0); + } +} + +void vpx_idct4x4_1_add_sse2(const tran_low_t *input, uint8_t *dest, + int stride) { + __m128i dc_value; + const __m128i zero = _mm_setzero_si128(); + int a; + + a = (int)dct_const_round_shift(input[0] * cospi_16_64); + a = (int)dct_const_round_shift(a * cospi_16_64); + a = ROUND_POWER_OF_TWO(a, 4); + + dc_value = _mm_set1_epi16(a); + + RECON_AND_STORE4X4(dest + 0 * stride, dc_value); + RECON_AND_STORE4X4(dest + 1 * stride, dc_value); + RECON_AND_STORE4X4(dest + 2 * stride, dc_value); + RECON_AND_STORE4X4(dest + 3 * stride, dc_value); +} + +static INLINE void transpose_4x4(__m128i *res) { + const __m128i tr0_0 = _mm_unpacklo_epi16(res[0], res[1]); + const __m128i tr0_1 = _mm_unpackhi_epi16(res[0], res[1]); + + res[0] = _mm_unpacklo_epi16(tr0_0, tr0_1); + res[1] = _mm_unpackhi_epi16(tr0_0, tr0_1); +} + +void idct4_sse2(__m128i *in) { + const __m128i k__cospi_p16_p16 = pair_set_epi16(cospi_16_64, cospi_16_64); + const __m128i k__cospi_p16_m16 = pair_set_epi16(cospi_16_64, -cospi_16_64); + const __m128i k__cospi_p24_m08 = pair_set_epi16(cospi_24_64, -cospi_8_64); + const __m128i k__cospi_p08_p24 = pair_set_epi16(cospi_8_64, cospi_24_64); + const __m128i k__DCT_CONST_ROUNDING = _mm_set1_epi32(DCT_CONST_ROUNDING); + __m128i u[8], v[8]; + + transpose_4x4(in); + // stage 1 + u[0] = _mm_unpacklo_epi16(in[0], in[1]); + u[1] = _mm_unpackhi_epi16(in[0], in[1]); + v[0] = _mm_madd_epi16(u[0], k__cospi_p16_p16); + v[1] = _mm_madd_epi16(u[0], k__cospi_p16_m16); + v[2] = _mm_madd_epi16(u[1], k__cospi_p24_m08); + v[3] = _mm_madd_epi16(u[1], k__cospi_p08_p24); + + u[0] = _mm_add_epi32(v[0], k__DCT_CONST_ROUNDING); + u[1] = _mm_add_epi32(v[1], k__DCT_CONST_ROUNDING); + u[2] = _mm_add_epi32(v[2], k__DCT_CONST_ROUNDING); + u[3] = _mm_add_epi32(v[3], k__DCT_CONST_ROUNDING); + + v[0] = _mm_srai_epi32(u[0], DCT_CONST_BITS); + v[1] = _mm_srai_epi32(u[1], DCT_CONST_BITS); + v[2] = _mm_srai_epi32(u[2], DCT_CONST_BITS); + v[3] = _mm_srai_epi32(u[3], DCT_CONST_BITS); + + u[0] = _mm_packs_epi32(v[0], v[1]); + u[1] = _mm_packs_epi32(v[3], v[2]); + + // stage 2 + in[0] = _mm_add_epi16(u[0], u[1]); + in[1] = _mm_sub_epi16(u[0], u[1]); + in[1] = _mm_shuffle_epi32(in[1], 0x4E); +} + +void iadst4_sse2(__m128i *in) { + const __m128i k__sinpi_p01_p04 = pair_set_epi16(sinpi_1_9, sinpi_4_9); + const __m128i k__sinpi_p03_p02 = pair_set_epi16(sinpi_3_9, sinpi_2_9); + const __m128i k__sinpi_p02_m01 = pair_set_epi16(sinpi_2_9, -sinpi_1_9); + const __m128i k__sinpi_p03_m04 = pair_set_epi16(sinpi_3_9, -sinpi_4_9); + const __m128i k__sinpi_p03_p03 = _mm_set1_epi16((int16_t)sinpi_3_9); + const __m128i kZero = _mm_set1_epi16(0); + const __m128i k__DCT_CONST_ROUNDING = _mm_set1_epi32(DCT_CONST_ROUNDING); + __m128i u[8], v[8], in7; + + transpose_4x4(in); + in7 = _mm_srli_si128(in[1], 8); + in7 = _mm_add_epi16(in7, in[0]); + in7 = _mm_sub_epi16(in7, in[1]); + + u[0] = _mm_unpacklo_epi16(in[0], in[1]); + u[1] = _mm_unpackhi_epi16(in[0], in[1]); + u[2] = _mm_unpacklo_epi16(in7, kZero); + u[3] = _mm_unpackhi_epi16(in[0], kZero); + + v[0] = _mm_madd_epi16(u[0], k__sinpi_p01_p04); // s0 + s3 + v[1] = _mm_madd_epi16(u[1], k__sinpi_p03_p02); // s2 + s5 + v[2] = _mm_madd_epi16(u[2], k__sinpi_p03_p03); // x2 + v[3] = _mm_madd_epi16(u[0], k__sinpi_p02_m01); // s1 - s4 + v[4] = _mm_madd_epi16(u[1], k__sinpi_p03_m04); // s2 - s6 + v[5] = _mm_madd_epi16(u[3], k__sinpi_p03_p03); // s2 + + u[0] = _mm_add_epi32(v[0], v[1]); + u[1] = _mm_add_epi32(v[3], v[4]); + u[2] = v[2]; + u[3] = _mm_add_epi32(u[0], u[1]); + u[4] = _mm_slli_epi32(v[5], 2); + u[5] = _mm_add_epi32(u[3], v[5]); + u[6] = _mm_sub_epi32(u[5], u[4]); + + v[0] = _mm_add_epi32(u[0], k__DCT_CONST_ROUNDING); + v[1] = _mm_add_epi32(u[1], k__DCT_CONST_ROUNDING); + v[2] = _mm_add_epi32(u[2], k__DCT_CONST_ROUNDING); + v[3] = _mm_add_epi32(u[6], k__DCT_CONST_ROUNDING); + + u[0] = _mm_srai_epi32(v[0], DCT_CONST_BITS); + u[1] = _mm_srai_epi32(v[1], DCT_CONST_BITS); + u[2] = _mm_srai_epi32(v[2], DCT_CONST_BITS); + u[3] = _mm_srai_epi32(v[3], DCT_CONST_BITS); + + in[0] = _mm_packs_epi32(u[0], u[1]); + in[1] = _mm_packs_epi32(u[2], u[3]); +} + +#define TRANSPOSE_8X8(in0, in1, in2, in3, in4, in5, in6, in7, \ + out0, out1, out2, out3, out4, out5, out6, out7) \ + { \ + const __m128i tr0_0 = _mm_unpacklo_epi16(in0, in1); \ + const __m128i tr0_1 = _mm_unpacklo_epi16(in2, in3); \ + const __m128i tr0_2 = _mm_unpackhi_epi16(in0, in1); \ + const __m128i tr0_3 = _mm_unpackhi_epi16(in2, in3); \ + const __m128i tr0_4 = _mm_unpacklo_epi16(in4, in5); \ + const __m128i tr0_5 = _mm_unpacklo_epi16(in6, in7); \ + const __m128i tr0_6 = _mm_unpackhi_epi16(in4, in5); \ + const __m128i tr0_7 = _mm_unpackhi_epi16(in6, in7); \ + \ + const __m128i tr1_0 = _mm_unpacklo_epi32(tr0_0, tr0_1); \ + const __m128i tr1_1 = _mm_unpacklo_epi32(tr0_2, tr0_3); \ + const __m128i tr1_2 = _mm_unpackhi_epi32(tr0_0, tr0_1); \ + const __m128i tr1_3 = _mm_unpackhi_epi32(tr0_2, tr0_3); \ + const __m128i tr1_4 = _mm_unpacklo_epi32(tr0_4, tr0_5); \ + const __m128i tr1_5 = _mm_unpacklo_epi32(tr0_6, tr0_7); \ + const __m128i tr1_6 = _mm_unpackhi_epi32(tr0_4, tr0_5); \ + const __m128i tr1_7 = _mm_unpackhi_epi32(tr0_6, tr0_7); \ + \ + out0 = _mm_unpacklo_epi64(tr1_0, tr1_4); \ + out1 = _mm_unpackhi_epi64(tr1_0, tr1_4); \ + out2 = _mm_unpacklo_epi64(tr1_2, tr1_6); \ + out3 = _mm_unpackhi_epi64(tr1_2, tr1_6); \ + out4 = _mm_unpacklo_epi64(tr1_1, tr1_5); \ + out5 = _mm_unpackhi_epi64(tr1_1, tr1_5); \ + out6 = _mm_unpacklo_epi64(tr1_3, tr1_7); \ + out7 = _mm_unpackhi_epi64(tr1_3, tr1_7); \ + } + +#define TRANSPOSE_4X8_10(tmp0, tmp1, tmp2, tmp3, \ + out0, out1, out2, out3) \ + { \ + const __m128i tr0_0 = _mm_unpackhi_epi16(tmp0, tmp1); \ + const __m128i tr0_1 = _mm_unpacklo_epi16(tmp1, tmp0); \ + const __m128i tr0_4 = _mm_unpacklo_epi16(tmp2, tmp3); \ + const __m128i tr0_5 = _mm_unpackhi_epi16(tmp3, tmp2); \ + \ + const __m128i tr1_0 = _mm_unpacklo_epi32(tr0_0, tr0_1); \ + const __m128i tr1_2 = _mm_unpackhi_epi32(tr0_0, tr0_1); \ + const __m128i tr1_4 = _mm_unpacklo_epi32(tr0_4, tr0_5); \ + const __m128i tr1_6 = _mm_unpackhi_epi32(tr0_4, tr0_5); \ + \ + out0 = _mm_unpacklo_epi64(tr1_0, tr1_4); \ + out1 = _mm_unpackhi_epi64(tr1_0, tr1_4); \ + out2 = _mm_unpacklo_epi64(tr1_2, tr1_6); \ + out3 = _mm_unpackhi_epi64(tr1_2, tr1_6); \ + } + +#define TRANSPOSE_8X8_10(in0, in1, in2, in3, out0, out1) \ + { \ + const __m128i tr0_0 = _mm_unpacklo_epi16(in0, in1); \ + const __m128i tr0_1 = _mm_unpacklo_epi16(in2, in3); \ + out0 = _mm_unpacklo_epi32(tr0_0, tr0_1); \ + out1 = _mm_unpackhi_epi32(tr0_0, tr0_1); \ + } + +// Define Macro for multiplying elements by constants and adding them together. +#define MULTIPLICATION_AND_ADD(lo_0, hi_0, lo_1, hi_1, \ + cst0, cst1, cst2, cst3, res0, res1, res2, res3) \ + { \ + tmp0 = _mm_madd_epi16(lo_0, cst0); \ + tmp1 = _mm_madd_epi16(hi_0, cst0); \ + tmp2 = _mm_madd_epi16(lo_0, cst1); \ + tmp3 = _mm_madd_epi16(hi_0, cst1); \ + tmp4 = _mm_madd_epi16(lo_1, cst2); \ + tmp5 = _mm_madd_epi16(hi_1, cst2); \ + tmp6 = _mm_madd_epi16(lo_1, cst3); \ + tmp7 = _mm_madd_epi16(hi_1, cst3); \ + \ + tmp0 = _mm_add_epi32(tmp0, rounding); \ + tmp1 = _mm_add_epi32(tmp1, rounding); \ + tmp2 = _mm_add_epi32(tmp2, rounding); \ + tmp3 = _mm_add_epi32(tmp3, rounding); \ + tmp4 = _mm_add_epi32(tmp4, rounding); \ + tmp5 = _mm_add_epi32(tmp5, rounding); \ + tmp6 = _mm_add_epi32(tmp6, rounding); \ + tmp7 = _mm_add_epi32(tmp7, rounding); \ + \ + tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS); \ + tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS); \ + tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS); \ + tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS); \ + tmp4 = _mm_srai_epi32(tmp4, DCT_CONST_BITS); \ + tmp5 = _mm_srai_epi32(tmp5, DCT_CONST_BITS); \ + tmp6 = _mm_srai_epi32(tmp6, DCT_CONST_BITS); \ + tmp7 = _mm_srai_epi32(tmp7, DCT_CONST_BITS); \ + \ + res0 = _mm_packs_epi32(tmp0, tmp1); \ + res1 = _mm_packs_epi32(tmp2, tmp3); \ + res2 = _mm_packs_epi32(tmp4, tmp5); \ + res3 = _mm_packs_epi32(tmp6, tmp7); \ + } + +#define MULTIPLICATION_AND_ADD_2(lo_0, hi_0, cst0, cst1, res0, res1) \ + { \ + tmp0 = _mm_madd_epi16(lo_0, cst0); \ + tmp1 = _mm_madd_epi16(hi_0, cst0); \ + tmp2 = _mm_madd_epi16(lo_0, cst1); \ + tmp3 = _mm_madd_epi16(hi_0, cst1); \ + \ + tmp0 = _mm_add_epi32(tmp0, rounding); \ + tmp1 = _mm_add_epi32(tmp1, rounding); \ + tmp2 = _mm_add_epi32(tmp2, rounding); \ + tmp3 = _mm_add_epi32(tmp3, rounding); \ + \ + tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS); \ + tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS); \ + tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS); \ + tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS); \ + \ + res0 = _mm_packs_epi32(tmp0, tmp1); \ + res1 = _mm_packs_epi32(tmp2, tmp3); \ + } + +#define IDCT8(in0, in1, in2, in3, in4, in5, in6, in7, \ + out0, out1, out2, out3, out4, out5, out6, out7) \ + { \ + /* Stage1 */ \ + { \ + const __m128i lo_17 = _mm_unpacklo_epi16(in1, in7); \ + const __m128i hi_17 = _mm_unpackhi_epi16(in1, in7); \ + const __m128i lo_35 = _mm_unpacklo_epi16(in3, in5); \ + const __m128i hi_35 = _mm_unpackhi_epi16(in3, in5); \ + \ + MULTIPLICATION_AND_ADD(lo_17, hi_17, lo_35, hi_35, stg1_0, \ + stg1_1, stg1_2, stg1_3, stp1_4, \ + stp1_7, stp1_5, stp1_6) \ + } \ + \ + /* Stage2 */ \ + { \ + const __m128i lo_04 = _mm_unpacklo_epi16(in0, in4); \ + const __m128i hi_04 = _mm_unpackhi_epi16(in0, in4); \ + const __m128i lo_26 = _mm_unpacklo_epi16(in2, in6); \ + const __m128i hi_26 = _mm_unpackhi_epi16(in2, in6); \ + \ + MULTIPLICATION_AND_ADD(lo_04, hi_04, lo_26, hi_26, stg2_0, \ + stg2_1, stg2_2, stg2_3, stp2_0, \ + stp2_1, stp2_2, stp2_3) \ + \ + stp2_4 = _mm_adds_epi16(stp1_4, stp1_5); \ + stp2_5 = _mm_subs_epi16(stp1_4, stp1_5); \ + stp2_6 = _mm_subs_epi16(stp1_7, stp1_6); \ + stp2_7 = _mm_adds_epi16(stp1_7, stp1_6); \ + } \ + \ + /* Stage3 */ \ + { \ + const __m128i lo_56 = _mm_unpacklo_epi16(stp2_6, stp2_5); \ + const __m128i hi_56 = _mm_unpackhi_epi16(stp2_6, stp2_5); \ + \ + stp1_0 = _mm_adds_epi16(stp2_0, stp2_3); \ + stp1_1 = _mm_adds_epi16(stp2_1, stp2_2); \ + stp1_2 = _mm_subs_epi16(stp2_1, stp2_2); \ + stp1_3 = _mm_subs_epi16(stp2_0, stp2_3); \ + \ + tmp0 = _mm_madd_epi16(lo_56, stg2_1); \ + tmp1 = _mm_madd_epi16(hi_56, stg2_1); \ + tmp2 = _mm_madd_epi16(lo_56, stg2_0); \ + tmp3 = _mm_madd_epi16(hi_56, stg2_0); \ + \ + tmp0 = _mm_add_epi32(tmp0, rounding); \ + tmp1 = _mm_add_epi32(tmp1, rounding); \ + tmp2 = _mm_add_epi32(tmp2, rounding); \ + tmp3 = _mm_add_epi32(tmp3, rounding); \ + \ + tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS); \ + tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS); \ + tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS); \ + tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS); \ + \ + stp1_5 = _mm_packs_epi32(tmp0, tmp1); \ + stp1_6 = _mm_packs_epi32(tmp2, tmp3); \ + } \ + \ + /* Stage4 */ \ + out0 = _mm_adds_epi16(stp1_0, stp2_7); \ + out1 = _mm_adds_epi16(stp1_1, stp1_6); \ + out2 = _mm_adds_epi16(stp1_2, stp1_5); \ + out3 = _mm_adds_epi16(stp1_3, stp2_4); \ + out4 = _mm_subs_epi16(stp1_3, stp2_4); \ + out5 = _mm_subs_epi16(stp1_2, stp1_5); \ + out6 = _mm_subs_epi16(stp1_1, stp1_6); \ + out7 = _mm_subs_epi16(stp1_0, stp2_7); \ + } + +void vpx_idct8x8_64_add_sse2(const tran_low_t *input, uint8_t *dest, + int stride) { + const __m128i zero = _mm_setzero_si128(); + const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING); + const __m128i final_rounding = _mm_set1_epi16(1 << 4); + const __m128i stg1_0 = pair_set_epi16(cospi_28_64, -cospi_4_64); + const __m128i stg1_1 = pair_set_epi16(cospi_4_64, cospi_28_64); + const __m128i stg1_2 = pair_set_epi16(-cospi_20_64, cospi_12_64); + const __m128i stg1_3 = pair_set_epi16(cospi_12_64, cospi_20_64); + const __m128i stg2_0 = pair_set_epi16(cospi_16_64, cospi_16_64); + const __m128i stg2_1 = pair_set_epi16(cospi_16_64, -cospi_16_64); + const __m128i stg2_2 = pair_set_epi16(cospi_24_64, -cospi_8_64); + const __m128i stg2_3 = pair_set_epi16(cospi_8_64, cospi_24_64); + + __m128i in0, in1, in2, in3, in4, in5, in6, in7; + __m128i stp1_0, stp1_1, stp1_2, stp1_3, stp1_4, stp1_5, stp1_6, stp1_7; + __m128i stp2_0, stp2_1, stp2_2, stp2_3, stp2_4, stp2_5, stp2_6, stp2_7; + __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; + int i; + + // Load input data. + in0 = load_input_data(input); + in1 = load_input_data(input + 8 * 1); + in2 = load_input_data(input + 8 * 2); + in3 = load_input_data(input + 8 * 3); + in4 = load_input_data(input + 8 * 4); + in5 = load_input_data(input + 8 * 5); + in6 = load_input_data(input + 8 * 6); + in7 = load_input_data(input + 8 * 7); + + // 2-D + for (i = 0; i < 2; i++) { + // 8x8 Transpose is copied from vpx_fdct8x8_sse2() + TRANSPOSE_8X8(in0, in1, in2, in3, in4, in5, in6, in7, + in0, in1, in2, in3, in4, in5, in6, in7); + + // 4-stage 1D idct8x8 + IDCT8(in0, in1, in2, in3, in4, in5, in6, in7, + in0, in1, in2, in3, in4, in5, in6, in7); + } + + // Final rounding and shift + in0 = _mm_adds_epi16(in0, final_rounding); + in1 = _mm_adds_epi16(in1, final_rounding); + in2 = _mm_adds_epi16(in2, final_rounding); + in3 = _mm_adds_epi16(in3, final_rounding); + in4 = _mm_adds_epi16(in4, final_rounding); + in5 = _mm_adds_epi16(in5, final_rounding); + in6 = _mm_adds_epi16(in6, final_rounding); + in7 = _mm_adds_epi16(in7, final_rounding); + + in0 = _mm_srai_epi16(in0, 5); + in1 = _mm_srai_epi16(in1, 5); + in2 = _mm_srai_epi16(in2, 5); + in3 = _mm_srai_epi16(in3, 5); + in4 = _mm_srai_epi16(in4, 5); + in5 = _mm_srai_epi16(in5, 5); + in6 = _mm_srai_epi16(in6, 5); + in7 = _mm_srai_epi16(in7, 5); + + RECON_AND_STORE(dest + 0 * stride, in0); + RECON_AND_STORE(dest + 1 * stride, in1); + RECON_AND_STORE(dest + 2 * stride, in2); + RECON_AND_STORE(dest + 3 * stride, in3); + RECON_AND_STORE(dest + 4 * stride, in4); + RECON_AND_STORE(dest + 5 * stride, in5); + RECON_AND_STORE(dest + 6 * stride, in6); + RECON_AND_STORE(dest + 7 * stride, in7); +} + +void vpx_idct8x8_1_add_sse2(const tran_low_t *input, uint8_t *dest, + int stride) { + __m128i dc_value; + const __m128i zero = _mm_setzero_si128(); + int a; + + a = (int)dct_const_round_shift(input[0] * cospi_16_64); + a = (int)dct_const_round_shift(a * cospi_16_64); + a = ROUND_POWER_OF_TWO(a, 5); + + dc_value = _mm_set1_epi16(a); + + RECON_AND_STORE(dest + 0 * stride, dc_value); + RECON_AND_STORE(dest + 1 * stride, dc_value); + RECON_AND_STORE(dest + 2 * stride, dc_value); + RECON_AND_STORE(dest + 3 * stride, dc_value); + RECON_AND_STORE(dest + 4 * stride, dc_value); + RECON_AND_STORE(dest + 5 * stride, dc_value); + RECON_AND_STORE(dest + 6 * stride, dc_value); + RECON_AND_STORE(dest + 7 * stride, dc_value); +} + +void idct8_sse2(__m128i *in) { + const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING); + const __m128i stg1_0 = pair_set_epi16(cospi_28_64, -cospi_4_64); + const __m128i stg1_1 = pair_set_epi16(cospi_4_64, cospi_28_64); + const __m128i stg1_2 = pair_set_epi16(-cospi_20_64, cospi_12_64); + const __m128i stg1_3 = pair_set_epi16(cospi_12_64, cospi_20_64); + const __m128i stg2_0 = pair_set_epi16(cospi_16_64, cospi_16_64); + const __m128i stg2_1 = pair_set_epi16(cospi_16_64, -cospi_16_64); + const __m128i stg2_2 = pair_set_epi16(cospi_24_64, -cospi_8_64); + const __m128i stg2_3 = pair_set_epi16(cospi_8_64, cospi_24_64); + + __m128i in0, in1, in2, in3, in4, in5, in6, in7; + __m128i stp1_0, stp1_1, stp1_2, stp1_3, stp1_4, stp1_5, stp1_6, stp1_7; + __m128i stp2_0, stp2_1, stp2_2, stp2_3, stp2_4, stp2_5, stp2_6, stp2_7; + __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; + + // 8x8 Transpose is copied from vpx_fdct8x8_sse2() + TRANSPOSE_8X8(in[0], in[1], in[2], in[3], in[4], in[5], in[6], in[7], + in0, in1, in2, in3, in4, in5, in6, in7); + + // 4-stage 1D idct8x8 + IDCT8(in0, in1, in2, in3, in4, in5, in6, in7, + in[0], in[1], in[2], in[3], in[4], in[5], in[6], in[7]); +} + +void iadst8_sse2(__m128i *in) { + const __m128i k__cospi_p02_p30 = pair_set_epi16(cospi_2_64, cospi_30_64); + const __m128i k__cospi_p30_m02 = pair_set_epi16(cospi_30_64, -cospi_2_64); + const __m128i k__cospi_p10_p22 = pair_set_epi16(cospi_10_64, cospi_22_64); + const __m128i k__cospi_p22_m10 = pair_set_epi16(cospi_22_64, -cospi_10_64); + const __m128i k__cospi_p18_p14 = pair_set_epi16(cospi_18_64, cospi_14_64); + const __m128i k__cospi_p14_m18 = pair_set_epi16(cospi_14_64, -cospi_18_64); + const __m128i k__cospi_p26_p06 = pair_set_epi16(cospi_26_64, cospi_6_64); + const __m128i k__cospi_p06_m26 = pair_set_epi16(cospi_6_64, -cospi_26_64); + const __m128i k__cospi_p08_p24 = pair_set_epi16(cospi_8_64, cospi_24_64); + const __m128i k__cospi_p24_m08 = pair_set_epi16(cospi_24_64, -cospi_8_64); + const __m128i k__cospi_m24_p08 = pair_set_epi16(-cospi_24_64, cospi_8_64); + const __m128i k__cospi_p16_m16 = pair_set_epi16(cospi_16_64, -cospi_16_64); + const __m128i k__cospi_p16_p16 = _mm_set1_epi16((int16_t)cospi_16_64); + const __m128i k__const_0 = _mm_set1_epi16(0); + const __m128i k__DCT_CONST_ROUNDING = _mm_set1_epi32(DCT_CONST_ROUNDING); + + __m128i u0, u1, u2, u3, u4, u5, u6, u7, u8, u9, u10, u11, u12, u13, u14, u15; + __m128i v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12, v13, v14, v15; + __m128i w0, w1, w2, w3, w4, w5, w6, w7, w8, w9, w10, w11, w12, w13, w14, w15; + __m128i s0, s1, s2, s3, s4, s5, s6, s7; + __m128i in0, in1, in2, in3, in4, in5, in6, in7; + + // transpose + array_transpose_8x8(in, in); + + // properly aligned for butterfly input + in0 = in[7]; + in1 = in[0]; + in2 = in[5]; + in3 = in[2]; + in4 = in[3]; + in5 = in[4]; + in6 = in[1]; + in7 = in[6]; + + // column transformation + // stage 1 + // interleave and multiply/add into 32-bit integer + s0 = _mm_unpacklo_epi16(in0, in1); + s1 = _mm_unpackhi_epi16(in0, in1); + s2 = _mm_unpacklo_epi16(in2, in3); + s3 = _mm_unpackhi_epi16(in2, in3); + s4 = _mm_unpacklo_epi16(in4, in5); + s5 = _mm_unpackhi_epi16(in4, in5); + s6 = _mm_unpacklo_epi16(in6, in7); + s7 = _mm_unpackhi_epi16(in6, in7); + + u0 = _mm_madd_epi16(s0, k__cospi_p02_p30); + u1 = _mm_madd_epi16(s1, k__cospi_p02_p30); + u2 = _mm_madd_epi16(s0, k__cospi_p30_m02); + u3 = _mm_madd_epi16(s1, k__cospi_p30_m02); + u4 = _mm_madd_epi16(s2, k__cospi_p10_p22); + u5 = _mm_madd_epi16(s3, k__cospi_p10_p22); + u6 = _mm_madd_epi16(s2, k__cospi_p22_m10); + u7 = _mm_madd_epi16(s3, k__cospi_p22_m10); + u8 = _mm_madd_epi16(s4, k__cospi_p18_p14); + u9 = _mm_madd_epi16(s5, k__cospi_p18_p14); + u10 = _mm_madd_epi16(s4, k__cospi_p14_m18); + u11 = _mm_madd_epi16(s5, k__cospi_p14_m18); + u12 = _mm_madd_epi16(s6, k__cospi_p26_p06); + u13 = _mm_madd_epi16(s7, k__cospi_p26_p06); + u14 = _mm_madd_epi16(s6, k__cospi_p06_m26); + u15 = _mm_madd_epi16(s7, k__cospi_p06_m26); + + // addition + w0 = _mm_add_epi32(u0, u8); + w1 = _mm_add_epi32(u1, u9); + w2 = _mm_add_epi32(u2, u10); + w3 = _mm_add_epi32(u3, u11); + w4 = _mm_add_epi32(u4, u12); + w5 = _mm_add_epi32(u5, u13); + w6 = _mm_add_epi32(u6, u14); + w7 = _mm_add_epi32(u7, u15); + w8 = _mm_sub_epi32(u0, u8); + w9 = _mm_sub_epi32(u1, u9); + w10 = _mm_sub_epi32(u2, u10); + w11 = _mm_sub_epi32(u3, u11); + w12 = _mm_sub_epi32(u4, u12); + w13 = _mm_sub_epi32(u5, u13); + w14 = _mm_sub_epi32(u6, u14); + w15 = _mm_sub_epi32(u7, u15); + + // shift and rounding + v0 = _mm_add_epi32(w0, k__DCT_CONST_ROUNDING); + v1 = _mm_add_epi32(w1, k__DCT_CONST_ROUNDING); + v2 = _mm_add_epi32(w2, k__DCT_CONST_ROUNDING); + v3 = _mm_add_epi32(w3, k__DCT_CONST_ROUNDING); + v4 = _mm_add_epi32(w4, k__DCT_CONST_ROUNDING); + v5 = _mm_add_epi32(w5, k__DCT_CONST_ROUNDING); + v6 = _mm_add_epi32(w6, k__DCT_CONST_ROUNDING); + v7 = _mm_add_epi32(w7, k__DCT_CONST_ROUNDING); + v8 = _mm_add_epi32(w8, k__DCT_CONST_ROUNDING); + v9 = _mm_add_epi32(w9, k__DCT_CONST_ROUNDING); + v10 = _mm_add_epi32(w10, k__DCT_CONST_ROUNDING); + v11 = _mm_add_epi32(w11, k__DCT_CONST_ROUNDING); + v12 = _mm_add_epi32(w12, k__DCT_CONST_ROUNDING); + v13 = _mm_add_epi32(w13, k__DCT_CONST_ROUNDING); + v14 = _mm_add_epi32(w14, k__DCT_CONST_ROUNDING); + v15 = _mm_add_epi32(w15, k__DCT_CONST_ROUNDING); + + u0 = _mm_srai_epi32(v0, DCT_CONST_BITS); + u1 = _mm_srai_epi32(v1, DCT_CONST_BITS); + u2 = _mm_srai_epi32(v2, DCT_CONST_BITS); + u3 = _mm_srai_epi32(v3, DCT_CONST_BITS); + u4 = _mm_srai_epi32(v4, DCT_CONST_BITS); + u5 = _mm_srai_epi32(v5, DCT_CONST_BITS); + u6 = _mm_srai_epi32(v6, DCT_CONST_BITS); + u7 = _mm_srai_epi32(v7, DCT_CONST_BITS); + u8 = _mm_srai_epi32(v8, DCT_CONST_BITS); + u9 = _mm_srai_epi32(v9, DCT_CONST_BITS); + u10 = _mm_srai_epi32(v10, DCT_CONST_BITS); + u11 = _mm_srai_epi32(v11, DCT_CONST_BITS); + u12 = _mm_srai_epi32(v12, DCT_CONST_BITS); + u13 = _mm_srai_epi32(v13, DCT_CONST_BITS); + u14 = _mm_srai_epi32(v14, DCT_CONST_BITS); + u15 = _mm_srai_epi32(v15, DCT_CONST_BITS); + + // back to 16-bit and pack 8 integers into __m128i + in[0] = _mm_packs_epi32(u0, u1); + in[1] = _mm_packs_epi32(u2, u3); + in[2] = _mm_packs_epi32(u4, u5); + in[3] = _mm_packs_epi32(u6, u7); + in[4] = _mm_packs_epi32(u8, u9); + in[5] = _mm_packs_epi32(u10, u11); + in[6] = _mm_packs_epi32(u12, u13); + in[7] = _mm_packs_epi32(u14, u15); + + // stage 2 + s0 = _mm_add_epi16(in[0], in[2]); + s1 = _mm_add_epi16(in[1], in[3]); + s2 = _mm_sub_epi16(in[0], in[2]); + s3 = _mm_sub_epi16(in[1], in[3]); + u0 = _mm_unpacklo_epi16(in[4], in[5]); + u1 = _mm_unpackhi_epi16(in[4], in[5]); + u2 = _mm_unpacklo_epi16(in[6], in[7]); + u3 = _mm_unpackhi_epi16(in[6], in[7]); + + v0 = _mm_madd_epi16(u0, k__cospi_p08_p24); + v1 = _mm_madd_epi16(u1, k__cospi_p08_p24); + v2 = _mm_madd_epi16(u0, k__cospi_p24_m08); + v3 = _mm_madd_epi16(u1, k__cospi_p24_m08); + v4 = _mm_madd_epi16(u2, k__cospi_m24_p08); + v5 = _mm_madd_epi16(u3, k__cospi_m24_p08); + v6 = _mm_madd_epi16(u2, k__cospi_p08_p24); + v7 = _mm_madd_epi16(u3, k__cospi_p08_p24); + + w0 = _mm_add_epi32(v0, v4); + w1 = _mm_add_epi32(v1, v5); + w2 = _mm_add_epi32(v2, v6); + w3 = _mm_add_epi32(v3, v7); + w4 = _mm_sub_epi32(v0, v4); + w5 = _mm_sub_epi32(v1, v5); + w6 = _mm_sub_epi32(v2, v6); + w7 = _mm_sub_epi32(v3, v7); + + v0 = _mm_add_epi32(w0, k__DCT_CONST_ROUNDING); + v1 = _mm_add_epi32(w1, k__DCT_CONST_ROUNDING); + v2 = _mm_add_epi32(w2, k__DCT_CONST_ROUNDING); + v3 = _mm_add_epi32(w3, k__DCT_CONST_ROUNDING); + v4 = _mm_add_epi32(w4, k__DCT_CONST_ROUNDING); + v5 = _mm_add_epi32(w5, k__DCT_CONST_ROUNDING); + v6 = _mm_add_epi32(w6, k__DCT_CONST_ROUNDING); + v7 = _mm_add_epi32(w7, k__DCT_CONST_ROUNDING); + + u0 = _mm_srai_epi32(v0, DCT_CONST_BITS); + u1 = _mm_srai_epi32(v1, DCT_CONST_BITS); + u2 = _mm_srai_epi32(v2, DCT_CONST_BITS); + u3 = _mm_srai_epi32(v3, DCT_CONST_BITS); + u4 = _mm_srai_epi32(v4, DCT_CONST_BITS); + u5 = _mm_srai_epi32(v5, DCT_CONST_BITS); + u6 = _mm_srai_epi32(v6, DCT_CONST_BITS); + u7 = _mm_srai_epi32(v7, DCT_CONST_BITS); + + // back to 16-bit intergers + s4 = _mm_packs_epi32(u0, u1); + s5 = _mm_packs_epi32(u2, u3); + s6 = _mm_packs_epi32(u4, u5); + s7 = _mm_packs_epi32(u6, u7); + + // stage 3 + u0 = _mm_unpacklo_epi16(s2, s3); + u1 = _mm_unpackhi_epi16(s2, s3); + u2 = _mm_unpacklo_epi16(s6, s7); + u3 = _mm_unpackhi_epi16(s6, s7); + + v0 = _mm_madd_epi16(u0, k__cospi_p16_p16); + v1 = _mm_madd_epi16(u1, k__cospi_p16_p16); + v2 = _mm_madd_epi16(u0, k__cospi_p16_m16); + v3 = _mm_madd_epi16(u1, k__cospi_p16_m16); + v4 = _mm_madd_epi16(u2, k__cospi_p16_p16); + v5 = _mm_madd_epi16(u3, k__cospi_p16_p16); + v6 = _mm_madd_epi16(u2, k__cospi_p16_m16); + v7 = _mm_madd_epi16(u3, k__cospi_p16_m16); + + u0 = _mm_add_epi32(v0, k__DCT_CONST_ROUNDING); + u1 = _mm_add_epi32(v1, k__DCT_CONST_ROUNDING); + u2 = _mm_add_epi32(v2, k__DCT_CONST_ROUNDING); + u3 = _mm_add_epi32(v3, k__DCT_CONST_ROUNDING); + u4 = _mm_add_epi32(v4, k__DCT_CONST_ROUNDING); + u5 = _mm_add_epi32(v5, k__DCT_CONST_ROUNDING); + u6 = _mm_add_epi32(v6, k__DCT_CONST_ROUNDING); + u7 = _mm_add_epi32(v7, k__DCT_CONST_ROUNDING); + + v0 = _mm_srai_epi32(u0, DCT_CONST_BITS); + v1 = _mm_srai_epi32(u1, DCT_CONST_BITS); + v2 = _mm_srai_epi32(u2, DCT_CONST_BITS); + v3 = _mm_srai_epi32(u3, DCT_CONST_BITS); + v4 = _mm_srai_epi32(u4, DCT_CONST_BITS); + v5 = _mm_srai_epi32(u5, DCT_CONST_BITS); + v6 = _mm_srai_epi32(u6, DCT_CONST_BITS); + v7 = _mm_srai_epi32(u7, DCT_CONST_BITS); + + s2 = _mm_packs_epi32(v0, v1); + s3 = _mm_packs_epi32(v2, v3); + s6 = _mm_packs_epi32(v4, v5); + s7 = _mm_packs_epi32(v6, v7); + + in[0] = s0; + in[1] = _mm_sub_epi16(k__const_0, s4); + in[2] = s6; + in[3] = _mm_sub_epi16(k__const_0, s2); + in[4] = s3; + in[5] = _mm_sub_epi16(k__const_0, s7); + in[6] = s5; + in[7] = _mm_sub_epi16(k__const_0, s1); +} + +void vpx_idct8x8_12_add_sse2(const tran_low_t *input, uint8_t *dest, + int stride) { + const __m128i zero = _mm_setzero_si128(); + const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING); + const __m128i final_rounding = _mm_set1_epi16(1 << 4); + const __m128i stg1_0 = pair_set_epi16(cospi_28_64, -cospi_4_64); + const __m128i stg1_1 = pair_set_epi16(cospi_4_64, cospi_28_64); + const __m128i stg1_2 = pair_set_epi16(-cospi_20_64, cospi_12_64); + const __m128i stg1_3 = pair_set_epi16(cospi_12_64, cospi_20_64); + const __m128i stg2_0 = pair_set_epi16(cospi_16_64, cospi_16_64); + const __m128i stg2_1 = pair_set_epi16(cospi_16_64, -cospi_16_64); + const __m128i stg2_2 = pair_set_epi16(cospi_24_64, -cospi_8_64); + const __m128i stg2_3 = pair_set_epi16(cospi_8_64, cospi_24_64); + const __m128i stg3_0 = pair_set_epi16(-cospi_16_64, cospi_16_64); + + __m128i in0, in1, in2, in3, in4, in5, in6, in7; + __m128i stp1_0, stp1_1, stp1_2, stp1_3, stp1_4, stp1_5, stp1_6, stp1_7; + __m128i stp2_0, stp2_1, stp2_2, stp2_3, stp2_4, stp2_5, stp2_6, stp2_7; + __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; + + // Rows. Load 4-row input data. + in0 = load_input_data(input); + in1 = load_input_data(input + 8 * 1); + in2 = load_input_data(input + 8 * 2); + in3 = load_input_data(input + 8 * 3); + + // 8x4 Transpose + TRANSPOSE_8X8_10(in0, in1, in2, in3, in0, in1); + // Stage1 + { + const __m128i lo_17 = _mm_unpackhi_epi16(in0, zero); + const __m128i lo_35 = _mm_unpackhi_epi16(in1, zero); + + tmp0 = _mm_madd_epi16(lo_17, stg1_0); + tmp2 = _mm_madd_epi16(lo_17, stg1_1); + tmp4 = _mm_madd_epi16(lo_35, stg1_2); + tmp6 = _mm_madd_epi16(lo_35, stg1_3); + + tmp0 = _mm_add_epi32(tmp0, rounding); + tmp2 = _mm_add_epi32(tmp2, rounding); + tmp4 = _mm_add_epi32(tmp4, rounding); + tmp6 = _mm_add_epi32(tmp6, rounding); + tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS); + tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS); + tmp4 = _mm_srai_epi32(tmp4, DCT_CONST_BITS); + tmp6 = _mm_srai_epi32(tmp6, DCT_CONST_BITS); + + stp1_4 = _mm_packs_epi32(tmp0, tmp2); + stp1_5 = _mm_packs_epi32(tmp4, tmp6); + } + + // Stage2 + { + const __m128i lo_04 = _mm_unpacklo_epi16(in0, zero); + const __m128i lo_26 = _mm_unpacklo_epi16(in1, zero); + + tmp0 = _mm_madd_epi16(lo_04, stg2_0); + tmp2 = _mm_madd_epi16(lo_04, stg2_1); + tmp4 = _mm_madd_epi16(lo_26, stg2_2); + tmp6 = _mm_madd_epi16(lo_26, stg2_3); + + tmp0 = _mm_add_epi32(tmp0, rounding); + tmp2 = _mm_add_epi32(tmp2, rounding); + tmp4 = _mm_add_epi32(tmp4, rounding); + tmp6 = _mm_add_epi32(tmp6, rounding); + tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS); + tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS); + tmp4 = _mm_srai_epi32(tmp4, DCT_CONST_BITS); + tmp6 = _mm_srai_epi32(tmp6, DCT_CONST_BITS); + + stp2_0 = _mm_packs_epi32(tmp0, tmp2); + stp2_2 = _mm_packs_epi32(tmp6, tmp4); + + tmp0 = _mm_adds_epi16(stp1_4, stp1_5); + tmp1 = _mm_subs_epi16(stp1_4, stp1_5); + + stp2_4 = tmp0; + stp2_5 = _mm_unpacklo_epi64(tmp1, zero); + stp2_6 = _mm_unpackhi_epi64(tmp1, zero); + } + + // Stage3 + { + const __m128i lo_56 = _mm_unpacklo_epi16(stp2_5, stp2_6); + + tmp4 = _mm_adds_epi16(stp2_0, stp2_2); + tmp6 = _mm_subs_epi16(stp2_0, stp2_2); + + stp1_2 = _mm_unpackhi_epi64(tmp6, tmp4); + stp1_3 = _mm_unpacklo_epi64(tmp6, tmp4); + + tmp0 = _mm_madd_epi16(lo_56, stg3_0); + tmp2 = _mm_madd_epi16(lo_56, stg2_0); // stg3_1 = stg2_0 + + tmp0 = _mm_add_epi32(tmp0, rounding); + tmp2 = _mm_add_epi32(tmp2, rounding); + tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS); + tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS); + + stp1_5 = _mm_packs_epi32(tmp0, tmp2); + } + + // Stage4 + tmp0 = _mm_adds_epi16(stp1_3, stp2_4); + tmp1 = _mm_adds_epi16(stp1_2, stp1_5); + tmp2 = _mm_subs_epi16(stp1_3, stp2_4); + tmp3 = _mm_subs_epi16(stp1_2, stp1_5); + + TRANSPOSE_4X8_10(tmp0, tmp1, tmp2, tmp3, in0, in1, in2, in3) + + IDCT8(in0, in1, in2, in3, zero, zero, zero, zero, + in0, in1, in2, in3, in4, in5, in6, in7); + // Final rounding and shift + in0 = _mm_adds_epi16(in0, final_rounding); + in1 = _mm_adds_epi16(in1, final_rounding); + in2 = _mm_adds_epi16(in2, final_rounding); + in3 = _mm_adds_epi16(in3, final_rounding); + in4 = _mm_adds_epi16(in4, final_rounding); + in5 = _mm_adds_epi16(in5, final_rounding); + in6 = _mm_adds_epi16(in6, final_rounding); + in7 = _mm_adds_epi16(in7, final_rounding); + + in0 = _mm_srai_epi16(in0, 5); + in1 = _mm_srai_epi16(in1, 5); + in2 = _mm_srai_epi16(in2, 5); + in3 = _mm_srai_epi16(in3, 5); + in4 = _mm_srai_epi16(in4, 5); + in5 = _mm_srai_epi16(in5, 5); + in6 = _mm_srai_epi16(in6, 5); + in7 = _mm_srai_epi16(in7, 5); + + RECON_AND_STORE(dest + 0 * stride, in0); + RECON_AND_STORE(dest + 1 * stride, in1); + RECON_AND_STORE(dest + 2 * stride, in2); + RECON_AND_STORE(dest + 3 * stride, in3); + RECON_AND_STORE(dest + 4 * stride, in4); + RECON_AND_STORE(dest + 5 * stride, in5); + RECON_AND_STORE(dest + 6 * stride, in6); + RECON_AND_STORE(dest + 7 * stride, in7); +} + +#define IDCT16 \ + /* Stage2 */ \ + { \ + const __m128i lo_1_15 = _mm_unpacklo_epi16(in[1], in[15]); \ + const __m128i hi_1_15 = _mm_unpackhi_epi16(in[1], in[15]); \ + const __m128i lo_9_7 = _mm_unpacklo_epi16(in[9], in[7]); \ + const __m128i hi_9_7 = _mm_unpackhi_epi16(in[9], in[7]); \ + const __m128i lo_5_11 = _mm_unpacklo_epi16(in[5], in[11]); \ + const __m128i hi_5_11 = _mm_unpackhi_epi16(in[5], in[11]); \ + const __m128i lo_13_3 = _mm_unpacklo_epi16(in[13], in[3]); \ + const __m128i hi_13_3 = _mm_unpackhi_epi16(in[13], in[3]); \ + \ + MULTIPLICATION_AND_ADD(lo_1_15, hi_1_15, lo_9_7, hi_9_7, \ + stg2_0, stg2_1, stg2_2, stg2_3, \ + stp2_8, stp2_15, stp2_9, stp2_14) \ + \ + MULTIPLICATION_AND_ADD(lo_5_11, hi_5_11, lo_13_3, hi_13_3, \ + stg2_4, stg2_5, stg2_6, stg2_7, \ + stp2_10, stp2_13, stp2_11, stp2_12) \ + } \ + \ + /* Stage3 */ \ + { \ + const __m128i lo_2_14 = _mm_unpacklo_epi16(in[2], in[14]); \ + const __m128i hi_2_14 = _mm_unpackhi_epi16(in[2], in[14]); \ + const __m128i lo_10_6 = _mm_unpacklo_epi16(in[10], in[6]); \ + const __m128i hi_10_6 = _mm_unpackhi_epi16(in[10], in[6]); \ + \ + MULTIPLICATION_AND_ADD(lo_2_14, hi_2_14, lo_10_6, hi_10_6, \ + stg3_0, stg3_1, stg3_2, stg3_3, \ + stp1_4, stp1_7, stp1_5, stp1_6) \ + \ + stp1_8_0 = _mm_add_epi16(stp2_8, stp2_9); \ + stp1_9 = _mm_sub_epi16(stp2_8, stp2_9); \ + stp1_10 = _mm_sub_epi16(stp2_11, stp2_10); \ + stp1_11 = _mm_add_epi16(stp2_11, stp2_10); \ + \ + stp1_12_0 = _mm_add_epi16(stp2_12, stp2_13); \ + stp1_13 = _mm_sub_epi16(stp2_12, stp2_13); \ + stp1_14 = _mm_sub_epi16(stp2_15, stp2_14); \ + stp1_15 = _mm_add_epi16(stp2_15, stp2_14); \ + } \ + \ + /* Stage4 */ \ + { \ + const __m128i lo_0_8 = _mm_unpacklo_epi16(in[0], in[8]); \ + const __m128i hi_0_8 = _mm_unpackhi_epi16(in[0], in[8]); \ + const __m128i lo_4_12 = _mm_unpacklo_epi16(in[4], in[12]); \ + const __m128i hi_4_12 = _mm_unpackhi_epi16(in[4], in[12]); \ + \ + const __m128i lo_9_14 = _mm_unpacklo_epi16(stp1_9, stp1_14); \ + const __m128i hi_9_14 = _mm_unpackhi_epi16(stp1_9, stp1_14); \ + const __m128i lo_10_13 = _mm_unpacklo_epi16(stp1_10, stp1_13); \ + const __m128i hi_10_13 = _mm_unpackhi_epi16(stp1_10, stp1_13); \ + \ + MULTIPLICATION_AND_ADD(lo_0_8, hi_0_8, lo_4_12, hi_4_12, \ + stg4_0, stg4_1, stg4_2, stg4_3, \ + stp2_0, stp2_1, stp2_2, stp2_3) \ + \ + stp2_4 = _mm_add_epi16(stp1_4, stp1_5); \ + stp2_5 = _mm_sub_epi16(stp1_4, stp1_5); \ + stp2_6 = _mm_sub_epi16(stp1_7, stp1_6); \ + stp2_7 = _mm_add_epi16(stp1_7, stp1_6); \ + \ + MULTIPLICATION_AND_ADD(lo_9_14, hi_9_14, lo_10_13, hi_10_13, \ + stg4_4, stg4_5, stg4_6, stg4_7, \ + stp2_9, stp2_14, stp2_10, stp2_13) \ + } \ + \ + /* Stage5 */ \ + { \ + const __m128i lo_6_5 = _mm_unpacklo_epi16(stp2_6, stp2_5); \ + const __m128i hi_6_5 = _mm_unpackhi_epi16(stp2_6, stp2_5); \ + \ + stp1_0 = _mm_add_epi16(stp2_0, stp2_3); \ + stp1_1 = _mm_add_epi16(stp2_1, stp2_2); \ + stp1_2 = _mm_sub_epi16(stp2_1, stp2_2); \ + stp1_3 = _mm_sub_epi16(stp2_0, stp2_3); \ + \ + tmp0 = _mm_madd_epi16(lo_6_5, stg4_1); \ + tmp1 = _mm_madd_epi16(hi_6_5, stg4_1); \ + tmp2 = _mm_madd_epi16(lo_6_5, stg4_0); \ + tmp3 = _mm_madd_epi16(hi_6_5, stg4_0); \ + \ + tmp0 = _mm_add_epi32(tmp0, rounding); \ + tmp1 = _mm_add_epi32(tmp1, rounding); \ + tmp2 = _mm_add_epi32(tmp2, rounding); \ + tmp3 = _mm_add_epi32(tmp3, rounding); \ + \ + tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS); \ + tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS); \ + tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS); \ + tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS); \ + \ + stp1_5 = _mm_packs_epi32(tmp0, tmp1); \ + stp1_6 = _mm_packs_epi32(tmp2, tmp3); \ + \ + stp1_8 = _mm_add_epi16(stp1_8_0, stp1_11); \ + stp1_9 = _mm_add_epi16(stp2_9, stp2_10); \ + stp1_10 = _mm_sub_epi16(stp2_9, stp2_10); \ + stp1_11 = _mm_sub_epi16(stp1_8_0, stp1_11); \ + \ + stp1_12 = _mm_sub_epi16(stp1_15, stp1_12_0); \ + stp1_13 = _mm_sub_epi16(stp2_14, stp2_13); \ + stp1_14 = _mm_add_epi16(stp2_14, stp2_13); \ + stp1_15 = _mm_add_epi16(stp1_15, stp1_12_0); \ + } \ + \ + /* Stage6 */ \ + { \ + const __m128i lo_10_13 = _mm_unpacklo_epi16(stp1_10, stp1_13); \ + const __m128i hi_10_13 = _mm_unpackhi_epi16(stp1_10, stp1_13); \ + const __m128i lo_11_12 = _mm_unpacklo_epi16(stp1_11, stp1_12); \ + const __m128i hi_11_12 = _mm_unpackhi_epi16(stp1_11, stp1_12); \ + \ + stp2_0 = _mm_add_epi16(stp1_0, stp2_7); \ + stp2_1 = _mm_add_epi16(stp1_1, stp1_6); \ + stp2_2 = _mm_add_epi16(stp1_2, stp1_5); \ + stp2_3 = _mm_add_epi16(stp1_3, stp2_4); \ + stp2_4 = _mm_sub_epi16(stp1_3, stp2_4); \ + stp2_5 = _mm_sub_epi16(stp1_2, stp1_5); \ + stp2_6 = _mm_sub_epi16(stp1_1, stp1_6); \ + stp2_7 = _mm_sub_epi16(stp1_0, stp2_7); \ + \ + MULTIPLICATION_AND_ADD(lo_10_13, hi_10_13, lo_11_12, hi_11_12, \ + stg6_0, stg4_0, stg6_0, stg4_0, \ + stp2_10, stp2_13, stp2_11, stp2_12) \ + } + +#define IDCT16_10 \ + /* Stage2 */ \ + { \ + const __m128i lo_1_15 = _mm_unpacklo_epi16(in[1], zero); \ + const __m128i hi_1_15 = _mm_unpackhi_epi16(in[1], zero); \ + const __m128i lo_13_3 = _mm_unpacklo_epi16(zero, in[3]); \ + const __m128i hi_13_3 = _mm_unpackhi_epi16(zero, in[3]); \ + \ + MULTIPLICATION_AND_ADD(lo_1_15, hi_1_15, lo_13_3, hi_13_3, \ + stg2_0, stg2_1, stg2_6, stg2_7, \ + stp1_8_0, stp1_15, stp1_11, stp1_12_0) \ + } \ + \ + /* Stage3 */ \ + { \ + const __m128i lo_2_14 = _mm_unpacklo_epi16(in[2], zero); \ + const __m128i hi_2_14 = _mm_unpackhi_epi16(in[2], zero); \ + \ + MULTIPLICATION_AND_ADD_2(lo_2_14, hi_2_14, \ + stg3_0, stg3_1, \ + stp2_4, stp2_7) \ + \ + stp1_9 = stp1_8_0; \ + stp1_10 = stp1_11; \ + \ + stp1_13 = stp1_12_0; \ + stp1_14 = stp1_15; \ + } \ + \ + /* Stage4 */ \ + { \ + const __m128i lo_0_8 = _mm_unpacklo_epi16(in[0], zero); \ + const __m128i hi_0_8 = _mm_unpackhi_epi16(in[0], zero); \ + \ + const __m128i lo_9_14 = _mm_unpacklo_epi16(stp1_9, stp1_14); \ + const __m128i hi_9_14 = _mm_unpackhi_epi16(stp1_9, stp1_14); \ + const __m128i lo_10_13 = _mm_unpacklo_epi16(stp1_10, stp1_13); \ + const __m128i hi_10_13 = _mm_unpackhi_epi16(stp1_10, stp1_13); \ + \ + MULTIPLICATION_AND_ADD_2(lo_0_8, hi_0_8, \ + stg4_0, stg4_1, \ + stp1_0, stp1_1) \ + stp2_5 = stp2_4; \ + stp2_6 = stp2_7; \ + \ + MULTIPLICATION_AND_ADD(lo_9_14, hi_9_14, lo_10_13, hi_10_13, \ + stg4_4, stg4_5, stg4_6, stg4_7, \ + stp2_9, stp2_14, stp2_10, stp2_13) \ + } \ + \ + /* Stage5 */ \ + { \ + const __m128i lo_6_5 = _mm_unpacklo_epi16(stp2_6, stp2_5); \ + const __m128i hi_6_5 = _mm_unpackhi_epi16(stp2_6, stp2_5); \ + \ + stp1_2 = stp1_1; \ + stp1_3 = stp1_0; \ + \ + tmp0 = _mm_madd_epi16(lo_6_5, stg4_1); \ + tmp1 = _mm_madd_epi16(hi_6_5, stg4_1); \ + tmp2 = _mm_madd_epi16(lo_6_5, stg4_0); \ + tmp3 = _mm_madd_epi16(hi_6_5, stg4_0); \ + \ + tmp0 = _mm_add_epi32(tmp0, rounding); \ + tmp1 = _mm_add_epi32(tmp1, rounding); \ + tmp2 = _mm_add_epi32(tmp2, rounding); \ + tmp3 = _mm_add_epi32(tmp3, rounding); \ + \ + tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS); \ + tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS); \ + tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS); \ + tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS); \ + \ + stp1_5 = _mm_packs_epi32(tmp0, tmp1); \ + stp1_6 = _mm_packs_epi32(tmp2, tmp3); \ + \ + stp1_8 = _mm_add_epi16(stp1_8_0, stp1_11); \ + stp1_9 = _mm_add_epi16(stp2_9, stp2_10); \ + stp1_10 = _mm_sub_epi16(stp2_9, stp2_10); \ + stp1_11 = _mm_sub_epi16(stp1_8_0, stp1_11); \ + \ + stp1_12 = _mm_sub_epi16(stp1_15, stp1_12_0); \ + stp1_13 = _mm_sub_epi16(stp2_14, stp2_13); \ + stp1_14 = _mm_add_epi16(stp2_14, stp2_13); \ + stp1_15 = _mm_add_epi16(stp1_15, stp1_12_0); \ + } \ + \ + /* Stage6 */ \ + { \ + const __m128i lo_10_13 = _mm_unpacklo_epi16(stp1_10, stp1_13); \ + const __m128i hi_10_13 = _mm_unpackhi_epi16(stp1_10, stp1_13); \ + const __m128i lo_11_12 = _mm_unpacklo_epi16(stp1_11, stp1_12); \ + const __m128i hi_11_12 = _mm_unpackhi_epi16(stp1_11, stp1_12); \ + \ + stp2_0 = _mm_add_epi16(stp1_0, stp2_7); \ + stp2_1 = _mm_add_epi16(stp1_1, stp1_6); \ + stp2_2 = _mm_add_epi16(stp1_2, stp1_5); \ + stp2_3 = _mm_add_epi16(stp1_3, stp2_4); \ + stp2_4 = _mm_sub_epi16(stp1_3, stp2_4); \ + stp2_5 = _mm_sub_epi16(stp1_2, stp1_5); \ + stp2_6 = _mm_sub_epi16(stp1_1, stp1_6); \ + stp2_7 = _mm_sub_epi16(stp1_0, stp2_7); \ + \ + MULTIPLICATION_AND_ADD(lo_10_13, hi_10_13, lo_11_12, hi_11_12, \ + stg6_0, stg4_0, stg6_0, stg4_0, \ + stp2_10, stp2_13, stp2_11, stp2_12) \ + } + +void vpx_idct16x16_256_add_sse2(const tran_low_t *input, uint8_t *dest, + int stride) { + const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING); + const __m128i final_rounding = _mm_set1_epi16(1 << 5); + const __m128i zero = _mm_setzero_si128(); + + const __m128i stg2_0 = pair_set_epi16(cospi_30_64, -cospi_2_64); + const __m128i stg2_1 = pair_set_epi16(cospi_2_64, cospi_30_64); + const __m128i stg2_2 = pair_set_epi16(cospi_14_64, -cospi_18_64); + const __m128i stg2_3 = pair_set_epi16(cospi_18_64, cospi_14_64); + const __m128i stg2_4 = pair_set_epi16(cospi_22_64, -cospi_10_64); + const __m128i stg2_5 = pair_set_epi16(cospi_10_64, cospi_22_64); + const __m128i stg2_6 = pair_set_epi16(cospi_6_64, -cospi_26_64); + const __m128i stg2_7 = pair_set_epi16(cospi_26_64, cospi_6_64); + + const __m128i stg3_0 = pair_set_epi16(cospi_28_64, -cospi_4_64); + const __m128i stg3_1 = pair_set_epi16(cospi_4_64, cospi_28_64); + const __m128i stg3_2 = pair_set_epi16(cospi_12_64, -cospi_20_64); + const __m128i stg3_3 = pair_set_epi16(cospi_20_64, cospi_12_64); + + const __m128i stg4_0 = pair_set_epi16(cospi_16_64, cospi_16_64); + const __m128i stg4_1 = pair_set_epi16(cospi_16_64, -cospi_16_64); + const __m128i stg4_2 = pair_set_epi16(cospi_24_64, -cospi_8_64); + const __m128i stg4_3 = pair_set_epi16(cospi_8_64, cospi_24_64); + const __m128i stg4_4 = pair_set_epi16(-cospi_8_64, cospi_24_64); + const __m128i stg4_5 = pair_set_epi16(cospi_24_64, cospi_8_64); + const __m128i stg4_6 = pair_set_epi16(-cospi_24_64, -cospi_8_64); + const __m128i stg4_7 = pair_set_epi16(-cospi_8_64, cospi_24_64); + + const __m128i stg6_0 = pair_set_epi16(-cospi_16_64, cospi_16_64); + + __m128i in[16], l[16], r[16], *curr1; + __m128i stp1_0, stp1_1, stp1_2, stp1_3, stp1_4, stp1_5, stp1_6, stp1_7, + stp1_8, stp1_9, stp1_10, stp1_11, stp1_12, stp1_13, stp1_14, stp1_15, + stp1_8_0, stp1_12_0; + __m128i stp2_0, stp2_1, stp2_2, stp2_3, stp2_4, stp2_5, stp2_6, stp2_7, + stp2_8, stp2_9, stp2_10, stp2_11, stp2_12, stp2_13, stp2_14, stp2_15; + __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; + int i; + + curr1 = l; + for (i = 0; i < 2; i++) { + // 1-D idct + + // Load input data. + in[0] = load_input_data(input); + in[8] = load_input_data(input + 8 * 1); + in[1] = load_input_data(input + 8 * 2); + in[9] = load_input_data(input + 8 * 3); + in[2] = load_input_data(input + 8 * 4); + in[10] = load_input_data(input + 8 * 5); + in[3] = load_input_data(input + 8 * 6); + in[11] = load_input_data(input + 8 * 7); + in[4] = load_input_data(input + 8 * 8); + in[12] = load_input_data(input + 8 * 9); + in[5] = load_input_data(input + 8 * 10); + in[13] = load_input_data(input + 8 * 11); + in[6] = load_input_data(input + 8 * 12); + in[14] = load_input_data(input + 8 * 13); + in[7] = load_input_data(input + 8 * 14); + in[15] = load_input_data(input + 8 * 15); + + array_transpose_8x8(in, in); + array_transpose_8x8(in + 8, in + 8); + + IDCT16 + + // Stage7 + curr1[0] = _mm_add_epi16(stp2_0, stp1_15); + curr1[1] = _mm_add_epi16(stp2_1, stp1_14); + curr1[2] = _mm_add_epi16(stp2_2, stp2_13); + curr1[3] = _mm_add_epi16(stp2_3, stp2_12); + curr1[4] = _mm_add_epi16(stp2_4, stp2_11); + curr1[5] = _mm_add_epi16(stp2_5, stp2_10); + curr1[6] = _mm_add_epi16(stp2_6, stp1_9); + curr1[7] = _mm_add_epi16(stp2_7, stp1_8); + curr1[8] = _mm_sub_epi16(stp2_7, stp1_8); + curr1[9] = _mm_sub_epi16(stp2_6, stp1_9); + curr1[10] = _mm_sub_epi16(stp2_5, stp2_10); + curr1[11] = _mm_sub_epi16(stp2_4, stp2_11); + curr1[12] = _mm_sub_epi16(stp2_3, stp2_12); + curr1[13] = _mm_sub_epi16(stp2_2, stp2_13); + curr1[14] = _mm_sub_epi16(stp2_1, stp1_14); + curr1[15] = _mm_sub_epi16(stp2_0, stp1_15); + + curr1 = r; + input += 128; + } + for (i = 0; i < 2; i++) { + int j; + // 1-D idct + array_transpose_8x8(l + i * 8, in); + array_transpose_8x8(r + i * 8, in + 8); + + IDCT16 + + // 2-D + in[0] = _mm_add_epi16(stp2_0, stp1_15); + in[1] = _mm_add_epi16(stp2_1, stp1_14); + in[2] = _mm_add_epi16(stp2_2, stp2_13); + in[3] = _mm_add_epi16(stp2_3, stp2_12); + in[4] = _mm_add_epi16(stp2_4, stp2_11); + in[5] = _mm_add_epi16(stp2_5, stp2_10); + in[6] = _mm_add_epi16(stp2_6, stp1_9); + in[7] = _mm_add_epi16(stp2_7, stp1_8); + in[8] = _mm_sub_epi16(stp2_7, stp1_8); + in[9] = _mm_sub_epi16(stp2_6, stp1_9); + in[10] = _mm_sub_epi16(stp2_5, stp2_10); + in[11] = _mm_sub_epi16(stp2_4, stp2_11); + in[12] = _mm_sub_epi16(stp2_3, stp2_12); + in[13] = _mm_sub_epi16(stp2_2, stp2_13); + in[14] = _mm_sub_epi16(stp2_1, stp1_14); + in[15] = _mm_sub_epi16(stp2_0, stp1_15); + + for (j = 0; j < 16; ++j) { + // Final rounding and shift + in[j] = _mm_adds_epi16(in[j], final_rounding); + in[j] = _mm_srai_epi16(in[j], 6); + RECON_AND_STORE(dest + j * stride, in[j]); + } + + dest += 8; + } +} + +void vpx_idct16x16_1_add_sse2(const tran_low_t *input, uint8_t *dest, + int stride) { + __m128i dc_value; + const __m128i zero = _mm_setzero_si128(); + int a, i; + + a = (int)dct_const_round_shift(input[0] * cospi_16_64); + a = (int)dct_const_round_shift(a * cospi_16_64); + a = ROUND_POWER_OF_TWO(a, 6); + + dc_value = _mm_set1_epi16(a); + + for (i = 0; i < 16; ++i) { + RECON_AND_STORE(dest + 0, dc_value); + RECON_AND_STORE(dest + 8, dc_value); + dest += stride; + } +} + +static void iadst16_8col(__m128i *in) { + // perform 16x16 1-D ADST for 8 columns + __m128i s[16], x[16], u[32], v[32]; + const __m128i k__cospi_p01_p31 = pair_set_epi16(cospi_1_64, cospi_31_64); + const __m128i k__cospi_p31_m01 = pair_set_epi16(cospi_31_64, -cospi_1_64); + const __m128i k__cospi_p05_p27 = pair_set_epi16(cospi_5_64, cospi_27_64); + const __m128i k__cospi_p27_m05 = pair_set_epi16(cospi_27_64, -cospi_5_64); + const __m128i k__cospi_p09_p23 = pair_set_epi16(cospi_9_64, cospi_23_64); + const __m128i k__cospi_p23_m09 = pair_set_epi16(cospi_23_64, -cospi_9_64); + const __m128i k__cospi_p13_p19 = pair_set_epi16(cospi_13_64, cospi_19_64); + const __m128i k__cospi_p19_m13 = pair_set_epi16(cospi_19_64, -cospi_13_64); + const __m128i k__cospi_p17_p15 = pair_set_epi16(cospi_17_64, cospi_15_64); + const __m128i k__cospi_p15_m17 = pair_set_epi16(cospi_15_64, -cospi_17_64); + const __m128i k__cospi_p21_p11 = pair_set_epi16(cospi_21_64, cospi_11_64); + const __m128i k__cospi_p11_m21 = pair_set_epi16(cospi_11_64, -cospi_21_64); + const __m128i k__cospi_p25_p07 = pair_set_epi16(cospi_25_64, cospi_7_64); + const __m128i k__cospi_p07_m25 = pair_set_epi16(cospi_7_64, -cospi_25_64); + const __m128i k__cospi_p29_p03 = pair_set_epi16(cospi_29_64, cospi_3_64); + const __m128i k__cospi_p03_m29 = pair_set_epi16(cospi_3_64, -cospi_29_64); + const __m128i k__cospi_p04_p28 = pair_set_epi16(cospi_4_64, cospi_28_64); + const __m128i k__cospi_p28_m04 = pair_set_epi16(cospi_28_64, -cospi_4_64); + const __m128i k__cospi_p20_p12 = pair_set_epi16(cospi_20_64, cospi_12_64); + const __m128i k__cospi_p12_m20 = pair_set_epi16(cospi_12_64, -cospi_20_64); + const __m128i k__cospi_m28_p04 = pair_set_epi16(-cospi_28_64, cospi_4_64); + const __m128i k__cospi_m12_p20 = pair_set_epi16(-cospi_12_64, cospi_20_64); + const __m128i k__cospi_p08_p24 = pair_set_epi16(cospi_8_64, cospi_24_64); + const __m128i k__cospi_p24_m08 = pair_set_epi16(cospi_24_64, -cospi_8_64); + const __m128i k__cospi_m24_p08 = pair_set_epi16(-cospi_24_64, cospi_8_64); + const __m128i k__cospi_m16_m16 = _mm_set1_epi16((int16_t)-cospi_16_64); + const __m128i k__cospi_p16_p16 = _mm_set1_epi16((int16_t)cospi_16_64); + const __m128i k__cospi_p16_m16 = pair_set_epi16(cospi_16_64, -cospi_16_64); + const __m128i k__cospi_m16_p16 = pair_set_epi16(-cospi_16_64, cospi_16_64); + const __m128i k__DCT_CONST_ROUNDING = _mm_set1_epi32(DCT_CONST_ROUNDING); + const __m128i kZero = _mm_set1_epi16(0); + + u[0] = _mm_unpacklo_epi16(in[15], in[0]); + u[1] = _mm_unpackhi_epi16(in[15], in[0]); + u[2] = _mm_unpacklo_epi16(in[13], in[2]); + u[3] = _mm_unpackhi_epi16(in[13], in[2]); + u[4] = _mm_unpacklo_epi16(in[11], in[4]); + u[5] = _mm_unpackhi_epi16(in[11], in[4]); + u[6] = _mm_unpacklo_epi16(in[9], in[6]); + u[7] = _mm_unpackhi_epi16(in[9], in[6]); + u[8] = _mm_unpacklo_epi16(in[7], in[8]); + u[9] = _mm_unpackhi_epi16(in[7], in[8]); + u[10] = _mm_unpacklo_epi16(in[5], in[10]); + u[11] = _mm_unpackhi_epi16(in[5], in[10]); + u[12] = _mm_unpacklo_epi16(in[3], in[12]); + u[13] = _mm_unpackhi_epi16(in[3], in[12]); + u[14] = _mm_unpacklo_epi16(in[1], in[14]); + u[15] = _mm_unpackhi_epi16(in[1], in[14]); + + v[0] = _mm_madd_epi16(u[0], k__cospi_p01_p31); + v[1] = _mm_madd_epi16(u[1], k__cospi_p01_p31); + v[2] = _mm_madd_epi16(u[0], k__cospi_p31_m01); + v[3] = _mm_madd_epi16(u[1], k__cospi_p31_m01); + v[4] = _mm_madd_epi16(u[2], k__cospi_p05_p27); + v[5] = _mm_madd_epi16(u[3], k__cospi_p05_p27); + v[6] = _mm_madd_epi16(u[2], k__cospi_p27_m05); + v[7] = _mm_madd_epi16(u[3], k__cospi_p27_m05); + v[8] = _mm_madd_epi16(u[4], k__cospi_p09_p23); + v[9] = _mm_madd_epi16(u[5], k__cospi_p09_p23); + v[10] = _mm_madd_epi16(u[4], k__cospi_p23_m09); + v[11] = _mm_madd_epi16(u[5], k__cospi_p23_m09); + v[12] = _mm_madd_epi16(u[6], k__cospi_p13_p19); + v[13] = _mm_madd_epi16(u[7], k__cospi_p13_p19); + v[14] = _mm_madd_epi16(u[6], k__cospi_p19_m13); + v[15] = _mm_madd_epi16(u[7], k__cospi_p19_m13); + v[16] = _mm_madd_epi16(u[8], k__cospi_p17_p15); + v[17] = _mm_madd_epi16(u[9], k__cospi_p17_p15); + v[18] = _mm_madd_epi16(u[8], k__cospi_p15_m17); + v[19] = _mm_madd_epi16(u[9], k__cospi_p15_m17); + v[20] = _mm_madd_epi16(u[10], k__cospi_p21_p11); + v[21] = _mm_madd_epi16(u[11], k__cospi_p21_p11); + v[22] = _mm_madd_epi16(u[10], k__cospi_p11_m21); + v[23] = _mm_madd_epi16(u[11], k__cospi_p11_m21); + v[24] = _mm_madd_epi16(u[12], k__cospi_p25_p07); + v[25] = _mm_madd_epi16(u[13], k__cospi_p25_p07); + v[26] = _mm_madd_epi16(u[12], k__cospi_p07_m25); + v[27] = _mm_madd_epi16(u[13], k__cospi_p07_m25); + v[28] = _mm_madd_epi16(u[14], k__cospi_p29_p03); + v[29] = _mm_madd_epi16(u[15], k__cospi_p29_p03); + v[30] = _mm_madd_epi16(u[14], k__cospi_p03_m29); + v[31] = _mm_madd_epi16(u[15], k__cospi_p03_m29); + + u[0] = _mm_add_epi32(v[0], v[16]); + u[1] = _mm_add_epi32(v[1], v[17]); + u[2] = _mm_add_epi32(v[2], v[18]); + u[3] = _mm_add_epi32(v[3], v[19]); + u[4] = _mm_add_epi32(v[4], v[20]); + u[5] = _mm_add_epi32(v[5], v[21]); + u[6] = _mm_add_epi32(v[6], v[22]); + u[7] = _mm_add_epi32(v[7], v[23]); + u[8] = _mm_add_epi32(v[8], v[24]); + u[9] = _mm_add_epi32(v[9], v[25]); + u[10] = _mm_add_epi32(v[10], v[26]); + u[11] = _mm_add_epi32(v[11], v[27]); + u[12] = _mm_add_epi32(v[12], v[28]); + u[13] = _mm_add_epi32(v[13], v[29]); + u[14] = _mm_add_epi32(v[14], v[30]); + u[15] = _mm_add_epi32(v[15], v[31]); + u[16] = _mm_sub_epi32(v[0], v[16]); + u[17] = _mm_sub_epi32(v[1], v[17]); + u[18] = _mm_sub_epi32(v[2], v[18]); + u[19] = _mm_sub_epi32(v[3], v[19]); + u[20] = _mm_sub_epi32(v[4], v[20]); + u[21] = _mm_sub_epi32(v[5], v[21]); + u[22] = _mm_sub_epi32(v[6], v[22]); + u[23] = _mm_sub_epi32(v[7], v[23]); + u[24] = _mm_sub_epi32(v[8], v[24]); + u[25] = _mm_sub_epi32(v[9], v[25]); + u[26] = _mm_sub_epi32(v[10], v[26]); + u[27] = _mm_sub_epi32(v[11], v[27]); + u[28] = _mm_sub_epi32(v[12], v[28]); + u[29] = _mm_sub_epi32(v[13], v[29]); + u[30] = _mm_sub_epi32(v[14], v[30]); + u[31] = _mm_sub_epi32(v[15], v[31]); + + v[0] = _mm_add_epi32(u[0], k__DCT_CONST_ROUNDING); + v[1] = _mm_add_epi32(u[1], k__DCT_CONST_ROUNDING); + v[2] = _mm_add_epi32(u[2], k__DCT_CONST_ROUNDING); + v[3] = _mm_add_epi32(u[3], k__DCT_CONST_ROUNDING); + v[4] = _mm_add_epi32(u[4], k__DCT_CONST_ROUNDING); + v[5] = _mm_add_epi32(u[5], k__DCT_CONST_ROUNDING); + v[6] = _mm_add_epi32(u[6], k__DCT_CONST_ROUNDING); + v[7] = _mm_add_epi32(u[7], k__DCT_CONST_ROUNDING); + v[8] = _mm_add_epi32(u[8], k__DCT_CONST_ROUNDING); + v[9] = _mm_add_epi32(u[9], k__DCT_CONST_ROUNDING); + v[10] = _mm_add_epi32(u[10], k__DCT_CONST_ROUNDING); + v[11] = _mm_add_epi32(u[11], k__DCT_CONST_ROUNDING); + v[12] = _mm_add_epi32(u[12], k__DCT_CONST_ROUNDING); + v[13] = _mm_add_epi32(u[13], k__DCT_CONST_ROUNDING); + v[14] = _mm_add_epi32(u[14], k__DCT_CONST_ROUNDING); + v[15] = _mm_add_epi32(u[15], k__DCT_CONST_ROUNDING); + v[16] = _mm_add_epi32(u[16], k__DCT_CONST_ROUNDING); + v[17] = _mm_add_epi32(u[17], k__DCT_CONST_ROUNDING); + v[18] = _mm_add_epi32(u[18], k__DCT_CONST_ROUNDING); + v[19] = _mm_add_epi32(u[19], k__DCT_CONST_ROUNDING); + v[20] = _mm_add_epi32(u[20], k__DCT_CONST_ROUNDING); + v[21] = _mm_add_epi32(u[21], k__DCT_CONST_ROUNDING); + v[22] = _mm_add_epi32(u[22], k__DCT_CONST_ROUNDING); + v[23] = _mm_add_epi32(u[23], k__DCT_CONST_ROUNDING); + v[24] = _mm_add_epi32(u[24], k__DCT_CONST_ROUNDING); + v[25] = _mm_add_epi32(u[25], k__DCT_CONST_ROUNDING); + v[26] = _mm_add_epi32(u[26], k__DCT_CONST_ROUNDING); + v[27] = _mm_add_epi32(u[27], k__DCT_CONST_ROUNDING); + v[28] = _mm_add_epi32(u[28], k__DCT_CONST_ROUNDING); + v[29] = _mm_add_epi32(u[29], k__DCT_CONST_ROUNDING); + v[30] = _mm_add_epi32(u[30], k__DCT_CONST_ROUNDING); + v[31] = _mm_add_epi32(u[31], k__DCT_CONST_ROUNDING); + + u[0] = _mm_srai_epi32(v[0], DCT_CONST_BITS); + u[1] = _mm_srai_epi32(v[1], DCT_CONST_BITS); + u[2] = _mm_srai_epi32(v[2], DCT_CONST_BITS); + u[3] = _mm_srai_epi32(v[3], DCT_CONST_BITS); + u[4] = _mm_srai_epi32(v[4], DCT_CONST_BITS); + u[5] = _mm_srai_epi32(v[5], DCT_CONST_BITS); + u[6] = _mm_srai_epi32(v[6], DCT_CONST_BITS); + u[7] = _mm_srai_epi32(v[7], DCT_CONST_BITS); + u[8] = _mm_srai_epi32(v[8], DCT_CONST_BITS); + u[9] = _mm_srai_epi32(v[9], DCT_CONST_BITS); + u[10] = _mm_srai_epi32(v[10], DCT_CONST_BITS); + u[11] = _mm_srai_epi32(v[11], DCT_CONST_BITS); + u[12] = _mm_srai_epi32(v[12], DCT_CONST_BITS); + u[13] = _mm_srai_epi32(v[13], DCT_CONST_BITS); + u[14] = _mm_srai_epi32(v[14], DCT_CONST_BITS); + u[15] = _mm_srai_epi32(v[15], DCT_CONST_BITS); + u[16] = _mm_srai_epi32(v[16], DCT_CONST_BITS); + u[17] = _mm_srai_epi32(v[17], DCT_CONST_BITS); + u[18] = _mm_srai_epi32(v[18], DCT_CONST_BITS); + u[19] = _mm_srai_epi32(v[19], DCT_CONST_BITS); + u[20] = _mm_srai_epi32(v[20], DCT_CONST_BITS); + u[21] = _mm_srai_epi32(v[21], DCT_CONST_BITS); + u[22] = _mm_srai_epi32(v[22], DCT_CONST_BITS); + u[23] = _mm_srai_epi32(v[23], DCT_CONST_BITS); + u[24] = _mm_srai_epi32(v[24], DCT_CONST_BITS); + u[25] = _mm_srai_epi32(v[25], DCT_CONST_BITS); + u[26] = _mm_srai_epi32(v[26], DCT_CONST_BITS); + u[27] = _mm_srai_epi32(v[27], DCT_CONST_BITS); + u[28] = _mm_srai_epi32(v[28], DCT_CONST_BITS); + u[29] = _mm_srai_epi32(v[29], DCT_CONST_BITS); + u[30] = _mm_srai_epi32(v[30], DCT_CONST_BITS); + u[31] = _mm_srai_epi32(v[31], DCT_CONST_BITS); + + s[0] = _mm_packs_epi32(u[0], u[1]); + s[1] = _mm_packs_epi32(u[2], u[3]); + s[2] = _mm_packs_epi32(u[4], u[5]); + s[3] = _mm_packs_epi32(u[6], u[7]); + s[4] = _mm_packs_epi32(u[8], u[9]); + s[5] = _mm_packs_epi32(u[10], u[11]); + s[6] = _mm_packs_epi32(u[12], u[13]); + s[7] = _mm_packs_epi32(u[14], u[15]); + s[8] = _mm_packs_epi32(u[16], u[17]); + s[9] = _mm_packs_epi32(u[18], u[19]); + s[10] = _mm_packs_epi32(u[20], u[21]); + s[11] = _mm_packs_epi32(u[22], u[23]); + s[12] = _mm_packs_epi32(u[24], u[25]); + s[13] = _mm_packs_epi32(u[26], u[27]); + s[14] = _mm_packs_epi32(u[28], u[29]); + s[15] = _mm_packs_epi32(u[30], u[31]); + + // stage 2 + u[0] = _mm_unpacklo_epi16(s[8], s[9]); + u[1] = _mm_unpackhi_epi16(s[8], s[9]); + u[2] = _mm_unpacklo_epi16(s[10], s[11]); + u[3] = _mm_unpackhi_epi16(s[10], s[11]); + u[4] = _mm_unpacklo_epi16(s[12], s[13]); + u[5] = _mm_unpackhi_epi16(s[12], s[13]); + u[6] = _mm_unpacklo_epi16(s[14], s[15]); + u[7] = _mm_unpackhi_epi16(s[14], s[15]); + + v[0] = _mm_madd_epi16(u[0], k__cospi_p04_p28); + v[1] = _mm_madd_epi16(u[1], k__cospi_p04_p28); + v[2] = _mm_madd_epi16(u[0], k__cospi_p28_m04); + v[3] = _mm_madd_epi16(u[1], k__cospi_p28_m04); + v[4] = _mm_madd_epi16(u[2], k__cospi_p20_p12); + v[5] = _mm_madd_epi16(u[3], k__cospi_p20_p12); + v[6] = _mm_madd_epi16(u[2], k__cospi_p12_m20); + v[7] = _mm_madd_epi16(u[3], k__cospi_p12_m20); + v[8] = _mm_madd_epi16(u[4], k__cospi_m28_p04); + v[9] = _mm_madd_epi16(u[5], k__cospi_m28_p04); + v[10] = _mm_madd_epi16(u[4], k__cospi_p04_p28); + v[11] = _mm_madd_epi16(u[5], k__cospi_p04_p28); + v[12] = _mm_madd_epi16(u[6], k__cospi_m12_p20); + v[13] = _mm_madd_epi16(u[7], k__cospi_m12_p20); + v[14] = _mm_madd_epi16(u[6], k__cospi_p20_p12); + v[15] = _mm_madd_epi16(u[7], k__cospi_p20_p12); + + u[0] = _mm_add_epi32(v[0], v[8]); + u[1] = _mm_add_epi32(v[1], v[9]); + u[2] = _mm_add_epi32(v[2], v[10]); + u[3] = _mm_add_epi32(v[3], v[11]); + u[4] = _mm_add_epi32(v[4], v[12]); + u[5] = _mm_add_epi32(v[5], v[13]); + u[6] = _mm_add_epi32(v[6], v[14]); + u[7] = _mm_add_epi32(v[7], v[15]); + u[8] = _mm_sub_epi32(v[0], v[8]); + u[9] = _mm_sub_epi32(v[1], v[9]); + u[10] = _mm_sub_epi32(v[2], v[10]); + u[11] = _mm_sub_epi32(v[3], v[11]); + u[12] = _mm_sub_epi32(v[4], v[12]); + u[13] = _mm_sub_epi32(v[5], v[13]); + u[14] = _mm_sub_epi32(v[6], v[14]); + u[15] = _mm_sub_epi32(v[7], v[15]); + + v[0] = _mm_add_epi32(u[0], k__DCT_CONST_ROUNDING); + v[1] = _mm_add_epi32(u[1], k__DCT_CONST_ROUNDING); + v[2] = _mm_add_epi32(u[2], k__DCT_CONST_ROUNDING); + v[3] = _mm_add_epi32(u[3], k__DCT_CONST_ROUNDING); + v[4] = _mm_add_epi32(u[4], k__DCT_CONST_ROUNDING); + v[5] = _mm_add_epi32(u[5], k__DCT_CONST_ROUNDING); + v[6] = _mm_add_epi32(u[6], k__DCT_CONST_ROUNDING); + v[7] = _mm_add_epi32(u[7], k__DCT_CONST_ROUNDING); + v[8] = _mm_add_epi32(u[8], k__DCT_CONST_ROUNDING); + v[9] = _mm_add_epi32(u[9], k__DCT_CONST_ROUNDING); + v[10] = _mm_add_epi32(u[10], k__DCT_CONST_ROUNDING); + v[11] = _mm_add_epi32(u[11], k__DCT_CONST_ROUNDING); + v[12] = _mm_add_epi32(u[12], k__DCT_CONST_ROUNDING); + v[13] = _mm_add_epi32(u[13], k__DCT_CONST_ROUNDING); + v[14] = _mm_add_epi32(u[14], k__DCT_CONST_ROUNDING); + v[15] = _mm_add_epi32(u[15], k__DCT_CONST_ROUNDING); + + u[0] = _mm_srai_epi32(v[0], DCT_CONST_BITS); + u[1] = _mm_srai_epi32(v[1], DCT_CONST_BITS); + u[2] = _mm_srai_epi32(v[2], DCT_CONST_BITS); + u[3] = _mm_srai_epi32(v[3], DCT_CONST_BITS); + u[4] = _mm_srai_epi32(v[4], DCT_CONST_BITS); + u[5] = _mm_srai_epi32(v[5], DCT_CONST_BITS); + u[6] = _mm_srai_epi32(v[6], DCT_CONST_BITS); + u[7] = _mm_srai_epi32(v[7], DCT_CONST_BITS); + u[8] = _mm_srai_epi32(v[8], DCT_CONST_BITS); + u[9] = _mm_srai_epi32(v[9], DCT_CONST_BITS); + u[10] = _mm_srai_epi32(v[10], DCT_CONST_BITS); + u[11] = _mm_srai_epi32(v[11], DCT_CONST_BITS); + u[12] = _mm_srai_epi32(v[12], DCT_CONST_BITS); + u[13] = _mm_srai_epi32(v[13], DCT_CONST_BITS); + u[14] = _mm_srai_epi32(v[14], DCT_CONST_BITS); + u[15] = _mm_srai_epi32(v[15], DCT_CONST_BITS); + + x[0] = _mm_add_epi16(s[0], s[4]); + x[1] = _mm_add_epi16(s[1], s[5]); + x[2] = _mm_add_epi16(s[2], s[6]); + x[3] = _mm_add_epi16(s[3], s[7]); + x[4] = _mm_sub_epi16(s[0], s[4]); + x[5] = _mm_sub_epi16(s[1], s[5]); + x[6] = _mm_sub_epi16(s[2], s[6]); + x[7] = _mm_sub_epi16(s[3], s[7]); + x[8] = _mm_packs_epi32(u[0], u[1]); + x[9] = _mm_packs_epi32(u[2], u[3]); + x[10] = _mm_packs_epi32(u[4], u[5]); + x[11] = _mm_packs_epi32(u[6], u[7]); + x[12] = _mm_packs_epi32(u[8], u[9]); + x[13] = _mm_packs_epi32(u[10], u[11]); + x[14] = _mm_packs_epi32(u[12], u[13]); + x[15] = _mm_packs_epi32(u[14], u[15]); + + // stage 3 + u[0] = _mm_unpacklo_epi16(x[4], x[5]); + u[1] = _mm_unpackhi_epi16(x[4], x[5]); + u[2] = _mm_unpacklo_epi16(x[6], x[7]); + u[3] = _mm_unpackhi_epi16(x[6], x[7]); + u[4] = _mm_unpacklo_epi16(x[12], x[13]); + u[5] = _mm_unpackhi_epi16(x[12], x[13]); + u[6] = _mm_unpacklo_epi16(x[14], x[15]); + u[7] = _mm_unpackhi_epi16(x[14], x[15]); + + v[0] = _mm_madd_epi16(u[0], k__cospi_p08_p24); + v[1] = _mm_madd_epi16(u[1], k__cospi_p08_p24); + v[2] = _mm_madd_epi16(u[0], k__cospi_p24_m08); + v[3] = _mm_madd_epi16(u[1], k__cospi_p24_m08); + v[4] = _mm_madd_epi16(u[2], k__cospi_m24_p08); + v[5] = _mm_madd_epi16(u[3], k__cospi_m24_p08); + v[6] = _mm_madd_epi16(u[2], k__cospi_p08_p24); + v[7] = _mm_madd_epi16(u[3], k__cospi_p08_p24); + v[8] = _mm_madd_epi16(u[4], k__cospi_p08_p24); + v[9] = _mm_madd_epi16(u[5], k__cospi_p08_p24); + v[10] = _mm_madd_epi16(u[4], k__cospi_p24_m08); + v[11] = _mm_madd_epi16(u[5], k__cospi_p24_m08); + v[12] = _mm_madd_epi16(u[6], k__cospi_m24_p08); + v[13] = _mm_madd_epi16(u[7], k__cospi_m24_p08); + v[14] = _mm_madd_epi16(u[6], k__cospi_p08_p24); + v[15] = _mm_madd_epi16(u[7], k__cospi_p08_p24); + + u[0] = _mm_add_epi32(v[0], v[4]); + u[1] = _mm_add_epi32(v[1], v[5]); + u[2] = _mm_add_epi32(v[2], v[6]); + u[3] = _mm_add_epi32(v[3], v[7]); + u[4] = _mm_sub_epi32(v[0], v[4]); + u[5] = _mm_sub_epi32(v[1], v[5]); + u[6] = _mm_sub_epi32(v[2], v[6]); + u[7] = _mm_sub_epi32(v[3], v[7]); + u[8] = _mm_add_epi32(v[8], v[12]); + u[9] = _mm_add_epi32(v[9], v[13]); + u[10] = _mm_add_epi32(v[10], v[14]); + u[11] = _mm_add_epi32(v[11], v[15]); + u[12] = _mm_sub_epi32(v[8], v[12]); + u[13] = _mm_sub_epi32(v[9], v[13]); + u[14] = _mm_sub_epi32(v[10], v[14]); + u[15] = _mm_sub_epi32(v[11], v[15]); + + u[0] = _mm_add_epi32(u[0], k__DCT_CONST_ROUNDING); + u[1] = _mm_add_epi32(u[1], k__DCT_CONST_ROUNDING); + u[2] = _mm_add_epi32(u[2], k__DCT_CONST_ROUNDING); + u[3] = _mm_add_epi32(u[3], k__DCT_CONST_ROUNDING); + u[4] = _mm_add_epi32(u[4], k__DCT_CONST_ROUNDING); + u[5] = _mm_add_epi32(u[5], k__DCT_CONST_ROUNDING); + u[6] = _mm_add_epi32(u[6], k__DCT_CONST_ROUNDING); + u[7] = _mm_add_epi32(u[7], k__DCT_CONST_ROUNDING); + u[8] = _mm_add_epi32(u[8], k__DCT_CONST_ROUNDING); + u[9] = _mm_add_epi32(u[9], k__DCT_CONST_ROUNDING); + u[10] = _mm_add_epi32(u[10], k__DCT_CONST_ROUNDING); + u[11] = _mm_add_epi32(u[11], k__DCT_CONST_ROUNDING); + u[12] = _mm_add_epi32(u[12], k__DCT_CONST_ROUNDING); + u[13] = _mm_add_epi32(u[13], k__DCT_CONST_ROUNDING); + u[14] = _mm_add_epi32(u[14], k__DCT_CONST_ROUNDING); + u[15] = _mm_add_epi32(u[15], k__DCT_CONST_ROUNDING); + + v[0] = _mm_srai_epi32(u[0], DCT_CONST_BITS); + v[1] = _mm_srai_epi32(u[1], DCT_CONST_BITS); + v[2] = _mm_srai_epi32(u[2], DCT_CONST_BITS); + v[3] = _mm_srai_epi32(u[3], DCT_CONST_BITS); + v[4] = _mm_srai_epi32(u[4], DCT_CONST_BITS); + v[5] = _mm_srai_epi32(u[5], DCT_CONST_BITS); + v[6] = _mm_srai_epi32(u[6], DCT_CONST_BITS); + v[7] = _mm_srai_epi32(u[7], DCT_CONST_BITS); + v[8] = _mm_srai_epi32(u[8], DCT_CONST_BITS); + v[9] = _mm_srai_epi32(u[9], DCT_CONST_BITS); + v[10] = _mm_srai_epi32(u[10], DCT_CONST_BITS); + v[11] = _mm_srai_epi32(u[11], DCT_CONST_BITS); + v[12] = _mm_srai_epi32(u[12], DCT_CONST_BITS); + v[13] = _mm_srai_epi32(u[13], DCT_CONST_BITS); + v[14] = _mm_srai_epi32(u[14], DCT_CONST_BITS); + v[15] = _mm_srai_epi32(u[15], DCT_CONST_BITS); + + s[0] = _mm_add_epi16(x[0], x[2]); + s[1] = _mm_add_epi16(x[1], x[3]); + s[2] = _mm_sub_epi16(x[0], x[2]); + s[3] = _mm_sub_epi16(x[1], x[3]); + s[4] = _mm_packs_epi32(v[0], v[1]); + s[5] = _mm_packs_epi32(v[2], v[3]); + s[6] = _mm_packs_epi32(v[4], v[5]); + s[7] = _mm_packs_epi32(v[6], v[7]); + s[8] = _mm_add_epi16(x[8], x[10]); + s[9] = _mm_add_epi16(x[9], x[11]); + s[10] = _mm_sub_epi16(x[8], x[10]); + s[11] = _mm_sub_epi16(x[9], x[11]); + s[12] = _mm_packs_epi32(v[8], v[9]); + s[13] = _mm_packs_epi32(v[10], v[11]); + s[14] = _mm_packs_epi32(v[12], v[13]); + s[15] = _mm_packs_epi32(v[14], v[15]); + + // stage 4 + u[0] = _mm_unpacklo_epi16(s[2], s[3]); + u[1] = _mm_unpackhi_epi16(s[2], s[3]); + u[2] = _mm_unpacklo_epi16(s[6], s[7]); + u[3] = _mm_unpackhi_epi16(s[6], s[7]); + u[4] = _mm_unpacklo_epi16(s[10], s[11]); + u[5] = _mm_unpackhi_epi16(s[10], s[11]); + u[6] = _mm_unpacklo_epi16(s[14], s[15]); + u[7] = _mm_unpackhi_epi16(s[14], s[15]); + + v[0] = _mm_madd_epi16(u[0], k__cospi_m16_m16); + v[1] = _mm_madd_epi16(u[1], k__cospi_m16_m16); + v[2] = _mm_madd_epi16(u[0], k__cospi_p16_m16); + v[3] = _mm_madd_epi16(u[1], k__cospi_p16_m16); + v[4] = _mm_madd_epi16(u[2], k__cospi_p16_p16); + v[5] = _mm_madd_epi16(u[3], k__cospi_p16_p16); + v[6] = _mm_madd_epi16(u[2], k__cospi_m16_p16); + v[7] = _mm_madd_epi16(u[3], k__cospi_m16_p16); + v[8] = _mm_madd_epi16(u[4], k__cospi_p16_p16); + v[9] = _mm_madd_epi16(u[5], k__cospi_p16_p16); + v[10] = _mm_madd_epi16(u[4], k__cospi_m16_p16); + v[11] = _mm_madd_epi16(u[5], k__cospi_m16_p16); + v[12] = _mm_madd_epi16(u[6], k__cospi_m16_m16); + v[13] = _mm_madd_epi16(u[7], k__cospi_m16_m16); + v[14] = _mm_madd_epi16(u[6], k__cospi_p16_m16); + v[15] = _mm_madd_epi16(u[7], k__cospi_p16_m16); + + u[0] = _mm_add_epi32(v[0], k__DCT_CONST_ROUNDING); + u[1] = _mm_add_epi32(v[1], k__DCT_CONST_ROUNDING); + u[2] = _mm_add_epi32(v[2], k__DCT_CONST_ROUNDING); + u[3] = _mm_add_epi32(v[3], k__DCT_CONST_ROUNDING); + u[4] = _mm_add_epi32(v[4], k__DCT_CONST_ROUNDING); + u[5] = _mm_add_epi32(v[5], k__DCT_CONST_ROUNDING); + u[6] = _mm_add_epi32(v[6], k__DCT_CONST_ROUNDING); + u[7] = _mm_add_epi32(v[7], k__DCT_CONST_ROUNDING); + u[8] = _mm_add_epi32(v[8], k__DCT_CONST_ROUNDING); + u[9] = _mm_add_epi32(v[9], k__DCT_CONST_ROUNDING); + u[10] = _mm_add_epi32(v[10], k__DCT_CONST_ROUNDING); + u[11] = _mm_add_epi32(v[11], k__DCT_CONST_ROUNDING); + u[12] = _mm_add_epi32(v[12], k__DCT_CONST_ROUNDING); + u[13] = _mm_add_epi32(v[13], k__DCT_CONST_ROUNDING); + u[14] = _mm_add_epi32(v[14], k__DCT_CONST_ROUNDING); + u[15] = _mm_add_epi32(v[15], k__DCT_CONST_ROUNDING); + + v[0] = _mm_srai_epi32(u[0], DCT_CONST_BITS); + v[1] = _mm_srai_epi32(u[1], DCT_CONST_BITS); + v[2] = _mm_srai_epi32(u[2], DCT_CONST_BITS); + v[3] = _mm_srai_epi32(u[3], DCT_CONST_BITS); + v[4] = _mm_srai_epi32(u[4], DCT_CONST_BITS); + v[5] = _mm_srai_epi32(u[5], DCT_CONST_BITS); + v[6] = _mm_srai_epi32(u[6], DCT_CONST_BITS); + v[7] = _mm_srai_epi32(u[7], DCT_CONST_BITS); + v[8] = _mm_srai_epi32(u[8], DCT_CONST_BITS); + v[9] = _mm_srai_epi32(u[9], DCT_CONST_BITS); + v[10] = _mm_srai_epi32(u[10], DCT_CONST_BITS); + v[11] = _mm_srai_epi32(u[11], DCT_CONST_BITS); + v[12] = _mm_srai_epi32(u[12], DCT_CONST_BITS); + v[13] = _mm_srai_epi32(u[13], DCT_CONST_BITS); + v[14] = _mm_srai_epi32(u[14], DCT_CONST_BITS); + v[15] = _mm_srai_epi32(u[15], DCT_CONST_BITS); + + in[0] = s[0]; + in[1] = _mm_sub_epi16(kZero, s[8]); + in[2] = s[12]; + in[3] = _mm_sub_epi16(kZero, s[4]); + in[4] = _mm_packs_epi32(v[4], v[5]); + in[5] = _mm_packs_epi32(v[12], v[13]); + in[6] = _mm_packs_epi32(v[8], v[9]); + in[7] = _mm_packs_epi32(v[0], v[1]); + in[8] = _mm_packs_epi32(v[2], v[3]); + in[9] = _mm_packs_epi32(v[10], v[11]); + in[10] = _mm_packs_epi32(v[14], v[15]); + in[11] = _mm_packs_epi32(v[6], v[7]); + in[12] = s[5]; + in[13] = _mm_sub_epi16(kZero, s[13]); + in[14] = s[9]; + in[15] = _mm_sub_epi16(kZero, s[1]); +} + +static void idct16_8col(__m128i *in) { + const __m128i k__cospi_p30_m02 = pair_set_epi16(cospi_30_64, -cospi_2_64); + const __m128i k__cospi_p02_p30 = pair_set_epi16(cospi_2_64, cospi_30_64); + const __m128i k__cospi_p14_m18 = pair_set_epi16(cospi_14_64, -cospi_18_64); + const __m128i k__cospi_p18_p14 = pair_set_epi16(cospi_18_64, cospi_14_64); + const __m128i k__cospi_p22_m10 = pair_set_epi16(cospi_22_64, -cospi_10_64); + const __m128i k__cospi_p10_p22 = pair_set_epi16(cospi_10_64, cospi_22_64); + const __m128i k__cospi_p06_m26 = pair_set_epi16(cospi_6_64, -cospi_26_64); + const __m128i k__cospi_p26_p06 = pair_set_epi16(cospi_26_64, cospi_6_64); + const __m128i k__cospi_p28_m04 = pair_set_epi16(cospi_28_64, -cospi_4_64); + const __m128i k__cospi_p04_p28 = pair_set_epi16(cospi_4_64, cospi_28_64); + const __m128i k__cospi_p12_m20 = pair_set_epi16(cospi_12_64, -cospi_20_64); + const __m128i k__cospi_p20_p12 = pair_set_epi16(cospi_20_64, cospi_12_64); + const __m128i k__cospi_p16_p16 = _mm_set1_epi16((int16_t)cospi_16_64); + const __m128i k__cospi_p16_m16 = pair_set_epi16(cospi_16_64, -cospi_16_64); + const __m128i k__cospi_p24_m08 = pair_set_epi16(cospi_24_64, -cospi_8_64); + const __m128i k__cospi_p08_p24 = pair_set_epi16(cospi_8_64, cospi_24_64); + const __m128i k__cospi_m08_p24 = pair_set_epi16(-cospi_8_64, cospi_24_64); + const __m128i k__cospi_p24_p08 = pair_set_epi16(cospi_24_64, cospi_8_64); + const __m128i k__cospi_m24_m08 = pair_set_epi16(-cospi_24_64, -cospi_8_64); + const __m128i k__cospi_m16_p16 = pair_set_epi16(-cospi_16_64, cospi_16_64); + const __m128i k__DCT_CONST_ROUNDING = _mm_set1_epi32(DCT_CONST_ROUNDING); + __m128i v[16], u[16], s[16], t[16]; + + // stage 1 + s[0] = in[0]; + s[1] = in[8]; + s[2] = in[4]; + s[3] = in[12]; + s[4] = in[2]; + s[5] = in[10]; + s[6] = in[6]; + s[7] = in[14]; + s[8] = in[1]; + s[9] = in[9]; + s[10] = in[5]; + s[11] = in[13]; + s[12] = in[3]; + s[13] = in[11]; + s[14] = in[7]; + s[15] = in[15]; + + // stage 2 + u[0] = _mm_unpacklo_epi16(s[8], s[15]); + u[1] = _mm_unpackhi_epi16(s[8], s[15]); + u[2] = _mm_unpacklo_epi16(s[9], s[14]); + u[3] = _mm_unpackhi_epi16(s[9], s[14]); + u[4] = _mm_unpacklo_epi16(s[10], s[13]); + u[5] = _mm_unpackhi_epi16(s[10], s[13]); + u[6] = _mm_unpacklo_epi16(s[11], s[12]); + u[7] = _mm_unpackhi_epi16(s[11], s[12]); + + v[0] = _mm_madd_epi16(u[0], k__cospi_p30_m02); + v[1] = _mm_madd_epi16(u[1], k__cospi_p30_m02); + v[2] = _mm_madd_epi16(u[0], k__cospi_p02_p30); + v[3] = _mm_madd_epi16(u[1], k__cospi_p02_p30); + v[4] = _mm_madd_epi16(u[2], k__cospi_p14_m18); + v[5] = _mm_madd_epi16(u[3], k__cospi_p14_m18); + v[6] = _mm_madd_epi16(u[2], k__cospi_p18_p14); + v[7] = _mm_madd_epi16(u[3], k__cospi_p18_p14); + v[8] = _mm_madd_epi16(u[4], k__cospi_p22_m10); + v[9] = _mm_madd_epi16(u[5], k__cospi_p22_m10); + v[10] = _mm_madd_epi16(u[4], k__cospi_p10_p22); + v[11] = _mm_madd_epi16(u[5], k__cospi_p10_p22); + v[12] = _mm_madd_epi16(u[6], k__cospi_p06_m26); + v[13] = _mm_madd_epi16(u[7], k__cospi_p06_m26); + v[14] = _mm_madd_epi16(u[6], k__cospi_p26_p06); + v[15] = _mm_madd_epi16(u[7], k__cospi_p26_p06); + + u[0] = _mm_add_epi32(v[0], k__DCT_CONST_ROUNDING); + u[1] = _mm_add_epi32(v[1], k__DCT_CONST_ROUNDING); + u[2] = _mm_add_epi32(v[2], k__DCT_CONST_ROUNDING); + u[3] = _mm_add_epi32(v[3], k__DCT_CONST_ROUNDING); + u[4] = _mm_add_epi32(v[4], k__DCT_CONST_ROUNDING); + u[5] = _mm_add_epi32(v[5], k__DCT_CONST_ROUNDING); + u[6] = _mm_add_epi32(v[6], k__DCT_CONST_ROUNDING); + u[7] = _mm_add_epi32(v[7], k__DCT_CONST_ROUNDING); + u[8] = _mm_add_epi32(v[8], k__DCT_CONST_ROUNDING); + u[9] = _mm_add_epi32(v[9], k__DCT_CONST_ROUNDING); + u[10] = _mm_add_epi32(v[10], k__DCT_CONST_ROUNDING); + u[11] = _mm_add_epi32(v[11], k__DCT_CONST_ROUNDING); + u[12] = _mm_add_epi32(v[12], k__DCT_CONST_ROUNDING); + u[13] = _mm_add_epi32(v[13], k__DCT_CONST_ROUNDING); + u[14] = _mm_add_epi32(v[14], k__DCT_CONST_ROUNDING); + u[15] = _mm_add_epi32(v[15], k__DCT_CONST_ROUNDING); + + u[0] = _mm_srai_epi32(u[0], DCT_CONST_BITS); + u[1] = _mm_srai_epi32(u[1], DCT_CONST_BITS); + u[2] = _mm_srai_epi32(u[2], DCT_CONST_BITS); + u[3] = _mm_srai_epi32(u[3], DCT_CONST_BITS); + u[4] = _mm_srai_epi32(u[4], DCT_CONST_BITS); + u[5] = _mm_srai_epi32(u[5], DCT_CONST_BITS); + u[6] = _mm_srai_epi32(u[6], DCT_CONST_BITS); + u[7] = _mm_srai_epi32(u[7], DCT_CONST_BITS); + u[8] = _mm_srai_epi32(u[8], DCT_CONST_BITS); + u[9] = _mm_srai_epi32(u[9], DCT_CONST_BITS); + u[10] = _mm_srai_epi32(u[10], DCT_CONST_BITS); + u[11] = _mm_srai_epi32(u[11], DCT_CONST_BITS); + u[12] = _mm_srai_epi32(u[12], DCT_CONST_BITS); + u[13] = _mm_srai_epi32(u[13], DCT_CONST_BITS); + u[14] = _mm_srai_epi32(u[14], DCT_CONST_BITS); + u[15] = _mm_srai_epi32(u[15], DCT_CONST_BITS); + + s[8] = _mm_packs_epi32(u[0], u[1]); + s[15] = _mm_packs_epi32(u[2], u[3]); + s[9] = _mm_packs_epi32(u[4], u[5]); + s[14] = _mm_packs_epi32(u[6], u[7]); + s[10] = _mm_packs_epi32(u[8], u[9]); + s[13] = _mm_packs_epi32(u[10], u[11]); + s[11] = _mm_packs_epi32(u[12], u[13]); + s[12] = _mm_packs_epi32(u[14], u[15]); + + // stage 3 + t[0] = s[0]; + t[1] = s[1]; + t[2] = s[2]; + t[3] = s[3]; + u[0] = _mm_unpacklo_epi16(s[4], s[7]); + u[1] = _mm_unpackhi_epi16(s[4], s[7]); + u[2] = _mm_unpacklo_epi16(s[5], s[6]); + u[3] = _mm_unpackhi_epi16(s[5], s[6]); + + v[0] = _mm_madd_epi16(u[0], k__cospi_p28_m04); + v[1] = _mm_madd_epi16(u[1], k__cospi_p28_m04); + v[2] = _mm_madd_epi16(u[0], k__cospi_p04_p28); + v[3] = _mm_madd_epi16(u[1], k__cospi_p04_p28); + v[4] = _mm_madd_epi16(u[2], k__cospi_p12_m20); + v[5] = _mm_madd_epi16(u[3], k__cospi_p12_m20); + v[6] = _mm_madd_epi16(u[2], k__cospi_p20_p12); + v[7] = _mm_madd_epi16(u[3], k__cospi_p20_p12); + + u[0] = _mm_add_epi32(v[0], k__DCT_CONST_ROUNDING); + u[1] = _mm_add_epi32(v[1], k__DCT_CONST_ROUNDING); + u[2] = _mm_add_epi32(v[2], k__DCT_CONST_ROUNDING); + u[3] = _mm_add_epi32(v[3], k__DCT_CONST_ROUNDING); + u[4] = _mm_add_epi32(v[4], k__DCT_CONST_ROUNDING); + u[5] = _mm_add_epi32(v[5], k__DCT_CONST_ROUNDING); + u[6] = _mm_add_epi32(v[6], k__DCT_CONST_ROUNDING); + u[7] = _mm_add_epi32(v[7], k__DCT_CONST_ROUNDING); + + u[0] = _mm_srai_epi32(u[0], DCT_CONST_BITS); + u[1] = _mm_srai_epi32(u[1], DCT_CONST_BITS); + u[2] = _mm_srai_epi32(u[2], DCT_CONST_BITS); + u[3] = _mm_srai_epi32(u[3], DCT_CONST_BITS); + u[4] = _mm_srai_epi32(u[4], DCT_CONST_BITS); + u[5] = _mm_srai_epi32(u[5], DCT_CONST_BITS); + u[6] = _mm_srai_epi32(u[6], DCT_CONST_BITS); + u[7] = _mm_srai_epi32(u[7], DCT_CONST_BITS); + + t[4] = _mm_packs_epi32(u[0], u[1]); + t[7] = _mm_packs_epi32(u[2], u[3]); + t[5] = _mm_packs_epi32(u[4], u[5]); + t[6] = _mm_packs_epi32(u[6], u[7]); + t[8] = _mm_add_epi16(s[8], s[9]); + t[9] = _mm_sub_epi16(s[8], s[9]); + t[10] = _mm_sub_epi16(s[11], s[10]); + t[11] = _mm_add_epi16(s[10], s[11]); + t[12] = _mm_add_epi16(s[12], s[13]); + t[13] = _mm_sub_epi16(s[12], s[13]); + t[14] = _mm_sub_epi16(s[15], s[14]); + t[15] = _mm_add_epi16(s[14], s[15]); + + // stage 4 + u[0] = _mm_unpacklo_epi16(t[0], t[1]); + u[1] = _mm_unpackhi_epi16(t[0], t[1]); + u[2] = _mm_unpacklo_epi16(t[2], t[3]); + u[3] = _mm_unpackhi_epi16(t[2], t[3]); + u[4] = _mm_unpacklo_epi16(t[9], t[14]); + u[5] = _mm_unpackhi_epi16(t[9], t[14]); + u[6] = _mm_unpacklo_epi16(t[10], t[13]); + u[7] = _mm_unpackhi_epi16(t[10], t[13]); + + v[0] = _mm_madd_epi16(u[0], k__cospi_p16_p16); + v[1] = _mm_madd_epi16(u[1], k__cospi_p16_p16); + v[2] = _mm_madd_epi16(u[0], k__cospi_p16_m16); + v[3] = _mm_madd_epi16(u[1], k__cospi_p16_m16); + v[4] = _mm_madd_epi16(u[2], k__cospi_p24_m08); + v[5] = _mm_madd_epi16(u[3], k__cospi_p24_m08); + v[6] = _mm_madd_epi16(u[2], k__cospi_p08_p24); + v[7] = _mm_madd_epi16(u[3], k__cospi_p08_p24); + v[8] = _mm_madd_epi16(u[4], k__cospi_m08_p24); + v[9] = _mm_madd_epi16(u[5], k__cospi_m08_p24); + v[10] = _mm_madd_epi16(u[4], k__cospi_p24_p08); + v[11] = _mm_madd_epi16(u[5], k__cospi_p24_p08); + v[12] = _mm_madd_epi16(u[6], k__cospi_m24_m08); + v[13] = _mm_madd_epi16(u[7], k__cospi_m24_m08); + v[14] = _mm_madd_epi16(u[6], k__cospi_m08_p24); + v[15] = _mm_madd_epi16(u[7], k__cospi_m08_p24); + + u[0] = _mm_add_epi32(v[0], k__DCT_CONST_ROUNDING); + u[1] = _mm_add_epi32(v[1], k__DCT_CONST_ROUNDING); + u[2] = _mm_add_epi32(v[2], k__DCT_CONST_ROUNDING); + u[3] = _mm_add_epi32(v[3], k__DCT_CONST_ROUNDING); + u[4] = _mm_add_epi32(v[4], k__DCT_CONST_ROUNDING); + u[5] = _mm_add_epi32(v[5], k__DCT_CONST_ROUNDING); + u[6] = _mm_add_epi32(v[6], k__DCT_CONST_ROUNDING); + u[7] = _mm_add_epi32(v[7], k__DCT_CONST_ROUNDING); + u[8] = _mm_add_epi32(v[8], k__DCT_CONST_ROUNDING); + u[9] = _mm_add_epi32(v[9], k__DCT_CONST_ROUNDING); + u[10] = _mm_add_epi32(v[10], k__DCT_CONST_ROUNDING); + u[11] = _mm_add_epi32(v[11], k__DCT_CONST_ROUNDING); + u[12] = _mm_add_epi32(v[12], k__DCT_CONST_ROUNDING); + u[13] = _mm_add_epi32(v[13], k__DCT_CONST_ROUNDING); + u[14] = _mm_add_epi32(v[14], k__DCT_CONST_ROUNDING); + u[15] = _mm_add_epi32(v[15], k__DCT_CONST_ROUNDING); + + u[0] = _mm_srai_epi32(u[0], DCT_CONST_BITS); + u[1] = _mm_srai_epi32(u[1], DCT_CONST_BITS); + u[2] = _mm_srai_epi32(u[2], DCT_CONST_BITS); + u[3] = _mm_srai_epi32(u[3], DCT_CONST_BITS); + u[4] = _mm_srai_epi32(u[4], DCT_CONST_BITS); + u[5] = _mm_srai_epi32(u[5], DCT_CONST_BITS); + u[6] = _mm_srai_epi32(u[6], DCT_CONST_BITS); + u[7] = _mm_srai_epi32(u[7], DCT_CONST_BITS); + u[8] = _mm_srai_epi32(u[8], DCT_CONST_BITS); + u[9] = _mm_srai_epi32(u[9], DCT_CONST_BITS); + u[10] = _mm_srai_epi32(u[10], DCT_CONST_BITS); + u[11] = _mm_srai_epi32(u[11], DCT_CONST_BITS); + u[12] = _mm_srai_epi32(u[12], DCT_CONST_BITS); + u[13] = _mm_srai_epi32(u[13], DCT_CONST_BITS); + u[14] = _mm_srai_epi32(u[14], DCT_CONST_BITS); + u[15] = _mm_srai_epi32(u[15], DCT_CONST_BITS); + + s[0] = _mm_packs_epi32(u[0], u[1]); + s[1] = _mm_packs_epi32(u[2], u[3]); + s[2] = _mm_packs_epi32(u[4], u[5]); + s[3] = _mm_packs_epi32(u[6], u[7]); + s[4] = _mm_add_epi16(t[4], t[5]); + s[5] = _mm_sub_epi16(t[4], t[5]); + s[6] = _mm_sub_epi16(t[7], t[6]); + s[7] = _mm_add_epi16(t[6], t[7]); + s[8] = t[8]; + s[15] = t[15]; + s[9] = _mm_packs_epi32(u[8], u[9]); + s[14] = _mm_packs_epi32(u[10], u[11]); + s[10] = _mm_packs_epi32(u[12], u[13]); + s[13] = _mm_packs_epi32(u[14], u[15]); + s[11] = t[11]; + s[12] = t[12]; + + // stage 5 + t[0] = _mm_add_epi16(s[0], s[3]); + t[1] = _mm_add_epi16(s[1], s[2]); + t[2] = _mm_sub_epi16(s[1], s[2]); + t[3] = _mm_sub_epi16(s[0], s[3]); + t[4] = s[4]; + t[7] = s[7]; + + u[0] = _mm_unpacklo_epi16(s[5], s[6]); + u[1] = _mm_unpackhi_epi16(s[5], s[6]); + v[0] = _mm_madd_epi16(u[0], k__cospi_m16_p16); + v[1] = _mm_madd_epi16(u[1], k__cospi_m16_p16); + v[2] = _mm_madd_epi16(u[0], k__cospi_p16_p16); + v[3] = _mm_madd_epi16(u[1], k__cospi_p16_p16); + u[0] = _mm_add_epi32(v[0], k__DCT_CONST_ROUNDING); + u[1] = _mm_add_epi32(v[1], k__DCT_CONST_ROUNDING); + u[2] = _mm_add_epi32(v[2], k__DCT_CONST_ROUNDING); + u[3] = _mm_add_epi32(v[3], k__DCT_CONST_ROUNDING); + u[0] = _mm_srai_epi32(u[0], DCT_CONST_BITS); + u[1] = _mm_srai_epi32(u[1], DCT_CONST_BITS); + u[2] = _mm_srai_epi32(u[2], DCT_CONST_BITS); + u[3] = _mm_srai_epi32(u[3], DCT_CONST_BITS); + t[5] = _mm_packs_epi32(u[0], u[1]); + t[6] = _mm_packs_epi32(u[2], u[3]); + + t[8] = _mm_add_epi16(s[8], s[11]); + t[9] = _mm_add_epi16(s[9], s[10]); + t[10] = _mm_sub_epi16(s[9], s[10]); + t[11] = _mm_sub_epi16(s[8], s[11]); + t[12] = _mm_sub_epi16(s[15], s[12]); + t[13] = _mm_sub_epi16(s[14], s[13]); + t[14] = _mm_add_epi16(s[13], s[14]); + t[15] = _mm_add_epi16(s[12], s[15]); + + // stage 6 + s[0] = _mm_add_epi16(t[0], t[7]); + s[1] = _mm_add_epi16(t[1], t[6]); + s[2] = _mm_add_epi16(t[2], t[5]); + s[3] = _mm_add_epi16(t[3], t[4]); + s[4] = _mm_sub_epi16(t[3], t[4]); + s[5] = _mm_sub_epi16(t[2], t[5]); + s[6] = _mm_sub_epi16(t[1], t[6]); + s[7] = _mm_sub_epi16(t[0], t[7]); + s[8] = t[8]; + s[9] = t[9]; + + u[0] = _mm_unpacklo_epi16(t[10], t[13]); + u[1] = _mm_unpackhi_epi16(t[10], t[13]); + u[2] = _mm_unpacklo_epi16(t[11], t[12]); + u[3] = _mm_unpackhi_epi16(t[11], t[12]); + + v[0] = _mm_madd_epi16(u[0], k__cospi_m16_p16); + v[1] = _mm_madd_epi16(u[1], k__cospi_m16_p16); + v[2] = _mm_madd_epi16(u[0], k__cospi_p16_p16); + v[3] = _mm_madd_epi16(u[1], k__cospi_p16_p16); + v[4] = _mm_madd_epi16(u[2], k__cospi_m16_p16); + v[5] = _mm_madd_epi16(u[3], k__cospi_m16_p16); + v[6] = _mm_madd_epi16(u[2], k__cospi_p16_p16); + v[7] = _mm_madd_epi16(u[3], k__cospi_p16_p16); + + u[0] = _mm_add_epi32(v[0], k__DCT_CONST_ROUNDING); + u[1] = _mm_add_epi32(v[1], k__DCT_CONST_ROUNDING); + u[2] = _mm_add_epi32(v[2], k__DCT_CONST_ROUNDING); + u[3] = _mm_add_epi32(v[3], k__DCT_CONST_ROUNDING); + u[4] = _mm_add_epi32(v[4], k__DCT_CONST_ROUNDING); + u[5] = _mm_add_epi32(v[5], k__DCT_CONST_ROUNDING); + u[6] = _mm_add_epi32(v[6], k__DCT_CONST_ROUNDING); + u[7] = _mm_add_epi32(v[7], k__DCT_CONST_ROUNDING); + + u[0] = _mm_srai_epi32(u[0], DCT_CONST_BITS); + u[1] = _mm_srai_epi32(u[1], DCT_CONST_BITS); + u[2] = _mm_srai_epi32(u[2], DCT_CONST_BITS); + u[3] = _mm_srai_epi32(u[3], DCT_CONST_BITS); + u[4] = _mm_srai_epi32(u[4], DCT_CONST_BITS); + u[5] = _mm_srai_epi32(u[5], DCT_CONST_BITS); + u[6] = _mm_srai_epi32(u[6], DCT_CONST_BITS); + u[7] = _mm_srai_epi32(u[7], DCT_CONST_BITS); + + s[10] = _mm_packs_epi32(u[0], u[1]); + s[13] = _mm_packs_epi32(u[2], u[3]); + s[11] = _mm_packs_epi32(u[4], u[5]); + s[12] = _mm_packs_epi32(u[6], u[7]); + s[14] = t[14]; + s[15] = t[15]; + + // stage 7 + in[0] = _mm_add_epi16(s[0], s[15]); + in[1] = _mm_add_epi16(s[1], s[14]); + in[2] = _mm_add_epi16(s[2], s[13]); + in[3] = _mm_add_epi16(s[3], s[12]); + in[4] = _mm_add_epi16(s[4], s[11]); + in[5] = _mm_add_epi16(s[5], s[10]); + in[6] = _mm_add_epi16(s[6], s[9]); + in[7] = _mm_add_epi16(s[7], s[8]); + in[8] = _mm_sub_epi16(s[7], s[8]); + in[9] = _mm_sub_epi16(s[6], s[9]); + in[10] = _mm_sub_epi16(s[5], s[10]); + in[11] = _mm_sub_epi16(s[4], s[11]); + in[12] = _mm_sub_epi16(s[3], s[12]); + in[13] = _mm_sub_epi16(s[2], s[13]); + in[14] = _mm_sub_epi16(s[1], s[14]); + in[15] = _mm_sub_epi16(s[0], s[15]); +} + +void idct16_sse2(__m128i *in0, __m128i *in1) { + array_transpose_16x16(in0, in1); + idct16_8col(in0); + idct16_8col(in1); +} + +void iadst16_sse2(__m128i *in0, __m128i *in1) { + array_transpose_16x16(in0, in1); + iadst16_8col(in0); + iadst16_8col(in1); +} + +void vpx_idct16x16_10_add_sse2(const tran_low_t *input, uint8_t *dest, + int stride) { + const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING); + const __m128i final_rounding = _mm_set1_epi16(1 << 5); + const __m128i zero = _mm_setzero_si128(); + + const __m128i stg2_0 = pair_set_epi16(cospi_30_64, -cospi_2_64); + const __m128i stg2_1 = pair_set_epi16(cospi_2_64, cospi_30_64); + const __m128i stg2_6 = pair_set_epi16(cospi_6_64, -cospi_26_64); + const __m128i stg2_7 = pair_set_epi16(cospi_26_64, cospi_6_64); + + const __m128i stg3_0 = pair_set_epi16(cospi_28_64, -cospi_4_64); + const __m128i stg3_1 = pair_set_epi16(cospi_4_64, cospi_28_64); + + const __m128i stg4_0 = pair_set_epi16(cospi_16_64, cospi_16_64); + const __m128i stg4_1 = pair_set_epi16(cospi_16_64, -cospi_16_64); + const __m128i stg4_4 = pair_set_epi16(-cospi_8_64, cospi_24_64); + const __m128i stg4_5 = pair_set_epi16(cospi_24_64, cospi_8_64); + const __m128i stg4_6 = pair_set_epi16(-cospi_24_64, -cospi_8_64); + const __m128i stg4_7 = pair_set_epi16(-cospi_8_64, cospi_24_64); + + const __m128i stg6_0 = pair_set_epi16(-cospi_16_64, cospi_16_64); + __m128i in[16], l[16]; + __m128i stp1_0, stp1_1, stp1_2, stp1_3, stp1_4, stp1_5, stp1_6, + stp1_8, stp1_9, stp1_10, stp1_11, stp1_12, stp1_13, stp1_14, stp1_15, + stp1_8_0, stp1_12_0; + __m128i stp2_0, stp2_1, stp2_2, stp2_3, stp2_4, stp2_5, stp2_6, stp2_7, + stp2_8, stp2_9, stp2_10, stp2_11, stp2_12, stp2_13, stp2_14; + __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; + int i; + // First 1-D inverse DCT + // Load input data. + in[0] = load_input_data(input); + in[1] = load_input_data(input + 8 * 2); + in[2] = load_input_data(input + 8 * 4); + in[3] = load_input_data(input + 8 * 6); + + TRANSPOSE_8X4(in[0], in[1], in[2], in[3], in[0], in[1]); + + // Stage2 + { + const __m128i lo_1_15 = _mm_unpackhi_epi16(in[0], zero); + const __m128i lo_13_3 = _mm_unpackhi_epi16(zero, in[1]); + + tmp0 = _mm_madd_epi16(lo_1_15, stg2_0); + tmp2 = _mm_madd_epi16(lo_1_15, stg2_1); + tmp5 = _mm_madd_epi16(lo_13_3, stg2_6); + tmp7 = _mm_madd_epi16(lo_13_3, stg2_7); + + tmp0 = _mm_add_epi32(tmp0, rounding); + tmp2 = _mm_add_epi32(tmp2, rounding); + tmp5 = _mm_add_epi32(tmp5, rounding); + tmp7 = _mm_add_epi32(tmp7, rounding); + + tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS); + tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS); + tmp5 = _mm_srai_epi32(tmp5, DCT_CONST_BITS); + tmp7 = _mm_srai_epi32(tmp7, DCT_CONST_BITS); + + stp2_8 = _mm_packs_epi32(tmp0, tmp2); + stp2_11 = _mm_packs_epi32(tmp5, tmp7); + } + + // Stage3 + { + const __m128i lo_2_14 = _mm_unpacklo_epi16(in[1], zero); + + tmp0 = _mm_madd_epi16(lo_2_14, stg3_0); + tmp2 = _mm_madd_epi16(lo_2_14, stg3_1); + + tmp0 = _mm_add_epi32(tmp0, rounding); + tmp2 = _mm_add_epi32(tmp2, rounding); + tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS); + tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS); + + stp1_13 = _mm_unpackhi_epi64(stp2_11, zero); + stp1_14 = _mm_unpackhi_epi64(stp2_8, zero); + + stp1_4 = _mm_packs_epi32(tmp0, tmp2); + } + + // Stage4 + { + const __m128i lo_0_8 = _mm_unpacklo_epi16(in[0], zero); + const __m128i lo_9_14 = _mm_unpacklo_epi16(stp2_8, stp1_14); + const __m128i lo_10_13 = _mm_unpacklo_epi16(stp2_11, stp1_13); + + tmp0 = _mm_madd_epi16(lo_0_8, stg4_0); + tmp2 = _mm_madd_epi16(lo_0_8, stg4_1); + tmp1 = _mm_madd_epi16(lo_9_14, stg4_4); + tmp3 = _mm_madd_epi16(lo_9_14, stg4_5); + tmp5 = _mm_madd_epi16(lo_10_13, stg4_6); + tmp7 = _mm_madd_epi16(lo_10_13, stg4_7); + + tmp0 = _mm_add_epi32(tmp0, rounding); + tmp2 = _mm_add_epi32(tmp2, rounding); + tmp1 = _mm_add_epi32(tmp1, rounding); + tmp3 = _mm_add_epi32(tmp3, rounding); + tmp5 = _mm_add_epi32(tmp5, rounding); + tmp7 = _mm_add_epi32(tmp7, rounding); + + tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS); + tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS); + tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS); + tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS); + tmp5 = _mm_srai_epi32(tmp5, DCT_CONST_BITS); + tmp7 = _mm_srai_epi32(tmp7, DCT_CONST_BITS); + + stp1_0 = _mm_packs_epi32(tmp0, tmp0); + stp1_1 = _mm_packs_epi32(tmp2, tmp2); + stp2_9 = _mm_packs_epi32(tmp1, tmp3); + stp2_10 = _mm_packs_epi32(tmp5, tmp7); + + stp2_6 = _mm_unpackhi_epi64(stp1_4, zero); + } + + // Stage5 and Stage6 + { + tmp0 = _mm_add_epi16(stp2_8, stp2_11); + tmp1 = _mm_sub_epi16(stp2_8, stp2_11); + tmp2 = _mm_add_epi16(stp2_9, stp2_10); + tmp3 = _mm_sub_epi16(stp2_9, stp2_10); + + stp1_9 = _mm_unpacklo_epi64(tmp2, zero); + stp1_10 = _mm_unpacklo_epi64(tmp3, zero); + stp1_8 = _mm_unpacklo_epi64(tmp0, zero); + stp1_11 = _mm_unpacklo_epi64(tmp1, zero); + + stp1_13 = _mm_unpackhi_epi64(tmp3, zero); + stp1_14 = _mm_unpackhi_epi64(tmp2, zero); + stp1_12 = _mm_unpackhi_epi64(tmp1, zero); + stp1_15 = _mm_unpackhi_epi64(tmp0, zero); + } + + // Stage6 + { + const __m128i lo_6_5 = _mm_unpacklo_epi16(stp2_6, stp1_4); + const __m128i lo_10_13 = _mm_unpacklo_epi16(stp1_10, stp1_13); + const __m128i lo_11_12 = _mm_unpacklo_epi16(stp1_11, stp1_12); + + tmp1 = _mm_madd_epi16(lo_6_5, stg4_1); + tmp3 = _mm_madd_epi16(lo_6_5, stg4_0); + tmp0 = _mm_madd_epi16(lo_10_13, stg6_0); + tmp2 = _mm_madd_epi16(lo_10_13, stg4_0); + tmp4 = _mm_madd_epi16(lo_11_12, stg6_0); + tmp6 = _mm_madd_epi16(lo_11_12, stg4_0); + + tmp1 = _mm_add_epi32(tmp1, rounding); + tmp3 = _mm_add_epi32(tmp3, rounding); + tmp0 = _mm_add_epi32(tmp0, rounding); + tmp2 = _mm_add_epi32(tmp2, rounding); + tmp4 = _mm_add_epi32(tmp4, rounding); + tmp6 = _mm_add_epi32(tmp6, rounding); + + tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS); + tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS); + tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS); + tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS); + tmp4 = _mm_srai_epi32(tmp4, DCT_CONST_BITS); + tmp6 = _mm_srai_epi32(tmp6, DCT_CONST_BITS); + + stp1_6 = _mm_packs_epi32(tmp3, tmp1); + + stp2_10 = _mm_packs_epi32(tmp0, zero); + stp2_13 = _mm_packs_epi32(tmp2, zero); + stp2_11 = _mm_packs_epi32(tmp4, zero); + stp2_12 = _mm_packs_epi32(tmp6, zero); + + tmp0 = _mm_add_epi16(stp1_0, stp1_4); + tmp1 = _mm_sub_epi16(stp1_0, stp1_4); + tmp2 = _mm_add_epi16(stp1_1, stp1_6); + tmp3 = _mm_sub_epi16(stp1_1, stp1_6); + + stp2_0 = _mm_unpackhi_epi64(tmp0, zero); + stp2_1 = _mm_unpacklo_epi64(tmp2, zero); + stp2_2 = _mm_unpackhi_epi64(tmp2, zero); + stp2_3 = _mm_unpacklo_epi64(tmp0, zero); + stp2_4 = _mm_unpacklo_epi64(tmp1, zero); + stp2_5 = _mm_unpackhi_epi64(tmp3, zero); + stp2_6 = _mm_unpacklo_epi64(tmp3, zero); + stp2_7 = _mm_unpackhi_epi64(tmp1, zero); + } + + // Stage7. Left 8x16 only. + l[0] = _mm_add_epi16(stp2_0, stp1_15); + l[1] = _mm_add_epi16(stp2_1, stp1_14); + l[2] = _mm_add_epi16(stp2_2, stp2_13); + l[3] = _mm_add_epi16(stp2_3, stp2_12); + l[4] = _mm_add_epi16(stp2_4, stp2_11); + l[5] = _mm_add_epi16(stp2_5, stp2_10); + l[6] = _mm_add_epi16(stp2_6, stp1_9); + l[7] = _mm_add_epi16(stp2_7, stp1_8); + l[8] = _mm_sub_epi16(stp2_7, stp1_8); + l[9] = _mm_sub_epi16(stp2_6, stp1_9); + l[10] = _mm_sub_epi16(stp2_5, stp2_10); + l[11] = _mm_sub_epi16(stp2_4, stp2_11); + l[12] = _mm_sub_epi16(stp2_3, stp2_12); + l[13] = _mm_sub_epi16(stp2_2, stp2_13); + l[14] = _mm_sub_epi16(stp2_1, stp1_14); + l[15] = _mm_sub_epi16(stp2_0, stp1_15); + + // Second 1-D inverse transform, performed per 8x16 block + for (i = 0; i < 2; i++) { + int j; + array_transpose_4X8(l + 8 * i, in); + + IDCT16_10 + + // Stage7 + in[0] = _mm_add_epi16(stp2_0, stp1_15); + in[1] = _mm_add_epi16(stp2_1, stp1_14); + in[2] = _mm_add_epi16(stp2_2, stp2_13); + in[3] = _mm_add_epi16(stp2_3, stp2_12); + in[4] = _mm_add_epi16(stp2_4, stp2_11); + in[5] = _mm_add_epi16(stp2_5, stp2_10); + in[6] = _mm_add_epi16(stp2_6, stp1_9); + in[7] = _mm_add_epi16(stp2_7, stp1_8); + in[8] = _mm_sub_epi16(stp2_7, stp1_8); + in[9] = _mm_sub_epi16(stp2_6, stp1_9); + in[10] = _mm_sub_epi16(stp2_5, stp2_10); + in[11] = _mm_sub_epi16(stp2_4, stp2_11); + in[12] = _mm_sub_epi16(stp2_3, stp2_12); + in[13] = _mm_sub_epi16(stp2_2, stp2_13); + in[14] = _mm_sub_epi16(stp2_1, stp1_14); + in[15] = _mm_sub_epi16(stp2_0, stp1_15); + + for (j = 0; j < 16; ++j) { + // Final rounding and shift + in[j] = _mm_adds_epi16(in[j], final_rounding); + in[j] = _mm_srai_epi16(in[j], 6); + RECON_AND_STORE(dest + j * stride, in[j]); + } + + dest += 8; + } +} + +#define LOAD_DQCOEFF(reg, input) \ + { \ + reg = load_input_data(input); \ + input += 8; \ + } \ + +#define IDCT32_34 \ +/* Stage1 */ \ +{ \ + const __m128i zero = _mm_setzero_si128();\ + const __m128i lo_1_31 = _mm_unpacklo_epi16(in[1], zero); \ + const __m128i hi_1_31 = _mm_unpackhi_epi16(in[1], zero); \ + \ + const __m128i lo_25_7= _mm_unpacklo_epi16(zero, in[7]); \ + const __m128i hi_25_7 = _mm_unpackhi_epi16(zero, in[7]); \ + \ + const __m128i lo_5_27 = _mm_unpacklo_epi16(in[5], zero); \ + const __m128i hi_5_27 = _mm_unpackhi_epi16(in[5], zero); \ + \ + const __m128i lo_29_3 = _mm_unpacklo_epi16(zero, in[3]); \ + const __m128i hi_29_3 = _mm_unpackhi_epi16(zero, in[3]); \ + \ + MULTIPLICATION_AND_ADD_2(lo_1_31, hi_1_31, stg1_0, \ + stg1_1, stp1_16, stp1_31); \ + MULTIPLICATION_AND_ADD_2(lo_25_7, hi_25_7, stg1_6, \ + stg1_7, stp1_19, stp1_28); \ + MULTIPLICATION_AND_ADD_2(lo_5_27, hi_5_27, stg1_8, \ + stg1_9, stp1_20, stp1_27); \ + MULTIPLICATION_AND_ADD_2(lo_29_3, hi_29_3, stg1_14, \ + stg1_15, stp1_23, stp1_24); \ +} \ +\ +/* Stage2 */ \ +{ \ + const __m128i zero = _mm_setzero_si128();\ + const __m128i lo_2_30 = _mm_unpacklo_epi16(in[2], zero); \ + const __m128i hi_2_30 = _mm_unpackhi_epi16(in[2], zero); \ + \ + const __m128i lo_26_6 = _mm_unpacklo_epi16(zero, in[6]); \ + const __m128i hi_26_6 = _mm_unpackhi_epi16(zero, in[6]); \ + \ + MULTIPLICATION_AND_ADD_2(lo_2_30, hi_2_30, stg2_0, \ + stg2_1, stp2_8, stp2_15); \ + MULTIPLICATION_AND_ADD_2(lo_26_6, hi_26_6, stg2_6, \ + stg2_7, stp2_11, stp2_12); \ + \ + stp2_16 = stp1_16; \ + stp2_19 = stp1_19; \ + \ + stp2_20 = stp1_20; \ + stp2_23 = stp1_23; \ + \ + stp2_24 = stp1_24; \ + stp2_27 = stp1_27; \ + \ + stp2_28 = stp1_28; \ + stp2_31 = stp1_31; \ +} \ +\ +/* Stage3 */ \ +{ \ + const __m128i zero = _mm_setzero_si128();\ + const __m128i lo_4_28 = _mm_unpacklo_epi16(in[4], zero); \ + const __m128i hi_4_28 = _mm_unpackhi_epi16(in[4], zero); \ + \ + const __m128i lo_17_30 = _mm_unpacklo_epi16(stp1_16, stp1_31); \ + const __m128i hi_17_30 = _mm_unpackhi_epi16(stp1_16, stp1_31); \ + const __m128i lo_18_29 = _mm_unpacklo_epi16(stp1_19, stp1_28); \ + const __m128i hi_18_29 = _mm_unpackhi_epi16(stp1_19, stp1_28); \ + \ + const __m128i lo_21_26 = _mm_unpacklo_epi16(stp1_20, stp1_27); \ + const __m128i hi_21_26 = _mm_unpackhi_epi16(stp1_20, stp1_27); \ + const __m128i lo_22_25 = _mm_unpacklo_epi16(stp1_23, stp1_24); \ + const __m128i hi_22_25 = _mm_unpackhi_epi16(stp1_23, stp2_24); \ + \ + MULTIPLICATION_AND_ADD_2(lo_4_28, hi_4_28, stg3_0, \ + stg3_1, stp1_4, stp1_7); \ + \ + stp1_8 = stp2_8; \ + stp1_11 = stp2_11; \ + stp1_12 = stp2_12; \ + stp1_15 = stp2_15; \ + \ + MULTIPLICATION_AND_ADD(lo_17_30, hi_17_30, lo_18_29, hi_18_29, stg3_4, \ + stg3_5, stg3_6, stg3_4, stp1_17, stp1_30, \ + stp1_18, stp1_29) \ + MULTIPLICATION_AND_ADD(lo_21_26, hi_21_26, lo_22_25, hi_22_25, stg3_8, \ + stg3_9, stg3_10, stg3_8, stp1_21, stp1_26, \ + stp1_22, stp1_25) \ + \ + stp1_16 = stp2_16; \ + stp1_31 = stp2_31; \ + stp1_19 = stp2_19; \ + stp1_20 = stp2_20; \ + stp1_23 = stp2_23; \ + stp1_24 = stp2_24; \ + stp1_27 = stp2_27; \ + stp1_28 = stp2_28; \ +} \ +\ +/* Stage4 */ \ +{ \ + const __m128i zero = _mm_setzero_si128();\ + const __m128i lo_0_16 = _mm_unpacklo_epi16(in[0], zero); \ + const __m128i hi_0_16 = _mm_unpackhi_epi16(in[0], zero); \ + \ + const __m128i lo_9_14 = _mm_unpacklo_epi16(stp2_8, stp2_15); \ + const __m128i hi_9_14 = _mm_unpackhi_epi16(stp2_8, stp2_15); \ + const __m128i lo_10_13 = _mm_unpacklo_epi16(stp2_11, stp2_12); \ + const __m128i hi_10_13 = _mm_unpackhi_epi16(stp2_11, stp2_12); \ + \ + MULTIPLICATION_AND_ADD_2(lo_0_16, hi_0_16, stg4_0, \ + stg4_1, stp2_0, stp2_1); \ + \ + stp2_4 = stp1_4; \ + stp2_5 = stp1_4; \ + stp2_6 = stp1_7; \ + stp2_7 = stp1_7; \ + \ + MULTIPLICATION_AND_ADD(lo_9_14, hi_9_14, lo_10_13, hi_10_13, stg4_4, \ + stg4_5, stg4_6, stg4_4, stp2_9, stp2_14, \ + stp2_10, stp2_13) \ + \ + stp2_8 = stp1_8; \ + stp2_15 = stp1_15; \ + stp2_11 = stp1_11; \ + stp2_12 = stp1_12; \ + \ + stp2_16 = _mm_add_epi16(stp1_16, stp1_19); \ + stp2_17 = _mm_add_epi16(stp1_17, stp1_18); \ + stp2_18 = _mm_sub_epi16(stp1_17, stp1_18); \ + stp2_19 = _mm_sub_epi16(stp1_16, stp1_19); \ + stp2_20 = _mm_sub_epi16(stp1_23, stp1_20); \ + stp2_21 = _mm_sub_epi16(stp1_22, stp1_21); \ + stp2_22 = _mm_add_epi16(stp1_22, stp1_21); \ + stp2_23 = _mm_add_epi16(stp1_23, stp1_20); \ + \ + stp2_24 = _mm_add_epi16(stp1_24, stp1_27); \ + stp2_25 = _mm_add_epi16(stp1_25, stp1_26); \ + stp2_26 = _mm_sub_epi16(stp1_25, stp1_26); \ + stp2_27 = _mm_sub_epi16(stp1_24, stp1_27); \ + stp2_28 = _mm_sub_epi16(stp1_31, stp1_28); \ + stp2_29 = _mm_sub_epi16(stp1_30, stp1_29); \ + stp2_30 = _mm_add_epi16(stp1_29, stp1_30); \ + stp2_31 = _mm_add_epi16(stp1_28, stp1_31); \ +} \ +\ +/* Stage5 */ \ +{ \ + const __m128i lo_6_5 = _mm_unpacklo_epi16(stp2_6, stp2_5); \ + const __m128i hi_6_5 = _mm_unpackhi_epi16(stp2_6, stp2_5); \ + const __m128i lo_18_29 = _mm_unpacklo_epi16(stp2_18, stp2_29); \ + const __m128i hi_18_29 = _mm_unpackhi_epi16(stp2_18, stp2_29); \ + \ + const __m128i lo_19_28 = _mm_unpacklo_epi16(stp2_19, stp2_28); \ + const __m128i hi_19_28 = _mm_unpackhi_epi16(stp2_19, stp2_28); \ + const __m128i lo_20_27 = _mm_unpacklo_epi16(stp2_20, stp2_27); \ + const __m128i hi_20_27 = _mm_unpackhi_epi16(stp2_20, stp2_27); \ + \ + const __m128i lo_21_26 = _mm_unpacklo_epi16(stp2_21, stp2_26); \ + const __m128i hi_21_26 = _mm_unpackhi_epi16(stp2_21, stp2_26); \ + \ + stp1_0 = stp2_0; \ + stp1_1 = stp2_1; \ + stp1_2 = stp2_1; \ + stp1_3 = stp2_0; \ + \ + tmp0 = _mm_madd_epi16(lo_6_5, stg4_1); \ + tmp1 = _mm_madd_epi16(hi_6_5, stg4_1); \ + tmp2 = _mm_madd_epi16(lo_6_5, stg4_0); \ + tmp3 = _mm_madd_epi16(hi_6_5, stg4_0); \ + \ + tmp0 = _mm_add_epi32(tmp0, rounding); \ + tmp1 = _mm_add_epi32(tmp1, rounding); \ + tmp2 = _mm_add_epi32(tmp2, rounding); \ + tmp3 = _mm_add_epi32(tmp3, rounding); \ + \ + tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS); \ + tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS); \ + tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS); \ + tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS); \ + \ + stp1_5 = _mm_packs_epi32(tmp0, tmp1); \ + stp1_6 = _mm_packs_epi32(tmp2, tmp3); \ + \ + stp1_4 = stp2_4; \ + stp1_7 = stp2_7; \ + \ + stp1_8 = _mm_add_epi16(stp2_8, stp2_11); \ + stp1_9 = _mm_add_epi16(stp2_9, stp2_10); \ + stp1_10 = _mm_sub_epi16(stp2_9, stp2_10); \ + stp1_11 = _mm_sub_epi16(stp2_8, stp2_11); \ + stp1_12 = _mm_sub_epi16(stp2_15, stp2_12); \ + stp1_13 = _mm_sub_epi16(stp2_14, stp2_13); \ + stp1_14 = _mm_add_epi16(stp2_14, stp2_13); \ + stp1_15 = _mm_add_epi16(stp2_15, stp2_12); \ + \ + stp1_16 = stp2_16; \ + stp1_17 = stp2_17; \ + \ + MULTIPLICATION_AND_ADD(lo_18_29, hi_18_29, lo_19_28, hi_19_28, stg4_4, \ + stg4_5, stg4_4, stg4_5, stp1_18, stp1_29, \ + stp1_19, stp1_28) \ + MULTIPLICATION_AND_ADD(lo_20_27, hi_20_27, lo_21_26, hi_21_26, stg4_6, \ + stg4_4, stg4_6, stg4_4, stp1_20, stp1_27, \ + stp1_21, stp1_26) \ + \ + stp1_22 = stp2_22; \ + stp1_23 = stp2_23; \ + stp1_24 = stp2_24; \ + stp1_25 = stp2_25; \ + stp1_30 = stp2_30; \ + stp1_31 = stp2_31; \ +} \ +\ +/* Stage6 */ \ +{ \ + const __m128i lo_10_13 = _mm_unpacklo_epi16(stp1_10, stp1_13); \ + const __m128i hi_10_13 = _mm_unpackhi_epi16(stp1_10, stp1_13); \ + const __m128i lo_11_12 = _mm_unpacklo_epi16(stp1_11, stp1_12); \ + const __m128i hi_11_12 = _mm_unpackhi_epi16(stp1_11, stp1_12); \ + \ + stp2_0 = _mm_add_epi16(stp1_0, stp1_7); \ + stp2_1 = _mm_add_epi16(stp1_1, stp1_6); \ + stp2_2 = _mm_add_epi16(stp1_2, stp1_5); \ + stp2_3 = _mm_add_epi16(stp1_3, stp1_4); \ + stp2_4 = _mm_sub_epi16(stp1_3, stp1_4); \ + stp2_5 = _mm_sub_epi16(stp1_2, stp1_5); \ + stp2_6 = _mm_sub_epi16(stp1_1, stp1_6); \ + stp2_7 = _mm_sub_epi16(stp1_0, stp1_7); \ + \ + stp2_8 = stp1_8; \ + stp2_9 = stp1_9; \ + stp2_14 = stp1_14; \ + stp2_15 = stp1_15; \ + \ + MULTIPLICATION_AND_ADD(lo_10_13, hi_10_13, lo_11_12, hi_11_12, \ + stg6_0, stg4_0, stg6_0, stg4_0, stp2_10, \ + stp2_13, stp2_11, stp2_12) \ + \ + stp2_16 = _mm_add_epi16(stp1_16, stp1_23); \ + stp2_17 = _mm_add_epi16(stp1_17, stp1_22); \ + stp2_18 = _mm_add_epi16(stp1_18, stp1_21); \ + stp2_19 = _mm_add_epi16(stp1_19, stp1_20); \ + stp2_20 = _mm_sub_epi16(stp1_19, stp1_20); \ + stp2_21 = _mm_sub_epi16(stp1_18, stp1_21); \ + stp2_22 = _mm_sub_epi16(stp1_17, stp1_22); \ + stp2_23 = _mm_sub_epi16(stp1_16, stp1_23); \ + \ + stp2_24 = _mm_sub_epi16(stp1_31, stp1_24); \ + stp2_25 = _mm_sub_epi16(stp1_30, stp1_25); \ + stp2_26 = _mm_sub_epi16(stp1_29, stp1_26); \ + stp2_27 = _mm_sub_epi16(stp1_28, stp1_27); \ + stp2_28 = _mm_add_epi16(stp1_27, stp1_28); \ + stp2_29 = _mm_add_epi16(stp1_26, stp1_29); \ + stp2_30 = _mm_add_epi16(stp1_25, stp1_30); \ + stp2_31 = _mm_add_epi16(stp1_24, stp1_31); \ +} \ +\ +/* Stage7 */ \ +{ \ + const __m128i lo_20_27 = _mm_unpacklo_epi16(stp2_20, stp2_27); \ + const __m128i hi_20_27 = _mm_unpackhi_epi16(stp2_20, stp2_27); \ + const __m128i lo_21_26 = _mm_unpacklo_epi16(stp2_21, stp2_26); \ + const __m128i hi_21_26 = _mm_unpackhi_epi16(stp2_21, stp2_26); \ + \ + const __m128i lo_22_25 = _mm_unpacklo_epi16(stp2_22, stp2_25); \ + const __m128i hi_22_25 = _mm_unpackhi_epi16(stp2_22, stp2_25); \ + const __m128i lo_23_24 = _mm_unpacklo_epi16(stp2_23, stp2_24); \ + const __m128i hi_23_24 = _mm_unpackhi_epi16(stp2_23, stp2_24); \ + \ + stp1_0 = _mm_add_epi16(stp2_0, stp2_15); \ + stp1_1 = _mm_add_epi16(stp2_1, stp2_14); \ + stp1_2 = _mm_add_epi16(stp2_2, stp2_13); \ + stp1_3 = _mm_add_epi16(stp2_3, stp2_12); \ + stp1_4 = _mm_add_epi16(stp2_4, stp2_11); \ + stp1_5 = _mm_add_epi16(stp2_5, stp2_10); \ + stp1_6 = _mm_add_epi16(stp2_6, stp2_9); \ + stp1_7 = _mm_add_epi16(stp2_7, stp2_8); \ + stp1_8 = _mm_sub_epi16(stp2_7, stp2_8); \ + stp1_9 = _mm_sub_epi16(stp2_6, stp2_9); \ + stp1_10 = _mm_sub_epi16(stp2_5, stp2_10); \ + stp1_11 = _mm_sub_epi16(stp2_4, stp2_11); \ + stp1_12 = _mm_sub_epi16(stp2_3, stp2_12); \ + stp1_13 = _mm_sub_epi16(stp2_2, stp2_13); \ + stp1_14 = _mm_sub_epi16(stp2_1, stp2_14); \ + stp1_15 = _mm_sub_epi16(stp2_0, stp2_15); \ + \ + stp1_16 = stp2_16; \ + stp1_17 = stp2_17; \ + stp1_18 = stp2_18; \ + stp1_19 = stp2_19; \ + \ + MULTIPLICATION_AND_ADD(lo_20_27, hi_20_27, lo_21_26, hi_21_26, stg6_0, \ + stg4_0, stg6_0, stg4_0, stp1_20, stp1_27, \ + stp1_21, stp1_26) \ + MULTIPLICATION_AND_ADD(lo_22_25, hi_22_25, lo_23_24, hi_23_24, stg6_0, \ + stg4_0, stg6_0, stg4_0, stp1_22, stp1_25, \ + stp1_23, stp1_24) \ + \ + stp1_28 = stp2_28; \ + stp1_29 = stp2_29; \ + stp1_30 = stp2_30; \ + stp1_31 = stp2_31; \ +} + + +#define IDCT32 \ +/* Stage1 */ \ +{ \ + const __m128i lo_1_31 = _mm_unpacklo_epi16(in[1], in[31]); \ + const __m128i hi_1_31 = _mm_unpackhi_epi16(in[1], in[31]); \ + const __m128i lo_17_15 = _mm_unpacklo_epi16(in[17], in[15]); \ + const __m128i hi_17_15 = _mm_unpackhi_epi16(in[17], in[15]); \ + \ + const __m128i lo_9_23 = _mm_unpacklo_epi16(in[9], in[23]); \ + const __m128i hi_9_23 = _mm_unpackhi_epi16(in[9], in[23]); \ + const __m128i lo_25_7= _mm_unpacklo_epi16(in[25], in[7]); \ + const __m128i hi_25_7 = _mm_unpackhi_epi16(in[25], in[7]); \ + \ + const __m128i lo_5_27 = _mm_unpacklo_epi16(in[5], in[27]); \ + const __m128i hi_5_27 = _mm_unpackhi_epi16(in[5], in[27]); \ + const __m128i lo_21_11 = _mm_unpacklo_epi16(in[21], in[11]); \ + const __m128i hi_21_11 = _mm_unpackhi_epi16(in[21], in[11]); \ + \ + const __m128i lo_13_19 = _mm_unpacklo_epi16(in[13], in[19]); \ + const __m128i hi_13_19 = _mm_unpackhi_epi16(in[13], in[19]); \ + const __m128i lo_29_3 = _mm_unpacklo_epi16(in[29], in[3]); \ + const __m128i hi_29_3 = _mm_unpackhi_epi16(in[29], in[3]); \ + \ + MULTIPLICATION_AND_ADD(lo_1_31, hi_1_31, lo_17_15, hi_17_15, stg1_0, \ + stg1_1, stg1_2, stg1_3, stp1_16, stp1_31, \ + stp1_17, stp1_30) \ + MULTIPLICATION_AND_ADD(lo_9_23, hi_9_23, lo_25_7, hi_25_7, stg1_4, \ + stg1_5, stg1_6, stg1_7, stp1_18, stp1_29, \ + stp1_19, stp1_28) \ + MULTIPLICATION_AND_ADD(lo_5_27, hi_5_27, lo_21_11, hi_21_11, stg1_8, \ + stg1_9, stg1_10, stg1_11, stp1_20, stp1_27, \ + stp1_21, stp1_26) \ + MULTIPLICATION_AND_ADD(lo_13_19, hi_13_19, lo_29_3, hi_29_3, stg1_12, \ + stg1_13, stg1_14, stg1_15, stp1_22, stp1_25, \ + stp1_23, stp1_24) \ +} \ +\ +/* Stage2 */ \ +{ \ + const __m128i lo_2_30 = _mm_unpacklo_epi16(in[2], in[30]); \ + const __m128i hi_2_30 = _mm_unpackhi_epi16(in[2], in[30]); \ + const __m128i lo_18_14 = _mm_unpacklo_epi16(in[18], in[14]); \ + const __m128i hi_18_14 = _mm_unpackhi_epi16(in[18], in[14]); \ + \ + const __m128i lo_10_22 = _mm_unpacklo_epi16(in[10], in[22]); \ + const __m128i hi_10_22 = _mm_unpackhi_epi16(in[10], in[22]); \ + const __m128i lo_26_6 = _mm_unpacklo_epi16(in[26], in[6]); \ + const __m128i hi_26_6 = _mm_unpackhi_epi16(in[26], in[6]); \ + \ + MULTIPLICATION_AND_ADD(lo_2_30, hi_2_30, lo_18_14, hi_18_14, stg2_0, \ + stg2_1, stg2_2, stg2_3, stp2_8, stp2_15, stp2_9, \ + stp2_14) \ + MULTIPLICATION_AND_ADD(lo_10_22, hi_10_22, lo_26_6, hi_26_6, stg2_4, \ + stg2_5, stg2_6, stg2_7, stp2_10, stp2_13, \ + stp2_11, stp2_12) \ + \ + stp2_16 = _mm_add_epi16(stp1_16, stp1_17); \ + stp2_17 = _mm_sub_epi16(stp1_16, stp1_17); \ + stp2_18 = _mm_sub_epi16(stp1_19, stp1_18); \ + stp2_19 = _mm_add_epi16(stp1_19, stp1_18); \ + \ + stp2_20 = _mm_add_epi16(stp1_20, stp1_21); \ + stp2_21 = _mm_sub_epi16(stp1_20, stp1_21); \ + stp2_22 = _mm_sub_epi16(stp1_23, stp1_22); \ + stp2_23 = _mm_add_epi16(stp1_23, stp1_22); \ + \ + stp2_24 = _mm_add_epi16(stp1_24, stp1_25); \ + stp2_25 = _mm_sub_epi16(stp1_24, stp1_25); \ + stp2_26 = _mm_sub_epi16(stp1_27, stp1_26); \ + stp2_27 = _mm_add_epi16(stp1_27, stp1_26); \ + \ + stp2_28 = _mm_add_epi16(stp1_28, stp1_29); \ + stp2_29 = _mm_sub_epi16(stp1_28, stp1_29); \ + stp2_30 = _mm_sub_epi16(stp1_31, stp1_30); \ + stp2_31 = _mm_add_epi16(stp1_31, stp1_30); \ +} \ +\ +/* Stage3 */ \ +{ \ + const __m128i lo_4_28 = _mm_unpacklo_epi16(in[4], in[28]); \ + const __m128i hi_4_28 = _mm_unpackhi_epi16(in[4], in[28]); \ + const __m128i lo_20_12 = _mm_unpacklo_epi16(in[20], in[12]); \ + const __m128i hi_20_12 = _mm_unpackhi_epi16(in[20], in[12]); \ + \ + const __m128i lo_17_30 = _mm_unpacklo_epi16(stp2_17, stp2_30); \ + const __m128i hi_17_30 = _mm_unpackhi_epi16(stp2_17, stp2_30); \ + const __m128i lo_18_29 = _mm_unpacklo_epi16(stp2_18, stp2_29); \ + const __m128i hi_18_29 = _mm_unpackhi_epi16(stp2_18, stp2_29); \ + \ + const __m128i lo_21_26 = _mm_unpacklo_epi16(stp2_21, stp2_26); \ + const __m128i hi_21_26 = _mm_unpackhi_epi16(stp2_21, stp2_26); \ + const __m128i lo_22_25 = _mm_unpacklo_epi16(stp2_22, stp2_25); \ + const __m128i hi_22_25 = _mm_unpackhi_epi16(stp2_22, stp2_25); \ + \ + MULTIPLICATION_AND_ADD(lo_4_28, hi_4_28, lo_20_12, hi_20_12, stg3_0, \ + stg3_1, stg3_2, stg3_3, stp1_4, stp1_7, stp1_5, \ + stp1_6) \ + \ + stp1_8 = _mm_add_epi16(stp2_8, stp2_9); \ + stp1_9 = _mm_sub_epi16(stp2_8, stp2_9); \ + stp1_10 = _mm_sub_epi16(stp2_11, stp2_10); \ + stp1_11 = _mm_add_epi16(stp2_11, stp2_10); \ + stp1_12 = _mm_add_epi16(stp2_12, stp2_13); \ + stp1_13 = _mm_sub_epi16(stp2_12, stp2_13); \ + stp1_14 = _mm_sub_epi16(stp2_15, stp2_14); \ + stp1_15 = _mm_add_epi16(stp2_15, stp2_14); \ + \ + MULTIPLICATION_AND_ADD(lo_17_30, hi_17_30, lo_18_29, hi_18_29, stg3_4, \ + stg3_5, stg3_6, stg3_4, stp1_17, stp1_30, \ + stp1_18, stp1_29) \ + MULTIPLICATION_AND_ADD(lo_21_26, hi_21_26, lo_22_25, hi_22_25, stg3_8, \ + stg3_9, stg3_10, stg3_8, stp1_21, stp1_26, \ + stp1_22, stp1_25) \ + \ + stp1_16 = stp2_16; \ + stp1_31 = stp2_31; \ + stp1_19 = stp2_19; \ + stp1_20 = stp2_20; \ + stp1_23 = stp2_23; \ + stp1_24 = stp2_24; \ + stp1_27 = stp2_27; \ + stp1_28 = stp2_28; \ +} \ +\ +/* Stage4 */ \ +{ \ + const __m128i lo_0_16 = _mm_unpacklo_epi16(in[0], in[16]); \ + const __m128i hi_0_16 = _mm_unpackhi_epi16(in[0], in[16]); \ + const __m128i lo_8_24 = _mm_unpacklo_epi16(in[8], in[24]); \ + const __m128i hi_8_24 = _mm_unpackhi_epi16(in[8], in[24]); \ + \ + const __m128i lo_9_14 = _mm_unpacklo_epi16(stp1_9, stp1_14); \ + const __m128i hi_9_14 = _mm_unpackhi_epi16(stp1_9, stp1_14); \ + const __m128i lo_10_13 = _mm_unpacklo_epi16(stp1_10, stp1_13); \ + const __m128i hi_10_13 = _mm_unpackhi_epi16(stp1_10, stp1_13); \ + \ + MULTIPLICATION_AND_ADD(lo_0_16, hi_0_16, lo_8_24, hi_8_24, stg4_0, \ + stg4_1, stg4_2, stg4_3, stp2_0, stp2_1, \ + stp2_2, stp2_3) \ + \ + stp2_4 = _mm_add_epi16(stp1_4, stp1_5); \ + stp2_5 = _mm_sub_epi16(stp1_4, stp1_5); \ + stp2_6 = _mm_sub_epi16(stp1_7, stp1_6); \ + stp2_7 = _mm_add_epi16(stp1_7, stp1_6); \ + \ + MULTIPLICATION_AND_ADD(lo_9_14, hi_9_14, lo_10_13, hi_10_13, stg4_4, \ + stg4_5, stg4_6, stg4_4, stp2_9, stp2_14, \ + stp2_10, stp2_13) \ + \ + stp2_8 = stp1_8; \ + stp2_15 = stp1_15; \ + stp2_11 = stp1_11; \ + stp2_12 = stp1_12; \ + \ + stp2_16 = _mm_add_epi16(stp1_16, stp1_19); \ + stp2_17 = _mm_add_epi16(stp1_17, stp1_18); \ + stp2_18 = _mm_sub_epi16(stp1_17, stp1_18); \ + stp2_19 = _mm_sub_epi16(stp1_16, stp1_19); \ + stp2_20 = _mm_sub_epi16(stp1_23, stp1_20); \ + stp2_21 = _mm_sub_epi16(stp1_22, stp1_21); \ + stp2_22 = _mm_add_epi16(stp1_22, stp1_21); \ + stp2_23 = _mm_add_epi16(stp1_23, stp1_20); \ + \ + stp2_24 = _mm_add_epi16(stp1_24, stp1_27); \ + stp2_25 = _mm_add_epi16(stp1_25, stp1_26); \ + stp2_26 = _mm_sub_epi16(stp1_25, stp1_26); \ + stp2_27 = _mm_sub_epi16(stp1_24, stp1_27); \ + stp2_28 = _mm_sub_epi16(stp1_31, stp1_28); \ + stp2_29 = _mm_sub_epi16(stp1_30, stp1_29); \ + stp2_30 = _mm_add_epi16(stp1_29, stp1_30); \ + stp2_31 = _mm_add_epi16(stp1_28, stp1_31); \ +} \ +\ +/* Stage5 */ \ +{ \ + const __m128i lo_6_5 = _mm_unpacklo_epi16(stp2_6, stp2_5); \ + const __m128i hi_6_5 = _mm_unpackhi_epi16(stp2_6, stp2_5); \ + const __m128i lo_18_29 = _mm_unpacklo_epi16(stp2_18, stp2_29); \ + const __m128i hi_18_29 = _mm_unpackhi_epi16(stp2_18, stp2_29); \ + \ + const __m128i lo_19_28 = _mm_unpacklo_epi16(stp2_19, stp2_28); \ + const __m128i hi_19_28 = _mm_unpackhi_epi16(stp2_19, stp2_28); \ + const __m128i lo_20_27 = _mm_unpacklo_epi16(stp2_20, stp2_27); \ + const __m128i hi_20_27 = _mm_unpackhi_epi16(stp2_20, stp2_27); \ + \ + const __m128i lo_21_26 = _mm_unpacklo_epi16(stp2_21, stp2_26); \ + const __m128i hi_21_26 = _mm_unpackhi_epi16(stp2_21, stp2_26); \ + \ + stp1_0 = _mm_add_epi16(stp2_0, stp2_3); \ + stp1_1 = _mm_add_epi16(stp2_1, stp2_2); \ + stp1_2 = _mm_sub_epi16(stp2_1, stp2_2); \ + stp1_3 = _mm_sub_epi16(stp2_0, stp2_3); \ + \ + tmp0 = _mm_madd_epi16(lo_6_5, stg4_1); \ + tmp1 = _mm_madd_epi16(hi_6_5, stg4_1); \ + tmp2 = _mm_madd_epi16(lo_6_5, stg4_0); \ + tmp3 = _mm_madd_epi16(hi_6_5, stg4_0); \ + \ + tmp0 = _mm_add_epi32(tmp0, rounding); \ + tmp1 = _mm_add_epi32(tmp1, rounding); \ + tmp2 = _mm_add_epi32(tmp2, rounding); \ + tmp3 = _mm_add_epi32(tmp3, rounding); \ + \ + tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS); \ + tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS); \ + tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS); \ + tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS); \ + \ + stp1_5 = _mm_packs_epi32(tmp0, tmp1); \ + stp1_6 = _mm_packs_epi32(tmp2, tmp3); \ + \ + stp1_4 = stp2_4; \ + stp1_7 = stp2_7; \ + \ + stp1_8 = _mm_add_epi16(stp2_8, stp2_11); \ + stp1_9 = _mm_add_epi16(stp2_9, stp2_10); \ + stp1_10 = _mm_sub_epi16(stp2_9, stp2_10); \ + stp1_11 = _mm_sub_epi16(stp2_8, stp2_11); \ + stp1_12 = _mm_sub_epi16(stp2_15, stp2_12); \ + stp1_13 = _mm_sub_epi16(stp2_14, stp2_13); \ + stp1_14 = _mm_add_epi16(stp2_14, stp2_13); \ + stp1_15 = _mm_add_epi16(stp2_15, stp2_12); \ + \ + stp1_16 = stp2_16; \ + stp1_17 = stp2_17; \ + \ + MULTIPLICATION_AND_ADD(lo_18_29, hi_18_29, lo_19_28, hi_19_28, stg4_4, \ + stg4_5, stg4_4, stg4_5, stp1_18, stp1_29, \ + stp1_19, stp1_28) \ + MULTIPLICATION_AND_ADD(lo_20_27, hi_20_27, lo_21_26, hi_21_26, stg4_6, \ + stg4_4, stg4_6, stg4_4, stp1_20, stp1_27, \ + stp1_21, stp1_26) \ + \ + stp1_22 = stp2_22; \ + stp1_23 = stp2_23; \ + stp1_24 = stp2_24; \ + stp1_25 = stp2_25; \ + stp1_30 = stp2_30; \ + stp1_31 = stp2_31; \ +} \ +\ +/* Stage6 */ \ +{ \ + const __m128i lo_10_13 = _mm_unpacklo_epi16(stp1_10, stp1_13); \ + const __m128i hi_10_13 = _mm_unpackhi_epi16(stp1_10, stp1_13); \ + const __m128i lo_11_12 = _mm_unpacklo_epi16(stp1_11, stp1_12); \ + const __m128i hi_11_12 = _mm_unpackhi_epi16(stp1_11, stp1_12); \ + \ + stp2_0 = _mm_add_epi16(stp1_0, stp1_7); \ + stp2_1 = _mm_add_epi16(stp1_1, stp1_6); \ + stp2_2 = _mm_add_epi16(stp1_2, stp1_5); \ + stp2_3 = _mm_add_epi16(stp1_3, stp1_4); \ + stp2_4 = _mm_sub_epi16(stp1_3, stp1_4); \ + stp2_5 = _mm_sub_epi16(stp1_2, stp1_5); \ + stp2_6 = _mm_sub_epi16(stp1_1, stp1_6); \ + stp2_7 = _mm_sub_epi16(stp1_0, stp1_7); \ + \ + stp2_8 = stp1_8; \ + stp2_9 = stp1_9; \ + stp2_14 = stp1_14; \ + stp2_15 = stp1_15; \ + \ + MULTIPLICATION_AND_ADD(lo_10_13, hi_10_13, lo_11_12, hi_11_12, \ + stg6_0, stg4_0, stg6_0, stg4_0, stp2_10, \ + stp2_13, stp2_11, stp2_12) \ + \ + stp2_16 = _mm_add_epi16(stp1_16, stp1_23); \ + stp2_17 = _mm_add_epi16(stp1_17, stp1_22); \ + stp2_18 = _mm_add_epi16(stp1_18, stp1_21); \ + stp2_19 = _mm_add_epi16(stp1_19, stp1_20); \ + stp2_20 = _mm_sub_epi16(stp1_19, stp1_20); \ + stp2_21 = _mm_sub_epi16(stp1_18, stp1_21); \ + stp2_22 = _mm_sub_epi16(stp1_17, stp1_22); \ + stp2_23 = _mm_sub_epi16(stp1_16, stp1_23); \ + \ + stp2_24 = _mm_sub_epi16(stp1_31, stp1_24); \ + stp2_25 = _mm_sub_epi16(stp1_30, stp1_25); \ + stp2_26 = _mm_sub_epi16(stp1_29, stp1_26); \ + stp2_27 = _mm_sub_epi16(stp1_28, stp1_27); \ + stp2_28 = _mm_add_epi16(stp1_27, stp1_28); \ + stp2_29 = _mm_add_epi16(stp1_26, stp1_29); \ + stp2_30 = _mm_add_epi16(stp1_25, stp1_30); \ + stp2_31 = _mm_add_epi16(stp1_24, stp1_31); \ +} \ +\ +/* Stage7 */ \ +{ \ + const __m128i lo_20_27 = _mm_unpacklo_epi16(stp2_20, stp2_27); \ + const __m128i hi_20_27 = _mm_unpackhi_epi16(stp2_20, stp2_27); \ + const __m128i lo_21_26 = _mm_unpacklo_epi16(stp2_21, stp2_26); \ + const __m128i hi_21_26 = _mm_unpackhi_epi16(stp2_21, stp2_26); \ + \ + const __m128i lo_22_25 = _mm_unpacklo_epi16(stp2_22, stp2_25); \ + const __m128i hi_22_25 = _mm_unpackhi_epi16(stp2_22, stp2_25); \ + const __m128i lo_23_24 = _mm_unpacklo_epi16(stp2_23, stp2_24); \ + const __m128i hi_23_24 = _mm_unpackhi_epi16(stp2_23, stp2_24); \ + \ + stp1_0 = _mm_add_epi16(stp2_0, stp2_15); \ + stp1_1 = _mm_add_epi16(stp2_1, stp2_14); \ + stp1_2 = _mm_add_epi16(stp2_2, stp2_13); \ + stp1_3 = _mm_add_epi16(stp2_3, stp2_12); \ + stp1_4 = _mm_add_epi16(stp2_4, stp2_11); \ + stp1_5 = _mm_add_epi16(stp2_5, stp2_10); \ + stp1_6 = _mm_add_epi16(stp2_6, stp2_9); \ + stp1_7 = _mm_add_epi16(stp2_7, stp2_8); \ + stp1_8 = _mm_sub_epi16(stp2_7, stp2_8); \ + stp1_9 = _mm_sub_epi16(stp2_6, stp2_9); \ + stp1_10 = _mm_sub_epi16(stp2_5, stp2_10); \ + stp1_11 = _mm_sub_epi16(stp2_4, stp2_11); \ + stp1_12 = _mm_sub_epi16(stp2_3, stp2_12); \ + stp1_13 = _mm_sub_epi16(stp2_2, stp2_13); \ + stp1_14 = _mm_sub_epi16(stp2_1, stp2_14); \ + stp1_15 = _mm_sub_epi16(stp2_0, stp2_15); \ + \ + stp1_16 = stp2_16; \ + stp1_17 = stp2_17; \ + stp1_18 = stp2_18; \ + stp1_19 = stp2_19; \ + \ + MULTIPLICATION_AND_ADD(lo_20_27, hi_20_27, lo_21_26, hi_21_26, stg6_0, \ + stg4_0, stg6_0, stg4_0, stp1_20, stp1_27, \ + stp1_21, stp1_26) \ + MULTIPLICATION_AND_ADD(lo_22_25, hi_22_25, lo_23_24, hi_23_24, stg6_0, \ + stg4_0, stg6_0, stg4_0, stp1_22, stp1_25, \ + stp1_23, stp1_24) \ + \ + stp1_28 = stp2_28; \ + stp1_29 = stp2_29; \ + stp1_30 = stp2_30; \ + stp1_31 = stp2_31; \ +} + +// Only upper-left 8x8 has non-zero coeff +void vpx_idct32x32_34_add_sse2(const tran_low_t *input, uint8_t *dest, + int stride) { + const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING); + const __m128i final_rounding = _mm_set1_epi16(1<<5); + + // idct constants for each stage + const __m128i stg1_0 = pair_set_epi16(cospi_31_64, -cospi_1_64); + const __m128i stg1_1 = pair_set_epi16(cospi_1_64, cospi_31_64); + const __m128i stg1_6 = pair_set_epi16(cospi_7_64, -cospi_25_64); + const __m128i stg1_7 = pair_set_epi16(cospi_25_64, cospi_7_64); + const __m128i stg1_8 = pair_set_epi16(cospi_27_64, -cospi_5_64); + const __m128i stg1_9 = pair_set_epi16(cospi_5_64, cospi_27_64); + const __m128i stg1_14 = pair_set_epi16(cospi_3_64, -cospi_29_64); + const __m128i stg1_15 = pair_set_epi16(cospi_29_64, cospi_3_64); + + const __m128i stg2_0 = pair_set_epi16(cospi_30_64, -cospi_2_64); + const __m128i stg2_1 = pair_set_epi16(cospi_2_64, cospi_30_64); + const __m128i stg2_6 = pair_set_epi16(cospi_6_64, -cospi_26_64); + const __m128i stg2_7 = pair_set_epi16(cospi_26_64, cospi_6_64); + + const __m128i stg3_0 = pair_set_epi16(cospi_28_64, -cospi_4_64); + const __m128i stg3_1 = pair_set_epi16(cospi_4_64, cospi_28_64); + const __m128i stg3_4 = pair_set_epi16(-cospi_4_64, cospi_28_64); + const __m128i stg3_5 = pair_set_epi16(cospi_28_64, cospi_4_64); + const __m128i stg3_6 = pair_set_epi16(-cospi_28_64, -cospi_4_64); + const __m128i stg3_8 = pair_set_epi16(-cospi_20_64, cospi_12_64); + const __m128i stg3_9 = pair_set_epi16(cospi_12_64, cospi_20_64); + const __m128i stg3_10 = pair_set_epi16(-cospi_12_64, -cospi_20_64); + + const __m128i stg4_0 = pair_set_epi16(cospi_16_64, cospi_16_64); + const __m128i stg4_1 = pair_set_epi16(cospi_16_64, -cospi_16_64); + const __m128i stg4_4 = pair_set_epi16(-cospi_8_64, cospi_24_64); + const __m128i stg4_5 = pair_set_epi16(cospi_24_64, cospi_8_64); + const __m128i stg4_6 = pair_set_epi16(-cospi_24_64, -cospi_8_64); + + const __m128i stg6_0 = pair_set_epi16(-cospi_16_64, cospi_16_64); + + __m128i in[32], col[32]; + __m128i stp1_0, stp1_1, stp1_2, stp1_3, stp1_4, stp1_5, stp1_6, stp1_7, + stp1_8, stp1_9, stp1_10, stp1_11, stp1_12, stp1_13, stp1_14, stp1_15, + stp1_16, stp1_17, stp1_18, stp1_19, stp1_20, stp1_21, stp1_22, + stp1_23, stp1_24, stp1_25, stp1_26, stp1_27, stp1_28, stp1_29, + stp1_30, stp1_31; + __m128i stp2_0, stp2_1, stp2_2, stp2_3, stp2_4, stp2_5, stp2_6, stp2_7, + stp2_8, stp2_9, stp2_10, stp2_11, stp2_12, stp2_13, stp2_14, stp2_15, + stp2_16, stp2_17, stp2_18, stp2_19, stp2_20, stp2_21, stp2_22, + stp2_23, stp2_24, stp2_25, stp2_26, stp2_27, stp2_28, stp2_29, + stp2_30, stp2_31; + __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; + int i; + + // Load input data. Only need to load the top left 8x8 block. + in[0] = load_input_data(input); + in[1] = load_input_data(input + 32); + in[2] = load_input_data(input + 64); + in[3] = load_input_data(input + 96); + in[4] = load_input_data(input + 128); + in[5] = load_input_data(input + 160); + in[6] = load_input_data(input + 192); + in[7] = load_input_data(input + 224); + + for (i = 8; i < 32; ++i) { + in[i] = _mm_setzero_si128(); + } + + array_transpose_8x8(in, in); + // TODO(hkuang): Following transposes are unnecessary. But remove them will + // lead to performance drop on some devices. + array_transpose_8x8(in + 8, in + 8); + array_transpose_8x8(in + 16, in + 16); + array_transpose_8x8(in + 24, in + 24); + + IDCT32_34 + + // 1_D: Store 32 intermediate results for each 8x32 block. + col[0] = _mm_add_epi16(stp1_0, stp1_31); + col[1] = _mm_add_epi16(stp1_1, stp1_30); + col[2] = _mm_add_epi16(stp1_2, stp1_29); + col[3] = _mm_add_epi16(stp1_3, stp1_28); + col[4] = _mm_add_epi16(stp1_4, stp1_27); + col[5] = _mm_add_epi16(stp1_5, stp1_26); + col[6] = _mm_add_epi16(stp1_6, stp1_25); + col[7] = _mm_add_epi16(stp1_7, stp1_24); + col[8] = _mm_add_epi16(stp1_8, stp1_23); + col[9] = _mm_add_epi16(stp1_9, stp1_22); + col[10] = _mm_add_epi16(stp1_10, stp1_21); + col[11] = _mm_add_epi16(stp1_11, stp1_20); + col[12] = _mm_add_epi16(stp1_12, stp1_19); + col[13] = _mm_add_epi16(stp1_13, stp1_18); + col[14] = _mm_add_epi16(stp1_14, stp1_17); + col[15] = _mm_add_epi16(stp1_15, stp1_16); + col[16] = _mm_sub_epi16(stp1_15, stp1_16); + col[17] = _mm_sub_epi16(stp1_14, stp1_17); + col[18] = _mm_sub_epi16(stp1_13, stp1_18); + col[19] = _mm_sub_epi16(stp1_12, stp1_19); + col[20] = _mm_sub_epi16(stp1_11, stp1_20); + col[21] = _mm_sub_epi16(stp1_10, stp1_21); + col[22] = _mm_sub_epi16(stp1_9, stp1_22); + col[23] = _mm_sub_epi16(stp1_8, stp1_23); + col[24] = _mm_sub_epi16(stp1_7, stp1_24); + col[25] = _mm_sub_epi16(stp1_6, stp1_25); + col[26] = _mm_sub_epi16(stp1_5, stp1_26); + col[27] = _mm_sub_epi16(stp1_4, stp1_27); + col[28] = _mm_sub_epi16(stp1_3, stp1_28); + col[29] = _mm_sub_epi16(stp1_2, stp1_29); + col[30] = _mm_sub_epi16(stp1_1, stp1_30); + col[31] = _mm_sub_epi16(stp1_0, stp1_31); + for (i = 0; i < 4; i++) { + int j; + const __m128i zero = _mm_setzero_si128(); + // Transpose 32x8 block to 8x32 block + array_transpose_8x8(col + i * 8, in); + IDCT32_34 + + // 2_D: Calculate the results and store them to destination. + in[0] = _mm_add_epi16(stp1_0, stp1_31); + in[1] = _mm_add_epi16(stp1_1, stp1_30); + in[2] = _mm_add_epi16(stp1_2, stp1_29); + in[3] = _mm_add_epi16(stp1_3, stp1_28); + in[4] = _mm_add_epi16(stp1_4, stp1_27); + in[5] = _mm_add_epi16(stp1_5, stp1_26); + in[6] = _mm_add_epi16(stp1_6, stp1_25); + in[7] = _mm_add_epi16(stp1_7, stp1_24); + in[8] = _mm_add_epi16(stp1_8, stp1_23); + in[9] = _mm_add_epi16(stp1_9, stp1_22); + in[10] = _mm_add_epi16(stp1_10, stp1_21); + in[11] = _mm_add_epi16(stp1_11, stp1_20); + in[12] = _mm_add_epi16(stp1_12, stp1_19); + in[13] = _mm_add_epi16(stp1_13, stp1_18); + in[14] = _mm_add_epi16(stp1_14, stp1_17); + in[15] = _mm_add_epi16(stp1_15, stp1_16); + in[16] = _mm_sub_epi16(stp1_15, stp1_16); + in[17] = _mm_sub_epi16(stp1_14, stp1_17); + in[18] = _mm_sub_epi16(stp1_13, stp1_18); + in[19] = _mm_sub_epi16(stp1_12, stp1_19); + in[20] = _mm_sub_epi16(stp1_11, stp1_20); + in[21] = _mm_sub_epi16(stp1_10, stp1_21); + in[22] = _mm_sub_epi16(stp1_9, stp1_22); + in[23] = _mm_sub_epi16(stp1_8, stp1_23); + in[24] = _mm_sub_epi16(stp1_7, stp1_24); + in[25] = _mm_sub_epi16(stp1_6, stp1_25); + in[26] = _mm_sub_epi16(stp1_5, stp1_26); + in[27] = _mm_sub_epi16(stp1_4, stp1_27); + in[28] = _mm_sub_epi16(stp1_3, stp1_28); + in[29] = _mm_sub_epi16(stp1_2, stp1_29); + in[30] = _mm_sub_epi16(stp1_1, stp1_30); + in[31] = _mm_sub_epi16(stp1_0, stp1_31); + + for (j = 0; j < 32; ++j) { + // Final rounding and shift + in[j] = _mm_adds_epi16(in[j], final_rounding); + in[j] = _mm_srai_epi16(in[j], 6); + RECON_AND_STORE(dest + j * stride, in[j]); + } + + dest += 8; + } +} + +void vpx_idct32x32_1024_add_sse2(const tran_low_t *input, uint8_t *dest, + int stride) { + const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING); + const __m128i final_rounding = _mm_set1_epi16(1 << 5); + const __m128i zero = _mm_setzero_si128(); + + // idct constants for each stage + const __m128i stg1_0 = pair_set_epi16(cospi_31_64, -cospi_1_64); + const __m128i stg1_1 = pair_set_epi16(cospi_1_64, cospi_31_64); + const __m128i stg1_2 = pair_set_epi16(cospi_15_64, -cospi_17_64); + const __m128i stg1_3 = pair_set_epi16(cospi_17_64, cospi_15_64); + const __m128i stg1_4 = pair_set_epi16(cospi_23_64, -cospi_9_64); + const __m128i stg1_5 = pair_set_epi16(cospi_9_64, cospi_23_64); + const __m128i stg1_6 = pair_set_epi16(cospi_7_64, -cospi_25_64); + const __m128i stg1_7 = pair_set_epi16(cospi_25_64, cospi_7_64); + const __m128i stg1_8 = pair_set_epi16(cospi_27_64, -cospi_5_64); + const __m128i stg1_9 = pair_set_epi16(cospi_5_64, cospi_27_64); + const __m128i stg1_10 = pair_set_epi16(cospi_11_64, -cospi_21_64); + const __m128i stg1_11 = pair_set_epi16(cospi_21_64, cospi_11_64); + const __m128i stg1_12 = pair_set_epi16(cospi_19_64, -cospi_13_64); + const __m128i stg1_13 = pair_set_epi16(cospi_13_64, cospi_19_64); + const __m128i stg1_14 = pair_set_epi16(cospi_3_64, -cospi_29_64); + const __m128i stg1_15 = pair_set_epi16(cospi_29_64, cospi_3_64); + + const __m128i stg2_0 = pair_set_epi16(cospi_30_64, -cospi_2_64); + const __m128i stg2_1 = pair_set_epi16(cospi_2_64, cospi_30_64); + const __m128i stg2_2 = pair_set_epi16(cospi_14_64, -cospi_18_64); + const __m128i stg2_3 = pair_set_epi16(cospi_18_64, cospi_14_64); + const __m128i stg2_4 = pair_set_epi16(cospi_22_64, -cospi_10_64); + const __m128i stg2_5 = pair_set_epi16(cospi_10_64, cospi_22_64); + const __m128i stg2_6 = pair_set_epi16(cospi_6_64, -cospi_26_64); + const __m128i stg2_7 = pair_set_epi16(cospi_26_64, cospi_6_64); + + const __m128i stg3_0 = pair_set_epi16(cospi_28_64, -cospi_4_64); + const __m128i stg3_1 = pair_set_epi16(cospi_4_64, cospi_28_64); + const __m128i stg3_2 = pair_set_epi16(cospi_12_64, -cospi_20_64); + const __m128i stg3_3 = pair_set_epi16(cospi_20_64, cospi_12_64); + const __m128i stg3_4 = pair_set_epi16(-cospi_4_64, cospi_28_64); + const __m128i stg3_5 = pair_set_epi16(cospi_28_64, cospi_4_64); + const __m128i stg3_6 = pair_set_epi16(-cospi_28_64, -cospi_4_64); + const __m128i stg3_8 = pair_set_epi16(-cospi_20_64, cospi_12_64); + const __m128i stg3_9 = pair_set_epi16(cospi_12_64, cospi_20_64); + const __m128i stg3_10 = pair_set_epi16(-cospi_12_64, -cospi_20_64); + + const __m128i stg4_0 = pair_set_epi16(cospi_16_64, cospi_16_64); + const __m128i stg4_1 = pair_set_epi16(cospi_16_64, -cospi_16_64); + const __m128i stg4_2 = pair_set_epi16(cospi_24_64, -cospi_8_64); + const __m128i stg4_3 = pair_set_epi16(cospi_8_64, cospi_24_64); + const __m128i stg4_4 = pair_set_epi16(-cospi_8_64, cospi_24_64); + const __m128i stg4_5 = pair_set_epi16(cospi_24_64, cospi_8_64); + const __m128i stg4_6 = pair_set_epi16(-cospi_24_64, -cospi_8_64); + + const __m128i stg6_0 = pair_set_epi16(-cospi_16_64, cospi_16_64); + + __m128i in[32], col[128], zero_idx[16]; + __m128i stp1_0, stp1_1, stp1_2, stp1_3, stp1_4, stp1_5, stp1_6, stp1_7, + stp1_8, stp1_9, stp1_10, stp1_11, stp1_12, stp1_13, stp1_14, stp1_15, + stp1_16, stp1_17, stp1_18, stp1_19, stp1_20, stp1_21, stp1_22, + stp1_23, stp1_24, stp1_25, stp1_26, stp1_27, stp1_28, stp1_29, + stp1_30, stp1_31; + __m128i stp2_0, stp2_1, stp2_2, stp2_3, stp2_4, stp2_5, stp2_6, stp2_7, + stp2_8, stp2_9, stp2_10, stp2_11, stp2_12, stp2_13, stp2_14, stp2_15, + stp2_16, stp2_17, stp2_18, stp2_19, stp2_20, stp2_21, stp2_22, + stp2_23, stp2_24, stp2_25, stp2_26, stp2_27, stp2_28, stp2_29, + stp2_30, stp2_31; + __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; + int i, j, i32; + + for (i = 0; i < 4; i++) { + i32 = (i << 5); + // First 1-D idct + // Load input data. + LOAD_DQCOEFF(in[0], input); + LOAD_DQCOEFF(in[8], input); + LOAD_DQCOEFF(in[16], input); + LOAD_DQCOEFF(in[24], input); + LOAD_DQCOEFF(in[1], input); + LOAD_DQCOEFF(in[9], input); + LOAD_DQCOEFF(in[17], input); + LOAD_DQCOEFF(in[25], input); + LOAD_DQCOEFF(in[2], input); + LOAD_DQCOEFF(in[10], input); + LOAD_DQCOEFF(in[18], input); + LOAD_DQCOEFF(in[26], input); + LOAD_DQCOEFF(in[3], input); + LOAD_DQCOEFF(in[11], input); + LOAD_DQCOEFF(in[19], input); + LOAD_DQCOEFF(in[27], input); + + LOAD_DQCOEFF(in[4], input); + LOAD_DQCOEFF(in[12], input); + LOAD_DQCOEFF(in[20], input); + LOAD_DQCOEFF(in[28], input); + LOAD_DQCOEFF(in[5], input); + LOAD_DQCOEFF(in[13], input); + LOAD_DQCOEFF(in[21], input); + LOAD_DQCOEFF(in[29], input); + LOAD_DQCOEFF(in[6], input); + LOAD_DQCOEFF(in[14], input); + LOAD_DQCOEFF(in[22], input); + LOAD_DQCOEFF(in[30], input); + LOAD_DQCOEFF(in[7], input); + LOAD_DQCOEFF(in[15], input); + LOAD_DQCOEFF(in[23], input); + LOAD_DQCOEFF(in[31], input); + + // checking if all entries are zero + zero_idx[0] = _mm_or_si128(in[0], in[1]); + zero_idx[1] = _mm_or_si128(in[2], in[3]); + zero_idx[2] = _mm_or_si128(in[4], in[5]); + zero_idx[3] = _mm_or_si128(in[6], in[7]); + zero_idx[4] = _mm_or_si128(in[8], in[9]); + zero_idx[5] = _mm_or_si128(in[10], in[11]); + zero_idx[6] = _mm_or_si128(in[12], in[13]); + zero_idx[7] = _mm_or_si128(in[14], in[15]); + zero_idx[8] = _mm_or_si128(in[16], in[17]); + zero_idx[9] = _mm_or_si128(in[18], in[19]); + zero_idx[10] = _mm_or_si128(in[20], in[21]); + zero_idx[11] = _mm_or_si128(in[22], in[23]); + zero_idx[12] = _mm_or_si128(in[24], in[25]); + zero_idx[13] = _mm_or_si128(in[26], in[27]); + zero_idx[14] = _mm_or_si128(in[28], in[29]); + zero_idx[15] = _mm_or_si128(in[30], in[31]); + + zero_idx[0] = _mm_or_si128(zero_idx[0], zero_idx[1]); + zero_idx[1] = _mm_or_si128(zero_idx[2], zero_idx[3]); + zero_idx[2] = _mm_or_si128(zero_idx[4], zero_idx[5]); + zero_idx[3] = _mm_or_si128(zero_idx[6], zero_idx[7]); + zero_idx[4] = _mm_or_si128(zero_idx[8], zero_idx[9]); + zero_idx[5] = _mm_or_si128(zero_idx[10], zero_idx[11]); + zero_idx[6] = _mm_or_si128(zero_idx[12], zero_idx[13]); + zero_idx[7] = _mm_or_si128(zero_idx[14], zero_idx[15]); + + zero_idx[8] = _mm_or_si128(zero_idx[0], zero_idx[1]); + zero_idx[9] = _mm_or_si128(zero_idx[2], zero_idx[3]); + zero_idx[10] = _mm_or_si128(zero_idx[4], zero_idx[5]); + zero_idx[11] = _mm_or_si128(zero_idx[6], zero_idx[7]); + zero_idx[12] = _mm_or_si128(zero_idx[8], zero_idx[9]); + zero_idx[13] = _mm_or_si128(zero_idx[10], zero_idx[11]); + zero_idx[14] = _mm_or_si128(zero_idx[12], zero_idx[13]); + + if (_mm_movemask_epi8(_mm_cmpeq_epi32(zero_idx[14], zero)) == 0xFFFF) { + col[i32 + 0] = _mm_setzero_si128(); + col[i32 + 1] = _mm_setzero_si128(); + col[i32 + 2] = _mm_setzero_si128(); + col[i32 + 3] = _mm_setzero_si128(); + col[i32 + 4] = _mm_setzero_si128(); + col[i32 + 5] = _mm_setzero_si128(); + col[i32 + 6] = _mm_setzero_si128(); + col[i32 + 7] = _mm_setzero_si128(); + col[i32 + 8] = _mm_setzero_si128(); + col[i32 + 9] = _mm_setzero_si128(); + col[i32 + 10] = _mm_setzero_si128(); + col[i32 + 11] = _mm_setzero_si128(); + col[i32 + 12] = _mm_setzero_si128(); + col[i32 + 13] = _mm_setzero_si128(); + col[i32 + 14] = _mm_setzero_si128(); + col[i32 + 15] = _mm_setzero_si128(); + col[i32 + 16] = _mm_setzero_si128(); + col[i32 + 17] = _mm_setzero_si128(); + col[i32 + 18] = _mm_setzero_si128(); + col[i32 + 19] = _mm_setzero_si128(); + col[i32 + 20] = _mm_setzero_si128(); + col[i32 + 21] = _mm_setzero_si128(); + col[i32 + 22] = _mm_setzero_si128(); + col[i32 + 23] = _mm_setzero_si128(); + col[i32 + 24] = _mm_setzero_si128(); + col[i32 + 25] = _mm_setzero_si128(); + col[i32 + 26] = _mm_setzero_si128(); + col[i32 + 27] = _mm_setzero_si128(); + col[i32 + 28] = _mm_setzero_si128(); + col[i32 + 29] = _mm_setzero_si128(); + col[i32 + 30] = _mm_setzero_si128(); + col[i32 + 31] = _mm_setzero_si128(); + continue; + } + + // Transpose 32x8 block to 8x32 block + array_transpose_8x8(in, in); + array_transpose_8x8(in + 8, in + 8); + array_transpose_8x8(in + 16, in + 16); + array_transpose_8x8(in + 24, in + 24); + + IDCT32 + + // 1_D: Store 32 intermediate results for each 8x32 block. + col[i32 + 0] = _mm_add_epi16(stp1_0, stp1_31); + col[i32 + 1] = _mm_add_epi16(stp1_1, stp1_30); + col[i32 + 2] = _mm_add_epi16(stp1_2, stp1_29); + col[i32 + 3] = _mm_add_epi16(stp1_3, stp1_28); + col[i32 + 4] = _mm_add_epi16(stp1_4, stp1_27); + col[i32 + 5] = _mm_add_epi16(stp1_5, stp1_26); + col[i32 + 6] = _mm_add_epi16(stp1_6, stp1_25); + col[i32 + 7] = _mm_add_epi16(stp1_7, stp1_24); + col[i32 + 8] = _mm_add_epi16(stp1_8, stp1_23); + col[i32 + 9] = _mm_add_epi16(stp1_9, stp1_22); + col[i32 + 10] = _mm_add_epi16(stp1_10, stp1_21); + col[i32 + 11] = _mm_add_epi16(stp1_11, stp1_20); + col[i32 + 12] = _mm_add_epi16(stp1_12, stp1_19); + col[i32 + 13] = _mm_add_epi16(stp1_13, stp1_18); + col[i32 + 14] = _mm_add_epi16(stp1_14, stp1_17); + col[i32 + 15] = _mm_add_epi16(stp1_15, stp1_16); + col[i32 + 16] = _mm_sub_epi16(stp1_15, stp1_16); + col[i32 + 17] = _mm_sub_epi16(stp1_14, stp1_17); + col[i32 + 18] = _mm_sub_epi16(stp1_13, stp1_18); + col[i32 + 19] = _mm_sub_epi16(stp1_12, stp1_19); + col[i32 + 20] = _mm_sub_epi16(stp1_11, stp1_20); + col[i32 + 21] = _mm_sub_epi16(stp1_10, stp1_21); + col[i32 + 22] = _mm_sub_epi16(stp1_9, stp1_22); + col[i32 + 23] = _mm_sub_epi16(stp1_8, stp1_23); + col[i32 + 24] = _mm_sub_epi16(stp1_7, stp1_24); + col[i32 + 25] = _mm_sub_epi16(stp1_6, stp1_25); + col[i32 + 26] = _mm_sub_epi16(stp1_5, stp1_26); + col[i32 + 27] = _mm_sub_epi16(stp1_4, stp1_27); + col[i32 + 28] = _mm_sub_epi16(stp1_3, stp1_28); + col[i32 + 29] = _mm_sub_epi16(stp1_2, stp1_29); + col[i32 + 30] = _mm_sub_epi16(stp1_1, stp1_30); + col[i32 + 31] = _mm_sub_epi16(stp1_0, stp1_31); + } + for (i = 0; i < 4; i++) { + // Second 1-D idct + j = i << 3; + + // Transpose 32x8 block to 8x32 block + array_transpose_8x8(col + j, in); + array_transpose_8x8(col + j + 32, in + 8); + array_transpose_8x8(col + j + 64, in + 16); + array_transpose_8x8(col + j + 96, in + 24); + + IDCT32 + + // 2_D: Calculate the results and store them to destination. + in[0] = _mm_add_epi16(stp1_0, stp1_31); + in[1] = _mm_add_epi16(stp1_1, stp1_30); + in[2] = _mm_add_epi16(stp1_2, stp1_29); + in[3] = _mm_add_epi16(stp1_3, stp1_28); + in[4] = _mm_add_epi16(stp1_4, stp1_27); + in[5] = _mm_add_epi16(stp1_5, stp1_26); + in[6] = _mm_add_epi16(stp1_6, stp1_25); + in[7] = _mm_add_epi16(stp1_7, stp1_24); + in[8] = _mm_add_epi16(stp1_8, stp1_23); + in[9] = _mm_add_epi16(stp1_9, stp1_22); + in[10] = _mm_add_epi16(stp1_10, stp1_21); + in[11] = _mm_add_epi16(stp1_11, stp1_20); + in[12] = _mm_add_epi16(stp1_12, stp1_19); + in[13] = _mm_add_epi16(stp1_13, stp1_18); + in[14] = _mm_add_epi16(stp1_14, stp1_17); + in[15] = _mm_add_epi16(stp1_15, stp1_16); + in[16] = _mm_sub_epi16(stp1_15, stp1_16); + in[17] = _mm_sub_epi16(stp1_14, stp1_17); + in[18] = _mm_sub_epi16(stp1_13, stp1_18); + in[19] = _mm_sub_epi16(stp1_12, stp1_19); + in[20] = _mm_sub_epi16(stp1_11, stp1_20); + in[21] = _mm_sub_epi16(stp1_10, stp1_21); + in[22] = _mm_sub_epi16(stp1_9, stp1_22); + in[23] = _mm_sub_epi16(stp1_8, stp1_23); + in[24] = _mm_sub_epi16(stp1_7, stp1_24); + in[25] = _mm_sub_epi16(stp1_6, stp1_25); + in[26] = _mm_sub_epi16(stp1_5, stp1_26); + in[27] = _mm_sub_epi16(stp1_4, stp1_27); + in[28] = _mm_sub_epi16(stp1_3, stp1_28); + in[29] = _mm_sub_epi16(stp1_2, stp1_29); + in[30] = _mm_sub_epi16(stp1_1, stp1_30); + in[31] = _mm_sub_epi16(stp1_0, stp1_31); + + for (j = 0; j < 32; ++j) { + // Final rounding and shift + in[j] = _mm_adds_epi16(in[j], final_rounding); + in[j] = _mm_srai_epi16(in[j], 6); + RECON_AND_STORE(dest + j * stride, in[j]); + } + + dest += 8; + } +} + +void vpx_idct32x32_1_add_sse2(const tran_low_t *input, uint8_t *dest, + int stride) { + __m128i dc_value; + const __m128i zero = _mm_setzero_si128(); + int a, j; + + a = (int)dct_const_round_shift(input[0] * cospi_16_64); + a = (int)dct_const_round_shift(a * cospi_16_64); + a = ROUND_POWER_OF_TWO(a, 6); + + dc_value = _mm_set1_epi16(a); + + for (j = 0; j < 32; ++j) { + RECON_AND_STORE(dest + 0 + j * stride, dc_value); + RECON_AND_STORE(dest + 8 + j * stride, dc_value); + RECON_AND_STORE(dest + 16 + j * stride, dc_value); + RECON_AND_STORE(dest + 24 + j * stride, dc_value); + } +} + +#if CONFIG_VP9_HIGHBITDEPTH +static INLINE __m128i clamp_high_sse2(__m128i value, int bd) { + __m128i ubounded, retval; + const __m128i zero = _mm_set1_epi16(0); + const __m128i one = _mm_set1_epi16(1); + const __m128i max = _mm_subs_epi16(_mm_slli_epi16(one, bd), one); + ubounded = _mm_cmpgt_epi16(value, max); + retval = _mm_andnot_si128(ubounded, value); + ubounded = _mm_and_si128(ubounded, max); + retval = _mm_or_si128(retval, ubounded); + retval = _mm_and_si128(retval, _mm_cmpgt_epi16(retval, zero)); + return retval; +} + +void vpx_highbd_idct4x4_16_add_sse2(const tran_low_t *input, uint8_t *dest8, + int stride, int bd) { + tran_low_t out[4 * 4]; + tran_low_t *outptr = out; + int i, j; + __m128i inptr[4]; + __m128i sign_bits[2]; + __m128i temp_mm, min_input, max_input; + int test; + uint16_t *dest = CONVERT_TO_SHORTPTR(dest8); + int optimised_cols = 0; + const __m128i zero = _mm_set1_epi16(0); + const __m128i eight = _mm_set1_epi16(8); + const __m128i max = _mm_set1_epi16(12043); + const __m128i min = _mm_set1_epi16(-12043); + // Load input into __m128i + inptr[0] = _mm_loadu_si128((const __m128i *)input); + inptr[1] = _mm_loadu_si128((const __m128i *)(input + 4)); + inptr[2] = _mm_loadu_si128((const __m128i *)(input + 8)); + inptr[3] = _mm_loadu_si128((const __m128i *)(input + 12)); + + // Pack to 16 bits + inptr[0] = _mm_packs_epi32(inptr[0], inptr[1]); + inptr[1] = _mm_packs_epi32(inptr[2], inptr[3]); + + max_input = _mm_max_epi16(inptr[0], inptr[1]); + min_input = _mm_min_epi16(inptr[0], inptr[1]); + max_input = _mm_cmpgt_epi16(max_input, max); + min_input = _mm_cmplt_epi16(min_input, min); + temp_mm = _mm_or_si128(max_input, min_input); + test = _mm_movemask_epi8(temp_mm); + + if (!test) { + // Do the row transform + idct4_sse2(inptr); + + // Check the min & max values + max_input = _mm_max_epi16(inptr[0], inptr[1]); + min_input = _mm_min_epi16(inptr[0], inptr[1]); + max_input = _mm_cmpgt_epi16(max_input, max); + min_input = _mm_cmplt_epi16(min_input, min); + temp_mm = _mm_or_si128(max_input, min_input); + test = _mm_movemask_epi8(temp_mm); + + if (test) { + transpose_4x4(inptr); + sign_bits[0] = _mm_cmplt_epi16(inptr[0], zero); + sign_bits[1] = _mm_cmplt_epi16(inptr[1], zero); + inptr[3] = _mm_unpackhi_epi16(inptr[1], sign_bits[1]); + inptr[2] = _mm_unpacklo_epi16(inptr[1], sign_bits[1]); + inptr[1] = _mm_unpackhi_epi16(inptr[0], sign_bits[0]); + inptr[0] = _mm_unpacklo_epi16(inptr[0], sign_bits[0]); + _mm_storeu_si128((__m128i *)outptr, inptr[0]); + _mm_storeu_si128((__m128i *)(outptr + 4), inptr[1]); + _mm_storeu_si128((__m128i *)(outptr + 8), inptr[2]); + _mm_storeu_si128((__m128i *)(outptr + 12), inptr[3]); + } else { + // Set to use the optimised transform for the column + optimised_cols = 1; + } + } else { + // Run the un-optimised row transform + for (i = 0; i < 4; ++i) { + vpx_highbd_idct4_c(input, outptr, bd); + input += 4; + outptr += 4; + } + } + + if (optimised_cols) { + idct4_sse2(inptr); + + // Final round and shift + inptr[0] = _mm_add_epi16(inptr[0], eight); + inptr[1] = _mm_add_epi16(inptr[1], eight); + + inptr[0] = _mm_srai_epi16(inptr[0], 4); + inptr[1] = _mm_srai_epi16(inptr[1], 4); + + // Reconstruction and Store + { + __m128i d0 = _mm_loadl_epi64((const __m128i *)dest); + __m128i d2 = _mm_loadl_epi64((const __m128i *)(dest + stride * 2)); + d0 = _mm_unpacklo_epi64( + d0, _mm_loadl_epi64((const __m128i *)(dest + stride))); + d2 = _mm_unpacklo_epi64( + d2, _mm_loadl_epi64((const __m128i *)(dest + stride * 3))); + d0 = clamp_high_sse2(_mm_adds_epi16(d0, inptr[0]), bd); + d2 = clamp_high_sse2(_mm_adds_epi16(d2, inptr[1]), bd); + // store input0 + _mm_storel_epi64((__m128i *)dest, d0); + // store input1 + d0 = _mm_srli_si128(d0, 8); + _mm_storel_epi64((__m128i *)(dest + stride), d0); + // store input2 + _mm_storel_epi64((__m128i *)(dest + stride * 2), d2); + // store input3 + d2 = _mm_srli_si128(d2, 8); + _mm_storel_epi64((__m128i *)(dest + stride * 3), d2); + } + } else { + // Run the un-optimised column transform + tran_low_t temp_in[4], temp_out[4]; + // Columns + for (i = 0; i < 4; ++i) { + for (j = 0; j < 4; ++j) + temp_in[j] = out[j * 4 + i]; + vpx_highbd_idct4_c(temp_in, temp_out, bd); + for (j = 0; j < 4; ++j) { + dest[j * stride + i] = highbd_clip_pixel_add( + dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 4), bd); + } + } + } +} + +void vpx_highbd_idct8x8_64_add_sse2(const tran_low_t *input, uint8_t *dest8, + int stride, int bd) { + tran_low_t out[8 * 8]; + tran_low_t *outptr = out; + int i, j, test; + __m128i inptr[8]; + __m128i min_input, max_input, temp1, temp2, sign_bits; + uint16_t *dest = CONVERT_TO_SHORTPTR(dest8); + const __m128i zero = _mm_set1_epi16(0); + const __m128i sixteen = _mm_set1_epi16(16); + const __m128i max = _mm_set1_epi16(6201); + const __m128i min = _mm_set1_epi16(-6201); + int optimised_cols = 0; + + // Load input into __m128i & pack to 16 bits + for (i = 0; i < 8; i++) { + temp1 = _mm_loadu_si128((const __m128i *)(input + 8 * i)); + temp2 = _mm_loadu_si128((const __m128i *)(input + 8 * i + 4)); + inptr[i] = _mm_packs_epi32(temp1, temp2); + } + + // Find the min & max for the row transform + max_input = _mm_max_epi16(inptr[0], inptr[1]); + min_input = _mm_min_epi16(inptr[0], inptr[1]); + for (i = 2; i < 8; i++) { + max_input = _mm_max_epi16(max_input, inptr[i]); + min_input = _mm_min_epi16(min_input, inptr[i]); + } + max_input = _mm_cmpgt_epi16(max_input, max); + min_input = _mm_cmplt_epi16(min_input, min); + temp1 = _mm_or_si128(max_input, min_input); + test = _mm_movemask_epi8(temp1); + + if (!test) { + // Do the row transform + idct8_sse2(inptr); + + // Find the min & max for the column transform + max_input = _mm_max_epi16(inptr[0], inptr[1]); + min_input = _mm_min_epi16(inptr[0], inptr[1]); + for (i = 2; i < 8; i++) { + max_input = _mm_max_epi16(max_input, inptr[i]); + min_input = _mm_min_epi16(min_input, inptr[i]); + } + max_input = _mm_cmpgt_epi16(max_input, max); + min_input = _mm_cmplt_epi16(min_input, min); + temp1 = _mm_or_si128(max_input, min_input); + test = _mm_movemask_epi8(temp1); + + if (test) { + array_transpose_8x8(inptr, inptr); + for (i = 0; i < 8; i++) { + sign_bits = _mm_cmplt_epi16(inptr[i], zero); + temp1 = _mm_unpackhi_epi16(inptr[i], sign_bits); + temp2 = _mm_unpacklo_epi16(inptr[i], sign_bits); + _mm_storeu_si128((__m128i *)(outptr + 4 * (2 * i + 1)), temp1); + _mm_storeu_si128((__m128i *)(outptr + 4 * (2 * i)), temp2); + } + } else { + // Set to use the optimised transform for the column + optimised_cols = 1; + } + } else { + // Run the un-optimised row transform + for (i = 0; i < 8; ++i) { + vpx_highbd_idct8_c(input, outptr, bd); + input += 8; + outptr += 8; + } + } + + if (optimised_cols) { + idct8_sse2(inptr); + + // Final round & shift and Reconstruction and Store + { + __m128i d[8]; + for (i = 0; i < 8; i++) { + inptr[i] = _mm_add_epi16(inptr[i], sixteen); + d[i] = _mm_loadu_si128((const __m128i *)(dest + stride*i)); + inptr[i] = _mm_srai_epi16(inptr[i], 5); + d[i] = clamp_high_sse2(_mm_adds_epi16(d[i], inptr[i]), bd); + // Store + _mm_storeu_si128((__m128i *)(dest + stride*i), d[i]); + } + } + } else { + // Run the un-optimised column transform + tran_low_t temp_in[8], temp_out[8]; + for (i = 0; i < 8; ++i) { + for (j = 0; j < 8; ++j) + temp_in[j] = out[j * 8 + i]; + vpx_highbd_idct8_c(temp_in, temp_out, bd); + for (j = 0; j < 8; ++j) { + dest[j * stride + i] = highbd_clip_pixel_add( + dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 5), bd); + } + } + } +} + +void vpx_highbd_idct8x8_10_add_sse2(const tran_low_t *input, uint8_t *dest8, + int stride, int bd) { + tran_low_t out[8 * 8] = { 0 }; + tran_low_t *outptr = out; + int i, j, test; + __m128i inptr[8]; + __m128i min_input, max_input, temp1, temp2, sign_bits; + uint16_t *dest = CONVERT_TO_SHORTPTR(dest8); + const __m128i zero = _mm_set1_epi16(0); + const __m128i sixteen = _mm_set1_epi16(16); + const __m128i max = _mm_set1_epi16(6201); + const __m128i min = _mm_set1_epi16(-6201); + int optimised_cols = 0; + + // Load input into __m128i & pack to 16 bits + for (i = 0; i < 8; i++) { + temp1 = _mm_loadu_si128((const __m128i *)(input + 8 * i)); + temp2 = _mm_loadu_si128((const __m128i *)(input + 8 * i + 4)); + inptr[i] = _mm_packs_epi32(temp1, temp2); + } + + // Find the min & max for the row transform + // only first 4 row has non-zero coefs + max_input = _mm_max_epi16(inptr[0], inptr[1]); + min_input = _mm_min_epi16(inptr[0], inptr[1]); + for (i = 2; i < 4; i++) { + max_input = _mm_max_epi16(max_input, inptr[i]); + min_input = _mm_min_epi16(min_input, inptr[i]); + } + max_input = _mm_cmpgt_epi16(max_input, max); + min_input = _mm_cmplt_epi16(min_input, min); + temp1 = _mm_or_si128(max_input, min_input); + test = _mm_movemask_epi8(temp1); + + if (!test) { + // Do the row transform + idct8_sse2(inptr); + + // Find the min & max for the column transform + // N.B. Only first 4 cols contain non-zero coeffs + max_input = _mm_max_epi16(inptr[0], inptr[1]); + min_input = _mm_min_epi16(inptr[0], inptr[1]); + for (i = 2; i < 8; i++) { + max_input = _mm_max_epi16(max_input, inptr[i]); + min_input = _mm_min_epi16(min_input, inptr[i]); + } + max_input = _mm_cmpgt_epi16(max_input, max); + min_input = _mm_cmplt_epi16(min_input, min); + temp1 = _mm_or_si128(max_input, min_input); + test = _mm_movemask_epi8(temp1); + + if (test) { + // Use fact only first 4 rows contain non-zero coeffs + array_transpose_4X8(inptr, inptr); + for (i = 0; i < 4; i++) { + sign_bits = _mm_cmplt_epi16(inptr[i], zero); + temp1 = _mm_unpackhi_epi16(inptr[i], sign_bits); + temp2 = _mm_unpacklo_epi16(inptr[i], sign_bits); + _mm_storeu_si128((__m128i *)(outptr + 4 * (2 * i + 1)), temp1); + _mm_storeu_si128((__m128i *)(outptr + 4 * (2 * i)), temp2); + } + } else { + // Set to use the optimised transform for the column + optimised_cols = 1; + } + } else { + // Run the un-optimised row transform + for (i = 0; i < 4; ++i) { + vpx_highbd_idct8_c(input, outptr, bd); + input += 8; + outptr += 8; + } + } + + if (optimised_cols) { + idct8_sse2(inptr); + + // Final round & shift and Reconstruction and Store + { + __m128i d[8]; + for (i = 0; i < 8; i++) { + inptr[i] = _mm_add_epi16(inptr[i], sixteen); + d[i] = _mm_loadu_si128((const __m128i *)(dest + stride*i)); + inptr[i] = _mm_srai_epi16(inptr[i], 5); + d[i] = clamp_high_sse2(_mm_adds_epi16(d[i], inptr[i]), bd); + // Store + _mm_storeu_si128((__m128i *)(dest + stride*i), d[i]); + } + } + } else { + // Run the un-optimised column transform + tran_low_t temp_in[8], temp_out[8]; + for (i = 0; i < 8; ++i) { + for (j = 0; j < 8; ++j) + temp_in[j] = out[j * 8 + i]; + vpx_highbd_idct8_c(temp_in, temp_out, bd); + for (j = 0; j < 8; ++j) { + dest[j * stride + i] = highbd_clip_pixel_add( + dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 5), bd); + } + } + } +} + +void vpx_highbd_idct16x16_256_add_sse2(const tran_low_t *input, uint8_t *dest8, + int stride, int bd) { + tran_low_t out[16 * 16]; + tran_low_t *outptr = out; + int i, j, test; + __m128i inptr[32]; + __m128i min_input, max_input, temp1, temp2, sign_bits; + uint16_t *dest = CONVERT_TO_SHORTPTR(dest8); + const __m128i zero = _mm_set1_epi16(0); + const __m128i rounding = _mm_set1_epi16(32); + const __m128i max = _mm_set1_epi16(3155); + const __m128i min = _mm_set1_epi16(-3155); + int optimised_cols = 0; + + // Load input into __m128i & pack to 16 bits + for (i = 0; i < 16; i++) { + temp1 = _mm_loadu_si128((const __m128i *)(input + 16 * i)); + temp2 = _mm_loadu_si128((const __m128i *)(input + 16 * i + 4)); + inptr[i] = _mm_packs_epi32(temp1, temp2); + temp1 = _mm_loadu_si128((const __m128i *)(input + 16 * i + 8)); + temp2 = _mm_loadu_si128((const __m128i *)(input + 16 * i + 12)); + inptr[i + 16] = _mm_packs_epi32(temp1, temp2); + } + + // Find the min & max for the row transform + max_input = _mm_max_epi16(inptr[0], inptr[1]); + min_input = _mm_min_epi16(inptr[0], inptr[1]); + for (i = 2; i < 32; i++) { + max_input = _mm_max_epi16(max_input, inptr[i]); + min_input = _mm_min_epi16(min_input, inptr[i]); + } + max_input = _mm_cmpgt_epi16(max_input, max); + min_input = _mm_cmplt_epi16(min_input, min); + temp1 = _mm_or_si128(max_input, min_input); + test = _mm_movemask_epi8(temp1); + + if (!test) { + // Do the row transform + idct16_sse2(inptr, inptr + 16); + + // Find the min & max for the column transform + max_input = _mm_max_epi16(inptr[0], inptr[1]); + min_input = _mm_min_epi16(inptr[0], inptr[1]); + for (i = 2; i < 32; i++) { + max_input = _mm_max_epi16(max_input, inptr[i]); + min_input = _mm_min_epi16(min_input, inptr[i]); + } + max_input = _mm_cmpgt_epi16(max_input, max); + min_input = _mm_cmplt_epi16(min_input, min); + temp1 = _mm_or_si128(max_input, min_input); + test = _mm_movemask_epi8(temp1); + + if (test) { + array_transpose_16x16(inptr, inptr + 16); + for (i = 0; i < 16; i++) { + sign_bits = _mm_cmplt_epi16(inptr[i], zero); + temp1 = _mm_unpacklo_epi16(inptr[i], sign_bits); + temp2 = _mm_unpackhi_epi16(inptr[i], sign_bits); + _mm_storeu_si128((__m128i *)(outptr + 4 * (i * 4)), temp1); + _mm_storeu_si128((__m128i *)(outptr + 4 * (i * 4 + 1)), temp2); + sign_bits = _mm_cmplt_epi16(inptr[i + 16], zero); + temp1 = _mm_unpacklo_epi16(inptr[i + 16], sign_bits); + temp2 = _mm_unpackhi_epi16(inptr[i + 16], sign_bits); + _mm_storeu_si128((__m128i *)(outptr + 4 * (i * 4 + 2)), temp1); + _mm_storeu_si128((__m128i *)(outptr + 4 * (i * 4 + 3)), temp2); + } + } else { + // Set to use the optimised transform for the column + optimised_cols = 1; + } + } else { + // Run the un-optimised row transform + for (i = 0; i < 16; ++i) { + vpx_highbd_idct16_c(input, outptr, bd); + input += 16; + outptr += 16; + } + } + + if (optimised_cols) { + idct16_sse2(inptr, inptr + 16); + + // Final round & shift and Reconstruction and Store + { + __m128i d[2]; + for (i = 0; i < 16; i++) { + inptr[i ] = _mm_add_epi16(inptr[i ], rounding); + inptr[i+16] = _mm_add_epi16(inptr[i+16], rounding); + d[0] = _mm_loadu_si128((const __m128i *)(dest + stride*i)); + d[1] = _mm_loadu_si128((const __m128i *)(dest + stride*i + 8)); + inptr[i ] = _mm_srai_epi16(inptr[i ], 6); + inptr[i+16] = _mm_srai_epi16(inptr[i+16], 6); + d[0] = clamp_high_sse2(_mm_add_epi16(d[0], inptr[i ]), bd); + d[1] = clamp_high_sse2(_mm_add_epi16(d[1], inptr[i+16]), bd); + // Store + _mm_storeu_si128((__m128i *)(dest + stride*i), d[0]); + _mm_storeu_si128((__m128i *)(dest + stride*i + 8), d[1]); + } + } + } else { + // Run the un-optimised column transform + tran_low_t temp_in[16], temp_out[16]; + for (i = 0; i < 16; ++i) { + for (j = 0; j < 16; ++j) + temp_in[j] = out[j * 16 + i]; + vpx_highbd_idct16_c(temp_in, temp_out, bd); + for (j = 0; j < 16; ++j) { + dest[j * stride + i] = highbd_clip_pixel_add( + dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 6), bd); + } + } + } +} + +void vpx_highbd_idct16x16_10_add_sse2(const tran_low_t *input, uint8_t *dest8, + int stride, int bd) { + tran_low_t out[16 * 16] = { 0 }; + tran_low_t *outptr = out; + int i, j, test; + __m128i inptr[32]; + __m128i min_input, max_input, temp1, temp2, sign_bits; + uint16_t *dest = CONVERT_TO_SHORTPTR(dest8); + const __m128i zero = _mm_set1_epi16(0); + const __m128i rounding = _mm_set1_epi16(32); + const __m128i max = _mm_set1_epi16(3155); + const __m128i min = _mm_set1_epi16(-3155); + int optimised_cols = 0; + + // Load input into __m128i & pack to 16 bits + for (i = 0; i < 16; i++) { + temp1 = _mm_loadu_si128((const __m128i *)(input + 16 * i)); + temp2 = _mm_loadu_si128((const __m128i *)(input + 16 * i + 4)); + inptr[i] = _mm_packs_epi32(temp1, temp2); + temp1 = _mm_loadu_si128((const __m128i *)(input + 16 * i + 8)); + temp2 = _mm_loadu_si128((const __m128i *)(input + 16 * i + 12)); + inptr[i + 16] = _mm_packs_epi32(temp1, temp2); + } + + // Find the min & max for the row transform + // Since all non-zero dct coefficients are in upper-left 4x4 area, + // we only need to consider first 4 rows here. + max_input = _mm_max_epi16(inptr[0], inptr[1]); + min_input = _mm_min_epi16(inptr[0], inptr[1]); + for (i = 2; i < 4; i++) { + max_input = _mm_max_epi16(max_input, inptr[i]); + min_input = _mm_min_epi16(min_input, inptr[i]); + } + max_input = _mm_cmpgt_epi16(max_input, max); + min_input = _mm_cmplt_epi16(min_input, min); + temp1 = _mm_or_si128(max_input, min_input); + test = _mm_movemask_epi8(temp1); + + if (!test) { + // Do the row transform (N.B. This transposes inptr) + idct16_sse2(inptr, inptr + 16); + + // Find the min & max for the column transform + // N.B. Only first 4 cols contain non-zero coeffs + max_input = _mm_max_epi16(inptr[0], inptr[1]); + min_input = _mm_min_epi16(inptr[0], inptr[1]); + for (i = 2; i < 16; i++) { + max_input = _mm_max_epi16(max_input, inptr[i]); + min_input = _mm_min_epi16(min_input, inptr[i]); + } + max_input = _mm_cmpgt_epi16(max_input, max); + min_input = _mm_cmplt_epi16(min_input, min); + temp1 = _mm_or_si128(max_input, min_input); + test = _mm_movemask_epi8(temp1); + + if (test) { + // Use fact only first 4 rows contain non-zero coeffs + array_transpose_8x8(inptr, inptr); + array_transpose_8x8(inptr + 8, inptr + 16); + for (i = 0; i < 4; i++) { + sign_bits = _mm_cmplt_epi16(inptr[i], zero); + temp1 = _mm_unpacklo_epi16(inptr[i], sign_bits); + temp2 = _mm_unpackhi_epi16(inptr[i], sign_bits); + _mm_storeu_si128((__m128i *)(outptr + 4 * (i * 4)), temp1); + _mm_storeu_si128((__m128i *)(outptr + 4 * (i * 4 + 1)), temp2); + sign_bits = _mm_cmplt_epi16(inptr[i + 16], zero); + temp1 = _mm_unpacklo_epi16(inptr[i + 16], sign_bits); + temp2 = _mm_unpackhi_epi16(inptr[i + 16], sign_bits); + _mm_storeu_si128((__m128i *)(outptr + 4 * (i * 4 + 2)), temp1); + _mm_storeu_si128((__m128i *)(outptr + 4 * (i * 4 + 3)), temp2); + } + } else { + // Set to use the optimised transform for the column + optimised_cols = 1; + } + } else { + // Run the un-optimised row transform + for (i = 0; i < 4; ++i) { + vpx_highbd_idct16_c(input, outptr, bd); + input += 16; + outptr += 16; + } + } + + if (optimised_cols) { + idct16_sse2(inptr, inptr + 16); + + // Final round & shift and Reconstruction and Store + { + __m128i d[2]; + for (i = 0; i < 16; i++) { + inptr[i ] = _mm_add_epi16(inptr[i ], rounding); + inptr[i+16] = _mm_add_epi16(inptr[i+16], rounding); + d[0] = _mm_loadu_si128((const __m128i *)(dest + stride*i)); + d[1] = _mm_loadu_si128((const __m128i *)(dest + stride*i + 8)); + inptr[i ] = _mm_srai_epi16(inptr[i ], 6); + inptr[i+16] = _mm_srai_epi16(inptr[i+16], 6); + d[0] = clamp_high_sse2(_mm_add_epi16(d[0], inptr[i ]), bd); + d[1] = clamp_high_sse2(_mm_add_epi16(d[1], inptr[i+16]), bd); + // Store + _mm_storeu_si128((__m128i *)(dest + stride*i), d[0]); + _mm_storeu_si128((__m128i *)(dest + stride*i + 8), d[1]); + } + } + } else { + // Run the un-optimised column transform + tran_low_t temp_in[16], temp_out[16]; + for (i = 0; i < 16; ++i) { + for (j = 0; j < 16; ++j) + temp_in[j] = out[j * 16 + i]; + vpx_highbd_idct16_c(temp_in, temp_out, bd); + for (j = 0; j < 16; ++j) { + dest[j * stride + i] = highbd_clip_pixel_add( + dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 6), bd); + } + } + } +} +#endif // CONFIG_VP9_HIGHBITDEPTH diff --git a/thirdparty/libvpx/vpx_dsp/x86/inv_txfm_sse2.h b/thirdparty/libvpx/vpx_dsp/x86/inv_txfm_sse2.h new file mode 100644 index 0000000000..bd520c18e5 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/x86/inv_txfm_sse2.h @@ -0,0 +1,196 @@ +/* + * Copyright (c) 2015 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VPX_DSP_X86_INV_TXFM_SSE2_H_ +#define VPX_DSP_X86_INV_TXFM_SSE2_H_ + +#include <emmintrin.h> // SSE2 +#include "./vpx_config.h" +#include "vpx/vpx_integer.h" +#include "vpx_dsp/inv_txfm.h" +#include "vpx_dsp/x86/txfm_common_sse2.h" + +// perform 8x8 transpose +static INLINE void array_transpose_8x8(__m128i *in, __m128i *res) { + const __m128i tr0_0 = _mm_unpacklo_epi16(in[0], in[1]); + const __m128i tr0_1 = _mm_unpacklo_epi16(in[2], in[3]); + const __m128i tr0_2 = _mm_unpackhi_epi16(in[0], in[1]); + const __m128i tr0_3 = _mm_unpackhi_epi16(in[2], in[3]); + const __m128i tr0_4 = _mm_unpacklo_epi16(in[4], in[5]); + const __m128i tr0_5 = _mm_unpacklo_epi16(in[6], in[7]); + const __m128i tr0_6 = _mm_unpackhi_epi16(in[4], in[5]); + const __m128i tr0_7 = _mm_unpackhi_epi16(in[6], in[7]); + + const __m128i tr1_0 = _mm_unpacklo_epi32(tr0_0, tr0_1); + const __m128i tr1_1 = _mm_unpacklo_epi32(tr0_4, tr0_5); + const __m128i tr1_2 = _mm_unpackhi_epi32(tr0_0, tr0_1); + const __m128i tr1_3 = _mm_unpackhi_epi32(tr0_4, tr0_5); + const __m128i tr1_4 = _mm_unpacklo_epi32(tr0_2, tr0_3); + const __m128i tr1_5 = _mm_unpacklo_epi32(tr0_6, tr0_7); + const __m128i tr1_6 = _mm_unpackhi_epi32(tr0_2, tr0_3); + const __m128i tr1_7 = _mm_unpackhi_epi32(tr0_6, tr0_7); + + res[0] = _mm_unpacklo_epi64(tr1_0, tr1_1); + res[1] = _mm_unpackhi_epi64(tr1_0, tr1_1); + res[2] = _mm_unpacklo_epi64(tr1_2, tr1_3); + res[3] = _mm_unpackhi_epi64(tr1_2, tr1_3); + res[4] = _mm_unpacklo_epi64(tr1_4, tr1_5); + res[5] = _mm_unpackhi_epi64(tr1_4, tr1_5); + res[6] = _mm_unpacklo_epi64(tr1_6, tr1_7); + res[7] = _mm_unpackhi_epi64(tr1_6, tr1_7); +} + +#define TRANSPOSE_8X4(in0, in1, in2, in3, out0, out1) \ + { \ + const __m128i tr0_0 = _mm_unpacklo_epi16(in0, in1); \ + const __m128i tr0_1 = _mm_unpacklo_epi16(in2, in3); \ + \ + in0 = _mm_unpacklo_epi32(tr0_0, tr0_1); /* i1 i0 */ \ + in1 = _mm_unpackhi_epi32(tr0_0, tr0_1); /* i3 i2 */ \ + } + +static INLINE void array_transpose_4X8(__m128i *in, __m128i * out) { + const __m128i tr0_0 = _mm_unpacklo_epi16(in[0], in[1]); + const __m128i tr0_1 = _mm_unpacklo_epi16(in[2], in[3]); + const __m128i tr0_4 = _mm_unpacklo_epi16(in[4], in[5]); + const __m128i tr0_5 = _mm_unpacklo_epi16(in[6], in[7]); + + const __m128i tr1_0 = _mm_unpacklo_epi32(tr0_0, tr0_1); + const __m128i tr1_2 = _mm_unpackhi_epi32(tr0_0, tr0_1); + const __m128i tr1_4 = _mm_unpacklo_epi32(tr0_4, tr0_5); + const __m128i tr1_6 = _mm_unpackhi_epi32(tr0_4, tr0_5); + + out[0] = _mm_unpacklo_epi64(tr1_0, tr1_4); + out[1] = _mm_unpackhi_epi64(tr1_0, tr1_4); + out[2] = _mm_unpacklo_epi64(tr1_2, tr1_6); + out[3] = _mm_unpackhi_epi64(tr1_2, tr1_6); +} + +static INLINE void array_transpose_16x16(__m128i *res0, __m128i *res1) { + __m128i tbuf[8]; + array_transpose_8x8(res0, res0); + array_transpose_8x8(res1, tbuf); + array_transpose_8x8(res0 + 8, res1); + array_transpose_8x8(res1 + 8, res1 + 8); + + res0[8] = tbuf[0]; + res0[9] = tbuf[1]; + res0[10] = tbuf[2]; + res0[11] = tbuf[3]; + res0[12] = tbuf[4]; + res0[13] = tbuf[5]; + res0[14] = tbuf[6]; + res0[15] = tbuf[7]; +} + +// Function to allow 8 bit optimisations to be used when profile 0 is used with +// highbitdepth enabled +static INLINE __m128i load_input_data(const tran_low_t *data) { +#if CONFIG_VP9_HIGHBITDEPTH + return octa_set_epi16(data[0], data[1], data[2], data[3], data[4], data[5], + data[6], data[7]); +#else + return _mm_load_si128((const __m128i *)data); +#endif +} + +static INLINE void load_buffer_8x16(const tran_low_t *input, __m128i *in) { + in[0] = load_input_data(input + 0 * 16); + in[1] = load_input_data(input + 1 * 16); + in[2] = load_input_data(input + 2 * 16); + in[3] = load_input_data(input + 3 * 16); + in[4] = load_input_data(input + 4 * 16); + in[5] = load_input_data(input + 5 * 16); + in[6] = load_input_data(input + 6 * 16); + in[7] = load_input_data(input + 7 * 16); + + in[8] = load_input_data(input + 8 * 16); + in[9] = load_input_data(input + 9 * 16); + in[10] = load_input_data(input + 10 * 16); + in[11] = load_input_data(input + 11 * 16); + in[12] = load_input_data(input + 12 * 16); + in[13] = load_input_data(input + 13 * 16); + in[14] = load_input_data(input + 14 * 16); + in[15] = load_input_data(input + 15 * 16); +} + +#define RECON_AND_STORE(dest, in_x) \ + { \ + __m128i d0 = _mm_loadl_epi64((__m128i *)(dest)); \ + d0 = _mm_unpacklo_epi8(d0, zero); \ + d0 = _mm_add_epi16(in_x, d0); \ + d0 = _mm_packus_epi16(d0, d0); \ + _mm_storel_epi64((__m128i *)(dest), d0); \ + } + +static INLINE void write_buffer_8x16(uint8_t *dest, __m128i *in, int stride) { + const __m128i final_rounding = _mm_set1_epi16(1<<5); + const __m128i zero = _mm_setzero_si128(); + // Final rounding and shift + in[0] = _mm_adds_epi16(in[0], final_rounding); + in[1] = _mm_adds_epi16(in[1], final_rounding); + in[2] = _mm_adds_epi16(in[2], final_rounding); + in[3] = _mm_adds_epi16(in[3], final_rounding); + in[4] = _mm_adds_epi16(in[4], final_rounding); + in[5] = _mm_adds_epi16(in[5], final_rounding); + in[6] = _mm_adds_epi16(in[6], final_rounding); + in[7] = _mm_adds_epi16(in[7], final_rounding); + in[8] = _mm_adds_epi16(in[8], final_rounding); + in[9] = _mm_adds_epi16(in[9], final_rounding); + in[10] = _mm_adds_epi16(in[10], final_rounding); + in[11] = _mm_adds_epi16(in[11], final_rounding); + in[12] = _mm_adds_epi16(in[12], final_rounding); + in[13] = _mm_adds_epi16(in[13], final_rounding); + in[14] = _mm_adds_epi16(in[14], final_rounding); + in[15] = _mm_adds_epi16(in[15], final_rounding); + + in[0] = _mm_srai_epi16(in[0], 6); + in[1] = _mm_srai_epi16(in[1], 6); + in[2] = _mm_srai_epi16(in[2], 6); + in[3] = _mm_srai_epi16(in[3], 6); + in[4] = _mm_srai_epi16(in[4], 6); + in[5] = _mm_srai_epi16(in[5], 6); + in[6] = _mm_srai_epi16(in[6], 6); + in[7] = _mm_srai_epi16(in[7], 6); + in[8] = _mm_srai_epi16(in[8], 6); + in[9] = _mm_srai_epi16(in[9], 6); + in[10] = _mm_srai_epi16(in[10], 6); + in[11] = _mm_srai_epi16(in[11], 6); + in[12] = _mm_srai_epi16(in[12], 6); + in[13] = _mm_srai_epi16(in[13], 6); + in[14] = _mm_srai_epi16(in[14], 6); + in[15] = _mm_srai_epi16(in[15], 6); + + RECON_AND_STORE(dest + 0 * stride, in[0]); + RECON_AND_STORE(dest + 1 * stride, in[1]); + RECON_AND_STORE(dest + 2 * stride, in[2]); + RECON_AND_STORE(dest + 3 * stride, in[3]); + RECON_AND_STORE(dest + 4 * stride, in[4]); + RECON_AND_STORE(dest + 5 * stride, in[5]); + RECON_AND_STORE(dest + 6 * stride, in[6]); + RECON_AND_STORE(dest + 7 * stride, in[7]); + RECON_AND_STORE(dest + 8 * stride, in[8]); + RECON_AND_STORE(dest + 9 * stride, in[9]); + RECON_AND_STORE(dest + 10 * stride, in[10]); + RECON_AND_STORE(dest + 11 * stride, in[11]); + RECON_AND_STORE(dest + 12 * stride, in[12]); + RECON_AND_STORE(dest + 13 * stride, in[13]); + RECON_AND_STORE(dest + 14 * stride, in[14]); + RECON_AND_STORE(dest + 15 * stride, in[15]); +} + +void idct4_sse2(__m128i *in); +void idct8_sse2(__m128i *in); +void idct16_sse2(__m128i *in0, __m128i *in1); +void iadst4_sse2(__m128i *in); +void iadst8_sse2(__m128i *in); +void iadst16_sse2(__m128i *in0, __m128i *in1); + +#endif // VPX_DSP_X86_INV_TXFM_SSE2_H_ diff --git a/thirdparty/libvpx/vpx_dsp/x86/inv_txfm_ssse3_x86_64.asm b/thirdparty/libvpx/vpx_dsp/x86/inv_txfm_ssse3_x86_64.asm new file mode 100644 index 0000000000..20baf820f6 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/x86/inv_txfm_ssse3_x86_64.asm @@ -0,0 +1,1793 @@ +; +; Copyright (c) 2014 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + +%include "third_party/x86inc/x86inc.asm" + +; This file provides SSSE3 version of the inverse transformation. Part +; of the functions are originally derived from the ffmpeg project. +; Note that the current version applies to x86 64-bit only. + +SECTION_RODATA + +pw_11585x2: times 8 dw 23170 + +pw_m2404x2: times 8 dw -2404*2 +pw_m4756x2: times 8 dw -4756*2 +pw_m5520x2: times 8 dw -5520*2 +pw_m8423x2: times 8 dw -8423*2 +pw_m9102x2: times 8 dw -9102*2 +pw_m10394x2: times 8 dw -10394*2 +pw_m11003x2: times 8 dw -11003*2 + +pw_16364x2: times 8 dw 16364*2 +pw_16305x2: times 8 dw 16305*2 +pw_16207x2: times 8 dw 16207*2 +pw_16069x2: times 8 dw 16069*2 +pw_15893x2: times 8 dw 15893*2 +pw_15679x2: times 8 dw 15679*2 +pw_15426x2: times 8 dw 15426*2 +pw_15137x2: times 8 dw 15137*2 +pw_14811x2: times 8 dw 14811*2 +pw_14449x2: times 8 dw 14449*2 +pw_14053x2: times 8 dw 14053*2 +pw_13623x2: times 8 dw 13623*2 +pw_13160x2: times 8 dw 13160*2 +pw_12665x2: times 8 dw 12665*2 +pw_12140x2: times 8 dw 12140*2 +pw__9760x2: times 8 dw 9760*2 +pw__7723x2: times 8 dw 7723*2 +pw__7005x2: times 8 dw 7005*2 +pw__6270x2: times 8 dw 6270*2 +pw__3981x2: times 8 dw 3981*2 +pw__3196x2: times 8 dw 3196*2 +pw__1606x2: times 8 dw 1606*2 +pw___804x2: times 8 dw 804*2 + +pd_8192: times 4 dd 8192 +pw_32: times 8 dw 32 +pw_16: times 8 dw 16 + +%macro TRANSFORM_COEFFS 2 +pw_%1_%2: dw %1, %2, %1, %2, %1, %2, %1, %2 +pw_m%2_%1: dw -%2, %1, -%2, %1, -%2, %1, -%2, %1 +pw_m%1_m%2: dw -%1, -%2, -%1, -%2, -%1, -%2, -%1, -%2 +%endmacro + +TRANSFORM_COEFFS 6270, 15137 +TRANSFORM_COEFFS 3196, 16069 +TRANSFORM_COEFFS 13623, 9102 + +; constants for 32x32_34 +TRANSFORM_COEFFS 804, 16364 +TRANSFORM_COEFFS 15426, 5520 +TRANSFORM_COEFFS 3981, 15893 +TRANSFORM_COEFFS 16207, 2404 +TRANSFORM_COEFFS 1606, 16305 +TRANSFORM_COEFFS 15679, 4756 +TRANSFORM_COEFFS 11585, 11585 + +; constants for 32x32_1024 +TRANSFORM_COEFFS 12140, 11003 +TRANSFORM_COEFFS 7005, 14811 +TRANSFORM_COEFFS 14053, 8423 +TRANSFORM_COEFFS 9760, 13160 +TRANSFORM_COEFFS 12665, 10394 +TRANSFORM_COEFFS 7723, 14449 + +%macro PAIR_PP_COEFFS 2 +dpw_%1_%2: dw %1, %1, %1, %1, %2, %2, %2, %2 +%endmacro + +%macro PAIR_MP_COEFFS 2 +dpw_m%1_%2: dw -%1, -%1, -%1, -%1, %2, %2, %2, %2 +%endmacro + +%macro PAIR_MM_COEFFS 2 +dpw_m%1_m%2: dw -%1, -%1, -%1, -%1, -%2, -%2, -%2, -%2 +%endmacro + +PAIR_PP_COEFFS 30274, 12540 +PAIR_PP_COEFFS 6392, 32138 +PAIR_MP_COEFFS 18204, 27246 + +PAIR_PP_COEFFS 12540, 12540 +PAIR_PP_COEFFS 30274, 30274 +PAIR_PP_COEFFS 6392, 6392 +PAIR_PP_COEFFS 32138, 32138 +PAIR_MM_COEFFS 18204, 18204 +PAIR_PP_COEFFS 27246, 27246 + +SECTION .text + +%if ARCH_X86_64 +%macro SUM_SUB 3 + psubw m%3, m%1, m%2 + paddw m%1, m%2 + SWAP %2, %3 +%endmacro + +; butterfly operation +%macro MUL_ADD_2X 6 ; dst1, dst2, src, round, coefs1, coefs2 + pmaddwd m%1, m%3, %5 + pmaddwd m%2, m%3, %6 + paddd m%1, %4 + paddd m%2, %4 + psrad m%1, 14 + psrad m%2, 14 +%endmacro + +%macro BUTTERFLY_4X 7 ; dst1, dst2, coef1, coef2, round, tmp1, tmp2 + punpckhwd m%6, m%2, m%1 + MUL_ADD_2X %7, %6, %6, %5, [pw_m%4_%3], [pw_%3_%4] + punpcklwd m%2, m%1 + MUL_ADD_2X %1, %2, %2, %5, [pw_m%4_%3], [pw_%3_%4] + packssdw m%1, m%7 + packssdw m%2, m%6 +%endmacro + +%macro BUTTERFLY_4Xmm 7 ; dst1, dst2, coef1, coef2, round, tmp1, tmp2 + punpckhwd m%6, m%2, m%1 + MUL_ADD_2X %7, %6, %6, %5, [pw_m%4_%3], [pw_m%3_m%4] + punpcklwd m%2, m%1 + MUL_ADD_2X %1, %2, %2, %5, [pw_m%4_%3], [pw_m%3_m%4] + packssdw m%1, m%7 + packssdw m%2, m%6 +%endmacro + +; matrix transpose +%macro INTERLEAVE_2X 4 + punpckh%1 m%4, m%2, m%3 + punpckl%1 m%2, m%3 + SWAP %3, %4 +%endmacro + +%macro TRANSPOSE8X8 9 + INTERLEAVE_2X wd, %1, %2, %9 + INTERLEAVE_2X wd, %3, %4, %9 + INTERLEAVE_2X wd, %5, %6, %9 + INTERLEAVE_2X wd, %7, %8, %9 + + INTERLEAVE_2X dq, %1, %3, %9 + INTERLEAVE_2X dq, %2, %4, %9 + INTERLEAVE_2X dq, %5, %7, %9 + INTERLEAVE_2X dq, %6, %8, %9 + + INTERLEAVE_2X qdq, %1, %5, %9 + INTERLEAVE_2X qdq, %3, %7, %9 + INTERLEAVE_2X qdq, %2, %6, %9 + INTERLEAVE_2X qdq, %4, %8, %9 + + SWAP %2, %5 + SWAP %4, %7 +%endmacro + +%macro IDCT8_1D 0 + SUM_SUB 0, 4, 9 + BUTTERFLY_4X 2, 6, 6270, 15137, m8, 9, 10 + pmulhrsw m0, m12 + pmulhrsw m4, m12 + BUTTERFLY_4X 1, 7, 3196, 16069, m8, 9, 10 + BUTTERFLY_4X 5, 3, 13623, 9102, m8, 9, 10 + + SUM_SUB 1, 5, 9 + SUM_SUB 7, 3, 9 + SUM_SUB 0, 6, 9 + SUM_SUB 4, 2, 9 + SUM_SUB 3, 5, 9 + pmulhrsw m3, m12 + pmulhrsw m5, m12 + + SUM_SUB 0, 7, 9 + SUM_SUB 4, 3, 9 + SUM_SUB 2, 5, 9 + SUM_SUB 6, 1, 9 + + SWAP 3, 6 + SWAP 1, 4 +%endmacro + +; This macro handles 8 pixels per line +%macro ADD_STORE_8P_2X 5; src1, src2, tmp1, tmp2, zero + paddw m%1, m11 + paddw m%2, m11 + psraw m%1, 5 + psraw m%2, 5 + + movh m%3, [outputq] + movh m%4, [outputq + strideq] + punpcklbw m%3, m%5 + punpcklbw m%4, m%5 + paddw m%3, m%1 + paddw m%4, m%2 + packuswb m%3, m%5 + packuswb m%4, m%5 + movh [outputq], m%3 + movh [outputq + strideq], m%4 +%endmacro + +INIT_XMM ssse3 +; full inverse 8x8 2D-DCT transform +cglobal idct8x8_64_add, 3, 5, 13, input, output, stride + mova m8, [pd_8192] + mova m11, [pw_16] + mova m12, [pw_11585x2] + + lea r3, [2 * strideq] +%if CONFIG_VP9_HIGHBITDEPTH + mova m0, [inputq + 0] + packssdw m0, [inputq + 16] + mova m1, [inputq + 32] + packssdw m1, [inputq + 48] + mova m2, [inputq + 64] + packssdw m2, [inputq + 80] + mova m3, [inputq + 96] + packssdw m3, [inputq + 112] + mova m4, [inputq + 128] + packssdw m4, [inputq + 144] + mova m5, [inputq + 160] + packssdw m5, [inputq + 176] + mova m6, [inputq + 192] + packssdw m6, [inputq + 208] + mova m7, [inputq + 224] + packssdw m7, [inputq + 240] +%else + mova m0, [inputq + 0] + mova m1, [inputq + 16] + mova m2, [inputq + 32] + mova m3, [inputq + 48] + mova m4, [inputq + 64] + mova m5, [inputq + 80] + mova m6, [inputq + 96] + mova m7, [inputq + 112] +%endif + TRANSPOSE8X8 0, 1, 2, 3, 4, 5, 6, 7, 9 + IDCT8_1D + TRANSPOSE8X8 0, 1, 2, 3, 4, 5, 6, 7, 9 + IDCT8_1D + + pxor m12, m12 + ADD_STORE_8P_2X 0, 1, 9, 10, 12 + lea outputq, [outputq + r3] + ADD_STORE_8P_2X 2, 3, 9, 10, 12 + lea outputq, [outputq + r3] + ADD_STORE_8P_2X 4, 5, 9, 10, 12 + lea outputq, [outputq + r3] + ADD_STORE_8P_2X 6, 7, 9, 10, 12 + + RET + +; inverse 8x8 2D-DCT transform with only first 10 coeffs non-zero +cglobal idct8x8_12_add, 3, 5, 13, input, output, stride + mova m8, [pd_8192] + mova m11, [pw_16] + mova m12, [pw_11585x2] + + lea r3, [2 * strideq] + +%if CONFIG_VP9_HIGHBITDEPTH + mova m0, [inputq + 0] + packssdw m0, [inputq + 16] + mova m1, [inputq + 32] + packssdw m1, [inputq + 48] + mova m2, [inputq + 64] + packssdw m2, [inputq + 80] + mova m3, [inputq + 96] + packssdw m3, [inputq + 112] +%else + mova m0, [inputq + 0] + mova m1, [inputq + 16] + mova m2, [inputq + 32] + mova m3, [inputq + 48] +%endif + + punpcklwd m0, m1 + punpcklwd m2, m3 + punpckhdq m9, m0, m2 + punpckldq m0, m2 + SWAP 2, 9 + + ; m0 -> [0], [0] + ; m1 -> [1], [1] + ; m2 -> [2], [2] + ; m3 -> [3], [3] + punpckhqdq m10, m0, m0 + punpcklqdq m0, m0 + punpckhqdq m9, m2, m2 + punpcklqdq m2, m2 + SWAP 1, 10 + SWAP 3, 9 + + pmulhrsw m0, m12 + pmulhrsw m2, [dpw_30274_12540] + pmulhrsw m1, [dpw_6392_32138] + pmulhrsw m3, [dpw_m18204_27246] + + SUM_SUB 0, 2, 9 + SUM_SUB 1, 3, 9 + + punpcklqdq m9, m3, m3 + punpckhqdq m5, m3, m9 + + SUM_SUB 3, 5, 9 + punpckhqdq m5, m3 + pmulhrsw m5, m12 + + punpckhqdq m9, m1, m5 + punpcklqdq m1, m5 + SWAP 5, 9 + + SUM_SUB 0, 5, 9 + SUM_SUB 2, 1, 9 + + punpckhqdq m3, m0, m0 + punpckhqdq m4, m1, m1 + punpckhqdq m6, m5, m5 + punpckhqdq m7, m2, m2 + + punpcklwd m0, m3 + punpcklwd m7, m2 + punpcklwd m1, m4 + punpcklwd m6, m5 + + punpckhdq m4, m0, m7 + punpckldq m0, m7 + punpckhdq m10, m1, m6 + punpckldq m5, m1, m6 + + punpckhqdq m1, m0, m5 + punpcklqdq m0, m5 + punpckhqdq m3, m4, m10 + punpcklqdq m2, m4, m10 + + + pmulhrsw m0, m12 + pmulhrsw m6, m2, [dpw_30274_30274] + pmulhrsw m4, m2, [dpw_12540_12540] + + pmulhrsw m7, m1, [dpw_32138_32138] + pmulhrsw m1, [dpw_6392_6392] + pmulhrsw m5, m3, [dpw_m18204_m18204] + pmulhrsw m3, [dpw_27246_27246] + + mova m2, m0 + SUM_SUB 0, 6, 9 + SUM_SUB 2, 4, 9 + SUM_SUB 1, 5, 9 + SUM_SUB 7, 3, 9 + + SUM_SUB 3, 5, 9 + pmulhrsw m3, m12 + pmulhrsw m5, m12 + + SUM_SUB 0, 7, 9 + SUM_SUB 2, 3, 9 + SUM_SUB 4, 5, 9 + SUM_SUB 6, 1, 9 + + SWAP 3, 6 + SWAP 1, 2 + SWAP 2, 4 + + + pxor m12, m12 + ADD_STORE_8P_2X 0, 1, 9, 10, 12 + lea outputq, [outputq + r3] + ADD_STORE_8P_2X 2, 3, 9, 10, 12 + lea outputq, [outputq + r3] + ADD_STORE_8P_2X 4, 5, 9, 10, 12 + lea outputq, [outputq + r3] + ADD_STORE_8P_2X 6, 7, 9, 10, 12 + + RET + +%define idx0 16 * 0 +%define idx1 16 * 1 +%define idx2 16 * 2 +%define idx3 16 * 3 +%define idx4 16 * 4 +%define idx5 16 * 5 +%define idx6 16 * 6 +%define idx7 16 * 7 +%define idx8 16 * 0 +%define idx9 16 * 1 +%define idx10 16 * 2 +%define idx11 16 * 3 +%define idx12 16 * 4 +%define idx13 16 * 5 +%define idx14 16 * 6 +%define idx15 16 * 7 +%define idx16 16 * 0 +%define idx17 16 * 1 +%define idx18 16 * 2 +%define idx19 16 * 3 +%define idx20 16 * 4 +%define idx21 16 * 5 +%define idx22 16 * 6 +%define idx23 16 * 7 +%define idx24 16 * 0 +%define idx25 16 * 1 +%define idx26 16 * 2 +%define idx27 16 * 3 +%define idx28 16 * 4 +%define idx29 16 * 5 +%define idx30 16 * 6 +%define idx31 16 * 7 + +; FROM idct32x32_add_neon.asm +; +; Instead of doing the transforms stage by stage, it is done by loading +; some input values and doing as many stages as possible to minimize the +; storing/loading of intermediate results. To fit within registers, the +; final coefficients are cut into four blocks: +; BLOCK A: 16-19,28-31 +; BLOCK B: 20-23,24-27 +; BLOCK C: 8-11,12-15 +; BLOCK D: 0-3,4-7 +; Blocks A and C are straight calculation through the various stages. In +; block B, further calculations are performed using the results from +; block A. In block D, further calculations are performed using the results +; from block C and then the final calculations are done using results from +; block A and B which have been combined at the end of block B. +; + +%macro IDCT32X32_34 4 + ; BLOCK A STAGE 1 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova m11, m1 + pmulhrsw m1, [pw___804x2] ; stp1_16 + mova [r4 + 0], m0 + pmulhrsw m11, [pw_16364x2] ; stp2_31 + mova [r4 + 16 * 2], m2 + mova m12, m7 + pmulhrsw m7, [pw_15426x2] ; stp1_28 + mova [r4 + 16 * 4], m4 + pmulhrsw m12, [pw_m5520x2] ; stp2_19 + mova [r4 + 16 * 6], m6 + + ; BLOCK A STAGE 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova m2, m1 ; stp1_16 + mova m0, m11 ; stp1_31 + mova m4, m7 ; stp1_28 + mova m15, m12 ; stp1_19 + + ; BLOCK A STAGE 3 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + BUTTERFLY_4X 0, 2, 3196, 16069, m8, 9, 10 ; stp1_17, stp1_30 + BUTTERFLY_4Xmm 4, 15, 3196, 16069, m8, 9, 10 ; stp1_29, stp1_18 + + ; BLOCK A STAGE 4 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + SUM_SUB 1, 12, 9 ; stp2_16, stp2_19 + SUM_SUB 0, 15, 9 ; stp2_17, stp2_18 + SUM_SUB 11, 7, 9 ; stp2_31, stp2_28 + SUM_SUB 2, 4, 9 ; stp2_30, stp2_29 + + ; BLOCK A STAGE 5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + BUTTERFLY_4X 4, 15, 6270, 15137, m8, 9, 10 ; stp1_18, stp1_29 + BUTTERFLY_4X 7, 12, 6270, 15137, m8, 9, 10 ; stp1_19, stp1_28 + + ; BLOCK B STAGE 1 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova m6, m5 + pmulhrsw m5, [pw__3981x2] ; stp1_20 + mova [stp + %4 + idx28], m12 + mova [stp + %4 + idx29], m15 + pmulhrsw m6, [pw_15893x2] ; stp2_27 + mova [stp + %4 + idx30], m2 + mova m2, m3 + pmulhrsw m3, [pw_m2404x2] ; stp1_23 + mova [stp + %4 + idx31], m11 + pmulhrsw m2, [pw_16207x2] ; stp2_24 + + ; BLOCK B STAGE 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova m13, m5 ; stp1_20 + mova m14, m6 ; stp1_27 + mova m15, m3 ; stp1_23 + mova m11, m2 ; stp1_24 + + ; BLOCK B STAGE 3 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + BUTTERFLY_4X 14, 13, 13623, 9102, m8, 9, 10 ; stp1_21, stp1_26 + BUTTERFLY_4Xmm 11, 15, 13623, 9102, m8, 9, 10 ; stp1_25, stp1_22 + + ; BLOCK B STAGE 4 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + SUM_SUB 3, 5, 9 ; stp2_23, stp2_20 + SUM_SUB 15, 14, 9 ; stp2_22, stp2_21 + SUM_SUB 2, 6, 9 ; stp2_24, stp2_27 + SUM_SUB 11, 13, 9 ; stp2_25, stp2_26 + + ; BLOCK B STAGE 5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + BUTTERFLY_4Xmm 6, 5, 6270, 15137, m8, 9, 10 ; stp1_27, stp1_20 + BUTTERFLY_4Xmm 13, 14, 6270, 15137, m8, 9, 10 ; stp1_26, stp1_21 + + ; BLOCK B STAGE 6 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + SUM_SUB 1, 3, 9 ; stp2_16, stp2_23 + SUM_SUB 0, 15, 9 ; stp2_17, stp2_22 + SUM_SUB 4, 14, 9 ; stp2_18, stp2_21 + SUM_SUB 7, 5, 9 ; stp2_19, stp2_20 + mova [stp + %3 + idx16], m1 + mova [stp + %3 + idx17], m0 + mova [stp + %3 + idx18], m4 + mova [stp + %3 + idx19], m7 + + mova m4, [stp + %4 + idx28] + mova m7, [stp + %4 + idx29] + mova m10, [stp + %4 + idx30] + mova m12, [stp + %4 + idx31] + SUM_SUB 4, 6, 9 ; stp2_28, stp2_27 + SUM_SUB 7, 13, 9 ; stp2_29, stp2_26 + SUM_SUB 10, 11, 9 ; stp2_30, stp2_25 + SUM_SUB 12, 2, 9 ; stp2_31, stp2_24 + mova [stp + %4 + idx28], m4 + mova [stp + %4 + idx29], m7 + mova [stp + %4 + idx30], m10 + mova [stp + %4 + idx31], m12 + + ; BLOCK B STAGE 7 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ +%if 0 ; overflow occurs in SUM_SUB when using test streams + mova m10, [pw_11585x2] + SUM_SUB 6, 5, 9 + pmulhrsw m6, m10 ; stp1_27 + pmulhrsw m5, m10 ; stp1_20 + SUM_SUB 13, 14, 9 + pmulhrsw m13, m10 ; stp1_26 + pmulhrsw m14, m10 ; stp1_21 + SUM_SUB 11, 15, 9 + pmulhrsw m11, m10 ; stp1_25 + pmulhrsw m15, m10 ; stp1_22 + SUM_SUB 2, 3, 9 + pmulhrsw m2, m10 ; stp1_24 + pmulhrsw m3, m10 ; stp1_23 +%else + BUTTERFLY_4X 6, 5, 11585, 11585, m8, 9, 10 ; stp1_20, stp1_27 + SWAP 6, 5 + BUTTERFLY_4X 13, 14, 11585, 11585, m8, 9, 10 ; stp1_21, stp1_26 + SWAP 13, 14 + BUTTERFLY_4X 11, 15, 11585, 11585, m8, 9, 10 ; stp1_22, stp1_25 + SWAP 11, 15 + BUTTERFLY_4X 2, 3, 11585, 11585, m8, 9, 10 ; stp1_23, stp1_24 + SWAP 2, 3 +%endif + + mova [stp + %4 + idx24], m2 + mova [stp + %4 + idx25], m11 + mova [stp + %4 + idx26], m13 + mova [stp + %4 + idx27], m6 + + ; BLOCK C STAGE 1 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + ; + ; BLOCK C STAGE 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova m0, [rsp + transposed_in + 16 * 2] + mova m6, [rsp + transposed_in + 16 * 6] + + mova m1, m0 + pmulhrsw m0, [pw__1606x2] ; stp1_8 + mova [stp + %3 + idx20], m5 + mova [stp + %3 + idx21], m14 + pmulhrsw m1, [pw_16305x2] ; stp2_15 + mova [stp + %3 + idx22], m15 + mova m7, m6 + pmulhrsw m7, [pw_m4756x2] ; stp2_11 + mova [stp + %3 + idx23], m3 + pmulhrsw m6, [pw_15679x2] ; stp1_12 + + ; BLOCK C STAGE 3 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova m3, m0 ; stp1_8 + mova m2, m1 ; stp1_15 + + ; BLOCK C STAGE 4 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + BUTTERFLY_4X 2, 3, 6270, 15137, m8, 9, 10 ; stp1_9, stp1_14 + mova m4, m7 ; stp1_11 + mova m5, m6 ; stp1_12 + BUTTERFLY_4Xmm 5, 4, 6270, 15137, m8, 9, 10 ; stp1_13, stp1_10 + + ; BLOCK C STAGE 5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + SUM_SUB 0, 7, 9 ; stp1_8, stp1_11 + SUM_SUB 2, 4, 9 ; stp1_9, stp1_10 + SUM_SUB 1, 6, 9 ; stp1_15, stp1_12 + SUM_SUB 3, 5, 9 ; stp1_14, stp1_13 + + ; BLOCK C STAGE 6 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ +%if 0 ; overflow occurs in SUM_SUB when using test streams + mova m10, [pw_11585x2] + SUM_SUB 5, 4, 9 + pmulhrsw m5, m10 ; stp1_13 + pmulhrsw m4, m10 ; stp1_10 + SUM_SUB 6, 7, 9 + pmulhrsw m6, m10 ; stp1_12 + pmulhrsw m7, m10 ; stp1_11 +%else + BUTTERFLY_4X 5, 4, 11585, 11585, m8, 9, 10 ; stp1_10, stp1_13 + SWAP 5, 4 + BUTTERFLY_4X 6, 7, 11585, 11585, m8, 9, 10 ; stp1_11, stp1_12 + SWAP 6, 7 +%endif + + ; BLOCK C STAGE 7 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova [stp + %2 + idx8], m0 + mova [stp + %2 + idx9], m2 + mova [stp + %2 + idx10], m4 + mova [stp + %2 + idx11], m7 + + ; BLOCK D STAGE 1 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + ; + ; BLOCK D STAGE 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + ; + ; BLOCK D STAGE 3 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova m11, [rsp + transposed_in + 16 * 4] + mova m12, m11 + pmulhrsw m11, [pw__3196x2] ; stp1_4 + pmulhrsw m12, [pw_16069x2] ; stp1_7 + + ; BLOCK D STAGE 4 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova m0, [rsp + transposed_in + 16 * 0] + mova m10, [pw_11585x2] + pmulhrsw m0, m10 ; stp1_1 + + mova m14, m11 ; stp1_4 + mova m13, m12 ; stp1_7 + + ; BLOCK D STAGE 5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ +%if 0 ; overflow occurs in SUM_SUB when using test streams + SUM_SUB 13, 14, 9 + pmulhrsw m13, m10 ; stp1_6 + pmulhrsw m14, m10 ; stp1_5 +%else + BUTTERFLY_4X 13, 14, 11585, 11585, m8, 9, 10 ; stp1_5, stp1_6 + SWAP 13, 14 +%endif + mova m7, m0 ; stp1_0 = stp1_1 + mova m4, m0 ; stp1_1 + mova m2, m7 ; stp1_0 + + ; BLOCK D STAGE 6 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + SUM_SUB 0, 12, 9 ; stp1_0, stp1_7 + SUM_SUB 7, 13, 9 ; stp1_1, stp1_6 + SUM_SUB 2, 14, 9 ; stp1_2, stp1_5 + SUM_SUB 4, 11, 9 ; stp1_3, stp1_4 + + ; BLOCK D STAGE 7 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + SUM_SUB 0, 1, 9 ; stp1_0, stp1_15 + SUM_SUB 7, 3, 9 ; stp1_1, stp1_14 + SUM_SUB 2, 5, 9 ; stp1_2, stp1_13 + SUM_SUB 4, 6, 9 ; stp1_3, stp1_12 + + ; 0-3, 28-31 final stage + mova m15, [stp + %4 + idx30] + mova m10, [stp + %4 + idx31] + SUM_SUB 0, 10, 9 ; stp1_0, stp1_31 + SUM_SUB 7, 15, 9 ; stp1_1, stp1_30 + mova [stp + %1 + idx0], m0 + mova [stp + %1 + idx1], m7 + mova [stp + %4 + idx30], m15 + mova [stp + %4 + idx31], m10 + mova m7, [stp + %4 + idx28] + mova m0, [stp + %4 + idx29] + SUM_SUB 2, 0, 9 ; stp1_2, stp1_29 + SUM_SUB 4, 7, 9 ; stp1_3, stp1_28 + mova [stp + %1 + idx2], m2 + mova [stp + %1 + idx3], m4 + mova [stp + %4 + idx28], m7 + mova [stp + %4 + idx29], m0 + + ; 12-15, 16-19 final stage + mova m0, [stp + %3 + idx16] + mova m7, [stp + %3 + idx17] + mova m2, [stp + %3 + idx18] + mova m4, [stp + %3 + idx19] + SUM_SUB 1, 0, 9 ; stp1_15, stp1_16 + SUM_SUB 3, 7, 9 ; stp1_14, stp1_17 + SUM_SUB 5, 2, 9 ; stp1_13, stp1_18 + SUM_SUB 6, 4, 9 ; stp1_12, stp1_19 + mova [stp + %2 + idx12], m6 + mova [stp + %2 + idx13], m5 + mova [stp + %2 + idx14], m3 + mova [stp + %2 + idx15], m1 + mova [stp + %3 + idx16], m0 + mova [stp + %3 + idx17], m7 + mova [stp + %3 + idx18], m2 + mova [stp + %3 + idx19], m4 + + mova m4, [stp + %2 + idx8] + mova m5, [stp + %2 + idx9] + mova m6, [stp + %2 + idx10] + mova m7, [stp + %2 + idx11] + SUM_SUB 11, 7, 9 ; stp1_4, stp1_11 + SUM_SUB 14, 6, 9 ; stp1_5, stp1_10 + SUM_SUB 13, 5, 9 ; stp1_6, stp1_9 + SUM_SUB 12, 4, 9 ; stp1_7, stp1_8 + + ; 4-7, 24-27 final stage + mova m0, [stp + %4 + idx27] + mova m1, [stp + %4 + idx26] + mova m2, [stp + %4 + idx25] + mova m3, [stp + %4 + idx24] + SUM_SUB 11, 0, 9 ; stp1_4, stp1_27 + SUM_SUB 14, 1, 9 ; stp1_5, stp1_26 + SUM_SUB 13, 2, 9 ; stp1_6, stp1_25 + SUM_SUB 12, 3, 9 ; stp1_7, stp1_24 + mova [stp + %4 + idx27], m0 + mova [stp + %4 + idx26], m1 + mova [stp + %4 + idx25], m2 + mova [stp + %4 + idx24], m3 + mova [stp + %1 + idx4], m11 + mova [stp + %1 + idx5], m14 + mova [stp + %1 + idx6], m13 + mova [stp + %1 + idx7], m12 + + ; 8-11, 20-23 final stage + mova m0, [stp + %3 + idx20] + mova m1, [stp + %3 + idx21] + mova m2, [stp + %3 + idx22] + mova m3, [stp + %3 + idx23] + SUM_SUB 7, 0, 9 ; stp1_11, stp_20 + SUM_SUB 6, 1, 9 ; stp1_10, stp_21 + SUM_SUB 5, 2, 9 ; stp1_9, stp_22 + SUM_SUB 4, 3, 9 ; stp1_8, stp_23 + mova [stp + %2 + idx8], m4 + mova [stp + %2 + idx9], m5 + mova [stp + %2 + idx10], m6 + mova [stp + %2 + idx11], m7 + mova [stp + %3 + idx20], m0 + mova [stp + %3 + idx21], m1 + mova [stp + %3 + idx22], m2 + mova [stp + %3 + idx23], m3 +%endmacro + +%macro RECON_AND_STORE 1 + mova m11, [pw_32] + lea stp, [rsp + %1] + mov r6, 32 + pxor m8, m8 +%%recon_and_store: + mova m0, [stp + 16 * 32 * 0] + mova m1, [stp + 16 * 32 * 1] + mova m2, [stp + 16 * 32 * 2] + mova m3, [stp + 16 * 32 * 3] + add stp, 16 + + paddw m0, m11 + paddw m1, m11 + paddw m2, m11 + paddw m3, m11 + psraw m0, 6 + psraw m1, 6 + psraw m2, 6 + psraw m3, 6 + movh m4, [outputq + 0] + movh m5, [outputq + 8] + movh m6, [outputq + 16] + movh m7, [outputq + 24] + punpcklbw m4, m8 + punpcklbw m5, m8 + punpcklbw m6, m8 + punpcklbw m7, m8 + paddw m0, m4 + paddw m1, m5 + paddw m2, m6 + paddw m3, m7 + packuswb m0, m1 + packuswb m2, m3 + mova [outputq + 0], m0 + mova [outputq + 16], m2 + lea outputq, [outputq + strideq] + dec r6 + jnz %%recon_and_store +%endmacro + +%define i32x32_size 16*32*5 +%define pass_two_start 16*32*0 +%define transposed_in 16*32*4 +%define pass_one_start 16*32*0 +%define stp r8 + +INIT_XMM ssse3 +cglobal idct32x32_34_add, 3, 11, 16, i32x32_size, input, output, stride + mova m8, [pd_8192] + lea stp, [rsp + pass_one_start] + +idct32x32_34: + mov r3, inputq + lea r4, [rsp + transposed_in] + +idct32x32_34_transpose: +%if CONFIG_VP9_HIGHBITDEPTH + mova m0, [r3 + 0] + packssdw m0, [r3 + 16] + mova m1, [r3 + 32 * 4] + packssdw m1, [r3 + 32 * 4 + 16] + mova m2, [r3 + 32 * 8] + packssdw m2, [r3 + 32 * 8 + 16] + mova m3, [r3 + 32 * 12] + packssdw m3, [r3 + 32 * 12 + 16] + mova m4, [r3 + 32 * 16] + packssdw m4, [r3 + 32 * 16 + 16] + mova m5, [r3 + 32 * 20] + packssdw m5, [r3 + 32 * 20 + 16] + mova m6, [r3 + 32 * 24] + packssdw m6, [r3 + 32 * 24 + 16] + mova m7, [r3 + 32 * 28] + packssdw m7, [r3 + 32 * 28 + 16] +%else + mova m0, [r3 + 0] + mova m1, [r3 + 16 * 4] + mova m2, [r3 + 16 * 8] + mova m3, [r3 + 16 * 12] + mova m4, [r3 + 16 * 16] + mova m5, [r3 + 16 * 20] + mova m6, [r3 + 16 * 24] + mova m7, [r3 + 16 * 28] +%endif + + TRANSPOSE8X8 0, 1, 2, 3, 4, 5, 6, 7, 9 + + IDCT32X32_34 16*0, 16*32, 16*64, 16*96 + lea stp, [stp + 16 * 8] + mov r6, 4 + lea stp, [rsp + pass_one_start] + lea r9, [rsp + pass_one_start] + +idct32x32_34_2: + lea r4, [rsp + transposed_in] + mov r3, r9 + +idct32x32_34_transpose_2: + mova m0, [r3 + 0] + mova m1, [r3 + 16 * 1] + mova m2, [r3 + 16 * 2] + mova m3, [r3 + 16 * 3] + mova m4, [r3 + 16 * 4] + mova m5, [r3 + 16 * 5] + mova m6, [r3 + 16 * 6] + mova m7, [r3 + 16 * 7] + + TRANSPOSE8X8 0, 1, 2, 3, 4, 5, 6, 7, 9 + + IDCT32X32_34 16*0, 16*8, 16*16, 16*24 + + lea stp, [stp + 16 * 32] + add r9, 16 * 32 + dec r6 + jnz idct32x32_34_2 + + RECON_AND_STORE pass_two_start + + RET + +%macro IDCT32X32_135 4 + ; BLOCK A STAGE 1 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova m1, [rsp + transposed_in + 16 * 1] + mova m11, m1 + pmulhrsw m1, [pw___804x2] ; stp1_16 + pmulhrsw m11, [pw_16364x2] ; stp2_31 + + mova m7, [rsp + transposed_in + 16 * 7] + mova m12, m7 + pmulhrsw m7, [pw_15426x2] ; stp1_28 + pmulhrsw m12, [pw_m5520x2] ; stp2_19 + + mova m3, [rsp + transposed_in + 16 * 9] + mova m4, m3 + pmulhrsw m3, [pw__7005x2] ; stp1_18 + pmulhrsw m4, [pw_14811x2] ; stp2_29 + + mova m0, [rsp + transposed_in + 16 * 15] + mova m2, m0 + pmulhrsw m0, [pw_12140x2] ; stp1_30 + pmulhrsw m2, [pw_m11003x2] ; stp2_17 + + ; BLOCK A STAGE 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + SUM_SUB 1, 2, 9 ; stp2_16, stp2_17 + SUM_SUB 12, 3, 9 ; stp2_19, stp2_18 + SUM_SUB 7, 4, 9 ; stp2_28, stp2_29 + SUM_SUB 11, 0, 9 ; stp2_31, stp2_30 + + ; BLOCK A STAGE 3 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + BUTTERFLY_4X 0, 2, 3196, 16069, m8, 9, 10 ; stp1_17, stp1_30 + BUTTERFLY_4Xmm 4, 3, 3196, 16069, m8, 9, 10 ; stp1_29, stp1_18 + + ; BLOCK A STAGE 4 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + SUM_SUB 1, 12, 9 ; stp2_16, stp2_19 + SUM_SUB 0, 3, 9 ; stp2_17, stp2_18 + SUM_SUB 11, 7, 9 ; stp2_31, stp2_28 + SUM_SUB 2, 4, 9 ; stp2_30, stp2_29 + + ; BLOCK A STAGE 5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + BUTTERFLY_4X 4, 3, 6270, 15137, m8, 9, 10 ; stp1_18, stp1_29 + BUTTERFLY_4X 7, 12, 6270, 15137, m8, 9, 10 ; stp1_19, stp1_28 + + mova [stp + %3 + idx16], m1 + mova [stp + %3 + idx17], m0 + mova [stp + %3 + idx18], m4 + mova [stp + %3 + idx19], m7 + mova [stp + %4 + idx28], m12 + mova [stp + %4 + idx29], m3 + mova [stp + %4 + idx30], m2 + mova [stp + %4 + idx31], m11 + + ; BLOCK B STAGE 1 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova m2, [rsp + transposed_in + 16 * 3] + mova m3, m2 + pmulhrsw m3, [pw_m2404x2] ; stp1_23 + pmulhrsw m2, [pw_16207x2] ; stp2_24 + + mova m5, [rsp + transposed_in + 16 * 5] + mova m6, m5 + pmulhrsw m5, [pw__3981x2] ; stp1_20 + pmulhrsw m6, [pw_15893x2] ; stp2_27 + + mova m14, [rsp + transposed_in + 16 * 11] + mova m13, m14 + pmulhrsw m13, [pw_m8423x2] ; stp1_21 + pmulhrsw m14, [pw_14053x2] ; stp2_26 + + mova m0, [rsp + transposed_in + 16 * 13] + mova m1, m0 + pmulhrsw m0, [pw__9760x2] ; stp1_22 + pmulhrsw m1, [pw_13160x2] ; stp2_25 + + ; BLOCK B STAGE 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + SUM_SUB 5, 13, 9 ; stp2_20, stp2_21 + SUM_SUB 3, 0, 9 ; stp2_23, stp2_22 + SUM_SUB 2, 1, 9 ; stp2_24, stp2_25 + SUM_SUB 6, 14, 9 ; stp2_27, stp2_26 + + ; BLOCK B STAGE 3 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + BUTTERFLY_4X 14, 13, 13623, 9102, m8, 9, 10 ; stp1_21, stp1_26 + BUTTERFLY_4Xmm 1, 0, 13623, 9102, m8, 9, 10 ; stp1_25, stp1_22 + + ; BLOCK B STAGE 4 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + SUM_SUB 3, 5, 9 ; stp2_23, stp2_20 + SUM_SUB 0, 14, 9 ; stp2_22, stp2_21 + SUM_SUB 2, 6, 9 ; stp2_24, stp2_27 + SUM_SUB 1, 13, 9 ; stp2_25, stp2_26 + + ; BLOCK B STAGE 5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + BUTTERFLY_4Xmm 6, 5, 6270, 15137, m8, 9, 10 ; stp1_27, stp1_20 + BUTTERFLY_4Xmm 13, 14, 6270, 15137, m8, 9, 10 ; stp1_26, stp1_21 + + ; BLOCK B STAGE 6 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova m4, [stp + %3 + idx16] + mova m7, [stp + %3 + idx17] + mova m11, [stp + %3 + idx18] + mova m12, [stp + %3 + idx19] + SUM_SUB 4, 3, 9 ; stp2_16, stp2_23 + SUM_SUB 7, 0, 9 ; stp2_17, stp2_22 + SUM_SUB 11, 14, 9 ; stp2_18, stp2_21 + SUM_SUB 12, 5, 9 ; stp2_19, stp2_20 + mova [stp + %3 + idx16], m4 + mova [stp + %3 + idx17], m7 + mova [stp + %3 + idx18], m11 + mova [stp + %3 + idx19], m12 + + mova m4, [stp + %4 + idx28] + mova m7, [stp + %4 + idx29] + mova m11, [stp + %4 + idx30] + mova m12, [stp + %4 + idx31] + SUM_SUB 4, 6, 9 ; stp2_28, stp2_27 + SUM_SUB 7, 13, 9 ; stp2_29, stp2_26 + SUM_SUB 11, 1, 9 ; stp2_30, stp2_25 + SUM_SUB 12, 2, 9 ; stp2_31, stp2_24 + mova [stp + %4 + idx28], m4 + mova [stp + %4 + idx29], m7 + mova [stp + %4 + idx30], m11 + mova [stp + %4 + idx31], m12 + + ; BLOCK B STAGE 7 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ +%if 0 ; overflow occurs in SUM_SUB when using test streams + mova m10, [pw_11585x2] + SUM_SUB 6, 5, 9 + pmulhrsw m6, m10 ; stp1_27 + pmulhrsw m5, m10 ; stp1_20 + SUM_SUB 13, 14, 9 + pmulhrsw m13, m10 ; stp1_26 + pmulhrsw m14, m10 ; stp1_21 + SUM_SUB 1, 0, 9 + pmulhrsw m1, m10 ; stp1_25 + pmulhrsw m0, m10 ; stp1_22 + SUM_SUB 2, 3, 9 + pmulhrsw m2, m10 ; stp1_25 + pmulhrsw m3, m10 ; stp1_22 +%else + BUTTERFLY_4X 6, 5, 11585, 11585, m8, 9, 10 ; stp1_20, stp1_27 + SWAP 6, 5 + BUTTERFLY_4X 13, 14, 11585, 11585, m8, 9, 10 ; stp1_21, stp1_26 + SWAP 13, 14 + BUTTERFLY_4X 1, 0, 11585, 11585, m8, 9, 10 ; stp1_22, stp1_25 + SWAP 1, 0 + BUTTERFLY_4X 2, 3, 11585, 11585, m8, 9, 10 ; stp1_23, stp1_24 + SWAP 2, 3 +%endif + mova [stp + %3 + idx20], m5 + mova [stp + %3 + idx21], m14 + mova [stp + %3 + idx22], m0 + mova [stp + %3 + idx23], m3 + mova [stp + %4 + idx24], m2 + mova [stp + %4 + idx25], m1 + mova [stp + %4 + idx26], m13 + mova [stp + %4 + idx27], m6 + + ; BLOCK C STAGE 1 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + ; + ; BLOCK C STAGE 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova m0, [rsp + transposed_in + 16 * 2] + mova m1, m0 + pmulhrsw m0, [pw__1606x2] ; stp1_8 + pmulhrsw m1, [pw_16305x2] ; stp2_15 + + mova m6, [rsp + transposed_in + 16 * 6] + mova m7, m6 + pmulhrsw m7, [pw_m4756x2] ; stp2_11 + pmulhrsw m6, [pw_15679x2] ; stp1_12 + + mova m4, [rsp + transposed_in + 16 * 10] + mova m5, m4 + pmulhrsw m4, [pw__7723x2] ; stp1_10 + pmulhrsw m5, [pw_14449x2] ; stp2_13 + + mova m2, [rsp + transposed_in + 16 * 14] + mova m3, m2 + pmulhrsw m3, [pw_m10394x2] ; stp1_9 + pmulhrsw m2, [pw_12665x2] ; stp2_14 + + ; BLOCK C STAGE 3 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + SUM_SUB 0, 3, 9 ; stp1_8, stp1_9 + SUM_SUB 7, 4, 9 ; stp1_11, stp1_10 + SUM_SUB 6, 5, 9 ; stp1_12, stp1_13 + SUM_SUB 1, 2, 9 ; stp1_15, stp1_14 + + ; BLOCK C STAGE 4 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + BUTTERFLY_4X 2, 3, 6270, 15137, m8, 9, 10 ; stp1_9, stp1_14 + BUTTERFLY_4Xmm 5, 4, 6270, 15137, m8, 9, 10 ; stp1_13, stp1_10 + + ; BLOCK C STAGE 5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + SUM_SUB 0, 7, 9 ; stp1_8, stp1_11 + SUM_SUB 2, 4, 9 ; stp1_9, stp1_10 + SUM_SUB 1, 6, 9 ; stp1_15, stp1_12 + SUM_SUB 3, 5, 9 ; stp1_14, stp1_13 + + ; BLOCK C STAGE 6 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ +%if 0 ; overflow occurs in SUM_SUB when using test streams + mova m10, [pw_11585x2] + SUM_SUB 5, 4, 9 + pmulhrsw m5, m10 ; stp1_13 + pmulhrsw m4, m10 ; stp1_10 + SUM_SUB 6, 7, 9 + pmulhrsw m6, m10 ; stp1_12 + pmulhrsw m7, m10 ; stp1_11 +%else + BUTTERFLY_4X 5, 4, 11585, 11585, m8, 9, 10 ; stp1_10, stp1_13 + SWAP 5, 4 + BUTTERFLY_4X 6, 7, 11585, 11585, m8, 9, 10 ; stp1_11, stp1_12 + SWAP 6, 7 +%endif + ; BLOCK C STAGE 7 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova [stp + %2 + idx8], m0 + mova [stp + %2 + idx9], m2 + mova [stp + %2 + idx10], m4 + mova [stp + %2 + idx11], m7 + mova [stp + %2 + idx12], m6 + mova [stp + %2 + idx13], m5 + mova [stp + %2 + idx14], m3 + mova [stp + %2 + idx15], m1 + + ; BLOCK D STAGE 1 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + ; + ; BLOCK D STAGE 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + ; + ; BLOCK D STAGE 3 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova m11, [rsp + transposed_in + 16 * 4] + mova m12, m11 + pmulhrsw m11, [pw__3196x2] ; stp1_4 + pmulhrsw m12, [pw_16069x2] ; stp1_7 + + mova m13, [rsp + transposed_in + 16 * 12] + mova m14, m13 + pmulhrsw m13, [pw_13623x2] ; stp1_6 + pmulhrsw m14, [pw_m9102x2] ; stp1_5 + + ; BLOCK D STAGE 4 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova m0, [rsp + transposed_in + 16 * 0] + mova m2, [rsp + transposed_in + 16 * 8] + pmulhrsw m0, [pw_11585x2] ; stp1_1 + mova m3, m2 + pmulhrsw m2, [pw__6270x2] ; stp1_2 + pmulhrsw m3, [pw_15137x2] ; stp1_3 + + SUM_SUB 11, 14, 9 ; stp1_4, stp1_5 + SUM_SUB 12, 13, 9 ; stp1_7, stp1_6 + + ; BLOCK D STAGE 5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ +%if 0 ; overflow occurs in SUM_SUB when using test streams + mova m10, [pw_11585x2] + SUM_SUB 13, 14, 9 + pmulhrsw m13, m10 ; stp1_6 + pmulhrsw m14, m10 ; stp1_5 +%else + BUTTERFLY_4X 13, 14, 11585, 11585, m8, 9, 10 ; stp1_5, stp1_6 + SWAP 13, 14 +%endif + mova m1, m0 ; stp1_0 = stp1_1 + SUM_SUB 0, 3, 9 ; stp1_0, stp1_3 + SUM_SUB 1, 2, 9 ; stp1_1, stp1_2 + + ; BLOCK D STAGE 6 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + SUM_SUB 0, 12, 9 ; stp1_0, stp1_7 + SUM_SUB 1, 13, 9 ; stp1_1, stp1_6 + SUM_SUB 2, 14, 9 ; stp1_2, stp1_5 + SUM_SUB 3, 11, 9 ; stp1_3, stp1_4 + + ; BLOCK D STAGE 7 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova m4, [stp + %2 + idx12] + mova m5, [stp + %2 + idx13] + mova m6, [stp + %2 + idx14] + mova m7, [stp + %2 + idx15] + SUM_SUB 0, 7, 9 ; stp1_0, stp1_15 + SUM_SUB 1, 6, 9 ; stp1_1, stp1_14 + SUM_SUB 2, 5, 9 ; stp1_2, stp1_13 + SUM_SUB 3, 4, 9 ; stp1_3, stp1_12 + + ; 0-3, 28-31 final stage + mova m10, [stp + %4 + idx31] + mova m15, [stp + %4 + idx30] + SUM_SUB 0, 10, 9 ; stp1_0, stp1_31 + SUM_SUB 1, 15, 9 ; stp1_1, stp1_30 + mova [stp + %1 + idx0], m0 + mova [stp + %1 + idx1], m1 + mova [stp + %4 + idx31], m10 + mova [stp + %4 + idx30], m15 + mova m0, [stp + %4 + idx29] + mova m1, [stp + %4 + idx28] + SUM_SUB 2, 0, 9 ; stp1_2, stp1_29 + SUM_SUB 3, 1, 9 ; stp1_3, stp1_28 + mova [stp + %1 + idx2], m2 + mova [stp + %1 + idx3], m3 + mova [stp + %4 + idx29], m0 + mova [stp + %4 + idx28], m1 + + ; 12-15, 16-19 final stage + mova m0, [stp + %3 + idx16] + mova m1, [stp + %3 + idx17] + mova m2, [stp + %3 + idx18] + mova m3, [stp + %3 + idx19] + SUM_SUB 7, 0, 9 ; stp1_15, stp1_16 + SUM_SUB 6, 1, 9 ; stp1_14, stp1_17 + SUM_SUB 5, 2, 9 ; stp1_13, stp1_18 + SUM_SUB 4, 3, 9 ; stp1_12, stp1_19 + mova [stp + %2 + idx12], m4 + mova [stp + %2 + idx13], m5 + mova [stp + %2 + idx14], m6 + mova [stp + %2 + idx15], m7 + mova [stp + %3 + idx16], m0 + mova [stp + %3 + idx17], m1 + mova [stp + %3 + idx18], m2 + mova [stp + %3 + idx19], m3 + + mova m4, [stp + %2 + idx8] + mova m5, [stp + %2 + idx9] + mova m6, [stp + %2 + idx10] + mova m7, [stp + %2 + idx11] + SUM_SUB 11, 7, 9 ; stp1_4, stp1_11 + SUM_SUB 14, 6, 9 ; stp1_5, stp1_10 + SUM_SUB 13, 5, 9 ; stp1_6, stp1_9 + SUM_SUB 12, 4, 9 ; stp1_7, stp1_8 + + ; 4-7, 24-27 final stage + mova m3, [stp + %4 + idx24] + mova m2, [stp + %4 + idx25] + mova m1, [stp + %4 + idx26] + mova m0, [stp + %4 + idx27] + SUM_SUB 12, 3, 9 ; stp1_7, stp1_24 + SUM_SUB 13, 2, 9 ; stp1_6, stp1_25 + SUM_SUB 14, 1, 9 ; stp1_5, stp1_26 + SUM_SUB 11, 0, 9 ; stp1_4, stp1_27 + mova [stp + %4 + idx24], m3 + mova [stp + %4 + idx25], m2 + mova [stp + %4 + idx26], m1 + mova [stp + %4 + idx27], m0 + mova [stp + %1 + idx4], m11 + mova [stp + %1 + idx5], m14 + mova [stp + %1 + idx6], m13 + mova [stp + %1 + idx7], m12 + + ; 8-11, 20-23 final stage + mova m0, [stp + %3 + idx20] + mova m1, [stp + %3 + idx21] + mova m2, [stp + %3 + idx22] + mova m3, [stp + %3 + idx23] + SUM_SUB 7, 0, 9 ; stp1_11, stp_20 + SUM_SUB 6, 1, 9 ; stp1_10, stp_21 + SUM_SUB 5, 2, 9 ; stp1_9, stp_22 + SUM_SUB 4, 3, 9 ; stp1_8, stp_23 + mova [stp + %2 + idx8], m4 + mova [stp + %2 + idx9], m5 + mova [stp + %2 + idx10], m6 + mova [stp + %2 + idx11], m7 + mova [stp + %3 + idx20], m0 + mova [stp + %3 + idx21], m1 + mova [stp + %3 + idx22], m2 + mova [stp + %3 + idx23], m3 +%endmacro + +INIT_XMM ssse3 +cglobal idct32x32_135_add, 3, 11, 16, i32x32_size, input, output, stride + mova m8, [pd_8192] + mov r6, 2 + lea stp, [rsp + pass_one_start] + +idct32x32_135: + mov r3, inputq + lea r4, [rsp + transposed_in] + mov r7, 2 + +idct32x32_135_transpose: +%if CONFIG_VP9_HIGHBITDEPTH + mova m0, [r3 + 0] + packssdw m0, [r3 + 16] + mova m1, [r3 + 32 * 4] + packssdw m1, [r3 + 32 * 4 + 16] + mova m2, [r3 + 32 * 8] + packssdw m2, [r3 + 32 * 8 + 16] + mova m3, [r3 + 32 * 12] + packssdw m3, [r3 + 32 * 12 + 16] + mova m4, [r3 + 32 * 16] + packssdw m4, [r3 + 32 * 16 + 16] + mova m5, [r3 + 32 * 20] + packssdw m5, [r3 + 32 * 20 + 16] + mova m6, [r3 + 32 * 24] + packssdw m6, [r3 + 32 * 24 + 16] + mova m7, [r3 + 32 * 28] + packssdw m7, [r3 + 32 * 28 + 16] +%else + mova m0, [r3 + 0] + mova m1, [r3 + 16 * 4] + mova m2, [r3 + 16 * 8] + mova m3, [r3 + 16 * 12] + mova m4, [r3 + 16 * 16] + mova m5, [r3 + 16 * 20] + mova m6, [r3 + 16 * 24] + mova m7, [r3 + 16 * 28] +%endif + TRANSPOSE8X8 0, 1, 2, 3, 4, 5, 6, 7, 9 + + mova [r4 + 0], m0 + mova [r4 + 16 * 1], m1 + mova [r4 + 16 * 2], m2 + mova [r4 + 16 * 3], m3 + mova [r4 + 16 * 4], m4 + mova [r4 + 16 * 5], m5 + mova [r4 + 16 * 6], m6 + mova [r4 + 16 * 7], m7 + +%if CONFIG_VP9_HIGHBITDEPTH + add r3, 32 +%else + add r3, 16 +%endif + add r4, 16 * 8 + dec r7 + jne idct32x32_135_transpose + + IDCT32X32_135 16*0, 16*32, 16*64, 16*96 + lea stp, [stp + 16 * 8] +%if CONFIG_VP9_HIGHBITDEPTH + lea inputq, [inputq + 32 * 32] +%else + lea inputq, [inputq + 16 * 32] +%endif + dec r6 + jnz idct32x32_135 + + mov r6, 4 + lea stp, [rsp + pass_one_start] + lea r9, [rsp + pass_one_start] + +idct32x32_135_2: + lea r4, [rsp + transposed_in] + mov r3, r9 + mov r7, 2 + +idct32x32_135_transpose_2: + mova m0, [r3 + 0] + mova m1, [r3 + 16 * 1] + mova m2, [r3 + 16 * 2] + mova m3, [r3 + 16 * 3] + mova m4, [r3 + 16 * 4] + mova m5, [r3 + 16 * 5] + mova m6, [r3 + 16 * 6] + mova m7, [r3 + 16 * 7] + + TRANSPOSE8X8 0, 1, 2, 3, 4, 5, 6, 7, 9 + + mova [r4 + 0], m0 + mova [r4 + 16 * 1], m1 + mova [r4 + 16 * 2], m2 + mova [r4 + 16 * 3], m3 + mova [r4 + 16 * 4], m4 + mova [r4 + 16 * 5], m5 + mova [r4 + 16 * 6], m6 + mova [r4 + 16 * 7], m7 + + add r3, 16 * 8 + add r4, 16 * 8 + dec r7 + jne idct32x32_135_transpose_2 + + IDCT32X32_135 16*0, 16*8, 16*16, 16*24 + + lea stp, [stp + 16 * 32] + add r9, 16 * 32 + dec r6 + jnz idct32x32_135_2 + + RECON_AND_STORE pass_two_start + + RET + +%macro IDCT32X32_1024 4 + ; BLOCK A STAGE 1 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova m1, [rsp + transposed_in + 16 * 1] + mova m11, [rsp + transposed_in + 16 * 31] + BUTTERFLY_4X 1, 11, 804, 16364, m8, 9, 10 ; stp1_16, stp1_31 + + mova m0, [rsp + transposed_in + 16 * 15] + mova m2, [rsp + transposed_in + 16 * 17] + BUTTERFLY_4X 2, 0, 12140, 11003, m8, 9, 10 ; stp1_17, stp1_30 + + mova m7, [rsp + transposed_in + 16 * 7] + mova m12, [rsp + transposed_in + 16 * 25] + BUTTERFLY_4X 12, 7, 15426, 5520, m8, 9, 10 ; stp1_19, stp1_28 + + mova m3, [rsp + transposed_in + 16 * 9] + mova m4, [rsp + transposed_in + 16 * 23] + BUTTERFLY_4X 3, 4, 7005, 14811, m8, 9, 10 ; stp1_18, stp1_29 + + ; BLOCK A STAGE 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + SUM_SUB 1, 2, 9 ; stp2_16, stp2_17 + SUM_SUB 12, 3, 9 ; stp2_19, stp2_18 + SUM_SUB 7, 4, 9 ; stp2_28, stp2_29 + SUM_SUB 11, 0, 9 ; stp2_31, stp2_30 + + ; BLOCK A STAGE 3 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + BUTTERFLY_4X 0, 2, 3196, 16069, m8, 9, 10 ; stp1_17, stp1_30 + BUTTERFLY_4Xmm 4, 3, 3196, 16069, m8, 9, 10 ; stp1_29, stp1_18 + + ; BLOCK A STAGE 4 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + SUM_SUB 1, 12, 9 ; stp2_16, stp2_19 + SUM_SUB 0, 3, 9 ; stp2_17, stp2_18 + SUM_SUB 11, 7, 9 ; stp2_31, stp2_28 + SUM_SUB 2, 4, 9 ; stp2_30, stp2_29 + + ; BLOCK A STAGE 5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + BUTTERFLY_4X 4, 3, 6270, 15137, m8, 9, 10 ; stp1_18, stp1_29 + BUTTERFLY_4X 7, 12, 6270, 15137, m8, 9, 10 ; stp1_19, stp1_28 + + mova [stp + %3 + idx16], m1 + mova [stp + %3 + idx17], m0 + mova [stp + %3 + idx18], m4 + mova [stp + %3 + idx19], m7 + mova [stp + %4 + idx28], m12 + mova [stp + %4 + idx29], m3 + mova [stp + %4 + idx30], m2 + mova [stp + %4 + idx31], m11 + + ; BLOCK B STAGE 1 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova m5, [rsp + transposed_in + 16 * 5] + mova m6, [rsp + transposed_in + 16 * 27] + BUTTERFLY_4X 5, 6, 3981, 15893, m8, 9, 10 ; stp1_20, stp1_27 + + mova m13, [rsp + transposed_in + 16 * 21] + mova m14, [rsp + transposed_in + 16 * 11] + BUTTERFLY_4X 13, 14, 14053, 8423, m8, 9, 10 ; stp1_21, stp1_26 + + mova m0, [rsp + transposed_in + 16 * 13] + mova m1, [rsp + transposed_in + 16 * 19] + BUTTERFLY_4X 0, 1, 9760, 13160, m8, 9, 10 ; stp1_22, stp1_25 + + mova m2, [rsp + transposed_in + 16 * 3] + mova m3, [rsp + transposed_in + 16 * 29] + BUTTERFLY_4X 3, 2, 16207, 2404, m8, 9, 10 ; stp1_23, stp1_24 + + ; BLOCK B STAGE 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + SUM_SUB 5, 13, 9 ; stp2_20, stp2_21 + SUM_SUB 3, 0, 9 ; stp2_23, stp2_22 + SUM_SUB 2, 1, 9 ; stp2_24, stp2_25 + SUM_SUB 6, 14, 9 ; stp2_27, stp2_26 + + ; BLOCK B STAGE 3 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + BUTTERFLY_4X 14, 13, 13623, 9102, m8, 9, 10 ; stp1_21, stp1_26 + BUTTERFLY_4Xmm 1, 0, 13623, 9102, m8, 9, 10 ; stp1_25, stp1_22 + + ; BLOCK B STAGE 4 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + SUM_SUB 3, 5, 9 ; stp2_23, stp2_20 + SUM_SUB 0, 14, 9 ; stp2_22, stp2_21 + SUM_SUB 2, 6, 9 ; stp2_24, stp2_27 + SUM_SUB 1, 13, 9 ; stp2_25, stp2_26 + + ; BLOCK B STAGE 5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + BUTTERFLY_4Xmm 6, 5, 6270, 15137, m8, 9, 10 ; stp1_27, stp1_20 + BUTTERFLY_4Xmm 13, 14, 6270, 15137, m8, 9, 10 ; stp1_26, stp1_21 + + ; BLOCK B STAGE 6 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova m4, [stp + %3 + idx16] + mova m7, [stp + %3 + idx17] + mova m11, [stp + %3 + idx18] + mova m12, [stp + %3 + idx19] + SUM_SUB 4, 3, 9 ; stp2_16, stp2_23 + SUM_SUB 7, 0, 9 ; stp2_17, stp2_22 + SUM_SUB 11, 14, 9 ; stp2_18, stp2_21 + SUM_SUB 12, 5, 9 ; stp2_19, stp2_20 + mova [stp + %3 + idx16], m4 + mova [stp + %3 + idx17], m7 + mova [stp + %3 + idx18], m11 + mova [stp + %3 + idx19], m12 + + mova m4, [stp + %4 + idx28] + mova m7, [stp + %4 + idx29] + mova m11, [stp + %4 + idx30] + mova m12, [stp + %4 + idx31] + SUM_SUB 4, 6, 9 ; stp2_28, stp2_27 + SUM_SUB 7, 13, 9 ; stp2_29, stp2_26 + SUM_SUB 11, 1, 9 ; stp2_30, stp2_25 + SUM_SUB 12, 2, 9 ; stp2_31, stp2_24 + mova [stp + %4 + idx28], m4 + mova [stp + %4 + idx29], m7 + mova [stp + %4 + idx30], m11 + mova [stp + %4 + idx31], m12 + + ; BLOCK B STAGE 7 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ +%if 0 ; overflow occurs in SUM_SUB when using test streams + mova m10, [pw_11585x2] + SUM_SUB 6, 5, 9 + pmulhrsw m6, m10 ; stp1_27 + pmulhrsw m5, m10 ; stp1_20 + SUM_SUB 13, 14, 9 + pmulhrsw m13, m10 ; stp1_26 + pmulhrsw m14, m10 ; stp1_21 + SUM_SUB 1, 0, 9 + pmulhrsw m1, m10 ; stp1_25 + pmulhrsw m0, m10 ; stp1_22 + SUM_SUB 2, 3, 9 + pmulhrsw m2, m10 ; stp1_25 + pmulhrsw m3, m10 ; stp1_22 +%else + BUTTERFLY_4X 6, 5, 11585, 11585, m8, 9, 10 ; stp1_20, stp1_27 + SWAP 6, 5 + BUTTERFLY_4X 13, 14, 11585, 11585, m8, 9, 10 ; stp1_21, stp1_26 + SWAP 13, 14 + BUTTERFLY_4X 1, 0, 11585, 11585, m8, 9, 10 ; stp1_22, stp1_25 + SWAP 1, 0 + BUTTERFLY_4X 2, 3, 11585, 11585, m8, 9, 10 ; stp1_23, stp1_24 + SWAP 2, 3 +%endif + mova [stp + %3 + idx20], m5 + mova [stp + %3 + idx21], m14 + mova [stp + %3 + idx22], m0 + mova [stp + %3 + idx23], m3 + mova [stp + %4 + idx24], m2 + mova [stp + %4 + idx25], m1 + mova [stp + %4 + idx26], m13 + mova [stp + %4 + idx27], m6 + + ; BLOCK C STAGE 1 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + ; + ; BLOCK C STAGE 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova m0, [rsp + transposed_in + 16 * 2] + mova m1, [rsp + transposed_in + 16 * 30] + BUTTERFLY_4X 0, 1, 1606, 16305, m8, 9, 10 ; stp1_8, stp1_15 + + mova m2, [rsp + transposed_in + 16 * 14] + mova m3, [rsp + transposed_in + 16 * 18] + BUTTERFLY_4X 3, 2, 12665, 10394, m8, 9, 10 ; stp1_9, stp1_14 + + mova m4, [rsp + transposed_in + 16 * 10] + mova m5, [rsp + transposed_in + 16 * 22] + BUTTERFLY_4X 4, 5, 7723, 14449, m8, 9, 10 ; stp1_10, stp1_13 + + mova m6, [rsp + transposed_in + 16 * 6] + mova m7, [rsp + transposed_in + 16 * 26] + BUTTERFLY_4X 7, 6, 15679, 4756, m8, 9, 10 ; stp1_11, stp1_12 + + ; BLOCK C STAGE 3 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + SUM_SUB 0, 3, 9 ; stp1_8, stp1_9 + SUM_SUB 7, 4, 9 ; stp1_11, stp1_10 + SUM_SUB 6, 5, 9 ; stp1_12, stp1_13 + SUM_SUB 1, 2, 9 ; stp1_15, stp1_14 + + ; BLOCK C STAGE 4 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + BUTTERFLY_4X 2, 3, 6270, 15137, m8, 9, 10 ; stp1_9, stp1_14 + BUTTERFLY_4Xmm 5, 4, 6270, 15137, m8, 9, 10 ; stp1_13, stp1_10 + + ; BLOCK C STAGE 5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + SUM_SUB 0, 7, 9 ; stp1_8, stp1_11 + SUM_SUB 2, 4, 9 ; stp1_9, stp1_10 + SUM_SUB 1, 6, 9 ; stp1_15, stp1_12 + SUM_SUB 3, 5, 9 ; stp1_14, stp1_13 + + ; BLOCK C STAGE 6 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ +%if 0 ; overflow occurs in SUM_SUB when using test streams + mova m10, [pw_11585x2] + SUM_SUB 5, 4, 9 + pmulhrsw m5, m10 ; stp1_13 + pmulhrsw m4, m10 ; stp1_10 + SUM_SUB 6, 7, 9 + pmulhrsw m6, m10 ; stp1_12 + pmulhrsw m7, m10 ; stp1_11 +%else + BUTTERFLY_4X 5, 4, 11585, 11585, m8, 9, 10 ; stp1_10, stp1_13 + SWAP 5, 4 + BUTTERFLY_4X 6, 7, 11585, 11585, m8, 9, 10 ; stp1_11, stp1_12 + SWAP 6, 7 +%endif + ; BLOCK C STAGE 7 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova [stp + %2 + idx8], m0 + mova [stp + %2 + idx9], m2 + mova [stp + %2 + idx10], m4 + mova [stp + %2 + idx11], m7 + mova [stp + %2 + idx12], m6 + mova [stp + %2 + idx13], m5 + mova [stp + %2 + idx14], m3 + mova [stp + %2 + idx15], m1 + + ; BLOCK D STAGE 1 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + ; + ; BLOCK D STAGE 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + ; + ; BLOCK D STAGE 3 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova m11, [rsp + transposed_in + 16 * 4] + mova m12, [rsp + transposed_in + 16 * 28] + BUTTERFLY_4X 11, 12, 3196, 16069, m8, 9, 10 ; stp1_4, stp1_7 + + mova m13, [rsp + transposed_in + 16 * 12] + mova m14, [rsp + transposed_in + 16 * 20] + BUTTERFLY_4X 14, 13, 13623, 9102, m8, 9, 10 ; stp1_5, stp1_6 + + ; BLOCK D STAGE 4 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova m0, [rsp + transposed_in + 16 * 0] + mova m1, [rsp + transposed_in + 16 * 16] + +%if 0 ; overflow occurs in SUM_SUB when using test streams + mova m10, [pw_11585x2] + SUM_SUB 0, 1, 9 + pmulhrsw m0, m10 ; stp1_1 + pmulhrsw m1, m10 ; stp1_0 +%else + BUTTERFLY_4X 0, 1, 11585, 11585, m8, 9, 10 ; stp1_1, stp1_0 + SWAP 0, 1 +%endif + mova m2, [rsp + transposed_in + 16 * 8] + mova m3, [rsp + transposed_in + 16 * 24] + BUTTERFLY_4X 2, 3, 6270, 15137, m8, 9, 10 ; stp1_2, stp1_3 + + mova m10, [pw_11585x2] + SUM_SUB 11, 14, 9 ; stp1_4, stp1_5 + SUM_SUB 12, 13, 9 ; stp1_7, stp1_6 + + ; BLOCK D STAGE 5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ +%if 0 ; overflow occurs in SUM_SUB when using test streams + SUM_SUB 13, 14, 9 + pmulhrsw m13, m10 ; stp1_6 + pmulhrsw m14, m10 ; stp1_5 +%else + BUTTERFLY_4X 13, 14, 11585, 11585, m8, 9, 10 ; stp1_5, stp1_6 + SWAP 13, 14 +%endif + SUM_SUB 0, 3, 9 ; stp1_0, stp1_3 + SUM_SUB 1, 2, 9 ; stp1_1, stp1_2 + + ; BLOCK D STAGE 6 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + SUM_SUB 0, 12, 9 ; stp1_0, stp1_7 + SUM_SUB 1, 13, 9 ; stp1_1, stp1_6 + SUM_SUB 2, 14, 9 ; stp1_2, stp1_5 + SUM_SUB 3, 11, 9 ; stp1_3, stp1_4 + + ; BLOCK D STAGE 7 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + mova m4, [stp + %2 + idx12] + mova m5, [stp + %2 + idx13] + mova m6, [stp + %2 + idx14] + mova m7, [stp + %2 + idx15] + SUM_SUB 0, 7, 9 ; stp1_0, stp1_15 + SUM_SUB 1, 6, 9 ; stp1_1, stp1_14 + SUM_SUB 2, 5, 9 ; stp1_2, stp1_13 + SUM_SUB 3, 4, 9 ; stp1_3, stp1_12 + + ; 0-3, 28-31 final stage + mova m10, [stp + %4 + idx31] + mova m15, [stp + %4 + idx30] + SUM_SUB 0, 10, 9 ; stp1_0, stp1_31 + SUM_SUB 1, 15, 9 ; stp1_1, stp1_30 + mova [stp + %1 + idx0], m0 + mova [stp + %1 + idx1], m1 + mova [stp + %4 + idx31], m10 + mova [stp + %4 + idx30], m15 + mova m0, [stp + %4 + idx29] + mova m1, [stp + %4 + idx28] + SUM_SUB 2, 0, 9 ; stp1_2, stp1_29 + SUM_SUB 3, 1, 9 ; stp1_3, stp1_28 + mova [stp + %1 + idx2], m2 + mova [stp + %1 + idx3], m3 + mova [stp + %4 + idx29], m0 + mova [stp + %4 + idx28], m1 + + ; 12-15, 16-19 final stage + mova m0, [stp + %3 + idx16] + mova m1, [stp + %3 + idx17] + mova m2, [stp + %3 + idx18] + mova m3, [stp + %3 + idx19] + SUM_SUB 7, 0, 9 ; stp1_15, stp1_16 + SUM_SUB 6, 1, 9 ; stp1_14, stp1_17 + SUM_SUB 5, 2, 9 ; stp1_13, stp1_18 + SUM_SUB 4, 3, 9 ; stp1_12, stp1_19 + mova [stp + %2 + idx12], m4 + mova [stp + %2 + idx13], m5 + mova [stp + %2 + idx14], m6 + mova [stp + %2 + idx15], m7 + mova [stp + %3 + idx16], m0 + mova [stp + %3 + idx17], m1 + mova [stp + %3 + idx18], m2 + mova [stp + %3 + idx19], m3 + + mova m4, [stp + %2 + idx8] + mova m5, [stp + %2 + idx9] + mova m6, [stp + %2 + idx10] + mova m7, [stp + %2 + idx11] + SUM_SUB 11, 7, 9 ; stp1_4, stp1_11 + SUM_SUB 14, 6, 9 ; stp1_5, stp1_10 + SUM_SUB 13, 5, 9 ; stp1_6, stp1_9 + SUM_SUB 12, 4, 9 ; stp1_7, stp1_8 + + ; 4-7, 24-27 final stage + mova m3, [stp + %4 + idx24] + mova m2, [stp + %4 + idx25] + mova m1, [stp + %4 + idx26] + mova m0, [stp + %4 + idx27] + SUM_SUB 12, 3, 9 ; stp1_7, stp1_24 + SUM_SUB 13, 2, 9 ; stp1_6, stp1_25 + SUM_SUB 14, 1, 9 ; stp1_5, stp1_26 + SUM_SUB 11, 0, 9 ; stp1_4, stp1_27 + mova [stp + %4 + idx24], m3 + mova [stp + %4 + idx25], m2 + mova [stp + %4 + idx26], m1 + mova [stp + %4 + idx27], m0 + mova [stp + %1 + idx4], m11 + mova [stp + %1 + idx5], m14 + mova [stp + %1 + idx6], m13 + mova [stp + %1 + idx7], m12 + + ; 8-11, 20-23 final stage + mova m0, [stp + %3 + idx20] + mova m1, [stp + %3 + idx21] + mova m2, [stp + %3 + idx22] + mova m3, [stp + %3 + idx23] + SUM_SUB 7, 0, 9 ; stp1_11, stp_20 + SUM_SUB 6, 1, 9 ; stp1_10, stp_21 + SUM_SUB 5, 2, 9 ; stp1_9, stp_22 + SUM_SUB 4, 3, 9 ; stp1_8, stp_23 + mova [stp + %2 + idx8], m4 + mova [stp + %2 + idx9], m5 + mova [stp + %2 + idx10], m6 + mova [stp + %2 + idx11], m7 + mova [stp + %3 + idx20], m0 + mova [stp + %3 + idx21], m1 + mova [stp + %3 + idx22], m2 + mova [stp + %3 + idx23], m3 +%endmacro + +INIT_XMM ssse3 +cglobal idct32x32_1024_add, 3, 11, 16, i32x32_size, input, output, stride + mova m8, [pd_8192] + mov r6, 4 + lea stp, [rsp + pass_one_start] + +idct32x32_1024: + mov r3, inputq + lea r4, [rsp + transposed_in] + mov r7, 4 + +idct32x32_1024_transpose: +%if CONFIG_VP9_HIGHBITDEPTH + mova m0, [r3 + 0] + packssdw m0, [r3 + 16] + mova m1, [r3 + 32 * 4] + packssdw m1, [r3 + 32 * 4 + 16] + mova m2, [r3 + 32 * 8] + packssdw m2, [r3 + 32 * 8 + 16] + mova m3, [r3 + 32 * 12] + packssdw m3, [r3 + 32 * 12 + 16] + mova m4, [r3 + 32 * 16] + packssdw m4, [r3 + 32 * 16 + 16] + mova m5, [r3 + 32 * 20] + packssdw m5, [r3 + 32 * 20 + 16] + mova m6, [r3 + 32 * 24] + packssdw m6, [r3 + 32 * 24 + 16] + mova m7, [r3 + 32 * 28] + packssdw m7, [r3 + 32 * 28 + 16] +%else + mova m0, [r3 + 0] + mova m1, [r3 + 16 * 4] + mova m2, [r3 + 16 * 8] + mova m3, [r3 + 16 * 12] + mova m4, [r3 + 16 * 16] + mova m5, [r3 + 16 * 20] + mova m6, [r3 + 16 * 24] + mova m7, [r3 + 16 * 28] +%endif + + TRANSPOSE8X8 0, 1, 2, 3, 4, 5, 6, 7, 9 + + mova [r4 + 0], m0 + mova [r4 + 16 * 1], m1 + mova [r4 + 16 * 2], m2 + mova [r4 + 16 * 3], m3 + mova [r4 + 16 * 4], m4 + mova [r4 + 16 * 5], m5 + mova [r4 + 16 * 6], m6 + mova [r4 + 16 * 7], m7 +%if CONFIG_VP9_HIGHBITDEPTH + add r3, 32 +%else + add r3, 16 +%endif + add r4, 16 * 8 + dec r7 + jne idct32x32_1024_transpose + + IDCT32X32_1024 16*0, 16*32, 16*64, 16*96 + + lea stp, [stp + 16 * 8] +%if CONFIG_VP9_HIGHBITDEPTH + lea inputq, [inputq + 32 * 32] +%else + lea inputq, [inputq + 16 * 32] +%endif + dec r6 + jnz idct32x32_1024 + + mov r6, 4 + lea stp, [rsp + pass_one_start] + lea r9, [rsp + pass_one_start] + +idct32x32_1024_2: + lea r4, [rsp + transposed_in] + mov r3, r9 + mov r7, 4 + +idct32x32_1024_transpose_2: + mova m0, [r3 + 0] + mova m1, [r3 + 16 * 1] + mova m2, [r3 + 16 * 2] + mova m3, [r3 + 16 * 3] + mova m4, [r3 + 16 * 4] + mova m5, [r3 + 16 * 5] + mova m6, [r3 + 16 * 6] + mova m7, [r3 + 16 * 7] + + TRANSPOSE8X8 0, 1, 2, 3, 4, 5, 6, 7, 9 + + mova [r4 + 0], m0 + mova [r4 + 16 * 1], m1 + mova [r4 + 16 * 2], m2 + mova [r4 + 16 * 3], m3 + mova [r4 + 16 * 4], m4 + mova [r4 + 16 * 5], m5 + mova [r4 + 16 * 6], m6 + mova [r4 + 16 * 7], m7 + + add r3, 16 * 8 + add r4, 16 * 8 + dec r7 + jne idct32x32_1024_transpose_2 + + IDCT32X32_1024 16*0, 16*8, 16*16, 16*24 + + lea stp, [stp + 16 * 32] + add r9, 16 * 32 + dec r6 + jnz idct32x32_1024_2 + + RECON_AND_STORE pass_two_start + + RET +%endif diff --git a/thirdparty/libvpx/vpx_dsp/x86/inv_wht_sse2.asm b/thirdparty/libvpx/vpx_dsp/x86/inv_wht_sse2.asm new file mode 100644 index 0000000000..fbbcd76bd7 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/x86/inv_wht_sse2.asm @@ -0,0 +1,109 @@ +; +; Copyright (c) 2015 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + +%include "third_party/x86inc/x86inc.asm" + +SECTION .text + +%macro REORDER_INPUTS 0 + ; a c d b to a b c d + SWAP 1, 3, 2 +%endmacro + +%macro TRANSFORM_COLS 0 + ; input: + ; m0 a + ; m1 b + ; m2 c + ; m3 d + paddw m0, m2 + psubw m3, m1 + + ; wide subtract + punpcklwd m4, m0 + punpcklwd m5, m3 + psrad m4, 16 + psrad m5, 16 + psubd m4, m5 + psrad m4, 1 + packssdw m4, m4 ; e + + psubw m5, m4, m1 ; b + psubw m4, m2 ; c + psubw m0, m5 + paddw m3, m4 + ; m0 a + SWAP 1, 5 ; m1 b + SWAP 2, 4 ; m2 c + ; m3 d +%endmacro + +%macro TRANSPOSE_4X4 0 + punpcklwd m0, m2 + punpcklwd m1, m3 + mova m2, m0 + punpcklwd m0, m1 + punpckhwd m2, m1 + pshufd m1, m0, 0x0e + pshufd m3, m2, 0x0e +%endmacro + +; transpose a 4x4 int16 matrix in xmm0 and xmm1 to the bottom half of xmm0-xmm3 +%macro TRANSPOSE_4X4_WIDE 0 + mova m3, m0 + punpcklwd m0, m1 + punpckhwd m3, m1 + mova m2, m0 + punpcklwd m0, m3 + punpckhwd m2, m3 + pshufd m1, m0, 0x0e + pshufd m3, m2, 0x0e +%endmacro + +%macro ADD_STORE_4P_2X 5 ; src1, src2, tmp1, tmp2, zero + movd m%3, [outputq] + movd m%4, [outputq + strideq] + punpcklbw m%3, m%5 + punpcklbw m%4, m%5 + paddw m%1, m%3 + paddw m%2, m%4 + packuswb m%1, m%5 + packuswb m%2, m%5 + movd [outputq], m%1 + movd [outputq + strideq], m%2 +%endmacro + +INIT_XMM sse2 +cglobal iwht4x4_16_add, 3, 3, 7, input, output, stride +%if CONFIG_VP9_HIGHBITDEPTH + mova m0, [inputq + 0] + packssdw m0, [inputq + 16] + mova m1, [inputq + 32] + packssdw m1, [inputq + 48] +%else + mova m0, [inputq + 0] + mova m1, [inputq + 16] +%endif + psraw m0, 2 + psraw m1, 2 + + TRANSPOSE_4X4_WIDE + REORDER_INPUTS + TRANSFORM_COLS + TRANSPOSE_4X4 + REORDER_INPUTS + TRANSFORM_COLS + + pxor m4, m4 + ADD_STORE_4P_2X 0, 1, 5, 6, 4 + lea outputq, [outputq + 2 * strideq] + ADD_STORE_4P_2X 2, 3, 5, 6, 4 + + RET diff --git a/thirdparty/libvpx/vpx_dsp/x86/loopfilter_avx2.c b/thirdparty/libvpx/vpx_dsp/x86/loopfilter_avx2.c new file mode 100644 index 0000000000..be1087c1e9 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/x86/loopfilter_avx2.c @@ -0,0 +1,979 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <immintrin.h> /* AVX2 */ + +#include "./vpx_dsp_rtcd.h" +#include "vpx_ports/mem.h" + +void vpx_lpf_horizontal_edge_8_avx2(unsigned char *s, int p, + const unsigned char *_blimit, + const unsigned char *_limit, + const unsigned char *_thresh) { + __m128i mask, hev, flat, flat2; + const __m128i zero = _mm_set1_epi16(0); + const __m128i one = _mm_set1_epi8(1); + __m128i q7p7, q6p6, q5p5, q4p4, q3p3, q2p2, q1p1, q0p0, p0q0, p1q1; + __m128i abs_p1p0; + + const __m128i thresh = _mm_broadcastb_epi8( + _mm_cvtsi32_si128((int) _thresh[0])); + const __m128i limit = _mm_broadcastb_epi8( + _mm_cvtsi32_si128((int) _limit[0])); + const __m128i blimit = _mm_broadcastb_epi8( + _mm_cvtsi32_si128((int) _blimit[0])); + + q4p4 = _mm_loadl_epi64((__m128i *) (s - 5 * p)); + q4p4 = _mm_castps_si128( + _mm_loadh_pi(_mm_castsi128_ps(q4p4), (__m64 *) (s + 4 * p))); + q3p3 = _mm_loadl_epi64((__m128i *) (s - 4 * p)); + q3p3 = _mm_castps_si128( + _mm_loadh_pi(_mm_castsi128_ps(q3p3), (__m64 *) (s + 3 * p))); + q2p2 = _mm_loadl_epi64((__m128i *) (s - 3 * p)); + q2p2 = _mm_castps_si128( + _mm_loadh_pi(_mm_castsi128_ps(q2p2), (__m64 *) (s + 2 * p))); + q1p1 = _mm_loadl_epi64((__m128i *) (s - 2 * p)); + q1p1 = _mm_castps_si128( + _mm_loadh_pi(_mm_castsi128_ps(q1p1), (__m64 *) (s + 1 * p))); + p1q1 = _mm_shuffle_epi32(q1p1, 78); + q0p0 = _mm_loadl_epi64((__m128i *) (s - 1 * p)); + q0p0 = _mm_castps_si128( + _mm_loadh_pi(_mm_castsi128_ps(q0p0), (__m64 *) (s - 0 * p))); + p0q0 = _mm_shuffle_epi32(q0p0, 78); + + { + __m128i abs_p1q1, abs_p0q0, abs_q1q0, fe, ff, work; + abs_p1p0 = _mm_or_si128(_mm_subs_epu8(q1p1, q0p0), + _mm_subs_epu8(q0p0, q1p1)); + abs_q1q0 = _mm_srli_si128(abs_p1p0, 8); + fe = _mm_set1_epi8(0xfe); + ff = _mm_cmpeq_epi8(abs_p1p0, abs_p1p0); + abs_p0q0 = _mm_or_si128(_mm_subs_epu8(q0p0, p0q0), + _mm_subs_epu8(p0q0, q0p0)); + abs_p1q1 = _mm_or_si128(_mm_subs_epu8(q1p1, p1q1), + _mm_subs_epu8(p1q1, q1p1)); + flat = _mm_max_epu8(abs_p1p0, abs_q1q0); + hev = _mm_subs_epu8(flat, thresh); + hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff); + + abs_p0q0 = _mm_adds_epu8(abs_p0q0, abs_p0q0); + abs_p1q1 = _mm_srli_epi16(_mm_and_si128(abs_p1q1, fe), 1); + mask = _mm_subs_epu8(_mm_adds_epu8(abs_p0q0, abs_p1q1), blimit); + mask = _mm_xor_si128(_mm_cmpeq_epi8(mask, zero), ff); + // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2 > blimit) * -1; + mask = _mm_max_epu8(abs_p1p0, mask); + // mask |= (abs(p1 - p0) > limit) * -1; + // mask |= (abs(q1 - q0) > limit) * -1; + + work = _mm_max_epu8( + _mm_or_si128(_mm_subs_epu8(q2p2, q1p1), + _mm_subs_epu8(q1p1, q2p2)), + _mm_or_si128(_mm_subs_epu8(q3p3, q2p2), + _mm_subs_epu8(q2p2, q3p3))); + mask = _mm_max_epu8(work, mask); + mask = _mm_max_epu8(mask, _mm_srli_si128(mask, 8)); + mask = _mm_subs_epu8(mask, limit); + mask = _mm_cmpeq_epi8(mask, zero); + } + + // lp filter + { + const __m128i t4 = _mm_set1_epi8(4); + const __m128i t3 = _mm_set1_epi8(3); + const __m128i t80 = _mm_set1_epi8(0x80); + const __m128i t1 = _mm_set1_epi16(0x1); + __m128i qs1ps1 = _mm_xor_si128(q1p1, t80); + __m128i qs0ps0 = _mm_xor_si128(q0p0, t80); + __m128i qs0 = _mm_xor_si128(p0q0, t80); + __m128i qs1 = _mm_xor_si128(p1q1, t80); + __m128i filt; + __m128i work_a; + __m128i filter1, filter2; + __m128i flat2_q6p6, flat2_q5p5, flat2_q4p4, flat2_q3p3, flat2_q2p2; + __m128i flat2_q1p1, flat2_q0p0, flat_q2p2, flat_q1p1, flat_q0p0; + + filt = _mm_and_si128(_mm_subs_epi8(qs1ps1, qs1), hev); + work_a = _mm_subs_epi8(qs0, qs0ps0); + filt = _mm_adds_epi8(filt, work_a); + filt = _mm_adds_epi8(filt, work_a); + filt = _mm_adds_epi8(filt, work_a); + /* (vpx_filter + 3 * (qs0 - ps0)) & mask */ + filt = _mm_and_si128(filt, mask); + + filter1 = _mm_adds_epi8(filt, t4); + filter2 = _mm_adds_epi8(filt, t3); + + filter1 = _mm_unpacklo_epi8(zero, filter1); + filter1 = _mm_srai_epi16(filter1, 0xB); + filter2 = _mm_unpacklo_epi8(zero, filter2); + filter2 = _mm_srai_epi16(filter2, 0xB); + + /* Filter1 >> 3 */ + filt = _mm_packs_epi16(filter2, _mm_subs_epi16(zero, filter1)); + qs0ps0 = _mm_xor_si128(_mm_adds_epi8(qs0ps0, filt), t80); + + /* filt >> 1 */ + filt = _mm_adds_epi16(filter1, t1); + filt = _mm_srai_epi16(filt, 1); + filt = _mm_andnot_si128( + _mm_srai_epi16(_mm_unpacklo_epi8(zero, hev), 0x8), filt); + filt = _mm_packs_epi16(filt, _mm_subs_epi16(zero, filt)); + qs1ps1 = _mm_xor_si128(_mm_adds_epi8(qs1ps1, filt), t80); + // loopfilter done + + { + __m128i work; + flat = _mm_max_epu8( + _mm_or_si128(_mm_subs_epu8(q2p2, q0p0), + _mm_subs_epu8(q0p0, q2p2)), + _mm_or_si128(_mm_subs_epu8(q3p3, q0p0), + _mm_subs_epu8(q0p0, q3p3))); + flat = _mm_max_epu8(abs_p1p0, flat); + flat = _mm_max_epu8(flat, _mm_srli_si128(flat, 8)); + flat = _mm_subs_epu8(flat, one); + flat = _mm_cmpeq_epi8(flat, zero); + flat = _mm_and_si128(flat, mask); + + q5p5 = _mm_loadl_epi64((__m128i *) (s - 6 * p)); + q5p5 = _mm_castps_si128( + _mm_loadh_pi(_mm_castsi128_ps(q5p5), + (__m64 *) (s + 5 * p))); + + q6p6 = _mm_loadl_epi64((__m128i *) (s - 7 * p)); + q6p6 = _mm_castps_si128( + _mm_loadh_pi(_mm_castsi128_ps(q6p6), + (__m64 *) (s + 6 * p))); + + flat2 = _mm_max_epu8( + _mm_or_si128(_mm_subs_epu8(q4p4, q0p0), + _mm_subs_epu8(q0p0, q4p4)), + _mm_or_si128(_mm_subs_epu8(q5p5, q0p0), + _mm_subs_epu8(q0p0, q5p5))); + + q7p7 = _mm_loadl_epi64((__m128i *) (s - 8 * p)); + q7p7 = _mm_castps_si128( + _mm_loadh_pi(_mm_castsi128_ps(q7p7), + (__m64 *) (s + 7 * p))); + + work = _mm_max_epu8( + _mm_or_si128(_mm_subs_epu8(q6p6, q0p0), + _mm_subs_epu8(q0p0, q6p6)), + _mm_or_si128(_mm_subs_epu8(q7p7, q0p0), + _mm_subs_epu8(q0p0, q7p7))); + + flat2 = _mm_max_epu8(work, flat2); + flat2 = _mm_max_epu8(flat2, _mm_srli_si128(flat2, 8)); + flat2 = _mm_subs_epu8(flat2, one); + flat2 = _mm_cmpeq_epi8(flat2, zero); + flat2 = _mm_and_si128(flat2, flat); // flat2 & flat & mask + } + + // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + // flat and wide flat calculations + { + const __m128i eight = _mm_set1_epi16(8); + const __m128i four = _mm_set1_epi16(4); + __m128i p7_16, p6_16, p5_16, p4_16, p3_16, p2_16, p1_16, p0_16; + __m128i q7_16, q6_16, q5_16, q4_16, q3_16, q2_16, q1_16, q0_16; + __m128i pixelFilter_p, pixelFilter_q; + __m128i pixetFilter_p2p1p0, pixetFilter_q2q1q0; + __m128i sum_p7, sum_q7, sum_p3, sum_q3, res_p, res_q; + + p7_16 = _mm_unpacklo_epi8(q7p7, zero); + p6_16 = _mm_unpacklo_epi8(q6p6, zero); + p5_16 = _mm_unpacklo_epi8(q5p5, zero); + p4_16 = _mm_unpacklo_epi8(q4p4, zero); + p3_16 = _mm_unpacklo_epi8(q3p3, zero); + p2_16 = _mm_unpacklo_epi8(q2p2, zero); + p1_16 = _mm_unpacklo_epi8(q1p1, zero); + p0_16 = _mm_unpacklo_epi8(q0p0, zero); + q0_16 = _mm_unpackhi_epi8(q0p0, zero); + q1_16 = _mm_unpackhi_epi8(q1p1, zero); + q2_16 = _mm_unpackhi_epi8(q2p2, zero); + q3_16 = _mm_unpackhi_epi8(q3p3, zero); + q4_16 = _mm_unpackhi_epi8(q4p4, zero); + q5_16 = _mm_unpackhi_epi8(q5p5, zero); + q6_16 = _mm_unpackhi_epi8(q6p6, zero); + q7_16 = _mm_unpackhi_epi8(q7p7, zero); + + pixelFilter_p = _mm_add_epi16(_mm_add_epi16(p6_16, p5_16), + _mm_add_epi16(p4_16, p3_16)); + pixelFilter_q = _mm_add_epi16(_mm_add_epi16(q6_16, q5_16), + _mm_add_epi16(q4_16, q3_16)); + + pixetFilter_p2p1p0 = _mm_add_epi16(p0_16, + _mm_add_epi16(p2_16, p1_16)); + pixelFilter_p = _mm_add_epi16(pixelFilter_p, pixetFilter_p2p1p0); + + pixetFilter_q2q1q0 = _mm_add_epi16(q0_16, + _mm_add_epi16(q2_16, q1_16)); + pixelFilter_q = _mm_add_epi16(pixelFilter_q, pixetFilter_q2q1q0); + pixelFilter_p = _mm_add_epi16(eight, + _mm_add_epi16(pixelFilter_p, pixelFilter_q)); + pixetFilter_p2p1p0 = _mm_add_epi16(four, + _mm_add_epi16(pixetFilter_p2p1p0, pixetFilter_q2q1q0)); + res_p = _mm_srli_epi16( + _mm_add_epi16(pixelFilter_p, _mm_add_epi16(p7_16, p0_16)), + 4); + res_q = _mm_srli_epi16( + _mm_add_epi16(pixelFilter_p, _mm_add_epi16(q7_16, q0_16)), + 4); + flat2_q0p0 = _mm_packus_epi16(res_p, res_q); + res_p = _mm_srli_epi16( + _mm_add_epi16(pixetFilter_p2p1p0, + _mm_add_epi16(p3_16, p0_16)), 3); + res_q = _mm_srli_epi16( + _mm_add_epi16(pixetFilter_p2p1p0, + _mm_add_epi16(q3_16, q0_16)), 3); + + flat_q0p0 = _mm_packus_epi16(res_p, res_q); + + sum_p7 = _mm_add_epi16(p7_16, p7_16); + sum_q7 = _mm_add_epi16(q7_16, q7_16); + sum_p3 = _mm_add_epi16(p3_16, p3_16); + sum_q3 = _mm_add_epi16(q3_16, q3_16); + + pixelFilter_q = _mm_sub_epi16(pixelFilter_p, p6_16); + pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q6_16); + res_p = _mm_srli_epi16( + _mm_add_epi16(pixelFilter_p, _mm_add_epi16(sum_p7, p1_16)), + 4); + res_q = _mm_srli_epi16( + _mm_add_epi16(pixelFilter_q, _mm_add_epi16(sum_q7, q1_16)), + 4); + flat2_q1p1 = _mm_packus_epi16(res_p, res_q); + + pixetFilter_q2q1q0 = _mm_sub_epi16(pixetFilter_p2p1p0, p2_16); + pixetFilter_p2p1p0 = _mm_sub_epi16(pixetFilter_p2p1p0, q2_16); + res_p = _mm_srli_epi16( + _mm_add_epi16(pixetFilter_p2p1p0, + _mm_add_epi16(sum_p3, p1_16)), 3); + res_q = _mm_srli_epi16( + _mm_add_epi16(pixetFilter_q2q1q0, + _mm_add_epi16(sum_q3, q1_16)), 3); + flat_q1p1 = _mm_packus_epi16(res_p, res_q); + + sum_p7 = _mm_add_epi16(sum_p7, p7_16); + sum_q7 = _mm_add_epi16(sum_q7, q7_16); + sum_p3 = _mm_add_epi16(sum_p3, p3_16); + sum_q3 = _mm_add_epi16(sum_q3, q3_16); + + pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q5_16); + pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p5_16); + res_p = _mm_srli_epi16( + _mm_add_epi16(pixelFilter_p, _mm_add_epi16(sum_p7, p2_16)), + 4); + res_q = _mm_srli_epi16( + _mm_add_epi16(pixelFilter_q, _mm_add_epi16(sum_q7, q2_16)), + 4); + flat2_q2p2 = _mm_packus_epi16(res_p, res_q); + + pixetFilter_p2p1p0 = _mm_sub_epi16(pixetFilter_p2p1p0, q1_16); + pixetFilter_q2q1q0 = _mm_sub_epi16(pixetFilter_q2q1q0, p1_16); + + res_p = _mm_srli_epi16( + _mm_add_epi16(pixetFilter_p2p1p0, + _mm_add_epi16(sum_p3, p2_16)), 3); + res_q = _mm_srli_epi16( + _mm_add_epi16(pixetFilter_q2q1q0, + _mm_add_epi16(sum_q3, q2_16)), 3); + flat_q2p2 = _mm_packus_epi16(res_p, res_q); + + sum_p7 = _mm_add_epi16(sum_p7, p7_16); + sum_q7 = _mm_add_epi16(sum_q7, q7_16); + pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q4_16); + pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p4_16); + res_p = _mm_srli_epi16( + _mm_add_epi16(pixelFilter_p, _mm_add_epi16(sum_p7, p3_16)), + 4); + res_q = _mm_srli_epi16( + _mm_add_epi16(pixelFilter_q, _mm_add_epi16(sum_q7, q3_16)), + 4); + flat2_q3p3 = _mm_packus_epi16(res_p, res_q); + + sum_p7 = _mm_add_epi16(sum_p7, p7_16); + sum_q7 = _mm_add_epi16(sum_q7, q7_16); + pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q3_16); + pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p3_16); + res_p = _mm_srli_epi16( + _mm_add_epi16(pixelFilter_p, _mm_add_epi16(sum_p7, p4_16)), + 4); + res_q = _mm_srli_epi16( + _mm_add_epi16(pixelFilter_q, _mm_add_epi16(sum_q7, q4_16)), + 4); + flat2_q4p4 = _mm_packus_epi16(res_p, res_q); + + sum_p7 = _mm_add_epi16(sum_p7, p7_16); + sum_q7 = _mm_add_epi16(sum_q7, q7_16); + pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q2_16); + pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p2_16); + res_p = _mm_srli_epi16( + _mm_add_epi16(pixelFilter_p, _mm_add_epi16(sum_p7, p5_16)), + 4); + res_q = _mm_srli_epi16( + _mm_add_epi16(pixelFilter_q, _mm_add_epi16(sum_q7, q5_16)), + 4); + flat2_q5p5 = _mm_packus_epi16(res_p, res_q); + + sum_p7 = _mm_add_epi16(sum_p7, p7_16); + sum_q7 = _mm_add_epi16(sum_q7, q7_16); + pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q1_16); + pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p1_16); + res_p = _mm_srli_epi16( + _mm_add_epi16(pixelFilter_p, _mm_add_epi16(sum_p7, p6_16)), + 4); + res_q = _mm_srli_epi16( + _mm_add_epi16(pixelFilter_q, _mm_add_epi16(sum_q7, q6_16)), + 4); + flat2_q6p6 = _mm_packus_epi16(res_p, res_q); + } + // wide flat + // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + + flat = _mm_shuffle_epi32(flat, 68); + flat2 = _mm_shuffle_epi32(flat2, 68); + + q2p2 = _mm_andnot_si128(flat, q2p2); + flat_q2p2 = _mm_and_si128(flat, flat_q2p2); + q2p2 = _mm_or_si128(q2p2, flat_q2p2); + + qs1ps1 = _mm_andnot_si128(flat, qs1ps1); + flat_q1p1 = _mm_and_si128(flat, flat_q1p1); + q1p1 = _mm_or_si128(qs1ps1, flat_q1p1); + + qs0ps0 = _mm_andnot_si128(flat, qs0ps0); + flat_q0p0 = _mm_and_si128(flat, flat_q0p0); + q0p0 = _mm_or_si128(qs0ps0, flat_q0p0); + + q6p6 = _mm_andnot_si128(flat2, q6p6); + flat2_q6p6 = _mm_and_si128(flat2, flat2_q6p6); + q6p6 = _mm_or_si128(q6p6, flat2_q6p6); + _mm_storel_epi64((__m128i *) (s - 7 * p), q6p6); + _mm_storeh_pi((__m64 *) (s + 6 * p), _mm_castsi128_ps(q6p6)); + + q5p5 = _mm_andnot_si128(flat2, q5p5); + flat2_q5p5 = _mm_and_si128(flat2, flat2_q5p5); + q5p5 = _mm_or_si128(q5p5, flat2_q5p5); + _mm_storel_epi64((__m128i *) (s - 6 * p), q5p5); + _mm_storeh_pi((__m64 *) (s + 5 * p), _mm_castsi128_ps(q5p5)); + + q4p4 = _mm_andnot_si128(flat2, q4p4); + flat2_q4p4 = _mm_and_si128(flat2, flat2_q4p4); + q4p4 = _mm_or_si128(q4p4, flat2_q4p4); + _mm_storel_epi64((__m128i *) (s - 5 * p), q4p4); + _mm_storeh_pi((__m64 *) (s + 4 * p), _mm_castsi128_ps(q4p4)); + + q3p3 = _mm_andnot_si128(flat2, q3p3); + flat2_q3p3 = _mm_and_si128(flat2, flat2_q3p3); + q3p3 = _mm_or_si128(q3p3, flat2_q3p3); + _mm_storel_epi64((__m128i *) (s - 4 * p), q3p3); + _mm_storeh_pi((__m64 *) (s + 3 * p), _mm_castsi128_ps(q3p3)); + + q2p2 = _mm_andnot_si128(flat2, q2p2); + flat2_q2p2 = _mm_and_si128(flat2, flat2_q2p2); + q2p2 = _mm_or_si128(q2p2, flat2_q2p2); + _mm_storel_epi64((__m128i *) (s - 3 * p), q2p2); + _mm_storeh_pi((__m64 *) (s + 2 * p), _mm_castsi128_ps(q2p2)); + + q1p1 = _mm_andnot_si128(flat2, q1p1); + flat2_q1p1 = _mm_and_si128(flat2, flat2_q1p1); + q1p1 = _mm_or_si128(q1p1, flat2_q1p1); + _mm_storel_epi64((__m128i *) (s - 2 * p), q1p1); + _mm_storeh_pi((__m64 *) (s + 1 * p), _mm_castsi128_ps(q1p1)); + + q0p0 = _mm_andnot_si128(flat2, q0p0); + flat2_q0p0 = _mm_and_si128(flat2, flat2_q0p0); + q0p0 = _mm_or_si128(q0p0, flat2_q0p0); + _mm_storel_epi64((__m128i *) (s - 1 * p), q0p0); + _mm_storeh_pi((__m64 *) (s - 0 * p), _mm_castsi128_ps(q0p0)); + } +} + +DECLARE_ALIGNED(32, static const uint8_t, filt_loopfilter_avx2[32]) = { + 0, 128, 1, 128, 2, 128, 3, 128, 4, 128, 5, 128, 6, 128, 7, 128, + 8, 128, 9, 128, 10, 128, 11, 128, 12, 128, 13, 128, 14, 128, 15, 128 +}; + +void vpx_lpf_horizontal_edge_16_avx2(unsigned char *s, int p, + const unsigned char *_blimit, + const unsigned char *_limit, + const unsigned char *_thresh) { + __m128i mask, hev, flat, flat2; + const __m128i zero = _mm_set1_epi16(0); + const __m128i one = _mm_set1_epi8(1); + __m128i p7, p6, p5; + __m128i p4, p3, p2, p1, p0, q0, q1, q2, q3, q4; + __m128i q5, q6, q7; + __m256i p256_7, q256_7, p256_6, q256_6, p256_5, q256_5, p256_4, + q256_4, p256_3, q256_3, p256_2, q256_2, p256_1, q256_1, + p256_0, q256_0; + + const __m128i thresh = _mm_broadcastb_epi8( + _mm_cvtsi32_si128((int) _thresh[0])); + const __m128i limit = _mm_broadcastb_epi8( + _mm_cvtsi32_si128((int) _limit[0])); + const __m128i blimit = _mm_broadcastb_epi8( + _mm_cvtsi32_si128((int) _blimit[0])); + + p256_4 = _mm256_castpd_si256(_mm256_broadcast_pd( + (__m128d const *)(s - 5 * p))); + p256_3 = _mm256_castpd_si256(_mm256_broadcast_pd( + (__m128d const *)(s - 4 * p))); + p256_2 = _mm256_castpd_si256(_mm256_broadcast_pd( + (__m128d const *)(s - 3 * p))); + p256_1 = _mm256_castpd_si256(_mm256_broadcast_pd( + (__m128d const *)(s - 2 * p))); + p256_0 = _mm256_castpd_si256(_mm256_broadcast_pd( + (__m128d const *)(s - 1 * p))); + q256_0 = _mm256_castpd_si256(_mm256_broadcast_pd( + (__m128d const *)(s - 0 * p))); + q256_1 = _mm256_castpd_si256(_mm256_broadcast_pd( + (__m128d const *)(s + 1 * p))); + q256_2 = _mm256_castpd_si256(_mm256_broadcast_pd( + (__m128d const *)(s + 2 * p))); + q256_3 = _mm256_castpd_si256(_mm256_broadcast_pd( + (__m128d const *)(s + 3 * p))); + q256_4 = _mm256_castpd_si256(_mm256_broadcast_pd( + (__m128d const *)(s + 4 * p))); + + p4 = _mm256_castsi256_si128(p256_4); + p3 = _mm256_castsi256_si128(p256_3); + p2 = _mm256_castsi256_si128(p256_2); + p1 = _mm256_castsi256_si128(p256_1); + p0 = _mm256_castsi256_si128(p256_0); + q0 = _mm256_castsi256_si128(q256_0); + q1 = _mm256_castsi256_si128(q256_1); + q2 = _mm256_castsi256_si128(q256_2); + q3 = _mm256_castsi256_si128(q256_3); + q4 = _mm256_castsi256_si128(q256_4); + + { + const __m128i abs_p1p0 = _mm_or_si128(_mm_subs_epu8(p1, p0), + _mm_subs_epu8(p0, p1)); + const __m128i abs_q1q0 = _mm_or_si128(_mm_subs_epu8(q1, q0), + _mm_subs_epu8(q0, q1)); + const __m128i fe = _mm_set1_epi8(0xfe); + const __m128i ff = _mm_cmpeq_epi8(abs_p1p0, abs_p1p0); + __m128i abs_p0q0 = _mm_or_si128(_mm_subs_epu8(p0, q0), + _mm_subs_epu8(q0, p0)); + __m128i abs_p1q1 = _mm_or_si128(_mm_subs_epu8(p1, q1), + _mm_subs_epu8(q1, p1)); + __m128i work; + flat = _mm_max_epu8(abs_p1p0, abs_q1q0); + hev = _mm_subs_epu8(flat, thresh); + hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff); + + abs_p0q0 = _mm_adds_epu8(abs_p0q0, abs_p0q0); + abs_p1q1 = _mm_srli_epi16(_mm_and_si128(abs_p1q1, fe), 1); + mask = _mm_subs_epu8(_mm_adds_epu8(abs_p0q0, abs_p1q1), blimit); + mask = _mm_xor_si128(_mm_cmpeq_epi8(mask, zero), ff); + // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2 > blimit) * -1; + mask = _mm_max_epu8(flat, mask); + // mask |= (abs(p1 - p0) > limit) * -1; + // mask |= (abs(q1 - q0) > limit) * -1; + work = _mm_max_epu8( + _mm_or_si128(_mm_subs_epu8(p2, p1), _mm_subs_epu8(p1, p2)), + _mm_or_si128(_mm_subs_epu8(p3, p2), _mm_subs_epu8(p2, p3))); + mask = _mm_max_epu8(work, mask); + work = _mm_max_epu8( + _mm_or_si128(_mm_subs_epu8(q2, q1), _mm_subs_epu8(q1, q2)), + _mm_or_si128(_mm_subs_epu8(q3, q2), _mm_subs_epu8(q2, q3))); + mask = _mm_max_epu8(work, mask); + mask = _mm_subs_epu8(mask, limit); + mask = _mm_cmpeq_epi8(mask, zero); + } + + // lp filter + { + const __m128i t4 = _mm_set1_epi8(4); + const __m128i t3 = _mm_set1_epi8(3); + const __m128i t80 = _mm_set1_epi8(0x80); + const __m128i te0 = _mm_set1_epi8(0xe0); + const __m128i t1f = _mm_set1_epi8(0x1f); + const __m128i t1 = _mm_set1_epi8(0x1); + const __m128i t7f = _mm_set1_epi8(0x7f); + + __m128i ps1 = _mm_xor_si128(p1, t80); + __m128i ps0 = _mm_xor_si128(p0, t80); + __m128i qs0 = _mm_xor_si128(q0, t80); + __m128i qs1 = _mm_xor_si128(q1, t80); + __m128i filt; + __m128i work_a; + __m128i filter1, filter2; + __m128i flat2_p6, flat2_p5, flat2_p4, flat2_p3, flat2_p2, flat2_p1, + flat2_p0, flat2_q0, flat2_q1, flat2_q2, flat2_q3, flat2_q4, + flat2_q5, flat2_q6, flat_p2, flat_p1, flat_p0, flat_q0, flat_q1, + flat_q2; + + filt = _mm_and_si128(_mm_subs_epi8(ps1, qs1), hev); + work_a = _mm_subs_epi8(qs0, ps0); + filt = _mm_adds_epi8(filt, work_a); + filt = _mm_adds_epi8(filt, work_a); + filt = _mm_adds_epi8(filt, work_a); + /* (vpx_filter + 3 * (qs0 - ps0)) & mask */ + filt = _mm_and_si128(filt, mask); + + filter1 = _mm_adds_epi8(filt, t4); + filter2 = _mm_adds_epi8(filt, t3); + + /* Filter1 >> 3 */ + work_a = _mm_cmpgt_epi8(zero, filter1); + filter1 = _mm_srli_epi16(filter1, 3); + work_a = _mm_and_si128(work_a, te0); + filter1 = _mm_and_si128(filter1, t1f); + filter1 = _mm_or_si128(filter1, work_a); + qs0 = _mm_xor_si128(_mm_subs_epi8(qs0, filter1), t80); + + /* Filter2 >> 3 */ + work_a = _mm_cmpgt_epi8(zero, filter2); + filter2 = _mm_srli_epi16(filter2, 3); + work_a = _mm_and_si128(work_a, te0); + filter2 = _mm_and_si128(filter2, t1f); + filter2 = _mm_or_si128(filter2, work_a); + ps0 = _mm_xor_si128(_mm_adds_epi8(ps0, filter2), t80); + + /* filt >> 1 */ + filt = _mm_adds_epi8(filter1, t1); + work_a = _mm_cmpgt_epi8(zero, filt); + filt = _mm_srli_epi16(filt, 1); + work_a = _mm_and_si128(work_a, t80); + filt = _mm_and_si128(filt, t7f); + filt = _mm_or_si128(filt, work_a); + filt = _mm_andnot_si128(hev, filt); + ps1 = _mm_xor_si128(_mm_adds_epi8(ps1, filt), t80); + qs1 = _mm_xor_si128(_mm_subs_epi8(qs1, filt), t80); + // loopfilter done + + { + __m128i work; + work = _mm_max_epu8( + _mm_or_si128(_mm_subs_epu8(p2, p0), _mm_subs_epu8(p0, p2)), + _mm_or_si128(_mm_subs_epu8(q2, q0), _mm_subs_epu8(q0, q2))); + flat = _mm_max_epu8(work, flat); + work = _mm_max_epu8( + _mm_or_si128(_mm_subs_epu8(p3, p0), _mm_subs_epu8(p0, p3)), + _mm_or_si128(_mm_subs_epu8(q3, q0), _mm_subs_epu8(q0, q3))); + flat = _mm_max_epu8(work, flat); + work = _mm_max_epu8( + _mm_or_si128(_mm_subs_epu8(p4, p0), _mm_subs_epu8(p0, p4)), + _mm_or_si128(_mm_subs_epu8(q4, q0), _mm_subs_epu8(q0, q4))); + flat = _mm_subs_epu8(flat, one); + flat = _mm_cmpeq_epi8(flat, zero); + flat = _mm_and_si128(flat, mask); + + p256_5 = _mm256_castpd_si256(_mm256_broadcast_pd( + (__m128d const *)(s - 6 * p))); + q256_5 = _mm256_castpd_si256(_mm256_broadcast_pd( + (__m128d const *)(s + 5 * p))); + p5 = _mm256_castsi256_si128(p256_5); + q5 = _mm256_castsi256_si128(q256_5); + flat2 = _mm_max_epu8( + _mm_or_si128(_mm_subs_epu8(p5, p0), _mm_subs_epu8(p0, p5)), + _mm_or_si128(_mm_subs_epu8(q5, q0), _mm_subs_epu8(q0, q5))); + + flat2 = _mm_max_epu8(work, flat2); + p256_6 = _mm256_castpd_si256(_mm256_broadcast_pd( + (__m128d const *)(s - 7 * p))); + q256_6 = _mm256_castpd_si256(_mm256_broadcast_pd( + (__m128d const *)(s + 6 * p))); + p6 = _mm256_castsi256_si128(p256_6); + q6 = _mm256_castsi256_si128(q256_6); + work = _mm_max_epu8( + _mm_or_si128(_mm_subs_epu8(p6, p0), _mm_subs_epu8(p0, p6)), + _mm_or_si128(_mm_subs_epu8(q6, q0), _mm_subs_epu8(q0, q6))); + + flat2 = _mm_max_epu8(work, flat2); + + p256_7 = _mm256_castpd_si256(_mm256_broadcast_pd( + (__m128d const *)(s - 8 * p))); + q256_7 = _mm256_castpd_si256(_mm256_broadcast_pd( + (__m128d const *)(s + 7 * p))); + p7 = _mm256_castsi256_si128(p256_7); + q7 = _mm256_castsi256_si128(q256_7); + work = _mm_max_epu8( + _mm_or_si128(_mm_subs_epu8(p7, p0), _mm_subs_epu8(p0, p7)), + _mm_or_si128(_mm_subs_epu8(q7, q0), _mm_subs_epu8(q0, q7))); + + flat2 = _mm_max_epu8(work, flat2); + flat2 = _mm_subs_epu8(flat2, one); + flat2 = _mm_cmpeq_epi8(flat2, zero); + flat2 = _mm_and_si128(flat2, flat); // flat2 & flat & mask + } + + // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + // flat and wide flat calculations + { + const __m256i eight = _mm256_set1_epi16(8); + const __m256i four = _mm256_set1_epi16(4); + __m256i pixelFilter_p, pixelFilter_q, pixetFilter_p2p1p0, + pixetFilter_q2q1q0, sum_p7, sum_q7, sum_p3, sum_q3, res_p, + res_q; + + const __m256i filter = _mm256_load_si256( + (__m256i const *)filt_loopfilter_avx2); + p256_7 = _mm256_shuffle_epi8(p256_7, filter); + p256_6 = _mm256_shuffle_epi8(p256_6, filter); + p256_5 = _mm256_shuffle_epi8(p256_5, filter); + p256_4 = _mm256_shuffle_epi8(p256_4, filter); + p256_3 = _mm256_shuffle_epi8(p256_3, filter); + p256_2 = _mm256_shuffle_epi8(p256_2, filter); + p256_1 = _mm256_shuffle_epi8(p256_1, filter); + p256_0 = _mm256_shuffle_epi8(p256_0, filter); + q256_0 = _mm256_shuffle_epi8(q256_0, filter); + q256_1 = _mm256_shuffle_epi8(q256_1, filter); + q256_2 = _mm256_shuffle_epi8(q256_2, filter); + q256_3 = _mm256_shuffle_epi8(q256_3, filter); + q256_4 = _mm256_shuffle_epi8(q256_4, filter); + q256_5 = _mm256_shuffle_epi8(q256_5, filter); + q256_6 = _mm256_shuffle_epi8(q256_6, filter); + q256_7 = _mm256_shuffle_epi8(q256_7, filter); + + pixelFilter_p = _mm256_add_epi16(_mm256_add_epi16(p256_6, p256_5), + _mm256_add_epi16(p256_4, p256_3)); + pixelFilter_q = _mm256_add_epi16(_mm256_add_epi16(q256_6, q256_5), + _mm256_add_epi16(q256_4, q256_3)); + + pixetFilter_p2p1p0 = _mm256_add_epi16(p256_0, + _mm256_add_epi16(p256_2, p256_1)); + pixelFilter_p = _mm256_add_epi16(pixelFilter_p, pixetFilter_p2p1p0); + + pixetFilter_q2q1q0 = _mm256_add_epi16(q256_0, + _mm256_add_epi16(q256_2, q256_1)); + pixelFilter_q = _mm256_add_epi16(pixelFilter_q, pixetFilter_q2q1q0); + + pixelFilter_p = _mm256_add_epi16(eight, + _mm256_add_epi16(pixelFilter_p, pixelFilter_q)); + + pixetFilter_p2p1p0 = _mm256_add_epi16(four, + _mm256_add_epi16(pixetFilter_p2p1p0, pixetFilter_q2q1q0)); + + res_p = _mm256_srli_epi16( + _mm256_add_epi16(pixelFilter_p, + _mm256_add_epi16(p256_7, p256_0)), 4); + + flat2_p0 = _mm256_castsi256_si128( + _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p), + 168)); + + res_q = _mm256_srli_epi16( + _mm256_add_epi16(pixelFilter_p, + _mm256_add_epi16(q256_7, q256_0)), 4); + + flat2_q0 = _mm256_castsi256_si128( + _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q), + 168)); + + res_p = _mm256_srli_epi16( + _mm256_add_epi16(pixetFilter_p2p1p0, + _mm256_add_epi16(p256_3, p256_0)), 3); + + flat_p0 = _mm256_castsi256_si128( + _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p), + 168)); + + res_q = _mm256_srli_epi16( + _mm256_add_epi16(pixetFilter_p2p1p0, + _mm256_add_epi16(q256_3, q256_0)), 3); + + flat_q0 = _mm256_castsi256_si128( + _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q), + 168)); + + sum_p7 = _mm256_add_epi16(p256_7, p256_7); + + sum_q7 = _mm256_add_epi16(q256_7, q256_7); + + sum_p3 = _mm256_add_epi16(p256_3, p256_3); + + sum_q3 = _mm256_add_epi16(q256_3, q256_3); + + pixelFilter_q = _mm256_sub_epi16(pixelFilter_p, p256_6); + + pixelFilter_p = _mm256_sub_epi16(pixelFilter_p, q256_6); + + res_p = _mm256_srli_epi16( + _mm256_add_epi16(pixelFilter_p, + _mm256_add_epi16(sum_p7, p256_1)), 4); + + flat2_p1 = _mm256_castsi256_si128( + _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p), + 168)); + + res_q = _mm256_srli_epi16( + _mm256_add_epi16(pixelFilter_q, + _mm256_add_epi16(sum_q7, q256_1)), 4); + + flat2_q1 = _mm256_castsi256_si128( + _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q), + 168)); + + pixetFilter_q2q1q0 = _mm256_sub_epi16(pixetFilter_p2p1p0, p256_2); + + pixetFilter_p2p1p0 = _mm256_sub_epi16(pixetFilter_p2p1p0, q256_2); + + res_p = _mm256_srli_epi16( + _mm256_add_epi16(pixetFilter_p2p1p0, + _mm256_add_epi16(sum_p3, p256_1)), 3); + + flat_p1 = _mm256_castsi256_si128( + _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p), + 168)); + + res_q = _mm256_srli_epi16( + _mm256_add_epi16(pixetFilter_q2q1q0, + _mm256_add_epi16(sum_q3, q256_1)), 3); + + flat_q1 = _mm256_castsi256_si128( + _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q), + 168)); + + sum_p7 = _mm256_add_epi16(sum_p7, p256_7); + + sum_q7 = _mm256_add_epi16(sum_q7, q256_7); + + sum_p3 = _mm256_add_epi16(sum_p3, p256_3); + + sum_q3 = _mm256_add_epi16(sum_q3, q256_3); + + pixelFilter_p = _mm256_sub_epi16(pixelFilter_p, q256_5); + + pixelFilter_q = _mm256_sub_epi16(pixelFilter_q, p256_5); + + res_p = _mm256_srli_epi16( + _mm256_add_epi16(pixelFilter_p, + _mm256_add_epi16(sum_p7, p256_2)), 4); + + flat2_p2 = _mm256_castsi256_si128( + _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p), + 168)); + + res_q = _mm256_srli_epi16( + _mm256_add_epi16(pixelFilter_q, + _mm256_add_epi16(sum_q7, q256_2)), 4); + + flat2_q2 = _mm256_castsi256_si128( + _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q), + 168)); + + pixetFilter_p2p1p0 = _mm256_sub_epi16(pixetFilter_p2p1p0, q256_1); + + pixetFilter_q2q1q0 = _mm256_sub_epi16(pixetFilter_q2q1q0, p256_1); + + res_p = _mm256_srli_epi16( + _mm256_add_epi16(pixetFilter_p2p1p0, + _mm256_add_epi16(sum_p3, p256_2)), 3); + + flat_p2 = _mm256_castsi256_si128( + _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p), + 168)); + + res_q = _mm256_srli_epi16( + _mm256_add_epi16(pixetFilter_q2q1q0, + _mm256_add_epi16(sum_q3, q256_2)), 3); + + flat_q2 = _mm256_castsi256_si128( + _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q), + 168)); + + sum_p7 = _mm256_add_epi16(sum_p7, p256_7); + + sum_q7 = _mm256_add_epi16(sum_q7, q256_7); + + pixelFilter_p = _mm256_sub_epi16(pixelFilter_p, q256_4); + + pixelFilter_q = _mm256_sub_epi16(pixelFilter_q, p256_4); + + res_p = _mm256_srli_epi16( + _mm256_add_epi16(pixelFilter_p, + _mm256_add_epi16(sum_p7, p256_3)), 4); + + flat2_p3 = _mm256_castsi256_si128( + _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p), + 168)); + + res_q = _mm256_srli_epi16( + _mm256_add_epi16(pixelFilter_q, + _mm256_add_epi16(sum_q7, q256_3)), 4); + + flat2_q3 = _mm256_castsi256_si128( + _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q), + 168)); + + sum_p7 = _mm256_add_epi16(sum_p7, p256_7); + + sum_q7 = _mm256_add_epi16(sum_q7, q256_7); + + pixelFilter_p = _mm256_sub_epi16(pixelFilter_p, q256_3); + + pixelFilter_q = _mm256_sub_epi16(pixelFilter_q, p256_3); + + res_p = _mm256_srli_epi16( + _mm256_add_epi16(pixelFilter_p, + _mm256_add_epi16(sum_p7, p256_4)), 4); + + flat2_p4 = _mm256_castsi256_si128( + _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p), + 168)); + + res_q = _mm256_srli_epi16( + _mm256_add_epi16(pixelFilter_q, + _mm256_add_epi16(sum_q7, q256_4)), 4); + + flat2_q4 = _mm256_castsi256_si128( + _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q), + 168)); + + sum_p7 = _mm256_add_epi16(sum_p7, p256_7); + + sum_q7 = _mm256_add_epi16(sum_q7, q256_7); + + pixelFilter_p = _mm256_sub_epi16(pixelFilter_p, q256_2); + + pixelFilter_q = _mm256_sub_epi16(pixelFilter_q, p256_2); + + res_p = _mm256_srli_epi16( + _mm256_add_epi16(pixelFilter_p, + _mm256_add_epi16(sum_p7, p256_5)), 4); + + flat2_p5 = _mm256_castsi256_si128( + _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p), + 168)); + + res_q = _mm256_srli_epi16( + _mm256_add_epi16(pixelFilter_q, + _mm256_add_epi16(sum_q7, q256_5)), 4); + + flat2_q5 = _mm256_castsi256_si128( + _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q), + 168)); + + sum_p7 = _mm256_add_epi16(sum_p7, p256_7); + + sum_q7 = _mm256_add_epi16(sum_q7, q256_7); + + pixelFilter_p = _mm256_sub_epi16(pixelFilter_p, q256_1); + + pixelFilter_q = _mm256_sub_epi16(pixelFilter_q, p256_1); + + res_p = _mm256_srli_epi16( + _mm256_add_epi16(pixelFilter_p, + _mm256_add_epi16(sum_p7, p256_6)), 4); + + flat2_p6 = _mm256_castsi256_si128( + _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p), + 168)); + + res_q = _mm256_srli_epi16( + _mm256_add_epi16(pixelFilter_q, + _mm256_add_epi16(sum_q7, q256_6)), 4); + + flat2_q6 = _mm256_castsi256_si128( + _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q), + 168)); + } + + // wide flat + // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + + p2 = _mm_andnot_si128(flat, p2); + flat_p2 = _mm_and_si128(flat, flat_p2); + p2 = _mm_or_si128(flat_p2, p2); + + p1 = _mm_andnot_si128(flat, ps1); + flat_p1 = _mm_and_si128(flat, flat_p1); + p1 = _mm_or_si128(flat_p1, p1); + + p0 = _mm_andnot_si128(flat, ps0); + flat_p0 = _mm_and_si128(flat, flat_p0); + p0 = _mm_or_si128(flat_p0, p0); + + q0 = _mm_andnot_si128(flat, qs0); + flat_q0 = _mm_and_si128(flat, flat_q0); + q0 = _mm_or_si128(flat_q0, q0); + + q1 = _mm_andnot_si128(flat, qs1); + flat_q1 = _mm_and_si128(flat, flat_q1); + q1 = _mm_or_si128(flat_q1, q1); + + q2 = _mm_andnot_si128(flat, q2); + flat_q2 = _mm_and_si128(flat, flat_q2); + q2 = _mm_or_si128(flat_q2, q2); + + p6 = _mm_andnot_si128(flat2, p6); + flat2_p6 = _mm_and_si128(flat2, flat2_p6); + p6 = _mm_or_si128(flat2_p6, p6); + _mm_storeu_si128((__m128i *) (s - 7 * p), p6); + + p5 = _mm_andnot_si128(flat2, p5); + flat2_p5 = _mm_and_si128(flat2, flat2_p5); + p5 = _mm_or_si128(flat2_p5, p5); + _mm_storeu_si128((__m128i *) (s - 6 * p), p5); + + p4 = _mm_andnot_si128(flat2, p4); + flat2_p4 = _mm_and_si128(flat2, flat2_p4); + p4 = _mm_or_si128(flat2_p4, p4); + _mm_storeu_si128((__m128i *) (s - 5 * p), p4); + + p3 = _mm_andnot_si128(flat2, p3); + flat2_p3 = _mm_and_si128(flat2, flat2_p3); + p3 = _mm_or_si128(flat2_p3, p3); + _mm_storeu_si128((__m128i *) (s - 4 * p), p3); + + p2 = _mm_andnot_si128(flat2, p2); + flat2_p2 = _mm_and_si128(flat2, flat2_p2); + p2 = _mm_or_si128(flat2_p2, p2); + _mm_storeu_si128((__m128i *) (s - 3 * p), p2); + + p1 = _mm_andnot_si128(flat2, p1); + flat2_p1 = _mm_and_si128(flat2, flat2_p1); + p1 = _mm_or_si128(flat2_p1, p1); + _mm_storeu_si128((__m128i *) (s - 2 * p), p1); + + p0 = _mm_andnot_si128(flat2, p0); + flat2_p0 = _mm_and_si128(flat2, flat2_p0); + p0 = _mm_or_si128(flat2_p0, p0); + _mm_storeu_si128((__m128i *) (s - 1 * p), p0); + + q0 = _mm_andnot_si128(flat2, q0); + flat2_q0 = _mm_and_si128(flat2, flat2_q0); + q0 = _mm_or_si128(flat2_q0, q0); + _mm_storeu_si128((__m128i *) (s - 0 * p), q0); + + q1 = _mm_andnot_si128(flat2, q1); + flat2_q1 = _mm_and_si128(flat2, flat2_q1); + q1 = _mm_or_si128(flat2_q1, q1); + _mm_storeu_si128((__m128i *) (s + 1 * p), q1); + + q2 = _mm_andnot_si128(flat2, q2); + flat2_q2 = _mm_and_si128(flat2, flat2_q2); + q2 = _mm_or_si128(flat2_q2, q2); + _mm_storeu_si128((__m128i *) (s + 2 * p), q2); + + q3 = _mm_andnot_si128(flat2, q3); + flat2_q3 = _mm_and_si128(flat2, flat2_q3); + q3 = _mm_or_si128(flat2_q3, q3); + _mm_storeu_si128((__m128i *) (s + 3 * p), q3); + + q4 = _mm_andnot_si128(flat2, q4); + flat2_q4 = _mm_and_si128(flat2, flat2_q4); + q4 = _mm_or_si128(flat2_q4, q4); + _mm_storeu_si128((__m128i *) (s + 4 * p), q4); + + q5 = _mm_andnot_si128(flat2, q5); + flat2_q5 = _mm_and_si128(flat2, flat2_q5); + q5 = _mm_or_si128(flat2_q5, q5); + _mm_storeu_si128((__m128i *) (s + 5 * p), q5); + + q6 = _mm_andnot_si128(flat2, q6); + flat2_q6 = _mm_and_si128(flat2, flat2_q6); + q6 = _mm_or_si128(flat2_q6, q6); + _mm_storeu_si128((__m128i *) (s + 6 * p), q6); + } +} diff --git a/thirdparty/libvpx/vpx_dsp/x86/loopfilter_sse2.c b/thirdparty/libvpx/vpx_dsp/x86/loopfilter_sse2.c new file mode 100644 index 0000000000..739adf31d0 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/x86/loopfilter_sse2.c @@ -0,0 +1,1776 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <emmintrin.h> // SSE2 + +#include "./vpx_dsp_rtcd.h" +#include "vpx_ports/mem.h" +#include "vpx_ports/emmintrin_compat.h" + +static INLINE __m128i abs_diff(__m128i a, __m128i b) { + return _mm_or_si128(_mm_subs_epu8(a, b), _mm_subs_epu8(b, a)); +} + +// filter_mask and hev_mask +#define FILTER_HEV_MASK do { \ + /* (abs(q1 - q0), abs(p1 - p0) */ \ + __m128i flat = abs_diff(q1p1, q0p0); \ + /* abs(p1 - q1), abs(p0 - q0) */ \ + const __m128i abs_p1q1p0q0 = abs_diff(p1p0, q1q0); \ + __m128i abs_p0q0, abs_p1q1, work; \ + \ + /* const uint8_t hev = hev_mask(thresh, *op1, *op0, *oq0, *oq1); */ \ + hev = _mm_unpacklo_epi8(_mm_max_epu8(flat, _mm_srli_si128(flat, 8)), zero); \ + hev = _mm_cmpgt_epi16(hev, thresh); \ + hev = _mm_packs_epi16(hev, hev); \ + \ + /* const int8_t mask = filter_mask(*limit, *blimit, */ \ + /* p3, p2, p1, p0, q0, q1, q2, q3); */ \ + abs_p0q0 = _mm_adds_epu8(abs_p1q1p0q0, abs_p1q1p0q0); /* abs(p0 - q0) * 2 */\ + abs_p1q1 = _mm_unpackhi_epi8(abs_p1q1p0q0, abs_p1q1p0q0); /* abs(p1 - q1) */\ + abs_p1q1 = _mm_srli_epi16(abs_p1q1, 9); \ + abs_p1q1 = _mm_packs_epi16(abs_p1q1, abs_p1q1); /* abs(p1 - q1) / 2 */ \ + /* abs(p0 - q0) * 2 + abs(p1 - q1) / 2 */ \ + mask = _mm_adds_epu8(abs_p0q0, abs_p1q1); \ + /* abs(p3 - p2), abs(p2 - p1) */ \ + work = abs_diff(p3p2, p2p1); \ + flat = _mm_max_epu8(work, flat); \ + /* abs(q3 - q2), abs(q2 - q1) */ \ + work = abs_diff(q3q2, q2q1); \ + flat = _mm_max_epu8(work, flat); \ + flat = _mm_max_epu8(flat, _mm_srli_si128(flat, 8)); \ + mask = _mm_unpacklo_epi64(mask, flat); \ + mask = _mm_subs_epu8(mask, limit); \ + mask = _mm_cmpeq_epi8(mask, zero); \ + mask = _mm_and_si128(mask, _mm_srli_si128(mask, 8)); \ +} while (0) + +#define FILTER4 do { \ + const __m128i t3t4 = _mm_set_epi8(3, 3, 3, 3, 3, 3, 3, 3, \ + 4, 4, 4, 4, 4, 4, 4, 4); \ + const __m128i t80 = _mm_set1_epi8(0x80); \ + __m128i filter, filter2filter1, work; \ + \ + ps1ps0 = _mm_xor_si128(p1p0, t80); /* ^ 0x80 */ \ + qs1qs0 = _mm_xor_si128(q1q0, t80); \ + \ + /* int8_t filter = signed_char_clamp(ps1 - qs1) & hev; */ \ + work = _mm_subs_epi8(ps1ps0, qs1qs0); \ + filter = _mm_and_si128(_mm_srli_si128(work, 8), hev); \ + /* filter = signed_char_clamp(filter + 3 * (qs0 - ps0)) & mask; */ \ + filter = _mm_subs_epi8(filter, work); \ + filter = _mm_subs_epi8(filter, work); \ + filter = _mm_subs_epi8(filter, work); /* + 3 * (qs0 - ps0) */ \ + filter = _mm_and_si128(filter, mask); /* & mask */ \ + filter = _mm_unpacklo_epi64(filter, filter); \ + \ + /* filter1 = signed_char_clamp(filter + 4) >> 3; */ \ + /* filter2 = signed_char_clamp(filter + 3) >> 3; */ \ + filter2filter1 = _mm_adds_epi8(filter, t3t4); /* signed_char_clamp */ \ + filter = _mm_unpackhi_epi8(filter2filter1, filter2filter1); \ + filter2filter1 = _mm_unpacklo_epi8(filter2filter1, filter2filter1); \ + filter2filter1 = _mm_srai_epi16(filter2filter1, 11); /* >> 3 */ \ + filter = _mm_srai_epi16(filter, 11); /* >> 3 */ \ + filter2filter1 = _mm_packs_epi16(filter2filter1, filter); \ + \ + /* filter = ROUND_POWER_OF_TWO(filter1, 1) & ~hev; */ \ + filter = _mm_subs_epi8(filter2filter1, ff); /* + 1 */ \ + filter = _mm_unpacklo_epi8(filter, filter); \ + filter = _mm_srai_epi16(filter, 9); /* round */ \ + filter = _mm_packs_epi16(filter, filter); \ + filter = _mm_andnot_si128(hev, filter); \ + \ + hev = _mm_unpackhi_epi64(filter2filter1, filter); \ + filter2filter1 = _mm_unpacklo_epi64(filter2filter1, filter); \ + \ + /* signed_char_clamp(qs1 - filter), signed_char_clamp(qs0 - filter1) */ \ + qs1qs0 = _mm_subs_epi8(qs1qs0, filter2filter1); \ + /* signed_char_clamp(ps1 + filter), signed_char_clamp(ps0 + filter2) */ \ + ps1ps0 = _mm_adds_epi8(ps1ps0, hev); \ + qs1qs0 = _mm_xor_si128(qs1qs0, t80); /* ^ 0x80 */ \ + ps1ps0 = _mm_xor_si128(ps1ps0, t80); /* ^ 0x80 */ \ +} while (0) + +void vpx_lpf_horizontal_4_sse2(uint8_t *s, int p /* pitch */, + const uint8_t *_blimit, const uint8_t *_limit, + const uint8_t *_thresh) { + const __m128i zero = _mm_set1_epi16(0); + const __m128i limit = + _mm_unpacklo_epi64(_mm_loadl_epi64((const __m128i *)_blimit), + _mm_loadl_epi64((const __m128i *)_limit)); + const __m128i thresh = + _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)_thresh), zero); + const __m128i ff = _mm_cmpeq_epi8(zero, zero); + __m128i q1p1, q0p0, p3p2, p2p1, p1p0, q3q2, q2q1, q1q0, ps1ps0, qs1qs0; + __m128i mask, hev; + + p3p2 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(s - 3 * p)), + _mm_loadl_epi64((__m128i *)(s - 4 * p))); + q1p1 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(s - 2 * p)), + _mm_loadl_epi64((__m128i *)(s + 1 * p))); + q0p0 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(s - 1 * p)), + _mm_loadl_epi64((__m128i *)(s + 0 * p))); + q3q2 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(s + 2 * p)), + _mm_loadl_epi64((__m128i *)(s + 3 * p))); + p1p0 = _mm_unpacklo_epi64(q0p0, q1p1); + p2p1 = _mm_unpacklo_epi64(q1p1, p3p2); + q1q0 = _mm_unpackhi_epi64(q0p0, q1p1); + q2q1 = _mm_unpacklo_epi64(_mm_srli_si128(q1p1, 8), q3q2); + + FILTER_HEV_MASK; + FILTER4; + + _mm_storeh_pi((__m64 *)(s - 2 * p), _mm_castsi128_ps(ps1ps0)); // *op1 + _mm_storel_epi64((__m128i *)(s - 1 * p), ps1ps0); // *op0 + _mm_storel_epi64((__m128i *)(s + 0 * p), qs1qs0); // *oq0 + _mm_storeh_pi((__m64 *)(s + 1 * p), _mm_castsi128_ps(qs1qs0)); // *oq1 +} + +void vpx_lpf_vertical_4_sse2(uint8_t *s, int p /* pitch */, + const uint8_t *_blimit, const uint8_t *_limit, + const uint8_t *_thresh) { + const __m128i zero = _mm_set1_epi16(0); + const __m128i limit = + _mm_unpacklo_epi64(_mm_loadl_epi64((const __m128i *)_blimit), + _mm_loadl_epi64((const __m128i *)_limit)); + const __m128i thresh = + _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)_thresh), zero); + const __m128i ff = _mm_cmpeq_epi8(zero, zero); + __m128i x0, x1, x2, x3; + __m128i q1p1, q0p0, p3p2, p2p1, p1p0, q3q2, q2q1, q1q0, ps1ps0, qs1qs0; + __m128i mask, hev; + + // 00 10 01 11 02 12 03 13 04 14 05 15 06 16 07 17 + q1q0 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(s + 0 * p - 4)), + _mm_loadl_epi64((__m128i *)(s + 1 * p - 4))); + + // 20 30 21 31 22 32 23 33 24 34 25 35 26 36 27 37 + x1 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(s + 2 * p - 4)), + _mm_loadl_epi64((__m128i *)(s + 3 * p - 4))); + + // 40 50 41 51 42 52 43 53 44 54 45 55 46 56 47 57 + x2 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(s + 4 * p - 4)), + _mm_loadl_epi64((__m128i *)(s + 5 * p - 4))); + + // 60 70 61 71 62 72 63 73 64 74 65 75 66 76 67 77 + x3 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(s + 6 * p - 4)), + _mm_loadl_epi64((__m128i *)(s + 7 * p - 4))); + + // Transpose 8x8 + // 00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33 + p1p0 = _mm_unpacklo_epi16(q1q0, x1); + // 40 50 60 70 41 51 61 71 42 52 62 72 43 53 63 73 + x0 = _mm_unpacklo_epi16(x2, x3); + // 00 10 20 30 40 50 60 70 01 11 21 31 41 51 61 71 + p3p2 = _mm_unpacklo_epi32(p1p0, x0); + // 02 12 22 32 42 52 62 72 03 13 23 33 43 53 63 73 + p1p0 = _mm_unpackhi_epi32(p1p0, x0); + p3p2 = _mm_unpackhi_epi64(p3p2, _mm_slli_si128(p3p2, 8)); // swap lo and high + p1p0 = _mm_unpackhi_epi64(p1p0, _mm_slli_si128(p1p0, 8)); // swap lo and high + + // 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37 + q1q0 = _mm_unpackhi_epi16(q1q0, x1); + // 44 54 64 74 45 55 65 75 46 56 66 76 47 57 67 77 + x2 = _mm_unpackhi_epi16(x2, x3); + // 06 16 26 36 46 56 66 76 07 17 27 37 47 57 67 77 + q3q2 = _mm_unpackhi_epi32(q1q0, x2); + // 04 14 24 34 44 54 64 74 05 15 25 35 45 55 65 75 + q1q0 = _mm_unpacklo_epi32(q1q0, x2); + + q0p0 = _mm_unpacklo_epi64(p1p0, q1q0); + q1p1 = _mm_unpackhi_epi64(p1p0, q1q0); + p1p0 = _mm_unpacklo_epi64(q0p0, q1p1); + p2p1 = _mm_unpacklo_epi64(q1p1, p3p2); + q2q1 = _mm_unpacklo_epi64(_mm_srli_si128(q1p1, 8), q3q2); + + FILTER_HEV_MASK; + FILTER4; + + // Transpose 8x4 to 4x8 + // qs1qs0: 20 21 22 23 24 25 26 27 30 31 32 33 34 34 36 37 + // ps1ps0: 10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07 + // 00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17 + ps1ps0 = _mm_unpackhi_epi64(ps1ps0, _mm_slli_si128(ps1ps0, 8)); + // 10 30 11 31 12 32 13 33 14 34 15 35 16 36 17 37 + x0 = _mm_unpackhi_epi8(ps1ps0, qs1qs0); + // 00 20 01 21 02 22 03 23 04 24 05 25 06 26 07 27 + ps1ps0 = _mm_unpacklo_epi8(ps1ps0, qs1qs0); + // 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37 + qs1qs0 = _mm_unpackhi_epi8(ps1ps0, x0); + // 00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33 + ps1ps0 = _mm_unpacklo_epi8(ps1ps0, x0); + + *(int *)(s + 0 * p - 2) = _mm_cvtsi128_si32(ps1ps0); + ps1ps0 = _mm_srli_si128(ps1ps0, 4); + *(int *)(s + 1 * p - 2) = _mm_cvtsi128_si32(ps1ps0); + ps1ps0 = _mm_srli_si128(ps1ps0, 4); + *(int *)(s + 2 * p - 2) = _mm_cvtsi128_si32(ps1ps0); + ps1ps0 = _mm_srli_si128(ps1ps0, 4); + *(int *)(s + 3 * p - 2) = _mm_cvtsi128_si32(ps1ps0); + + *(int *)(s + 4 * p - 2) = _mm_cvtsi128_si32(qs1qs0); + qs1qs0 = _mm_srli_si128(qs1qs0, 4); + *(int *)(s + 5 * p - 2) = _mm_cvtsi128_si32(qs1qs0); + qs1qs0 = _mm_srli_si128(qs1qs0, 4); + *(int *)(s + 6 * p - 2) = _mm_cvtsi128_si32(qs1qs0); + qs1qs0 = _mm_srli_si128(qs1qs0, 4); + *(int *)(s + 7 * p - 2) = _mm_cvtsi128_si32(qs1qs0); +} + +void vpx_lpf_horizontal_edge_8_sse2(unsigned char *s, int p, + const unsigned char *_blimit, + const unsigned char *_limit, + const unsigned char *_thresh) { + const __m128i zero = _mm_set1_epi16(0); + const __m128i one = _mm_set1_epi8(1); + const __m128i blimit = _mm_load_si128((const __m128i *)_blimit); + const __m128i limit = _mm_load_si128((const __m128i *)_limit); + const __m128i thresh = _mm_load_si128((const __m128i *)_thresh); + __m128i mask, hev, flat, flat2; + __m128i q7p7, q6p6, q5p5, q4p4, q3p3, q2p2, q1p1, q0p0, p0q0, p1q1; + __m128i abs_p1p0; + + q4p4 = _mm_loadl_epi64((__m128i *)(s - 5 * p)); + q4p4 = _mm_castps_si128(_mm_loadh_pi(_mm_castsi128_ps(q4p4), + (__m64 *)(s + 4 * p))); + q3p3 = _mm_loadl_epi64((__m128i *)(s - 4 * p)); + q3p3 = _mm_castps_si128(_mm_loadh_pi(_mm_castsi128_ps(q3p3), + (__m64 *)(s + 3 * p))); + q2p2 = _mm_loadl_epi64((__m128i *)(s - 3 * p)); + q2p2 = _mm_castps_si128(_mm_loadh_pi(_mm_castsi128_ps(q2p2), + (__m64 *)(s + 2 * p))); + q1p1 = _mm_loadl_epi64((__m128i *)(s - 2 * p)); + q1p1 = _mm_castps_si128(_mm_loadh_pi(_mm_castsi128_ps(q1p1), + (__m64 *)(s + 1 * p))); + p1q1 = _mm_shuffle_epi32(q1p1, 78); + q0p0 = _mm_loadl_epi64((__m128i *)(s - 1 * p)); + q0p0 = _mm_castps_si128(_mm_loadh_pi(_mm_castsi128_ps(q0p0), + (__m64 *)(s - 0 * p))); + p0q0 = _mm_shuffle_epi32(q0p0, 78); + + { + __m128i abs_p1q1, abs_p0q0, abs_q1q0, fe, ff, work; + abs_p1p0 = abs_diff(q1p1, q0p0); + abs_q1q0 = _mm_srli_si128(abs_p1p0, 8); + fe = _mm_set1_epi8(0xfe); + ff = _mm_cmpeq_epi8(abs_p1p0, abs_p1p0); + abs_p0q0 = abs_diff(q0p0, p0q0); + abs_p1q1 = abs_diff(q1p1, p1q1); + flat = _mm_max_epu8(abs_p1p0, abs_q1q0); + hev = _mm_subs_epu8(flat, thresh); + hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff); + + abs_p0q0 =_mm_adds_epu8(abs_p0q0, abs_p0q0); + abs_p1q1 = _mm_srli_epi16(_mm_and_si128(abs_p1q1, fe), 1); + mask = _mm_subs_epu8(_mm_adds_epu8(abs_p0q0, abs_p1q1), blimit); + mask = _mm_xor_si128(_mm_cmpeq_epi8(mask, zero), ff); + // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2 > blimit) * -1; + mask = _mm_max_epu8(abs_p1p0, mask); + // mask |= (abs(p1 - p0) > limit) * -1; + // mask |= (abs(q1 - q0) > limit) * -1; + + work = _mm_max_epu8(abs_diff(q2p2, q1p1), + abs_diff(q3p3, q2p2)); + mask = _mm_max_epu8(work, mask); + mask = _mm_max_epu8(mask, _mm_srli_si128(mask, 8)); + mask = _mm_subs_epu8(mask, limit); + mask = _mm_cmpeq_epi8(mask, zero); + } + + // lp filter + { + const __m128i t4 = _mm_set1_epi8(4); + const __m128i t3 = _mm_set1_epi8(3); + const __m128i t80 = _mm_set1_epi8(0x80); + const __m128i t1 = _mm_set1_epi16(0x1); + __m128i qs1ps1 = _mm_xor_si128(q1p1, t80); + __m128i qs0ps0 = _mm_xor_si128(q0p0, t80); + __m128i qs0 = _mm_xor_si128(p0q0, t80); + __m128i qs1 = _mm_xor_si128(p1q1, t80); + __m128i filt; + __m128i work_a; + __m128i filter1, filter2; + __m128i flat2_q6p6, flat2_q5p5, flat2_q4p4, flat2_q3p3, flat2_q2p2; + __m128i flat2_q1p1, flat2_q0p0, flat_q2p2, flat_q1p1, flat_q0p0; + + filt = _mm_and_si128(_mm_subs_epi8(qs1ps1, qs1), hev); + work_a = _mm_subs_epi8(qs0, qs0ps0); + filt = _mm_adds_epi8(filt, work_a); + filt = _mm_adds_epi8(filt, work_a); + filt = _mm_adds_epi8(filt, work_a); + // (vpx_filter + 3 * (qs0 - ps0)) & mask + filt = _mm_and_si128(filt, mask); + + filter1 = _mm_adds_epi8(filt, t4); + filter2 = _mm_adds_epi8(filt, t3); + + filter1 = _mm_unpacklo_epi8(zero, filter1); + filter1 = _mm_srai_epi16(filter1, 0xB); + filter2 = _mm_unpacklo_epi8(zero, filter2); + filter2 = _mm_srai_epi16(filter2, 0xB); + + // Filter1 >> 3 + filt = _mm_packs_epi16(filter2, _mm_subs_epi16(zero, filter1)); + qs0ps0 = _mm_xor_si128(_mm_adds_epi8(qs0ps0, filt), t80); + + // filt >> 1 + filt = _mm_adds_epi16(filter1, t1); + filt = _mm_srai_epi16(filt, 1); + filt = _mm_andnot_si128(_mm_srai_epi16(_mm_unpacklo_epi8(zero, hev), 0x8), + filt); + filt = _mm_packs_epi16(filt, _mm_subs_epi16(zero, filt)); + qs1ps1 = _mm_xor_si128(_mm_adds_epi8(qs1ps1, filt), t80); + // loopfilter done + + { + __m128i work; + flat = _mm_max_epu8(abs_diff(q2p2, q0p0), abs_diff(q3p3, q0p0)); + flat = _mm_max_epu8(abs_p1p0, flat); + flat = _mm_max_epu8(flat, _mm_srli_si128(flat, 8)); + flat = _mm_subs_epu8(flat, one); + flat = _mm_cmpeq_epi8(flat, zero); + flat = _mm_and_si128(flat, mask); + + q5p5 = _mm_loadl_epi64((__m128i *)(s - 6 * p)); + q5p5 = _mm_castps_si128(_mm_loadh_pi(_mm_castsi128_ps(q5p5), + (__m64 *)(s + 5 * p))); + + q6p6 = _mm_loadl_epi64((__m128i *)(s - 7 * p)); + q6p6 = _mm_castps_si128(_mm_loadh_pi(_mm_castsi128_ps(q6p6), + (__m64 *)(s + 6 * p))); + flat2 = _mm_max_epu8(abs_diff(q4p4, q0p0), abs_diff(q5p5, q0p0)); + + q7p7 = _mm_loadl_epi64((__m128i *)(s - 8 * p)); + q7p7 = _mm_castps_si128(_mm_loadh_pi(_mm_castsi128_ps(q7p7), + (__m64 *)(s + 7 * p))); + work = _mm_max_epu8(abs_diff(q6p6, q0p0), abs_diff(q7p7, q0p0)); + flat2 = _mm_max_epu8(work, flat2); + flat2 = _mm_max_epu8(flat2, _mm_srli_si128(flat2, 8)); + flat2 = _mm_subs_epu8(flat2, one); + flat2 = _mm_cmpeq_epi8(flat2, zero); + flat2 = _mm_and_si128(flat2, flat); // flat2 & flat & mask + } + + // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + // flat and wide flat calculations + { + const __m128i eight = _mm_set1_epi16(8); + const __m128i four = _mm_set1_epi16(4); + __m128i p7_16, p6_16, p5_16, p4_16, p3_16, p2_16, p1_16, p0_16; + __m128i q7_16, q6_16, q5_16, q4_16, q3_16, q2_16, q1_16, q0_16; + __m128i pixelFilter_p, pixelFilter_q; + __m128i pixetFilter_p2p1p0, pixetFilter_q2q1q0; + __m128i sum_p7, sum_q7, sum_p3, sum_q3, res_p, res_q; + + p7_16 = _mm_unpacklo_epi8(q7p7, zero);; + p6_16 = _mm_unpacklo_epi8(q6p6, zero); + p5_16 = _mm_unpacklo_epi8(q5p5, zero); + p4_16 = _mm_unpacklo_epi8(q4p4, zero); + p3_16 = _mm_unpacklo_epi8(q3p3, zero); + p2_16 = _mm_unpacklo_epi8(q2p2, zero); + p1_16 = _mm_unpacklo_epi8(q1p1, zero); + p0_16 = _mm_unpacklo_epi8(q0p0, zero); + q0_16 = _mm_unpackhi_epi8(q0p0, zero); + q1_16 = _mm_unpackhi_epi8(q1p1, zero); + q2_16 = _mm_unpackhi_epi8(q2p2, zero); + q3_16 = _mm_unpackhi_epi8(q3p3, zero); + q4_16 = _mm_unpackhi_epi8(q4p4, zero); + q5_16 = _mm_unpackhi_epi8(q5p5, zero); + q6_16 = _mm_unpackhi_epi8(q6p6, zero); + q7_16 = _mm_unpackhi_epi8(q7p7, zero); + + pixelFilter_p = _mm_add_epi16(_mm_add_epi16(p6_16, p5_16), + _mm_add_epi16(p4_16, p3_16)); + pixelFilter_q = _mm_add_epi16(_mm_add_epi16(q6_16, q5_16), + _mm_add_epi16(q4_16, q3_16)); + + pixetFilter_p2p1p0 = _mm_add_epi16(p0_16, _mm_add_epi16(p2_16, p1_16)); + pixelFilter_p = _mm_add_epi16(pixelFilter_p, pixetFilter_p2p1p0); + + pixetFilter_q2q1q0 = _mm_add_epi16(q0_16, _mm_add_epi16(q2_16, q1_16)); + pixelFilter_q = _mm_add_epi16(pixelFilter_q, pixetFilter_q2q1q0); + pixelFilter_p = _mm_add_epi16(eight, _mm_add_epi16(pixelFilter_p, + pixelFilter_q)); + pixetFilter_p2p1p0 = _mm_add_epi16(four, + _mm_add_epi16(pixetFilter_p2p1p0, + pixetFilter_q2q1q0)); + res_p = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p, + _mm_add_epi16(p7_16, p0_16)), 4); + res_q = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p, + _mm_add_epi16(q7_16, q0_16)), 4); + flat2_q0p0 = _mm_packus_epi16(res_p, res_q); + res_p = _mm_srli_epi16(_mm_add_epi16(pixetFilter_p2p1p0, + _mm_add_epi16(p3_16, p0_16)), 3); + res_q = _mm_srli_epi16(_mm_add_epi16(pixetFilter_p2p1p0, + _mm_add_epi16(q3_16, q0_16)), 3); + + flat_q0p0 = _mm_packus_epi16(res_p, res_q); + + sum_p7 = _mm_add_epi16(p7_16, p7_16); + sum_q7 = _mm_add_epi16(q7_16, q7_16); + sum_p3 = _mm_add_epi16(p3_16, p3_16); + sum_q3 = _mm_add_epi16(q3_16, q3_16); + + pixelFilter_q = _mm_sub_epi16(pixelFilter_p, p6_16); + pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q6_16); + res_p = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p, + _mm_add_epi16(sum_p7, p1_16)), 4); + res_q = _mm_srli_epi16(_mm_add_epi16(pixelFilter_q, + _mm_add_epi16(sum_q7, q1_16)), 4); + flat2_q1p1 = _mm_packus_epi16(res_p, res_q); + + pixetFilter_q2q1q0 = _mm_sub_epi16(pixetFilter_p2p1p0, p2_16); + pixetFilter_p2p1p0 = _mm_sub_epi16(pixetFilter_p2p1p0, q2_16); + res_p = _mm_srli_epi16(_mm_add_epi16(pixetFilter_p2p1p0, + _mm_add_epi16(sum_p3, p1_16)), 3); + res_q = _mm_srli_epi16(_mm_add_epi16(pixetFilter_q2q1q0, + _mm_add_epi16(sum_q3, q1_16)), 3); + flat_q1p1 = _mm_packus_epi16(res_p, res_q); + + sum_p7 = _mm_add_epi16(sum_p7, p7_16); + sum_q7 = _mm_add_epi16(sum_q7, q7_16); + sum_p3 = _mm_add_epi16(sum_p3, p3_16); + sum_q3 = _mm_add_epi16(sum_q3, q3_16); + + pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q5_16); + pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p5_16); + res_p = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p, + _mm_add_epi16(sum_p7, p2_16)), 4); + res_q = _mm_srli_epi16(_mm_add_epi16(pixelFilter_q, + _mm_add_epi16(sum_q7, q2_16)), 4); + flat2_q2p2 = _mm_packus_epi16(res_p, res_q); + + pixetFilter_p2p1p0 = _mm_sub_epi16(pixetFilter_p2p1p0, q1_16); + pixetFilter_q2q1q0 = _mm_sub_epi16(pixetFilter_q2q1q0, p1_16); + + res_p = _mm_srli_epi16(_mm_add_epi16(pixetFilter_p2p1p0, + _mm_add_epi16(sum_p3, p2_16)), 3); + res_q = _mm_srli_epi16(_mm_add_epi16(pixetFilter_q2q1q0, + _mm_add_epi16(sum_q3, q2_16)), 3); + flat_q2p2 = _mm_packus_epi16(res_p, res_q); + + sum_p7 = _mm_add_epi16(sum_p7, p7_16); + sum_q7 = _mm_add_epi16(sum_q7, q7_16); + pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q4_16); + pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p4_16); + res_p = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p, + _mm_add_epi16(sum_p7, p3_16)), 4); + res_q = _mm_srli_epi16(_mm_add_epi16(pixelFilter_q, + _mm_add_epi16(sum_q7, q3_16)), 4); + flat2_q3p3 = _mm_packus_epi16(res_p, res_q); + + sum_p7 = _mm_add_epi16(sum_p7, p7_16); + sum_q7 = _mm_add_epi16(sum_q7, q7_16); + pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q3_16); + pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p3_16); + res_p = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p, + _mm_add_epi16(sum_p7, p4_16)), 4); + res_q = _mm_srli_epi16(_mm_add_epi16(pixelFilter_q, + _mm_add_epi16(sum_q7, q4_16)), 4); + flat2_q4p4 = _mm_packus_epi16(res_p, res_q); + + sum_p7 = _mm_add_epi16(sum_p7, p7_16); + sum_q7 = _mm_add_epi16(sum_q7, q7_16); + pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q2_16); + pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p2_16); + res_p = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p, + _mm_add_epi16(sum_p7, p5_16)), 4); + res_q = _mm_srli_epi16(_mm_add_epi16(pixelFilter_q, + _mm_add_epi16(sum_q7, q5_16)), 4); + flat2_q5p5 = _mm_packus_epi16(res_p, res_q); + + sum_p7 = _mm_add_epi16(sum_p7, p7_16); + sum_q7 = _mm_add_epi16(sum_q7, q7_16); + pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q1_16); + pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p1_16); + res_p = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p, + _mm_add_epi16(sum_p7, p6_16)), 4); + res_q = _mm_srli_epi16(_mm_add_epi16(pixelFilter_q, + _mm_add_epi16(sum_q7, q6_16)), 4); + flat2_q6p6 = _mm_packus_epi16(res_p, res_q); + } + // wide flat + // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + + flat = _mm_shuffle_epi32(flat, 68); + flat2 = _mm_shuffle_epi32(flat2, 68); + + q2p2 = _mm_andnot_si128(flat, q2p2); + flat_q2p2 = _mm_and_si128(flat, flat_q2p2); + q2p2 = _mm_or_si128(q2p2, flat_q2p2); + + qs1ps1 = _mm_andnot_si128(flat, qs1ps1); + flat_q1p1 = _mm_and_si128(flat, flat_q1p1); + q1p1 = _mm_or_si128(qs1ps1, flat_q1p1); + + qs0ps0 = _mm_andnot_si128(flat, qs0ps0); + flat_q0p0 = _mm_and_si128(flat, flat_q0p0); + q0p0 = _mm_or_si128(qs0ps0, flat_q0p0); + + q6p6 = _mm_andnot_si128(flat2, q6p6); + flat2_q6p6 = _mm_and_si128(flat2, flat2_q6p6); + q6p6 = _mm_or_si128(q6p6, flat2_q6p6); + _mm_storel_epi64((__m128i *)(s - 7 * p), q6p6); + _mm_storeh_pi((__m64 *)(s + 6 * p), _mm_castsi128_ps(q6p6)); + + q5p5 = _mm_andnot_si128(flat2, q5p5); + flat2_q5p5 = _mm_and_si128(flat2, flat2_q5p5); + q5p5 = _mm_or_si128(q5p5, flat2_q5p5); + _mm_storel_epi64((__m128i *)(s - 6 * p), q5p5); + _mm_storeh_pi((__m64 *)(s + 5 * p), _mm_castsi128_ps(q5p5)); + + q4p4 = _mm_andnot_si128(flat2, q4p4); + flat2_q4p4 = _mm_and_si128(flat2, flat2_q4p4); + q4p4 = _mm_or_si128(q4p4, flat2_q4p4); + _mm_storel_epi64((__m128i *)(s - 5 * p), q4p4); + _mm_storeh_pi((__m64 *)(s + 4 * p), _mm_castsi128_ps(q4p4)); + + q3p3 = _mm_andnot_si128(flat2, q3p3); + flat2_q3p3 = _mm_and_si128(flat2, flat2_q3p3); + q3p3 = _mm_or_si128(q3p3, flat2_q3p3); + _mm_storel_epi64((__m128i *)(s - 4 * p), q3p3); + _mm_storeh_pi((__m64 *)(s + 3 * p), _mm_castsi128_ps(q3p3)); + + q2p2 = _mm_andnot_si128(flat2, q2p2); + flat2_q2p2 = _mm_and_si128(flat2, flat2_q2p2); + q2p2 = _mm_or_si128(q2p2, flat2_q2p2); + _mm_storel_epi64((__m128i *)(s - 3 * p), q2p2); + _mm_storeh_pi((__m64 *)(s + 2 * p), _mm_castsi128_ps(q2p2)); + + q1p1 = _mm_andnot_si128(flat2, q1p1); + flat2_q1p1 = _mm_and_si128(flat2, flat2_q1p1); + q1p1 = _mm_or_si128(q1p1, flat2_q1p1); + _mm_storel_epi64((__m128i *)(s - 2 * p), q1p1); + _mm_storeh_pi((__m64 *)(s + 1 * p), _mm_castsi128_ps(q1p1)); + + q0p0 = _mm_andnot_si128(flat2, q0p0); + flat2_q0p0 = _mm_and_si128(flat2, flat2_q0p0); + q0p0 = _mm_or_si128(q0p0, flat2_q0p0); + _mm_storel_epi64((__m128i *)(s - 1 * p), q0p0); + _mm_storeh_pi((__m64 *)(s - 0 * p), _mm_castsi128_ps(q0p0)); + } +} + +static INLINE __m128i filter_add2_sub2(const __m128i *const total, + const __m128i *const a1, + const __m128i *const a2, + const __m128i *const s1, + const __m128i *const s2) { + __m128i x = _mm_add_epi16(*a1, *total); + x = _mm_add_epi16(_mm_sub_epi16(x, _mm_add_epi16(*s1, *s2)), *a2); + return x; +} + +static INLINE __m128i filter8_mask(const __m128i *const flat, + const __m128i *const other_filt, + const __m128i *const f8_lo, + const __m128i *const f8_hi) { + const __m128i f8 = _mm_packus_epi16(_mm_srli_epi16(*f8_lo, 3), + _mm_srli_epi16(*f8_hi, 3)); + const __m128i result = _mm_and_si128(*flat, f8); + return _mm_or_si128(_mm_andnot_si128(*flat, *other_filt), result); +} + +static INLINE __m128i filter16_mask(const __m128i *const flat, + const __m128i *const other_filt, + const __m128i *const f_lo, + const __m128i *const f_hi) { + const __m128i f = _mm_packus_epi16(_mm_srli_epi16(*f_lo, 4), + _mm_srli_epi16(*f_hi, 4)); + const __m128i result = _mm_and_si128(*flat, f); + return _mm_or_si128(_mm_andnot_si128(*flat, *other_filt), result); +} + +void vpx_lpf_horizontal_edge_16_sse2(unsigned char *s, int p, + const unsigned char *_blimit, + const unsigned char *_limit, + const unsigned char *_thresh) { + const __m128i zero = _mm_set1_epi16(0); + const __m128i one = _mm_set1_epi8(1); + const __m128i blimit = _mm_load_si128((const __m128i *)_blimit); + const __m128i limit = _mm_load_si128((const __m128i *)_limit); + const __m128i thresh = _mm_load_si128((const __m128i *)_thresh); + __m128i mask, hev, flat, flat2; + __m128i p7, p6, p5; + __m128i p4, p3, p2, p1, p0, q0, q1, q2, q3, q4; + __m128i q5, q6, q7; + + __m128i op2, op1, op0, oq0, oq1, oq2; + + __m128i max_abs_p1p0q1q0; + + p7 = _mm_loadu_si128((__m128i *)(s - 8 * p)); + p6 = _mm_loadu_si128((__m128i *)(s - 7 * p)); + p5 = _mm_loadu_si128((__m128i *)(s - 6 * p)); + p4 = _mm_loadu_si128((__m128i *)(s - 5 * p)); + p3 = _mm_loadu_si128((__m128i *)(s - 4 * p)); + p2 = _mm_loadu_si128((__m128i *)(s - 3 * p)); + p1 = _mm_loadu_si128((__m128i *)(s - 2 * p)); + p0 = _mm_loadu_si128((__m128i *)(s - 1 * p)); + q0 = _mm_loadu_si128((__m128i *)(s - 0 * p)); + q1 = _mm_loadu_si128((__m128i *)(s + 1 * p)); + q2 = _mm_loadu_si128((__m128i *)(s + 2 * p)); + q3 = _mm_loadu_si128((__m128i *)(s + 3 * p)); + q4 = _mm_loadu_si128((__m128i *)(s + 4 * p)); + q5 = _mm_loadu_si128((__m128i *)(s + 5 * p)); + q6 = _mm_loadu_si128((__m128i *)(s + 6 * p)); + q7 = _mm_loadu_si128((__m128i *)(s + 7 * p)); + + { + const __m128i abs_p1p0 = abs_diff(p1, p0); + const __m128i abs_q1q0 = abs_diff(q1, q0); + const __m128i fe = _mm_set1_epi8(0xfe); + const __m128i ff = _mm_cmpeq_epi8(zero, zero); + __m128i abs_p0q0 = abs_diff(p0, q0); + __m128i abs_p1q1 = abs_diff(p1, q1); + __m128i work; + max_abs_p1p0q1q0 = _mm_max_epu8(abs_p1p0, abs_q1q0); + + abs_p0q0 =_mm_adds_epu8(abs_p0q0, abs_p0q0); + abs_p1q1 = _mm_srli_epi16(_mm_and_si128(abs_p1q1, fe), 1); + mask = _mm_subs_epu8(_mm_adds_epu8(abs_p0q0, abs_p1q1), blimit); + mask = _mm_xor_si128(_mm_cmpeq_epi8(mask, zero), ff); + // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2 > blimit) * -1; + mask = _mm_max_epu8(max_abs_p1p0q1q0, mask); + // mask |= (abs(p1 - p0) > limit) * -1; + // mask |= (abs(q1 - q0) > limit) * -1; + work = _mm_max_epu8(abs_diff(p2, p1), abs_diff(p3, p2)); + mask = _mm_max_epu8(work, mask); + work = _mm_max_epu8(abs_diff(q2, q1), abs_diff(q3, q2)); + mask = _mm_max_epu8(work, mask); + mask = _mm_subs_epu8(mask, limit); + mask = _mm_cmpeq_epi8(mask, zero); + } + + { + __m128i work; + work = _mm_max_epu8(abs_diff(p2, p0), abs_diff(q2, q0)); + flat = _mm_max_epu8(work, max_abs_p1p0q1q0); + work = _mm_max_epu8(abs_diff(p3, p0), abs_diff(q3, q0)); + flat = _mm_max_epu8(work, flat); + work = _mm_max_epu8(abs_diff(p4, p0), abs_diff(q4, q0)); + flat = _mm_subs_epu8(flat, one); + flat = _mm_cmpeq_epi8(flat, zero); + flat = _mm_and_si128(flat, mask); + flat2 = _mm_max_epu8(abs_diff(p5, p0), abs_diff(q5, q0)); + flat2 = _mm_max_epu8(work, flat2); + work = _mm_max_epu8(abs_diff(p6, p0), abs_diff(q6, q0)); + flat2 = _mm_max_epu8(work, flat2); + work = _mm_max_epu8(abs_diff(p7, p0), abs_diff(q7, q0)); + flat2 = _mm_max_epu8(work, flat2); + flat2 = _mm_subs_epu8(flat2, one); + flat2 = _mm_cmpeq_epi8(flat2, zero); + flat2 = _mm_and_si128(flat2, flat); // flat2 & flat & mask + } + + // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + // filter4 + { + const __m128i t4 = _mm_set1_epi8(4); + const __m128i t3 = _mm_set1_epi8(3); + const __m128i t80 = _mm_set1_epi8(0x80); + const __m128i te0 = _mm_set1_epi8(0xe0); + const __m128i t1f = _mm_set1_epi8(0x1f); + const __m128i t1 = _mm_set1_epi8(0x1); + const __m128i t7f = _mm_set1_epi8(0x7f); + const __m128i ff = _mm_cmpeq_epi8(t4, t4); + + __m128i filt; + __m128i work_a; + __m128i filter1, filter2; + + op1 = _mm_xor_si128(p1, t80); + op0 = _mm_xor_si128(p0, t80); + oq0 = _mm_xor_si128(q0, t80); + oq1 = _mm_xor_si128(q1, t80); + + hev = _mm_subs_epu8(max_abs_p1p0q1q0, thresh); + hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff); + filt = _mm_and_si128(_mm_subs_epi8(op1, oq1), hev); + + work_a = _mm_subs_epi8(oq0, op0); + filt = _mm_adds_epi8(filt, work_a); + filt = _mm_adds_epi8(filt, work_a); + filt = _mm_adds_epi8(filt, work_a); + // (vpx_filter + 3 * (qs0 - ps0)) & mask + filt = _mm_and_si128(filt, mask); + filter1 = _mm_adds_epi8(filt, t4); + filter2 = _mm_adds_epi8(filt, t3); + + // Filter1 >> 3 + work_a = _mm_cmpgt_epi8(zero, filter1); + filter1 = _mm_srli_epi16(filter1, 3); + work_a = _mm_and_si128(work_a, te0); + filter1 = _mm_and_si128(filter1, t1f); + filter1 = _mm_or_si128(filter1, work_a); + oq0 = _mm_xor_si128(_mm_subs_epi8(oq0, filter1), t80); + + // Filter2 >> 3 + work_a = _mm_cmpgt_epi8(zero, filter2); + filter2 = _mm_srli_epi16(filter2, 3); + work_a = _mm_and_si128(work_a, te0); + filter2 = _mm_and_si128(filter2, t1f); + filter2 = _mm_or_si128(filter2, work_a); + op0 = _mm_xor_si128(_mm_adds_epi8(op0, filter2), t80); + + // filt >> 1 + filt = _mm_adds_epi8(filter1, t1); + work_a = _mm_cmpgt_epi8(zero, filt); + filt = _mm_srli_epi16(filt, 1); + work_a = _mm_and_si128(work_a, t80); + filt = _mm_and_si128(filt, t7f); + filt = _mm_or_si128(filt, work_a); + filt = _mm_andnot_si128(hev, filt); + op1 = _mm_xor_si128(_mm_adds_epi8(op1, filt), t80); + oq1 = _mm_xor_si128(_mm_subs_epi8(oq1, filt), t80); + // loopfilter done + + // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + // filter8 + { + const __m128i four = _mm_set1_epi16(4); + const __m128i p3_lo = _mm_unpacklo_epi8(p3, zero); + const __m128i p2_lo = _mm_unpacklo_epi8(p2, zero); + const __m128i p1_lo = _mm_unpacklo_epi8(p1, zero); + const __m128i p0_lo = _mm_unpacklo_epi8(p0, zero); + const __m128i q0_lo = _mm_unpacklo_epi8(q0, zero); + const __m128i q1_lo = _mm_unpacklo_epi8(q1, zero); + const __m128i q2_lo = _mm_unpacklo_epi8(q2, zero); + const __m128i q3_lo = _mm_unpacklo_epi8(q3, zero); + + const __m128i p3_hi = _mm_unpackhi_epi8(p3, zero); + const __m128i p2_hi = _mm_unpackhi_epi8(p2, zero); + const __m128i p1_hi = _mm_unpackhi_epi8(p1, zero); + const __m128i p0_hi = _mm_unpackhi_epi8(p0, zero); + const __m128i q0_hi = _mm_unpackhi_epi8(q0, zero); + const __m128i q1_hi = _mm_unpackhi_epi8(q1, zero); + const __m128i q2_hi = _mm_unpackhi_epi8(q2, zero); + const __m128i q3_hi = _mm_unpackhi_epi8(q3, zero); + __m128i f8_lo, f8_hi; + + f8_lo = _mm_add_epi16(_mm_add_epi16(p3_lo, four), + _mm_add_epi16(p3_lo, p2_lo)); + f8_lo = _mm_add_epi16(_mm_add_epi16(p3_lo, f8_lo), + _mm_add_epi16(p2_lo, p1_lo)); + f8_lo = _mm_add_epi16(_mm_add_epi16(p0_lo, q0_lo), f8_lo); + + f8_hi = _mm_add_epi16(_mm_add_epi16(p3_hi, four), + _mm_add_epi16(p3_hi, p2_hi)); + f8_hi = _mm_add_epi16(_mm_add_epi16(p3_hi, f8_hi), + _mm_add_epi16(p2_hi, p1_hi)); + f8_hi = _mm_add_epi16(_mm_add_epi16(p0_hi, q0_hi), f8_hi); + + op2 = filter8_mask(&flat, &p2, &f8_lo, &f8_hi); + + f8_lo = filter_add2_sub2(&f8_lo, &q1_lo, &p1_lo, &p2_lo, &p3_lo); + f8_hi = filter_add2_sub2(&f8_hi, &q1_hi, &p1_hi, &p2_hi, &p3_hi); + op1 = filter8_mask(&flat, &op1, &f8_lo, &f8_hi); + + f8_lo = filter_add2_sub2(&f8_lo, &q2_lo, &p0_lo, &p1_lo, &p3_lo); + f8_hi = filter_add2_sub2(&f8_hi, &q2_hi, &p0_hi, &p1_hi, &p3_hi); + op0 = filter8_mask(&flat, &op0, &f8_lo, &f8_hi); + + f8_lo = filter_add2_sub2(&f8_lo, &q3_lo, &q0_lo, &p0_lo, &p3_lo); + f8_hi = filter_add2_sub2(&f8_hi, &q3_hi, &q0_hi, &p0_hi, &p3_hi); + oq0 = filter8_mask(&flat, &oq0, &f8_lo, &f8_hi); + + f8_lo = filter_add2_sub2(&f8_lo, &q3_lo, &q1_lo, &q0_lo, &p2_lo); + f8_hi = filter_add2_sub2(&f8_hi, &q3_hi, &q1_hi, &q0_hi, &p2_hi); + oq1 = filter8_mask(&flat, &oq1, &f8_lo, &f8_hi); + + f8_lo = filter_add2_sub2(&f8_lo, &q3_lo, &q2_lo, &q1_lo, &p1_lo); + f8_hi = filter_add2_sub2(&f8_hi, &q3_hi, &q2_hi, &q1_hi, &p1_hi); + oq2 = filter8_mask(&flat, &q2, &f8_lo, &f8_hi); + } + + // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + // wide flat calculations + { + const __m128i eight = _mm_set1_epi16(8); + const __m128i p7_lo = _mm_unpacklo_epi8(p7, zero); + const __m128i p6_lo = _mm_unpacklo_epi8(p6, zero); + const __m128i p5_lo = _mm_unpacklo_epi8(p5, zero); + const __m128i p4_lo = _mm_unpacklo_epi8(p4, zero); + const __m128i p3_lo = _mm_unpacklo_epi8(p3, zero); + const __m128i p2_lo = _mm_unpacklo_epi8(p2, zero); + const __m128i p1_lo = _mm_unpacklo_epi8(p1, zero); + const __m128i p0_lo = _mm_unpacklo_epi8(p0, zero); + const __m128i q0_lo = _mm_unpacklo_epi8(q0, zero); + const __m128i q1_lo = _mm_unpacklo_epi8(q1, zero); + const __m128i q2_lo = _mm_unpacklo_epi8(q2, zero); + const __m128i q3_lo = _mm_unpacklo_epi8(q3, zero); + const __m128i q4_lo = _mm_unpacklo_epi8(q4, zero); + const __m128i q5_lo = _mm_unpacklo_epi8(q5, zero); + const __m128i q6_lo = _mm_unpacklo_epi8(q6, zero); + const __m128i q7_lo = _mm_unpacklo_epi8(q7, zero); + + const __m128i p7_hi = _mm_unpackhi_epi8(p7, zero); + const __m128i p6_hi = _mm_unpackhi_epi8(p6, zero); + const __m128i p5_hi = _mm_unpackhi_epi8(p5, zero); + const __m128i p4_hi = _mm_unpackhi_epi8(p4, zero); + const __m128i p3_hi = _mm_unpackhi_epi8(p3, zero); + const __m128i p2_hi = _mm_unpackhi_epi8(p2, zero); + const __m128i p1_hi = _mm_unpackhi_epi8(p1, zero); + const __m128i p0_hi = _mm_unpackhi_epi8(p0, zero); + const __m128i q0_hi = _mm_unpackhi_epi8(q0, zero); + const __m128i q1_hi = _mm_unpackhi_epi8(q1, zero); + const __m128i q2_hi = _mm_unpackhi_epi8(q2, zero); + const __m128i q3_hi = _mm_unpackhi_epi8(q3, zero); + const __m128i q4_hi = _mm_unpackhi_epi8(q4, zero); + const __m128i q5_hi = _mm_unpackhi_epi8(q5, zero); + const __m128i q6_hi = _mm_unpackhi_epi8(q6, zero); + const __m128i q7_hi = _mm_unpackhi_epi8(q7, zero); + + __m128i f_lo; + __m128i f_hi; + + f_lo = _mm_sub_epi16(_mm_slli_epi16(p7_lo, 3), p7_lo); // p7 * 7 + f_lo = _mm_add_epi16(_mm_slli_epi16(p6_lo, 1), + _mm_add_epi16(p4_lo, f_lo)); + f_lo = _mm_add_epi16(_mm_add_epi16(p3_lo, f_lo), + _mm_add_epi16(p2_lo, p1_lo)); + f_lo = _mm_add_epi16(_mm_add_epi16(p0_lo, q0_lo), f_lo); + f_lo = _mm_add_epi16(_mm_add_epi16(p5_lo, eight), f_lo); + + f_hi = _mm_sub_epi16(_mm_slli_epi16(p7_hi, 3), p7_hi); // p7 * 7 + f_hi = _mm_add_epi16(_mm_slli_epi16(p6_hi, 1), + _mm_add_epi16(p4_hi, f_hi)); + f_hi = _mm_add_epi16(_mm_add_epi16(p3_hi, f_hi), + _mm_add_epi16(p2_hi, p1_hi)); + f_hi = _mm_add_epi16(_mm_add_epi16(p0_hi, q0_hi), f_hi); + f_hi = _mm_add_epi16(_mm_add_epi16(p5_hi, eight), f_hi); + + p6 = filter16_mask(&flat2, &p6, &f_lo, &f_hi); + _mm_storeu_si128((__m128i *)(s - 7 * p), p6); + + f_lo = filter_add2_sub2(&f_lo, &q1_lo, &p5_lo, &p6_lo, &p7_lo); + f_hi = filter_add2_sub2(&f_hi, &q1_hi, &p5_hi, &p6_hi, &p7_hi); + p5 = filter16_mask(&flat2, &p5, &f_lo, &f_hi); + _mm_storeu_si128((__m128i *)(s - 6 * p), p5); + + f_lo = filter_add2_sub2(&f_lo, &q2_lo, &p4_lo, &p5_lo, &p7_lo); + f_hi = filter_add2_sub2(&f_hi, &q2_hi, &p4_hi, &p5_hi, &p7_hi); + p4 = filter16_mask(&flat2, &p4, &f_lo, &f_hi); + _mm_storeu_si128((__m128i *)(s - 5 * p), p4); + + f_lo = filter_add2_sub2(&f_lo, &q3_lo, &p3_lo, &p4_lo, &p7_lo); + f_hi = filter_add2_sub2(&f_hi, &q3_hi, &p3_hi, &p4_hi, &p7_hi); + p3 = filter16_mask(&flat2, &p3, &f_lo, &f_hi); + _mm_storeu_si128((__m128i *)(s - 4 * p), p3); + + f_lo = filter_add2_sub2(&f_lo, &q4_lo, &p2_lo, &p3_lo, &p7_lo); + f_hi = filter_add2_sub2(&f_hi, &q4_hi, &p2_hi, &p3_hi, &p7_hi); + op2 = filter16_mask(&flat2, &op2, &f_lo, &f_hi); + _mm_storeu_si128((__m128i *)(s - 3 * p), op2); + + f_lo = filter_add2_sub2(&f_lo, &q5_lo, &p1_lo, &p2_lo, &p7_lo); + f_hi = filter_add2_sub2(&f_hi, &q5_hi, &p1_hi, &p2_hi, &p7_hi); + op1 = filter16_mask(&flat2, &op1, &f_lo, &f_hi); + _mm_storeu_si128((__m128i *)(s - 2 * p), op1); + + f_lo = filter_add2_sub2(&f_lo, &q6_lo, &p0_lo, &p1_lo, &p7_lo); + f_hi = filter_add2_sub2(&f_hi, &q6_hi, &p0_hi, &p1_hi, &p7_hi); + op0 = filter16_mask(&flat2, &op0, &f_lo, &f_hi); + _mm_storeu_si128((__m128i *)(s - 1 * p), op0); + + f_lo = filter_add2_sub2(&f_lo, &q7_lo, &q0_lo, &p0_lo, &p7_lo); + f_hi = filter_add2_sub2(&f_hi, &q7_hi, &q0_hi, &p0_hi, &p7_hi); + oq0 = filter16_mask(&flat2, &oq0, &f_lo, &f_hi); + _mm_storeu_si128((__m128i *)(s - 0 * p), oq0); + + f_lo = filter_add2_sub2(&f_lo, &q7_lo, &q1_lo, &p6_lo, &q0_lo); + f_hi = filter_add2_sub2(&f_hi, &q7_hi, &q1_hi, &p6_hi, &q0_hi); + oq1 = filter16_mask(&flat2, &oq1, &f_lo, &f_hi); + _mm_storeu_si128((__m128i *)(s + 1 * p), oq1); + + f_lo = filter_add2_sub2(&f_lo, &q7_lo, &q2_lo, &p5_lo, &q1_lo); + f_hi = filter_add2_sub2(&f_hi, &q7_hi, &q2_hi, &p5_hi, &q1_hi); + oq2 = filter16_mask(&flat2, &oq2, &f_lo, &f_hi); + _mm_storeu_si128((__m128i *)(s + 2 * p), oq2); + + f_lo = filter_add2_sub2(&f_lo, &q7_lo, &q3_lo, &p4_lo, &q2_lo); + f_hi = filter_add2_sub2(&f_hi, &q7_hi, &q3_hi, &p4_hi, &q2_hi); + q3 = filter16_mask(&flat2, &q3, &f_lo, &f_hi); + _mm_storeu_si128((__m128i *)(s + 3 * p), q3); + + f_lo = filter_add2_sub2(&f_lo, &q7_lo, &q4_lo, &p3_lo, &q3_lo); + f_hi = filter_add2_sub2(&f_hi, &q7_hi, &q4_hi, &p3_hi, &q3_hi); + q4 = filter16_mask(&flat2, &q4, &f_lo, &f_hi); + _mm_storeu_si128((__m128i *)(s + 4 * p), q4); + + f_lo = filter_add2_sub2(&f_lo, &q7_lo, &q5_lo, &p2_lo, &q4_lo); + f_hi = filter_add2_sub2(&f_hi, &q7_hi, &q5_hi, &p2_hi, &q4_hi); + q5 = filter16_mask(&flat2, &q5, &f_lo, &f_hi); + _mm_storeu_si128((__m128i *)(s + 5 * p), q5); + + f_lo = filter_add2_sub2(&f_lo, &q7_lo, &q6_lo, &p1_lo, &q5_lo); + f_hi = filter_add2_sub2(&f_hi, &q7_hi, &q6_hi, &p1_hi, &q5_hi); + q6 = filter16_mask(&flat2, &q6, &f_lo, &f_hi); + _mm_storeu_si128((__m128i *)(s + 6 * p), q6); + } + // wide flat + // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + } +} + +void vpx_lpf_horizontal_8_sse2(unsigned char *s, int p, + const unsigned char *_blimit, + const unsigned char *_limit, + const unsigned char *_thresh) { + DECLARE_ALIGNED(16, unsigned char, flat_op2[16]); + DECLARE_ALIGNED(16, unsigned char, flat_op1[16]); + DECLARE_ALIGNED(16, unsigned char, flat_op0[16]); + DECLARE_ALIGNED(16, unsigned char, flat_oq2[16]); + DECLARE_ALIGNED(16, unsigned char, flat_oq1[16]); + DECLARE_ALIGNED(16, unsigned char, flat_oq0[16]); + const __m128i zero = _mm_set1_epi16(0); + const __m128i blimit = _mm_load_si128((const __m128i *)_blimit); + const __m128i limit = _mm_load_si128((const __m128i *)_limit); + const __m128i thresh = _mm_load_si128((const __m128i *)_thresh); + __m128i mask, hev, flat; + __m128i p3, p2, p1, p0, q0, q1, q2, q3; + __m128i q3p3, q2p2, q1p1, q0p0, p1q1, p0q0; + + q3p3 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(s - 4 * p)), + _mm_loadl_epi64((__m128i *)(s + 3 * p))); + q2p2 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(s - 3 * p)), + _mm_loadl_epi64((__m128i *)(s + 2 * p))); + q1p1 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(s - 2 * p)), + _mm_loadl_epi64((__m128i *)(s + 1 * p))); + q0p0 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(s - 1 * p)), + _mm_loadl_epi64((__m128i *)(s - 0 * p))); + p1q1 = _mm_shuffle_epi32(q1p1, 78); + p0q0 = _mm_shuffle_epi32(q0p0, 78); + + { + // filter_mask and hev_mask + const __m128i one = _mm_set1_epi8(1); + const __m128i fe = _mm_set1_epi8(0xfe); + const __m128i ff = _mm_cmpeq_epi8(fe, fe); + __m128i abs_p1q1, abs_p0q0, abs_q1q0, abs_p1p0, work; + abs_p1p0 = abs_diff(q1p1, q0p0); + abs_q1q0 = _mm_srli_si128(abs_p1p0, 8); + + abs_p0q0 = abs_diff(q0p0, p0q0); + abs_p1q1 = abs_diff(q1p1, p1q1); + flat = _mm_max_epu8(abs_p1p0, abs_q1q0); + hev = _mm_subs_epu8(flat, thresh); + hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff); + + abs_p0q0 =_mm_adds_epu8(abs_p0q0, abs_p0q0); + abs_p1q1 = _mm_srli_epi16(_mm_and_si128(abs_p1q1, fe), 1); + mask = _mm_subs_epu8(_mm_adds_epu8(abs_p0q0, abs_p1q1), blimit); + mask = _mm_xor_si128(_mm_cmpeq_epi8(mask, zero), ff); + // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2 > blimit) * -1; + mask = _mm_max_epu8(abs_p1p0, mask); + // mask |= (abs(p1 - p0) > limit) * -1; + // mask |= (abs(q1 - q0) > limit) * -1; + + work = _mm_max_epu8(abs_diff(q2p2, q1p1), + abs_diff(q3p3, q2p2)); + mask = _mm_max_epu8(work, mask); + mask = _mm_max_epu8(mask, _mm_srli_si128(mask, 8)); + mask = _mm_subs_epu8(mask, limit); + mask = _mm_cmpeq_epi8(mask, zero); + + // flat_mask4 + + flat = _mm_max_epu8(abs_diff(q2p2, q0p0), + abs_diff(q3p3, q0p0)); + flat = _mm_max_epu8(abs_p1p0, flat); + flat = _mm_max_epu8(flat, _mm_srli_si128(flat, 8)); + flat = _mm_subs_epu8(flat, one); + flat = _mm_cmpeq_epi8(flat, zero); + flat = _mm_and_si128(flat, mask); + } + + { + const __m128i four = _mm_set1_epi16(4); + unsigned char *src = s; + { + __m128i workp_a, workp_b, workp_shft; + p3 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 4 * p)), zero); + p2 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 3 * p)), zero); + p1 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 2 * p)), zero); + p0 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 1 * p)), zero); + q0 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 0 * p)), zero); + q1 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src + 1 * p)), zero); + q2 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src + 2 * p)), zero); + q3 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src + 3 * p)), zero); + + workp_a = _mm_add_epi16(_mm_add_epi16(p3, p3), _mm_add_epi16(p2, p1)); + workp_a = _mm_add_epi16(_mm_add_epi16(workp_a, four), p0); + workp_b = _mm_add_epi16(_mm_add_epi16(q0, p2), p3); + workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3); + _mm_storel_epi64((__m128i *)&flat_op2[0], + _mm_packus_epi16(workp_shft, workp_shft)); + + workp_b = _mm_add_epi16(_mm_add_epi16(q0, q1), p1); + workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3); + _mm_storel_epi64((__m128i *)&flat_op1[0], + _mm_packus_epi16(workp_shft, workp_shft)); + + workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p3), q2); + workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, p1), p0); + workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3); + _mm_storel_epi64((__m128i *)&flat_op0[0], + _mm_packus_epi16(workp_shft, workp_shft)); + + workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p3), q3); + workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, p0), q0); + workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3); + _mm_storel_epi64((__m128i *)&flat_oq0[0], + _mm_packus_epi16(workp_shft, workp_shft)); + + workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p2), q3); + workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, q0), q1); + workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3); + _mm_storel_epi64((__m128i *)&flat_oq1[0], + _mm_packus_epi16(workp_shft, workp_shft)); + + workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p1), q3); + workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, q1), q2); + workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3); + _mm_storel_epi64((__m128i *)&flat_oq2[0], + _mm_packus_epi16(workp_shft, workp_shft)); + } + } + // lp filter + { + const __m128i t4 = _mm_set1_epi8(4); + const __m128i t3 = _mm_set1_epi8(3); + const __m128i t80 = _mm_set1_epi8(0x80); + const __m128i t1 = _mm_set1_epi8(0x1); + const __m128i ps1 = _mm_xor_si128(_mm_loadl_epi64((__m128i *)(s - 2 * p)), + t80); + const __m128i ps0 = _mm_xor_si128(_mm_loadl_epi64((__m128i *)(s - 1 * p)), + t80); + const __m128i qs0 = _mm_xor_si128(_mm_loadl_epi64((__m128i *)(s + 0 * p)), + t80); + const __m128i qs1 = _mm_xor_si128(_mm_loadl_epi64((__m128i *)(s + 1 * p)), + t80); + __m128i filt; + __m128i work_a; + __m128i filter1, filter2; + + filt = _mm_and_si128(_mm_subs_epi8(ps1, qs1), hev); + work_a = _mm_subs_epi8(qs0, ps0); + filt = _mm_adds_epi8(filt, work_a); + filt = _mm_adds_epi8(filt, work_a); + filt = _mm_adds_epi8(filt, work_a); + // (vpx_filter + 3 * (qs0 - ps0)) & mask + filt = _mm_and_si128(filt, mask); + + filter1 = _mm_adds_epi8(filt, t4); + filter2 = _mm_adds_epi8(filt, t3); + + // Filter1 >> 3 + filter1 = _mm_unpacklo_epi8(zero, filter1); + filter1 = _mm_srai_epi16(filter1, 11); + filter1 = _mm_packs_epi16(filter1, filter1); + + // Filter2 >> 3 + filter2 = _mm_unpacklo_epi8(zero, filter2); + filter2 = _mm_srai_epi16(filter2, 11); + filter2 = _mm_packs_epi16(filter2, zero); + + // filt >> 1 + filt = _mm_adds_epi8(filter1, t1); + filt = _mm_unpacklo_epi8(zero, filt); + filt = _mm_srai_epi16(filt, 9); + filt = _mm_packs_epi16(filt, zero); + + filt = _mm_andnot_si128(hev, filt); + + work_a = _mm_xor_si128(_mm_subs_epi8(qs0, filter1), t80); + q0 = _mm_loadl_epi64((__m128i *)flat_oq0); + work_a = _mm_andnot_si128(flat, work_a); + q0 = _mm_and_si128(flat, q0); + q0 = _mm_or_si128(work_a, q0); + + work_a = _mm_xor_si128(_mm_subs_epi8(qs1, filt), t80); + q1 = _mm_loadl_epi64((__m128i *)flat_oq1); + work_a = _mm_andnot_si128(flat, work_a); + q1 = _mm_and_si128(flat, q1); + q1 = _mm_or_si128(work_a, q1); + + work_a = _mm_loadu_si128((__m128i *)(s + 2 * p)); + q2 = _mm_loadl_epi64((__m128i *)flat_oq2); + work_a = _mm_andnot_si128(flat, work_a); + q2 = _mm_and_si128(flat, q2); + q2 = _mm_or_si128(work_a, q2); + + work_a = _mm_xor_si128(_mm_adds_epi8(ps0, filter2), t80); + p0 = _mm_loadl_epi64((__m128i *)flat_op0); + work_a = _mm_andnot_si128(flat, work_a); + p0 = _mm_and_si128(flat, p0); + p0 = _mm_or_si128(work_a, p0); + + work_a = _mm_xor_si128(_mm_adds_epi8(ps1, filt), t80); + p1 = _mm_loadl_epi64((__m128i *)flat_op1); + work_a = _mm_andnot_si128(flat, work_a); + p1 = _mm_and_si128(flat, p1); + p1 = _mm_or_si128(work_a, p1); + + work_a = _mm_loadu_si128((__m128i *)(s - 3 * p)); + p2 = _mm_loadl_epi64((__m128i *)flat_op2); + work_a = _mm_andnot_si128(flat, work_a); + p2 = _mm_and_si128(flat, p2); + p2 = _mm_or_si128(work_a, p2); + + _mm_storel_epi64((__m128i *)(s - 3 * p), p2); + _mm_storel_epi64((__m128i *)(s - 2 * p), p1); + _mm_storel_epi64((__m128i *)(s - 1 * p), p0); + _mm_storel_epi64((__m128i *)(s + 0 * p), q0); + _mm_storel_epi64((__m128i *)(s + 1 * p), q1); + _mm_storel_epi64((__m128i *)(s + 2 * p), q2); + } +} + +void vpx_lpf_horizontal_8_dual_sse2(uint8_t *s, int p, + const uint8_t *_blimit0, + const uint8_t *_limit0, + const uint8_t *_thresh0, + const uint8_t *_blimit1, + const uint8_t *_limit1, + const uint8_t *_thresh1) { + DECLARE_ALIGNED(16, unsigned char, flat_op2[16]); + DECLARE_ALIGNED(16, unsigned char, flat_op1[16]); + DECLARE_ALIGNED(16, unsigned char, flat_op0[16]); + DECLARE_ALIGNED(16, unsigned char, flat_oq2[16]); + DECLARE_ALIGNED(16, unsigned char, flat_oq1[16]); + DECLARE_ALIGNED(16, unsigned char, flat_oq0[16]); + const __m128i zero = _mm_set1_epi16(0); + const __m128i blimit = + _mm_unpacklo_epi64(_mm_load_si128((const __m128i *)_blimit0), + _mm_load_si128((const __m128i *)_blimit1)); + const __m128i limit = + _mm_unpacklo_epi64(_mm_load_si128((const __m128i *)_limit0), + _mm_load_si128((const __m128i *)_limit1)); + const __m128i thresh = + _mm_unpacklo_epi64(_mm_load_si128((const __m128i *)_thresh0), + _mm_load_si128((const __m128i *)_thresh1)); + + __m128i mask, hev, flat; + __m128i p3, p2, p1, p0, q0, q1, q2, q3; + + p3 = _mm_loadu_si128((__m128i *)(s - 4 * p)); + p2 = _mm_loadu_si128((__m128i *)(s - 3 * p)); + p1 = _mm_loadu_si128((__m128i *)(s - 2 * p)); + p0 = _mm_loadu_si128((__m128i *)(s - 1 * p)); + q0 = _mm_loadu_si128((__m128i *)(s - 0 * p)); + q1 = _mm_loadu_si128((__m128i *)(s + 1 * p)); + q2 = _mm_loadu_si128((__m128i *)(s + 2 * p)); + q3 = _mm_loadu_si128((__m128i *)(s + 3 * p)); + { + const __m128i abs_p1p0 = _mm_or_si128(_mm_subs_epu8(p1, p0), + _mm_subs_epu8(p0, p1)); + const __m128i abs_q1q0 = _mm_or_si128(_mm_subs_epu8(q1, q0), + _mm_subs_epu8(q0, q1)); + const __m128i one = _mm_set1_epi8(1); + const __m128i fe = _mm_set1_epi8(0xfe); + const __m128i ff = _mm_cmpeq_epi8(abs_p1p0, abs_p1p0); + __m128i abs_p0q0 = _mm_or_si128(_mm_subs_epu8(p0, q0), + _mm_subs_epu8(q0, p0)); + __m128i abs_p1q1 = _mm_or_si128(_mm_subs_epu8(p1, q1), + _mm_subs_epu8(q1, p1)); + __m128i work; + + // filter_mask and hev_mask + flat = _mm_max_epu8(abs_p1p0, abs_q1q0); + hev = _mm_subs_epu8(flat, thresh); + hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff); + + abs_p0q0 =_mm_adds_epu8(abs_p0q0, abs_p0q0); + abs_p1q1 = _mm_srli_epi16(_mm_and_si128(abs_p1q1, fe), 1); + mask = _mm_subs_epu8(_mm_adds_epu8(abs_p0q0, abs_p1q1), blimit); + mask = _mm_xor_si128(_mm_cmpeq_epi8(mask, zero), ff); + // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2 > blimit) * -1; + mask = _mm_max_epu8(flat, mask); + // mask |= (abs(p1 - p0) > limit) * -1; + // mask |= (abs(q1 - q0) > limit) * -1; + work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p2, p1), + _mm_subs_epu8(p1, p2)), + _mm_or_si128(_mm_subs_epu8(p3, p2), + _mm_subs_epu8(p2, p3))); + mask = _mm_max_epu8(work, mask); + work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(q2, q1), + _mm_subs_epu8(q1, q2)), + _mm_or_si128(_mm_subs_epu8(q3, q2), + _mm_subs_epu8(q2, q3))); + mask = _mm_max_epu8(work, mask); + mask = _mm_subs_epu8(mask, limit); + mask = _mm_cmpeq_epi8(mask, zero); + + // flat_mask4 + work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p2, p0), + _mm_subs_epu8(p0, p2)), + _mm_or_si128(_mm_subs_epu8(q2, q0), + _mm_subs_epu8(q0, q2))); + flat = _mm_max_epu8(work, flat); + work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p3, p0), + _mm_subs_epu8(p0, p3)), + _mm_or_si128(_mm_subs_epu8(q3, q0), + _mm_subs_epu8(q0, q3))); + flat = _mm_max_epu8(work, flat); + flat = _mm_subs_epu8(flat, one); + flat = _mm_cmpeq_epi8(flat, zero); + flat = _mm_and_si128(flat, mask); + } + { + const __m128i four = _mm_set1_epi16(4); + unsigned char *src = s; + int i = 0; + + do { + __m128i workp_a, workp_b, workp_shft; + p3 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 4 * p)), zero); + p2 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 3 * p)), zero); + p1 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 2 * p)), zero); + p0 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 1 * p)), zero); + q0 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 0 * p)), zero); + q1 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src + 1 * p)), zero); + q2 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src + 2 * p)), zero); + q3 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src + 3 * p)), zero); + + workp_a = _mm_add_epi16(_mm_add_epi16(p3, p3), _mm_add_epi16(p2, p1)); + workp_a = _mm_add_epi16(_mm_add_epi16(workp_a, four), p0); + workp_b = _mm_add_epi16(_mm_add_epi16(q0, p2), p3); + workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3); + _mm_storel_epi64((__m128i *)&flat_op2[i * 8], + _mm_packus_epi16(workp_shft, workp_shft)); + + workp_b = _mm_add_epi16(_mm_add_epi16(q0, q1), p1); + workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3); + _mm_storel_epi64((__m128i *)&flat_op1[i * 8], + _mm_packus_epi16(workp_shft, workp_shft)); + + workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p3), q2); + workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, p1), p0); + workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3); + _mm_storel_epi64((__m128i *)&flat_op0[i * 8], + _mm_packus_epi16(workp_shft, workp_shft)); + + workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p3), q3); + workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, p0), q0); + workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3); + _mm_storel_epi64((__m128i *)&flat_oq0[i * 8], + _mm_packus_epi16(workp_shft, workp_shft)); + + workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p2), q3); + workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, q0), q1); + workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3); + _mm_storel_epi64((__m128i *)&flat_oq1[i * 8], + _mm_packus_epi16(workp_shft, workp_shft)); + + workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p1), q3); + workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, q1), q2); + workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3); + _mm_storel_epi64((__m128i *)&flat_oq2[i * 8], + _mm_packus_epi16(workp_shft, workp_shft)); + + src += 8; + } while (++i < 2); + } + // lp filter + { + const __m128i t4 = _mm_set1_epi8(4); + const __m128i t3 = _mm_set1_epi8(3); + const __m128i t80 = _mm_set1_epi8(0x80); + const __m128i te0 = _mm_set1_epi8(0xe0); + const __m128i t1f = _mm_set1_epi8(0x1f); + const __m128i t1 = _mm_set1_epi8(0x1); + const __m128i t7f = _mm_set1_epi8(0x7f); + + const __m128i ps1 = _mm_xor_si128(_mm_loadu_si128((__m128i *)(s - 2 * p)), + t80); + const __m128i ps0 = _mm_xor_si128(_mm_loadu_si128((__m128i *)(s - 1 * p)), + t80); + const __m128i qs0 = _mm_xor_si128(_mm_loadu_si128((__m128i *)(s + 0 * p)), + t80); + const __m128i qs1 = _mm_xor_si128(_mm_loadu_si128((__m128i *)(s + 1 * p)), + t80); + __m128i filt; + __m128i work_a; + __m128i filter1, filter2; + + filt = _mm_and_si128(_mm_subs_epi8(ps1, qs1), hev); + work_a = _mm_subs_epi8(qs0, ps0); + filt = _mm_adds_epi8(filt, work_a); + filt = _mm_adds_epi8(filt, work_a); + filt = _mm_adds_epi8(filt, work_a); + // (vpx_filter + 3 * (qs0 - ps0)) & mask + filt = _mm_and_si128(filt, mask); + + filter1 = _mm_adds_epi8(filt, t4); + filter2 = _mm_adds_epi8(filt, t3); + + // Filter1 >> 3 + work_a = _mm_cmpgt_epi8(zero, filter1); + filter1 = _mm_srli_epi16(filter1, 3); + work_a = _mm_and_si128(work_a, te0); + filter1 = _mm_and_si128(filter1, t1f); + filter1 = _mm_or_si128(filter1, work_a); + + // Filter2 >> 3 + work_a = _mm_cmpgt_epi8(zero, filter2); + filter2 = _mm_srli_epi16(filter2, 3); + work_a = _mm_and_si128(work_a, te0); + filter2 = _mm_and_si128(filter2, t1f); + filter2 = _mm_or_si128(filter2, work_a); + + // filt >> 1 + filt = _mm_adds_epi8(filter1, t1); + work_a = _mm_cmpgt_epi8(zero, filt); + filt = _mm_srli_epi16(filt, 1); + work_a = _mm_and_si128(work_a, t80); + filt = _mm_and_si128(filt, t7f); + filt = _mm_or_si128(filt, work_a); + + filt = _mm_andnot_si128(hev, filt); + + work_a = _mm_xor_si128(_mm_subs_epi8(qs0, filter1), t80); + q0 = _mm_load_si128((__m128i *)flat_oq0); + work_a = _mm_andnot_si128(flat, work_a); + q0 = _mm_and_si128(flat, q0); + q0 = _mm_or_si128(work_a, q0); + + work_a = _mm_xor_si128(_mm_subs_epi8(qs1, filt), t80); + q1 = _mm_load_si128((__m128i *)flat_oq1); + work_a = _mm_andnot_si128(flat, work_a); + q1 = _mm_and_si128(flat, q1); + q1 = _mm_or_si128(work_a, q1); + + work_a = _mm_loadu_si128((__m128i *)(s + 2 * p)); + q2 = _mm_load_si128((__m128i *)flat_oq2); + work_a = _mm_andnot_si128(flat, work_a); + q2 = _mm_and_si128(flat, q2); + q2 = _mm_or_si128(work_a, q2); + + work_a = _mm_xor_si128(_mm_adds_epi8(ps0, filter2), t80); + p0 = _mm_load_si128((__m128i *)flat_op0); + work_a = _mm_andnot_si128(flat, work_a); + p0 = _mm_and_si128(flat, p0); + p0 = _mm_or_si128(work_a, p0); + + work_a = _mm_xor_si128(_mm_adds_epi8(ps1, filt), t80); + p1 = _mm_load_si128((__m128i *)flat_op1); + work_a = _mm_andnot_si128(flat, work_a); + p1 = _mm_and_si128(flat, p1); + p1 = _mm_or_si128(work_a, p1); + + work_a = _mm_loadu_si128((__m128i *)(s - 3 * p)); + p2 = _mm_load_si128((__m128i *)flat_op2); + work_a = _mm_andnot_si128(flat, work_a); + p2 = _mm_and_si128(flat, p2); + p2 = _mm_or_si128(work_a, p2); + + _mm_storeu_si128((__m128i *)(s - 3 * p), p2); + _mm_storeu_si128((__m128i *)(s - 2 * p), p1); + _mm_storeu_si128((__m128i *)(s - 1 * p), p0); + _mm_storeu_si128((__m128i *)(s + 0 * p), q0); + _mm_storeu_si128((__m128i *)(s + 1 * p), q1); + _mm_storeu_si128((__m128i *)(s + 2 * p), q2); + } +} + +void vpx_lpf_horizontal_4_dual_sse2(unsigned char *s, int p, + const unsigned char *_blimit0, + const unsigned char *_limit0, + const unsigned char *_thresh0, + const unsigned char *_blimit1, + const unsigned char *_limit1, + const unsigned char *_thresh1) { + const __m128i blimit = + _mm_unpacklo_epi64(_mm_load_si128((const __m128i *)_blimit0), + _mm_load_si128((const __m128i *)_blimit1)); + const __m128i limit = + _mm_unpacklo_epi64(_mm_load_si128((const __m128i *)_limit0), + _mm_load_si128((const __m128i *)_limit1)); + const __m128i thresh = + _mm_unpacklo_epi64(_mm_load_si128((const __m128i *)_thresh0), + _mm_load_si128((const __m128i *)_thresh1)); + const __m128i zero = _mm_set1_epi16(0); + __m128i p3, p2, p1, p0, q0, q1, q2, q3; + __m128i mask, hev, flat; + + p3 = _mm_loadu_si128((__m128i *)(s - 4 * p)); + p2 = _mm_loadu_si128((__m128i *)(s - 3 * p)); + p1 = _mm_loadu_si128((__m128i *)(s - 2 * p)); + p0 = _mm_loadu_si128((__m128i *)(s - 1 * p)); + q0 = _mm_loadu_si128((__m128i *)(s - 0 * p)); + q1 = _mm_loadu_si128((__m128i *)(s + 1 * p)); + q2 = _mm_loadu_si128((__m128i *)(s + 2 * p)); + q3 = _mm_loadu_si128((__m128i *)(s + 3 * p)); + + // filter_mask and hev_mask + { + const __m128i abs_p1p0 = _mm_or_si128(_mm_subs_epu8(p1, p0), + _mm_subs_epu8(p0, p1)); + const __m128i abs_q1q0 = _mm_or_si128(_mm_subs_epu8(q1, q0), + _mm_subs_epu8(q0, q1)); + const __m128i fe = _mm_set1_epi8(0xfe); + const __m128i ff = _mm_cmpeq_epi8(abs_p1p0, abs_p1p0); + __m128i abs_p0q0 = _mm_or_si128(_mm_subs_epu8(p0, q0), + _mm_subs_epu8(q0, p0)); + __m128i abs_p1q1 = _mm_or_si128(_mm_subs_epu8(p1, q1), + _mm_subs_epu8(q1, p1)); + __m128i work; + + flat = _mm_max_epu8(abs_p1p0, abs_q1q0); + hev = _mm_subs_epu8(flat, thresh); + hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff); + + abs_p0q0 =_mm_adds_epu8(abs_p0q0, abs_p0q0); + abs_p1q1 = _mm_srli_epi16(_mm_and_si128(abs_p1q1, fe), 1); + mask = _mm_subs_epu8(_mm_adds_epu8(abs_p0q0, abs_p1q1), blimit); + mask = _mm_xor_si128(_mm_cmpeq_epi8(mask, zero), ff); + // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2 > blimit) * -1; + mask = _mm_max_epu8(flat, mask); + // mask |= (abs(p1 - p0) > limit) * -1; + // mask |= (abs(q1 - q0) > limit) * -1; + work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p2, p1), + _mm_subs_epu8(p1, p2)), + _mm_or_si128(_mm_subs_epu8(p3, p2), + _mm_subs_epu8(p2, p3))); + mask = _mm_max_epu8(work, mask); + work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(q2, q1), + _mm_subs_epu8(q1, q2)), + _mm_or_si128(_mm_subs_epu8(q3, q2), + _mm_subs_epu8(q2, q3))); + mask = _mm_max_epu8(work, mask); + mask = _mm_subs_epu8(mask, limit); + mask = _mm_cmpeq_epi8(mask, zero); + } + + // filter4 + { + const __m128i t4 = _mm_set1_epi8(4); + const __m128i t3 = _mm_set1_epi8(3); + const __m128i t80 = _mm_set1_epi8(0x80); + const __m128i te0 = _mm_set1_epi8(0xe0); + const __m128i t1f = _mm_set1_epi8(0x1f); + const __m128i t1 = _mm_set1_epi8(0x1); + const __m128i t7f = _mm_set1_epi8(0x7f); + + const __m128i ps1 = _mm_xor_si128(_mm_loadu_si128((__m128i *)(s - 2 * p)), + t80); + const __m128i ps0 = _mm_xor_si128(_mm_loadu_si128((__m128i *)(s - 1 * p)), + t80); + const __m128i qs0 = _mm_xor_si128(_mm_loadu_si128((__m128i *)(s + 0 * p)), + t80); + const __m128i qs1 = _mm_xor_si128(_mm_loadu_si128((__m128i *)(s + 1 * p)), + t80); + __m128i filt; + __m128i work_a; + __m128i filter1, filter2; + + filt = _mm_and_si128(_mm_subs_epi8(ps1, qs1), hev); + work_a = _mm_subs_epi8(qs0, ps0); + filt = _mm_adds_epi8(filt, work_a); + filt = _mm_adds_epi8(filt, work_a); + filt = _mm_adds_epi8(filt, work_a); + // (vpx_filter + 3 * (qs0 - ps0)) & mask + filt = _mm_and_si128(filt, mask); + + filter1 = _mm_adds_epi8(filt, t4); + filter2 = _mm_adds_epi8(filt, t3); + + // Filter1 >> 3 + work_a = _mm_cmpgt_epi8(zero, filter1); + filter1 = _mm_srli_epi16(filter1, 3); + work_a = _mm_and_si128(work_a, te0); + filter1 = _mm_and_si128(filter1, t1f); + filter1 = _mm_or_si128(filter1, work_a); + + // Filter2 >> 3 + work_a = _mm_cmpgt_epi8(zero, filter2); + filter2 = _mm_srli_epi16(filter2, 3); + work_a = _mm_and_si128(work_a, te0); + filter2 = _mm_and_si128(filter2, t1f); + filter2 = _mm_or_si128(filter2, work_a); + + // filt >> 1 + filt = _mm_adds_epi8(filter1, t1); + work_a = _mm_cmpgt_epi8(zero, filt); + filt = _mm_srli_epi16(filt, 1); + work_a = _mm_and_si128(work_a, t80); + filt = _mm_and_si128(filt, t7f); + filt = _mm_or_si128(filt, work_a); + + filt = _mm_andnot_si128(hev, filt); + + q0 = _mm_xor_si128(_mm_subs_epi8(qs0, filter1), t80); + q1 = _mm_xor_si128(_mm_subs_epi8(qs1, filt), t80); + p0 = _mm_xor_si128(_mm_adds_epi8(ps0, filter2), t80); + p1 = _mm_xor_si128(_mm_adds_epi8(ps1, filt), t80); + + _mm_storeu_si128((__m128i *)(s - 2 * p), p1); + _mm_storeu_si128((__m128i *)(s - 1 * p), p0); + _mm_storeu_si128((__m128i *)(s + 0 * p), q0); + _mm_storeu_si128((__m128i *)(s + 1 * p), q1); + } +} + +static INLINE void transpose8x16(unsigned char *in0, unsigned char *in1, + int in_p, unsigned char *out, int out_p) { + __m128i x0, x1, x2, x3, x4, x5, x6, x7; + __m128i x8, x9, x10, x11, x12, x13, x14, x15; + + // 2-way interleave w/hoisting of unpacks + x0 = _mm_loadl_epi64((__m128i *)in0); // 1 + x1 = _mm_loadl_epi64((__m128i *)(in0 + in_p)); // 3 + x0 = _mm_unpacklo_epi8(x0, x1); // 1 + + x2 = _mm_loadl_epi64((__m128i *)(in0 + 2 * in_p)); // 5 + x3 = _mm_loadl_epi64((__m128i *)(in0 + 3*in_p)); // 7 + x1 = _mm_unpacklo_epi8(x2, x3); // 2 + + x4 = _mm_loadl_epi64((__m128i *)(in0 + 4*in_p)); // 9 + x5 = _mm_loadl_epi64((__m128i *)(in0 + 5*in_p)); // 11 + x2 = _mm_unpacklo_epi8(x4, x5); // 3 + + x6 = _mm_loadl_epi64((__m128i *)(in0 + 6*in_p)); // 13 + x7 = _mm_loadl_epi64((__m128i *)(in0 + 7*in_p)); // 15 + x3 = _mm_unpacklo_epi8(x6, x7); // 4 + x4 = _mm_unpacklo_epi16(x0, x1); // 9 + + x8 = _mm_loadl_epi64((__m128i *)in1); // 2 + x9 = _mm_loadl_epi64((__m128i *)(in1 + in_p)); // 4 + x8 = _mm_unpacklo_epi8(x8, x9); // 5 + x5 = _mm_unpacklo_epi16(x2, x3); // 10 + + x10 = _mm_loadl_epi64((__m128i *)(in1 + 2 * in_p)); // 6 + x11 = _mm_loadl_epi64((__m128i *)(in1 + 3*in_p)); // 8 + x9 = _mm_unpacklo_epi8(x10, x11); // 6 + + x12 = _mm_loadl_epi64((__m128i *)(in1 + 4*in_p)); // 10 + x13 = _mm_loadl_epi64((__m128i *)(in1 + 5*in_p)); // 12 + x10 = _mm_unpacklo_epi8(x12, x13); // 7 + x12 = _mm_unpacklo_epi16(x8, x9); // 11 + + x14 = _mm_loadl_epi64((__m128i *)(in1 + 6*in_p)); // 14 + x15 = _mm_loadl_epi64((__m128i *)(in1 + 7*in_p)); // 16 + x11 = _mm_unpacklo_epi8(x14, x15); // 8 + x13 = _mm_unpacklo_epi16(x10, x11); // 12 + + x6 = _mm_unpacklo_epi32(x4, x5); // 13 + x7 = _mm_unpackhi_epi32(x4, x5); // 14 + x14 = _mm_unpacklo_epi32(x12, x13); // 15 + x15 = _mm_unpackhi_epi32(x12, x13); // 16 + + // Store first 4-line result + _mm_storeu_si128((__m128i *)out, _mm_unpacklo_epi64(x6, x14)); + _mm_storeu_si128((__m128i *)(out + out_p), _mm_unpackhi_epi64(x6, x14)); + _mm_storeu_si128((__m128i *)(out + 2 * out_p), _mm_unpacklo_epi64(x7, x15)); + _mm_storeu_si128((__m128i *)(out + 3 * out_p), _mm_unpackhi_epi64(x7, x15)); + + x4 = _mm_unpackhi_epi16(x0, x1); + x5 = _mm_unpackhi_epi16(x2, x3); + x12 = _mm_unpackhi_epi16(x8, x9); + x13 = _mm_unpackhi_epi16(x10, x11); + + x6 = _mm_unpacklo_epi32(x4, x5); + x7 = _mm_unpackhi_epi32(x4, x5); + x14 = _mm_unpacklo_epi32(x12, x13); + x15 = _mm_unpackhi_epi32(x12, x13); + + // Store second 4-line result + _mm_storeu_si128((__m128i *)(out + 4 * out_p), _mm_unpacklo_epi64(x6, x14)); + _mm_storeu_si128((__m128i *)(out + 5 * out_p), _mm_unpackhi_epi64(x6, x14)); + _mm_storeu_si128((__m128i *)(out + 6 * out_p), _mm_unpacklo_epi64(x7, x15)); + _mm_storeu_si128((__m128i *)(out + 7 * out_p), _mm_unpackhi_epi64(x7, x15)); +} + +static INLINE void transpose(unsigned char *src[], int in_p, + unsigned char *dst[], int out_p, + int num_8x8_to_transpose) { + int idx8x8 = 0; + __m128i x0, x1, x2, x3, x4, x5, x6, x7; + do { + unsigned char *in = src[idx8x8]; + unsigned char *out = dst[idx8x8]; + + x0 = _mm_loadl_epi64((__m128i *)(in + 0*in_p)); // 00 01 02 03 04 05 06 07 + x1 = _mm_loadl_epi64((__m128i *)(in + 1*in_p)); // 10 11 12 13 14 15 16 17 + // 00 10 01 11 02 12 03 13 04 14 05 15 06 16 07 17 + x0 = _mm_unpacklo_epi8(x0, x1); + + x2 = _mm_loadl_epi64((__m128i *)(in + 2*in_p)); // 20 21 22 23 24 25 26 27 + x3 = _mm_loadl_epi64((__m128i *)(in + 3*in_p)); // 30 31 32 33 34 35 36 37 + // 20 30 21 31 22 32 23 33 24 34 25 35 26 36 27 37 + x1 = _mm_unpacklo_epi8(x2, x3); + + x4 = _mm_loadl_epi64((__m128i *)(in + 4*in_p)); // 40 41 42 43 44 45 46 47 + x5 = _mm_loadl_epi64((__m128i *)(in + 5*in_p)); // 50 51 52 53 54 55 56 57 + // 40 50 41 51 42 52 43 53 44 54 45 55 46 56 47 57 + x2 = _mm_unpacklo_epi8(x4, x5); + + x6 = _mm_loadl_epi64((__m128i *)(in + 6*in_p)); // 60 61 62 63 64 65 66 67 + x7 = _mm_loadl_epi64((__m128i *)(in + 7*in_p)); // 70 71 72 73 74 75 76 77 + // 60 70 61 71 62 72 63 73 64 74 65 75 66 76 67 77 + x3 = _mm_unpacklo_epi8(x6, x7); + + // 00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33 + x4 = _mm_unpacklo_epi16(x0, x1); + // 40 50 60 70 41 51 61 71 42 52 62 72 43 53 63 73 + x5 = _mm_unpacklo_epi16(x2, x3); + // 00 10 20 30 40 50 60 70 01 11 21 31 41 51 61 71 + x6 = _mm_unpacklo_epi32(x4, x5); + _mm_storel_pd((double *)(out + 0*out_p), + _mm_castsi128_pd(x6)); // 00 10 20 30 40 50 60 70 + _mm_storeh_pd((double *)(out + 1*out_p), + _mm_castsi128_pd(x6)); // 01 11 21 31 41 51 61 71 + // 02 12 22 32 42 52 62 72 03 13 23 33 43 53 63 73 + x7 = _mm_unpackhi_epi32(x4, x5); + _mm_storel_pd((double *)(out + 2*out_p), + _mm_castsi128_pd(x7)); // 02 12 22 32 42 52 62 72 + _mm_storeh_pd((double *)(out + 3*out_p), + _mm_castsi128_pd(x7)); // 03 13 23 33 43 53 63 73 + + // 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37 + x4 = _mm_unpackhi_epi16(x0, x1); + // 44 54 64 74 45 55 65 75 46 56 66 76 47 57 67 77 + x5 = _mm_unpackhi_epi16(x2, x3); + // 04 14 24 34 44 54 64 74 05 15 25 35 45 55 65 75 + x6 = _mm_unpacklo_epi32(x4, x5); + _mm_storel_pd((double *)(out + 4*out_p), + _mm_castsi128_pd(x6)); // 04 14 24 34 44 54 64 74 + _mm_storeh_pd((double *)(out + 5*out_p), + _mm_castsi128_pd(x6)); // 05 15 25 35 45 55 65 75 + // 06 16 26 36 46 56 66 76 07 17 27 37 47 57 67 77 + x7 = _mm_unpackhi_epi32(x4, x5); + + _mm_storel_pd((double *)(out + 6*out_p), + _mm_castsi128_pd(x7)); // 06 16 26 36 46 56 66 76 + _mm_storeh_pd((double *)(out + 7*out_p), + _mm_castsi128_pd(x7)); // 07 17 27 37 47 57 67 77 + } while (++idx8x8 < num_8x8_to_transpose); +} + +void vpx_lpf_vertical_4_dual_sse2(uint8_t *s, int p, const uint8_t *blimit0, + const uint8_t *limit0, + const uint8_t *thresh0, + const uint8_t *blimit1, + const uint8_t *limit1, + const uint8_t *thresh1) { + DECLARE_ALIGNED(16, unsigned char, t_dst[16 * 8]); + unsigned char *src[2]; + unsigned char *dst[2]; + + // Transpose 8x16 + transpose8x16(s - 4, s - 4 + p * 8, p, t_dst, 16); + + // Loop filtering + vpx_lpf_horizontal_4_dual_sse2(t_dst + 4 * 16, 16, blimit0, limit0, thresh0, + blimit1, limit1, thresh1); + src[0] = t_dst; + src[1] = t_dst + 8; + dst[0] = s - 4; + dst[1] = s - 4 + p * 8; + + // Transpose back + transpose(src, 16, dst, p, 2); +} + +void vpx_lpf_vertical_8_sse2(unsigned char *s, int p, + const unsigned char *blimit, + const unsigned char *limit, + const unsigned char *thresh) { + DECLARE_ALIGNED(8, unsigned char, t_dst[8 * 8]); + unsigned char *src[1]; + unsigned char *dst[1]; + + // Transpose 8x8 + src[0] = s - 4; + dst[0] = t_dst; + + transpose(src, p, dst, 8, 1); + + // Loop filtering + vpx_lpf_horizontal_8_sse2(t_dst + 4 * 8, 8, blimit, limit, thresh); + + src[0] = t_dst; + dst[0] = s - 4; + + // Transpose back + transpose(src, 8, dst, p, 1); +} + +void vpx_lpf_vertical_8_dual_sse2(uint8_t *s, int p, const uint8_t *blimit0, + const uint8_t *limit0, + const uint8_t *thresh0, + const uint8_t *blimit1, + const uint8_t *limit1, + const uint8_t *thresh1) { + DECLARE_ALIGNED(16, unsigned char, t_dst[16 * 8]); + unsigned char *src[2]; + unsigned char *dst[2]; + + // Transpose 8x16 + transpose8x16(s - 4, s - 4 + p * 8, p, t_dst, 16); + + // Loop filtering + vpx_lpf_horizontal_8_dual_sse2(t_dst + 4 * 16, 16, blimit0, limit0, thresh0, + blimit1, limit1, thresh1); + src[0] = t_dst; + src[1] = t_dst + 8; + + dst[0] = s - 4; + dst[1] = s - 4 + p * 8; + + // Transpose back + transpose(src, 16, dst, p, 2); +} + +void vpx_lpf_vertical_16_sse2(unsigned char *s, int p, + const unsigned char *blimit, + const unsigned char *limit, + const unsigned char *thresh) { + DECLARE_ALIGNED(8, unsigned char, t_dst[8 * 16]); + unsigned char *src[2]; + unsigned char *dst[2]; + + src[0] = s - 8; + src[1] = s; + dst[0] = t_dst; + dst[1] = t_dst + 8 * 8; + + // Transpose 16x8 + transpose(src, p, dst, 8, 2); + + // Loop filtering + vpx_lpf_horizontal_edge_8_sse2(t_dst + 8 * 8, 8, blimit, limit, thresh); + + src[0] = t_dst; + src[1] = t_dst + 8 * 8; + dst[0] = s - 8; + dst[1] = s; + + // Transpose back + transpose(src, 8, dst, p, 2); +} + +void vpx_lpf_vertical_16_dual_sse2(unsigned char *s, int p, + const uint8_t *blimit, const uint8_t *limit, + const uint8_t *thresh) { + DECLARE_ALIGNED(16, unsigned char, t_dst[256]); + + // Transpose 16x16 + transpose8x16(s - 8, s - 8 + 8 * p, p, t_dst, 16); + transpose8x16(s, s + 8 * p, p, t_dst + 8 * 16, 16); + + // Loop filtering + vpx_lpf_horizontal_edge_16_sse2(t_dst + 8 * 16, 16, blimit, limit, thresh); + + // Transpose back + transpose8x16(t_dst, t_dst + 8 * 16, 16, s - 8, p); + transpose8x16(t_dst + 8, t_dst + 8 + 8 * 16, 16, s - 8 + 8 * p, p); +} diff --git a/thirdparty/libvpx/vpx_dsp/x86/txfm_common_sse2.h b/thirdparty/libvpx/vpx_dsp/x86/txfm_common_sse2.h new file mode 100644 index 0000000000..536b206876 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/x86/txfm_common_sse2.h @@ -0,0 +1,29 @@ +/* + * Copyright (c) 2015 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VPX_DSP_X86_TXFM_COMMON_SSE2_H_ +#define VPX_DSP_X86_TXFM_COMMON_SSE2_H_ + +#include <emmintrin.h> +#include "vpx/vpx_integer.h" + +#define pair_set_epi16(a, b) \ + _mm_set_epi16((int16_t)(b), (int16_t)(a), (int16_t)(b), (int16_t)(a), \ + (int16_t)(b), (int16_t)(a), (int16_t)(b), (int16_t)(a)) + +#define dual_set_epi16(a, b) \ + _mm_set_epi16((int16_t)(b), (int16_t)(b), (int16_t)(b), (int16_t)(b), \ + (int16_t)(a), (int16_t)(a), (int16_t)(a), (int16_t)(a)) + +#define octa_set_epi16(a, b, c, d, e, f, g, h) \ + _mm_setr_epi16((int16_t)(a), (int16_t)(b), (int16_t)(c), (int16_t)(d), \ + (int16_t)(e), (int16_t)(f), (int16_t)(g), (int16_t)(h)) + +#endif // VPX_DSP_X86_TXFM_COMMON_SSE2_H_ diff --git a/thirdparty/libvpx/vpx_dsp/x86/vpx_asm_stubs.c b/thirdparty/libvpx/vpx_dsp/x86/vpx_asm_stubs.c new file mode 100644 index 0000000000..422b0fc422 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/x86/vpx_asm_stubs.c @@ -0,0 +1,162 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "./vpx_config.h" +#include "./vpx_dsp_rtcd.h" +#include "vpx_dsp/x86/convolve.h" + +#if HAVE_SSE2 +filter8_1dfunction vpx_filter_block1d16_v8_sse2; +filter8_1dfunction vpx_filter_block1d16_h8_sse2; +filter8_1dfunction vpx_filter_block1d8_v8_sse2; +filter8_1dfunction vpx_filter_block1d8_h8_sse2; +filter8_1dfunction vpx_filter_block1d4_v8_sse2; +filter8_1dfunction vpx_filter_block1d4_h8_sse2; +filter8_1dfunction vpx_filter_block1d16_v8_avg_sse2; +filter8_1dfunction vpx_filter_block1d16_h8_avg_sse2; +filter8_1dfunction vpx_filter_block1d8_v8_avg_sse2; +filter8_1dfunction vpx_filter_block1d8_h8_avg_sse2; +filter8_1dfunction vpx_filter_block1d4_v8_avg_sse2; +filter8_1dfunction vpx_filter_block1d4_h8_avg_sse2; + +filter8_1dfunction vpx_filter_block1d16_v2_sse2; +filter8_1dfunction vpx_filter_block1d16_h2_sse2; +filter8_1dfunction vpx_filter_block1d8_v2_sse2; +filter8_1dfunction vpx_filter_block1d8_h2_sse2; +filter8_1dfunction vpx_filter_block1d4_v2_sse2; +filter8_1dfunction vpx_filter_block1d4_h2_sse2; +filter8_1dfunction vpx_filter_block1d16_v2_avg_sse2; +filter8_1dfunction vpx_filter_block1d16_h2_avg_sse2; +filter8_1dfunction vpx_filter_block1d8_v2_avg_sse2; +filter8_1dfunction vpx_filter_block1d8_h2_avg_sse2; +filter8_1dfunction vpx_filter_block1d4_v2_avg_sse2; +filter8_1dfunction vpx_filter_block1d4_h2_avg_sse2; + +// void vpx_convolve8_horiz_sse2(const uint8_t *src, ptrdiff_t src_stride, +// uint8_t *dst, ptrdiff_t dst_stride, +// const int16_t *filter_x, int x_step_q4, +// const int16_t *filter_y, int y_step_q4, +// int w, int h); +// void vpx_convolve8_vert_sse2(const uint8_t *src, ptrdiff_t src_stride, +// uint8_t *dst, ptrdiff_t dst_stride, +// const int16_t *filter_x, int x_step_q4, +// const int16_t *filter_y, int y_step_q4, +// int w, int h); +// void vpx_convolve8_avg_horiz_sse2(const uint8_t *src, ptrdiff_t src_stride, +// uint8_t *dst, ptrdiff_t dst_stride, +// const int16_t *filter_x, int x_step_q4, +// const int16_t *filter_y, int y_step_q4, +// int w, int h); +// void vpx_convolve8_avg_vert_sse2(const uint8_t *src, ptrdiff_t src_stride, +// uint8_t *dst, ptrdiff_t dst_stride, +// const int16_t *filter_x, int x_step_q4, +// const int16_t *filter_y, int y_step_q4, +// int w, int h); +FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , sse2); +FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , sse2); +FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, sse2); +FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_, sse2); + +// void vpx_convolve8_sse2(const uint8_t *src, ptrdiff_t src_stride, +// uint8_t *dst, ptrdiff_t dst_stride, +// const int16_t *filter_x, int x_step_q4, +// const int16_t *filter_y, int y_step_q4, +// int w, int h); +// void vpx_convolve8_avg_sse2(const uint8_t *src, ptrdiff_t src_stride, +// uint8_t *dst, ptrdiff_t dst_stride, +// const int16_t *filter_x, int x_step_q4, +// const int16_t *filter_y, int y_step_q4, +// int w, int h); +FUN_CONV_2D(, sse2); +FUN_CONV_2D(avg_ , sse2); + +#if CONFIG_VP9_HIGHBITDEPTH && ARCH_X86_64 +highbd_filter8_1dfunction vpx_highbd_filter_block1d16_v8_sse2; +highbd_filter8_1dfunction vpx_highbd_filter_block1d16_h8_sse2; +highbd_filter8_1dfunction vpx_highbd_filter_block1d8_v8_sse2; +highbd_filter8_1dfunction vpx_highbd_filter_block1d8_h8_sse2; +highbd_filter8_1dfunction vpx_highbd_filter_block1d4_v8_sse2; +highbd_filter8_1dfunction vpx_highbd_filter_block1d4_h8_sse2; +highbd_filter8_1dfunction vpx_highbd_filter_block1d16_v8_avg_sse2; +highbd_filter8_1dfunction vpx_highbd_filter_block1d16_h8_avg_sse2; +highbd_filter8_1dfunction vpx_highbd_filter_block1d8_v8_avg_sse2; +highbd_filter8_1dfunction vpx_highbd_filter_block1d8_h8_avg_sse2; +highbd_filter8_1dfunction vpx_highbd_filter_block1d4_v8_avg_sse2; +highbd_filter8_1dfunction vpx_highbd_filter_block1d4_h8_avg_sse2; + +highbd_filter8_1dfunction vpx_highbd_filter_block1d16_v2_sse2; +highbd_filter8_1dfunction vpx_highbd_filter_block1d16_h2_sse2; +highbd_filter8_1dfunction vpx_highbd_filter_block1d8_v2_sse2; +highbd_filter8_1dfunction vpx_highbd_filter_block1d8_h2_sse2; +highbd_filter8_1dfunction vpx_highbd_filter_block1d4_v2_sse2; +highbd_filter8_1dfunction vpx_highbd_filter_block1d4_h2_sse2; +highbd_filter8_1dfunction vpx_highbd_filter_block1d16_v2_avg_sse2; +highbd_filter8_1dfunction vpx_highbd_filter_block1d16_h2_avg_sse2; +highbd_filter8_1dfunction vpx_highbd_filter_block1d8_v2_avg_sse2; +highbd_filter8_1dfunction vpx_highbd_filter_block1d8_h2_avg_sse2; +highbd_filter8_1dfunction vpx_highbd_filter_block1d4_v2_avg_sse2; +highbd_filter8_1dfunction vpx_highbd_filter_block1d4_h2_avg_sse2; + +// void vpx_highbd_convolve8_horiz_sse2(const uint8_t *src, +// ptrdiff_t src_stride, +// uint8_t *dst, +// ptrdiff_t dst_stride, +// const int16_t *filter_x, +// int x_step_q4, +// const int16_t *filter_y, +// int y_step_q4, +// int w, int h, int bd); +// void vpx_highbd_convolve8_vert_sse2(const uint8_t *src, +// ptrdiff_t src_stride, +// uint8_t *dst, +// ptrdiff_t dst_stride, +// const int16_t *filter_x, +// int x_step_q4, +// const int16_t *filter_y, +// int y_step_q4, +// int w, int h, int bd); +// void vpx_highbd_convolve8_avg_horiz_sse2(const uint8_t *src, +// ptrdiff_t src_stride, +// uint8_t *dst, +// ptrdiff_t dst_stride, +// const int16_t *filter_x, +// int x_step_q4, +// const int16_t *filter_y, +// int y_step_q4, +// int w, int h, int bd); +// void vpx_highbd_convolve8_avg_vert_sse2(const uint8_t *src, +// ptrdiff_t src_stride, +// uint8_t *dst, +// ptrdiff_t dst_stride, +// const int16_t *filter_x, +// int x_step_q4, +// const int16_t *filter_y, +// int y_step_q4, +// int w, int h, int bd); +HIGH_FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , sse2); +HIGH_FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , sse2); +HIGH_FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, sse2); +HIGH_FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_, + sse2); + +// void vpx_highbd_convolve8_sse2(const uint8_t *src, ptrdiff_t src_stride, +// uint8_t *dst, ptrdiff_t dst_stride, +// const int16_t *filter_x, int x_step_q4, +// const int16_t *filter_y, int y_step_q4, +// int w, int h, int bd); +// void vpx_highbd_convolve8_avg_sse2(const uint8_t *src, ptrdiff_t src_stride, +// uint8_t *dst, ptrdiff_t dst_stride, +// const int16_t *filter_x, int x_step_q4, +// const int16_t *filter_y, int y_step_q4, +// int w, int h, int bd); +HIGH_FUN_CONV_2D(, sse2); +HIGH_FUN_CONV_2D(avg_ , sse2); +#endif // CONFIG_VP9_HIGHBITDEPTH && ARCH_X86_64 +#endif // HAVE_SSE2 diff --git a/thirdparty/libvpx/vpx_dsp/x86/vpx_convolve_copy_sse2.asm b/thirdparty/libvpx/vpx_dsp/x86/vpx_convolve_copy_sse2.asm new file mode 100644 index 0000000000..abc0270655 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/x86/vpx_convolve_copy_sse2.asm @@ -0,0 +1,228 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + +%include "third_party/x86inc/x86inc.asm" + +SECTION .text + +%macro convolve_fn 1-2 +%ifidn %1, avg +%define AUX_XMM_REGS 4 +%else +%define AUX_XMM_REGS 0 +%endif +%ifidn %2, highbd +%define pavg pavgw +cglobal %2_convolve_%1, 4, 7, 4+AUX_XMM_REGS, src, src_stride, \ + dst, dst_stride, \ + fx, fxs, fy, fys, w, h, bd +%else +%define pavg pavgb +cglobal convolve_%1, 4, 7, 4+AUX_XMM_REGS, src, src_stride, \ + dst, dst_stride, \ + fx, fxs, fy, fys, w, h +%endif + mov r4d, dword wm +%ifidn %2, highbd + shl r4d, 1 + shl srcq, 1 + shl src_strideq, 1 + shl dstq, 1 + shl dst_strideq, 1 +%else + cmp r4d, 4 + je .w4 +%endif + cmp r4d, 8 + je .w8 + cmp r4d, 16 + je .w16 + cmp r4d, 32 + je .w32 +%ifidn %2, highbd + cmp r4d, 64 + je .w64 + + mov r4d, dword hm +.loop128: + movu m0, [srcq] + movu m1, [srcq+16] + movu m2, [srcq+32] + movu m3, [srcq+48] +%ifidn %1, avg + pavg m0, [dstq] + pavg m1, [dstq+16] + pavg m2, [dstq+32] + pavg m3, [dstq+48] +%endif + mova [dstq ], m0 + mova [dstq+16], m1 + mova [dstq+32], m2 + mova [dstq+48], m3 + movu m0, [srcq+64] + movu m1, [srcq+80] + movu m2, [srcq+96] + movu m3, [srcq+112] + add srcq, src_strideq +%ifidn %1, avg + pavg m0, [dstq+64] + pavg m1, [dstq+80] + pavg m2, [dstq+96] + pavg m3, [dstq+112] +%endif + mova [dstq+64], m0 + mova [dstq+80], m1 + mova [dstq+96], m2 + mova [dstq+112], m3 + add dstq, dst_strideq + dec r4d + jnz .loop128 + RET +%endif + +.w64 + mov r4d, dword hm +.loop64: + movu m0, [srcq] + movu m1, [srcq+16] + movu m2, [srcq+32] + movu m3, [srcq+48] + add srcq, src_strideq +%ifidn %1, avg + pavg m0, [dstq] + pavg m1, [dstq+16] + pavg m2, [dstq+32] + pavg m3, [dstq+48] +%endif + mova [dstq ], m0 + mova [dstq+16], m1 + mova [dstq+32], m2 + mova [dstq+48], m3 + add dstq, dst_strideq + dec r4d + jnz .loop64 + RET + +.w32: + mov r4d, dword hm +.loop32: + movu m0, [srcq] + movu m1, [srcq+16] + movu m2, [srcq+src_strideq] + movu m3, [srcq+src_strideq+16] + lea srcq, [srcq+src_strideq*2] +%ifidn %1, avg + pavg m0, [dstq] + pavg m1, [dstq +16] + pavg m2, [dstq+dst_strideq] + pavg m3, [dstq+dst_strideq+16] +%endif + mova [dstq ], m0 + mova [dstq +16], m1 + mova [dstq+dst_strideq ], m2 + mova [dstq+dst_strideq+16], m3 + lea dstq, [dstq+dst_strideq*2] + sub r4d, 2 + jnz .loop32 + RET + +.w16: + mov r4d, dword hm + lea r5q, [src_strideq*3] + lea r6q, [dst_strideq*3] +.loop16: + movu m0, [srcq] + movu m1, [srcq+src_strideq] + movu m2, [srcq+src_strideq*2] + movu m3, [srcq+r5q] + lea srcq, [srcq+src_strideq*4] +%ifidn %1, avg + pavg m0, [dstq] + pavg m1, [dstq+dst_strideq] + pavg m2, [dstq+dst_strideq*2] + pavg m3, [dstq+r6q] +%endif + mova [dstq ], m0 + mova [dstq+dst_strideq ], m1 + mova [dstq+dst_strideq*2], m2 + mova [dstq+r6q ], m3 + lea dstq, [dstq+dst_strideq*4] + sub r4d, 4 + jnz .loop16 + RET + +.w8: + mov r4d, dword hm + lea r5q, [src_strideq*3] + lea r6q, [dst_strideq*3] +.loop8: + movh m0, [srcq] + movh m1, [srcq+src_strideq] + movh m2, [srcq+src_strideq*2] + movh m3, [srcq+r5q] + lea srcq, [srcq+src_strideq*4] +%ifidn %1, avg + movh m4, [dstq] + movh m5, [dstq+dst_strideq] + movh m6, [dstq+dst_strideq*2] + movh m7, [dstq+r6q] + pavg m0, m4 + pavg m1, m5 + pavg m2, m6 + pavg m3, m7 +%endif + movh [dstq ], m0 + movh [dstq+dst_strideq ], m1 + movh [dstq+dst_strideq*2], m2 + movh [dstq+r6q ], m3 + lea dstq, [dstq+dst_strideq*4] + sub r4d, 4 + jnz .loop8 + RET + +%ifnidn %2, highbd +.w4: + mov r4d, dword hm + lea r5q, [src_strideq*3] + lea r6q, [dst_strideq*3] +.loop4: + movd m0, [srcq] + movd m1, [srcq+src_strideq] + movd m2, [srcq+src_strideq*2] + movd m3, [srcq+r5q] + lea srcq, [srcq+src_strideq*4] +%ifidn %1, avg + movd m4, [dstq] + movd m5, [dstq+dst_strideq] + movd m6, [dstq+dst_strideq*2] + movd m7, [dstq+r6q] + pavg m0, m4 + pavg m1, m5 + pavg m2, m6 + pavg m3, m7 +%endif + movd [dstq ], m0 + movd [dstq+dst_strideq ], m1 + movd [dstq+dst_strideq*2], m2 + movd [dstq+r6q ], m3 + lea dstq, [dstq+dst_strideq*4] + sub r4d, 4 + jnz .loop4 + RET +%endif +%endmacro + +INIT_XMM sse2 +convolve_fn copy +convolve_fn avg +%if CONFIG_VP9_HIGHBITDEPTH +convolve_fn copy, highbd +convolve_fn avg, highbd +%endif diff --git a/thirdparty/libvpx/vpx_dsp/x86/vpx_subpixel_8t_intrin_avx2.c b/thirdparty/libvpx/vpx_dsp/x86/vpx_subpixel_8t_intrin_avx2.c new file mode 100644 index 0000000000..b718678537 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/x86/vpx_subpixel_8t_intrin_avx2.c @@ -0,0 +1,605 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +// Due to a header conflict between math.h and intrinsics includes with ceil() +// in certain configurations under vs9 this include needs to precede +// immintrin.h. + +#include <immintrin.h> + +#include "./vpx_dsp_rtcd.h" +#include "vpx_dsp/x86/convolve.h" +#include "vpx_ports/mem.h" + +// filters for 16_h8 and 16_v8 +DECLARE_ALIGNED(32, static const uint8_t, filt1_global_avx2[32]) = { + 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, + 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 +}; + +DECLARE_ALIGNED(32, static const uint8_t, filt2_global_avx2[32]) = { + 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, + 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10 +}; + +DECLARE_ALIGNED(32, static const uint8_t, filt3_global_avx2[32]) = { + 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, + 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12 +}; + +DECLARE_ALIGNED(32, static const uint8_t, filt4_global_avx2[32]) = { + 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14, + 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14 +}; + +#if defined(__clang__) +# if __clang_major__ < 3 || (__clang_major__ == 3 && __clang_minor__ <= 3) || \ + (defined(__APPLE__) && \ + ((__clang_major__ == 4 && __clang_minor__ <= 2) || \ + (__clang_major__ == 5 && __clang_minor__ == 0))) + +# define MM256_BROADCASTSI128_SI256(x) \ + _mm_broadcastsi128_si256((__m128i const *)&(x)) +# else // clang > 3.3, and not 5.0 on macosx. +# define MM256_BROADCASTSI128_SI256(x) _mm256_broadcastsi128_si256(x) +# endif // clang <= 3.3 +#elif defined(__GNUC__) +# if __GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ <= 6) +# define MM256_BROADCASTSI128_SI256(x) \ + _mm_broadcastsi128_si256((__m128i const *)&(x)) +# elif __GNUC__ == 4 && __GNUC_MINOR__ == 7 +# define MM256_BROADCASTSI128_SI256(x) _mm_broadcastsi128_si256(x) +# else // gcc > 4.7 +# define MM256_BROADCASTSI128_SI256(x) _mm256_broadcastsi128_si256(x) +# endif // gcc <= 4.6 +#else // !(gcc || clang) +# define MM256_BROADCASTSI128_SI256(x) _mm256_broadcastsi128_si256(x) +#endif // __clang__ + +static void vpx_filter_block1d16_h8_avx2(const uint8_t *src_ptr, + ptrdiff_t src_pixels_per_line, + uint8_t *output_ptr, + ptrdiff_t output_pitch, + uint32_t output_height, + const int16_t *filter) { + __m128i filtersReg; + __m256i addFilterReg64, filt1Reg, filt2Reg, filt3Reg, filt4Reg; + __m256i firstFilters, secondFilters, thirdFilters, forthFilters; + __m256i srcRegFilt32b1_1, srcRegFilt32b2_1, srcRegFilt32b2, srcRegFilt32b3; + __m256i srcReg32b1, srcReg32b2, filtersReg32; + unsigned int i; + ptrdiff_t src_stride, dst_stride; + + // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64 + addFilterReg64 = _mm256_set1_epi32((int)0x0400040u); + filtersReg = _mm_loadu_si128((const __m128i *)filter); + // converting the 16 bit (short) to 8 bit (byte) and have the same data + // in both lanes of 128 bit register. + filtersReg =_mm_packs_epi16(filtersReg, filtersReg); + // have the same data in both lanes of a 256 bit register + filtersReg32 = MM256_BROADCASTSI128_SI256(filtersReg); + + // duplicate only the first 16 bits (first and second byte) + // across 256 bit register + firstFilters = _mm256_shuffle_epi8(filtersReg32, + _mm256_set1_epi16(0x100u)); + // duplicate only the second 16 bits (third and forth byte) + // across 256 bit register + secondFilters = _mm256_shuffle_epi8(filtersReg32, + _mm256_set1_epi16(0x302u)); + // duplicate only the third 16 bits (fifth and sixth byte) + // across 256 bit register + thirdFilters = _mm256_shuffle_epi8(filtersReg32, + _mm256_set1_epi16(0x504u)); + // duplicate only the forth 16 bits (seventh and eighth byte) + // across 256 bit register + forthFilters = _mm256_shuffle_epi8(filtersReg32, + _mm256_set1_epi16(0x706u)); + + filt1Reg = _mm256_load_si256((__m256i const *)filt1_global_avx2); + filt2Reg = _mm256_load_si256((__m256i const *)filt2_global_avx2); + filt3Reg = _mm256_load_si256((__m256i const *)filt3_global_avx2); + filt4Reg = _mm256_load_si256((__m256i const *)filt4_global_avx2); + + // multiple the size of the source and destination stride by two + src_stride = src_pixels_per_line << 1; + dst_stride = output_pitch << 1; + for (i = output_height; i > 1; i-=2) { + // load the 2 strides of source + srcReg32b1 = _mm256_castsi128_si256( + _mm_loadu_si128((const __m128i *)(src_ptr - 3))); + srcReg32b1 = _mm256_inserti128_si256(srcReg32b1, + _mm_loadu_si128((const __m128i *) + (src_ptr+src_pixels_per_line-3)), 1); + + // filter the source buffer + srcRegFilt32b1_1= _mm256_shuffle_epi8(srcReg32b1, filt1Reg); + srcRegFilt32b2= _mm256_shuffle_epi8(srcReg32b1, filt4Reg); + + // multiply 2 adjacent elements with the filter and add the result + srcRegFilt32b1_1 = _mm256_maddubs_epi16(srcRegFilt32b1_1, firstFilters); + srcRegFilt32b2 = _mm256_maddubs_epi16(srcRegFilt32b2, forthFilters); + + // add and saturate the results together + srcRegFilt32b1_1 = _mm256_adds_epi16(srcRegFilt32b1_1, srcRegFilt32b2); + + // filter the source buffer + srcRegFilt32b3= _mm256_shuffle_epi8(srcReg32b1, filt2Reg); + srcRegFilt32b2= _mm256_shuffle_epi8(srcReg32b1, filt3Reg); + + // multiply 2 adjacent elements with the filter and add the result + srcRegFilt32b3 = _mm256_maddubs_epi16(srcRegFilt32b3, secondFilters); + srcRegFilt32b2 = _mm256_maddubs_epi16(srcRegFilt32b2, thirdFilters); + + // add and saturate the results together + srcRegFilt32b1_1 = _mm256_adds_epi16(srcRegFilt32b1_1, + _mm256_min_epi16(srcRegFilt32b3, srcRegFilt32b2)); + + // reading 2 strides of the next 16 bytes + // (part of it was being read by earlier read) + srcReg32b2 = _mm256_castsi128_si256( + _mm_loadu_si128((const __m128i *)(src_ptr + 5))); + srcReg32b2 = _mm256_inserti128_si256(srcReg32b2, + _mm_loadu_si128((const __m128i *) + (src_ptr+src_pixels_per_line+5)), 1); + + // add and saturate the results together + srcRegFilt32b1_1 = _mm256_adds_epi16(srcRegFilt32b1_1, + _mm256_max_epi16(srcRegFilt32b3, srcRegFilt32b2)); + + // filter the source buffer + srcRegFilt32b2_1 = _mm256_shuffle_epi8(srcReg32b2, filt1Reg); + srcRegFilt32b2 = _mm256_shuffle_epi8(srcReg32b2, filt4Reg); + + // multiply 2 adjacent elements with the filter and add the result + srcRegFilt32b2_1 = _mm256_maddubs_epi16(srcRegFilt32b2_1, firstFilters); + srcRegFilt32b2 = _mm256_maddubs_epi16(srcRegFilt32b2, forthFilters); + + // add and saturate the results together + srcRegFilt32b2_1 = _mm256_adds_epi16(srcRegFilt32b2_1, srcRegFilt32b2); + + // filter the source buffer + srcRegFilt32b3= _mm256_shuffle_epi8(srcReg32b2, filt2Reg); + srcRegFilt32b2= _mm256_shuffle_epi8(srcReg32b2, filt3Reg); + + // multiply 2 adjacent elements with the filter and add the result + srcRegFilt32b3 = _mm256_maddubs_epi16(srcRegFilt32b3, secondFilters); + srcRegFilt32b2 = _mm256_maddubs_epi16(srcRegFilt32b2, thirdFilters); + + // add and saturate the results together + srcRegFilt32b2_1 = _mm256_adds_epi16(srcRegFilt32b2_1, + _mm256_min_epi16(srcRegFilt32b3, srcRegFilt32b2)); + srcRegFilt32b2_1 = _mm256_adds_epi16(srcRegFilt32b2_1, + _mm256_max_epi16(srcRegFilt32b3, srcRegFilt32b2)); + + + srcRegFilt32b1_1 = _mm256_adds_epi16(srcRegFilt32b1_1, addFilterReg64); + + srcRegFilt32b2_1 = _mm256_adds_epi16(srcRegFilt32b2_1, addFilterReg64); + + // shift by 7 bit each 16 bit + srcRegFilt32b1_1 = _mm256_srai_epi16(srcRegFilt32b1_1, 7); + srcRegFilt32b2_1 = _mm256_srai_epi16(srcRegFilt32b2_1, 7); + + // shrink to 8 bit each 16 bits, the first lane contain the first + // convolve result and the second lane contain the second convolve + // result + srcRegFilt32b1_1 = _mm256_packus_epi16(srcRegFilt32b1_1, + srcRegFilt32b2_1); + + src_ptr+=src_stride; + + // save 16 bytes + _mm_store_si128((__m128i*)output_ptr, + _mm256_castsi256_si128(srcRegFilt32b1_1)); + + // save the next 16 bits + _mm_store_si128((__m128i*)(output_ptr+output_pitch), + _mm256_extractf128_si256(srcRegFilt32b1_1, 1)); + output_ptr+=dst_stride; + } + + // if the number of strides is odd. + // process only 16 bytes + if (i > 0) { + __m128i srcReg1, srcReg2, srcRegFilt1_1, srcRegFilt2_1; + __m128i srcRegFilt2, srcRegFilt3; + + srcReg1 = _mm_loadu_si128((const __m128i *)(src_ptr - 3)); + + // filter the source buffer + srcRegFilt1_1 = _mm_shuffle_epi8(srcReg1, + _mm256_castsi256_si128(filt1Reg)); + srcRegFilt2 = _mm_shuffle_epi8(srcReg1, + _mm256_castsi256_si128(filt4Reg)); + + // multiply 2 adjacent elements with the filter and add the result + srcRegFilt1_1 = _mm_maddubs_epi16(srcRegFilt1_1, + _mm256_castsi256_si128(firstFilters)); + srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, + _mm256_castsi256_si128(forthFilters)); + + // add and saturate the results together + srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1, srcRegFilt2); + + // filter the source buffer + srcRegFilt3= _mm_shuffle_epi8(srcReg1, + _mm256_castsi256_si128(filt2Reg)); + srcRegFilt2= _mm_shuffle_epi8(srcReg1, + _mm256_castsi256_si128(filt3Reg)); + + // multiply 2 adjacent elements with the filter and add the result + srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, + _mm256_castsi256_si128(secondFilters)); + srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, + _mm256_castsi256_si128(thirdFilters)); + + // add and saturate the results together + srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1, + _mm_min_epi16(srcRegFilt3, srcRegFilt2)); + + // reading the next 16 bytes + // (part of it was being read by earlier read) + srcReg2 = _mm_loadu_si128((const __m128i *)(src_ptr + 5)); + + // add and saturate the results together + srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1, + _mm_max_epi16(srcRegFilt3, srcRegFilt2)); + + // filter the source buffer + srcRegFilt2_1 = _mm_shuffle_epi8(srcReg2, + _mm256_castsi256_si128(filt1Reg)); + srcRegFilt2 = _mm_shuffle_epi8(srcReg2, + _mm256_castsi256_si128(filt4Reg)); + + // multiply 2 adjacent elements with the filter and add the result + srcRegFilt2_1 = _mm_maddubs_epi16(srcRegFilt2_1, + _mm256_castsi256_si128(firstFilters)); + srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, + _mm256_castsi256_si128(forthFilters)); + + // add and saturate the results together + srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1, srcRegFilt2); + + // filter the source buffer + srcRegFilt3 = _mm_shuffle_epi8(srcReg2, + _mm256_castsi256_si128(filt2Reg)); + srcRegFilt2 = _mm_shuffle_epi8(srcReg2, + _mm256_castsi256_si128(filt3Reg)); + + // multiply 2 adjacent elements with the filter and add the result + srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, + _mm256_castsi256_si128(secondFilters)); + srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, + _mm256_castsi256_si128(thirdFilters)); + + // add and saturate the results together + srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1, + _mm_min_epi16(srcRegFilt3, srcRegFilt2)); + srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1, + _mm_max_epi16(srcRegFilt3, srcRegFilt2)); + + + srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1, + _mm256_castsi256_si128(addFilterReg64)); + + srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1, + _mm256_castsi256_si128(addFilterReg64)); + + // shift by 7 bit each 16 bit + srcRegFilt1_1 = _mm_srai_epi16(srcRegFilt1_1, 7); + srcRegFilt2_1 = _mm_srai_epi16(srcRegFilt2_1, 7); + + // shrink to 8 bit each 16 bits, the first lane contain the first + // convolve result and the second lane contain the second convolve + // result + srcRegFilt1_1 = _mm_packus_epi16(srcRegFilt1_1, srcRegFilt2_1); + + // save 16 bytes + _mm_store_si128((__m128i*)output_ptr, srcRegFilt1_1); + } +} + +static void vpx_filter_block1d16_v8_avx2(const uint8_t *src_ptr, + ptrdiff_t src_pitch, + uint8_t *output_ptr, + ptrdiff_t out_pitch, + uint32_t output_height, + const int16_t *filter) { + __m128i filtersReg; + __m256i addFilterReg64; + __m256i srcReg32b1, srcReg32b2, srcReg32b3, srcReg32b4, srcReg32b5; + __m256i srcReg32b6, srcReg32b7, srcReg32b8, srcReg32b9, srcReg32b10; + __m256i srcReg32b11, srcReg32b12, filtersReg32; + __m256i firstFilters, secondFilters, thirdFilters, forthFilters; + unsigned int i; + ptrdiff_t src_stride, dst_stride; + + // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64 + addFilterReg64 = _mm256_set1_epi32((int)0x0400040u); + filtersReg = _mm_loadu_si128((const __m128i *)filter); + // converting the 16 bit (short) to 8 bit (byte) and have the + // same data in both lanes of 128 bit register. + filtersReg =_mm_packs_epi16(filtersReg, filtersReg); + // have the same data in both lanes of a 256 bit register + filtersReg32 = MM256_BROADCASTSI128_SI256(filtersReg); + + // duplicate only the first 16 bits (first and second byte) + // across 256 bit register + firstFilters = _mm256_shuffle_epi8(filtersReg32, + _mm256_set1_epi16(0x100u)); + // duplicate only the second 16 bits (third and forth byte) + // across 256 bit register + secondFilters = _mm256_shuffle_epi8(filtersReg32, + _mm256_set1_epi16(0x302u)); + // duplicate only the third 16 bits (fifth and sixth byte) + // across 256 bit register + thirdFilters = _mm256_shuffle_epi8(filtersReg32, + _mm256_set1_epi16(0x504u)); + // duplicate only the forth 16 bits (seventh and eighth byte) + // across 256 bit register + forthFilters = _mm256_shuffle_epi8(filtersReg32, + _mm256_set1_epi16(0x706u)); + + // multiple the size of the source and destination stride by two + src_stride = src_pitch << 1; + dst_stride = out_pitch << 1; + + // load 16 bytes 7 times in stride of src_pitch + srcReg32b1 = _mm256_castsi128_si256( + _mm_loadu_si128((const __m128i *)(src_ptr))); + srcReg32b2 = _mm256_castsi128_si256( + _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch))); + srcReg32b3 = _mm256_castsi128_si256( + _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 2))); + srcReg32b4 = _mm256_castsi128_si256( + _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 3))); + srcReg32b5 = _mm256_castsi128_si256( + _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 4))); + srcReg32b6 = _mm256_castsi128_si256( + _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 5))); + srcReg32b7 = _mm256_castsi128_si256( + _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 6))); + + // have each consecutive loads on the same 256 register + srcReg32b1 = _mm256_inserti128_si256(srcReg32b1, + _mm256_castsi256_si128(srcReg32b2), 1); + srcReg32b2 = _mm256_inserti128_si256(srcReg32b2, + _mm256_castsi256_si128(srcReg32b3), 1); + srcReg32b3 = _mm256_inserti128_si256(srcReg32b3, + _mm256_castsi256_si128(srcReg32b4), 1); + srcReg32b4 = _mm256_inserti128_si256(srcReg32b4, + _mm256_castsi256_si128(srcReg32b5), 1); + srcReg32b5 = _mm256_inserti128_si256(srcReg32b5, + _mm256_castsi256_si128(srcReg32b6), 1); + srcReg32b6 = _mm256_inserti128_si256(srcReg32b6, + _mm256_castsi256_si128(srcReg32b7), 1); + + // merge every two consecutive registers except the last one + srcReg32b10 = _mm256_unpacklo_epi8(srcReg32b1, srcReg32b2); + srcReg32b1 = _mm256_unpackhi_epi8(srcReg32b1, srcReg32b2); + + // save + srcReg32b11 = _mm256_unpacklo_epi8(srcReg32b3, srcReg32b4); + + // save + srcReg32b3 = _mm256_unpackhi_epi8(srcReg32b3, srcReg32b4); + + // save + srcReg32b2 = _mm256_unpacklo_epi8(srcReg32b5, srcReg32b6); + + // save + srcReg32b5 = _mm256_unpackhi_epi8(srcReg32b5, srcReg32b6); + + + for (i = output_height; i > 1; i-=2) { + // load the last 2 loads of 16 bytes and have every two + // consecutive loads in the same 256 bit register + srcReg32b8 = _mm256_castsi128_si256( + _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 7))); + srcReg32b7 = _mm256_inserti128_si256(srcReg32b7, + _mm256_castsi256_si128(srcReg32b8), 1); + srcReg32b9 = _mm256_castsi128_si256( + _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 8))); + srcReg32b8 = _mm256_inserti128_si256(srcReg32b8, + _mm256_castsi256_si128(srcReg32b9), 1); + + // merge every two consecutive registers + // save + srcReg32b4 = _mm256_unpacklo_epi8(srcReg32b7, srcReg32b8); + srcReg32b7 = _mm256_unpackhi_epi8(srcReg32b7, srcReg32b8); + + // multiply 2 adjacent elements with the filter and add the result + srcReg32b10 = _mm256_maddubs_epi16(srcReg32b10, firstFilters); + srcReg32b6 = _mm256_maddubs_epi16(srcReg32b4, forthFilters); + + // add and saturate the results together + srcReg32b10 = _mm256_adds_epi16(srcReg32b10, srcReg32b6); + + // multiply 2 adjacent elements with the filter and add the result + srcReg32b8 = _mm256_maddubs_epi16(srcReg32b11, secondFilters); + srcReg32b12 = _mm256_maddubs_epi16(srcReg32b2, thirdFilters); + + // add and saturate the results together + srcReg32b10 = _mm256_adds_epi16(srcReg32b10, + _mm256_min_epi16(srcReg32b8, srcReg32b12)); + srcReg32b10 = _mm256_adds_epi16(srcReg32b10, + _mm256_max_epi16(srcReg32b8, srcReg32b12)); + + // multiply 2 adjacent elements with the filter and add the result + srcReg32b1 = _mm256_maddubs_epi16(srcReg32b1, firstFilters); + srcReg32b6 = _mm256_maddubs_epi16(srcReg32b7, forthFilters); + + srcReg32b1 = _mm256_adds_epi16(srcReg32b1, srcReg32b6); + + // multiply 2 adjacent elements with the filter and add the result + srcReg32b8 = _mm256_maddubs_epi16(srcReg32b3, secondFilters); + srcReg32b12 = _mm256_maddubs_epi16(srcReg32b5, thirdFilters); + + // add and saturate the results together + srcReg32b1 = _mm256_adds_epi16(srcReg32b1, + _mm256_min_epi16(srcReg32b8, srcReg32b12)); + srcReg32b1 = _mm256_adds_epi16(srcReg32b1, + _mm256_max_epi16(srcReg32b8, srcReg32b12)); + + srcReg32b10 = _mm256_adds_epi16(srcReg32b10, addFilterReg64); + srcReg32b1 = _mm256_adds_epi16(srcReg32b1, addFilterReg64); + + // shift by 7 bit each 16 bit + srcReg32b10 = _mm256_srai_epi16(srcReg32b10, 7); + srcReg32b1 = _mm256_srai_epi16(srcReg32b1, 7); + + // shrink to 8 bit each 16 bits, the first lane contain the first + // convolve result and the second lane contain the second convolve + // result + srcReg32b1 = _mm256_packus_epi16(srcReg32b10, srcReg32b1); + + src_ptr+=src_stride; + + // save 16 bytes + _mm_store_si128((__m128i*)output_ptr, + _mm256_castsi256_si128(srcReg32b1)); + + // save the next 16 bits + _mm_store_si128((__m128i*)(output_ptr+out_pitch), + _mm256_extractf128_si256(srcReg32b1, 1)); + + output_ptr+=dst_stride; + + // save part of the registers for next strides + srcReg32b10 = srcReg32b11; + srcReg32b1 = srcReg32b3; + srcReg32b11 = srcReg32b2; + srcReg32b3 = srcReg32b5; + srcReg32b2 = srcReg32b4; + srcReg32b5 = srcReg32b7; + srcReg32b7 = srcReg32b9; + } + if (i > 0) { + __m128i srcRegFilt1, srcRegFilt3, srcRegFilt4, srcRegFilt5; + __m128i srcRegFilt6, srcRegFilt7, srcRegFilt8; + // load the last 16 bytes + srcRegFilt8 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 7)); + + // merge the last 2 results together + srcRegFilt4 = _mm_unpacklo_epi8( + _mm256_castsi256_si128(srcReg32b7), srcRegFilt8); + srcRegFilt7 = _mm_unpackhi_epi8( + _mm256_castsi256_si128(srcReg32b7), srcRegFilt8); + + // multiply 2 adjacent elements with the filter and add the result + srcRegFilt1 = _mm_maddubs_epi16(_mm256_castsi256_si128(srcReg32b10), + _mm256_castsi256_si128(firstFilters)); + srcRegFilt4 = _mm_maddubs_epi16(srcRegFilt4, + _mm256_castsi256_si128(forthFilters)); + srcRegFilt3 = _mm_maddubs_epi16(_mm256_castsi256_si128(srcReg32b1), + _mm256_castsi256_si128(firstFilters)); + srcRegFilt7 = _mm_maddubs_epi16(srcRegFilt7, + _mm256_castsi256_si128(forthFilters)); + + // add and saturate the results together + srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt4); + srcRegFilt3 = _mm_adds_epi16(srcRegFilt3, srcRegFilt7); + + + // multiply 2 adjacent elements with the filter and add the result + srcRegFilt4 = _mm_maddubs_epi16(_mm256_castsi256_si128(srcReg32b11), + _mm256_castsi256_si128(secondFilters)); + srcRegFilt5 = _mm_maddubs_epi16(_mm256_castsi256_si128(srcReg32b3), + _mm256_castsi256_si128(secondFilters)); + + // multiply 2 adjacent elements with the filter and add the result + srcRegFilt6 = _mm_maddubs_epi16(_mm256_castsi256_si128(srcReg32b2), + _mm256_castsi256_si128(thirdFilters)); + srcRegFilt7 = _mm_maddubs_epi16(_mm256_castsi256_si128(srcReg32b5), + _mm256_castsi256_si128(thirdFilters)); + + // add and saturate the results together + srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, + _mm_min_epi16(srcRegFilt4, srcRegFilt6)); + srcRegFilt3 = _mm_adds_epi16(srcRegFilt3, + _mm_min_epi16(srcRegFilt5, srcRegFilt7)); + + // add and saturate the results together + srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, + _mm_max_epi16(srcRegFilt4, srcRegFilt6)); + srcRegFilt3 = _mm_adds_epi16(srcRegFilt3, + _mm_max_epi16(srcRegFilt5, srcRegFilt7)); + + + srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, + _mm256_castsi256_si128(addFilterReg64)); + srcRegFilt3 = _mm_adds_epi16(srcRegFilt3, + _mm256_castsi256_si128(addFilterReg64)); + + // shift by 7 bit each 16 bit + srcRegFilt1 = _mm_srai_epi16(srcRegFilt1, 7); + srcRegFilt3 = _mm_srai_epi16(srcRegFilt3, 7); + + // shrink to 8 bit each 16 bits, the first lane contain the first + // convolve result and the second lane contain the second convolve + // result + srcRegFilt1 = _mm_packus_epi16(srcRegFilt1, srcRegFilt3); + + // save 16 bytes + _mm_store_si128((__m128i*)output_ptr, srcRegFilt1); + } +} + +#if HAVE_AVX2 && HAVE_SSSE3 +filter8_1dfunction vpx_filter_block1d4_v8_ssse3; +#if ARCH_X86_64 +filter8_1dfunction vpx_filter_block1d8_v8_intrin_ssse3; +filter8_1dfunction vpx_filter_block1d8_h8_intrin_ssse3; +filter8_1dfunction vpx_filter_block1d4_h8_intrin_ssse3; +#define vpx_filter_block1d8_v8_avx2 vpx_filter_block1d8_v8_intrin_ssse3 +#define vpx_filter_block1d8_h8_avx2 vpx_filter_block1d8_h8_intrin_ssse3 +#define vpx_filter_block1d4_h8_avx2 vpx_filter_block1d4_h8_intrin_ssse3 +#else // ARCH_X86 +filter8_1dfunction vpx_filter_block1d8_v8_ssse3; +filter8_1dfunction vpx_filter_block1d8_h8_ssse3; +filter8_1dfunction vpx_filter_block1d4_h8_ssse3; +#define vpx_filter_block1d8_v8_avx2 vpx_filter_block1d8_v8_ssse3 +#define vpx_filter_block1d8_h8_avx2 vpx_filter_block1d8_h8_ssse3 +#define vpx_filter_block1d4_h8_avx2 vpx_filter_block1d4_h8_ssse3 +#endif // ARCH_X86_64 +filter8_1dfunction vpx_filter_block1d16_v2_ssse3; +filter8_1dfunction vpx_filter_block1d16_h2_ssse3; +filter8_1dfunction vpx_filter_block1d8_v2_ssse3; +filter8_1dfunction vpx_filter_block1d8_h2_ssse3; +filter8_1dfunction vpx_filter_block1d4_v2_ssse3; +filter8_1dfunction vpx_filter_block1d4_h2_ssse3; +#define vpx_filter_block1d4_v8_avx2 vpx_filter_block1d4_v8_ssse3 +#define vpx_filter_block1d16_v2_avx2 vpx_filter_block1d16_v2_ssse3 +#define vpx_filter_block1d16_h2_avx2 vpx_filter_block1d16_h2_ssse3 +#define vpx_filter_block1d8_v2_avx2 vpx_filter_block1d8_v2_ssse3 +#define vpx_filter_block1d8_h2_avx2 vpx_filter_block1d8_h2_ssse3 +#define vpx_filter_block1d4_v2_avx2 vpx_filter_block1d4_v2_ssse3 +#define vpx_filter_block1d4_h2_avx2 vpx_filter_block1d4_h2_ssse3 +// void vpx_convolve8_horiz_avx2(const uint8_t *src, ptrdiff_t src_stride, +// uint8_t *dst, ptrdiff_t dst_stride, +// const int16_t *filter_x, int x_step_q4, +// const int16_t *filter_y, int y_step_q4, +// int w, int h); +// void vpx_convolve8_vert_avx2(const uint8_t *src, ptrdiff_t src_stride, +// uint8_t *dst, ptrdiff_t dst_stride, +// const int16_t *filter_x, int x_step_q4, +// const int16_t *filter_y, int y_step_q4, +// int w, int h); +FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , avx2); +FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , avx2); + +// void vpx_convolve8_avx2(const uint8_t *src, ptrdiff_t src_stride, +// uint8_t *dst, ptrdiff_t dst_stride, +// const int16_t *filter_x, int x_step_q4, +// const int16_t *filter_y, int y_step_q4, +// int w, int h); +FUN_CONV_2D(, avx2); +#endif // HAVE_AX2 && HAVE_SSSE3 diff --git a/thirdparty/libvpx/vpx_dsp/x86/vpx_subpixel_8t_intrin_ssse3.c b/thirdparty/libvpx/vpx_dsp/x86/vpx_subpixel_8t_intrin_ssse3.c new file mode 100644 index 0000000000..6fd52087c7 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/x86/vpx_subpixel_8t_intrin_ssse3.c @@ -0,0 +1,915 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +// Due to a header conflict between math.h and intrinsics includes with ceil() +// in certain configurations under vs9 this include needs to precede +// tmmintrin.h. + +#include <tmmintrin.h> + +#include "./vpx_dsp_rtcd.h" +#include "vpx_dsp/vpx_filter.h" +#include "vpx_dsp/x86/convolve.h" +#include "vpx_mem/vpx_mem.h" +#include "vpx_ports/mem.h" +#include "vpx_ports/emmintrin_compat.h" + +// filters only for the 4_h8 convolution +DECLARE_ALIGNED(16, static const uint8_t, filt1_4_h8[16]) = { + 0, 1, 1, 2, 2, 3, 3, 4, 2, 3, 3, 4, 4, 5, 5, 6 +}; + +DECLARE_ALIGNED(16, static const uint8_t, filt2_4_h8[16]) = { + 4, 5, 5, 6, 6, 7, 7, 8, 6, 7, 7, 8, 8, 9, 9, 10 +}; + +// filters for 8_h8 and 16_h8 +DECLARE_ALIGNED(16, static const uint8_t, filt1_global[16]) = { + 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 +}; + +DECLARE_ALIGNED(16, static const uint8_t, filt2_global[16]) = { + 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10 +}; + +DECLARE_ALIGNED(16, static const uint8_t, filt3_global[16]) = { + 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12 +}; + +DECLARE_ALIGNED(16, static const uint8_t, filt4_global[16]) = { + 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14 +}; + +// These are reused by the avx2 intrinsics. +filter8_1dfunction vpx_filter_block1d8_v8_intrin_ssse3; +filter8_1dfunction vpx_filter_block1d8_h8_intrin_ssse3; +filter8_1dfunction vpx_filter_block1d4_h8_intrin_ssse3; + +void vpx_filter_block1d4_h8_intrin_ssse3(const uint8_t *src_ptr, + ptrdiff_t src_pixels_per_line, + uint8_t *output_ptr, + ptrdiff_t output_pitch, + uint32_t output_height, + const int16_t *filter) { + __m128i firstFilters, secondFilters, shuffle1, shuffle2; + __m128i srcRegFilt1, srcRegFilt2, srcRegFilt3, srcRegFilt4; + __m128i addFilterReg64, filtersReg, srcReg, minReg; + unsigned int i; + + // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64 + addFilterReg64 =_mm_set1_epi32((int)0x0400040u); + filtersReg = _mm_loadu_si128((const __m128i *)filter); + // converting the 16 bit (short) to 8 bit (byte) and have the same data + // in both lanes of 128 bit register. + filtersReg =_mm_packs_epi16(filtersReg, filtersReg); + + // duplicate only the first 16 bits in the filter into the first lane + firstFilters = _mm_shufflelo_epi16(filtersReg, 0); + // duplicate only the third 16 bit in the filter into the first lane + secondFilters = _mm_shufflelo_epi16(filtersReg, 0xAAu); + // duplicate only the seconds 16 bits in the filter into the second lane + // firstFilters: k0 k1 k0 k1 k0 k1 k0 k1 k2 k3 k2 k3 k2 k3 k2 k3 + firstFilters = _mm_shufflehi_epi16(firstFilters, 0x55u); + // duplicate only the forth 16 bits in the filter into the second lane + // secondFilters: k4 k5 k4 k5 k4 k5 k4 k5 k6 k7 k6 k7 k6 k7 k6 k7 + secondFilters = _mm_shufflehi_epi16(secondFilters, 0xFFu); + + // loading the local filters + shuffle1 =_mm_load_si128((__m128i const *)filt1_4_h8); + shuffle2 = _mm_load_si128((__m128i const *)filt2_4_h8); + + for (i = 0; i < output_height; i++) { + srcReg = _mm_loadu_si128((const __m128i *)(src_ptr - 3)); + + // filter the source buffer + srcRegFilt1= _mm_shuffle_epi8(srcReg, shuffle1); + srcRegFilt2= _mm_shuffle_epi8(srcReg, shuffle2); + + // multiply 2 adjacent elements with the filter and add the result + srcRegFilt1 = _mm_maddubs_epi16(srcRegFilt1, firstFilters); + srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, secondFilters); + + // extract the higher half of the lane + srcRegFilt3 = _mm_srli_si128(srcRegFilt1, 8); + srcRegFilt4 = _mm_srli_si128(srcRegFilt2, 8); + + minReg = _mm_min_epi16(srcRegFilt3, srcRegFilt2); + + // add and saturate all the results together + srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt4); + srcRegFilt3 = _mm_max_epi16(srcRegFilt3, srcRegFilt2); + srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, minReg); + srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt3); + srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, addFilterReg64); + + // shift by 7 bit each 16 bits + srcRegFilt1 = _mm_srai_epi16(srcRegFilt1, 7); + + // shrink to 8 bit each 16 bits + srcRegFilt1 = _mm_packus_epi16(srcRegFilt1, srcRegFilt1); + src_ptr+=src_pixels_per_line; + + // save only 4 bytes + *((int*)&output_ptr[0])= _mm_cvtsi128_si32(srcRegFilt1); + + output_ptr+=output_pitch; + } +} + +void vpx_filter_block1d8_h8_intrin_ssse3(const uint8_t *src_ptr, + ptrdiff_t src_pixels_per_line, + uint8_t *output_ptr, + ptrdiff_t output_pitch, + uint32_t output_height, + const int16_t *filter) { + __m128i firstFilters, secondFilters, thirdFilters, forthFilters, srcReg; + __m128i filt1Reg, filt2Reg, filt3Reg, filt4Reg; + __m128i srcRegFilt1, srcRegFilt2, srcRegFilt3, srcRegFilt4; + __m128i addFilterReg64, filtersReg, minReg; + unsigned int i; + + // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64 + addFilterReg64 = _mm_set1_epi32((int)0x0400040u); + filtersReg = _mm_loadu_si128((const __m128i *)filter); + // converting the 16 bit (short) to 8 bit (byte) and have the same data + // in both lanes of 128 bit register. + filtersReg =_mm_packs_epi16(filtersReg, filtersReg); + + // duplicate only the first 16 bits (first and second byte) + // across 128 bit register + firstFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x100u)); + // duplicate only the second 16 bits (third and forth byte) + // across 128 bit register + secondFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x302u)); + // duplicate only the third 16 bits (fifth and sixth byte) + // across 128 bit register + thirdFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x504u)); + // duplicate only the forth 16 bits (seventh and eighth byte) + // across 128 bit register + forthFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x706u)); + + filt1Reg = _mm_load_si128((__m128i const *)filt1_global); + filt2Reg = _mm_load_si128((__m128i const *)filt2_global); + filt3Reg = _mm_load_si128((__m128i const *)filt3_global); + filt4Reg = _mm_load_si128((__m128i const *)filt4_global); + + for (i = 0; i < output_height; i++) { + srcReg = _mm_loadu_si128((const __m128i *)(src_ptr - 3)); + + // filter the source buffer + srcRegFilt1= _mm_shuffle_epi8(srcReg, filt1Reg); + srcRegFilt2= _mm_shuffle_epi8(srcReg, filt2Reg); + + // multiply 2 adjacent elements with the filter and add the result + srcRegFilt1 = _mm_maddubs_epi16(srcRegFilt1, firstFilters); + srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, secondFilters); + + // filter the source buffer + srcRegFilt3= _mm_shuffle_epi8(srcReg, filt3Reg); + srcRegFilt4= _mm_shuffle_epi8(srcReg, filt4Reg); + + // multiply 2 adjacent elements with the filter and add the result + srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, thirdFilters); + srcRegFilt4 = _mm_maddubs_epi16(srcRegFilt4, forthFilters); + + // add and saturate all the results together + minReg = _mm_min_epi16(srcRegFilt2, srcRegFilt3); + srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt4); + + srcRegFilt2= _mm_max_epi16(srcRegFilt2, srcRegFilt3); + srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, minReg); + srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt2); + srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, addFilterReg64); + + // shift by 7 bit each 16 bits + srcRegFilt1 = _mm_srai_epi16(srcRegFilt1, 7); + + // shrink to 8 bit each 16 bits + srcRegFilt1 = _mm_packus_epi16(srcRegFilt1, srcRegFilt1); + + src_ptr+=src_pixels_per_line; + + // save only 8 bytes + _mm_storel_epi64((__m128i*)&output_ptr[0], srcRegFilt1); + + output_ptr+=output_pitch; + } +} + +void vpx_filter_block1d8_v8_intrin_ssse3(const uint8_t *src_ptr, + ptrdiff_t src_pitch, + uint8_t *output_ptr, + ptrdiff_t out_pitch, + uint32_t output_height, + const int16_t *filter) { + __m128i addFilterReg64, filtersReg, minReg; + __m128i firstFilters, secondFilters, thirdFilters, forthFilters; + __m128i srcRegFilt1, srcRegFilt2, srcRegFilt3, srcRegFilt5; + __m128i srcReg1, srcReg2, srcReg3, srcReg4, srcReg5, srcReg6, srcReg7; + __m128i srcReg8; + unsigned int i; + + // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64 + addFilterReg64 = _mm_set1_epi32((int)0x0400040u); + filtersReg = _mm_loadu_si128((const __m128i *)filter); + // converting the 16 bit (short) to 8 bit (byte) and have the same data + // in both lanes of 128 bit register. + filtersReg =_mm_packs_epi16(filtersReg, filtersReg); + + // duplicate only the first 16 bits in the filter + firstFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x100u)); + // duplicate only the second 16 bits in the filter + secondFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x302u)); + // duplicate only the third 16 bits in the filter + thirdFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x504u)); + // duplicate only the forth 16 bits in the filter + forthFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x706u)); + + // load the first 7 rows of 8 bytes + srcReg1 = _mm_loadl_epi64((const __m128i *)src_ptr); + srcReg2 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch)); + srcReg3 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 2)); + srcReg4 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 3)); + srcReg5 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 4)); + srcReg6 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 5)); + srcReg7 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 6)); + + for (i = 0; i < output_height; i++) { + // load the last 8 bytes + srcReg8 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 7)); + + // merge the result together + srcRegFilt1 = _mm_unpacklo_epi8(srcReg1, srcReg2); + srcRegFilt3 = _mm_unpacklo_epi8(srcReg3, srcReg4); + + // merge the result together + srcRegFilt2 = _mm_unpacklo_epi8(srcReg5, srcReg6); + srcRegFilt5 = _mm_unpacklo_epi8(srcReg7, srcReg8); + + // multiply 2 adjacent elements with the filter and add the result + srcRegFilt1 = _mm_maddubs_epi16(srcRegFilt1, firstFilters); + srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, secondFilters); + srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, thirdFilters); + srcRegFilt5 = _mm_maddubs_epi16(srcRegFilt5, forthFilters); + + // add and saturate the results together + minReg = _mm_min_epi16(srcRegFilt2, srcRegFilt3); + srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt5); + srcRegFilt2 = _mm_max_epi16(srcRegFilt2, srcRegFilt3); + srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, minReg); + srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt2); + srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, addFilterReg64); + + // shift by 7 bit each 16 bit + srcRegFilt1 = _mm_srai_epi16(srcRegFilt1, 7); + + // shrink to 8 bit each 16 bits + srcRegFilt1 = _mm_packus_epi16(srcRegFilt1, srcRegFilt1); + + src_ptr+=src_pitch; + + // shift down a row + srcReg1 = srcReg2; + srcReg2 = srcReg3; + srcReg3 = srcReg4; + srcReg4 = srcReg5; + srcReg5 = srcReg6; + srcReg6 = srcReg7; + srcReg7 = srcReg8; + + // save only 8 bytes convolve result + _mm_storel_epi64((__m128i*)&output_ptr[0], srcRegFilt1); + + output_ptr+=out_pitch; + } +} + +filter8_1dfunction vpx_filter_block1d16_v8_ssse3; +filter8_1dfunction vpx_filter_block1d16_h8_ssse3; +filter8_1dfunction vpx_filter_block1d8_v8_ssse3; +filter8_1dfunction vpx_filter_block1d8_h8_ssse3; +filter8_1dfunction vpx_filter_block1d4_v8_ssse3; +filter8_1dfunction vpx_filter_block1d4_h8_ssse3; +filter8_1dfunction vpx_filter_block1d16_v8_avg_ssse3; +filter8_1dfunction vpx_filter_block1d16_h8_avg_ssse3; +filter8_1dfunction vpx_filter_block1d8_v8_avg_ssse3; +filter8_1dfunction vpx_filter_block1d8_h8_avg_ssse3; +filter8_1dfunction vpx_filter_block1d4_v8_avg_ssse3; +filter8_1dfunction vpx_filter_block1d4_h8_avg_ssse3; + +filter8_1dfunction vpx_filter_block1d16_v2_ssse3; +filter8_1dfunction vpx_filter_block1d16_h2_ssse3; +filter8_1dfunction vpx_filter_block1d8_v2_ssse3; +filter8_1dfunction vpx_filter_block1d8_h2_ssse3; +filter8_1dfunction vpx_filter_block1d4_v2_ssse3; +filter8_1dfunction vpx_filter_block1d4_h2_ssse3; +filter8_1dfunction vpx_filter_block1d16_v2_avg_ssse3; +filter8_1dfunction vpx_filter_block1d16_h2_avg_ssse3; +filter8_1dfunction vpx_filter_block1d8_v2_avg_ssse3; +filter8_1dfunction vpx_filter_block1d8_h2_avg_ssse3; +filter8_1dfunction vpx_filter_block1d4_v2_avg_ssse3; +filter8_1dfunction vpx_filter_block1d4_h2_avg_ssse3; + +// void vpx_convolve8_horiz_ssse3(const uint8_t *src, ptrdiff_t src_stride, +// uint8_t *dst, ptrdiff_t dst_stride, +// const int16_t *filter_x, int x_step_q4, +// const int16_t *filter_y, int y_step_q4, +// int w, int h); +// void vpx_convolve8_vert_ssse3(const uint8_t *src, ptrdiff_t src_stride, +// uint8_t *dst, ptrdiff_t dst_stride, +// const int16_t *filter_x, int x_step_q4, +// const int16_t *filter_y, int y_step_q4, +// int w, int h); +// void vpx_convolve8_avg_horiz_ssse3(const uint8_t *src, ptrdiff_t src_stride, +// uint8_t *dst, ptrdiff_t dst_stride, +// const int16_t *filter_x, int x_step_q4, +// const int16_t *filter_y, int y_step_q4, +// int w, int h); +// void vpx_convolve8_avg_vert_ssse3(const uint8_t *src, ptrdiff_t src_stride, +// uint8_t *dst, ptrdiff_t dst_stride, +// const int16_t *filter_x, int x_step_q4, +// const int16_t *filter_y, int y_step_q4, +// int w, int h); +FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , ssse3); +FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , ssse3); +FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, ssse3); +FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_, + ssse3); + +#define TRANSPOSE_8X8(in0, in1, in2, in3, in4, in5, in6, in7, \ + out0, out1, out2, out3, out4, out5, out6, out7) { \ + const __m128i tr0_0 = _mm_unpacklo_epi8(in0, in1); \ + const __m128i tr0_1 = _mm_unpacklo_epi8(in2, in3); \ + const __m128i tr0_2 = _mm_unpacklo_epi8(in4, in5); \ + const __m128i tr0_3 = _mm_unpacklo_epi8(in6, in7); \ + \ + const __m128i tr1_0 = _mm_unpacklo_epi16(tr0_0, tr0_1); \ + const __m128i tr1_1 = _mm_unpackhi_epi16(tr0_0, tr0_1); \ + const __m128i tr1_2 = _mm_unpacklo_epi16(tr0_2, tr0_3); \ + const __m128i tr1_3 = _mm_unpackhi_epi16(tr0_2, tr0_3); \ + \ + const __m128i tr2_0 = _mm_unpacklo_epi32(tr1_0, tr1_2); \ + const __m128i tr2_1 = _mm_unpackhi_epi32(tr1_0, tr1_2); \ + const __m128i tr2_2 = _mm_unpacklo_epi32(tr1_1, tr1_3); \ + const __m128i tr2_3 = _mm_unpackhi_epi32(tr1_1, tr1_3); \ + \ + out0 = _mm_unpacklo_epi64(tr2_0, tr2_0); \ + out1 = _mm_unpackhi_epi64(tr2_0, tr2_0); \ + out2 = _mm_unpacklo_epi64(tr2_1, tr2_1); \ + out3 = _mm_unpackhi_epi64(tr2_1, tr2_1); \ + out4 = _mm_unpacklo_epi64(tr2_2, tr2_2); \ + out5 = _mm_unpackhi_epi64(tr2_2, tr2_2); \ + out6 = _mm_unpacklo_epi64(tr2_3, tr2_3); \ + out7 = _mm_unpackhi_epi64(tr2_3, tr2_3); \ +} + +static void filter_horiz_w8_ssse3(const uint8_t *src_x, ptrdiff_t src_pitch, + uint8_t *dst, const int16_t *x_filter) { + const __m128i k_256 = _mm_set1_epi16(1 << 8); + const __m128i f_values = _mm_load_si128((const __m128i *)x_filter); + // pack and duplicate the filter values + const __m128i f1f0 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0200u)); + const __m128i f3f2 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0604u)); + const __m128i f5f4 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0a08u)); + const __m128i f7f6 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0e0cu)); + const __m128i A = _mm_loadl_epi64((const __m128i *)src_x); + const __m128i B = _mm_loadl_epi64((const __m128i *)(src_x + src_pitch)); + const __m128i C = _mm_loadl_epi64((const __m128i *)(src_x + src_pitch * 2)); + const __m128i D = _mm_loadl_epi64((const __m128i *)(src_x + src_pitch * 3)); + const __m128i E = _mm_loadl_epi64((const __m128i *)(src_x + src_pitch * 4)); + const __m128i F = _mm_loadl_epi64((const __m128i *)(src_x + src_pitch * 5)); + const __m128i G = _mm_loadl_epi64((const __m128i *)(src_x + src_pitch * 6)); + const __m128i H = _mm_loadl_epi64((const __m128i *)(src_x + src_pitch * 7)); + // 00 01 10 11 02 03 12 13 04 05 14 15 06 07 16 17 + const __m128i tr0_0 = _mm_unpacklo_epi16(A, B); + // 20 21 30 31 22 23 32 33 24 25 34 35 26 27 36 37 + const __m128i tr0_1 = _mm_unpacklo_epi16(C, D); + // 40 41 50 51 42 43 52 53 44 45 54 55 46 47 56 57 + const __m128i tr0_2 = _mm_unpacklo_epi16(E, F); + // 60 61 70 71 62 63 72 73 64 65 74 75 66 67 76 77 + const __m128i tr0_3 = _mm_unpacklo_epi16(G, H); + // 00 01 10 11 20 21 30 31 02 03 12 13 22 23 32 33 + const __m128i tr1_0 = _mm_unpacklo_epi32(tr0_0, tr0_1); + // 04 05 14 15 24 25 34 35 06 07 16 17 26 27 36 37 + const __m128i tr1_1 = _mm_unpackhi_epi32(tr0_0, tr0_1); + // 40 41 50 51 60 61 70 71 42 43 52 53 62 63 72 73 + const __m128i tr1_2 = _mm_unpacklo_epi32(tr0_2, tr0_3); + // 44 45 54 55 64 65 74 75 46 47 56 57 66 67 76 77 + const __m128i tr1_3 = _mm_unpackhi_epi32(tr0_2, tr0_3); + // 00 01 10 11 20 21 30 31 40 41 50 51 60 61 70 71 + const __m128i s1s0 = _mm_unpacklo_epi64(tr1_0, tr1_2); + const __m128i s3s2 = _mm_unpackhi_epi64(tr1_0, tr1_2); + const __m128i s5s4 = _mm_unpacklo_epi64(tr1_1, tr1_3); + const __m128i s7s6 = _mm_unpackhi_epi64(tr1_1, tr1_3); + // multiply 2 adjacent elements with the filter and add the result + const __m128i x0 = _mm_maddubs_epi16(s1s0, f1f0); + const __m128i x1 = _mm_maddubs_epi16(s3s2, f3f2); + const __m128i x2 = _mm_maddubs_epi16(s5s4, f5f4); + const __m128i x3 = _mm_maddubs_epi16(s7s6, f7f6); + // add and saturate the results together + const __m128i min_x2x1 = _mm_min_epi16(x2, x1); + const __m128i max_x2x1 = _mm_max_epi16(x2, x1); + __m128i temp = _mm_adds_epi16(x0, x3); + temp = _mm_adds_epi16(temp, min_x2x1); + temp = _mm_adds_epi16(temp, max_x2x1); + // round and shift by 7 bit each 16 bit + temp = _mm_mulhrs_epi16(temp, k_256); + // shrink to 8 bit each 16 bits + temp = _mm_packus_epi16(temp, temp); + // save only 8 bytes convolve result + _mm_storel_epi64((__m128i*)dst, temp); +} + +static void transpose8x8_to_dst(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride) { + __m128i A, B, C, D, E, F, G, H; + + A = _mm_loadl_epi64((const __m128i *)src); + B = _mm_loadl_epi64((const __m128i *)(src + src_stride)); + C = _mm_loadl_epi64((const __m128i *)(src + src_stride * 2)); + D = _mm_loadl_epi64((const __m128i *)(src + src_stride * 3)); + E = _mm_loadl_epi64((const __m128i *)(src + src_stride * 4)); + F = _mm_loadl_epi64((const __m128i *)(src + src_stride * 5)); + G = _mm_loadl_epi64((const __m128i *)(src + src_stride * 6)); + H = _mm_loadl_epi64((const __m128i *)(src + src_stride * 7)); + + TRANSPOSE_8X8(A, B, C, D, E, F, G, H, + A, B, C, D, E, F, G, H); + + _mm_storel_epi64((__m128i*)dst, A); + _mm_storel_epi64((__m128i*)(dst + dst_stride * 1), B); + _mm_storel_epi64((__m128i*)(dst + dst_stride * 2), C); + _mm_storel_epi64((__m128i*)(dst + dst_stride * 3), D); + _mm_storel_epi64((__m128i*)(dst + dst_stride * 4), E); + _mm_storel_epi64((__m128i*)(dst + dst_stride * 5), F); + _mm_storel_epi64((__m128i*)(dst + dst_stride * 6), G); + _mm_storel_epi64((__m128i*)(dst + dst_stride * 7), H); +} + +static void scaledconvolve_horiz_w8(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const InterpKernel *x_filters, + int x0_q4, int x_step_q4, int w, int h) { + DECLARE_ALIGNED(16, uint8_t, temp[8 * 8]); + int x, y, z; + src -= SUBPEL_TAPS / 2 - 1; + + // This function processes 8x8 areas. The intermediate height is not always + // a multiple of 8, so force it to be a multiple of 8 here. + y = h + (8 - (h & 0x7)); + + do { + int x_q4 = x0_q4; + for (x = 0; x < w; x += 8) { + // process 8 src_x steps + for (z = 0; z < 8; ++z) { + const uint8_t *const src_x = &src[x_q4 >> SUBPEL_BITS]; + const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK]; + if (x_q4 & SUBPEL_MASK) { + filter_horiz_w8_ssse3(src_x, src_stride, temp + (z * 8), x_filter); + } else { + int i; + for (i = 0; i < 8; ++i) { + temp[z * 8 + i] = src_x[i * src_stride + 3]; + } + } + x_q4 += x_step_q4; + } + + // transpose the 8x8 filters values back to dst + transpose8x8_to_dst(temp, 8, dst + x, dst_stride); + } + + src += src_stride * 8; + dst += dst_stride * 8; + } while (y -= 8); +} + +static void filter_horiz_w4_ssse3(const uint8_t *src_ptr, ptrdiff_t src_pitch, + uint8_t *dst, const int16_t *filter) { + const __m128i k_256 = _mm_set1_epi16(1 << 8); + const __m128i f_values = _mm_load_si128((const __m128i *)filter); + // pack and duplicate the filter values + const __m128i f1f0 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0200u)); + const __m128i f3f2 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0604u)); + const __m128i f5f4 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0a08u)); + const __m128i f7f6 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0e0cu)); + const __m128i A = _mm_loadl_epi64((const __m128i *)src_ptr); + const __m128i B = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch)); + const __m128i C = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 2)); + const __m128i D = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 3)); + // TRANSPOSE... + // 00 01 02 03 04 05 06 07 + // 10 11 12 13 14 15 16 17 + // 20 21 22 23 24 25 26 27 + // 30 31 32 33 34 35 36 37 + // + // TO + // + // 00 10 20 30 + // 01 11 21 31 + // 02 12 22 32 + // 03 13 23 33 + // 04 14 24 34 + // 05 15 25 35 + // 06 16 26 36 + // 07 17 27 37 + // + // 00 01 10 11 02 03 12 13 04 05 14 15 06 07 16 17 + const __m128i tr0_0 = _mm_unpacklo_epi16(A, B); + // 20 21 30 31 22 23 32 33 24 25 34 35 26 27 36 37 + const __m128i tr0_1 = _mm_unpacklo_epi16(C, D); + // 00 01 10 11 20 21 30 31 02 03 12 13 22 23 32 33 + const __m128i s1s0 = _mm_unpacklo_epi32(tr0_0, tr0_1); + // 04 05 14 15 24 25 34 35 06 07 16 17 26 27 36 37 + const __m128i s5s4 = _mm_unpackhi_epi32(tr0_0, tr0_1); + // 02 03 12 13 22 23 32 33 + const __m128i s3s2 = _mm_srli_si128(s1s0, 8); + // 06 07 16 17 26 27 36 37 + const __m128i s7s6 = _mm_srli_si128(s5s4, 8); + // multiply 2 adjacent elements with the filter and add the result + const __m128i x0 = _mm_maddubs_epi16(s1s0, f1f0); + const __m128i x1 = _mm_maddubs_epi16(s3s2, f3f2); + const __m128i x2 = _mm_maddubs_epi16(s5s4, f5f4); + const __m128i x3 = _mm_maddubs_epi16(s7s6, f7f6); + // add and saturate the results together + const __m128i min_x2x1 = _mm_min_epi16(x2, x1); + const __m128i max_x2x1 = _mm_max_epi16(x2, x1); + __m128i temp = _mm_adds_epi16(x0, x3); + temp = _mm_adds_epi16(temp, min_x2x1); + temp = _mm_adds_epi16(temp, max_x2x1); + // round and shift by 7 bit each 16 bit + temp = _mm_mulhrs_epi16(temp, k_256); + // shrink to 8 bit each 16 bits + temp = _mm_packus_epi16(temp, temp); + // save only 4 bytes + *(int *)dst = _mm_cvtsi128_si32(temp); +} + +static void transpose4x4_to_dst(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride) { + __m128i A = _mm_cvtsi32_si128(*(const int *)src); + __m128i B = _mm_cvtsi32_si128(*(const int *)(src + src_stride)); + __m128i C = _mm_cvtsi32_si128(*(const int *)(src + src_stride * 2)); + __m128i D = _mm_cvtsi32_si128(*(const int *)(src + src_stride * 3)); + // 00 10 01 11 02 12 03 13 + const __m128i tr0_0 = _mm_unpacklo_epi8(A, B); + // 20 30 21 31 22 32 23 33 + const __m128i tr0_1 = _mm_unpacklo_epi8(C, D); + // 00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33 + A = _mm_unpacklo_epi16(tr0_0, tr0_1); + B = _mm_srli_si128(A, 4); + C = _mm_srli_si128(A, 8); + D = _mm_srli_si128(A, 12); + + *(int *)(dst) = _mm_cvtsi128_si32(A); + *(int *)(dst + dst_stride) = _mm_cvtsi128_si32(B); + *(int *)(dst + dst_stride * 2) = _mm_cvtsi128_si32(C); + *(int *)(dst + dst_stride * 3) = _mm_cvtsi128_si32(D); +} + +static void scaledconvolve_horiz_w4(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const InterpKernel *x_filters, + int x0_q4, int x_step_q4, int w, int h) { + DECLARE_ALIGNED(16, uint8_t, temp[4 * 4]); + int x, y, z; + src -= SUBPEL_TAPS / 2 - 1; + + for (y = 0; y < h; y += 4) { + int x_q4 = x0_q4; + for (x = 0; x < w; x += 4) { + // process 4 src_x steps + for (z = 0; z < 4; ++z) { + const uint8_t *const src_x = &src[x_q4 >> SUBPEL_BITS]; + const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK]; + if (x_q4 & SUBPEL_MASK) { + filter_horiz_w4_ssse3(src_x, src_stride, temp + (z * 4), x_filter); + } else { + int i; + for (i = 0; i < 4; ++i) { + temp[z * 4 + i] = src_x[i * src_stride + 3]; + } + } + x_q4 += x_step_q4; + } + + // transpose the 4x4 filters values back to dst + transpose4x4_to_dst(temp, 4, dst + x, dst_stride); + } + + src += src_stride * 4; + dst += dst_stride * 4; + } +} + +static void filter_vert_w4_ssse3(const uint8_t *src_ptr, ptrdiff_t src_pitch, + uint8_t *dst, const int16_t *filter) { + const __m128i k_256 = _mm_set1_epi16(1 << 8); + const __m128i f_values = _mm_load_si128((const __m128i *)filter); + // pack and duplicate the filter values + const __m128i f1f0 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0200u)); + const __m128i f3f2 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0604u)); + const __m128i f5f4 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0a08u)); + const __m128i f7f6 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0e0cu)); + const __m128i A = _mm_cvtsi32_si128(*(const int *)src_ptr); + const __m128i B = _mm_cvtsi32_si128(*(const int *)(src_ptr + src_pitch)); + const __m128i C = _mm_cvtsi32_si128(*(const int *)(src_ptr + src_pitch * 2)); + const __m128i D = _mm_cvtsi32_si128(*(const int *)(src_ptr + src_pitch * 3)); + const __m128i E = _mm_cvtsi32_si128(*(const int *)(src_ptr + src_pitch * 4)); + const __m128i F = _mm_cvtsi32_si128(*(const int *)(src_ptr + src_pitch * 5)); + const __m128i G = _mm_cvtsi32_si128(*(const int *)(src_ptr + src_pitch * 6)); + const __m128i H = _mm_cvtsi32_si128(*(const int *)(src_ptr + src_pitch * 7)); + const __m128i s1s0 = _mm_unpacklo_epi8(A, B); + const __m128i s3s2 = _mm_unpacklo_epi8(C, D); + const __m128i s5s4 = _mm_unpacklo_epi8(E, F); + const __m128i s7s6 = _mm_unpacklo_epi8(G, H); + // multiply 2 adjacent elements with the filter and add the result + const __m128i x0 = _mm_maddubs_epi16(s1s0, f1f0); + const __m128i x1 = _mm_maddubs_epi16(s3s2, f3f2); + const __m128i x2 = _mm_maddubs_epi16(s5s4, f5f4); + const __m128i x3 = _mm_maddubs_epi16(s7s6, f7f6); + // add and saturate the results together + const __m128i min_x2x1 = _mm_min_epi16(x2, x1); + const __m128i max_x2x1 = _mm_max_epi16(x2, x1); + __m128i temp = _mm_adds_epi16(x0, x3); + temp = _mm_adds_epi16(temp, min_x2x1); + temp = _mm_adds_epi16(temp, max_x2x1); + // round and shift by 7 bit each 16 bit + temp = _mm_mulhrs_epi16(temp, k_256); + // shrink to 8 bit each 16 bits + temp = _mm_packus_epi16(temp, temp); + // save only 4 bytes + *(int *)dst = _mm_cvtsi128_si32(temp); +} + +static void scaledconvolve_vert_w4(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const InterpKernel *y_filters, + int y0_q4, int y_step_q4, int w, int h) { + int y; + int y_q4 = y0_q4; + + src -= src_stride * (SUBPEL_TAPS / 2 - 1); + for (y = 0; y < h; ++y) { + const unsigned char *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride]; + const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK]; + + if (y_q4 & SUBPEL_MASK) { + filter_vert_w4_ssse3(src_y, src_stride, &dst[y * dst_stride], y_filter); + } else { + memcpy(&dst[y * dst_stride], &src_y[3 * src_stride], w); + } + + y_q4 += y_step_q4; + } +} + +static void filter_vert_w8_ssse3(const uint8_t *src_ptr, ptrdiff_t src_pitch, + uint8_t *dst, const int16_t *filter) { + const __m128i k_256 = _mm_set1_epi16(1 << 8); + const __m128i f_values = _mm_load_si128((const __m128i *)filter); + // pack and duplicate the filter values + const __m128i f1f0 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0200u)); + const __m128i f3f2 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0604u)); + const __m128i f5f4 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0a08u)); + const __m128i f7f6 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0e0cu)); + const __m128i A = _mm_loadl_epi64((const __m128i *)src_ptr); + const __m128i B = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch)); + const __m128i C = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 2)); + const __m128i D = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 3)); + const __m128i E = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 4)); + const __m128i F = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 5)); + const __m128i G = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 6)); + const __m128i H = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 7)); + const __m128i s1s0 = _mm_unpacklo_epi8(A, B); + const __m128i s3s2 = _mm_unpacklo_epi8(C, D); + const __m128i s5s4 = _mm_unpacklo_epi8(E, F); + const __m128i s7s6 = _mm_unpacklo_epi8(G, H); + // multiply 2 adjacent elements with the filter and add the result + const __m128i x0 = _mm_maddubs_epi16(s1s0, f1f0); + const __m128i x1 = _mm_maddubs_epi16(s3s2, f3f2); + const __m128i x2 = _mm_maddubs_epi16(s5s4, f5f4); + const __m128i x3 = _mm_maddubs_epi16(s7s6, f7f6); + // add and saturate the results together + const __m128i min_x2x1 = _mm_min_epi16(x2, x1); + const __m128i max_x2x1 = _mm_max_epi16(x2, x1); + __m128i temp = _mm_adds_epi16(x0, x3); + temp = _mm_adds_epi16(temp, min_x2x1); + temp = _mm_adds_epi16(temp, max_x2x1); + // round and shift by 7 bit each 16 bit + temp = _mm_mulhrs_epi16(temp, k_256); + // shrink to 8 bit each 16 bits + temp = _mm_packus_epi16(temp, temp); + // save only 8 bytes convolve result + _mm_storel_epi64((__m128i*)dst, temp); +} + +static void scaledconvolve_vert_w8(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const InterpKernel *y_filters, + int y0_q4, int y_step_q4, int w, int h) { + int y; + int y_q4 = y0_q4; + + src -= src_stride * (SUBPEL_TAPS / 2 - 1); + for (y = 0; y < h; ++y) { + const unsigned char *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride]; + const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK]; + if (y_q4 & SUBPEL_MASK) { + filter_vert_w8_ssse3(src_y, src_stride, &dst[y * dst_stride], y_filter); + } else { + memcpy(&dst[y * dst_stride], &src_y[3 * src_stride], w); + } + y_q4 += y_step_q4; + } +} + +static void filter_vert_w16_ssse3(const uint8_t *src_ptr, ptrdiff_t src_pitch, + uint8_t *dst, const int16_t *filter, int w) { + const __m128i k_256 = _mm_set1_epi16(1 << 8); + const __m128i f_values = _mm_load_si128((const __m128i *)filter); + // pack and duplicate the filter values + const __m128i f1f0 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0200u)); + const __m128i f3f2 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0604u)); + const __m128i f5f4 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0a08u)); + const __m128i f7f6 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0e0cu)); + int i; + + for (i = 0; i < w; i += 16) { + const __m128i A = _mm_loadu_si128((const __m128i *)src_ptr); + const __m128i B = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch)); + const __m128i C = + _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 2)); + const __m128i D = + _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 3)); + const __m128i E = + _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 4)); + const __m128i F = + _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 5)); + const __m128i G = + _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 6)); + const __m128i H = + _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 7)); + // merge the result together + const __m128i s1s0_lo = _mm_unpacklo_epi8(A, B); + const __m128i s7s6_lo = _mm_unpacklo_epi8(G, H); + const __m128i s1s0_hi = _mm_unpackhi_epi8(A, B); + const __m128i s7s6_hi = _mm_unpackhi_epi8(G, H); + // multiply 2 adjacent elements with the filter and add the result + const __m128i x0_lo = _mm_maddubs_epi16(s1s0_lo, f1f0); + const __m128i x3_lo = _mm_maddubs_epi16(s7s6_lo, f7f6); + const __m128i x0_hi = _mm_maddubs_epi16(s1s0_hi, f1f0); + const __m128i x3_hi = _mm_maddubs_epi16(s7s6_hi, f7f6); + // add and saturate the results together + const __m128i x3x0_lo = _mm_adds_epi16(x0_lo, x3_lo); + const __m128i x3x0_hi = _mm_adds_epi16(x0_hi, x3_hi); + // merge the result together + const __m128i s3s2_lo = _mm_unpacklo_epi8(C, D); + const __m128i s3s2_hi = _mm_unpackhi_epi8(C, D); + // multiply 2 adjacent elements with the filter and add the result + const __m128i x1_lo = _mm_maddubs_epi16(s3s2_lo, f3f2); + const __m128i x1_hi = _mm_maddubs_epi16(s3s2_hi, f3f2); + // merge the result together + const __m128i s5s4_lo = _mm_unpacklo_epi8(E, F); + const __m128i s5s4_hi = _mm_unpackhi_epi8(E, F); + // multiply 2 adjacent elements with the filter and add the result + const __m128i x2_lo = _mm_maddubs_epi16(s5s4_lo, f5f4); + const __m128i x2_hi = _mm_maddubs_epi16(s5s4_hi, f5f4); + // add and saturate the results together + __m128i temp_lo = _mm_adds_epi16(x3x0_lo, _mm_min_epi16(x1_lo, x2_lo)); + __m128i temp_hi = _mm_adds_epi16(x3x0_hi, _mm_min_epi16(x1_hi, x2_hi)); + + // add and saturate the results together + temp_lo = _mm_adds_epi16(temp_lo, _mm_max_epi16(x1_lo, x2_lo)); + temp_hi = _mm_adds_epi16(temp_hi, _mm_max_epi16(x1_hi, x2_hi)); + // round and shift by 7 bit each 16 bit + temp_lo = _mm_mulhrs_epi16(temp_lo, k_256); + temp_hi = _mm_mulhrs_epi16(temp_hi, k_256); + // shrink to 8 bit each 16 bits, the first lane contain the first + // convolve result and the second lane contain the second convolve + // result + temp_hi = _mm_packus_epi16(temp_lo, temp_hi); + src_ptr += 16; + // save 16 bytes convolve result + _mm_store_si128((__m128i*)&dst[i], temp_hi); + } +} + +static void scaledconvolve_vert_w16(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const InterpKernel *y_filters, + int y0_q4, int y_step_q4, int w, int h) { + int y; + int y_q4 = y0_q4; + + src -= src_stride * (SUBPEL_TAPS / 2 - 1); + for (y = 0; y < h; ++y) { + const unsigned char *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride]; + const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK]; + if (y_q4 & SUBPEL_MASK) { + filter_vert_w16_ssse3(src_y, src_stride, &dst[y * dst_stride], y_filter, + w); + } else { + memcpy(&dst[y * dst_stride], &src_y[3 * src_stride], w); + } + y_q4 += y_step_q4; + } +} + +static void scaledconvolve2d(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const InterpKernel *const x_filters, + int x0_q4, int x_step_q4, + const InterpKernel *const y_filters, + int y0_q4, int y_step_q4, + int w, int h) { + // Note: Fixed size intermediate buffer, temp, places limits on parameters. + // 2d filtering proceeds in 2 steps: + // (1) Interpolate horizontally into an intermediate buffer, temp. + // (2) Interpolate temp vertically to derive the sub-pixel result. + // Deriving the maximum number of rows in the temp buffer (135): + // --Smallest scaling factor is x1/2 ==> y_step_q4 = 32 (Normative). + // --Largest block size is 64x64 pixels. + // --64 rows in the downscaled frame span a distance of (64 - 1) * 32 in the + // original frame (in 1/16th pixel units). + // --Must round-up because block may be located at sub-pixel position. + // --Require an additional SUBPEL_TAPS rows for the 8-tap filter tails. + // --((64 - 1) * 32 + 15) >> 4 + 8 = 135. + // --Require an additional 8 rows for the horiz_w8 transpose tail. + DECLARE_ALIGNED(16, uint8_t, temp[(135 + 8) * 64]); + const int intermediate_height = + (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS; + + assert(w <= 64); + assert(h <= 64); + assert(y_step_q4 <= 32); + assert(x_step_q4 <= 32); + + if (w >= 8) { + scaledconvolve_horiz_w8(src - src_stride * (SUBPEL_TAPS / 2 - 1), + src_stride, temp, 64, x_filters, x0_q4, x_step_q4, + w, intermediate_height); + } else { + scaledconvolve_horiz_w4(src - src_stride * (SUBPEL_TAPS / 2 - 1), + src_stride, temp, 64, x_filters, x0_q4, x_step_q4, + w, intermediate_height); + } + + if (w >= 16) { + scaledconvolve_vert_w16(temp + 64 * (SUBPEL_TAPS / 2 - 1), 64, dst, + dst_stride, y_filters, y0_q4, y_step_q4, w, h); + } else if (w == 8) { + scaledconvolve_vert_w8(temp + 64 * (SUBPEL_TAPS / 2 - 1), 64, dst, + dst_stride, y_filters, y0_q4, y_step_q4, w, h); + } else { + scaledconvolve_vert_w4(temp + 64 * (SUBPEL_TAPS / 2 - 1), 64, dst, + dst_stride, y_filters, y0_q4, y_step_q4, w, h); + } +} + +static const InterpKernel *get_filter_base(const int16_t *filter) { + // NOTE: This assumes that the filter table is 256-byte aligned. + // TODO(agrange) Modify to make independent of table alignment. + return (const InterpKernel *)(((intptr_t)filter) & ~((intptr_t)0xFF)); +} + +static int get_filter_offset(const int16_t *f, const InterpKernel *base) { + return (int)((const InterpKernel *)(intptr_t)f - base); +} + +void vpx_scaled_2d_ssse3(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h) { + const InterpKernel *const filters_x = get_filter_base(filter_x); + const int x0_q4 = get_filter_offset(filter_x, filters_x); + + const InterpKernel *const filters_y = get_filter_base(filter_y); + const int y0_q4 = get_filter_offset(filter_y, filters_y); + + scaledconvolve2d(src, src_stride, dst, dst_stride, + filters_x, x0_q4, x_step_q4, + filters_y, y0_q4, y_step_q4, w, h); +} + +// void vp9_convolve8_ssse3(const uint8_t *src, ptrdiff_t src_stride, +// uint8_t *dst, ptrdiff_t dst_stride, +// const int16_t *filter_x, int x_step_q4, +// const int16_t *filter_y, int y_step_q4, +// int w, int h); +// void vpx_convolve8_avg_ssse3(const uint8_t *src, ptrdiff_t src_stride, +// uint8_t *dst, ptrdiff_t dst_stride, +// const int16_t *filter_x, int x_step_q4, +// const int16_t *filter_y, int y_step_q4, +// int w, int h); +FUN_CONV_2D(, ssse3); +FUN_CONV_2D(avg_ , ssse3); diff --git a/thirdparty/libvpx/vpx_dsp/x86/vpx_subpixel_8t_sse2.asm b/thirdparty/libvpx/vpx_dsp/x86/vpx_subpixel_8t_sse2.asm new file mode 100644 index 0000000000..08f3d6a6cf --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/x86/vpx_subpixel_8t_sse2.asm @@ -0,0 +1,987 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + + +%include "vpx_ports/x86_abi_support.asm" + +;Note: tap3 and tap4 have to be applied and added after other taps to avoid +;overflow. + +%macro GET_FILTERS_4 0 + mov rdx, arg(5) ;filter ptr + mov rcx, 0x0400040 + + movdqa xmm7, [rdx] ;load filters + pshuflw xmm0, xmm7, 0b ;k0 + pshuflw xmm1, xmm7, 01010101b ;k1 + pshuflw xmm2, xmm7, 10101010b ;k2 + pshuflw xmm3, xmm7, 11111111b ;k3 + psrldq xmm7, 8 + pshuflw xmm4, xmm7, 0b ;k4 + pshuflw xmm5, xmm7, 01010101b ;k5 + pshuflw xmm6, xmm7, 10101010b ;k6 + pshuflw xmm7, xmm7, 11111111b ;k7 + + punpcklqdq xmm0, xmm1 + punpcklqdq xmm2, xmm3 + punpcklqdq xmm5, xmm4 + punpcklqdq xmm6, xmm7 + + movdqa k0k1, xmm0 + movdqa k2k3, xmm2 + movdqa k5k4, xmm5 + movdqa k6k7, xmm6 + + movq xmm6, rcx + pshufd xmm6, xmm6, 0 + movdqa krd, xmm6 + + pxor xmm7, xmm7 + movdqa zero, xmm7 +%endm + +%macro APPLY_FILTER_4 1 + punpckldq xmm0, xmm1 ;two row in one register + punpckldq xmm6, xmm7 + punpckldq xmm2, xmm3 + punpckldq xmm5, xmm4 + + punpcklbw xmm0, zero ;unpack to word + punpcklbw xmm6, zero + punpcklbw xmm2, zero + punpcklbw xmm5, zero + + pmullw xmm0, k0k1 ;multiply the filter factors + pmullw xmm6, k6k7 + pmullw xmm2, k2k3 + pmullw xmm5, k5k4 + + paddsw xmm0, xmm6 ;sum + movdqa xmm1, xmm0 + psrldq xmm1, 8 + paddsw xmm0, xmm1 + paddsw xmm0, xmm2 + psrldq xmm2, 8 + paddsw xmm0, xmm5 + psrldq xmm5, 8 + paddsw xmm0, xmm2 + paddsw xmm0, xmm5 + + paddsw xmm0, krd ;rounding + psraw xmm0, 7 ;shift + packuswb xmm0, xmm0 ;pack to byte + +%if %1 + movd xmm1, [rdi] + pavgb xmm0, xmm1 +%endif + movd [rdi], xmm0 +%endm + +%macro GET_FILTERS 0 + mov rdx, arg(5) ;filter ptr + mov rsi, arg(0) ;src_ptr + mov rdi, arg(2) ;output_ptr + mov rcx, 0x0400040 + + movdqa xmm7, [rdx] ;load filters + pshuflw xmm0, xmm7, 0b ;k0 + pshuflw xmm1, xmm7, 01010101b ;k1 + pshuflw xmm2, xmm7, 10101010b ;k2 + pshuflw xmm3, xmm7, 11111111b ;k3 + pshufhw xmm4, xmm7, 0b ;k4 + pshufhw xmm5, xmm7, 01010101b ;k5 + pshufhw xmm6, xmm7, 10101010b ;k6 + pshufhw xmm7, xmm7, 11111111b ;k7 + + punpcklwd xmm0, xmm0 + punpcklwd xmm1, xmm1 + punpcklwd xmm2, xmm2 + punpcklwd xmm3, xmm3 + punpckhwd xmm4, xmm4 + punpckhwd xmm5, xmm5 + punpckhwd xmm6, xmm6 + punpckhwd xmm7, xmm7 + + movdqa k0, xmm0 ;store filter factors on stack + movdqa k1, xmm1 + movdqa k2, xmm2 + movdqa k3, xmm3 + movdqa k4, xmm4 + movdqa k5, xmm5 + movdqa k6, xmm6 + movdqa k7, xmm7 + + movq xmm6, rcx + pshufd xmm6, xmm6, 0 + movdqa krd, xmm6 ;rounding + + pxor xmm7, xmm7 + movdqa zero, xmm7 +%endm + +%macro LOAD_VERT_8 1 + movq xmm0, [rsi + %1] ;0 + movq xmm1, [rsi + rax + %1] ;1 + movq xmm6, [rsi + rdx * 2 + %1] ;6 + lea rsi, [rsi + rax] + movq xmm7, [rsi + rdx * 2 + %1] ;7 + movq xmm2, [rsi + rax + %1] ;2 + movq xmm3, [rsi + rax * 2 + %1] ;3 + movq xmm4, [rsi + rdx + %1] ;4 + movq xmm5, [rsi + rax * 4 + %1] ;5 +%endm + +%macro APPLY_FILTER_8 2 + punpcklbw xmm0, zero + punpcklbw xmm1, zero + punpcklbw xmm6, zero + punpcklbw xmm7, zero + punpcklbw xmm2, zero + punpcklbw xmm5, zero + punpcklbw xmm3, zero + punpcklbw xmm4, zero + + pmullw xmm0, k0 + pmullw xmm1, k1 + pmullw xmm6, k6 + pmullw xmm7, k7 + pmullw xmm2, k2 + pmullw xmm5, k5 + pmullw xmm3, k3 + pmullw xmm4, k4 + + paddsw xmm0, xmm1 + paddsw xmm0, xmm6 + paddsw xmm0, xmm7 + paddsw xmm0, xmm2 + paddsw xmm0, xmm5 + paddsw xmm0, xmm3 + paddsw xmm0, xmm4 + + paddsw xmm0, krd ;rounding + psraw xmm0, 7 ;shift + packuswb xmm0, xmm0 ;pack back to byte +%if %1 + movq xmm1, [rdi + %2] + pavgb xmm0, xmm1 +%endif + movq [rdi + %2], xmm0 +%endm + +;void vpx_filter_block1d4_v8_sse2 +;( +; unsigned char *src_ptr, +; unsigned int src_pitch, +; unsigned char *output_ptr, +; unsigned int out_pitch, +; unsigned int output_height, +; short *filter +;) +global sym(vpx_filter_block1d4_v8_sse2) PRIVATE +sym(vpx_filter_block1d4_v8_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + push rbx + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, 16 * 6 + %define k0k1 [rsp + 16 * 0] + %define k2k3 [rsp + 16 * 1] + %define k5k4 [rsp + 16 * 2] + %define k6k7 [rsp + 16 * 3] + %define krd [rsp + 16 * 4] + %define zero [rsp + 16 * 5] + + GET_FILTERS_4 + + mov rsi, arg(0) ;src_ptr + mov rdi, arg(2) ;output_ptr + + movsxd rax, DWORD PTR arg(1) ;pixels_per_line + movsxd rbx, DWORD PTR arg(3) ;out_pitch + lea rdx, [rax + rax * 2] + movsxd rcx, DWORD PTR arg(4) ;output_height + +.loop: + movd xmm0, [rsi] ;load src: row 0 + movd xmm1, [rsi + rax] ;1 + movd xmm6, [rsi + rdx * 2] ;6 + lea rsi, [rsi + rax] + movd xmm7, [rsi + rdx * 2] ;7 + movd xmm2, [rsi + rax] ;2 + movd xmm3, [rsi + rax * 2] ;3 + movd xmm4, [rsi + rdx] ;4 + movd xmm5, [rsi + rax * 4] ;5 + + APPLY_FILTER_4 0 + + lea rdi, [rdi + rbx] + dec rcx + jnz .loop + + add rsp, 16 * 6 + pop rsp + pop rbx + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +;void vpx_filter_block1d8_v8_sse2 +;( +; unsigned char *src_ptr, +; unsigned int src_pitch, +; unsigned char *output_ptr, +; unsigned int out_pitch, +; unsigned int output_height, +; short *filter +;) +global sym(vpx_filter_block1d8_v8_sse2) PRIVATE +sym(vpx_filter_block1d8_v8_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + push rbx + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, 16 * 10 + %define k0 [rsp + 16 * 0] + %define k1 [rsp + 16 * 1] + %define k2 [rsp + 16 * 2] + %define k3 [rsp + 16 * 3] + %define k4 [rsp + 16 * 4] + %define k5 [rsp + 16 * 5] + %define k6 [rsp + 16 * 6] + %define k7 [rsp + 16 * 7] + %define krd [rsp + 16 * 8] + %define zero [rsp + 16 * 9] + + GET_FILTERS + + movsxd rax, DWORD PTR arg(1) ;pixels_per_line + movsxd rbx, DWORD PTR arg(3) ;out_pitch + lea rdx, [rax + rax * 2] + movsxd rcx, DWORD PTR arg(4) ;output_height + +.loop: + LOAD_VERT_8 0 + APPLY_FILTER_8 0, 0 + + lea rdi, [rdi + rbx] + dec rcx + jnz .loop + + add rsp, 16 * 10 + pop rsp + pop rbx + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +;void vpx_filter_block1d16_v8_sse2 +;( +; unsigned char *src_ptr, +; unsigned int src_pitch, +; unsigned char *output_ptr, +; unsigned int out_pitch, +; unsigned int output_height, +; short *filter +;) +global sym(vpx_filter_block1d16_v8_sse2) PRIVATE +sym(vpx_filter_block1d16_v8_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + push rbx + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, 16 * 10 + %define k0 [rsp + 16 * 0] + %define k1 [rsp + 16 * 1] + %define k2 [rsp + 16 * 2] + %define k3 [rsp + 16 * 3] + %define k4 [rsp + 16 * 4] + %define k5 [rsp + 16 * 5] + %define k6 [rsp + 16 * 6] + %define k7 [rsp + 16 * 7] + %define krd [rsp + 16 * 8] + %define zero [rsp + 16 * 9] + + GET_FILTERS + + movsxd rax, DWORD PTR arg(1) ;pixels_per_line + movsxd rbx, DWORD PTR arg(3) ;out_pitch + lea rdx, [rax + rax * 2] + movsxd rcx, DWORD PTR arg(4) ;output_height + +.loop: + LOAD_VERT_8 0 + APPLY_FILTER_8 0, 0 + sub rsi, rax + + LOAD_VERT_8 8 + APPLY_FILTER_8 0, 8 + add rdi, rbx + + dec rcx + jnz .loop + + add rsp, 16 * 10 + pop rsp + pop rbx + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d4_v8_avg_sse2) PRIVATE +sym(vpx_filter_block1d4_v8_avg_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + push rbx + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, 16 * 6 + %define k0k1 [rsp + 16 * 0] + %define k2k3 [rsp + 16 * 1] + %define k5k4 [rsp + 16 * 2] + %define k6k7 [rsp + 16 * 3] + %define krd [rsp + 16 * 4] + %define zero [rsp + 16 * 5] + + GET_FILTERS_4 + + mov rsi, arg(0) ;src_ptr + mov rdi, arg(2) ;output_ptr + + movsxd rax, DWORD PTR arg(1) ;pixels_per_line + movsxd rbx, DWORD PTR arg(3) ;out_pitch + lea rdx, [rax + rax * 2] + movsxd rcx, DWORD PTR arg(4) ;output_height + +.loop: + movd xmm0, [rsi] ;load src: row 0 + movd xmm1, [rsi + rax] ;1 + movd xmm6, [rsi + rdx * 2] ;6 + lea rsi, [rsi + rax] + movd xmm7, [rsi + rdx * 2] ;7 + movd xmm2, [rsi + rax] ;2 + movd xmm3, [rsi + rax * 2] ;3 + movd xmm4, [rsi + rdx] ;4 + movd xmm5, [rsi + rax * 4] ;5 + + APPLY_FILTER_4 1 + + lea rdi, [rdi + rbx] + dec rcx + jnz .loop + + add rsp, 16 * 6 + pop rsp + pop rbx + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d8_v8_avg_sse2) PRIVATE +sym(vpx_filter_block1d8_v8_avg_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + push rbx + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, 16 * 10 + %define k0 [rsp + 16 * 0] + %define k1 [rsp + 16 * 1] + %define k2 [rsp + 16 * 2] + %define k3 [rsp + 16 * 3] + %define k4 [rsp + 16 * 4] + %define k5 [rsp + 16 * 5] + %define k6 [rsp + 16 * 6] + %define k7 [rsp + 16 * 7] + %define krd [rsp + 16 * 8] + %define zero [rsp + 16 * 9] + + GET_FILTERS + + movsxd rax, DWORD PTR arg(1) ;pixels_per_line + movsxd rbx, DWORD PTR arg(3) ;out_pitch + lea rdx, [rax + rax * 2] + movsxd rcx, DWORD PTR arg(4) ;output_height +.loop: + LOAD_VERT_8 0 + APPLY_FILTER_8 1, 0 + + lea rdi, [rdi + rbx] + dec rcx + jnz .loop + + add rsp, 16 * 10 + pop rsp + pop rbx + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d16_v8_avg_sse2) PRIVATE +sym(vpx_filter_block1d16_v8_avg_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + push rbx + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, 16 * 10 + %define k0 [rsp + 16 * 0] + %define k1 [rsp + 16 * 1] + %define k2 [rsp + 16 * 2] + %define k3 [rsp + 16 * 3] + %define k4 [rsp + 16 * 4] + %define k5 [rsp + 16 * 5] + %define k6 [rsp + 16 * 6] + %define k7 [rsp + 16 * 7] + %define krd [rsp + 16 * 8] + %define zero [rsp + 16 * 9] + + GET_FILTERS + + movsxd rax, DWORD PTR arg(1) ;pixels_per_line + movsxd rbx, DWORD PTR arg(3) ;out_pitch + lea rdx, [rax + rax * 2] + movsxd rcx, DWORD PTR arg(4) ;output_height +.loop: + LOAD_VERT_8 0 + APPLY_FILTER_8 1, 0 + sub rsi, rax + + LOAD_VERT_8 8 + APPLY_FILTER_8 1, 8 + add rdi, rbx + + dec rcx + jnz .loop + + add rsp, 16 * 10 + pop rsp + pop rbx + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +;void vpx_filter_block1d4_h8_sse2 +;( +; unsigned char *src_ptr, +; unsigned int src_pixels_per_line, +; unsigned char *output_ptr, +; unsigned int output_pitch, +; unsigned int output_height, +; short *filter +;) +global sym(vpx_filter_block1d4_h8_sse2) PRIVATE +sym(vpx_filter_block1d4_h8_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, 16 * 6 + %define k0k1 [rsp + 16 * 0] + %define k2k3 [rsp + 16 * 1] + %define k5k4 [rsp + 16 * 2] + %define k6k7 [rsp + 16 * 3] + %define krd [rsp + 16 * 4] + %define zero [rsp + 16 * 5] + + GET_FILTERS_4 + + mov rsi, arg(0) ;src_ptr + mov rdi, arg(2) ;output_ptr + + movsxd rax, DWORD PTR arg(1) ;pixels_per_line + movsxd rdx, DWORD PTR arg(3) ;out_pitch + movsxd rcx, DWORD PTR arg(4) ;output_height + +.loop: + movdqu xmm0, [rsi - 3] ;load src + + movdqa xmm1, xmm0 + movdqa xmm6, xmm0 + movdqa xmm7, xmm0 + movdqa xmm2, xmm0 + movdqa xmm3, xmm0 + movdqa xmm5, xmm0 + movdqa xmm4, xmm0 + + psrldq xmm1, 1 + psrldq xmm6, 6 + psrldq xmm7, 7 + psrldq xmm2, 2 + psrldq xmm3, 3 + psrldq xmm5, 5 + psrldq xmm4, 4 + + APPLY_FILTER_4 0 + + lea rsi, [rsi + rax] + lea rdi, [rdi + rdx] + dec rcx + jnz .loop + + add rsp, 16 * 6 + pop rsp + + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +;void vpx_filter_block1d8_h8_sse2 +;( +; unsigned char *src_ptr, +; unsigned int src_pixels_per_line, +; unsigned char *output_ptr, +; unsigned int output_pitch, +; unsigned int output_height, +; short *filter +;) +global sym(vpx_filter_block1d8_h8_sse2) PRIVATE +sym(vpx_filter_block1d8_h8_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, 16 * 10 + %define k0 [rsp + 16 * 0] + %define k1 [rsp + 16 * 1] + %define k2 [rsp + 16 * 2] + %define k3 [rsp + 16 * 3] + %define k4 [rsp + 16 * 4] + %define k5 [rsp + 16 * 5] + %define k6 [rsp + 16 * 6] + %define k7 [rsp + 16 * 7] + %define krd [rsp + 16 * 8] + %define zero [rsp + 16 * 9] + + GET_FILTERS + + movsxd rax, DWORD PTR arg(1) ;pixels_per_line + movsxd rdx, DWORD PTR arg(3) ;out_pitch + movsxd rcx, DWORD PTR arg(4) ;output_height + +.loop: + movdqu xmm0, [rsi - 3] ;load src + + movdqa xmm1, xmm0 + movdqa xmm6, xmm0 + movdqa xmm7, xmm0 + movdqa xmm2, xmm0 + movdqa xmm5, xmm0 + movdqa xmm3, xmm0 + movdqa xmm4, xmm0 + + psrldq xmm1, 1 + psrldq xmm6, 6 + psrldq xmm7, 7 + psrldq xmm2, 2 + psrldq xmm5, 5 + psrldq xmm3, 3 + psrldq xmm4, 4 + + APPLY_FILTER_8 0, 0 + + lea rsi, [rsi + rax] + lea rdi, [rdi + rdx] + dec rcx + jnz .loop + + add rsp, 16 * 10 + pop rsp + + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +;void vpx_filter_block1d16_h8_sse2 +;( +; unsigned char *src_ptr, +; unsigned int src_pixels_per_line, +; unsigned char *output_ptr, +; unsigned int output_pitch, +; unsigned int output_height, +; short *filter +;) +global sym(vpx_filter_block1d16_h8_sse2) PRIVATE +sym(vpx_filter_block1d16_h8_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, 16 * 10 + %define k0 [rsp + 16 * 0] + %define k1 [rsp + 16 * 1] + %define k2 [rsp + 16 * 2] + %define k3 [rsp + 16 * 3] + %define k4 [rsp + 16 * 4] + %define k5 [rsp + 16 * 5] + %define k6 [rsp + 16 * 6] + %define k7 [rsp + 16 * 7] + %define krd [rsp + 16 * 8] + %define zero [rsp + 16 * 9] + + GET_FILTERS + + movsxd rax, DWORD PTR arg(1) ;pixels_per_line + movsxd rdx, DWORD PTR arg(3) ;out_pitch + movsxd rcx, DWORD PTR arg(4) ;output_height + +.loop: + movdqu xmm0, [rsi - 3] ;load src + + movdqa xmm1, xmm0 + movdqa xmm6, xmm0 + movdqa xmm7, xmm0 + movdqa xmm2, xmm0 + movdqa xmm5, xmm0 + movdqa xmm3, xmm0 + movdqa xmm4, xmm0 + + psrldq xmm1, 1 + psrldq xmm6, 6 + psrldq xmm7, 7 + psrldq xmm2, 2 + psrldq xmm5, 5 + psrldq xmm3, 3 + psrldq xmm4, 4 + + APPLY_FILTER_8 0, 0 + + movdqu xmm0, [rsi + 5] ;load src + + movdqa xmm1, xmm0 + movdqa xmm6, xmm0 + movdqa xmm7, xmm0 + movdqa xmm2, xmm0 + movdqa xmm5, xmm0 + movdqa xmm3, xmm0 + movdqa xmm4, xmm0 + + psrldq xmm1, 1 + psrldq xmm6, 6 + psrldq xmm7, 7 + psrldq xmm2, 2 + psrldq xmm5, 5 + psrldq xmm3, 3 + psrldq xmm4, 4 + + APPLY_FILTER_8 0, 8 + + lea rsi, [rsi + rax] + lea rdi, [rdi + rdx] + dec rcx + jnz .loop + + add rsp, 16 * 10 + pop rsp + + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d4_h8_avg_sse2) PRIVATE +sym(vpx_filter_block1d4_h8_avg_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, 16 * 6 + %define k0k1 [rsp + 16 * 0] + %define k2k3 [rsp + 16 * 1] + %define k5k4 [rsp + 16 * 2] + %define k6k7 [rsp + 16 * 3] + %define krd [rsp + 16 * 4] + %define zero [rsp + 16 * 5] + + GET_FILTERS_4 + + mov rsi, arg(0) ;src_ptr + mov rdi, arg(2) ;output_ptr + + movsxd rax, DWORD PTR arg(1) ;pixels_per_line + movsxd rdx, DWORD PTR arg(3) ;out_pitch + movsxd rcx, DWORD PTR arg(4) ;output_height + +.loop: + movdqu xmm0, [rsi - 3] ;load src + + movdqa xmm1, xmm0 + movdqa xmm6, xmm0 + movdqa xmm7, xmm0 + movdqa xmm2, xmm0 + movdqa xmm3, xmm0 + movdqa xmm5, xmm0 + movdqa xmm4, xmm0 + + psrldq xmm1, 1 + psrldq xmm6, 6 + psrldq xmm7, 7 + psrldq xmm2, 2 + psrldq xmm3, 3 + psrldq xmm5, 5 + psrldq xmm4, 4 + + APPLY_FILTER_4 1 + + lea rsi, [rsi + rax] + lea rdi, [rdi + rdx] + dec rcx + jnz .loop + + add rsp, 16 * 6 + pop rsp + + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d8_h8_avg_sse2) PRIVATE +sym(vpx_filter_block1d8_h8_avg_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, 16 * 10 + %define k0 [rsp + 16 * 0] + %define k1 [rsp + 16 * 1] + %define k2 [rsp + 16 * 2] + %define k3 [rsp + 16 * 3] + %define k4 [rsp + 16 * 4] + %define k5 [rsp + 16 * 5] + %define k6 [rsp + 16 * 6] + %define k7 [rsp + 16 * 7] + %define krd [rsp + 16 * 8] + %define zero [rsp + 16 * 9] + + GET_FILTERS + + movsxd rax, DWORD PTR arg(1) ;pixels_per_line + movsxd rdx, DWORD PTR arg(3) ;out_pitch + movsxd rcx, DWORD PTR arg(4) ;output_height + +.loop: + movdqu xmm0, [rsi - 3] ;load src + + movdqa xmm1, xmm0 + movdqa xmm6, xmm0 + movdqa xmm7, xmm0 + movdqa xmm2, xmm0 + movdqa xmm5, xmm0 + movdqa xmm3, xmm0 + movdqa xmm4, xmm0 + + psrldq xmm1, 1 + psrldq xmm6, 6 + psrldq xmm7, 7 + psrldq xmm2, 2 + psrldq xmm5, 5 + psrldq xmm3, 3 + psrldq xmm4, 4 + + APPLY_FILTER_8 1, 0 + + lea rsi, [rsi + rax] + lea rdi, [rdi + rdx] + dec rcx + jnz .loop + + add rsp, 16 * 10 + pop rsp + + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d16_h8_avg_sse2) PRIVATE +sym(vpx_filter_block1d16_h8_avg_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + ALIGN_STACK 16, rax + sub rsp, 16 * 10 + %define k0 [rsp + 16 * 0] + %define k1 [rsp + 16 * 1] + %define k2 [rsp + 16 * 2] + %define k3 [rsp + 16 * 3] + %define k4 [rsp + 16 * 4] + %define k5 [rsp + 16 * 5] + %define k6 [rsp + 16 * 6] + %define k7 [rsp + 16 * 7] + %define krd [rsp + 16 * 8] + %define zero [rsp + 16 * 9] + + GET_FILTERS + + movsxd rax, DWORD PTR arg(1) ;pixels_per_line + movsxd rdx, DWORD PTR arg(3) ;out_pitch + movsxd rcx, DWORD PTR arg(4) ;output_height + +.loop: + movdqu xmm0, [rsi - 3] ;load src + + movdqa xmm1, xmm0 + movdqa xmm6, xmm0 + movdqa xmm7, xmm0 + movdqa xmm2, xmm0 + movdqa xmm5, xmm0 + movdqa xmm3, xmm0 + movdqa xmm4, xmm0 + + psrldq xmm1, 1 + psrldq xmm6, 6 + psrldq xmm7, 7 + psrldq xmm2, 2 + psrldq xmm5, 5 + psrldq xmm3, 3 + psrldq xmm4, 4 + + APPLY_FILTER_8 1, 0 + + movdqu xmm0, [rsi + 5] ;load src + + movdqa xmm1, xmm0 + movdqa xmm6, xmm0 + movdqa xmm7, xmm0 + movdqa xmm2, xmm0 + movdqa xmm5, xmm0 + movdqa xmm3, xmm0 + movdqa xmm4, xmm0 + + psrldq xmm1, 1 + psrldq xmm6, 6 + psrldq xmm7, 7 + psrldq xmm2, 2 + psrldq xmm5, 5 + psrldq xmm3, 3 + psrldq xmm4, 4 + + APPLY_FILTER_8 1, 8 + + lea rsi, [rsi + rax] + lea rdi, [rdi + rdx] + dec rcx + jnz .loop + + add rsp, 16 * 10 + pop rsp + + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret diff --git a/thirdparty/libvpx/vpx_dsp/x86/vpx_subpixel_8t_ssse3.asm b/thirdparty/libvpx/vpx_dsp/x86/vpx_subpixel_8t_ssse3.asm new file mode 100644 index 0000000000..d2cb8ea292 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/x86/vpx_subpixel_8t_ssse3.asm @@ -0,0 +1,629 @@ +; +; Copyright (c) 2015 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + +%include "third_party/x86inc/x86inc.asm" + +SECTION_RODATA +pw_64: times 8 dw 64 + +; %define USE_PMULHRSW +; NOTE: pmulhrsw has a latency of 5 cycles. Tests showed a performance loss +; when using this instruction. +; +; The add order below (based on ffvp9) must be followed to prevent outranges. +; x = k0k1 + k4k5 +; y = k2k3 + k6k7 +; z = signed SAT(x + y) + +SECTION .text +%if ARCH_X86_64 + %define LOCAL_VARS_SIZE 16*4 +%else + %define LOCAL_VARS_SIZE 16*6 +%endif + +%macro SETUP_LOCAL_VARS 0 + ; TODO(slavarnway): using xmm registers for these on ARCH_X86_64 + + ; pmaddubsw has a higher latency on some platforms, this might be eased by + ; interleaving the instructions. + %define k0k1 [rsp + 16*0] + %define k2k3 [rsp + 16*1] + %define k4k5 [rsp + 16*2] + %define k6k7 [rsp + 16*3] + packsswb m4, m4 + ; TODO(slavarnway): multiple pshufb instructions had a higher latency on + ; some platforms. + pshuflw m0, m4, 0b ;k0_k1 + pshuflw m1, m4, 01010101b ;k2_k3 + pshuflw m2, m4, 10101010b ;k4_k5 + pshuflw m3, m4, 11111111b ;k6_k7 + punpcklqdq m0, m0 + punpcklqdq m1, m1 + punpcklqdq m2, m2 + punpcklqdq m3, m3 + mova k0k1, m0 + mova k2k3, m1 + mova k4k5, m2 + mova k6k7, m3 +%if ARCH_X86_64 + %define krd m12 + %define tmp m13 + mova krd, [GLOBAL(pw_64)] +%else + %define tmp [rsp + 16*4] + %define krd [rsp + 16*5] +%if CONFIG_PIC=0 + mova m6, [GLOBAL(pw_64)] +%else + ; build constants without accessing global memory + pcmpeqb m6, m6 ;all ones + psrlw m6, 15 + psllw m6, 6 ;aka pw_64 +%endif + mova krd, m6 +%endif +%endm + +%macro HORIZx4_ROW 2 + mova %2, %1 + punpcklbw %1, %1 + punpckhbw %2, %2 + + mova m3, %2 + palignr %2, %1, 1 + palignr m3, %1, 5 + + pmaddubsw %2, k0k1k4k5 + pmaddubsw m3, k2k3k6k7 + mova m4, %2 ;k0k1 + mova m5, m3 ;k2k3 + psrldq %2, 8 ;k4k5 + psrldq m3, 8 ;k6k7 + paddsw %2, m4 + paddsw m5, m3 + paddsw %2, m5 + paddsw %2, krd + psraw %2, 7 + packuswb %2, %2 +%endm + +;------------------------------------------------------------------------------- +%macro SUBPIX_HFILTER4 1 +cglobal filter_block1d4_%1, 6, 6+(ARCH_X86_64*2), 11, LOCAL_VARS_SIZE, \ + src, sstride, dst, dstride, height, filter + mova m4, [filterq] + packsswb m4, m4 +%if ARCH_X86_64 + %define k0k1k4k5 m8 + %define k2k3k6k7 m9 + %define krd m10 + %define orig_height r7d + mova krd, [GLOBAL(pw_64)] + pshuflw k0k1k4k5, m4, 0b ;k0_k1 + pshufhw k0k1k4k5, k0k1k4k5, 10101010b ;k0_k1_k4_k5 + pshuflw k2k3k6k7, m4, 01010101b ;k2_k3 + pshufhw k2k3k6k7, k2k3k6k7, 11111111b ;k2_k3_k6_k7 +%else + %define k0k1k4k5 [rsp + 16*0] + %define k2k3k6k7 [rsp + 16*1] + %define krd [rsp + 16*2] + %define orig_height [rsp + 16*3] + pshuflw m6, m4, 0b ;k0_k1 + pshufhw m6, m6, 10101010b ;k0_k1_k4_k5 + pshuflw m7, m4, 01010101b ;k2_k3 + pshufhw m7, m7, 11111111b ;k2_k3_k6_k7 +%if CONFIG_PIC=0 + mova m1, [GLOBAL(pw_64)] +%else + ; build constants without accessing global memory + pcmpeqb m1, m1 ;all ones + psrlw m1, 15 + psllw m1, 6 ;aka pw_64 +%endif + mova k0k1k4k5, m6 + mova k2k3k6k7, m7 + mova krd, m1 +%endif + mov orig_height, heightd + shr heightd, 1 +.loop: + ;Do two rows at once + movh m0, [srcq - 3] + movh m1, [srcq + 5] + punpcklqdq m0, m1 + mova m1, m0 + movh m2, [srcq + sstrideq - 3] + movh m3, [srcq + sstrideq + 5] + punpcklqdq m2, m3 + mova m3, m2 + punpcklbw m0, m0 + punpckhbw m1, m1 + punpcklbw m2, m2 + punpckhbw m3, m3 + mova m4, m1 + palignr m4, m0, 1 + pmaddubsw m4, k0k1k4k5 + palignr m1, m0, 5 + pmaddubsw m1, k2k3k6k7 + mova m7, m3 + palignr m7, m2, 1 + pmaddubsw m7, k0k1k4k5 + palignr m3, m2, 5 + pmaddubsw m3, k2k3k6k7 + mova m0, m4 ;k0k1 + mova m5, m1 ;k2k3 + mova m2, m7 ;k0k1 upper + psrldq m4, 8 ;k4k5 + psrldq m1, 8 ;k6k7 + paddsw m4, m0 + paddsw m5, m1 + mova m1, m3 ;k2k3 upper + psrldq m7, 8 ;k4k5 upper + psrldq m3, 8 ;k6k7 upper + paddsw m7, m2 + paddsw m4, m5 + paddsw m1, m3 + paddsw m7, m1 + paddsw m4, krd + psraw m4, 7 + packuswb m4, m4 + paddsw m7, krd + psraw m7, 7 + packuswb m7, m7 + +%ifidn %1, h8_avg + movd m0, [dstq] + pavgb m4, m0 + movd m2, [dstq + dstrideq] + pavgb m7, m2 +%endif + movd [dstq], m4 + movd [dstq + dstrideq], m7 + + lea srcq, [srcq + sstrideq ] + prefetcht0 [srcq + 4 * sstrideq - 3] + lea srcq, [srcq + sstrideq ] + lea dstq, [dstq + 2 * dstrideq ] + prefetcht0 [srcq + 2 * sstrideq - 3] + + dec heightd + jnz .loop + + ; Do last row if output_height is odd + mov heightd, orig_height + and heightd, 1 + je .done + + movh m0, [srcq - 3] ; load src + movh m1, [srcq + 5] + punpcklqdq m0, m1 + + HORIZx4_ROW m0, m1 +%ifidn %1, h8_avg + movd m0, [dstq] + pavgb m1, m0 +%endif + movd [dstq], m1 +.done + RET +%endm + +%macro HORIZx8_ROW 5 + mova %2, %1 + punpcklbw %1, %1 + punpckhbw %2, %2 + + mova %3, %2 + mova %4, %2 + mova %5, %2 + + palignr %2, %1, 1 + palignr %3, %1, 5 + palignr %4, %1, 9 + palignr %5, %1, 13 + + pmaddubsw %2, k0k1 + pmaddubsw %3, k2k3 + pmaddubsw %4, k4k5 + pmaddubsw %5, k6k7 + paddsw %2, %4 + paddsw %5, %3 + paddsw %2, %5 + paddsw %2, krd + psraw %2, 7 + packuswb %2, %2 + SWAP %1, %2 +%endm + +;------------------------------------------------------------------------------- +%macro SUBPIX_HFILTER8 1 +cglobal filter_block1d8_%1, 6, 6+(ARCH_X86_64*1), 14, LOCAL_VARS_SIZE, \ + src, sstride, dst, dstride, height, filter + mova m4, [filterq] + SETUP_LOCAL_VARS +%if ARCH_X86_64 + %define orig_height r7d +%else + %define orig_height heightmp +%endif + mov orig_height, heightd + shr heightd, 1 + +.loop: + movh m0, [srcq - 3] + movh m3, [srcq + 5] + movh m4, [srcq + sstrideq - 3] + movh m7, [srcq + sstrideq + 5] + punpcklqdq m0, m3 + mova m1, m0 + punpcklbw m0, m0 + punpckhbw m1, m1 + mova m5, m1 + palignr m5, m0, 13 + pmaddubsw m5, k6k7 + mova m2, m1 + mova m3, m1 + palignr m1, m0, 1 + pmaddubsw m1, k0k1 + punpcklqdq m4, m7 + mova m6, m4 + punpcklbw m4, m4 + palignr m2, m0, 5 + punpckhbw m6, m6 + palignr m3, m0, 9 + mova m7, m6 + pmaddubsw m2, k2k3 + pmaddubsw m3, k4k5 + + palignr m7, m4, 13 + mova m0, m6 + palignr m0, m4, 5 + pmaddubsw m7, k6k7 + paddsw m1, m3 + paddsw m2, m5 + paddsw m1, m2 + mova m5, m6 + palignr m6, m4, 1 + pmaddubsw m0, k2k3 + pmaddubsw m6, k0k1 + palignr m5, m4, 9 + paddsw m1, krd + pmaddubsw m5, k4k5 + psraw m1, 7 + paddsw m0, m7 +%ifidn %1, h8_avg + movh m7, [dstq] + movh m2, [dstq + dstrideq] +%endif + packuswb m1, m1 + paddsw m6, m5 + paddsw m6, m0 + paddsw m6, krd + psraw m6, 7 + packuswb m6, m6 +%ifidn %1, h8_avg + pavgb m1, m7 + pavgb m6, m2 +%endif + movh [dstq], m1 + movh [dstq + dstrideq], m6 + + lea srcq, [srcq + sstrideq ] + prefetcht0 [srcq + 4 * sstrideq - 3] + lea srcq, [srcq + sstrideq ] + lea dstq, [dstq + 2 * dstrideq ] + prefetcht0 [srcq + 2 * sstrideq - 3] + dec heightd + jnz .loop + + ;Do last row if output_height is odd + mov heightd, orig_height + and heightd, 1 + je .done + + movh m0, [srcq - 3] + movh m3, [srcq + 5] + punpcklqdq m0, m3 + + HORIZx8_ROW m0, m1, m2, m3, m4 + +%ifidn %1, h8_avg + movh m1, [dstq] + pavgb m0, m1 +%endif + movh [dstq], m0 +.done: + RET +%endm + +;------------------------------------------------------------------------------- +%macro SUBPIX_HFILTER16 1 +cglobal filter_block1d16_%1, 6, 6+(ARCH_X86_64*0), 14, LOCAL_VARS_SIZE, \ + src, sstride, dst, dstride, height, filter + mova m4, [filterq] + SETUP_LOCAL_VARS +.loop: + prefetcht0 [srcq + 2 * sstrideq -3] + + movh m0, [srcq - 3] + movh m4, [srcq + 5] + movh m6, [srcq + 13] + punpcklqdq m0, m4 + mova m7, m0 + punpckhbw m0, m0 + mova m1, m0 + punpcklqdq m4, m6 + mova m3, m0 + punpcklbw m7, m7 + + palignr m3, m7, 13 + mova m2, m0 + pmaddubsw m3, k6k7 + palignr m0, m7, 1 + pmaddubsw m0, k0k1 + palignr m1, m7, 5 + pmaddubsw m1, k2k3 + palignr m2, m7, 9 + pmaddubsw m2, k4k5 + paddsw m1, m3 + mova m3, m4 + punpckhbw m4, m4 + mova m5, m4 + punpcklbw m3, m3 + mova m7, m4 + palignr m5, m3, 5 + mova m6, m4 + palignr m4, m3, 1 + pmaddubsw m4, k0k1 + pmaddubsw m5, k2k3 + palignr m6, m3, 9 + pmaddubsw m6, k4k5 + palignr m7, m3, 13 + pmaddubsw m7, k6k7 + paddsw m0, m2 + paddsw m0, m1 +%ifidn %1, h8_avg + mova m1, [dstq] +%endif + paddsw m4, m6 + paddsw m5, m7 + paddsw m4, m5 + paddsw m0, krd + paddsw m4, krd + psraw m0, 7 + psraw m4, 7 + packuswb m0, m4 +%ifidn %1, h8_avg + pavgb m0, m1 +%endif + lea srcq, [srcq + sstrideq] + mova [dstq], m0 + lea dstq, [dstq + dstrideq] + dec heightd + jnz .loop + RET +%endm + +INIT_XMM ssse3 +SUBPIX_HFILTER16 h8 +SUBPIX_HFILTER16 h8_avg +SUBPIX_HFILTER8 h8 +SUBPIX_HFILTER8 h8_avg +SUBPIX_HFILTER4 h8 +SUBPIX_HFILTER4 h8_avg + +;------------------------------------------------------------------------------- +%macro SUBPIX_VFILTER 2 +cglobal filter_block1d%2_%1, 6, 6+(ARCH_X86_64*3), 14, LOCAL_VARS_SIZE, \ + src, sstride, dst, dstride, height, filter + mova m4, [filterq] + SETUP_LOCAL_VARS +%if ARCH_X86_64 + %define src1q r7 + %define sstride6q r8 + %define dst_stride dstrideq +%else + %define src1q filterq + %define sstride6q dstrideq + %define dst_stride dstridemp +%endif + mov src1q, srcq + add src1q, sstrideq + lea sstride6q, [sstrideq + sstrideq * 4] + add sstride6q, sstrideq ;pitch * 6 + +%ifidn %2, 8 + %define movx movh +%else + %define movx movd +%endif +.loop: + movx m0, [srcq ] ;A + movx m1, [srcq + sstrideq ] ;B + punpcklbw m0, m1 ;A B + movx m2, [srcq + sstrideq * 2 ] ;C + pmaddubsw m0, k0k1 + mova m6, m2 + movx m3, [src1q + sstrideq * 2] ;D + punpcklbw m2, m3 ;C D + pmaddubsw m2, k2k3 + movx m4, [srcq + sstrideq * 4 ] ;E + mova m7, m4 + movx m5, [src1q + sstrideq * 4] ;F + punpcklbw m4, m5 ;E F + pmaddubsw m4, k4k5 + punpcklbw m1, m6 ;A B next iter + movx m6, [srcq + sstride6q ] ;G + punpcklbw m5, m6 ;E F next iter + punpcklbw m3, m7 ;C D next iter + pmaddubsw m5, k4k5 + movx m7, [src1q + sstride6q ] ;H + punpcklbw m6, m7 ;G H + pmaddubsw m6, k6k7 + pmaddubsw m3, k2k3 + pmaddubsw m1, k0k1 + paddsw m0, m4 + paddsw m2, m6 + movx m6, [srcq + sstrideq * 8 ] ;H next iter + punpcklbw m7, m6 + pmaddubsw m7, k6k7 + paddsw m0, m2 + paddsw m0, krd + psraw m0, 7 + paddsw m1, m5 + packuswb m0, m0 + + paddsw m3, m7 + paddsw m1, m3 + paddsw m1, krd + psraw m1, 7 + lea srcq, [srcq + sstrideq * 2 ] + lea src1q, [src1q + sstrideq * 2] + packuswb m1, m1 + +%ifidn %1, v8_avg + movx m2, [dstq] + pavgb m0, m2 +%endif + movx [dstq], m0 + add dstq, dst_stride +%ifidn %1, v8_avg + movx m3, [dstq] + pavgb m1, m3 +%endif + movx [dstq], m1 + add dstq, dst_stride + sub heightd, 2 + cmp heightd, 1 + jg .loop + + cmp heightd, 0 + je .done + + movx m0, [srcq ] ;A + movx m1, [srcq + sstrideq ] ;B + movx m6, [srcq + sstride6q ] ;G + punpcklbw m0, m1 ;A B + movx m7, [src1q + sstride6q ] ;H + pmaddubsw m0, k0k1 + movx m2, [srcq + sstrideq * 2 ] ;C + punpcklbw m6, m7 ;G H + movx m3, [src1q + sstrideq * 2] ;D + pmaddubsw m6, k6k7 + movx m4, [srcq + sstrideq * 4 ] ;E + punpcklbw m2, m3 ;C D + movx m5, [src1q + sstrideq * 4] ;F + punpcklbw m4, m5 ;E F + pmaddubsw m2, k2k3 + pmaddubsw m4, k4k5 + paddsw m2, m6 + paddsw m0, m4 + paddsw m0, m2 + paddsw m0, krd + psraw m0, 7 + packuswb m0, m0 +%ifidn %1, v8_avg + movx m1, [dstq] + pavgb m0, m1 +%endif + movx [dstq], m0 +.done: + RET +%endm + +;------------------------------------------------------------------------------- +%macro SUBPIX_VFILTER16 1 +cglobal filter_block1d16_%1, 6, 6+(ARCH_X86_64*3), 14, LOCAL_VARS_SIZE, \ + src, sstride, dst, dstride, height, filter + mova m4, [filterq] + SETUP_LOCAL_VARS +%if ARCH_X86_64 + %define src1q r7 + %define sstride6q r8 + %define dst_stride dstrideq +%else + %define src1q filterq + %define sstride6q dstrideq + %define dst_stride dstridemp +%endif + mov src1q, srcq + add src1q, sstrideq + lea sstride6q, [sstrideq + sstrideq * 4] + add sstride6q, sstrideq ;pitch * 6 + +.loop: + movh m0, [srcq ] ;A + movh m1, [srcq + sstrideq ] ;B + movh m2, [srcq + sstrideq * 2 ] ;C + movh m3, [src1q + sstrideq * 2] ;D + movh m4, [srcq + sstrideq * 4 ] ;E + movh m5, [src1q + sstrideq * 4] ;F + + punpcklbw m0, m1 ;A B + movh m6, [srcq + sstride6q] ;G + punpcklbw m2, m3 ;C D + movh m7, [src1q + sstride6q] ;H + punpcklbw m4, m5 ;E F + pmaddubsw m0, k0k1 + movh m3, [srcq + 8] ;A + pmaddubsw m2, k2k3 + punpcklbw m6, m7 ;G H + movh m5, [srcq + sstrideq + 8] ;B + pmaddubsw m4, k4k5 + punpcklbw m3, m5 ;A B + movh m7, [srcq + sstrideq * 2 + 8] ;C + pmaddubsw m6, k6k7 + movh m5, [src1q + sstrideq * 2 + 8] ;D + punpcklbw m7, m5 ;C D + paddsw m2, m6 + pmaddubsw m3, k0k1 + movh m1, [srcq + sstrideq * 4 + 8] ;E + paddsw m0, m4 + pmaddubsw m7, k2k3 + movh m6, [src1q + sstrideq * 4 + 8] ;F + punpcklbw m1, m6 ;E F + paddsw m0, m2 + paddsw m0, krd + movh m2, [srcq + sstride6q + 8] ;G + pmaddubsw m1, k4k5 + movh m5, [src1q + sstride6q + 8] ;H + psraw m0, 7 + punpcklbw m2, m5 ;G H + pmaddubsw m2, k6k7 +%ifidn %1, v8_avg + mova m4, [dstq] +%endif + movh [dstq], m0 + paddsw m7, m2 + paddsw m3, m1 + paddsw m3, m7 + paddsw m3, krd + psraw m3, 7 + packuswb m0, m3 + + add srcq, sstrideq + add src1q, sstrideq +%ifidn %1, v8_avg + pavgb m0, m4 +%endif + mova [dstq], m0 + add dstq, dst_stride + dec heightd + jnz .loop + RET +%endm + +INIT_XMM ssse3 +SUBPIX_VFILTER16 v8 +SUBPIX_VFILTER16 v8_avg +SUBPIX_VFILTER v8, 8 +SUBPIX_VFILTER v8_avg, 8 +SUBPIX_VFILTER v8, 4 +SUBPIX_VFILTER v8_avg, 4 diff --git a/thirdparty/libvpx/vpx_dsp/x86/vpx_subpixel_bilinear_sse2.asm b/thirdparty/libvpx/vpx_dsp/x86/vpx_subpixel_bilinear_sse2.asm new file mode 100644 index 0000000000..a378dd0402 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/x86/vpx_subpixel_bilinear_sse2.asm @@ -0,0 +1,448 @@ +; +; Copyright (c) 2014 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + +%include "vpx_ports/x86_abi_support.asm" + +%macro GET_PARAM_4 0 + mov rdx, arg(5) ;filter ptr + mov rsi, arg(0) ;src_ptr + mov rdi, arg(2) ;output_ptr + mov rcx, 0x0400040 + + movdqa xmm3, [rdx] ;load filters + pshuflw xmm4, xmm3, 11111111b ;k3 + psrldq xmm3, 8 + pshuflw xmm3, xmm3, 0b ;k4 + punpcklqdq xmm4, xmm3 ;k3k4 + + movq xmm3, rcx ;rounding + pshufd xmm3, xmm3, 0 + + pxor xmm2, xmm2 + + movsxd rax, DWORD PTR arg(1) ;pixels_per_line + movsxd rdx, DWORD PTR arg(3) ;out_pitch + movsxd rcx, DWORD PTR arg(4) ;output_height +%endm + +%macro APPLY_FILTER_4 1 + + punpckldq xmm0, xmm1 ;two row in one register + punpcklbw xmm0, xmm2 ;unpack to word + pmullw xmm0, xmm4 ;multiply the filter factors + + movdqa xmm1, xmm0 + psrldq xmm1, 8 + paddsw xmm0, xmm1 + + paddsw xmm0, xmm3 ;rounding + psraw xmm0, 7 ;shift + packuswb xmm0, xmm0 ;pack to byte + +%if %1 + movd xmm1, [rdi] + pavgb xmm0, xmm1 +%endif + + movd [rdi], xmm0 + lea rsi, [rsi + rax] + lea rdi, [rdi + rdx] + dec rcx +%endm + +%macro GET_PARAM 0 + mov rdx, arg(5) ;filter ptr + mov rsi, arg(0) ;src_ptr + mov rdi, arg(2) ;output_ptr + mov rcx, 0x0400040 + + movdqa xmm7, [rdx] ;load filters + + pshuflw xmm6, xmm7, 11111111b ;k3 + pshufhw xmm7, xmm7, 0b ;k4 + punpcklwd xmm6, xmm6 + punpckhwd xmm7, xmm7 + + movq xmm4, rcx ;rounding + pshufd xmm4, xmm4, 0 + + pxor xmm5, xmm5 + + movsxd rax, DWORD PTR arg(1) ;pixels_per_line + movsxd rdx, DWORD PTR arg(3) ;out_pitch + movsxd rcx, DWORD PTR arg(4) ;output_height +%endm + +%macro APPLY_FILTER_8 1 + punpcklbw xmm0, xmm5 + punpcklbw xmm1, xmm5 + + pmullw xmm0, xmm6 + pmullw xmm1, xmm7 + paddsw xmm0, xmm1 + paddsw xmm0, xmm4 ;rounding + psraw xmm0, 7 ;shift + packuswb xmm0, xmm0 ;pack back to byte +%if %1 + movq xmm1, [rdi] + pavgb xmm0, xmm1 +%endif + movq [rdi], xmm0 ;store the result + + lea rsi, [rsi + rax] + lea rdi, [rdi + rdx] + dec rcx +%endm + +%macro APPLY_FILTER_16 1 + punpcklbw xmm0, xmm5 + punpcklbw xmm1, xmm5 + punpckhbw xmm2, xmm5 + punpckhbw xmm3, xmm5 + + pmullw xmm0, xmm6 + pmullw xmm1, xmm7 + pmullw xmm2, xmm6 + pmullw xmm3, xmm7 + + paddsw xmm0, xmm1 + paddsw xmm2, xmm3 + + paddsw xmm0, xmm4 ;rounding + paddsw xmm2, xmm4 + psraw xmm0, 7 ;shift + psraw xmm2, 7 + packuswb xmm0, xmm2 ;pack back to byte +%if %1 + movdqu xmm1, [rdi] + pavgb xmm0, xmm1 +%endif + movdqu [rdi], xmm0 ;store the result + + lea rsi, [rsi + rax] + lea rdi, [rdi + rdx] + dec rcx +%endm + +global sym(vpx_filter_block1d4_v2_sse2) PRIVATE +sym(vpx_filter_block1d4_v2_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + push rsi + push rdi + ; end prolog + + GET_PARAM_4 +.loop: + movd xmm0, [rsi] ;load src + movd xmm1, [rsi + rax] + + APPLY_FILTER_4 0 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d8_v2_sse2) PRIVATE +sym(vpx_filter_block1d8_v2_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + GET_PARAM +.loop: + movq xmm0, [rsi] ;0 + movq xmm1, [rsi + rax] ;1 + + APPLY_FILTER_8 0 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d16_v2_sse2) PRIVATE +sym(vpx_filter_block1d16_v2_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + GET_PARAM +.loop: + movdqu xmm0, [rsi] ;0 + movdqu xmm1, [rsi + rax] ;1 + movdqa xmm2, xmm0 + movdqa xmm3, xmm1 + + APPLY_FILTER_16 0 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d4_v2_avg_sse2) PRIVATE +sym(vpx_filter_block1d4_v2_avg_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + push rsi + push rdi + ; end prolog + + GET_PARAM_4 +.loop: + movd xmm0, [rsi] ;load src + movd xmm1, [rsi + rax] + + APPLY_FILTER_4 1 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d8_v2_avg_sse2) PRIVATE +sym(vpx_filter_block1d8_v2_avg_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + GET_PARAM +.loop: + movq xmm0, [rsi] ;0 + movq xmm1, [rsi + rax] ;1 + + APPLY_FILTER_8 1 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d16_v2_avg_sse2) PRIVATE +sym(vpx_filter_block1d16_v2_avg_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + GET_PARAM +.loop: + movdqu xmm0, [rsi] ;0 + movdqu xmm1, [rsi + rax] ;1 + movdqa xmm2, xmm0 + movdqa xmm3, xmm1 + + APPLY_FILTER_16 1 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d4_h2_sse2) PRIVATE +sym(vpx_filter_block1d4_h2_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + push rsi + push rdi + ; end prolog + + GET_PARAM_4 +.loop: + movdqu xmm0, [rsi] ;load src + movdqa xmm1, xmm0 + psrldq xmm1, 1 + + APPLY_FILTER_4 0 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d8_h2_sse2) PRIVATE +sym(vpx_filter_block1d8_h2_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + GET_PARAM +.loop: + movdqu xmm0, [rsi] ;load src + movdqa xmm1, xmm0 + psrldq xmm1, 1 + + APPLY_FILTER_8 0 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d16_h2_sse2) PRIVATE +sym(vpx_filter_block1d16_h2_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + GET_PARAM +.loop: + movdqu xmm0, [rsi] ;load src + movdqu xmm1, [rsi + 1] + movdqa xmm2, xmm0 + movdqa xmm3, xmm1 + + APPLY_FILTER_16 0 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d4_h2_avg_sse2) PRIVATE +sym(vpx_filter_block1d4_h2_avg_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + push rsi + push rdi + ; end prolog + + GET_PARAM_4 +.loop: + movdqu xmm0, [rsi] ;load src + movdqa xmm1, xmm0 + psrldq xmm1, 1 + + APPLY_FILTER_4 1 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d8_h2_avg_sse2) PRIVATE +sym(vpx_filter_block1d8_h2_avg_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + GET_PARAM +.loop: + movdqu xmm0, [rsi] ;load src + movdqa xmm1, xmm0 + psrldq xmm1, 1 + + APPLY_FILTER_8 1 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d16_h2_avg_sse2) PRIVATE +sym(vpx_filter_block1d16_h2_avg_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + GET_PARAM +.loop: + movdqu xmm0, [rsi] ;load src + movdqu xmm1, [rsi + 1] + movdqa xmm2, xmm0 + movdqa xmm3, xmm1 + + APPLY_FILTER_16 1 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret diff --git a/thirdparty/libvpx/vpx_dsp/x86/vpx_subpixel_bilinear_ssse3.asm b/thirdparty/libvpx/vpx_dsp/x86/vpx_subpixel_bilinear_ssse3.asm new file mode 100644 index 0000000000..3c8cfd2253 --- /dev/null +++ b/thirdparty/libvpx/vpx_dsp/x86/vpx_subpixel_bilinear_ssse3.asm @@ -0,0 +1,422 @@ +; +; Copyright (c) 2014 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + +%include "vpx_ports/x86_abi_support.asm" + +%macro GET_PARAM_4 0 + mov rdx, arg(5) ;filter ptr + mov rsi, arg(0) ;src_ptr + mov rdi, arg(2) ;output_ptr + mov rcx, 0x0400040 + + movdqa xmm3, [rdx] ;load filters + psrldq xmm3, 6 + packsswb xmm3, xmm3 + pshuflw xmm3, xmm3, 0b ;k3_k4 + + movq xmm2, rcx ;rounding + pshufd xmm2, xmm2, 0 + + movsxd rax, DWORD PTR arg(1) ;pixels_per_line + movsxd rdx, DWORD PTR arg(3) ;out_pitch + movsxd rcx, DWORD PTR arg(4) ;output_height +%endm + +%macro APPLY_FILTER_4 1 + punpcklbw xmm0, xmm1 + pmaddubsw xmm0, xmm3 + + paddsw xmm0, xmm2 ;rounding + psraw xmm0, 7 ;shift + packuswb xmm0, xmm0 ;pack to byte + +%if %1 + movd xmm1, [rdi] + pavgb xmm0, xmm1 +%endif + movd [rdi], xmm0 + lea rsi, [rsi + rax] + lea rdi, [rdi + rdx] + dec rcx +%endm + +%macro GET_PARAM 0 + mov rdx, arg(5) ;filter ptr + mov rsi, arg(0) ;src_ptr + mov rdi, arg(2) ;output_ptr + mov rcx, 0x0400040 + + movdqa xmm7, [rdx] ;load filters + psrldq xmm7, 6 + packsswb xmm7, xmm7 + pshuflw xmm7, xmm7, 0b ;k3_k4 + punpcklwd xmm7, xmm7 + + movq xmm6, rcx ;rounding + pshufd xmm6, xmm6, 0 + + movsxd rax, DWORD PTR arg(1) ;pixels_per_line + movsxd rdx, DWORD PTR arg(3) ;out_pitch + movsxd rcx, DWORD PTR arg(4) ;output_height +%endm + +%macro APPLY_FILTER_8 1 + punpcklbw xmm0, xmm1 + pmaddubsw xmm0, xmm7 + + paddsw xmm0, xmm6 ;rounding + psraw xmm0, 7 ;shift + packuswb xmm0, xmm0 ;pack back to byte + +%if %1 + movq xmm1, [rdi] + pavgb xmm0, xmm1 +%endif + movq [rdi], xmm0 ;store the result + + lea rsi, [rsi + rax] + lea rdi, [rdi + rdx] + dec rcx +%endm + +%macro APPLY_FILTER_16 1 + punpcklbw xmm0, xmm1 + punpckhbw xmm2, xmm1 + pmaddubsw xmm0, xmm7 + pmaddubsw xmm2, xmm7 + + paddsw xmm0, xmm6 ;rounding + paddsw xmm2, xmm6 + psraw xmm0, 7 ;shift + psraw xmm2, 7 + packuswb xmm0, xmm2 ;pack back to byte + +%if %1 + movdqu xmm1, [rdi] + pavgb xmm0, xmm1 +%endif + movdqu [rdi], xmm0 ;store the result + + lea rsi, [rsi + rax] + lea rdi, [rdi + rdx] + dec rcx +%endm + +global sym(vpx_filter_block1d4_v2_ssse3) PRIVATE +sym(vpx_filter_block1d4_v2_ssse3): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + push rsi + push rdi + ; end prolog + + GET_PARAM_4 +.loop: + movd xmm0, [rsi] ;load src + movd xmm1, [rsi + rax] + + APPLY_FILTER_4 0 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d8_v2_ssse3) PRIVATE +sym(vpx_filter_block1d8_v2_ssse3): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + GET_PARAM +.loop: + movq xmm0, [rsi] ;0 + movq xmm1, [rsi + rax] ;1 + + APPLY_FILTER_8 0 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d16_v2_ssse3) PRIVATE +sym(vpx_filter_block1d16_v2_ssse3): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + GET_PARAM +.loop: + movdqu xmm0, [rsi] ;0 + movdqu xmm1, [rsi + rax] ;1 + movdqa xmm2, xmm0 + + APPLY_FILTER_16 0 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d4_v2_avg_ssse3) PRIVATE +sym(vpx_filter_block1d4_v2_avg_ssse3): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + push rsi + push rdi + ; end prolog + + GET_PARAM_4 +.loop: + movd xmm0, [rsi] ;load src + movd xmm1, [rsi + rax] + + APPLY_FILTER_4 1 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d8_v2_avg_ssse3) PRIVATE +sym(vpx_filter_block1d8_v2_avg_ssse3): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + GET_PARAM +.loop: + movq xmm0, [rsi] ;0 + movq xmm1, [rsi + rax] ;1 + + APPLY_FILTER_8 1 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d16_v2_avg_ssse3) PRIVATE +sym(vpx_filter_block1d16_v2_avg_ssse3): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + GET_PARAM +.loop: + movdqu xmm0, [rsi] ;0 + movdqu xmm1, [rsi + rax] ;1 + movdqa xmm2, xmm0 + + APPLY_FILTER_16 1 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d4_h2_ssse3) PRIVATE +sym(vpx_filter_block1d4_h2_ssse3): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + push rsi + push rdi + ; end prolog + + GET_PARAM_4 +.loop: + movdqu xmm0, [rsi] ;load src + movdqa xmm1, xmm0 + psrldq xmm1, 1 + + APPLY_FILTER_4 0 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d8_h2_ssse3) PRIVATE +sym(vpx_filter_block1d8_h2_ssse3): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + GET_PARAM +.loop: + movdqu xmm0, [rsi] ;load src + movdqa xmm1, xmm0 + psrldq xmm1, 1 + + APPLY_FILTER_8 0 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d16_h2_ssse3) PRIVATE +sym(vpx_filter_block1d16_h2_ssse3): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + GET_PARAM +.loop: + movdqu xmm0, [rsi] ;load src + movdqu xmm1, [rsi + 1] + movdqa xmm2, xmm0 + + APPLY_FILTER_16 0 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d4_h2_avg_ssse3) PRIVATE +sym(vpx_filter_block1d4_h2_avg_ssse3): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + push rsi + push rdi + ; end prolog + + GET_PARAM_4 +.loop: + movdqu xmm0, [rsi] ;load src + movdqa xmm1, xmm0 + psrldq xmm1, 1 + + APPLY_FILTER_4 1 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d8_h2_avg_ssse3) PRIVATE +sym(vpx_filter_block1d8_h2_avg_ssse3): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + GET_PARAM +.loop: + movdqu xmm0, [rsi] ;load src + movdqa xmm1, xmm0 + psrldq xmm1, 1 + + APPLY_FILTER_8 1 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + +global sym(vpx_filter_block1d16_h2_avg_ssse3) PRIVATE +sym(vpx_filter_block1d16_h2_avg_ssse3): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 6 + SAVE_XMM 7 + push rsi + push rdi + ; end prolog + + GET_PARAM +.loop: + movdqu xmm0, [rsi] ;load src + movdqu xmm1, [rsi + 1] + movdqa xmm2, xmm0 + + APPLY_FILTER_16 1 + jnz .loop + + ; begin epilog + pop rdi + pop rsi + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret diff --git a/thirdparty/libvpx/vpx_mem/include/vpx_mem_intrnl.h b/thirdparty/libvpx/vpx_mem/include/vpx_mem_intrnl.h new file mode 100644 index 0000000000..c4dd78550f --- /dev/null +++ b/thirdparty/libvpx/vpx_mem/include/vpx_mem_intrnl.h @@ -0,0 +1,31 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VPX_MEM_INCLUDE_VPX_MEM_INTRNL_H_ +#define VPX_MEM_INCLUDE_VPX_MEM_INTRNL_H_ +#include "./vpx_config.h" + +#define ADDRESS_STORAGE_SIZE sizeof(size_t) + +#ifndef DEFAULT_ALIGNMENT +# if defined(VXWORKS) +# define DEFAULT_ALIGNMENT 32 /*default addr alignment to use in +calls to vpx_* functions other +than vpx_memalign*/ +# else +# define DEFAULT_ALIGNMENT (2 * sizeof(void*)) /* NOLINT */ +# endif +#endif + +/*returns an addr aligned to the byte boundary specified by align*/ +#define align_addr(addr,align) (void*)(((size_t)(addr) + ((align) - 1)) & (size_t)-(align)) + +#endif // VPX_MEM_INCLUDE_VPX_MEM_INTRNL_H_ diff --git a/thirdparty/libvpx/vpx_mem/vpx_mem.c b/thirdparty/libvpx/vpx_mem/vpx_mem.c new file mode 100644 index 0000000000..b261fc0da1 --- /dev/null +++ b/thirdparty/libvpx/vpx_mem/vpx_mem.c @@ -0,0 +1,100 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "vpx_mem.h" +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include "include/vpx_mem_intrnl.h" +#include "vpx/vpx_integer.h" + +void *vpx_memalign(size_t align, size_t size) { + void *addr, + * x = NULL; + + addr = malloc(size + align - 1 + ADDRESS_STORAGE_SIZE); + + if (addr) { + x = align_addr((unsigned char *)addr + ADDRESS_STORAGE_SIZE, (int)align); + /* save the actual malloc address */ + ((size_t *)x)[-1] = (size_t)addr; + } + + return x; +} + +void *vpx_malloc(size_t size) { + return vpx_memalign(DEFAULT_ALIGNMENT, size); +} + +void *vpx_calloc(size_t num, size_t size) { + void *x; + + x = vpx_memalign(DEFAULT_ALIGNMENT, num * size); + + if (x) + memset(x, 0, num * size); + + return x; +} + +void *vpx_realloc(void *memblk, size_t size) { + void *addr, + * new_addr = NULL; + int align = DEFAULT_ALIGNMENT; + + /* + The realloc() function changes the size of the object pointed to by + ptr to the size specified by size, and returns a pointer to the + possibly moved block. The contents are unchanged up to the lesser + of the new and old sizes. If ptr is null, realloc() behaves like + malloc() for the specified size. If size is zero (0) and ptr is + not a null pointer, the object pointed to is freed. + */ + if (!memblk) + new_addr = vpx_malloc(size); + else if (!size) + vpx_free(memblk); + else { + addr = (void *)(((size_t *)memblk)[-1]); + memblk = NULL; + + new_addr = realloc(addr, size + align + ADDRESS_STORAGE_SIZE); + + if (new_addr) { + addr = new_addr; + new_addr = (void *)(((size_t) + ((unsigned char *)new_addr + ADDRESS_STORAGE_SIZE) + (align - 1)) & + (size_t) - align); + /* save the actual malloc address */ + ((size_t *)new_addr)[-1] = (size_t)addr; + } + } + + return new_addr; +} + +void vpx_free(void *memblk) { + if (memblk) { + void *addr = (void *)(((size_t *)memblk)[-1]); + free(addr); + } +} + +#if CONFIG_VP9_HIGHBITDEPTH +void *vpx_memset16(void *dest, int val, size_t length) { + size_t i; + uint16_t *dest16 = (uint16_t *)dest; + for (i = 0; i < length; i++) + *dest16++ = val; + return dest; +} +#endif // CONFIG_VP9_HIGHBITDEPTH diff --git a/thirdparty/libvpx/vpx_mem/vpx_mem.h b/thirdparty/libvpx/vpx_mem/vpx_mem.h new file mode 100644 index 0000000000..a006e0f00b --- /dev/null +++ b/thirdparty/libvpx/vpx_mem/vpx_mem.h @@ -0,0 +1,47 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VPX_MEM_VPX_MEM_H_ +#define VPX_MEM_VPX_MEM_H_ + +#include "vpx_config.h" +#if defined(__uClinux__) +# include <lddk.h> +#endif + +#include <stdlib.h> +#include <stddef.h> + +#if defined(__cplusplus) +extern "C" { +#endif + + void *vpx_memalign(size_t align, size_t size); + void *vpx_malloc(size_t size); + void *vpx_calloc(size_t num, size_t size); + void *vpx_realloc(void *memblk, size_t size); + void vpx_free(void *memblk); + +#if CONFIG_VP9_HIGHBITDEPTH + void *vpx_memset16(void *dest, int val, size_t length); +#endif + +#include <string.h> + +#ifdef VPX_MEM_PLTFRM +# include VPX_MEM_PLTFRM +#endif + +#if defined(__cplusplus) +} +#endif + +#endif // VPX_MEM_VPX_MEM_H_ diff --git a/thirdparty/libvpx/vpx_ports/arm.h b/thirdparty/libvpx/vpx_ports/arm.h new file mode 100644 index 0000000000..42c98f5a83 --- /dev/null +++ b/thirdparty/libvpx/vpx_ports/arm.h @@ -0,0 +1,41 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VPX_PORTS_ARM_H_ +#define VPX_PORTS_ARM_H_ +#include <stdlib.h> +#include "vpx_config.h" + +#ifdef __cplusplus +extern "C" { +#endif + +/*ARMv5TE "Enhanced DSP" instructions.*/ +#define HAS_EDSP 0x01 +/*ARMv6 "Parallel" or "Media" instructions.*/ +#define HAS_MEDIA 0x02 +/*ARMv7 optional NEON instructions.*/ +#define HAS_NEON 0x04 + +int arm_cpu_caps(void); + +// Earlier gcc compilers have issues with some neon intrinsics +#if !defined(__clang__) && defined(__GNUC__) && \ + __GNUC__ == 4 && __GNUC_MINOR__ <= 6 +#define VPX_INCOMPATIBLE_GCC +#endif + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VPX_PORTS_ARM_H_ + diff --git a/thirdparty/libvpx/vpx_ports/arm_cpudetect.c b/thirdparty/libvpx/vpx_ports/arm_cpudetect.c new file mode 100644 index 0000000000..8a4b8af964 --- /dev/null +++ b/thirdparty/libvpx/vpx_ports/arm_cpudetect.c @@ -0,0 +1,175 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <stdlib.h> +#include <string.h> +#include "vpx_ports/arm.h" +#include "./vpx_config.h" + +#ifdef WINAPI_FAMILY +#include <winapifamily.h> +#if !WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP) +#define getenv(x) NULL +#endif +#endif + +static int arm_cpu_env_flags(int *flags) { + char *env; + env = getenv("VPX_SIMD_CAPS"); + if (env && *env) { + *flags = (int)strtol(env, NULL, 0); + return 0; + } + *flags = 0; + return -1; +} + +static int arm_cpu_env_mask(void) { + char *env; + env = getenv("VPX_SIMD_CAPS_MASK"); + return env && *env ? (int)strtol(env, NULL, 0) : ~0; +} + +#if !CONFIG_RUNTIME_CPU_DETECT + +int arm_cpu_caps(void) { + /* This function should actually be a no-op. There is no way to adjust any of + * these because the RTCD tables do not exist: the functions are called + * statically */ + int flags; + int mask; + if (!arm_cpu_env_flags(&flags)) { + return flags; + } + mask = arm_cpu_env_mask(); +#if HAVE_MEDIA + flags |= HAS_MEDIA; +#endif /* HAVE_MEDIA */ +#if HAVE_NEON || HAVE_NEON_ASM + flags |= HAS_NEON; +#endif /* HAVE_NEON || HAVE_NEON_ASM */ + return flags & mask; +} + +#elif defined(_MSC_VER) /* end !CONFIG_RUNTIME_CPU_DETECT */ +/*For GetExceptionCode() and EXCEPTION_ILLEGAL_INSTRUCTION.*/ +#define WIN32_LEAN_AND_MEAN +#define WIN32_EXTRA_LEAN +#include <windows.h> + +int arm_cpu_caps(void) { + int flags; + int mask; + if (!arm_cpu_env_flags(&flags)) { + return flags; + } + mask = arm_cpu_env_mask(); + /* MSVC has no inline __asm support for ARM, but it does let you __emit + * instructions via their assembled hex code. + * All of these instructions should be essentially nops. + */ +#if HAVE_MEDIA + if (mask & HAS_MEDIA) + __try { + /*SHADD8 r3,r3,r3*/ + __emit(0xE6333F93); + flags |= HAS_MEDIA; + } __except (GetExceptionCode() == EXCEPTION_ILLEGAL_INSTRUCTION) { + /*Ignore exception.*/ + } +} +#endif /* HAVE_MEDIA */ +#if HAVE_NEON || HAVE_NEON_ASM +if (mask &HAS_NEON) { + __try { + /*VORR q0,q0,q0*/ + __emit(0xF2200150); + flags |= HAS_NEON; + } __except (GetExceptionCode() == EXCEPTION_ILLEGAL_INSTRUCTION) { + /*Ignore exception.*/ + } +} +#endif /* HAVE_NEON || HAVE_NEON_ASM */ +return flags & mask; +} + +#elif defined(__ANDROID__) /* end _MSC_VER */ +#include <cpu-features.h> + +int arm_cpu_caps(void) { + int flags; + int mask; + uint64_t features; + if (!arm_cpu_env_flags(&flags)) { + return flags; + } + mask = arm_cpu_env_mask(); + features = android_getCpuFeatures(); + +#if HAVE_MEDIA + flags |= HAS_MEDIA; +#endif /* HAVE_MEDIA */ +#if HAVE_NEON || HAVE_NEON_ASM + if (features & ANDROID_CPU_ARM_FEATURE_NEON) + flags |= HAS_NEON; +#endif /* HAVE_NEON || HAVE_NEON_ASM */ + return flags & mask; +} + +#elif defined(__linux__) /* end __ANDROID__ */ + +#include <stdio.h> + +int arm_cpu_caps(void) { + FILE *fin; + int flags; + int mask; + if (!arm_cpu_env_flags(&flags)) { + return flags; + } + mask = arm_cpu_env_mask(); + /* Reading /proc/self/auxv would be easier, but that doesn't work reliably + * on Android. + * This also means that detection will fail in Scratchbox. + */ + fin = fopen("/proc/cpuinfo", "r"); + if (fin != NULL) { + /* 512 should be enough for anybody (it's even enough for all the flags + * that x86 has accumulated... so far). + */ + char buf[512]; + while (fgets(buf, 511, fin) != NULL) { +#if HAVE_NEON || HAVE_NEON_ASM + if (memcmp(buf, "Features", 8) == 0) { + char *p; + p = strstr(buf, " neon"); + if (p != NULL && (p[5] == ' ' || p[5] == '\n')) { + flags |= HAS_NEON; + } + } +#endif /* HAVE_NEON || HAVE_NEON_ASM */ +#if HAVE_MEDIA + if (memcmp(buf, "CPU architecture:", 17) == 0) { + int version; + version = atoi(buf + 17); + if (version >= 6) { + flags |= HAS_MEDIA; + } + } +#endif /* HAVE_MEDIA */ + } + fclose(fin); + } + return flags & mask; +} +#else /* end __linux__ */ +#error "--enable-runtime-cpu-detect selected, but no CPU detection method " \ +"available for your platform. Reconfigure with --disable-runtime-cpu-detect." +#endif diff --git a/thirdparty/libvpx/vpx_ports/bitops.h b/thirdparty/libvpx/vpx_ports/bitops.h new file mode 100644 index 0000000000..84ff3659fe --- /dev/null +++ b/thirdparty/libvpx/vpx_ports/bitops.h @@ -0,0 +1,76 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VPX_PORTS_BITOPS_H_ +#define VPX_PORTS_BITOPS_H_ + +#include <assert.h> + +#include "vpx_ports/msvc.h" + +#ifdef _MSC_VER +# include <math.h> // the ceil() definition must precede intrin.h +# if _MSC_VER > 1310 && (defined(_M_X64) || defined(_M_IX86)) +# include <intrin.h> +# define USE_MSC_INTRINSICS +# endif +#endif + +#ifdef __cplusplus +extern "C" { +#endif + +// These versions of get_msb() are only valid when n != 0 because all +// of the optimized versions are undefined when n == 0: +// https://gcc.gnu.org/onlinedocs/gcc/Other-Builtins.html + +// use GNU builtins where available. +#if defined(__GNUC__) && \ + ((__GNUC__ == 3 && __GNUC_MINOR__ >= 4) || __GNUC__ >= 4) +static INLINE int get_msb(unsigned int n) { + assert(n != 0); + return 31 ^ __builtin_clz(n); +} +#elif defined(USE_MSC_INTRINSICS) +#pragma intrinsic(_BitScanReverse) + +static INLINE int get_msb(unsigned int n) { + unsigned long first_set_bit; + assert(n != 0); + _BitScanReverse(&first_set_bit, n); + return first_set_bit; +} +#undef USE_MSC_INTRINSICS +#else +// Returns (int)floor(log2(n)). n must be > 0. +static INLINE int get_msb(unsigned int n) { + int log = 0; + unsigned int value = n; + int i; + + assert(n != 0); + + for (i = 4; i >= 0; --i) { + const int shift = (1 << i); + const unsigned int x = value >> shift; + if (x != 0) { + value = x; + log += shift; + } + } + return log; +} +#endif + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VPX_PORTS_BITOPS_H_ diff --git a/thirdparty/libvpx/vpx_ports/config.h b/thirdparty/libvpx/vpx_ports/config.h new file mode 100644 index 0000000000..3c1ab99f4a --- /dev/null +++ b/thirdparty/libvpx/vpx_ports/config.h @@ -0,0 +1,16 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VPX_PORTS_CONFIG_H_ +#define VPX_PORTS_CONFIG_H_ + +#include "vpx_config.h" + +#endif // VPX_PORTS_CONFIG_H_ diff --git a/thirdparty/libvpx/vpx_ports/emmintrin_compat.h b/thirdparty/libvpx/vpx_ports/emmintrin_compat.h new file mode 100644 index 0000000000..16176383d2 --- /dev/null +++ b/thirdparty/libvpx/vpx_ports/emmintrin_compat.h @@ -0,0 +1,55 @@ +/* + * Copyright (c) 2012 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VPX_PORTS_EMMINTRIN_COMPAT_H_ +#define VPX_PORTS_EMMINTRIN_COMPAT_H_ + +#if defined(__GNUC__) && __GNUC__ < 4 +/* From emmintrin.h (gcc 4.5.3) */ +/* Casts between various SP, DP, INT vector types. Note that these do no + conversion of values, they just change the type. */ +extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) +_mm_castpd_ps(__m128d __A) +{ + return (__m128) __A; +} + +extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) +_mm_castpd_si128(__m128d __A) +{ + return (__m128i) __A; +} + +extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) +_mm_castps_pd(__m128 __A) +{ + return (__m128d) __A; +} + +extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) +_mm_castps_si128(__m128 __A) +{ + return (__m128i) __A; +} + +extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) +_mm_castsi128_ps(__m128i __A) +{ + return (__m128) __A; +} + +extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) +_mm_castsi128_pd(__m128i __A) +{ + return (__m128d) __A; +} +#endif + +#endif // VPX_PORTS_EMMINTRIN_COMPAT_H_ diff --git a/thirdparty/libvpx/vpx_ports/emms.asm b/thirdparty/libvpx/vpx_ports/emms.asm new file mode 100644 index 0000000000..db8da28737 --- /dev/null +++ b/thirdparty/libvpx/vpx_ports/emms.asm @@ -0,0 +1,38 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + + +%include "vpx_ports/x86_abi_support.asm" + +section .text +global sym(vpx_reset_mmx_state) PRIVATE +sym(vpx_reset_mmx_state): + emms + ret + + +%if LIBVPX_YASM_WIN64 +global sym(vpx_winx64_fldcw) PRIVATE +sym(vpx_winx64_fldcw): + sub rsp, 8 + mov [rsp], rcx ; win x64 specific + fldcw [rsp] + add rsp, 8 + ret + + +global sym(vpx_winx64_fstcw) PRIVATE +sym(vpx_winx64_fstcw): + sub rsp, 8 + fstcw [rsp] + mov rax, [rsp] + add rsp, 8 + ret +%endif diff --git a/thirdparty/libvpx/vpx_ports/mem.h b/thirdparty/libvpx/vpx_ports/mem.h new file mode 100644 index 0000000000..7502f90632 --- /dev/null +++ b/thirdparty/libvpx/vpx_ports/mem.h @@ -0,0 +1,53 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VPX_PORTS_MEM_H_ +#define VPX_PORTS_MEM_H_ + +#include "vpx_config.h" +#include "vpx/vpx_integer.h" + +#if (defined(__GNUC__) && __GNUC__) || defined(__SUNPRO_C) +#define DECLARE_ALIGNED(n,typ,val) typ val __attribute__ ((aligned (n))) +#elif defined(_MSC_VER) +#define DECLARE_ALIGNED(n,typ,val) __declspec(align(n)) typ val +#else +#warning No alignment directives known for this compiler. +#define DECLARE_ALIGNED(n,typ,val) typ val +#endif + +/* Indicates that the usage of the specified variable has been audited to assure + * that it's safe to use uninitialized. Silences 'may be used uninitialized' + * warnings on gcc. + */ +#if defined(__GNUC__) && __GNUC__ +#define UNINITIALIZED_IS_SAFE(x) x=x +#else +#define UNINITIALIZED_IS_SAFE(x) x +#endif + +#if HAVE_NEON && defined(_MSC_VER) +#define __builtin_prefetch(x) +#endif + +/* Shift down with rounding */ +#define ROUND_POWER_OF_TWO(value, n) \ + (((value) + (1 << ((n) - 1))) >> (n)) + +#define ALIGN_POWER_OF_TWO(value, n) \ + (((value) + ((1 << (n)) - 1)) & ~((1 << (n)) - 1)) + +#if CONFIG_VP9_HIGHBITDEPTH +#define CONVERT_TO_SHORTPTR(x) ((uint16_t*)(((uintptr_t)x) << 1)) +#define CONVERT_TO_BYTEPTR(x) ((uint8_t*)(((uintptr_t)x) >> 1)) +#endif // CONFIG_VP9_HIGHBITDEPTH + +#endif // VPX_PORTS_MEM_H_ diff --git a/thirdparty/libvpx/vpx_ports/mem_ops.h b/thirdparty/libvpx/vpx_ports/mem_ops.h new file mode 100644 index 0000000000..620df31b22 --- /dev/null +++ b/thirdparty/libvpx/vpx_ports/mem_ops.h @@ -0,0 +1,226 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VPX_PORTS_MEM_OPS_H_ +#define VPX_PORTS_MEM_OPS_H_ + +/* \file + * \brief Provides portable memory access primitives + * + * This function provides portable primitives for getting and setting of + * signed and unsigned integers in 16, 24, and 32 bit sizes. The operations + * can be performed on unaligned data regardless of hardware support for + * unaligned accesses. + * + * The type used to pass the integral values may be changed by defining + * MEM_VALUE_T with the appropriate type. The type given must be an integral + * numeric type. + * + * The actual functions instantiated have the MEM_VALUE_T type name pasted + * on to the symbol name. This allows the developer to instantiate these + * operations for multiple types within the same translation unit. This is + * of somewhat questionable utility, but the capability exists nonetheless. + * Users not making use of this functionality should call the functions + * without the type name appended, and the preprocessor will take care of + * it. + * + * NOTE: This code is not supported on platforms where char > 1 octet ATM. + */ + +#ifndef MAU_T +/* Minimum Access Unit for this target */ +#define MAU_T unsigned char +#endif + +#ifndef MEM_VALUE_T +#define MEM_VALUE_T int +#endif + +#undef MEM_VALUE_T_SZ_BITS +#define MEM_VALUE_T_SZ_BITS (sizeof(MEM_VALUE_T) << 3) + +#undef mem_ops_wrap_symbol +#define mem_ops_wrap_symbol(fn) mem_ops_wrap_symbol2(fn, MEM_VALUE_T) +#undef mem_ops_wrap_symbol2 +#define mem_ops_wrap_symbol2(fn,typ) mem_ops_wrap_symbol3(fn,typ) +#undef mem_ops_wrap_symbol3 +#define mem_ops_wrap_symbol3(fn,typ) fn##_as_##typ + +/* + * Include aligned access routines + */ +#define INCLUDED_BY_MEM_OPS_H +#include "mem_ops_aligned.h" +#undef INCLUDED_BY_MEM_OPS_H + +#undef mem_get_be16 +#define mem_get_be16 mem_ops_wrap_symbol(mem_get_be16) +static unsigned MEM_VALUE_T mem_get_be16(const void *vmem) { + unsigned MEM_VALUE_T val; + const MAU_T *mem = (const MAU_T *)vmem; + + val = mem[0] << 8; + val |= mem[1]; + return val; +} + +#undef mem_get_be24 +#define mem_get_be24 mem_ops_wrap_symbol(mem_get_be24) +static unsigned MEM_VALUE_T mem_get_be24(const void *vmem) { + unsigned MEM_VALUE_T val; + const MAU_T *mem = (const MAU_T *)vmem; + + val = mem[0] << 16; + val |= mem[1] << 8; + val |= mem[2]; + return val; +} + +#undef mem_get_be32 +#define mem_get_be32 mem_ops_wrap_symbol(mem_get_be32) +static unsigned MEM_VALUE_T mem_get_be32(const void *vmem) { + unsigned MEM_VALUE_T val; + const MAU_T *mem = (const MAU_T *)vmem; + + val = ((unsigned MEM_VALUE_T)mem[0]) << 24; + val |= mem[1] << 16; + val |= mem[2] << 8; + val |= mem[3]; + return val; +} + +#undef mem_get_le16 +#define mem_get_le16 mem_ops_wrap_symbol(mem_get_le16) +static unsigned MEM_VALUE_T mem_get_le16(const void *vmem) { + unsigned MEM_VALUE_T val; + const MAU_T *mem = (const MAU_T *)vmem; + + val = mem[1] << 8; + val |= mem[0]; + return val; +} + +#undef mem_get_le24 +#define mem_get_le24 mem_ops_wrap_symbol(mem_get_le24) +static unsigned MEM_VALUE_T mem_get_le24(const void *vmem) { + unsigned MEM_VALUE_T val; + const MAU_T *mem = (const MAU_T *)vmem; + + val = mem[2] << 16; + val |= mem[1] << 8; + val |= mem[0]; + return val; +} + +#undef mem_get_le32 +#define mem_get_le32 mem_ops_wrap_symbol(mem_get_le32) +static unsigned MEM_VALUE_T mem_get_le32(const void *vmem) { + unsigned MEM_VALUE_T val; + const MAU_T *mem = (const MAU_T *)vmem; + + val = ((unsigned MEM_VALUE_T)mem[3]) << 24; + val |= mem[2] << 16; + val |= mem[1] << 8; + val |= mem[0]; + return val; +} + +#define mem_get_s_generic(end,sz) \ + static VPX_INLINE signed MEM_VALUE_T mem_get_s##end##sz(const void *vmem) {\ + const MAU_T *mem = (const MAU_T*)vmem;\ + signed MEM_VALUE_T val = mem_get_##end##sz(mem);\ + return (val << (MEM_VALUE_T_SZ_BITS - sz)) >> (MEM_VALUE_T_SZ_BITS - sz);\ + } + +#undef mem_get_sbe16 +#define mem_get_sbe16 mem_ops_wrap_symbol(mem_get_sbe16) +mem_get_s_generic(be, 16) + +#undef mem_get_sbe24 +#define mem_get_sbe24 mem_ops_wrap_symbol(mem_get_sbe24) +mem_get_s_generic(be, 24) + +#undef mem_get_sbe32 +#define mem_get_sbe32 mem_ops_wrap_symbol(mem_get_sbe32) +mem_get_s_generic(be, 32) + +#undef mem_get_sle16 +#define mem_get_sle16 mem_ops_wrap_symbol(mem_get_sle16) +mem_get_s_generic(le, 16) + +#undef mem_get_sle24 +#define mem_get_sle24 mem_ops_wrap_symbol(mem_get_sle24) +mem_get_s_generic(le, 24) + +#undef mem_get_sle32 +#define mem_get_sle32 mem_ops_wrap_symbol(mem_get_sle32) +mem_get_s_generic(le, 32) + +#undef mem_put_be16 +#define mem_put_be16 mem_ops_wrap_symbol(mem_put_be16) +static VPX_INLINE void mem_put_be16(void *vmem, MEM_VALUE_T val) { + MAU_T *mem = (MAU_T *)vmem; + + mem[0] = (MAU_T)((val >> 8) & 0xff); + mem[1] = (MAU_T)((val >> 0) & 0xff); +} + +#undef mem_put_be24 +#define mem_put_be24 mem_ops_wrap_symbol(mem_put_be24) +static VPX_INLINE void mem_put_be24(void *vmem, MEM_VALUE_T val) { + MAU_T *mem = (MAU_T *)vmem; + + mem[0] = (MAU_T)((val >> 16) & 0xff); + mem[1] = (MAU_T)((val >> 8) & 0xff); + mem[2] = (MAU_T)((val >> 0) & 0xff); +} + +#undef mem_put_be32 +#define mem_put_be32 mem_ops_wrap_symbol(mem_put_be32) +static VPX_INLINE void mem_put_be32(void *vmem, MEM_VALUE_T val) { + MAU_T *mem = (MAU_T *)vmem; + + mem[0] = (MAU_T)((val >> 24) & 0xff); + mem[1] = (MAU_T)((val >> 16) & 0xff); + mem[2] = (MAU_T)((val >> 8) & 0xff); + mem[3] = (MAU_T)((val >> 0) & 0xff); +} + +#undef mem_put_le16 +#define mem_put_le16 mem_ops_wrap_symbol(mem_put_le16) +static VPX_INLINE void mem_put_le16(void *vmem, MEM_VALUE_T val) { + MAU_T *mem = (MAU_T *)vmem; + + mem[0] = (MAU_T)((val >> 0) & 0xff); + mem[1] = (MAU_T)((val >> 8) & 0xff); +} + +#undef mem_put_le24 +#define mem_put_le24 mem_ops_wrap_symbol(mem_put_le24) +static VPX_INLINE void mem_put_le24(void *vmem, MEM_VALUE_T val) { + MAU_T *mem = (MAU_T *)vmem; + + mem[0] = (MAU_T)((val >> 0) & 0xff); + mem[1] = (MAU_T)((val >> 8) & 0xff); + mem[2] = (MAU_T)((val >> 16) & 0xff); +} + +#undef mem_put_le32 +#define mem_put_le32 mem_ops_wrap_symbol(mem_put_le32) +static VPX_INLINE void mem_put_le32(void *vmem, MEM_VALUE_T val) { + MAU_T *mem = (MAU_T *)vmem; + + mem[0] = (MAU_T)((val >> 0) & 0xff); + mem[1] = (MAU_T)((val >> 8) & 0xff); + mem[2] = (MAU_T)((val >> 16) & 0xff); + mem[3] = (MAU_T)((val >> 24) & 0xff); +} + +#endif // VPX_PORTS_MEM_OPS_H_ diff --git a/thirdparty/libvpx/vpx_ports/mem_ops_aligned.h b/thirdparty/libvpx/vpx_ports/mem_ops_aligned.h new file mode 100644 index 0000000000..46f61738ba --- /dev/null +++ b/thirdparty/libvpx/vpx_ports/mem_ops_aligned.h @@ -0,0 +1,169 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VPX_PORTS_MEM_OPS_ALIGNED_H_ +#define VPX_PORTS_MEM_OPS_ALIGNED_H_ + +#include "vpx/vpx_integer.h" + +/* \file + * \brief Provides portable memory access primitives for operating on aligned + * data + * + * This file is split from mem_ops.h for easier maintenance. See mem_ops.h + * for a more detailed description of these primitives. + */ +#ifndef INCLUDED_BY_MEM_OPS_H +#error Include mem_ops.h, not mem_ops_aligned.h directly. +#endif + +/* Architectures that provide instructions for doing this byte swapping + * could redefine these macros. + */ +#define swap_endian_16(val,raw) do {\ + val = (uint16_t)(((raw>>8) & 0x00ff) \ + | ((raw<<8) & 0xff00));\ + } while(0) +#define swap_endian_32(val,raw) do {\ + val = ((raw>>24) & 0x000000ff) \ + | ((raw>>8) & 0x0000ff00) \ + | ((raw<<8) & 0x00ff0000) \ + | ((raw<<24) & 0xff000000); \ + } while(0) +#define swap_endian_16_se(val,raw) do {\ + swap_endian_16(val,raw);\ + val = ((val << 16) >> 16);\ + } while(0) +#define swap_endian_32_se(val,raw) swap_endian_32(val,raw) + +#define mem_get_ne_aligned_generic(end,sz) \ + static VPX_INLINE unsigned MEM_VALUE_T \ + mem_get_##end##sz##_aligned(const void *vmem) {\ + const uint##sz##_t *mem = (const uint##sz##_t *)vmem;\ + return *mem;\ + } + +#define mem_get_sne_aligned_generic(end,sz) \ + static VPX_INLINE signed MEM_VALUE_T \ + mem_get_s##end##sz##_aligned(const void *vmem) {\ + const int##sz##_t *mem = (const int##sz##_t *)vmem;\ + return *mem;\ + } + +#define mem_get_se_aligned_generic(end,sz) \ + static VPX_INLINE unsigned MEM_VALUE_T \ + mem_get_##end##sz##_aligned(const void *vmem) {\ + const uint##sz##_t *mem = (const uint##sz##_t *)vmem;\ + unsigned MEM_VALUE_T val, raw = *mem;\ + swap_endian_##sz(val,raw);\ + return val;\ + } + +#define mem_get_sse_aligned_generic(end,sz) \ + static VPX_INLINE signed MEM_VALUE_T \ + mem_get_s##end##sz##_aligned(const void *vmem) {\ + const int##sz##_t *mem = (const int##sz##_t *)vmem;\ + unsigned MEM_VALUE_T val, raw = *mem;\ + swap_endian_##sz##_se(val,raw);\ + return val;\ + } + +#define mem_put_ne_aligned_generic(end,sz) \ + static VPX_INLINE void \ + mem_put_##end##sz##_aligned(void *vmem, MEM_VALUE_T val) {\ + uint##sz##_t *mem = (uint##sz##_t *)vmem;\ + *mem = (uint##sz##_t)val;\ + } + +#define mem_put_se_aligned_generic(end,sz) \ + static VPX_INLINE void \ + mem_put_##end##sz##_aligned(void *vmem, MEM_VALUE_T val) {\ + uint##sz##_t *mem = (uint##sz##_t *)vmem, raw;\ + swap_endian_##sz(raw,val);\ + *mem = (uint##sz##_t)raw;\ + } + +#include "vpx_config.h" +#if CONFIG_BIG_ENDIAN +#define mem_get_be_aligned_generic(sz) mem_get_ne_aligned_generic(be,sz) +#define mem_get_sbe_aligned_generic(sz) mem_get_sne_aligned_generic(be,sz) +#define mem_get_le_aligned_generic(sz) mem_get_se_aligned_generic(le,sz) +#define mem_get_sle_aligned_generic(sz) mem_get_sse_aligned_generic(le,sz) +#define mem_put_be_aligned_generic(sz) mem_put_ne_aligned_generic(be,sz) +#define mem_put_le_aligned_generic(sz) mem_put_se_aligned_generic(le,sz) +#else +#define mem_get_be_aligned_generic(sz) mem_get_se_aligned_generic(be,sz) +#define mem_get_sbe_aligned_generic(sz) mem_get_sse_aligned_generic(be,sz) +#define mem_get_le_aligned_generic(sz) mem_get_ne_aligned_generic(le,sz) +#define mem_get_sle_aligned_generic(sz) mem_get_sne_aligned_generic(le,sz) +#define mem_put_be_aligned_generic(sz) mem_put_se_aligned_generic(be,sz) +#define mem_put_le_aligned_generic(sz) mem_put_ne_aligned_generic(le,sz) +#endif + +#undef mem_get_be16_aligned +#define mem_get_be16_aligned mem_ops_wrap_symbol(mem_get_be16_aligned) +mem_get_be_aligned_generic(16) + +#undef mem_get_be32_aligned +#define mem_get_be32_aligned mem_ops_wrap_symbol(mem_get_be32_aligned) +mem_get_be_aligned_generic(32) + +#undef mem_get_le16_aligned +#define mem_get_le16_aligned mem_ops_wrap_symbol(mem_get_le16_aligned) +mem_get_le_aligned_generic(16) + +#undef mem_get_le32_aligned +#define mem_get_le32_aligned mem_ops_wrap_symbol(mem_get_le32_aligned) +mem_get_le_aligned_generic(32) + +#undef mem_get_sbe16_aligned +#define mem_get_sbe16_aligned mem_ops_wrap_symbol(mem_get_sbe16_aligned) +mem_get_sbe_aligned_generic(16) + +#undef mem_get_sbe32_aligned +#define mem_get_sbe32_aligned mem_ops_wrap_symbol(mem_get_sbe32_aligned) +mem_get_sbe_aligned_generic(32) + +#undef mem_get_sle16_aligned +#define mem_get_sle16_aligned mem_ops_wrap_symbol(mem_get_sle16_aligned) +mem_get_sle_aligned_generic(16) + +#undef mem_get_sle32_aligned +#define mem_get_sle32_aligned mem_ops_wrap_symbol(mem_get_sle32_aligned) +mem_get_sle_aligned_generic(32) + +#undef mem_put_be16_aligned +#define mem_put_be16_aligned mem_ops_wrap_symbol(mem_put_be16_aligned) +mem_put_be_aligned_generic(16) + +#undef mem_put_be32_aligned +#define mem_put_be32_aligned mem_ops_wrap_symbol(mem_put_be32_aligned) +mem_put_be_aligned_generic(32) + +#undef mem_put_le16_aligned +#define mem_put_le16_aligned mem_ops_wrap_symbol(mem_put_le16_aligned) +mem_put_le_aligned_generic(16) + +#undef mem_put_le32_aligned +#define mem_put_le32_aligned mem_ops_wrap_symbol(mem_put_le32_aligned) +mem_put_le_aligned_generic(32) + +#undef mem_get_ne_aligned_generic +#undef mem_get_se_aligned_generic +#undef mem_get_sne_aligned_generic +#undef mem_get_sse_aligned_generic +#undef mem_put_ne_aligned_generic +#undef mem_put_se_aligned_generic +#undef swap_endian_16 +#undef swap_endian_32 +#undef swap_endian_16_se +#undef swap_endian_32_se + +#endif // VPX_PORTS_MEM_OPS_ALIGNED_H_ diff --git a/thirdparty/libvpx/vpx_ports/msvc.h b/thirdparty/libvpx/vpx_ports/msvc.h new file mode 100644 index 0000000000..cab77405f4 --- /dev/null +++ b/thirdparty/libvpx/vpx_ports/msvc.h @@ -0,0 +1,32 @@ +/* + * Copyright (c) 2015 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VPX_PORTS_MSVC_H_ +#define VPX_PORTS_MSVC_H_ +#ifdef _MSC_VER + +#include "./vpx_config.h" + +# if _MSC_VER < 1900 // VS2015 provides snprintf +# define snprintf _snprintf +# endif // _MSC_VER < 1900 + +#if _MSC_VER < 1800 // VS2013 provides round +#include <math.h> +static INLINE double round(double x) { + if (x < 0) + return ceil(x - 0.5); + else + return floor(x + 0.5); +} +#endif // _MSC_VER < 1800 + +#endif // _MSC_VER +#endif // VPX_PORTS_MSVC_H_ diff --git a/thirdparty/libvpx/vpx_ports/system_state.h b/thirdparty/libvpx/vpx_ports/system_state.h new file mode 100644 index 0000000000..086c64681f --- /dev/null +++ b/thirdparty/libvpx/vpx_ports/system_state.h @@ -0,0 +1,22 @@ +/* + * Copyright (c) 2015 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VPX_PORTS_SYSTEM_STATE_H_ +#define VPX_PORTS_SYSTEM_STATE_H_ + +#include "./vpx_config.h" + +#if ARCH_X86 || ARCH_X86_64 +void vpx_reset_mmx_state(void); +#define vpx_clear_system_state() vpx_reset_mmx_state() +#else +#define vpx_clear_system_state() +#endif // ARCH_X86 || ARCH_X86_64 +#endif // VPX_PORTS_SYSTEM_STATE_H_ diff --git a/thirdparty/libvpx/vpx_ports/vpx_once.h b/thirdparty/libvpx/vpx_ports/vpx_once.h new file mode 100644 index 0000000000..da04db4590 --- /dev/null +++ b/thirdparty/libvpx/vpx_ports/vpx_once.h @@ -0,0 +1,150 @@ +/* + * Copyright (c) 2015 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VPX_PORTS_VPX_ONCE_H_ +#define VPX_PORTS_VPX_ONCE_H_ + +#include "vpx_config.h" + +/* Implement a function wrapper to guarantee initialization + * thread-safety for library singletons. + * + * NOTE: These functions use static locks, and can only be + * used with one common argument per compilation unit. So + * + * file1.c: + * vpx_once(foo); + * ... + * vpx_once(foo); + * + * file2.c: + * vpx_once(bar); + * + * will ensure foo() and bar() are each called only once, but in + * + * file1.c: + * vpx_once(foo); + * vpx_once(bar): + * + * bar() will never be called because the lock is used up + * by the call to foo(). + */ + +#if CONFIG_MULTITHREAD && defined(_WIN32) +#include <windows.h> +#include <stdlib.h> +/* Declare a per-compilation-unit state variable to track the progress + * of calling func() only once. This must be at global scope because + * local initializers are not thread-safe in MSVC prior to Visual + * Studio 2015. + * + * As a static, once_state will be zero-initialized as program start. + */ +static LONG once_state; +static void once(void (*func)(void)) +{ + /* Try to advance once_state from its initial value of 0 to 1. + * Only one thread can succeed in doing so. + */ + if (InterlockedCompareExchange(&once_state, 1, 0) == 0) { + /* We're the winning thread, having set once_state to 1. + * Call our function. */ + func(); + /* Now advance once_state to 2, unblocking any other threads. */ + InterlockedIncrement(&once_state); + return; + } + + /* We weren't the winning thread, but we want to block on + * the state variable so we don't return before func() + * has finished executing elsewhere. + * + * Try to advance once_state from 2 to 2, which is only possible + * after the winning thead advances it from 1 to 2. + */ + while (InterlockedCompareExchange(&once_state, 2, 2) != 2) { + /* State isn't yet 2. Try again. + * + * We are used for singleton initialization functions, + * which should complete quickly. Contention will likewise + * be rare, so it's worthwhile to use a simple but cpu- + * intensive busy-wait instead of successive backoff, + * waiting on a kernel object, or another heavier-weight scheme. + * + * We can at least yield our timeslice. + */ + Sleep(0); + } + + /* We've seen once_state advance to 2, so we know func() + * has been called. And we've left once_state as we found it, + * so other threads will have the same experience. + * + * It's safe to return now. + */ + return; +} + + +#elif CONFIG_MULTITHREAD && defined(__OS2__) +#define INCL_DOS +#include <os2.h> +static void once(void (*func)(void)) +{ + static int done; + + /* If the initialization is complete, return early. */ + if(done) + return; + + /* Causes all other threads in the process to block themselves + * and give up their time slice. + */ + DosEnterCritSec(); + + if (!done) + { + func(); + done = 1; + } + + /* Restores normal thread dispatching for the current process. */ + DosExitCritSec(); +} + + +#elif CONFIG_MULTITHREAD && HAVE_PTHREAD_H +#include <pthread.h> +static void once(void (*func)(void)) +{ + static pthread_once_t lock = PTHREAD_ONCE_INIT; + pthread_once(&lock, func); +} + + +#else +/* No-op version that performs no synchronization. *_rtcd() is idempotent, + * so as long as your platform provides atomic loads/stores of pointers + * no synchronization is strictly necessary. + */ + +static void once(void (*func)(void)) +{ + static int done; + + if(!done) + { + func(); + done = 1; + } +} +#endif + +#endif // VPX_PORTS_VPX_ONCE_H_ diff --git a/thirdparty/libvpx/vpx_ports/vpx_timer.h b/thirdparty/libvpx/vpx_ports/vpx_timer.h new file mode 100644 index 0000000000..dd98e291c2 --- /dev/null +++ b/thirdparty/libvpx/vpx_ports/vpx_timer.h @@ -0,0 +1,120 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VPX_PORTS_VPX_TIMER_H_ +#define VPX_PORTS_VPX_TIMER_H_ + +#include "./vpx_config.h" + +#include "vpx/vpx_integer.h" + +#if CONFIG_OS_SUPPORT + +#if defined(_WIN32) +/* + * Win32 specific includes + */ +#ifndef WIN32_LEAN_AND_MEAN +#define WIN32_LEAN_AND_MEAN +#endif +#include <windows.h> +#else +/* + * POSIX specific includes + */ +#include <sys/time.h> + +/* timersub is not provided by msys at this time. */ +#ifndef timersub +#define timersub(a, b, result) \ + do { \ + (result)->tv_sec = (a)->tv_sec - (b)->tv_sec; \ + (result)->tv_usec = (a)->tv_usec - (b)->tv_usec; \ + if ((result)->tv_usec < 0) { \ + --(result)->tv_sec; \ + (result)->tv_usec += 1000000; \ + } \ + } while (0) +#endif +#endif + + +struct vpx_usec_timer { +#if defined(_WIN32) + LARGE_INTEGER begin, end; +#else + struct timeval begin, end; +#endif +}; + + +static INLINE void +vpx_usec_timer_start(struct vpx_usec_timer *t) { +#if defined(_WIN32) + QueryPerformanceCounter(&t->begin); +#else + gettimeofday(&t->begin, NULL); +#endif +} + + +static INLINE void +vpx_usec_timer_mark(struct vpx_usec_timer *t) { +#if defined(_WIN32) + QueryPerformanceCounter(&t->end); +#else + gettimeofday(&t->end, NULL); +#endif +} + + +static INLINE int64_t +vpx_usec_timer_elapsed(struct vpx_usec_timer *t) { +#if defined(_WIN32) + LARGE_INTEGER freq, diff; + + diff.QuadPart = t->end.QuadPart - t->begin.QuadPart; + + QueryPerformanceFrequency(&freq); + return diff.QuadPart * 1000000 / freq.QuadPart; +#else + struct timeval diff; + + timersub(&t->end, &t->begin, &diff); + return diff.tv_sec * 1000000 + diff.tv_usec; +#endif +} + +#else /* CONFIG_OS_SUPPORT = 0*/ + +/* Empty timer functions if CONFIG_OS_SUPPORT = 0 */ +#ifndef timersub +#define timersub(a, b, result) +#endif + +struct vpx_usec_timer { + void *dummy; +}; + +static INLINE void +vpx_usec_timer_start(struct vpx_usec_timer *t) { } + +static INLINE void +vpx_usec_timer_mark(struct vpx_usec_timer *t) { } + +static INLINE int +vpx_usec_timer_elapsed(struct vpx_usec_timer *t) { + return 0; +} + +#endif /* CONFIG_OS_SUPPORT */ + +#endif // VPX_PORTS_VPX_TIMER_H_ diff --git a/thirdparty/libvpx/vpx_ports/x86.h b/thirdparty/libvpx/vpx_ports/x86.h new file mode 100644 index 0000000000..bae25ac345 --- /dev/null +++ b/thirdparty/libvpx/vpx_ports/x86.h @@ -0,0 +1,330 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VPX_PORTS_X86_H_ +#define VPX_PORTS_X86_H_ +#include <stdlib.h> + +#if defined(_MSC_VER) +#include <intrin.h> /* For __cpuidex, __rdtsc */ +#endif + +#include "vpx_config.h" +#include "vpx/vpx_integer.h" + +#ifdef __cplusplus +extern "C" { +#endif + +typedef enum { + VPX_CPU_UNKNOWN = -1, + VPX_CPU_AMD, + VPX_CPU_AMD_OLD, + VPX_CPU_CENTAUR, + VPX_CPU_CYRIX, + VPX_CPU_INTEL, + VPX_CPU_NEXGEN, + VPX_CPU_NSC, + VPX_CPU_RISE, + VPX_CPU_SIS, + VPX_CPU_TRANSMETA, + VPX_CPU_TRANSMETA_OLD, + VPX_CPU_UMC, + VPX_CPU_VIA, + + VPX_CPU_LAST +} vpx_cpu_t; + +#if defined(__GNUC__) && __GNUC__ || defined(__ANDROID__) +#if ARCH_X86_64 +#define cpuid(func, func2, ax, bx, cx, dx)\ + __asm__ __volatile__ (\ + "cpuid \n\t" \ + : "=a" (ax), "=b" (bx), "=c" (cx), "=d" (dx) \ + : "a" (func), "c" (func2)); +#else +#define cpuid(func, func2, ax, bx, cx, dx)\ + __asm__ __volatile__ (\ + "mov %%ebx, %%edi \n\t" \ + "cpuid \n\t" \ + "xchg %%edi, %%ebx \n\t" \ + : "=a" (ax), "=D" (bx), "=c" (cx), "=d" (dx) \ + : "a" (func), "c" (func2)); +#endif +#elif defined(__SUNPRO_C) || defined(__SUNPRO_CC) /* end __GNUC__ or __ANDROID__*/ +#if ARCH_X86_64 +#define cpuid(func, func2, ax, bx, cx, dx)\ + asm volatile (\ + "xchg %rsi, %rbx \n\t" \ + "cpuid \n\t" \ + "movl %ebx, %edi \n\t" \ + "xchg %rsi, %rbx \n\t" \ + : "=a" (ax), "=D" (bx), "=c" (cx), "=d" (dx) \ + : "a" (func), "c" (func2)); +#else +#define cpuid(func, func2, ax, bx, cx, dx)\ + asm volatile (\ + "pushl %ebx \n\t" \ + "cpuid \n\t" \ + "movl %ebx, %edi \n\t" \ + "popl %ebx \n\t" \ + : "=a" (ax), "=D" (bx), "=c" (cx), "=d" (dx) \ + : "a" (func), "c" (func2)); +#endif +#else /* end __SUNPRO__ */ +#if ARCH_X86_64 +#if defined(_MSC_VER) && _MSC_VER > 1500 +#define cpuid(func, func2, a, b, c, d) do {\ + int regs[4];\ + __cpuidex(regs, func, func2); \ + a = regs[0]; b = regs[1]; c = regs[2]; d = regs[3];\ + } while(0) +#else +#define cpuid(func, func2, a, b, c, d) do {\ + int regs[4];\ + __cpuid(regs, func); \ + a = regs[0]; b = regs[1]; c = regs[2]; d = regs[3];\ + } while (0) +#endif +#else +#define cpuid(func, func2, a, b, c, d)\ + __asm mov eax, func\ + __asm mov ecx, func2\ + __asm cpuid\ + __asm mov a, eax\ + __asm mov b, ebx\ + __asm mov c, ecx\ + __asm mov d, edx +#endif +#endif /* end others */ + +// NaCl has no support for xgetbv or the raw opcode. +#if !defined(__native_client__) && (defined(__i386__) || defined(__x86_64__)) +static INLINE uint64_t xgetbv(void) { + const uint32_t ecx = 0; + uint32_t eax, edx; + // Use the raw opcode for xgetbv for compatibility with older toolchains. + __asm__ volatile ( + ".byte 0x0f, 0x01, 0xd0\n" + : "=a"(eax), "=d"(edx) : "c" (ecx)); + return ((uint64_t)edx << 32) | eax; +} +#elif (defined(_M_X64) || defined(_M_IX86)) && \ + defined(_MSC_FULL_VER) && _MSC_FULL_VER >= 160040219 // >= VS2010 SP1 +#include <immintrin.h> +#define xgetbv() _xgetbv(0) +#elif defined(_MSC_VER) && defined(_M_IX86) +static INLINE uint64_t xgetbv(void) { + uint32_t eax_, edx_; + __asm { + xor ecx, ecx // ecx = 0 + // Use the raw opcode for xgetbv for compatibility with older toolchains. + __asm _emit 0x0f __asm _emit 0x01 __asm _emit 0xd0 + mov eax_, eax + mov edx_, edx + } + return ((uint64_t)edx_ << 32) | eax_; +} +#else +#define xgetbv() 0U // no AVX for older x64 or unrecognized toolchains. +#endif + +#if defined(_MSC_VER) && _MSC_VER >= 1700 +#include <windows.h> +#if WINAPI_FAMILY_PARTITION(WINAPI_FAMILY_APP) +#define getenv(x) NULL +#endif +#endif + +#define HAS_MMX 0x01 +#define HAS_SSE 0x02 +#define HAS_SSE2 0x04 +#define HAS_SSE3 0x08 +#define HAS_SSSE3 0x10 +#define HAS_SSE4_1 0x20 +#define HAS_AVX 0x40 +#define HAS_AVX2 0x80 +#ifndef BIT +#define BIT(n) (1<<n) +#endif + +static INLINE int +x86_simd_caps(void) { + unsigned int flags = 0; + unsigned int mask = ~0; + unsigned int max_cpuid_val, reg_eax, reg_ebx, reg_ecx, reg_edx; + char *env; + (void)reg_ebx; + + /* See if the CPU capabilities are being overridden by the environment */ + env = getenv("VPX_SIMD_CAPS"); + + if (env && *env) + return (int)strtol(env, NULL, 0); + + env = getenv("VPX_SIMD_CAPS_MASK"); + + if (env && *env) + mask = (unsigned int)strtoul(env, NULL, 0); + + /* Ensure that the CPUID instruction supports extended features */ + cpuid(0, 0, max_cpuid_val, reg_ebx, reg_ecx, reg_edx); + + if (max_cpuid_val < 1) + return 0; + + /* Get the standard feature flags */ + cpuid(1, 0, reg_eax, reg_ebx, reg_ecx, reg_edx); + + if (reg_edx & BIT(23)) flags |= HAS_MMX; + + if (reg_edx & BIT(25)) flags |= HAS_SSE; /* aka xmm */ + + if (reg_edx & BIT(26)) flags |= HAS_SSE2; /* aka wmt */ + + if (reg_ecx & BIT(0)) flags |= HAS_SSE3; + + if (reg_ecx & BIT(9)) flags |= HAS_SSSE3; + + if (reg_ecx & BIT(19)) flags |= HAS_SSE4_1; + + // bits 27 (OSXSAVE) & 28 (256-bit AVX) + if ((reg_ecx & (BIT(27) | BIT(28))) == (BIT(27) | BIT(28))) { + if ((xgetbv() & 0x6) == 0x6) { + flags |= HAS_AVX; + + if (max_cpuid_val >= 7) { + /* Get the leaf 7 feature flags. Needed to check for AVX2 support */ + cpuid(7, 0, reg_eax, reg_ebx, reg_ecx, reg_edx); + + if (reg_ebx & BIT(5)) flags |= HAS_AVX2; + } + } + } + + return flags & mask; +} + +// Note: +// 32-bit CPU cycle counter is light-weighted for most function performance +// measurement. For large function (CPU time > a couple of seconds), 64-bit +// counter should be used. +// 32-bit CPU cycle counter +static INLINE unsigned int +x86_readtsc(void) { +#if defined(__GNUC__) && __GNUC__ + unsigned int tsc; + __asm__ __volatile__("rdtsc\n\t":"=a"(tsc):); + return tsc; +#elif defined(__SUNPRO_C) || defined(__SUNPRO_CC) + unsigned int tsc; + asm volatile("rdtsc\n\t":"=a"(tsc):); + return tsc; +#else +#if ARCH_X86_64 + return (unsigned int)__rdtsc(); +#else + __asm rdtsc; +#endif +#endif +} +// 64-bit CPU cycle counter +static INLINE uint64_t +x86_readtsc64(void) { +#if defined(__GNUC__) && __GNUC__ + uint32_t hi, lo; + __asm__ __volatile__("rdtsc" : "=a"(lo), "=d"(hi)); + return ((uint64_t)hi << 32) | lo; +#elif defined(__SUNPRO_C) || defined(__SUNPRO_CC) + uint_t hi, lo; + asm volatile("rdtsc\n\t" : "=a"(lo), "=d"(hi)); + return ((uint64_t)hi << 32) | lo; +#else +#if ARCH_X86_64 + return (uint64_t)__rdtsc(); +#else + __asm rdtsc; +#endif +#endif +} + +#if defined(__GNUC__) && __GNUC__ +#define x86_pause_hint()\ + __asm__ __volatile__ ("pause \n\t") +#elif defined(__SUNPRO_C) || defined(__SUNPRO_CC) +#define x86_pause_hint()\ + asm volatile ("pause \n\t") +#else +#if ARCH_X86_64 +#define x86_pause_hint()\ + _mm_pause(); +#else +#define x86_pause_hint()\ + __asm pause +#endif +#endif + +#if defined(__GNUC__) && __GNUC__ +static void +x87_set_control_word(unsigned short mode) { + __asm__ __volatile__("fldcw %0" : : "m"(*&mode)); +} +static unsigned short +x87_get_control_word(void) { + unsigned short mode; + __asm__ __volatile__("fstcw %0\n\t":"=m"(*&mode):); + return mode; +} +#elif defined(__SUNPRO_C) || defined(__SUNPRO_CC) +static void +x87_set_control_word(unsigned short mode) { + asm volatile("fldcw %0" : : "m"(*&mode)); +} +static unsigned short +x87_get_control_word(void) { + unsigned short mode; + asm volatile("fstcw %0\n\t":"=m"(*&mode):); + return mode; +} +#elif ARCH_X86_64 +/* No fldcw intrinsics on Windows x64, punt to external asm */ +extern void vpx_winx64_fldcw(unsigned short mode); +extern unsigned short vpx_winx64_fstcw(void); +#define x87_set_control_word vpx_winx64_fldcw +#define x87_get_control_word vpx_winx64_fstcw +#else +static void +x87_set_control_word(unsigned short mode) { + __asm { fldcw mode } +} +static unsigned short +x87_get_control_word(void) { + unsigned short mode; + __asm { fstcw mode } + return mode; +} +#endif + +static INLINE unsigned int +x87_set_double_precision(void) { + unsigned int mode = x87_get_control_word(); + x87_set_control_word((mode&~0x300) | 0x200); + return mode; +} + + +extern void vpx_reset_mmx_state(void); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VPX_PORTS_X86_H_ diff --git a/thirdparty/libvpx/vpx_ports/x86_abi_support.asm b/thirdparty/libvpx/vpx_ports/x86_abi_support.asm new file mode 100644 index 0000000000..708fa101c5 --- /dev/null +++ b/thirdparty/libvpx/vpx_ports/x86_abi_support.asm @@ -0,0 +1,404 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + + +%include "vpx_config.asm" + +; 32/64 bit compatibility macros +; +; In general, we make the source use 64 bit syntax, then twiddle with it using +; the preprocessor to get the 32 bit syntax on 32 bit platforms. +; +%ifidn __OUTPUT_FORMAT__,elf32 +%define ABI_IS_32BIT 1 +%elifidn __OUTPUT_FORMAT__,macho32 +%define ABI_IS_32BIT 1 +%elifidn __OUTPUT_FORMAT__,win32 +%define ABI_IS_32BIT 1 +%elifidn __OUTPUT_FORMAT__,aout +%define ABI_IS_32BIT 1 +%else +%define ABI_IS_32BIT 0 +%endif + +%if ABI_IS_32BIT +%define rax eax +%define rbx ebx +%define rcx ecx +%define rdx edx +%define rsi esi +%define rdi edi +%define rsp esp +%define rbp ebp +%define movsxd mov +%macro movq 2 + %ifidn %1,eax + movd %1,%2 + %elifidn %2,eax + movd %1,%2 + %elifidn %1,ebx + movd %1,%2 + %elifidn %2,ebx + movd %1,%2 + %elifidn %1,ecx + movd %1,%2 + %elifidn %2,ecx + movd %1,%2 + %elifidn %1,edx + movd %1,%2 + %elifidn %2,edx + movd %1,%2 + %elifidn %1,esi + movd %1,%2 + %elifidn %2,esi + movd %1,%2 + %elifidn %1,edi + movd %1,%2 + %elifidn %2,edi + movd %1,%2 + %elifidn %1,esp + movd %1,%2 + %elifidn %2,esp + movd %1,%2 + %elifidn %1,ebp + movd %1,%2 + %elifidn %2,ebp + movd %1,%2 + %else + movq %1,%2 + %endif +%endmacro +%endif + + +; LIBVPX_YASM_WIN64 +; Set LIBVPX_YASM_WIN64 if output is Windows 64bit so the code will work if x64 +; or win64 is defined on the Yasm command line. +%ifidn __OUTPUT_FORMAT__,win64 +%define LIBVPX_YASM_WIN64 1 +%elifidn __OUTPUT_FORMAT__,x64 +%define LIBVPX_YASM_WIN64 1 +%else +%define LIBVPX_YASM_WIN64 0 +%endif + +; sym() +; Return the proper symbol name for the target ABI. +; +; Certain ABIs, notably MS COFF and Darwin MACH-O, require that symbols +; with C linkage be prefixed with an underscore. +; +%ifidn __OUTPUT_FORMAT__,elf32 +%define sym(x) x +%elifidn __OUTPUT_FORMAT__,elf64 +%define sym(x) x +%elifidn __OUTPUT_FORMAT__,elfx32 +%define sym(x) x +%elif LIBVPX_YASM_WIN64 +%define sym(x) x +%else +%define sym(x) _ %+ x +%endif + +; PRIVATE +; Macro for the attribute to hide a global symbol for the target ABI. +; This is only active if CHROMIUM is defined. +; +; Chromium doesn't like exported global symbols due to symbol clashing with +; plugins among other things. +; +; Requires Chromium's patched copy of yasm: +; http://src.chromium.org/viewvc/chrome?view=rev&revision=73761 +; http://www.tortall.net/projects/yasm/ticket/236 +; +%ifdef CHROMIUM + %ifidn __OUTPUT_FORMAT__,elf32 + %define PRIVATE :hidden + %elifidn __OUTPUT_FORMAT__,elf64 + %define PRIVATE :hidden + %elifidn __OUTPUT_FORMAT__,elfx32 + %define PRIVATE :hidden + %elif LIBVPX_YASM_WIN64 + %define PRIVATE + %else + %define PRIVATE :private_extern + %endif +%else + %define PRIVATE +%endif + +; arg() +; Return the address specification of the given argument +; +%if ABI_IS_32BIT + %define arg(x) [ebp+8+4*x] +%else + ; 64 bit ABI passes arguments in registers. This is a workaround to get up + ; and running quickly. Relies on SHADOW_ARGS_TO_STACK + %if LIBVPX_YASM_WIN64 + %define arg(x) [rbp+16+8*x] + %else + %define arg(x) [rbp-8-8*x] + %endif +%endif + +; REG_SZ_BYTES, REG_SZ_BITS +; Size of a register +%if ABI_IS_32BIT +%define REG_SZ_BYTES 4 +%define REG_SZ_BITS 32 +%else +%define REG_SZ_BYTES 8 +%define REG_SZ_BITS 64 +%endif + + +; ALIGN_STACK <alignment> <register> +; This macro aligns the stack to the given alignment (in bytes). The stack +; is left such that the previous value of the stack pointer is the first +; argument on the stack (ie, the inverse of this macro is 'pop rsp.') +; This macro uses one temporary register, which is not preserved, and thus +; must be specified as an argument. +%macro ALIGN_STACK 2 + mov %2, rsp + and rsp, -%1 + lea rsp, [rsp - (%1 - REG_SZ_BYTES)] + push %2 +%endmacro + + +; +; The Microsoft assembler tries to impose a certain amount of type safety in +; its register usage. YASM doesn't recognize these directives, so we just +; %define them away to maintain as much compatibility as possible with the +; original inline assembler we're porting from. +; +%idefine PTR +%idefine XMMWORD +%idefine MMWORD + +; PIC macros +; +%if ABI_IS_32BIT + %if CONFIG_PIC=1 + %ifidn __OUTPUT_FORMAT__,elf32 + %define WRT_PLT wrt ..plt + %macro GET_GOT 1 + extern _GLOBAL_OFFSET_TABLE_ + push %1 + call %%get_got + %%sub_offset: + jmp %%exitGG + %%get_got: + mov %1, [esp] + add %1, _GLOBAL_OFFSET_TABLE_ + $$ - %%sub_offset wrt ..gotpc + ret + %%exitGG: + %undef GLOBAL + %define GLOBAL(x) x + %1 wrt ..gotoff + %undef RESTORE_GOT + %define RESTORE_GOT pop %1 + %endmacro + %elifidn __OUTPUT_FORMAT__,macho32 + %macro GET_GOT 1 + push %1 + call %%get_got + %%get_got: + pop %1 + %undef GLOBAL + %define GLOBAL(x) x + %1 - %%get_got + %undef RESTORE_GOT + %define RESTORE_GOT pop %1 + %endmacro + %endif + %endif + + %ifdef CHROMIUM + %ifidn __OUTPUT_FORMAT__,macho32 + %define HIDDEN_DATA(x) x:private_extern + %else + %define HIDDEN_DATA(x) x + %endif + %else + %define HIDDEN_DATA(x) x + %endif +%else + %macro GET_GOT 1 + %endmacro + %define GLOBAL(x) rel x + %ifidn __OUTPUT_FORMAT__,elf64 + %define WRT_PLT wrt ..plt + %define HIDDEN_DATA(x) x:data hidden + %elifidn __OUTPUT_FORMAT__,elfx32 + %define WRT_PLT wrt ..plt + %define HIDDEN_DATA(x) x:data hidden + %elifidn __OUTPUT_FORMAT__,macho64 + %ifdef CHROMIUM + %define HIDDEN_DATA(x) x:private_extern + %else + %define HIDDEN_DATA(x) x + %endif + %else + %define HIDDEN_DATA(x) x + %endif +%endif +%ifnmacro GET_GOT + %macro GET_GOT 1 + %endmacro + %define GLOBAL(x) x +%endif +%ifndef RESTORE_GOT +%define RESTORE_GOT +%endif +%ifndef WRT_PLT +%define WRT_PLT +%endif + +%if ABI_IS_32BIT + %macro SHADOW_ARGS_TO_STACK 1 + %endm + %define UNSHADOW_ARGS +%else +%if LIBVPX_YASM_WIN64 + %macro SHADOW_ARGS_TO_STACK 1 ; argc + %if %1 > 0 + mov arg(0),rcx + %endif + %if %1 > 1 + mov arg(1),rdx + %endif + %if %1 > 2 + mov arg(2),r8 + %endif + %if %1 > 3 + mov arg(3),r9 + %endif + %endm +%else + %macro SHADOW_ARGS_TO_STACK 1 ; argc + %if %1 > 0 + push rdi + %endif + %if %1 > 1 + push rsi + %endif + %if %1 > 2 + push rdx + %endif + %if %1 > 3 + push rcx + %endif + %if %1 > 4 + push r8 + %endif + %if %1 > 5 + push r9 + %endif + %if %1 > 6 + %assign i %1-6 + %assign off 16 + %rep i + mov rax,[rbp+off] + push rax + %assign off off+8 + %endrep + %endif + %endm +%endif + %define UNSHADOW_ARGS mov rsp, rbp +%endif + +; Win64 ABI requires that XMM6:XMM15 are callee saved +; SAVE_XMM n, [u] +; store registers 6-n on the stack +; if u is specified, use unaligned movs. +; Win64 ABI requires 16 byte stack alignment, but then pushes an 8 byte return +; value. Typically we follow this up with 'push rbp' - re-aligning the stack - +; but in some cases this is not done and unaligned movs must be used. +%if LIBVPX_YASM_WIN64 +%macro SAVE_XMM 1-2 a + %if %1 < 6 + %error Only xmm registers 6-15 must be preserved + %else + %assign last_xmm %1 + %define movxmm movdq %+ %2 + %assign xmm_stack_space ((last_xmm - 5) * 16) + sub rsp, xmm_stack_space + %assign i 6 + %rep (last_xmm - 5) + movxmm [rsp + ((i - 6) * 16)], xmm %+ i + %assign i i+1 + %endrep + %endif +%endmacro +%macro RESTORE_XMM 0 + %ifndef last_xmm + %error RESTORE_XMM must be paired with SAVE_XMM n + %else + %assign i last_xmm + %rep (last_xmm - 5) + movxmm xmm %+ i, [rsp +((i - 6) * 16)] + %assign i i-1 + %endrep + add rsp, xmm_stack_space + ; there are a couple functions which return from multiple places. + ; otherwise, we could uncomment these: + ; %undef last_xmm + ; %undef xmm_stack_space + ; %undef movxmm + %endif +%endmacro +%else +%macro SAVE_XMM 1-2 +%endmacro +%macro RESTORE_XMM 0 +%endmacro +%endif + +; Name of the rodata section +; +; .rodata seems to be an elf-ism, as it doesn't work on OSX. +; +%ifidn __OUTPUT_FORMAT__,macho64 +%define SECTION_RODATA section .text +%elifidn __OUTPUT_FORMAT__,macho32 +%macro SECTION_RODATA 0 +section .text +%endmacro +%elifidn __OUTPUT_FORMAT__,aout +%define SECTION_RODATA section .data +%else +%define SECTION_RODATA section .rodata +%endif + + +; Tell GNU ld that we don't require an executable stack. +%ifidn __OUTPUT_FORMAT__,elf32 +section .note.GNU-stack noalloc noexec nowrite progbits +section .text +%elifidn __OUTPUT_FORMAT__,elf64 +section .note.GNU-stack noalloc noexec nowrite progbits +section .text +%elifidn __OUTPUT_FORMAT__,elfx32 +section .note.GNU-stack noalloc noexec nowrite progbits +section .text +%endif + +; On Android platforms use lrand48 when building postproc routines. Prior to L +; rand() was not available. +%if CONFIG_POSTPROC=1 || CONFIG_VP9_POSTPROC=1 +%ifdef __ANDROID__ +extern sym(lrand48) +%define LIBVPX_RAND lrand48 +%else +extern sym(rand) +%define LIBVPX_RAND rand +%endif +%endif ; CONFIG_POSTPROC || CONFIG_VP9_POSTPROC diff --git a/thirdparty/libvpx/vpx_scale/generic/yv12config.c b/thirdparty/libvpx/vpx_scale/generic/yv12config.c new file mode 100644 index 0000000000..6bbb6d8d48 --- /dev/null +++ b/thirdparty/libvpx/vpx_scale/generic/yv12config.c @@ -0,0 +1,287 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <assert.h> + +#include "vpx_scale/yv12config.h" +#include "vpx_mem/vpx_mem.h" +#include "vpx_ports/mem.h" + +/**************************************************************************** +* Exports +****************************************************************************/ + +/**************************************************************************** + * + ****************************************************************************/ +#define yv12_align_addr(addr, align) \ + (void*)(((size_t)(addr) + ((align) - 1)) & (size_t)-(align)) + +int +vp8_yv12_de_alloc_frame_buffer(YV12_BUFFER_CONFIG *ybf) { + if (ybf) { + // If libvpx is using frame buffer callbacks then buffer_alloc_sz must + // not be set. + if (ybf->buffer_alloc_sz > 0) { + vpx_free(ybf->buffer_alloc); + } + + /* buffer_alloc isn't accessed by most functions. Rather y_buffer, + u_buffer and v_buffer point to buffer_alloc and are used. Clear out + all of this so that a freed pointer isn't inadvertently used */ + memset(ybf, 0, sizeof(YV12_BUFFER_CONFIG)); + } else { + return -1; + } + + return 0; +} + +int vp8_yv12_realloc_frame_buffer(YV12_BUFFER_CONFIG *ybf, + int width, int height, int border) { + if (ybf) { + int aligned_width = (width + 15) & ~15; + int aligned_height = (height + 15) & ~15; + int y_stride = ((aligned_width + 2 * border) + 31) & ~31; + int yplane_size = (aligned_height + 2 * border) * y_stride; + int uv_width = aligned_width >> 1; + int uv_height = aligned_height >> 1; + /** There is currently a bunch of code which assumes + * uv_stride == y_stride/2, so enforce this here. */ + int uv_stride = y_stride >> 1; + int uvplane_size = (uv_height + border) * uv_stride; + const int frame_size = yplane_size + 2 * uvplane_size; + + if (!ybf->buffer_alloc) { + ybf->buffer_alloc = (uint8_t *)vpx_memalign(32, frame_size); + ybf->buffer_alloc_sz = frame_size; + } + + if (!ybf->buffer_alloc || ybf->buffer_alloc_sz < frame_size) + return -1; + + /* Only support allocating buffers that have a border that's a multiple + * of 32. The border restriction is required to get 16-byte alignment of + * the start of the chroma rows without introducing an arbitrary gap + * between planes, which would break the semantics of things like + * vpx_img_set_rect(). */ + if (border & 0x1f) + return -3; + + ybf->y_crop_width = width; + ybf->y_crop_height = height; + ybf->y_width = aligned_width; + ybf->y_height = aligned_height; + ybf->y_stride = y_stride; + + ybf->uv_crop_width = (width + 1) / 2; + ybf->uv_crop_height = (height + 1) / 2; + ybf->uv_width = uv_width; + ybf->uv_height = uv_height; + ybf->uv_stride = uv_stride; + + ybf->alpha_width = 0; + ybf->alpha_height = 0; + ybf->alpha_stride = 0; + + ybf->border = border; + ybf->frame_size = frame_size; + + ybf->y_buffer = ybf->buffer_alloc + (border * y_stride) + border; + ybf->u_buffer = ybf->buffer_alloc + yplane_size + (border / 2 * uv_stride) + border / 2; + ybf->v_buffer = ybf->buffer_alloc + yplane_size + uvplane_size + (border / 2 * uv_stride) + border / 2; + ybf->alpha_buffer = NULL; + + ybf->corrupted = 0; /* assume not currupted by errors */ + return 0; + } + return -2; +} + +int vp8_yv12_alloc_frame_buffer(YV12_BUFFER_CONFIG *ybf, + int width, int height, int border) { + if (ybf) { + vp8_yv12_de_alloc_frame_buffer(ybf); + return vp8_yv12_realloc_frame_buffer(ybf, width, height, border); + } + return -2; +} + +#if CONFIG_VP9 +// TODO(jkoleszar): Maybe replace this with struct vpx_image + +int vpx_free_frame_buffer(YV12_BUFFER_CONFIG *ybf) { + if (ybf) { + if (ybf->buffer_alloc_sz > 0) { + vpx_free(ybf->buffer_alloc); + } + + /* buffer_alloc isn't accessed by most functions. Rather y_buffer, + u_buffer and v_buffer point to buffer_alloc and are used. Clear out + all of this so that a freed pointer isn't inadvertently used */ + memset(ybf, 0, sizeof(YV12_BUFFER_CONFIG)); + } else { + return -1; + } + + return 0; +} + +int vpx_realloc_frame_buffer(YV12_BUFFER_CONFIG *ybf, + int width, int height, + int ss_x, int ss_y, +#if CONFIG_VP9_HIGHBITDEPTH + int use_highbitdepth, +#endif + int border, + int byte_alignment, + vpx_codec_frame_buffer_t *fb, + vpx_get_frame_buffer_cb_fn_t cb, + void *cb_priv) { + if (ybf) { + const int vp9_byte_align = (byte_alignment == 0) ? 1 : byte_alignment; + const int aligned_width = (width + 7) & ~7; + const int aligned_height = (height + 7) & ~7; + const int y_stride = ((aligned_width + 2 * border) + 31) & ~31; + const uint64_t yplane_size = (aligned_height + 2 * border) * + (uint64_t)y_stride + byte_alignment; + const int uv_width = aligned_width >> ss_x; + const int uv_height = aligned_height >> ss_y; + const int uv_stride = y_stride >> ss_x; + const int uv_border_w = border >> ss_x; + const int uv_border_h = border >> ss_y; + const uint64_t uvplane_size = (uv_height + 2 * uv_border_h) * + (uint64_t)uv_stride + byte_alignment; + +#if CONFIG_VP9_HIGHBITDEPTH + const uint64_t frame_size = + (1 + use_highbitdepth) * (yplane_size + 2 * uvplane_size); +#else + const uint64_t frame_size = yplane_size + 2 * uvplane_size; +#endif // CONFIG_VP9_HIGHBITDEPTH + + uint8_t *buf = NULL; + + if (cb != NULL) { + const int align_addr_extra_size = 31; + const uint64_t external_frame_size = frame_size + align_addr_extra_size; + + assert(fb != NULL); + + if (external_frame_size != (size_t)external_frame_size) + return -1; + + // Allocation to hold larger frame, or first allocation. + if (cb(cb_priv, (size_t)external_frame_size, fb) < 0) + return -1; + + if (fb->data == NULL || fb->size < external_frame_size) + return -1; + + ybf->buffer_alloc = (uint8_t *)yv12_align_addr(fb->data, 32); + +#if defined(__has_feature) +#if __has_feature(memory_sanitizer) + // This memset is needed for fixing the issue of using uninitialized + // value in msan test. It will cause a perf loss, so only do this for + // msan test. + memset(ybf->buffer_alloc, 0, (int)frame_size); +#endif +#endif + } else if (frame_size > (size_t)ybf->buffer_alloc_sz) { + // Allocation to hold larger frame, or first allocation. + vpx_free(ybf->buffer_alloc); + ybf->buffer_alloc = NULL; + + if (frame_size != (size_t)frame_size) + return -1; + + ybf->buffer_alloc = (uint8_t *)vpx_memalign(32, (size_t)frame_size); + if (!ybf->buffer_alloc) + return -1; + + ybf->buffer_alloc_sz = (int)frame_size; + + // This memset is needed for fixing valgrind error from C loop filter + // due to access uninitialized memory in frame border. It could be + // removed if border is totally removed. + memset(ybf->buffer_alloc, 0, ybf->buffer_alloc_sz); + } + + /* Only support allocating buffers that have a border that's a multiple + * of 32. The border restriction is required to get 16-byte alignment of + * the start of the chroma rows without introducing an arbitrary gap + * between planes, which would break the semantics of things like + * vpx_img_set_rect(). */ + if (border & 0x1f) + return -3; + + ybf->y_crop_width = width; + ybf->y_crop_height = height; + ybf->y_width = aligned_width; + ybf->y_height = aligned_height; + ybf->y_stride = y_stride; + + ybf->uv_crop_width = (width + ss_x) >> ss_x; + ybf->uv_crop_height = (height + ss_y) >> ss_y; + ybf->uv_width = uv_width; + ybf->uv_height = uv_height; + ybf->uv_stride = uv_stride; + + ybf->border = border; + ybf->frame_size = (int)frame_size; + ybf->subsampling_x = ss_x; + ybf->subsampling_y = ss_y; + + buf = ybf->buffer_alloc; +#if CONFIG_VP9_HIGHBITDEPTH + if (use_highbitdepth) { + // Store uint16 addresses when using 16bit framebuffers + buf = CONVERT_TO_BYTEPTR(ybf->buffer_alloc); + ybf->flags = YV12_FLAG_HIGHBITDEPTH; + } else { + ybf->flags = 0; + } +#endif // CONFIG_VP9_HIGHBITDEPTH + + ybf->y_buffer = (uint8_t *)yv12_align_addr( + buf + (border * y_stride) + border, vp9_byte_align); + ybf->u_buffer = (uint8_t *)yv12_align_addr( + buf + yplane_size + (uv_border_h * uv_stride) + uv_border_w, + vp9_byte_align); + ybf->v_buffer = (uint8_t *)yv12_align_addr( + buf + yplane_size + uvplane_size + (uv_border_h * uv_stride) + + uv_border_w, vp9_byte_align); + + ybf->corrupted = 0; /* assume not corrupted by errors */ + return 0; + } + return -2; +} + +int vpx_alloc_frame_buffer(YV12_BUFFER_CONFIG *ybf, + int width, int height, + int ss_x, int ss_y, +#if CONFIG_VP9_HIGHBITDEPTH + int use_highbitdepth, +#endif + int border, + int byte_alignment) { + if (ybf) { + vpx_free_frame_buffer(ybf); + return vpx_realloc_frame_buffer(ybf, width, height, ss_x, ss_y, +#if CONFIG_VP9_HIGHBITDEPTH + use_highbitdepth, +#endif + border, byte_alignment, NULL, NULL, NULL); + } + return -2; +} +#endif diff --git a/thirdparty/libvpx/vpx_scale/generic/yv12extend.c b/thirdparty/libvpx/vpx_scale/generic/yv12extend.c new file mode 100644 index 0000000000..52f0aff1f2 --- /dev/null +++ b/thirdparty/libvpx/vpx_scale/generic/yv12extend.c @@ -0,0 +1,324 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <assert.h> +#include "./vpx_config.h" +#include "./vpx_scale_rtcd.h" +#include "vpx/vpx_integer.h" +#include "vpx_mem/vpx_mem.h" +#include "vpx_ports/mem.h" +#include "vpx_scale/yv12config.h" +#if CONFIG_VP9_HIGHBITDEPTH +#include "vp9/common/vp9_common.h" +#endif + +static void extend_plane(uint8_t *const src, int src_stride, + int width, int height, + int extend_top, int extend_left, + int extend_bottom, int extend_right) { + int i; + const int linesize = extend_left + extend_right + width; + + /* copy the left and right most columns out */ + uint8_t *src_ptr1 = src; + uint8_t *src_ptr2 = src + width - 1; + uint8_t *dst_ptr1 = src - extend_left; + uint8_t *dst_ptr2 = src + width; + + for (i = 0; i < height; ++i) { + memset(dst_ptr1, src_ptr1[0], extend_left); + memset(dst_ptr2, src_ptr2[0], extend_right); + src_ptr1 += src_stride; + src_ptr2 += src_stride; + dst_ptr1 += src_stride; + dst_ptr2 += src_stride; + } + + /* Now copy the top and bottom lines into each line of the respective + * borders + */ + src_ptr1 = src - extend_left; + src_ptr2 = src + src_stride * (height - 1) - extend_left; + dst_ptr1 = src + src_stride * -extend_top - extend_left; + dst_ptr2 = src + src_stride * height - extend_left; + + for (i = 0; i < extend_top; ++i) { + memcpy(dst_ptr1, src_ptr1, linesize); + dst_ptr1 += src_stride; + } + + for (i = 0; i < extend_bottom; ++i) { + memcpy(dst_ptr2, src_ptr2, linesize); + dst_ptr2 += src_stride; + } +} + +#if CONFIG_VP9_HIGHBITDEPTH +static void extend_plane_high(uint8_t *const src8, int src_stride, + int width, int height, + int extend_top, int extend_left, + int extend_bottom, int extend_right) { + int i; + const int linesize = extend_left + extend_right + width; + uint16_t *src = CONVERT_TO_SHORTPTR(src8); + + /* copy the left and right most columns out */ + uint16_t *src_ptr1 = src; + uint16_t *src_ptr2 = src + width - 1; + uint16_t *dst_ptr1 = src - extend_left; + uint16_t *dst_ptr2 = src + width; + + for (i = 0; i < height; ++i) { + vpx_memset16(dst_ptr1, src_ptr1[0], extend_left); + vpx_memset16(dst_ptr2, src_ptr2[0], extend_right); + src_ptr1 += src_stride; + src_ptr2 += src_stride; + dst_ptr1 += src_stride; + dst_ptr2 += src_stride; + } + + /* Now copy the top and bottom lines into each line of the respective + * borders + */ + src_ptr1 = src - extend_left; + src_ptr2 = src + src_stride * (height - 1) - extend_left; + dst_ptr1 = src + src_stride * -extend_top - extend_left; + dst_ptr2 = src + src_stride * height - extend_left; + + for (i = 0; i < extend_top; ++i) { + memcpy(dst_ptr1, src_ptr1, linesize * sizeof(uint16_t)); + dst_ptr1 += src_stride; + } + + for (i = 0; i < extend_bottom; ++i) { + memcpy(dst_ptr2, src_ptr2, linesize * sizeof(uint16_t)); + dst_ptr2 += src_stride; + } +} +#endif + +void vp8_yv12_extend_frame_borders_c(YV12_BUFFER_CONFIG *ybf) { + const int uv_border = ybf->border / 2; + + assert(ybf->border % 2 == 0); + assert(ybf->y_height - ybf->y_crop_height < 16); + assert(ybf->y_width - ybf->y_crop_width < 16); + assert(ybf->y_height - ybf->y_crop_height >= 0); + assert(ybf->y_width - ybf->y_crop_width >= 0); + +#if CONFIG_VP9_HIGHBITDEPTH + if (ybf->flags & YV12_FLAG_HIGHBITDEPTH) { + extend_plane_high( + ybf->y_buffer, ybf->y_stride, + ybf->y_crop_width, ybf->y_crop_height, + ybf->border, ybf->border, + ybf->border + ybf->y_height - ybf->y_crop_height, + ybf->border + ybf->y_width - ybf->y_crop_width); + + extend_plane_high( + ybf->u_buffer, ybf->uv_stride, + ybf->uv_crop_width, ybf->uv_crop_height, + uv_border, uv_border, + uv_border + ybf->uv_height - ybf->uv_crop_height, + uv_border + ybf->uv_width - ybf->uv_crop_width); + + extend_plane_high( + ybf->v_buffer, ybf->uv_stride, + ybf->uv_crop_width, ybf->uv_crop_height, + uv_border, uv_border, + uv_border + ybf->uv_height - ybf->uv_crop_height, + uv_border + ybf->uv_width - ybf->uv_crop_width); + return; + } +#endif + extend_plane(ybf->y_buffer, ybf->y_stride, + ybf->y_crop_width, ybf->y_crop_height, + ybf->border, ybf->border, + ybf->border + ybf->y_height - ybf->y_crop_height, + ybf->border + ybf->y_width - ybf->y_crop_width); + + extend_plane(ybf->u_buffer, ybf->uv_stride, + ybf->uv_crop_width, ybf->uv_crop_height, + uv_border, uv_border, + uv_border + ybf->uv_height - ybf->uv_crop_height, + uv_border + ybf->uv_width - ybf->uv_crop_width); + + extend_plane(ybf->v_buffer, ybf->uv_stride, + ybf->uv_crop_width, ybf->uv_crop_height, + uv_border, uv_border, + uv_border + ybf->uv_height - ybf->uv_crop_height, + uv_border + ybf->uv_width - ybf->uv_crop_width); +} + +#if CONFIG_VP9 +static void extend_frame(YV12_BUFFER_CONFIG *const ybf, int ext_size) { + const int c_w = ybf->uv_crop_width; + const int c_h = ybf->uv_crop_height; + const int ss_x = ybf->uv_width < ybf->y_width; + const int ss_y = ybf->uv_height < ybf->y_height; + const int c_et = ext_size >> ss_y; + const int c_el = ext_size >> ss_x; + const int c_eb = c_et + ybf->uv_height - ybf->uv_crop_height; + const int c_er = c_el + ybf->uv_width - ybf->uv_crop_width; + + assert(ybf->y_height - ybf->y_crop_height < 16); + assert(ybf->y_width - ybf->y_crop_width < 16); + assert(ybf->y_height - ybf->y_crop_height >= 0); + assert(ybf->y_width - ybf->y_crop_width >= 0); + +#if CONFIG_VP9_HIGHBITDEPTH + if (ybf->flags & YV12_FLAG_HIGHBITDEPTH) { + extend_plane_high(ybf->y_buffer, ybf->y_stride, + ybf->y_crop_width, ybf->y_crop_height, + ext_size, ext_size, + ext_size + ybf->y_height - ybf->y_crop_height, + ext_size + ybf->y_width - ybf->y_crop_width); + extend_plane_high(ybf->u_buffer, ybf->uv_stride, + c_w, c_h, c_et, c_el, c_eb, c_er); + extend_plane_high(ybf->v_buffer, ybf->uv_stride, + c_w, c_h, c_et, c_el, c_eb, c_er); + return; + } +#endif + extend_plane(ybf->y_buffer, ybf->y_stride, + ybf->y_crop_width, ybf->y_crop_height, + ext_size, ext_size, + ext_size + ybf->y_height - ybf->y_crop_height, + ext_size + ybf->y_width - ybf->y_crop_width); + + extend_plane(ybf->u_buffer, ybf->uv_stride, + c_w, c_h, c_et, c_el, c_eb, c_er); + + extend_plane(ybf->v_buffer, ybf->uv_stride, + c_w, c_h, c_et, c_el, c_eb, c_er); +} + +void vpx_extend_frame_borders_c(YV12_BUFFER_CONFIG *ybf) { + extend_frame(ybf, ybf->border); +} + +void vpx_extend_frame_inner_borders_c(YV12_BUFFER_CONFIG *ybf) { + const int inner_bw = (ybf->border > VP9INNERBORDERINPIXELS) ? + VP9INNERBORDERINPIXELS : ybf->border; + extend_frame(ybf, inner_bw); +} + +#if CONFIG_VP9_HIGHBITDEPTH +static void memcpy_short_addr(uint8_t *dst8, const uint8_t *src8, int num) { + uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); + uint16_t *src = CONVERT_TO_SHORTPTR(src8); + memcpy(dst, src, num * sizeof(uint16_t)); +} +#endif // CONFIG_VP9_HIGHBITDEPTH +#endif // CONFIG_VP9 + +// Copies the source image into the destination image and updates the +// destination's UMV borders. +// Note: The frames are assumed to be identical in size. +void vp8_yv12_copy_frame_c(const YV12_BUFFER_CONFIG *src_ybc, + YV12_BUFFER_CONFIG *dst_ybc) { + int row; + const uint8_t *src = src_ybc->y_buffer; + uint8_t *dst = dst_ybc->y_buffer; + +#if 0 + /* These assertions are valid in the codec, but the libvpx-tester uses + * this code slightly differently. + */ + assert(src_ybc->y_width == dst_ybc->y_width); + assert(src_ybc->y_height == dst_ybc->y_height); +#endif + +#if CONFIG_VP9_HIGHBITDEPTH + if (src_ybc->flags & YV12_FLAG_HIGHBITDEPTH) { + assert(dst_ybc->flags & YV12_FLAG_HIGHBITDEPTH); + for (row = 0; row < src_ybc->y_height; ++row) { + memcpy_short_addr(dst, src, src_ybc->y_width); + src += src_ybc->y_stride; + dst += dst_ybc->y_stride; + } + + src = src_ybc->u_buffer; + dst = dst_ybc->u_buffer; + + for (row = 0; row < src_ybc->uv_height; ++row) { + memcpy_short_addr(dst, src, src_ybc->uv_width); + src += src_ybc->uv_stride; + dst += dst_ybc->uv_stride; + } + + src = src_ybc->v_buffer; + dst = dst_ybc->v_buffer; + + for (row = 0; row < src_ybc->uv_height; ++row) { + memcpy_short_addr(dst, src, src_ybc->uv_width); + src += src_ybc->uv_stride; + dst += dst_ybc->uv_stride; + } + + vp8_yv12_extend_frame_borders_c(dst_ybc); + return; + } else { + assert(!(dst_ybc->flags & YV12_FLAG_HIGHBITDEPTH)); + } +#endif + + for (row = 0; row < src_ybc->y_height; ++row) { + memcpy(dst, src, src_ybc->y_width); + src += src_ybc->y_stride; + dst += dst_ybc->y_stride; + } + + src = src_ybc->u_buffer; + dst = dst_ybc->u_buffer; + + for (row = 0; row < src_ybc->uv_height; ++row) { + memcpy(dst, src, src_ybc->uv_width); + src += src_ybc->uv_stride; + dst += dst_ybc->uv_stride; + } + + src = src_ybc->v_buffer; + dst = dst_ybc->v_buffer; + + for (row = 0; row < src_ybc->uv_height; ++row) { + memcpy(dst, src, src_ybc->uv_width); + src += src_ybc->uv_stride; + dst += dst_ybc->uv_stride; + } + + vp8_yv12_extend_frame_borders_c(dst_ybc); +} + +void vpx_yv12_copy_y_c(const YV12_BUFFER_CONFIG *src_ybc, + YV12_BUFFER_CONFIG *dst_ybc) { + int row; + const uint8_t *src = src_ybc->y_buffer; + uint8_t *dst = dst_ybc->y_buffer; + +#if CONFIG_VP9_HIGHBITDEPTH + if (src_ybc->flags & YV12_FLAG_HIGHBITDEPTH) { + const uint16_t *src16 = CONVERT_TO_SHORTPTR(src); + uint16_t *dst16 = CONVERT_TO_SHORTPTR(dst); + for (row = 0; row < src_ybc->y_height; ++row) { + memcpy(dst16, src16, src_ybc->y_width * sizeof(uint16_t)); + src16 += src_ybc->y_stride; + dst16 += dst_ybc->y_stride; + } + return; + } +#endif + + for (row = 0; row < src_ybc->y_height; ++row) { + memcpy(dst, src, src_ybc->y_width); + src += src_ybc->y_stride; + dst += dst_ybc->y_stride; + } +} diff --git a/thirdparty/libvpx/vpx_scale/vpx_scale.h b/thirdparty/libvpx/vpx_scale/vpx_scale.h new file mode 100644 index 0000000000..43fcf9d66e --- /dev/null +++ b/thirdparty/libvpx/vpx_scale/vpx_scale.h @@ -0,0 +1,27 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#ifndef VPX_SCALE_VPX_SCALE_H_ +#define VPX_SCALE_VPX_SCALE_H_ + +#include "vpx_scale/yv12config.h" + +extern void vpx_scale_frame(YV12_BUFFER_CONFIG *src, + YV12_BUFFER_CONFIG *dst, + unsigned char *temp_area, + unsigned char temp_height, + unsigned int hscale, + unsigned int hratio, + unsigned int vscale, + unsigned int vratio, + unsigned int interlaced); + +#endif // VPX_SCALE_VPX_SCALE_H_ diff --git a/thirdparty/libvpx/vpx_scale/vpx_scale_rtcd.c b/thirdparty/libvpx/vpx_scale/vpx_scale_rtcd.c new file mode 100644 index 0000000000..bea603fd10 --- /dev/null +++ b/thirdparty/libvpx/vpx_scale/vpx_scale_rtcd.c @@ -0,0 +1,18 @@ +/* + * Copyright (c) 2011 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ +#include "./vpx_config.h" +#define RTCD_C +#include "./vpx_scale_rtcd.h" +#include "vpx_ports/vpx_once.h" + +void vpx_scale_rtcd() +{ + once(setup_rtcd_internal); +} diff --git a/thirdparty/libvpx/vpx_scale/yv12config.h b/thirdparty/libvpx/vpx_scale/yv12config.h new file mode 100644 index 0000000000..37b255d4d3 --- /dev/null +++ b/thirdparty/libvpx/vpx_scale/yv12config.h @@ -0,0 +1,105 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VPX_SCALE_YV12CONFIG_H_ +#define VPX_SCALE_YV12CONFIG_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +#include "./vpx_config.h" +#include "vpx/vpx_codec.h" +#include "vpx/vpx_frame_buffer.h" +#include "vpx/vpx_integer.h" + +#define VP8BORDERINPIXELS 32 +#define VP9INNERBORDERINPIXELS 96 +#define VP9_INTERP_EXTEND 4 +#define VP9_ENC_BORDER_IN_PIXELS 160 +#define VP9_DEC_BORDER_IN_PIXELS 32 + +typedef struct yv12_buffer_config { + int y_width; + int y_height; + int y_crop_width; + int y_crop_height; + int y_stride; + + int uv_width; + int uv_height; + int uv_crop_width; + int uv_crop_height; + int uv_stride; + + int alpha_width; + int alpha_height; + int alpha_stride; + + uint8_t *y_buffer; + uint8_t *u_buffer; + uint8_t *v_buffer; + uint8_t *alpha_buffer; + + uint8_t *buffer_alloc; + int buffer_alloc_sz; + int border; + int frame_size; + int subsampling_x; + int subsampling_y; + unsigned int bit_depth; + vpx_color_space_t color_space; + vpx_color_range_t color_range; + int render_width; + int render_height; + + int corrupted; + int flags; +} YV12_BUFFER_CONFIG; + +#define YV12_FLAG_HIGHBITDEPTH 8 + +int vp8_yv12_alloc_frame_buffer(YV12_BUFFER_CONFIG *ybf, + int width, int height, int border); +int vp8_yv12_realloc_frame_buffer(YV12_BUFFER_CONFIG *ybf, + int width, int height, int border); +int vp8_yv12_de_alloc_frame_buffer(YV12_BUFFER_CONFIG *ybf); + +int vpx_alloc_frame_buffer(YV12_BUFFER_CONFIG *ybf, + int width, int height, int ss_x, int ss_y, +#if CONFIG_VP9_HIGHBITDEPTH + int use_highbitdepth, +#endif + int border, int byte_alignment); + +// Updates the yv12 buffer config with the frame buffer. |byte_alignment| must +// be a power of 2, from 32 to 1024. 0 sets legacy alignment. If cb is not +// NULL, then libvpx is using the frame buffer callbacks to handle memory. +// If cb is not NULL, libvpx will call cb with minimum size in bytes needed +// to decode the current frame. If cb is NULL, libvpx will allocate memory +// internally to decode the current frame. Returns 0 on success. Returns < 0 +// on failure. +int vpx_realloc_frame_buffer(YV12_BUFFER_CONFIG *ybf, + int width, int height, int ss_x, int ss_y, +#if CONFIG_VP9_HIGHBITDEPTH + int use_highbitdepth, +#endif + int border, + int byte_alignment, + vpx_codec_frame_buffer_t *fb, + vpx_get_frame_buffer_cb_fn_t cb, + void *cb_priv); +int vpx_free_frame_buffer(YV12_BUFFER_CONFIG *ybf); + +#ifdef __cplusplus +} +#endif + +#endif // VPX_SCALE_YV12CONFIG_H_ diff --git a/thirdparty/libvpx/vpx_util/endian_inl.h b/thirdparty/libvpx/vpx_util/endian_inl.h new file mode 100644 index 0000000000..37bdce1ccd --- /dev/null +++ b/thirdparty/libvpx/vpx_util/endian_inl.h @@ -0,0 +1,120 @@ +// Copyright 2014 Google Inc. All Rights Reserved. +// +// Use of this source code is governed by a BSD-style license +// that can be found in the COPYING file in the root of the source +// tree. An additional intellectual property rights grant can be found +// in the file PATENTS. All contributing project authors may +// be found in the AUTHORS file in the root of the source tree. +// ----------------------------------------------------------------------------- +// +// Endian related functions. + +#ifndef VPX_UTIL_ENDIAN_INL_H_ +#define VPX_UTIL_ENDIAN_INL_H_ + +#include <stdlib.h> +#include "./vpx_config.h" +#include "vpx/vpx_integer.h" + +#if defined(__GNUC__) +# define LOCAL_GCC_VERSION ((__GNUC__ << 8) | __GNUC_MINOR__) +# define LOCAL_GCC_PREREQ(maj, min) \ + (LOCAL_GCC_VERSION >= (((maj) << 8) | (min))) +#else +# define LOCAL_GCC_VERSION 0 +# define LOCAL_GCC_PREREQ(maj, min) 0 +#endif + +// handle clang compatibility +#ifndef __has_builtin +# define __has_builtin(x) 0 +#endif + +// some endian fix (e.g.: mips-gcc doesn't define __BIG_ENDIAN__) +#if !defined(WORDS_BIGENDIAN) && \ + (defined(__BIG_ENDIAN__) || defined(_M_PPC) || \ + (defined(__BYTE_ORDER__) && (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__))) +#define WORDS_BIGENDIAN +#endif + +#if defined(WORDS_BIGENDIAN) +#define HToLE32 BSwap32 +#define HToLE16 BSwap16 +#define HToBE64(x) (x) +#define HToBE32(x) (x) +#else +#define HToLE32(x) (x) +#define HToLE16(x) (x) +#define HToBE64(X) BSwap64(X) +#define HToBE32(X) BSwap32(X) +#endif + +#if LOCAL_GCC_PREREQ(4, 8) || __has_builtin(__builtin_bswap16) +#define HAVE_BUILTIN_BSWAP16 +#endif + +#if LOCAL_GCC_PREREQ(4, 3) || __has_builtin(__builtin_bswap32) +#define HAVE_BUILTIN_BSWAP32 +#endif + +#if LOCAL_GCC_PREREQ(4, 3) || __has_builtin(__builtin_bswap64) +#define HAVE_BUILTIN_BSWAP64 +#endif + +#if HAVE_MIPS32 && defined(__mips__) && !defined(__mips64) && \ + defined(__mips_isa_rev) && (__mips_isa_rev >= 2) && (__mips_isa_rev < 6) +#define VPX_USE_MIPS32_R2 +#endif + +static INLINE uint16_t BSwap16(uint16_t x) { +#if defined(HAVE_BUILTIN_BSWAP16) + return __builtin_bswap16(x); +#elif defined(_MSC_VER) + return _byteswap_ushort(x); +#else + // gcc will recognize a 'rorw $8, ...' here: + return (x >> 8) | ((x & 0xff) << 8); +#endif // HAVE_BUILTIN_BSWAP16 +} + +static INLINE uint32_t BSwap32(uint32_t x) { +#if defined(VPX_USE_MIPS32_R2) + uint32_t ret; + __asm__ volatile ( + "wsbh %[ret], %[x] \n\t" + "rotr %[ret], %[ret], 16 \n\t" + : [ret]"=r"(ret) + : [x]"r"(x) + ); + return ret; +#elif defined(HAVE_BUILTIN_BSWAP32) + return __builtin_bswap32(x); +#elif defined(__i386__) || defined(__x86_64__) + uint32_t swapped_bytes; + __asm__ volatile("bswap %0" : "=r"(swapped_bytes) : "0"(x)); + return swapped_bytes; +#elif defined(_MSC_VER) + return (uint32_t)_byteswap_ulong(x); +#else + return (x >> 24) | ((x >> 8) & 0xff00) | ((x << 8) & 0xff0000) | (x << 24); +#endif // HAVE_BUILTIN_BSWAP32 +} + +static INLINE uint64_t BSwap64(uint64_t x) { +#if defined(HAVE_BUILTIN_BSWAP64) + return __builtin_bswap64(x); +#elif defined(__x86_64__) + uint64_t swapped_bytes; + __asm__ volatile("bswapq %0" : "=r"(swapped_bytes) : "0"(x)); + return swapped_bytes; +#elif defined(_MSC_VER) + return (uint64_t)_byteswap_uint64(x); +#else // generic code for swapping 64-bit values (suggested by bdb@) + x = ((x & 0xffffffff00000000ull) >> 32) | ((x & 0x00000000ffffffffull) << 32); + x = ((x & 0xffff0000ffff0000ull) >> 16) | ((x & 0x0000ffff0000ffffull) << 16); + x = ((x & 0xff00ff00ff00ff00ull) >> 8) | ((x & 0x00ff00ff00ff00ffull) << 8); + return x; +#endif // HAVE_BUILTIN_BSWAP64 +} + +#endif // VPX_UTIL_ENDIAN_INL_H_ diff --git a/thirdparty/libvpx/vpx_util/vpx_thread.c b/thirdparty/libvpx/vpx_util/vpx_thread.c new file mode 100644 index 0000000000..0bb0125bd4 --- /dev/null +++ b/thirdparty/libvpx/vpx_util/vpx_thread.c @@ -0,0 +1,184 @@ +// Copyright 2013 Google Inc. All Rights Reserved. +// +// Use of this source code is governed by a BSD-style license +// that can be found in the COPYING file in the root of the source +// tree. An additional intellectual property rights grant can be found +// in the file PATENTS. All contributing project authors may +// be found in the AUTHORS file in the root of the source tree. +// ----------------------------------------------------------------------------- +// +// Multi-threaded worker +// +// Original source: +// http://git.chromium.org/webm/libwebp.git +// 100644 blob 264210ba2807e4da47eb5d18c04cf869d89b9784 src/utils/thread.c + +#include <assert.h> +#include <string.h> // for memset() +#include "./vpx_thread.h" +#include "vpx_mem/vpx_mem.h" + +#if CONFIG_MULTITHREAD + +struct VPxWorkerImpl { + pthread_mutex_t mutex_; + pthread_cond_t condition_; + pthread_t thread_; +}; + +//------------------------------------------------------------------------------ + +static void execute(VPxWorker *const worker); // Forward declaration. + +static THREADFN thread_loop(void *ptr) { + VPxWorker *const worker = (VPxWorker*)ptr; + int done = 0; + while (!done) { + pthread_mutex_lock(&worker->impl_->mutex_); + while (worker->status_ == OK) { // wait in idling mode + pthread_cond_wait(&worker->impl_->condition_, &worker->impl_->mutex_); + } + if (worker->status_ == WORK) { + execute(worker); + worker->status_ = OK; + } else if (worker->status_ == NOT_OK) { // finish the worker + done = 1; + } + // signal to the main thread that we're done (for sync()) + pthread_cond_signal(&worker->impl_->condition_); + pthread_mutex_unlock(&worker->impl_->mutex_); + } + return THREAD_RETURN(NULL); // Thread is finished +} + +// main thread state control +static void change_state(VPxWorker *const worker, + VPxWorkerStatus new_status) { + // No-op when attempting to change state on a thread that didn't come up. + // Checking status_ without acquiring the lock first would result in a data + // race. + if (worker->impl_ == NULL) return; + + pthread_mutex_lock(&worker->impl_->mutex_); + if (worker->status_ >= OK) { + // wait for the worker to finish + while (worker->status_ != OK) { + pthread_cond_wait(&worker->impl_->condition_, &worker->impl_->mutex_); + } + // assign new status and release the working thread if needed + if (new_status != OK) { + worker->status_ = new_status; + pthread_cond_signal(&worker->impl_->condition_); + } + } + pthread_mutex_unlock(&worker->impl_->mutex_); +} + +#endif // CONFIG_MULTITHREAD + +//------------------------------------------------------------------------------ + +static void init(VPxWorker *const worker) { + memset(worker, 0, sizeof(*worker)); + worker->status_ = NOT_OK; +} + +static int sync(VPxWorker *const worker) { +#if CONFIG_MULTITHREAD + change_state(worker, OK); +#endif + assert(worker->status_ <= OK); + return !worker->had_error; +} + +static int reset(VPxWorker *const worker) { + int ok = 1; + worker->had_error = 0; + if (worker->status_ < OK) { +#if CONFIG_MULTITHREAD + worker->impl_ = (VPxWorkerImpl*)vpx_calloc(1, sizeof(*worker->impl_)); + if (worker->impl_ == NULL) { + return 0; + } + if (pthread_mutex_init(&worker->impl_->mutex_, NULL)) { + goto Error; + } + if (pthread_cond_init(&worker->impl_->condition_, NULL)) { + pthread_mutex_destroy(&worker->impl_->mutex_); + goto Error; + } + pthread_mutex_lock(&worker->impl_->mutex_); + ok = !pthread_create(&worker->impl_->thread_, NULL, thread_loop, worker); + if (ok) worker->status_ = OK; + pthread_mutex_unlock(&worker->impl_->mutex_); + if (!ok) { + pthread_mutex_destroy(&worker->impl_->mutex_); + pthread_cond_destroy(&worker->impl_->condition_); + Error: + vpx_free(worker->impl_); + worker->impl_ = NULL; + return 0; + } +#else + worker->status_ = OK; +#endif + } else if (worker->status_ > OK) { + ok = sync(worker); + } + assert(!ok || (worker->status_ == OK)); + return ok; +} + +static void execute(VPxWorker *const worker) { + if (worker->hook != NULL) { + worker->had_error |= !worker->hook(worker->data1, worker->data2); + } +} + +static void launch(VPxWorker *const worker) { +#if CONFIG_MULTITHREAD + change_state(worker, WORK); +#else + execute(worker); +#endif +} + +static void end(VPxWorker *const worker) { +#if CONFIG_MULTITHREAD + if (worker->impl_ != NULL) { + change_state(worker, NOT_OK); + pthread_join(worker->impl_->thread_, NULL); + pthread_mutex_destroy(&worker->impl_->mutex_); + pthread_cond_destroy(&worker->impl_->condition_); + vpx_free(worker->impl_); + worker->impl_ = NULL; + } +#else + worker->status_ = NOT_OK; + assert(worker->impl_ == NULL); +#endif + assert(worker->status_ == NOT_OK); +} + +//------------------------------------------------------------------------------ + +static VPxWorkerInterface g_worker_interface = { + init, reset, sync, launch, execute, end +}; + +int vpx_set_worker_interface(const VPxWorkerInterface* const winterface) { + if (winterface == NULL || + winterface->init == NULL || winterface->reset == NULL || + winterface->sync == NULL || winterface->launch == NULL || + winterface->execute == NULL || winterface->end == NULL) { + return 0; + } + g_worker_interface = *winterface; + return 1; +} + +const VPxWorkerInterface *vpx_get_worker_interface(void) { + return &g_worker_interface; +} + +//------------------------------------------------------------------------------ diff --git a/thirdparty/libvpx/vpx_util/vpx_thread.h b/thirdparty/libvpx/vpx_util/vpx_thread.h new file mode 100644 index 0000000000..2062abd75f --- /dev/null +++ b/thirdparty/libvpx/vpx_util/vpx_thread.h @@ -0,0 +1,369 @@ +// Copyright 2013 Google Inc. All Rights Reserved. +// +// Use of this source code is governed by a BSD-style license +// that can be found in the COPYING file in the root of the source +// tree. An additional intellectual property rights grant can be found +// in the file PATENTS. All contributing project authors may +// be found in the AUTHORS file in the root of the source tree. +// ----------------------------------------------------------------------------- +// +// Multi-threaded worker +// +// Original source: +// http://git.chromium.org/webm/libwebp.git +// 100644 blob 7bd451b124ae3b81596abfbcc823e3cb129d3a38 src/utils/thread.h + +#ifndef VPX_THREAD_H_ +#define VPX_THREAD_H_ + +#include "./vpx_config.h" + +#ifdef __cplusplus +extern "C" { +#endif + +// Set maximum decode threads to be 8 due to the limit of frame buffers +// and not enough semaphores in the emulation layer on windows. +#define MAX_DECODE_THREADS 8 + +#if CONFIG_MULTITHREAD + +#if defined(_WIN32) && !HAVE_PTHREAD_H +#include <errno.h> // NOLINT +#include <process.h> // NOLINT +#include <windows.h> // NOLINT +typedef HANDLE pthread_t; +typedef CRITICAL_SECTION pthread_mutex_t; +typedef struct { + HANDLE waiting_sem_; + HANDLE received_sem_; + HANDLE signal_event_; +} pthread_cond_t; + +//------------------------------------------------------------------------------ +// simplistic pthread emulation layer + +// _beginthreadex requires __stdcall +#define THREADFN unsigned int __stdcall +#define THREAD_RETURN(val) (unsigned int)((DWORD_PTR)val) + +static INLINE int pthread_create(pthread_t* const thread, const void* attr, + unsigned int (__stdcall *start)(void*), + void* arg) { + (void)attr; + *thread = (pthread_t)_beginthreadex(NULL, /* void *security */ + 0, /* unsigned stack_size */ + start, + arg, + 0, /* unsigned initflag */ + NULL); /* unsigned *thrdaddr */ + if (*thread == NULL) return 1; + SetThreadPriority(*thread, THREAD_PRIORITY_ABOVE_NORMAL); + return 0; +} + +static INLINE int pthread_join(pthread_t thread, void** value_ptr) { + (void)value_ptr; + return (WaitForSingleObject(thread, INFINITE) != WAIT_OBJECT_0 || + CloseHandle(thread) == 0); +} + +// Mutex +static INLINE int pthread_mutex_init(pthread_mutex_t *const mutex, + void* mutexattr) { + (void)mutexattr; + InitializeCriticalSection(mutex); + return 0; +} + +static INLINE int pthread_mutex_trylock(pthread_mutex_t *const mutex) { + return TryEnterCriticalSection(mutex) ? 0 : EBUSY; +} + +static INLINE int pthread_mutex_lock(pthread_mutex_t *const mutex) { + EnterCriticalSection(mutex); + return 0; +} + +static INLINE int pthread_mutex_unlock(pthread_mutex_t *const mutex) { + LeaveCriticalSection(mutex); + return 0; +} + +static INLINE int pthread_mutex_destroy(pthread_mutex_t *const mutex) { + DeleteCriticalSection(mutex); + return 0; +} + +// Condition +static INLINE int pthread_cond_destroy(pthread_cond_t *const condition) { + int ok = 1; + ok &= (CloseHandle(condition->waiting_sem_) != 0); + ok &= (CloseHandle(condition->received_sem_) != 0); + ok &= (CloseHandle(condition->signal_event_) != 0); + return !ok; +} + +static INLINE int pthread_cond_init(pthread_cond_t *const condition, + void* cond_attr) { + (void)cond_attr; + condition->waiting_sem_ = CreateSemaphore(NULL, 0, MAX_DECODE_THREADS, NULL); + condition->received_sem_ = CreateSemaphore(NULL, 0, MAX_DECODE_THREADS, NULL); + condition->signal_event_ = CreateEvent(NULL, FALSE, FALSE, NULL); + if (condition->waiting_sem_ == NULL || + condition->received_sem_ == NULL || + condition->signal_event_ == NULL) { + pthread_cond_destroy(condition); + return 1; + } + return 0; +} + +static INLINE int pthread_cond_signal(pthread_cond_t *const condition) { + int ok = 1; + if (WaitForSingleObject(condition->waiting_sem_, 0) == WAIT_OBJECT_0) { + // a thread is waiting in pthread_cond_wait: allow it to be notified + ok = SetEvent(condition->signal_event_); + // wait until the event is consumed so the signaler cannot consume + // the event via its own pthread_cond_wait. + ok &= (WaitForSingleObject(condition->received_sem_, INFINITE) != + WAIT_OBJECT_0); + } + return !ok; +} + +static INLINE int pthread_cond_wait(pthread_cond_t *const condition, + pthread_mutex_t *const mutex) { + int ok; + // note that there is a consumer available so the signal isn't dropped in + // pthread_cond_signal + if (!ReleaseSemaphore(condition->waiting_sem_, 1, NULL)) + return 1; + // now unlock the mutex so pthread_cond_signal may be issued + pthread_mutex_unlock(mutex); + ok = (WaitForSingleObject(condition->signal_event_, INFINITE) == + WAIT_OBJECT_0); + ok &= ReleaseSemaphore(condition->received_sem_, 1, NULL); + pthread_mutex_lock(mutex); + return !ok; +} +#elif defined(__OS2__) +#define INCL_DOS +#include <os2.h> // NOLINT + +#include <errno.h> // NOLINT +#include <stdlib.h> // NOLINT +#include <sys/builtin.h> // NOLINT + +#define pthread_t TID +#define pthread_mutex_t HMTX + +typedef struct { + HEV event_sem_; + HEV ack_sem_; + volatile unsigned wait_count_; +} pthread_cond_t; + +//------------------------------------------------------------------------------ +// simplistic pthread emulation layer + +#define THREADFN void * +#define THREAD_RETURN(val) (val) + +typedef struct { + void* (*start_)(void*); + void* arg_; +} thread_arg; + +static void thread_start(void* arg) { + thread_arg targ = *(thread_arg *)arg; + free(arg); + + targ.start_(targ.arg_); +} + +static INLINE int pthread_create(pthread_t* const thread, const void* attr, + void* (*start)(void*), + void* arg) { + int tid; + thread_arg *targ = (thread_arg *)malloc(sizeof(*targ)); + if (targ == NULL) return 1; + + (void)attr; + + targ->start_ = start; + targ->arg_ = arg; + tid = (pthread_t)_beginthread(thread_start, NULL, 1024 * 1024, targ); + if (tid == -1) { + free(targ); + return 1; + } + + *thread = tid; + return 0; +} + +static INLINE int pthread_join(pthread_t thread, void** value_ptr) { + (void)value_ptr; + return DosWaitThread(&thread, DCWW_WAIT) != 0; +} + +// Mutex +static INLINE int pthread_mutex_init(pthread_mutex_t *const mutex, + void* mutexattr) { + (void)mutexattr; + return DosCreateMutexSem(NULL, mutex, 0, FALSE) != 0; +} + +static INLINE int pthread_mutex_trylock(pthread_mutex_t *const mutex) { + return DosRequestMutexSem(*mutex, SEM_IMMEDIATE_RETURN) == 0 ? 0 : EBUSY; +} + +static INLINE int pthread_mutex_lock(pthread_mutex_t *const mutex) { + return DosRequestMutexSem(*mutex, SEM_INDEFINITE_WAIT) != 0; +} + +static INLINE int pthread_mutex_unlock(pthread_mutex_t *const mutex) { + return DosReleaseMutexSem(*mutex) != 0; +} + +static INLINE int pthread_mutex_destroy(pthread_mutex_t *const mutex) { + return DosCloseMutexSem(*mutex) != 0; +} + +// Condition +static INLINE int pthread_cond_destroy(pthread_cond_t *const condition) { + int ok = 1; + ok &= DosCloseEventSem(condition->event_sem_) == 0; + ok &= DosCloseEventSem(condition->ack_sem_) == 0; + return !ok; +} + +static INLINE int pthread_cond_init(pthread_cond_t *const condition, + void* cond_attr) { + int ok = 1; + (void)cond_attr; + + ok &= DosCreateEventSem(NULL, &condition->event_sem_, DCE_POSTONE, FALSE) + == 0; + ok &= DosCreateEventSem(NULL, &condition->ack_sem_, DCE_POSTONE, FALSE) == 0; + if (!ok) { + pthread_cond_destroy(condition); + return 1; + } + condition->wait_count_ = 0; + return 0; +} + +static INLINE int pthread_cond_signal(pthread_cond_t *const condition) { + int ok = 1; + + if (!__atomic_cmpxchg32(&condition->wait_count_, 0, 0)) { + ok &= DosPostEventSem(condition->event_sem_) == 0; + ok &= DosWaitEventSem(condition->ack_sem_, SEM_INDEFINITE_WAIT) == 0; + } + + return !ok; +} + +static INLINE int pthread_cond_broadcast(pthread_cond_t *const condition) { + int ok = 1; + + while (!__atomic_cmpxchg32(&condition->wait_count_, 0, 0)) + ok &= pthread_cond_signal(condition) == 0; + + return !ok; +} + +static INLINE int pthread_cond_wait(pthread_cond_t *const condition, + pthread_mutex_t *const mutex) { + int ok = 1; + + __atomic_increment(&condition->wait_count_); + + ok &= pthread_mutex_unlock(mutex) == 0; + + ok &= DosWaitEventSem(condition->event_sem_, SEM_INDEFINITE_WAIT) == 0; + + __atomic_decrement(&condition->wait_count_); + + ok &= DosPostEventSem(condition->ack_sem_) == 0; + + pthread_mutex_lock(mutex); + + return !ok; +} +#else // _WIN32 +#include <pthread.h> // NOLINT +# define THREADFN void* +# define THREAD_RETURN(val) val +#endif + +#endif // CONFIG_MULTITHREAD + +// State of the worker thread object +typedef enum { + NOT_OK = 0, // object is unusable + OK, // ready to work + WORK // busy finishing the current task +} VPxWorkerStatus; + +// Function to be called by the worker thread. Takes two opaque pointers as +// arguments (data1 and data2), and should return false in case of error. +typedef int (*VPxWorkerHook)(void*, void*); + +// Platform-dependent implementation details for the worker. +typedef struct VPxWorkerImpl VPxWorkerImpl; + +// Synchronization object used to launch job in the worker thread +typedef struct { + VPxWorkerImpl *impl_; + VPxWorkerStatus status_; + VPxWorkerHook hook; // hook to call + void *data1; // first argument passed to 'hook' + void *data2; // second argument passed to 'hook' + int had_error; // return value of the last call to 'hook' +} VPxWorker; + +// The interface for all thread-worker related functions. All these functions +// must be implemented. +typedef struct { + // Must be called first, before any other method. + void (*init)(VPxWorker *const worker); + // Must be called to initialize the object and spawn the thread. Re-entrant. + // Will potentially launch the thread. Returns false in case of error. + int (*reset)(VPxWorker *const worker); + // Makes sure the previous work is finished. Returns true if worker->had_error + // was not set and no error condition was triggered by the working thread. + int (*sync)(VPxWorker *const worker); + // Triggers the thread to call hook() with data1 and data2 arguments. These + // hook/data1/data2 values can be changed at any time before calling this + // function, but not be changed afterward until the next call to Sync(). + void (*launch)(VPxWorker *const worker); + // This function is similar to launch() except that it calls the + // hook directly instead of using a thread. Convenient to bypass the thread + // mechanism while still using the VPxWorker structs. sync() must + // still be called afterward (for error reporting). + void (*execute)(VPxWorker *const worker); + // Kill the thread and terminate the object. To use the object again, one + // must call reset() again. + void (*end)(VPxWorker *const worker); +} VPxWorkerInterface; + +// Install a new set of threading functions, overriding the defaults. This +// should be done before any workers are started, i.e., before any encoding or +// decoding takes place. The contents of the interface struct are copied, it +// is safe to free the corresponding memory after this call. This function is +// not thread-safe. Return false in case of invalid pointer or methods. +int vpx_set_worker_interface(const VPxWorkerInterface *const winterface); + +// Retrieve the currently set thread worker interface. +const VPxWorkerInterface *vpx_get_worker_interface(void); + +//------------------------------------------------------------------------------ + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VPX_THREAD_H_ diff --git a/thirdparty/libvpx/vpx_version.h b/thirdparty/libvpx/vpx_version.h new file mode 100644 index 0000000000..5cff3b429f --- /dev/null +++ b/thirdparty/libvpx/vpx_version.h @@ -0,0 +1,7 @@ +#define VERSION_MAJOR 1 +#define VERSION_MINOR 6 +#define VERSION_PATCH 0 +#define VERSION_EXTRA "" +#define VERSION_PACKED ((VERSION_MAJOR<<16)|(VERSION_MINOR<<8)|(VERSION_PATCH)) +#define VERSION_STRING_NOSP "v1.6.0" +#define VERSION_STRING " v1.6.0" |