diff options
Diffstat (limited to 'thirdparty/libwebp/src/dec')
| -rw-r--r-- | thirdparty/libwebp/src/dec/alpha_dec.c | 232 | ||||
| -rw-r--r-- | thirdparty/libwebp/src/dec/alphai_dec.h | 54 | ||||
| -rw-r--r-- | thirdparty/libwebp/src/dec/buffer_dec.c | 311 | ||||
| -rw-r--r-- | thirdparty/libwebp/src/dec/common_dec.h | 54 | ||||
| -rw-r--r-- | thirdparty/libwebp/src/dec/frame_dec.c | 812 | ||||
| -rw-r--r-- | thirdparty/libwebp/src/dec/idec_dec.c | 892 | ||||
| -rw-r--r-- | thirdparty/libwebp/src/dec/io_dec.c | 649 | ||||
| -rw-r--r-- | thirdparty/libwebp/src/dec/quant_dec.c | 110 | ||||
| -rw-r--r-- | thirdparty/libwebp/src/dec/tree_dec.c | 532 | ||||
| -rw-r--r-- | thirdparty/libwebp/src/dec/vp8_dec.c | 721 | ||||
| -rw-r--r-- | thirdparty/libwebp/src/dec/vp8_dec.h | 185 | ||||
| -rw-r--r-- | thirdparty/libwebp/src/dec/vp8i_dec.h | 319 | ||||
| -rw-r--r-- | thirdparty/libwebp/src/dec/vp8l_dec.c | 1689 | ||||
| -rw-r--r-- | thirdparty/libwebp/src/dec/vp8li_dec.h | 135 | ||||
| -rw-r--r-- | thirdparty/libwebp/src/dec/webp_dec.c | 845 | ||||
| -rw-r--r-- | thirdparty/libwebp/src/dec/webpi_dec.h | 133 |
16 files changed, 7673 insertions, 0 deletions
diff --git a/thirdparty/libwebp/src/dec/alpha_dec.c b/thirdparty/libwebp/src/dec/alpha_dec.c new file mode 100644 index 0000000000..bce735bfc2 --- /dev/null +++ b/thirdparty/libwebp/src/dec/alpha_dec.c @@ -0,0 +1,232 @@ +// Copyright 2011 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. +// ----------------------------------------------------------------------------- +// +// Alpha-plane decompression. +// +// Author: Skal (pascal.massimino@gmail.com) + +#include <stdlib.h> +#include "src/dec/alphai_dec.h" +#include "src/dec/vp8i_dec.h" +#include "src/dec/vp8li_dec.h" +#include "src/dsp/dsp.h" +#include "src/utils/quant_levels_dec_utils.h" +#include "src/utils/utils.h" +#include "src/webp/format_constants.h" + +//------------------------------------------------------------------------------ +// ALPHDecoder object. + +// Allocates a new alpha decoder instance. +static ALPHDecoder* ALPHNew(void) { + ALPHDecoder* const dec = (ALPHDecoder*)WebPSafeCalloc(1ULL, sizeof(*dec)); + return dec; +} + +// Clears and deallocates an alpha decoder instance. +static void ALPHDelete(ALPHDecoder* const dec) { + if (dec != NULL) { + VP8LDelete(dec->vp8l_dec_); + dec->vp8l_dec_ = NULL; + WebPSafeFree(dec); + } +} + +//------------------------------------------------------------------------------ +// Decoding. + +// Initialize alpha decoding by parsing the alpha header and decoding the image +// header for alpha data stored using lossless compression. +// Returns false in case of error in alpha header (data too short, invalid +// compression method or filter, error in lossless header data etc). +static int ALPHInit(ALPHDecoder* const dec, const uint8_t* data, + size_t data_size, const VP8Io* const src_io, + uint8_t* output) { + int ok = 0; + const uint8_t* const alpha_data = data + ALPHA_HEADER_LEN; + const size_t alpha_data_size = data_size - ALPHA_HEADER_LEN; + int rsrv; + VP8Io* const io = &dec->io_; + + assert(data != NULL && output != NULL && src_io != NULL); + + VP8FiltersInit(); + dec->output_ = output; + dec->width_ = src_io->width; + dec->height_ = src_io->height; + assert(dec->width_ > 0 && dec->height_ > 0); + + if (data_size <= ALPHA_HEADER_LEN) { + return 0; + } + + dec->method_ = (data[0] >> 0) & 0x03; + dec->filter_ = (WEBP_FILTER_TYPE)((data[0] >> 2) & 0x03); + dec->pre_processing_ = (data[0] >> 4) & 0x03; + rsrv = (data[0] >> 6) & 0x03; + if (dec->method_ < ALPHA_NO_COMPRESSION || + dec->method_ > ALPHA_LOSSLESS_COMPRESSION || + dec->filter_ >= WEBP_FILTER_LAST || + dec->pre_processing_ > ALPHA_PREPROCESSED_LEVELS || + rsrv != 0) { + return 0; + } + + // Copy the necessary parameters from src_io to io + VP8InitIo(io); + WebPInitCustomIo(NULL, io); + io->opaque = dec; + io->width = src_io->width; + io->height = src_io->height; + + io->use_cropping = src_io->use_cropping; + io->crop_left = src_io->crop_left; + io->crop_right = src_io->crop_right; + io->crop_top = src_io->crop_top; + io->crop_bottom = src_io->crop_bottom; + // No need to copy the scaling parameters. + + if (dec->method_ == ALPHA_NO_COMPRESSION) { + const size_t alpha_decoded_size = dec->width_ * dec->height_; + ok = (alpha_data_size >= alpha_decoded_size); + } else { + assert(dec->method_ == ALPHA_LOSSLESS_COMPRESSION); + ok = VP8LDecodeAlphaHeader(dec, alpha_data, alpha_data_size); + } + + return ok; +} + +// Decodes, unfilters and dequantizes *at least* 'num_rows' rows of alpha +// starting from row number 'row'. It assumes that rows up to (row - 1) have +// already been decoded. +// Returns false in case of bitstream error. +static int ALPHDecode(VP8Decoder* const dec, int row, int num_rows) { + ALPHDecoder* const alph_dec = dec->alph_dec_; + const int width = alph_dec->width_; + const int height = alph_dec->io_.crop_bottom; + if (alph_dec->method_ == ALPHA_NO_COMPRESSION) { + int y; + const uint8_t* prev_line = dec->alpha_prev_line_; + const uint8_t* deltas = dec->alpha_data_ + ALPHA_HEADER_LEN + row * width; + uint8_t* dst = dec->alpha_plane_ + row * width; + assert(deltas <= &dec->alpha_data_[dec->alpha_data_size_]); + if (alph_dec->filter_ != WEBP_FILTER_NONE) { + assert(WebPUnfilters[alph_dec->filter_] != NULL); + for (y = 0; y < num_rows; ++y) { + WebPUnfilters[alph_dec->filter_](prev_line, deltas, dst, width); + prev_line = dst; + dst += width; + deltas += width; + } + } else { + for (y = 0; y < num_rows; ++y) { + memcpy(dst, deltas, width * sizeof(*dst)); + prev_line = dst; + dst += width; + deltas += width; + } + } + dec->alpha_prev_line_ = prev_line; + } else { // alph_dec->method_ == ALPHA_LOSSLESS_COMPRESSION + assert(alph_dec->vp8l_dec_ != NULL); + if (!VP8LDecodeAlphaImageStream(alph_dec, row + num_rows)) { + return 0; + } + } + + if (row + num_rows >= height) { + dec->is_alpha_decoded_ = 1; + } + return 1; +} + +static int AllocateAlphaPlane(VP8Decoder* const dec, const VP8Io* const io) { + const int stride = io->width; + const int height = io->crop_bottom; + const uint64_t alpha_size = (uint64_t)stride * height; + assert(dec->alpha_plane_mem_ == NULL); + dec->alpha_plane_mem_ = + (uint8_t*)WebPSafeMalloc(alpha_size, sizeof(*dec->alpha_plane_)); + if (dec->alpha_plane_mem_ == NULL) { + return 0; + } + dec->alpha_plane_ = dec->alpha_plane_mem_; + dec->alpha_prev_line_ = NULL; + return 1; +} + +void WebPDeallocateAlphaMemory(VP8Decoder* const dec) { + assert(dec != NULL); + WebPSafeFree(dec->alpha_plane_mem_); + dec->alpha_plane_mem_ = NULL; + dec->alpha_plane_ = NULL; + ALPHDelete(dec->alph_dec_); + dec->alph_dec_ = NULL; +} + +//------------------------------------------------------------------------------ +// Main entry point. + +const uint8_t* VP8DecompressAlphaRows(VP8Decoder* const dec, + const VP8Io* const io, + int row, int num_rows) { + const int width = io->width; + const int height = io->crop_bottom; + + assert(dec != NULL && io != NULL); + + if (row < 0 || num_rows <= 0 || row + num_rows > height) { + return NULL; // sanity check. + } + + if (!dec->is_alpha_decoded_) { + if (dec->alph_dec_ == NULL) { // Initialize decoder. + dec->alph_dec_ = ALPHNew(); + if (dec->alph_dec_ == NULL) return NULL; + if (!AllocateAlphaPlane(dec, io)) goto Error; + if (!ALPHInit(dec->alph_dec_, dec->alpha_data_, dec->alpha_data_size_, + io, dec->alpha_plane_)) { + goto Error; + } + // if we allowed use of alpha dithering, check whether it's needed at all + if (dec->alph_dec_->pre_processing_ != ALPHA_PREPROCESSED_LEVELS) { + dec->alpha_dithering_ = 0; // disable dithering + } else { + num_rows = height - row; // decode everything in one pass + } + } + + assert(dec->alph_dec_ != NULL); + assert(row + num_rows <= height); + if (!ALPHDecode(dec, row, num_rows)) goto Error; + + if (dec->is_alpha_decoded_) { // finished? + ALPHDelete(dec->alph_dec_); + dec->alph_dec_ = NULL; + if (dec->alpha_dithering_ > 0) { + uint8_t* const alpha = dec->alpha_plane_ + io->crop_top * width + + io->crop_left; + if (!WebPDequantizeLevels(alpha, + io->crop_right - io->crop_left, + io->crop_bottom - io->crop_top, + width, dec->alpha_dithering_)) { + goto Error; + } + } + } + } + + // Return a pointer to the current decoded row. + return dec->alpha_plane_ + row * width; + + Error: + WebPDeallocateAlphaMemory(dec); + return NULL; +} diff --git a/thirdparty/libwebp/src/dec/alphai_dec.h b/thirdparty/libwebp/src/dec/alphai_dec.h new file mode 100644 index 0000000000..e0fa281a55 --- /dev/null +++ b/thirdparty/libwebp/src/dec/alphai_dec.h @@ -0,0 +1,54 @@ +// 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. +// ----------------------------------------------------------------------------- +// +// Alpha decoder: internal header. +// +// Author: Urvang (urvang@google.com) + +#ifndef WEBP_DEC_ALPHAI_DEC_H_ +#define WEBP_DEC_ALPHAI_DEC_H_ + +#include "src/dec/webpi_dec.h" +#include "src/utils/filters_utils.h" + +#ifdef __cplusplus +extern "C" { +#endif + +struct VP8LDecoder; // Defined in dec/vp8li.h. + +typedef struct ALPHDecoder ALPHDecoder; +struct ALPHDecoder { + int width_; + int height_; + int method_; + WEBP_FILTER_TYPE filter_; + int pre_processing_; + struct VP8LDecoder* vp8l_dec_; + VP8Io io_; + int use_8b_decode_; // Although alpha channel requires only 1 byte per + // pixel, sometimes VP8LDecoder may need to allocate + // 4 bytes per pixel internally during decode. + uint8_t* output_; + const uint8_t* prev_line_; // last output row (or NULL) +}; + +//------------------------------------------------------------------------------ +// internal functions. Not public. + +// Deallocate memory associated to dec->alpha_plane_ decoding +void WebPDeallocateAlphaMemory(VP8Decoder* const dec); + +//------------------------------------------------------------------------------ + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif /* WEBP_DEC_ALPHAI_DEC_H_ */ diff --git a/thirdparty/libwebp/src/dec/buffer_dec.c b/thirdparty/libwebp/src/dec/buffer_dec.c new file mode 100644 index 0000000000..75eb3c40b4 --- /dev/null +++ b/thirdparty/libwebp/src/dec/buffer_dec.c @@ -0,0 +1,311 @@ +// Copyright 2011 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. +// ----------------------------------------------------------------------------- +// +// Everything about WebPDecBuffer +// +// Author: Skal (pascal.massimino@gmail.com) + +#include <stdlib.h> + +#include "src/dec/vp8i_dec.h" +#include "src/dec/webpi_dec.h" +#include "src/utils/utils.h" + +//------------------------------------------------------------------------------ +// WebPDecBuffer + +// Number of bytes per pixel for the different color-spaces. +static const uint8_t kModeBpp[MODE_LAST] = { + 3, 4, 3, 4, 4, 2, 2, + 4, 4, 4, 2, // pre-multiplied modes + 1, 1 }; + +// Check that webp_csp_mode is within the bounds of WEBP_CSP_MODE. +// Convert to an integer to handle both the unsigned/signed enum cases +// without the need for casting to remove type limit warnings. +static int IsValidColorspace(int webp_csp_mode) { + return (webp_csp_mode >= MODE_RGB && webp_csp_mode < MODE_LAST); +} + +// strictly speaking, the very last (or first, if flipped) row +// doesn't require padding. +#define MIN_BUFFER_SIZE(WIDTH, HEIGHT, STRIDE) \ + ((uint64_t)(STRIDE) * ((HEIGHT) - 1) + (WIDTH)) + +static VP8StatusCode CheckDecBuffer(const WebPDecBuffer* const buffer) { + int ok = 1; + const WEBP_CSP_MODE mode = buffer->colorspace; + const int width = buffer->width; + const int height = buffer->height; + if (!IsValidColorspace(mode)) { + ok = 0; + } else if (!WebPIsRGBMode(mode)) { // YUV checks + const WebPYUVABuffer* const buf = &buffer->u.YUVA; + const int uv_width = (width + 1) / 2; + const int uv_height = (height + 1) / 2; + const int y_stride = abs(buf->y_stride); + const int u_stride = abs(buf->u_stride); + const int v_stride = abs(buf->v_stride); + const int a_stride = abs(buf->a_stride); + const uint64_t y_size = MIN_BUFFER_SIZE(width, height, y_stride); + const uint64_t u_size = MIN_BUFFER_SIZE(uv_width, uv_height, u_stride); + const uint64_t v_size = MIN_BUFFER_SIZE(uv_width, uv_height, v_stride); + const uint64_t a_size = MIN_BUFFER_SIZE(width, height, a_stride); + ok &= (y_size <= buf->y_size); + ok &= (u_size <= buf->u_size); + ok &= (v_size <= buf->v_size); + ok &= (y_stride >= width); + ok &= (u_stride >= uv_width); + ok &= (v_stride >= uv_width); + ok &= (buf->y != NULL); + ok &= (buf->u != NULL); + ok &= (buf->v != NULL); + if (mode == MODE_YUVA) { + ok &= (a_stride >= width); + ok &= (a_size <= buf->a_size); + ok &= (buf->a != NULL); + } + } else { // RGB checks + const WebPRGBABuffer* const buf = &buffer->u.RGBA; + const int stride = abs(buf->stride); + const uint64_t size = MIN_BUFFER_SIZE(width, height, stride); + ok &= (size <= buf->size); + ok &= (stride >= width * kModeBpp[mode]); + ok &= (buf->rgba != NULL); + } + return ok ? VP8_STATUS_OK : VP8_STATUS_INVALID_PARAM; +} +#undef MIN_BUFFER_SIZE + +static VP8StatusCode AllocateBuffer(WebPDecBuffer* const buffer) { + const int w = buffer->width; + const int h = buffer->height; + const WEBP_CSP_MODE mode = buffer->colorspace; + + if (w <= 0 || h <= 0 || !IsValidColorspace(mode)) { + return VP8_STATUS_INVALID_PARAM; + } + + if (buffer->is_external_memory <= 0 && buffer->private_memory == NULL) { + uint8_t* output; + int uv_stride = 0, a_stride = 0; + uint64_t uv_size = 0, a_size = 0, total_size; + // We need memory and it hasn't been allocated yet. + // => initialize output buffer, now that dimensions are known. + int stride; + uint64_t size; + + if ((uint64_t)w * kModeBpp[mode] >= (1ull << 32)) { + return VP8_STATUS_INVALID_PARAM; + } + stride = w * kModeBpp[mode]; + size = (uint64_t)stride * h; + if (!WebPIsRGBMode(mode)) { + uv_stride = (w + 1) / 2; + uv_size = (uint64_t)uv_stride * ((h + 1) / 2); + if (mode == MODE_YUVA) { + a_stride = w; + a_size = (uint64_t)a_stride * h; + } + } + total_size = size + 2 * uv_size + a_size; + + // Security/sanity checks + output = (uint8_t*)WebPSafeMalloc(total_size, sizeof(*output)); + if (output == NULL) { + return VP8_STATUS_OUT_OF_MEMORY; + } + buffer->private_memory = output; + + if (!WebPIsRGBMode(mode)) { // YUVA initialization + WebPYUVABuffer* const buf = &buffer->u.YUVA; + buf->y = output; + buf->y_stride = stride; + buf->y_size = (size_t)size; + buf->u = output + size; + buf->u_stride = uv_stride; + buf->u_size = (size_t)uv_size; + buf->v = output + size + uv_size; + buf->v_stride = uv_stride; + buf->v_size = (size_t)uv_size; + if (mode == MODE_YUVA) { + buf->a = output + size + 2 * uv_size; + } + buf->a_size = (size_t)a_size; + buf->a_stride = a_stride; + } else { // RGBA initialization + WebPRGBABuffer* const buf = &buffer->u.RGBA; + buf->rgba = output; + buf->stride = stride; + buf->size = (size_t)size; + } + } + return CheckDecBuffer(buffer); +} + +VP8StatusCode WebPFlipBuffer(WebPDecBuffer* const buffer) { + if (buffer == NULL) { + return VP8_STATUS_INVALID_PARAM; + } + if (WebPIsRGBMode(buffer->colorspace)) { + WebPRGBABuffer* const buf = &buffer->u.RGBA; + buf->rgba += (buffer->height - 1) * buf->stride; + buf->stride = -buf->stride; + } else { + WebPYUVABuffer* const buf = &buffer->u.YUVA; + const int H = buffer->height; + buf->y += (H - 1) * buf->y_stride; + buf->y_stride = -buf->y_stride; + buf->u += ((H - 1) >> 1) * buf->u_stride; + buf->u_stride = -buf->u_stride; + buf->v += ((H - 1) >> 1) * buf->v_stride; + buf->v_stride = -buf->v_stride; + if (buf->a != NULL) { + buf->a += (H - 1) * buf->a_stride; + buf->a_stride = -buf->a_stride; + } + } + return VP8_STATUS_OK; +} + +VP8StatusCode WebPAllocateDecBuffer(int width, int height, + const WebPDecoderOptions* const options, + WebPDecBuffer* const buffer) { + VP8StatusCode status; + if (buffer == NULL || width <= 0 || height <= 0) { + return VP8_STATUS_INVALID_PARAM; + } + if (options != NULL) { // First, apply options if there is any. + if (options->use_cropping) { + const int cw = options->crop_width; + const int ch = options->crop_height; + const int x = options->crop_left & ~1; + const int y = options->crop_top & ~1; + if (x < 0 || y < 0 || cw <= 0 || ch <= 0 || + x + cw > width || y + ch > height) { + return VP8_STATUS_INVALID_PARAM; // out of frame boundary. + } + width = cw; + height = ch; + } + + if (options->use_scaling) { +#if !defined(WEBP_REDUCE_SIZE) + int scaled_width = options->scaled_width; + int scaled_height = options->scaled_height; + if (!WebPRescalerGetScaledDimensions( + width, height, &scaled_width, &scaled_height)) { + return VP8_STATUS_INVALID_PARAM; + } + width = scaled_width; + height = scaled_height; +#else + return VP8_STATUS_INVALID_PARAM; // rescaling not supported +#endif + } + } + buffer->width = width; + buffer->height = height; + + // Then, allocate buffer for real. + status = AllocateBuffer(buffer); + if (status != VP8_STATUS_OK) return status; + + // Use the stride trick if vertical flip is needed. + if (options != NULL && options->flip) { + status = WebPFlipBuffer(buffer); + } + return status; +} + +//------------------------------------------------------------------------------ +// constructors / destructors + +int WebPInitDecBufferInternal(WebPDecBuffer* buffer, int version) { + if (WEBP_ABI_IS_INCOMPATIBLE(version, WEBP_DECODER_ABI_VERSION)) { + return 0; // version mismatch + } + if (buffer == NULL) return 0; + memset(buffer, 0, sizeof(*buffer)); + return 1; +} + +void WebPFreeDecBuffer(WebPDecBuffer* buffer) { + if (buffer != NULL) { + if (buffer->is_external_memory <= 0) { + WebPSafeFree(buffer->private_memory); + } + buffer->private_memory = NULL; + } +} + +void WebPCopyDecBuffer(const WebPDecBuffer* const src, + WebPDecBuffer* const dst) { + if (src != NULL && dst != NULL) { + *dst = *src; + if (src->private_memory != NULL) { + dst->is_external_memory = 1; // dst buffer doesn't own the memory. + dst->private_memory = NULL; + } + } +} + +// Copy and transfer ownership from src to dst (beware of parameter order!) +void WebPGrabDecBuffer(WebPDecBuffer* const src, WebPDecBuffer* const dst) { + if (src != NULL && dst != NULL) { + *dst = *src; + if (src->private_memory != NULL) { + src->is_external_memory = 1; // src relinquishes ownership + src->private_memory = NULL; + } + } +} + +VP8StatusCode WebPCopyDecBufferPixels(const WebPDecBuffer* const src_buf, + WebPDecBuffer* const dst_buf) { + assert(src_buf != NULL && dst_buf != NULL); + assert(src_buf->colorspace == dst_buf->colorspace); + + dst_buf->width = src_buf->width; + dst_buf->height = src_buf->height; + if (CheckDecBuffer(dst_buf) != VP8_STATUS_OK) { + return VP8_STATUS_INVALID_PARAM; + } + if (WebPIsRGBMode(src_buf->colorspace)) { + const WebPRGBABuffer* const src = &src_buf->u.RGBA; + const WebPRGBABuffer* const dst = &dst_buf->u.RGBA; + WebPCopyPlane(src->rgba, src->stride, dst->rgba, dst->stride, + src_buf->width * kModeBpp[src_buf->colorspace], + src_buf->height); + } else { + const WebPYUVABuffer* const src = &src_buf->u.YUVA; + const WebPYUVABuffer* const dst = &dst_buf->u.YUVA; + WebPCopyPlane(src->y, src->y_stride, dst->y, dst->y_stride, + src_buf->width, src_buf->height); + WebPCopyPlane(src->u, src->u_stride, dst->u, dst->u_stride, + (src_buf->width + 1) / 2, (src_buf->height + 1) / 2); + WebPCopyPlane(src->v, src->v_stride, dst->v, dst->v_stride, + (src_buf->width + 1) / 2, (src_buf->height + 1) / 2); + if (WebPIsAlphaMode(src_buf->colorspace)) { + WebPCopyPlane(src->a, src->a_stride, dst->a, dst->a_stride, + src_buf->width, src_buf->height); + } + } + return VP8_STATUS_OK; +} + +int WebPAvoidSlowMemory(const WebPDecBuffer* const output, + const WebPBitstreamFeatures* const features) { + assert(output != NULL); + return (output->is_external_memory >= 2) && + WebPIsPremultipliedMode(output->colorspace) && + (features != NULL && features->has_alpha); +} + +//------------------------------------------------------------------------------ diff --git a/thirdparty/libwebp/src/dec/common_dec.h b/thirdparty/libwebp/src/dec/common_dec.h new file mode 100644 index 0000000000..9995f1a51a --- /dev/null +++ b/thirdparty/libwebp/src/dec/common_dec.h @@ -0,0 +1,54 @@ +// Copyright 2015 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. +// ----------------------------------------------------------------------------- +// +// Definitions and macros common to encoding and decoding +// +// Author: Skal (pascal.massimino@gmail.com) + +#ifndef WEBP_DEC_COMMON_DEC_H_ +#define WEBP_DEC_COMMON_DEC_H_ + +// intra prediction modes +enum { B_DC_PRED = 0, // 4x4 modes + B_TM_PRED = 1, + B_VE_PRED = 2, + B_HE_PRED = 3, + B_RD_PRED = 4, + B_VR_PRED = 5, + B_LD_PRED = 6, + B_VL_PRED = 7, + B_HD_PRED = 8, + B_HU_PRED = 9, + NUM_BMODES = B_HU_PRED + 1 - B_DC_PRED, // = 10 + + // Luma16 or UV modes + DC_PRED = B_DC_PRED, V_PRED = B_VE_PRED, + H_PRED = B_HE_PRED, TM_PRED = B_TM_PRED, + B_PRED = NUM_BMODES, // refined I4x4 mode + NUM_PRED_MODES = 4, + + // special modes + B_DC_PRED_NOTOP = 4, + B_DC_PRED_NOLEFT = 5, + B_DC_PRED_NOTOPLEFT = 6, + NUM_B_DC_MODES = 7 }; + +enum { MB_FEATURE_TREE_PROBS = 3, + NUM_MB_SEGMENTS = 4, + NUM_REF_LF_DELTAS = 4, + NUM_MODE_LF_DELTAS = 4, // I4x4, ZERO, *, SPLIT + MAX_NUM_PARTITIONS = 8, + // Probabilities + NUM_TYPES = 4, // 0: i16-AC, 1: i16-DC, 2:chroma-AC, 3:i4-AC + NUM_BANDS = 8, + NUM_CTX = 3, + NUM_PROBAS = 11 + }; + +#endif // WEBP_DEC_COMMON_DEC_H_ diff --git a/thirdparty/libwebp/src/dec/frame_dec.c b/thirdparty/libwebp/src/dec/frame_dec.c new file mode 100644 index 0000000000..517d0f5850 --- /dev/null +++ b/thirdparty/libwebp/src/dec/frame_dec.c @@ -0,0 +1,812 @@ +// Copyright 2010 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. +// ----------------------------------------------------------------------------- +// +// Frame-reconstruction function. Memory allocation. +// +// Author: Skal (pascal.massimino@gmail.com) + +#include <stdlib.h> +#include "src/dec/vp8i_dec.h" +#include "src/utils/utils.h" + +//------------------------------------------------------------------------------ +// Main reconstruction function. + +static const uint16_t kScan[16] = { + 0 + 0 * BPS, 4 + 0 * BPS, 8 + 0 * BPS, 12 + 0 * BPS, + 0 + 4 * BPS, 4 + 4 * BPS, 8 + 4 * BPS, 12 + 4 * BPS, + 0 + 8 * BPS, 4 + 8 * BPS, 8 + 8 * BPS, 12 + 8 * BPS, + 0 + 12 * BPS, 4 + 12 * BPS, 8 + 12 * BPS, 12 + 12 * BPS +}; + +static int CheckMode(int mb_x, int mb_y, int mode) { + if (mode == B_DC_PRED) { + if (mb_x == 0) { + return (mb_y == 0) ? B_DC_PRED_NOTOPLEFT : B_DC_PRED_NOLEFT; + } else { + return (mb_y == 0) ? B_DC_PRED_NOTOP : B_DC_PRED; + } + } + return mode; +} + +static void Copy32b(uint8_t* const dst, const uint8_t* const src) { + memcpy(dst, src, 4); +} + +static WEBP_INLINE void DoTransform(uint32_t bits, const int16_t* const src, + uint8_t* const dst) { + switch (bits >> 30) { + case 3: + VP8Transform(src, dst, 0); + break; + case 2: + VP8TransformAC3(src, dst); + break; + case 1: + VP8TransformDC(src, dst); + break; + default: + break; + } +} + +static void DoUVTransform(uint32_t bits, const int16_t* const src, + uint8_t* const dst) { + if (bits & 0xff) { // any non-zero coeff at all? + if (bits & 0xaa) { // any non-zero AC coefficient? + VP8TransformUV(src, dst); // note we don't use the AC3 variant for U/V + } else { + VP8TransformDCUV(src, dst); + } + } +} + +static void ReconstructRow(const VP8Decoder* const dec, + const VP8ThreadContext* ctx) { + int j; + int mb_x; + const int mb_y = ctx->mb_y_; + const int cache_id = ctx->id_; + uint8_t* const y_dst = dec->yuv_b_ + Y_OFF; + uint8_t* const u_dst = dec->yuv_b_ + U_OFF; + uint8_t* const v_dst = dec->yuv_b_ + V_OFF; + + // Initialize left-most block. + for (j = 0; j < 16; ++j) { + y_dst[j * BPS - 1] = 129; + } + for (j = 0; j < 8; ++j) { + u_dst[j * BPS - 1] = 129; + v_dst[j * BPS - 1] = 129; + } + + // Init top-left sample on left column too. + if (mb_y > 0) { + y_dst[-1 - BPS] = u_dst[-1 - BPS] = v_dst[-1 - BPS] = 129; + } else { + // we only need to do this init once at block (0,0). + // Afterward, it remains valid for the whole topmost row. + memset(y_dst - BPS - 1, 127, 16 + 4 + 1); + memset(u_dst - BPS - 1, 127, 8 + 1); + memset(v_dst - BPS - 1, 127, 8 + 1); + } + + // Reconstruct one row. + for (mb_x = 0; mb_x < dec->mb_w_; ++mb_x) { + const VP8MBData* const block = ctx->mb_data_ + mb_x; + + // Rotate in the left samples from previously decoded block. We move four + // pixels at a time for alignment reason, and because of in-loop filter. + if (mb_x > 0) { + for (j = -1; j < 16; ++j) { + Copy32b(&y_dst[j * BPS - 4], &y_dst[j * BPS + 12]); + } + for (j = -1; j < 8; ++j) { + Copy32b(&u_dst[j * BPS - 4], &u_dst[j * BPS + 4]); + Copy32b(&v_dst[j * BPS - 4], &v_dst[j * BPS + 4]); + } + } + { + // bring top samples into the cache + VP8TopSamples* const top_yuv = dec->yuv_t_ + mb_x; + const int16_t* const coeffs = block->coeffs_; + uint32_t bits = block->non_zero_y_; + int n; + + if (mb_y > 0) { + memcpy(y_dst - BPS, top_yuv[0].y, 16); + memcpy(u_dst - BPS, top_yuv[0].u, 8); + memcpy(v_dst - BPS, top_yuv[0].v, 8); + } + + // predict and add residuals + if (block->is_i4x4_) { // 4x4 + uint32_t* const top_right = (uint32_t*)(y_dst - BPS + 16); + + if (mb_y > 0) { + if (mb_x >= dec->mb_w_ - 1) { // on rightmost border + memset(top_right, top_yuv[0].y[15], sizeof(*top_right)); + } else { + memcpy(top_right, top_yuv[1].y, sizeof(*top_right)); + } + } + // replicate the top-right pixels below + top_right[BPS] = top_right[2 * BPS] = top_right[3 * BPS] = top_right[0]; + + // predict and add residuals for all 4x4 blocks in turn. + for (n = 0; n < 16; ++n, bits <<= 2) { + uint8_t* const dst = y_dst + kScan[n]; + VP8PredLuma4[block->imodes_[n]](dst); + DoTransform(bits, coeffs + n * 16, dst); + } + } else { // 16x16 + const int pred_func = CheckMode(mb_x, mb_y, block->imodes_[0]); + VP8PredLuma16[pred_func](y_dst); + if (bits != 0) { + for (n = 0; n < 16; ++n, bits <<= 2) { + DoTransform(bits, coeffs + n * 16, y_dst + kScan[n]); + } + } + } + { + // Chroma + const uint32_t bits_uv = block->non_zero_uv_; + const int pred_func = CheckMode(mb_x, mb_y, block->uvmode_); + VP8PredChroma8[pred_func](u_dst); + VP8PredChroma8[pred_func](v_dst); + DoUVTransform(bits_uv >> 0, coeffs + 16 * 16, u_dst); + DoUVTransform(bits_uv >> 8, coeffs + 20 * 16, v_dst); + } + + // stash away top samples for next block + if (mb_y < dec->mb_h_ - 1) { + memcpy(top_yuv[0].y, y_dst + 15 * BPS, 16); + memcpy(top_yuv[0].u, u_dst + 7 * BPS, 8); + memcpy(top_yuv[0].v, v_dst + 7 * BPS, 8); + } + } + // Transfer reconstructed samples from yuv_b_ cache to final destination. + { + const int y_offset = cache_id * 16 * dec->cache_y_stride_; + const int uv_offset = cache_id * 8 * dec->cache_uv_stride_; + uint8_t* const y_out = dec->cache_y_ + mb_x * 16 + y_offset; + uint8_t* const u_out = dec->cache_u_ + mb_x * 8 + uv_offset; + uint8_t* const v_out = dec->cache_v_ + mb_x * 8 + uv_offset; + for (j = 0; j < 16; ++j) { + memcpy(y_out + j * dec->cache_y_stride_, y_dst + j * BPS, 16); + } + for (j = 0; j < 8; ++j) { + memcpy(u_out + j * dec->cache_uv_stride_, u_dst + j * BPS, 8); + memcpy(v_out + j * dec->cache_uv_stride_, v_dst + j * BPS, 8); + } + } + } +} + +//------------------------------------------------------------------------------ +// Filtering + +// kFilterExtraRows[] = How many extra lines are needed on the MB boundary +// for caching, given a filtering level. +// Simple filter: up to 2 luma samples are read and 1 is written. +// Complex filter: up to 4 luma samples are read and 3 are written. Same for +// U/V, so it's 8 samples total (because of the 2x upsampling). +static const uint8_t kFilterExtraRows[3] = { 0, 2, 8 }; + +static void DoFilter(const VP8Decoder* const dec, int mb_x, int mb_y) { + const VP8ThreadContext* const ctx = &dec->thread_ctx_; + const int cache_id = ctx->id_; + const int y_bps = dec->cache_y_stride_; + const VP8FInfo* const f_info = ctx->f_info_ + mb_x; + uint8_t* const y_dst = dec->cache_y_ + cache_id * 16 * y_bps + mb_x * 16; + const int ilevel = f_info->f_ilevel_; + const int limit = f_info->f_limit_; + if (limit == 0) { + return; + } + assert(limit >= 3); + if (dec->filter_type_ == 1) { // simple + if (mb_x > 0) { + VP8SimpleHFilter16(y_dst, y_bps, limit + 4); + } + if (f_info->f_inner_) { + VP8SimpleHFilter16i(y_dst, y_bps, limit); + } + if (mb_y > 0) { + VP8SimpleVFilter16(y_dst, y_bps, limit + 4); + } + if (f_info->f_inner_) { + VP8SimpleVFilter16i(y_dst, y_bps, limit); + } + } else { // complex + const int uv_bps = dec->cache_uv_stride_; + uint8_t* const u_dst = dec->cache_u_ + cache_id * 8 * uv_bps + mb_x * 8; + uint8_t* const v_dst = dec->cache_v_ + cache_id * 8 * uv_bps + mb_x * 8; + const int hev_thresh = f_info->hev_thresh_; + if (mb_x > 0) { + VP8HFilter16(y_dst, y_bps, limit + 4, ilevel, hev_thresh); + VP8HFilter8(u_dst, v_dst, uv_bps, limit + 4, ilevel, hev_thresh); + } + if (f_info->f_inner_) { + VP8HFilter16i(y_dst, y_bps, limit, ilevel, hev_thresh); + VP8HFilter8i(u_dst, v_dst, uv_bps, limit, ilevel, hev_thresh); + } + if (mb_y > 0) { + VP8VFilter16(y_dst, y_bps, limit + 4, ilevel, hev_thresh); + VP8VFilter8(u_dst, v_dst, uv_bps, limit + 4, ilevel, hev_thresh); + } + if (f_info->f_inner_) { + VP8VFilter16i(y_dst, y_bps, limit, ilevel, hev_thresh); + VP8VFilter8i(u_dst, v_dst, uv_bps, limit, ilevel, hev_thresh); + } + } +} + +// Filter the decoded macroblock row (if needed) +static void FilterRow(const VP8Decoder* const dec) { + int mb_x; + const int mb_y = dec->thread_ctx_.mb_y_; + assert(dec->thread_ctx_.filter_row_); + for (mb_x = dec->tl_mb_x_; mb_x < dec->br_mb_x_; ++mb_x) { + DoFilter(dec, mb_x, mb_y); + } +} + +//------------------------------------------------------------------------------ +// Precompute the filtering strength for each segment and each i4x4/i16x16 mode. + +static void PrecomputeFilterStrengths(VP8Decoder* const dec) { + if (dec->filter_type_ > 0) { + int s; + const VP8FilterHeader* const hdr = &dec->filter_hdr_; + for (s = 0; s < NUM_MB_SEGMENTS; ++s) { + int i4x4; + // First, compute the initial level + int base_level; + if (dec->segment_hdr_.use_segment_) { + base_level = dec->segment_hdr_.filter_strength_[s]; + if (!dec->segment_hdr_.absolute_delta_) { + base_level += hdr->level_; + } + } else { + base_level = hdr->level_; + } + for (i4x4 = 0; i4x4 <= 1; ++i4x4) { + VP8FInfo* const info = &dec->fstrengths_[s][i4x4]; + int level = base_level; + if (hdr->use_lf_delta_) { + level += hdr->ref_lf_delta_[0]; + if (i4x4) { + level += hdr->mode_lf_delta_[0]; + } + } + level = (level < 0) ? 0 : (level > 63) ? 63 : level; + if (level > 0) { + int ilevel = level; + if (hdr->sharpness_ > 0) { + if (hdr->sharpness_ > 4) { + ilevel >>= 2; + } else { + ilevel >>= 1; + } + if (ilevel > 9 - hdr->sharpness_) { + ilevel = 9 - hdr->sharpness_; + } + } + if (ilevel < 1) ilevel = 1; + info->f_ilevel_ = ilevel; + info->f_limit_ = 2 * level + ilevel; + info->hev_thresh_ = (level >= 40) ? 2 : (level >= 15) ? 1 : 0; + } else { + info->f_limit_ = 0; // no filtering + } + info->f_inner_ = i4x4; + } + } + } +} + +//------------------------------------------------------------------------------ +// Dithering + +// minimal amp that will provide a non-zero dithering effect +#define MIN_DITHER_AMP 4 + +#define DITHER_AMP_TAB_SIZE 12 +static const uint8_t kQuantToDitherAmp[DITHER_AMP_TAB_SIZE] = { + // roughly, it's dqm->uv_mat_[1] + 8, 7, 6, 4, 4, 2, 2, 2, 1, 1, 1, 1 +}; + +void VP8InitDithering(const WebPDecoderOptions* const options, + VP8Decoder* const dec) { + assert(dec != NULL); + if (options != NULL) { + const int d = options->dithering_strength; + const int max_amp = (1 << VP8_RANDOM_DITHER_FIX) - 1; + const int f = (d < 0) ? 0 : (d > 100) ? max_amp : (d * max_amp / 100); + if (f > 0) { + int s; + int all_amp = 0; + for (s = 0; s < NUM_MB_SEGMENTS; ++s) { + VP8QuantMatrix* const dqm = &dec->dqm_[s]; + if (dqm->uv_quant_ < DITHER_AMP_TAB_SIZE) { + // TODO(skal): should we specially dither more for uv_quant_ < 0? + const int idx = (dqm->uv_quant_ < 0) ? 0 : dqm->uv_quant_; + dqm->dither_ = (f * kQuantToDitherAmp[idx]) >> 3; + } + all_amp |= dqm->dither_; + } + if (all_amp != 0) { + VP8InitRandom(&dec->dithering_rg_, 1.0f); + dec->dither_ = 1; + } + } + // potentially allow alpha dithering + dec->alpha_dithering_ = options->alpha_dithering_strength; + if (dec->alpha_dithering_ > 100) { + dec->alpha_dithering_ = 100; + } else if (dec->alpha_dithering_ < 0) { + dec->alpha_dithering_ = 0; + } + } +} + +// Convert to range: [-2,2] for dither=50, [-4,4] for dither=100 +static void Dither8x8(VP8Random* const rg, uint8_t* dst, int bps, int amp) { + uint8_t dither[64]; + int i; + for (i = 0; i < 8 * 8; ++i) { + dither[i] = VP8RandomBits2(rg, VP8_DITHER_AMP_BITS + 1, amp); + } + VP8DitherCombine8x8(dither, dst, bps); +} + +static void DitherRow(VP8Decoder* const dec) { + int mb_x; + assert(dec->dither_); + for (mb_x = dec->tl_mb_x_; mb_x < dec->br_mb_x_; ++mb_x) { + const VP8ThreadContext* const ctx = &dec->thread_ctx_; + const VP8MBData* const data = ctx->mb_data_ + mb_x; + const int cache_id = ctx->id_; + const int uv_bps = dec->cache_uv_stride_; + if (data->dither_ >= MIN_DITHER_AMP) { + uint8_t* const u_dst = dec->cache_u_ + cache_id * 8 * uv_bps + mb_x * 8; + uint8_t* const v_dst = dec->cache_v_ + cache_id * 8 * uv_bps + mb_x * 8; + Dither8x8(&dec->dithering_rg_, u_dst, uv_bps, data->dither_); + Dither8x8(&dec->dithering_rg_, v_dst, uv_bps, data->dither_); + } + } +} + +//------------------------------------------------------------------------------ +// This function is called after a row of macroblocks is finished decoding. +// It also takes into account the following restrictions: +// * In case of in-loop filtering, we must hold off sending some of the bottom +// pixels as they are yet unfiltered. They will be when the next macroblock +// row is decoded. Meanwhile, we must preserve them by rotating them in the +// cache area. This doesn't hold for the very bottom row of the uncropped +// picture of course. +// * we must clip the remaining pixels against the cropping area. The VP8Io +// struct must have the following fields set correctly before calling put(): + +#define MACROBLOCK_VPOS(mb_y) ((mb_y) * 16) // vertical position of a MB + +// Finalize and transmit a complete row. Return false in case of user-abort. +static int FinishRow(VP8Decoder* const dec, VP8Io* const io) { + int ok = 1; + const VP8ThreadContext* const ctx = &dec->thread_ctx_; + const int cache_id = ctx->id_; + const int extra_y_rows = kFilterExtraRows[dec->filter_type_]; + const int ysize = extra_y_rows * dec->cache_y_stride_; + const int uvsize = (extra_y_rows / 2) * dec->cache_uv_stride_; + const int y_offset = cache_id * 16 * dec->cache_y_stride_; + const int uv_offset = cache_id * 8 * dec->cache_uv_stride_; + uint8_t* const ydst = dec->cache_y_ - ysize + y_offset; + uint8_t* const udst = dec->cache_u_ - uvsize + uv_offset; + uint8_t* const vdst = dec->cache_v_ - uvsize + uv_offset; + const int mb_y = ctx->mb_y_; + const int is_first_row = (mb_y == 0); + const int is_last_row = (mb_y >= dec->br_mb_y_ - 1); + + if (dec->mt_method_ == 2) { + ReconstructRow(dec, ctx); + } + + if (ctx->filter_row_) { + FilterRow(dec); + } + + if (dec->dither_) { + DitherRow(dec); + } + + if (io->put != NULL) { + int y_start = MACROBLOCK_VPOS(mb_y); + int y_end = MACROBLOCK_VPOS(mb_y + 1); + if (!is_first_row) { + y_start -= extra_y_rows; + io->y = ydst; + io->u = udst; + io->v = vdst; + } else { + io->y = dec->cache_y_ + y_offset; + io->u = dec->cache_u_ + uv_offset; + io->v = dec->cache_v_ + uv_offset; + } + + if (!is_last_row) { + y_end -= extra_y_rows; + } + if (y_end > io->crop_bottom) { + y_end = io->crop_bottom; // make sure we don't overflow on last row. + } + io->a = NULL; + if (dec->alpha_data_ != NULL && y_start < y_end) { + // TODO(skal): testing presence of alpha with dec->alpha_data_ is not a + // good idea. + io->a = VP8DecompressAlphaRows(dec, io, y_start, y_end - y_start); + if (io->a == NULL) { + return VP8SetError(dec, VP8_STATUS_BITSTREAM_ERROR, + "Could not decode alpha data."); + } + } + if (y_start < io->crop_top) { + const int delta_y = io->crop_top - y_start; + y_start = io->crop_top; + assert(!(delta_y & 1)); + io->y += dec->cache_y_stride_ * delta_y; + io->u += dec->cache_uv_stride_ * (delta_y >> 1); + io->v += dec->cache_uv_stride_ * (delta_y >> 1); + if (io->a != NULL) { + io->a += io->width * delta_y; + } + } + if (y_start < y_end) { + io->y += io->crop_left; + io->u += io->crop_left >> 1; + io->v += io->crop_left >> 1; + if (io->a != NULL) { + io->a += io->crop_left; + } + io->mb_y = y_start - io->crop_top; + io->mb_w = io->crop_right - io->crop_left; + io->mb_h = y_end - y_start; + ok = io->put(io); + } + } + // rotate top samples if needed + if (cache_id + 1 == dec->num_caches_) { + if (!is_last_row) { + memcpy(dec->cache_y_ - ysize, ydst + 16 * dec->cache_y_stride_, ysize); + memcpy(dec->cache_u_ - uvsize, udst + 8 * dec->cache_uv_stride_, uvsize); + memcpy(dec->cache_v_ - uvsize, vdst + 8 * dec->cache_uv_stride_, uvsize); + } + } + + return ok; +} + +#undef MACROBLOCK_VPOS + +//------------------------------------------------------------------------------ + +int VP8ProcessRow(VP8Decoder* const dec, VP8Io* const io) { + int ok = 1; + VP8ThreadContext* const ctx = &dec->thread_ctx_; + const int filter_row = + (dec->filter_type_ > 0) && + (dec->mb_y_ >= dec->tl_mb_y_) && (dec->mb_y_ <= dec->br_mb_y_); + if (dec->mt_method_ == 0) { + // ctx->id_ and ctx->f_info_ are already set + ctx->mb_y_ = dec->mb_y_; + ctx->filter_row_ = filter_row; + ReconstructRow(dec, ctx); + ok = FinishRow(dec, io); + } else { + WebPWorker* const worker = &dec->worker_; + // Finish previous job *before* updating context + ok &= WebPGetWorkerInterface()->Sync(worker); + assert(worker->status_ == OK); + if (ok) { // spawn a new deblocking/output job + ctx->io_ = *io; + ctx->id_ = dec->cache_id_; + ctx->mb_y_ = dec->mb_y_; + ctx->filter_row_ = filter_row; + if (dec->mt_method_ == 2) { // swap macroblock data + VP8MBData* const tmp = ctx->mb_data_; + ctx->mb_data_ = dec->mb_data_; + dec->mb_data_ = tmp; + } else { + // perform reconstruction directly in main thread + ReconstructRow(dec, ctx); + } + if (filter_row) { // swap filter info + VP8FInfo* const tmp = ctx->f_info_; + ctx->f_info_ = dec->f_info_; + dec->f_info_ = tmp; + } + // (reconstruct)+filter in parallel + WebPGetWorkerInterface()->Launch(worker); + if (++dec->cache_id_ == dec->num_caches_) { + dec->cache_id_ = 0; + } + } + } + return ok; +} + +//------------------------------------------------------------------------------ +// Finish setting up the decoding parameter once user's setup() is called. + +VP8StatusCode VP8EnterCritical(VP8Decoder* const dec, VP8Io* const io) { + // Call setup() first. This may trigger additional decoding features on 'io'. + // Note: Afterward, we must call teardown() no matter what. + if (io->setup != NULL && !io->setup(io)) { + VP8SetError(dec, VP8_STATUS_USER_ABORT, "Frame setup failed"); + return dec->status_; + } + + // Disable filtering per user request + if (io->bypass_filtering) { + dec->filter_type_ = 0; + } + // TODO(skal): filter type / strength / sharpness forcing + + // Define the area where we can skip in-loop filtering, in case of cropping. + // + // 'Simple' filter reads two luma samples outside of the macroblock + // and filters one. It doesn't filter the chroma samples. Hence, we can + // avoid doing the in-loop filtering before crop_top/crop_left position. + // For the 'Complex' filter, 3 samples are read and up to 3 are filtered. + // Means: there's a dependency chain that goes all the way up to the + // top-left corner of the picture (MB #0). We must filter all the previous + // macroblocks. + // TODO(skal): add an 'approximate_decoding' option, that won't produce + // a 1:1 bit-exactness for complex filtering? + { + const int extra_pixels = kFilterExtraRows[dec->filter_type_]; + if (dec->filter_type_ == 2) { + // For complex filter, we need to preserve the dependency chain. + dec->tl_mb_x_ = 0; + dec->tl_mb_y_ = 0; + } else { + // For simple filter, we can filter only the cropped region. + // We include 'extra_pixels' on the other side of the boundary, since + // vertical or horizontal filtering of the previous macroblock can + // modify some abutting pixels. + dec->tl_mb_x_ = (io->crop_left - extra_pixels) >> 4; + dec->tl_mb_y_ = (io->crop_top - extra_pixels) >> 4; + if (dec->tl_mb_x_ < 0) dec->tl_mb_x_ = 0; + if (dec->tl_mb_y_ < 0) dec->tl_mb_y_ = 0; + } + // We need some 'extra' pixels on the right/bottom. + dec->br_mb_y_ = (io->crop_bottom + 15 + extra_pixels) >> 4; + dec->br_mb_x_ = (io->crop_right + 15 + extra_pixels) >> 4; + if (dec->br_mb_x_ > dec->mb_w_) { + dec->br_mb_x_ = dec->mb_w_; + } + if (dec->br_mb_y_ > dec->mb_h_) { + dec->br_mb_y_ = dec->mb_h_; + } + } + PrecomputeFilterStrengths(dec); + return VP8_STATUS_OK; +} + +int VP8ExitCritical(VP8Decoder* const dec, VP8Io* const io) { + int ok = 1; + if (dec->mt_method_ > 0) { + ok = WebPGetWorkerInterface()->Sync(&dec->worker_); + } + + if (io->teardown != NULL) { + io->teardown(io); + } + return ok; +} + +//------------------------------------------------------------------------------ +// For multi-threaded decoding we need to use 3 rows of 16 pixels as delay line. +// +// Reason is: the deblocking filter cannot deblock the bottom horizontal edges +// immediately, and needs to wait for first few rows of the next macroblock to +// be decoded. Hence, deblocking is lagging behind by 4 or 8 pixels (depending +// on strength). +// With two threads, the vertical positions of the rows being decoded are: +// Decode: [ 0..15][16..31][32..47][48..63][64..79][... +// Deblock: [ 0..11][12..27][28..43][44..59][... +// If we use two threads and two caches of 16 pixels, the sequence would be: +// Decode: [ 0..15][16..31][ 0..15!!][16..31][ 0..15][... +// Deblock: [ 0..11][12..27!!][-4..11][12..27][... +// The problem occurs during row [12..15!!] that both the decoding and +// deblocking threads are writing simultaneously. +// With 3 cache lines, one get a safe write pattern: +// Decode: [ 0..15][16..31][32..47][ 0..15][16..31][32..47][0.. +// Deblock: [ 0..11][12..27][28..43][-4..11][12..27][28... +// Note that multi-threaded output _without_ deblocking can make use of two +// cache lines of 16 pixels only, since there's no lagging behind. The decoding +// and output process have non-concurrent writing: +// Decode: [ 0..15][16..31][ 0..15][16..31][... +// io->put: [ 0..15][16..31][ 0..15][... + +#define MT_CACHE_LINES 3 +#define ST_CACHE_LINES 1 // 1 cache row only for single-threaded case + +// Initialize multi/single-thread worker +static int InitThreadContext(VP8Decoder* const dec) { + dec->cache_id_ = 0; + if (dec->mt_method_ > 0) { + WebPWorker* const worker = &dec->worker_; + if (!WebPGetWorkerInterface()->Reset(worker)) { + return VP8SetError(dec, VP8_STATUS_OUT_OF_MEMORY, + "thread initialization failed."); + } + worker->data1 = dec; + worker->data2 = (void*)&dec->thread_ctx_.io_; + worker->hook = (WebPWorkerHook)FinishRow; + dec->num_caches_ = + (dec->filter_type_ > 0) ? MT_CACHE_LINES : MT_CACHE_LINES - 1; + } else { + dec->num_caches_ = ST_CACHE_LINES; + } + return 1; +} + +int VP8GetThreadMethod(const WebPDecoderOptions* const options, + const WebPHeaderStructure* const headers, + int width, int height) { + if (options == NULL || options->use_threads == 0) { + return 0; + } + (void)headers; + (void)width; + (void)height; + assert(headers == NULL || !headers->is_lossless); +#if defined(WEBP_USE_THREAD) + if (width < MIN_WIDTH_FOR_THREADS) return 0; + // TODO(skal): tune the heuristic further +#if 0 + if (height < 2 * width) return 2; +#endif + return 2; +#else // !WEBP_USE_THREAD + return 0; +#endif +} + +#undef MT_CACHE_LINES +#undef ST_CACHE_LINES + +//------------------------------------------------------------------------------ +// Memory setup + +static int AllocateMemory(VP8Decoder* const dec) { + const int num_caches = dec->num_caches_; + const int mb_w = dec->mb_w_; + // Note: we use 'size_t' when there's no overflow risk, uint64_t otherwise. + const size_t intra_pred_mode_size = 4 * mb_w * sizeof(uint8_t); + const size_t top_size = sizeof(VP8TopSamples) * mb_w; + const size_t mb_info_size = (mb_w + 1) * sizeof(VP8MB); + const size_t f_info_size = + (dec->filter_type_ > 0) ? + mb_w * (dec->mt_method_ > 0 ? 2 : 1) * sizeof(VP8FInfo) + : 0; + const size_t yuv_size = YUV_SIZE * sizeof(*dec->yuv_b_); + const size_t mb_data_size = + (dec->mt_method_ == 2 ? 2 : 1) * mb_w * sizeof(*dec->mb_data_); + const size_t cache_height = (16 * num_caches + + kFilterExtraRows[dec->filter_type_]) * 3 / 2; + const size_t cache_size = top_size * cache_height; + // alpha_size is the only one that scales as width x height. + const uint64_t alpha_size = (dec->alpha_data_ != NULL) ? + (uint64_t)dec->pic_hdr_.width_ * dec->pic_hdr_.height_ : 0ULL; + const uint64_t needed = (uint64_t)intra_pred_mode_size + + top_size + mb_info_size + f_info_size + + yuv_size + mb_data_size + + cache_size + alpha_size + WEBP_ALIGN_CST; + uint8_t* mem; + + if (needed != (size_t)needed) return 0; // check for overflow + if (needed > dec->mem_size_) { + WebPSafeFree(dec->mem_); + dec->mem_size_ = 0; + dec->mem_ = WebPSafeMalloc(needed, sizeof(uint8_t)); + if (dec->mem_ == NULL) { + return VP8SetError(dec, VP8_STATUS_OUT_OF_MEMORY, + "no memory during frame initialization."); + } + // down-cast is ok, thanks to WebPSafeMalloc() above. + dec->mem_size_ = (size_t)needed; + } + + mem = (uint8_t*)dec->mem_; + dec->intra_t_ = mem; + mem += intra_pred_mode_size; + + dec->yuv_t_ = (VP8TopSamples*)mem; + mem += top_size; + + dec->mb_info_ = ((VP8MB*)mem) + 1; + mem += mb_info_size; + + dec->f_info_ = f_info_size ? (VP8FInfo*)mem : NULL; + mem += f_info_size; + dec->thread_ctx_.id_ = 0; + dec->thread_ctx_.f_info_ = dec->f_info_; + if (dec->mt_method_ > 0) { + // secondary cache line. The deblocking process need to make use of the + // filtering strength from previous macroblock row, while the new ones + // are being decoded in parallel. We'll just swap the pointers. + dec->thread_ctx_.f_info_ += mb_w; + } + + mem = (uint8_t*)WEBP_ALIGN(mem); + assert((yuv_size & WEBP_ALIGN_CST) == 0); + dec->yuv_b_ = mem; + mem += yuv_size; + + dec->mb_data_ = (VP8MBData*)mem; + dec->thread_ctx_.mb_data_ = (VP8MBData*)mem; + if (dec->mt_method_ == 2) { + dec->thread_ctx_.mb_data_ += mb_w; + } + mem += mb_data_size; + + dec->cache_y_stride_ = 16 * mb_w; + dec->cache_uv_stride_ = 8 * mb_w; + { + const int extra_rows = kFilterExtraRows[dec->filter_type_]; + const int extra_y = extra_rows * dec->cache_y_stride_; + const int extra_uv = (extra_rows / 2) * dec->cache_uv_stride_; + dec->cache_y_ = mem + extra_y; + dec->cache_u_ = dec->cache_y_ + + 16 * num_caches * dec->cache_y_stride_ + extra_uv; + dec->cache_v_ = dec->cache_u_ + + 8 * num_caches * dec->cache_uv_stride_ + extra_uv; + dec->cache_id_ = 0; + } + mem += cache_size; + + // alpha plane + dec->alpha_plane_ = alpha_size ? mem : NULL; + mem += alpha_size; + assert(mem <= (uint8_t*)dec->mem_ + dec->mem_size_); + + // note: left/top-info is initialized once for all. + memset(dec->mb_info_ - 1, 0, mb_info_size); + VP8InitScanline(dec); // initialize left too. + + // initialize top + memset(dec->intra_t_, B_DC_PRED, intra_pred_mode_size); + + return 1; +} + +static void InitIo(VP8Decoder* const dec, VP8Io* io) { + // prepare 'io' + io->mb_y = 0; + io->y = dec->cache_y_; + io->u = dec->cache_u_; + io->v = dec->cache_v_; + io->y_stride = dec->cache_y_stride_; + io->uv_stride = dec->cache_uv_stride_; + io->a = NULL; +} + +int VP8InitFrame(VP8Decoder* const dec, VP8Io* const io) { + if (!InitThreadContext(dec)) return 0; // call first. Sets dec->num_caches_. + if (!AllocateMemory(dec)) return 0; + InitIo(dec, io); + VP8DspInit(); // Init critical function pointers and look-up tables. + return 1; +} + +//------------------------------------------------------------------------------ diff --git a/thirdparty/libwebp/src/dec/idec_dec.c b/thirdparty/libwebp/src/dec/idec_dec.c new file mode 100644 index 0000000000..a371ed7500 --- /dev/null +++ b/thirdparty/libwebp/src/dec/idec_dec.c @@ -0,0 +1,892 @@ +// Copyright 2011 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. +// ----------------------------------------------------------------------------- +// +// Incremental decoding +// +// Author: somnath@google.com (Somnath Banerjee) + +#include <assert.h> +#include <string.h> +#include <stdlib.h> + +#include "src/dec/alphai_dec.h" +#include "src/dec/webpi_dec.h" +#include "src/dec/vp8i_dec.h" +#include "src/utils/utils.h" + +// In append mode, buffer allocations increase as multiples of this value. +// Needs to be a power of 2. +#define CHUNK_SIZE 4096 +#define MAX_MB_SIZE 4096 + +//------------------------------------------------------------------------------ +// Data structures for memory and states + +// Decoding states. State normally flows as: +// WEBP_HEADER->VP8_HEADER->VP8_PARTS0->VP8_DATA->DONE for a lossy image, and +// WEBP_HEADER->VP8L_HEADER->VP8L_DATA->DONE for a lossless image. +// If there is any error the decoder goes into state ERROR. +typedef enum { + STATE_WEBP_HEADER, // All the data before that of the VP8/VP8L chunk. + STATE_VP8_HEADER, // The VP8 Frame header (within the VP8 chunk). + STATE_VP8_PARTS0, + STATE_VP8_DATA, + STATE_VP8L_HEADER, + STATE_VP8L_DATA, + STATE_DONE, + STATE_ERROR +} DecState; + +// Operating state for the MemBuffer +typedef enum { + MEM_MODE_NONE = 0, + MEM_MODE_APPEND, + MEM_MODE_MAP +} MemBufferMode; + +// storage for partition #0 and partial data (in a rolling fashion) +typedef struct { + MemBufferMode mode_; // Operation mode + size_t start_; // start location of the data to be decoded + size_t end_; // end location + size_t buf_size_; // size of the allocated buffer + uint8_t* buf_; // We don't own this buffer in case WebPIUpdate() + + size_t part0_size_; // size of partition #0 + const uint8_t* part0_buf_; // buffer to store partition #0 +} MemBuffer; + +struct WebPIDecoder { + DecState state_; // current decoding state + WebPDecParams params_; // Params to store output info + int is_lossless_; // for down-casting 'dec_'. + void* dec_; // either a VP8Decoder or a VP8LDecoder instance + VP8Io io_; + + MemBuffer mem_; // input memory buffer. + WebPDecBuffer output_; // output buffer (when no external one is supplied, + // or if the external one has slow-memory) + WebPDecBuffer* final_output_; // Slow-memory output to copy to eventually. + size_t chunk_size_; // Compressed VP8/VP8L size extracted from Header. + + int last_mb_y_; // last row reached for intra-mode decoding +}; + +// MB context to restore in case VP8DecodeMB() fails +typedef struct { + VP8MB left_; + VP8MB info_; + VP8BitReader token_br_; +} MBContext; + +//------------------------------------------------------------------------------ +// MemBuffer: incoming data handling + +static WEBP_INLINE size_t MemDataSize(const MemBuffer* mem) { + return (mem->end_ - mem->start_); +} + +// Check if we need to preserve the compressed alpha data, as it may not have +// been decoded yet. +static int NeedCompressedAlpha(const WebPIDecoder* const idec) { + if (idec->state_ == STATE_WEBP_HEADER) { + // We haven't parsed the headers yet, so we don't know whether the image is + // lossy or lossless. This also means that we haven't parsed the ALPH chunk. + return 0; + } + if (idec->is_lossless_) { + return 0; // ALPH chunk is not present for lossless images. + } else { + const VP8Decoder* const dec = (VP8Decoder*)idec->dec_; + assert(dec != NULL); // Must be true as idec->state_ != STATE_WEBP_HEADER. + return (dec->alpha_data_ != NULL) && !dec->is_alpha_decoded_; + } +} + +static void DoRemap(WebPIDecoder* const idec, ptrdiff_t offset) { + MemBuffer* const mem = &idec->mem_; + const uint8_t* const new_base = mem->buf_ + mem->start_; + // note: for VP8, setting up idec->io_ is only really needed at the beginning + // of the decoding, till partition #0 is complete. + idec->io_.data = new_base; + idec->io_.data_size = MemDataSize(mem); + + if (idec->dec_ != NULL) { + if (!idec->is_lossless_) { + VP8Decoder* const dec = (VP8Decoder*)idec->dec_; + const uint32_t last_part = dec->num_parts_minus_one_; + if (offset != 0) { + uint32_t p; + for (p = 0; p <= last_part; ++p) { + VP8RemapBitReader(dec->parts_ + p, offset); + } + // Remap partition #0 data pointer to new offset, but only in MAP + // mode (in APPEND mode, partition #0 is copied into a fixed memory). + if (mem->mode_ == MEM_MODE_MAP) { + VP8RemapBitReader(&dec->br_, offset); + } + } + { + const uint8_t* const last_start = dec->parts_[last_part].buf_; + VP8BitReaderSetBuffer(&dec->parts_[last_part], last_start, + mem->buf_ + mem->end_ - last_start); + } + if (NeedCompressedAlpha(idec)) { + ALPHDecoder* const alph_dec = dec->alph_dec_; + dec->alpha_data_ += offset; + if (alph_dec != NULL) { + if (alph_dec->method_ == ALPHA_LOSSLESS_COMPRESSION) { + VP8LDecoder* const alph_vp8l_dec = alph_dec->vp8l_dec_; + assert(alph_vp8l_dec != NULL); + assert(dec->alpha_data_size_ >= ALPHA_HEADER_LEN); + VP8LBitReaderSetBuffer(&alph_vp8l_dec->br_, + dec->alpha_data_ + ALPHA_HEADER_LEN, + dec->alpha_data_size_ - ALPHA_HEADER_LEN); + } else { // alph_dec->method_ == ALPHA_NO_COMPRESSION + // Nothing special to do in this case. + } + } + } + } else { // Resize lossless bitreader + VP8LDecoder* const dec = (VP8LDecoder*)idec->dec_; + VP8LBitReaderSetBuffer(&dec->br_, new_base, MemDataSize(mem)); + } + } +} + +// Appends data to the end of MemBuffer->buf_. It expands the allocated memory +// size if required and also updates VP8BitReader's if new memory is allocated. +static int AppendToMemBuffer(WebPIDecoder* const idec, + const uint8_t* const data, size_t data_size) { + VP8Decoder* const dec = (VP8Decoder*)idec->dec_; + MemBuffer* const mem = &idec->mem_; + const int need_compressed_alpha = NeedCompressedAlpha(idec); + const uint8_t* const old_start = mem->buf_ + mem->start_; + const uint8_t* const old_base = + need_compressed_alpha ? dec->alpha_data_ : old_start; + assert(mem->mode_ == MEM_MODE_APPEND); + if (data_size > MAX_CHUNK_PAYLOAD) { + // security safeguard: trying to allocate more than what the format + // allows for a chunk should be considered a smoke smell. + return 0; + } + + if (mem->end_ + data_size > mem->buf_size_) { // Need some free memory + const size_t new_mem_start = old_start - old_base; + const size_t current_size = MemDataSize(mem) + new_mem_start; + const uint64_t new_size = (uint64_t)current_size + data_size; + const uint64_t extra_size = (new_size + CHUNK_SIZE - 1) & ~(CHUNK_SIZE - 1); + uint8_t* const new_buf = + (uint8_t*)WebPSafeMalloc(extra_size, sizeof(*new_buf)); + if (new_buf == NULL) return 0; + memcpy(new_buf, old_base, current_size); + WebPSafeFree(mem->buf_); + mem->buf_ = new_buf; + mem->buf_size_ = (size_t)extra_size; + mem->start_ = new_mem_start; + mem->end_ = current_size; + } + + memcpy(mem->buf_ + mem->end_, data, data_size); + mem->end_ += data_size; + assert(mem->end_ <= mem->buf_size_); + + DoRemap(idec, mem->buf_ + mem->start_ - old_start); + return 1; +} + +static int RemapMemBuffer(WebPIDecoder* const idec, + const uint8_t* const data, size_t data_size) { + MemBuffer* const mem = &idec->mem_; + const uint8_t* const old_buf = mem->buf_; + const uint8_t* const old_start = old_buf + mem->start_; + assert(mem->mode_ == MEM_MODE_MAP); + + if (data_size < mem->buf_size_) return 0; // can't remap to a shorter buffer! + + mem->buf_ = (uint8_t*)data; + mem->end_ = mem->buf_size_ = data_size; + + DoRemap(idec, mem->buf_ + mem->start_ - old_start); + return 1; +} + +static void InitMemBuffer(MemBuffer* const mem) { + mem->mode_ = MEM_MODE_NONE; + mem->buf_ = NULL; + mem->buf_size_ = 0; + mem->part0_buf_ = NULL; + mem->part0_size_ = 0; +} + +static void ClearMemBuffer(MemBuffer* const mem) { + assert(mem); + if (mem->mode_ == MEM_MODE_APPEND) { + WebPSafeFree(mem->buf_); + WebPSafeFree((void*)mem->part0_buf_); + } +} + +static int CheckMemBufferMode(MemBuffer* const mem, MemBufferMode expected) { + if (mem->mode_ == MEM_MODE_NONE) { + mem->mode_ = expected; // switch to the expected mode + } else if (mem->mode_ != expected) { + return 0; // we mixed the modes => error + } + assert(mem->mode_ == expected); // mode is ok + return 1; +} + +// To be called last. +static VP8StatusCode FinishDecoding(WebPIDecoder* const idec) { + const WebPDecoderOptions* const options = idec->params_.options; + WebPDecBuffer* const output = idec->params_.output; + + idec->state_ = STATE_DONE; + if (options != NULL && options->flip) { + const VP8StatusCode status = WebPFlipBuffer(output); + if (status != VP8_STATUS_OK) return status; + } + if (idec->final_output_ != NULL) { + WebPCopyDecBufferPixels(output, idec->final_output_); // do the slow-copy + WebPFreeDecBuffer(&idec->output_); + *output = *idec->final_output_; + idec->final_output_ = NULL; + } + return VP8_STATUS_OK; +} + +//------------------------------------------------------------------------------ +// Macroblock-decoding contexts + +static void SaveContext(const VP8Decoder* dec, const VP8BitReader* token_br, + MBContext* const context) { + context->left_ = dec->mb_info_[-1]; + context->info_ = dec->mb_info_[dec->mb_x_]; + context->token_br_ = *token_br; +} + +static void RestoreContext(const MBContext* context, VP8Decoder* const dec, + VP8BitReader* const token_br) { + dec->mb_info_[-1] = context->left_; + dec->mb_info_[dec->mb_x_] = context->info_; + *token_br = context->token_br_; +} + +//------------------------------------------------------------------------------ + +static VP8StatusCode IDecError(WebPIDecoder* const idec, VP8StatusCode error) { + if (idec->state_ == STATE_VP8_DATA) { + VP8Io* const io = &idec->io_; + if (io->teardown != NULL) { + io->teardown(io); + } + } + idec->state_ = STATE_ERROR; + return error; +} + +static void ChangeState(WebPIDecoder* const idec, DecState new_state, + size_t consumed_bytes) { + MemBuffer* const mem = &idec->mem_; + idec->state_ = new_state; + mem->start_ += consumed_bytes; + assert(mem->start_ <= mem->end_); + idec->io_.data = mem->buf_ + mem->start_; + idec->io_.data_size = MemDataSize(mem); +} + +// Headers +static VP8StatusCode DecodeWebPHeaders(WebPIDecoder* const idec) { + MemBuffer* const mem = &idec->mem_; + const uint8_t* data = mem->buf_ + mem->start_; + size_t curr_size = MemDataSize(mem); + VP8StatusCode status; + WebPHeaderStructure headers; + + headers.data = data; + headers.data_size = curr_size; + headers.have_all_data = 0; + status = WebPParseHeaders(&headers); + if (status == VP8_STATUS_NOT_ENOUGH_DATA) { + return VP8_STATUS_SUSPENDED; // We haven't found a VP8 chunk yet. + } else if (status != VP8_STATUS_OK) { + return IDecError(idec, status); + } + + idec->chunk_size_ = headers.compressed_size; + idec->is_lossless_ = headers.is_lossless; + if (!idec->is_lossless_) { + VP8Decoder* const dec = VP8New(); + if (dec == NULL) { + return VP8_STATUS_OUT_OF_MEMORY; + } + idec->dec_ = dec; + dec->alpha_data_ = headers.alpha_data; + dec->alpha_data_size_ = headers.alpha_data_size; + ChangeState(idec, STATE_VP8_HEADER, headers.offset); + } else { + VP8LDecoder* const dec = VP8LNew(); + if (dec == NULL) { + return VP8_STATUS_OUT_OF_MEMORY; + } + idec->dec_ = dec; + ChangeState(idec, STATE_VP8L_HEADER, headers.offset); + } + return VP8_STATUS_OK; +} + +static VP8StatusCode DecodeVP8FrameHeader(WebPIDecoder* const idec) { + const uint8_t* data = idec->mem_.buf_ + idec->mem_.start_; + const size_t curr_size = MemDataSize(&idec->mem_); + int width, height; + uint32_t bits; + + if (curr_size < VP8_FRAME_HEADER_SIZE) { + // Not enough data bytes to extract VP8 Frame Header. + return VP8_STATUS_SUSPENDED; + } + if (!VP8GetInfo(data, curr_size, idec->chunk_size_, &width, &height)) { + return IDecError(idec, VP8_STATUS_BITSTREAM_ERROR); + } + + bits = data[0] | (data[1] << 8) | (data[2] << 16); + idec->mem_.part0_size_ = (bits >> 5) + VP8_FRAME_HEADER_SIZE; + + idec->io_.data = data; + idec->io_.data_size = curr_size; + idec->state_ = STATE_VP8_PARTS0; + return VP8_STATUS_OK; +} + +// Partition #0 +static VP8StatusCode CopyParts0Data(WebPIDecoder* const idec) { + VP8Decoder* const dec = (VP8Decoder*)idec->dec_; + VP8BitReader* const br = &dec->br_; + const size_t part_size = br->buf_end_ - br->buf_; + MemBuffer* const mem = &idec->mem_; + assert(!idec->is_lossless_); + assert(mem->part0_buf_ == NULL); + // the following is a format limitation, no need for runtime check: + assert(part_size <= mem->part0_size_); + if (part_size == 0) { // can't have zero-size partition #0 + return VP8_STATUS_BITSTREAM_ERROR; + } + if (mem->mode_ == MEM_MODE_APPEND) { + // We copy and grab ownership of the partition #0 data. + uint8_t* const part0_buf = (uint8_t*)WebPSafeMalloc(1ULL, part_size); + if (part0_buf == NULL) { + return VP8_STATUS_OUT_OF_MEMORY; + } + memcpy(part0_buf, br->buf_, part_size); + mem->part0_buf_ = part0_buf; + VP8BitReaderSetBuffer(br, part0_buf, part_size); + } else { + // Else: just keep pointers to the partition #0's data in dec_->br_. + } + mem->start_ += part_size; + return VP8_STATUS_OK; +} + +static VP8StatusCode DecodePartition0(WebPIDecoder* const idec) { + VP8Decoder* const dec = (VP8Decoder*)idec->dec_; + VP8Io* const io = &idec->io_; + const WebPDecParams* const params = &idec->params_; + WebPDecBuffer* const output = params->output; + + // Wait till we have enough data for the whole partition #0 + if (MemDataSize(&idec->mem_) < idec->mem_.part0_size_) { + return VP8_STATUS_SUSPENDED; + } + + if (!VP8GetHeaders(dec, io)) { + const VP8StatusCode status = dec->status_; + if (status == VP8_STATUS_SUSPENDED || + status == VP8_STATUS_NOT_ENOUGH_DATA) { + // treating NOT_ENOUGH_DATA as SUSPENDED state + return VP8_STATUS_SUSPENDED; + } + return IDecError(idec, status); + } + + // Allocate/Verify output buffer now + dec->status_ = WebPAllocateDecBuffer(io->width, io->height, params->options, + output); + if (dec->status_ != VP8_STATUS_OK) { + return IDecError(idec, dec->status_); + } + // This change must be done before calling VP8InitFrame() + dec->mt_method_ = VP8GetThreadMethod(params->options, NULL, + io->width, io->height); + VP8InitDithering(params->options, dec); + + dec->status_ = CopyParts0Data(idec); + if (dec->status_ != VP8_STATUS_OK) { + return IDecError(idec, dec->status_); + } + + // Finish setting up the decoding parameters. Will call io->setup(). + if (VP8EnterCritical(dec, io) != VP8_STATUS_OK) { + return IDecError(idec, dec->status_); + } + + // Note: past this point, teardown() must always be called + // in case of error. + idec->state_ = STATE_VP8_DATA; + // Allocate memory and prepare everything. + if (!VP8InitFrame(dec, io)) { + return IDecError(idec, dec->status_); + } + return VP8_STATUS_OK; +} + +// Remaining partitions +static VP8StatusCode DecodeRemaining(WebPIDecoder* const idec) { + VP8Decoder* const dec = (VP8Decoder*)idec->dec_; + VP8Io* const io = &idec->io_; + + assert(dec->ready_); + for (; dec->mb_y_ < dec->mb_h_; ++dec->mb_y_) { + if (idec->last_mb_y_ != dec->mb_y_) { + if (!VP8ParseIntraModeRow(&dec->br_, dec)) { + // note: normally, error shouldn't occur since we already have the whole + // partition0 available here in DecodeRemaining(). Reaching EOF while + // reading intra modes really means a BITSTREAM_ERROR. + return IDecError(idec, VP8_STATUS_BITSTREAM_ERROR); + } + idec->last_mb_y_ = dec->mb_y_; + } + for (; dec->mb_x_ < dec->mb_w_; ++dec->mb_x_) { + VP8BitReader* const token_br = + &dec->parts_[dec->mb_y_ & dec->num_parts_minus_one_]; + MBContext context; + SaveContext(dec, token_br, &context); + if (!VP8DecodeMB(dec, token_br)) { + // We shouldn't fail when MAX_MB data was available + if (dec->num_parts_minus_one_ == 0 && + MemDataSize(&idec->mem_) > MAX_MB_SIZE) { + return IDecError(idec, VP8_STATUS_BITSTREAM_ERROR); + } + RestoreContext(&context, dec, token_br); + return VP8_STATUS_SUSPENDED; + } + // Release buffer only if there is only one partition + if (dec->num_parts_minus_one_ == 0) { + idec->mem_.start_ = token_br->buf_ - idec->mem_.buf_; + assert(idec->mem_.start_ <= idec->mem_.end_); + } + } + VP8InitScanline(dec); // Prepare for next scanline + + // Reconstruct, filter and emit the row. + if (!VP8ProcessRow(dec, io)) { + return IDecError(idec, VP8_STATUS_USER_ABORT); + } + } + // Synchronize the thread and check for errors. + if (!VP8ExitCritical(dec, io)) { + return IDecError(idec, VP8_STATUS_USER_ABORT); + } + dec->ready_ = 0; + return FinishDecoding(idec); +} + +static VP8StatusCode ErrorStatusLossless(WebPIDecoder* const idec, + VP8StatusCode status) { + if (status == VP8_STATUS_SUSPENDED || status == VP8_STATUS_NOT_ENOUGH_DATA) { + return VP8_STATUS_SUSPENDED; + } + return IDecError(idec, status); +} + +static VP8StatusCode DecodeVP8LHeader(WebPIDecoder* const idec) { + VP8Io* const io = &idec->io_; + VP8LDecoder* const dec = (VP8LDecoder*)idec->dec_; + const WebPDecParams* const params = &idec->params_; + WebPDecBuffer* const output = params->output; + size_t curr_size = MemDataSize(&idec->mem_); + assert(idec->is_lossless_); + + // Wait until there's enough data for decoding header. + if (curr_size < (idec->chunk_size_ >> 3)) { + dec->status_ = VP8_STATUS_SUSPENDED; + return ErrorStatusLossless(idec, dec->status_); + } + + if (!VP8LDecodeHeader(dec, io)) { + if (dec->status_ == VP8_STATUS_BITSTREAM_ERROR && + curr_size < idec->chunk_size_) { + dec->status_ = VP8_STATUS_SUSPENDED; + } + return ErrorStatusLossless(idec, dec->status_); + } + // Allocate/verify output buffer now. + dec->status_ = WebPAllocateDecBuffer(io->width, io->height, params->options, + output); + if (dec->status_ != VP8_STATUS_OK) { + return IDecError(idec, dec->status_); + } + + idec->state_ = STATE_VP8L_DATA; + return VP8_STATUS_OK; +} + +static VP8StatusCode DecodeVP8LData(WebPIDecoder* const idec) { + VP8LDecoder* const dec = (VP8LDecoder*)idec->dec_; + const size_t curr_size = MemDataSize(&idec->mem_); + assert(idec->is_lossless_); + + // Switch to incremental decoding if we don't have all the bytes available. + dec->incremental_ = (curr_size < idec->chunk_size_); + + if (!VP8LDecodeImage(dec)) { + return ErrorStatusLossless(idec, dec->status_); + } + assert(dec->status_ == VP8_STATUS_OK || dec->status_ == VP8_STATUS_SUSPENDED); + return (dec->status_ == VP8_STATUS_SUSPENDED) ? dec->status_ + : FinishDecoding(idec); +} + + // Main decoding loop +static VP8StatusCode IDecode(WebPIDecoder* idec) { + VP8StatusCode status = VP8_STATUS_SUSPENDED; + + if (idec->state_ == STATE_WEBP_HEADER) { + status = DecodeWebPHeaders(idec); + } else { + if (idec->dec_ == NULL) { + return VP8_STATUS_SUSPENDED; // can't continue if we have no decoder. + } + } + if (idec->state_ == STATE_VP8_HEADER) { + status = DecodeVP8FrameHeader(idec); + } + if (idec->state_ == STATE_VP8_PARTS0) { + status = DecodePartition0(idec); + } + if (idec->state_ == STATE_VP8_DATA) { + status = DecodeRemaining(idec); + } + if (idec->state_ == STATE_VP8L_HEADER) { + status = DecodeVP8LHeader(idec); + } + if (idec->state_ == STATE_VP8L_DATA) { + status = DecodeVP8LData(idec); + } + return status; +} + +//------------------------------------------------------------------------------ +// Internal constructor + +static WebPIDecoder* NewDecoder(WebPDecBuffer* const output_buffer, + const WebPBitstreamFeatures* const features) { + WebPIDecoder* idec = (WebPIDecoder*)WebPSafeCalloc(1ULL, sizeof(*idec)); + if (idec == NULL) { + return NULL; + } + + idec->state_ = STATE_WEBP_HEADER; + idec->chunk_size_ = 0; + + idec->last_mb_y_ = -1; + + InitMemBuffer(&idec->mem_); + WebPInitDecBuffer(&idec->output_); + VP8InitIo(&idec->io_); + + WebPResetDecParams(&idec->params_); + if (output_buffer == NULL || WebPAvoidSlowMemory(output_buffer, features)) { + idec->params_.output = &idec->output_; + idec->final_output_ = output_buffer; + if (output_buffer != NULL) { + idec->params_.output->colorspace = output_buffer->colorspace; + } + } else { + idec->params_.output = output_buffer; + idec->final_output_ = NULL; + } + WebPInitCustomIo(&idec->params_, &idec->io_); // Plug the I/O functions. + + return idec; +} + +//------------------------------------------------------------------------------ +// Public functions + +WebPIDecoder* WebPINewDecoder(WebPDecBuffer* output_buffer) { + return NewDecoder(output_buffer, NULL); +} + +WebPIDecoder* WebPIDecode(const uint8_t* data, size_t data_size, + WebPDecoderConfig* config) { + WebPIDecoder* idec; + WebPBitstreamFeatures tmp_features; + WebPBitstreamFeatures* const features = + (config == NULL) ? &tmp_features : &config->input; + memset(&tmp_features, 0, sizeof(tmp_features)); + + // Parse the bitstream's features, if requested: + if (data != NULL && data_size > 0) { + if (WebPGetFeatures(data, data_size, features) != VP8_STATUS_OK) { + return NULL; + } + } + + // Create an instance of the incremental decoder + idec = (config != NULL) ? NewDecoder(&config->output, features) + : NewDecoder(NULL, features); + if (idec == NULL) { + return NULL; + } + // Finish initialization + if (config != NULL) { + idec->params_.options = &config->options; + } + return idec; +} + +void WebPIDelete(WebPIDecoder* idec) { + if (idec == NULL) return; + if (idec->dec_ != NULL) { + if (!idec->is_lossless_) { + if (idec->state_ == STATE_VP8_DATA) { + // Synchronize the thread, clean-up and check for errors. + VP8ExitCritical((VP8Decoder*)idec->dec_, &idec->io_); + } + VP8Delete((VP8Decoder*)idec->dec_); + } else { + VP8LDelete((VP8LDecoder*)idec->dec_); + } + } + ClearMemBuffer(&idec->mem_); + WebPFreeDecBuffer(&idec->output_); + WebPSafeFree(idec); +} + +//------------------------------------------------------------------------------ +// Wrapper toward WebPINewDecoder + +WebPIDecoder* WebPINewRGB(WEBP_CSP_MODE csp, uint8_t* output_buffer, + size_t output_buffer_size, int output_stride) { + const int is_external_memory = (output_buffer != NULL) ? 1 : 0; + WebPIDecoder* idec; + + if (csp >= MODE_YUV) return NULL; + if (is_external_memory == 0) { // Overwrite parameters to sane values. + output_buffer_size = 0; + output_stride = 0; + } else { // A buffer was passed. Validate the other params. + if (output_stride == 0 || output_buffer_size == 0) { + return NULL; // invalid parameter. + } + } + idec = WebPINewDecoder(NULL); + if (idec == NULL) return NULL; + idec->output_.colorspace = csp; + idec->output_.is_external_memory = is_external_memory; + idec->output_.u.RGBA.rgba = output_buffer; + idec->output_.u.RGBA.stride = output_stride; + idec->output_.u.RGBA.size = output_buffer_size; + return idec; +} + +WebPIDecoder* WebPINewYUVA(uint8_t* luma, size_t luma_size, int luma_stride, + uint8_t* u, size_t u_size, int u_stride, + uint8_t* v, size_t v_size, int v_stride, + uint8_t* a, size_t a_size, int a_stride) { + const int is_external_memory = (luma != NULL) ? 1 : 0; + WebPIDecoder* idec; + WEBP_CSP_MODE colorspace; + + if (is_external_memory == 0) { // Overwrite parameters to sane values. + luma_size = u_size = v_size = a_size = 0; + luma_stride = u_stride = v_stride = a_stride = 0; + u = v = a = NULL; + colorspace = MODE_YUVA; + } else { // A luma buffer was passed. Validate the other parameters. + if (u == NULL || v == NULL) return NULL; + if (luma_size == 0 || u_size == 0 || v_size == 0) return NULL; + if (luma_stride == 0 || u_stride == 0 || v_stride == 0) return NULL; + if (a != NULL) { + if (a_size == 0 || a_stride == 0) return NULL; + } + colorspace = (a == NULL) ? MODE_YUV : MODE_YUVA; + } + + idec = WebPINewDecoder(NULL); + if (idec == NULL) return NULL; + + idec->output_.colorspace = colorspace; + idec->output_.is_external_memory = is_external_memory; + idec->output_.u.YUVA.y = luma; + idec->output_.u.YUVA.y_stride = luma_stride; + idec->output_.u.YUVA.y_size = luma_size; + idec->output_.u.YUVA.u = u; + idec->output_.u.YUVA.u_stride = u_stride; + idec->output_.u.YUVA.u_size = u_size; + idec->output_.u.YUVA.v = v; + idec->output_.u.YUVA.v_stride = v_stride; + idec->output_.u.YUVA.v_size = v_size; + idec->output_.u.YUVA.a = a; + idec->output_.u.YUVA.a_stride = a_stride; + idec->output_.u.YUVA.a_size = a_size; + return idec; +} + +WebPIDecoder* WebPINewYUV(uint8_t* luma, size_t luma_size, int luma_stride, + uint8_t* u, size_t u_size, int u_stride, + uint8_t* v, size_t v_size, int v_stride) { + return WebPINewYUVA(luma, luma_size, luma_stride, + u, u_size, u_stride, + v, v_size, v_stride, + NULL, 0, 0); +} + +//------------------------------------------------------------------------------ + +static VP8StatusCode IDecCheckStatus(const WebPIDecoder* const idec) { + assert(idec); + if (idec->state_ == STATE_ERROR) { + return VP8_STATUS_BITSTREAM_ERROR; + } + if (idec->state_ == STATE_DONE) { + return VP8_STATUS_OK; + } + return VP8_STATUS_SUSPENDED; +} + +VP8StatusCode WebPIAppend(WebPIDecoder* idec, + const uint8_t* data, size_t data_size) { + VP8StatusCode status; + if (idec == NULL || data == NULL) { + return VP8_STATUS_INVALID_PARAM; + } + status = IDecCheckStatus(idec); + if (status != VP8_STATUS_SUSPENDED) { + return status; + } + // Check mixed calls between RemapMemBuffer and AppendToMemBuffer. + if (!CheckMemBufferMode(&idec->mem_, MEM_MODE_APPEND)) { + return VP8_STATUS_INVALID_PARAM; + } + // Append data to memory buffer + if (!AppendToMemBuffer(idec, data, data_size)) { + return VP8_STATUS_OUT_OF_MEMORY; + } + return IDecode(idec); +} + +VP8StatusCode WebPIUpdate(WebPIDecoder* idec, + const uint8_t* data, size_t data_size) { + VP8StatusCode status; + if (idec == NULL || data == NULL) { + return VP8_STATUS_INVALID_PARAM; + } + status = IDecCheckStatus(idec); + if (status != VP8_STATUS_SUSPENDED) { + return status; + } + // Check mixed calls between RemapMemBuffer and AppendToMemBuffer. + if (!CheckMemBufferMode(&idec->mem_, MEM_MODE_MAP)) { + return VP8_STATUS_INVALID_PARAM; + } + // Make the memory buffer point to the new buffer + if (!RemapMemBuffer(idec, data, data_size)) { + return VP8_STATUS_INVALID_PARAM; + } + return IDecode(idec); +} + +//------------------------------------------------------------------------------ + +static const WebPDecBuffer* GetOutputBuffer(const WebPIDecoder* const idec) { + if (idec == NULL || idec->dec_ == NULL) { + return NULL; + } + if (idec->state_ <= STATE_VP8_PARTS0) { + return NULL; + } + if (idec->final_output_ != NULL) { + return NULL; // not yet slow-copied + } + return idec->params_.output; +} + +const WebPDecBuffer* WebPIDecodedArea(const WebPIDecoder* idec, + int* left, int* top, + int* width, int* height) { + const WebPDecBuffer* const src = GetOutputBuffer(idec); + if (left != NULL) *left = 0; + if (top != NULL) *top = 0; + if (src != NULL) { + if (width != NULL) *width = src->width; + if (height != NULL) *height = idec->params_.last_y; + } else { + if (width != NULL) *width = 0; + if (height != NULL) *height = 0; + } + return src; +} + +uint8_t* WebPIDecGetRGB(const WebPIDecoder* idec, int* last_y, + int* width, int* height, int* stride) { + const WebPDecBuffer* const src = GetOutputBuffer(idec); + if (src == NULL) return NULL; + if (src->colorspace >= MODE_YUV) { + return NULL; + } + + if (last_y != NULL) *last_y = idec->params_.last_y; + if (width != NULL) *width = src->width; + if (height != NULL) *height = src->height; + if (stride != NULL) *stride = src->u.RGBA.stride; + + return src->u.RGBA.rgba; +} + +uint8_t* WebPIDecGetYUVA(const WebPIDecoder* idec, int* last_y, + uint8_t** u, uint8_t** v, uint8_t** a, + int* width, int* height, + int* stride, int* uv_stride, int* a_stride) { + const WebPDecBuffer* const src = GetOutputBuffer(idec); + if (src == NULL) return NULL; + if (src->colorspace < MODE_YUV) { + return NULL; + } + + if (last_y != NULL) *last_y = idec->params_.last_y; + if (u != NULL) *u = src->u.YUVA.u; + if (v != NULL) *v = src->u.YUVA.v; + if (a != NULL) *a = src->u.YUVA.a; + if (width != NULL) *width = src->width; + if (height != NULL) *height = src->height; + if (stride != NULL) *stride = src->u.YUVA.y_stride; + if (uv_stride != NULL) *uv_stride = src->u.YUVA.u_stride; + if (a_stride != NULL) *a_stride = src->u.YUVA.a_stride; + + return src->u.YUVA.y; +} + +int WebPISetIOHooks(WebPIDecoder* const idec, + VP8IoPutHook put, + VP8IoSetupHook setup, + VP8IoTeardownHook teardown, + void* user_data) { + if (idec == NULL || idec->state_ > STATE_WEBP_HEADER) { + return 0; + } + + idec->io_.put = put; + idec->io_.setup = setup; + idec->io_.teardown = teardown; + idec->io_.opaque = user_data; + + return 1; +} diff --git a/thirdparty/libwebp/src/dec/io_dec.c b/thirdparty/libwebp/src/dec/io_dec.c new file mode 100644 index 0000000000..e603f19c98 --- /dev/null +++ b/thirdparty/libwebp/src/dec/io_dec.c @@ -0,0 +1,649 @@ +// Copyright 2011 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. +// ----------------------------------------------------------------------------- +// +// functions for sample output. +// +// Author: Skal (pascal.massimino@gmail.com) + +#include <assert.h> +#include <stdlib.h> +#include "src/dec/vp8i_dec.h" +#include "src/dec/webpi_dec.h" +#include "src/dsp/dsp.h" +#include "src/dsp/yuv.h" +#include "src/utils/utils.h" + +//------------------------------------------------------------------------------ +// Main YUV<->RGB conversion functions + +static int EmitYUV(const VP8Io* const io, WebPDecParams* const p) { + WebPDecBuffer* output = p->output; + const WebPYUVABuffer* const buf = &output->u.YUVA; + uint8_t* const y_dst = buf->y + io->mb_y * buf->y_stride; + uint8_t* const u_dst = buf->u + (io->mb_y >> 1) * buf->u_stride; + uint8_t* const v_dst = buf->v + (io->mb_y >> 1) * buf->v_stride; + const int mb_w = io->mb_w; + const int mb_h = io->mb_h; + const int uv_w = (mb_w + 1) / 2; + const int uv_h = (mb_h + 1) / 2; + int j; + for (j = 0; j < mb_h; ++j) { + memcpy(y_dst + j * buf->y_stride, io->y + j * io->y_stride, mb_w); + } + for (j = 0; j < uv_h; ++j) { + memcpy(u_dst + j * buf->u_stride, io->u + j * io->uv_stride, uv_w); + memcpy(v_dst + j * buf->v_stride, io->v + j * io->uv_stride, uv_w); + } + return io->mb_h; +} + +// Point-sampling U/V sampler. +static int EmitSampledRGB(const VP8Io* const io, WebPDecParams* const p) { + WebPDecBuffer* const output = p->output; + WebPRGBABuffer* const buf = &output->u.RGBA; + uint8_t* const dst = buf->rgba + io->mb_y * buf->stride; + WebPSamplerProcessPlane(io->y, io->y_stride, + io->u, io->v, io->uv_stride, + dst, buf->stride, io->mb_w, io->mb_h, + WebPSamplers[output->colorspace]); + return io->mb_h; +} + +//------------------------------------------------------------------------------ +// Fancy upsampling + +#ifdef FANCY_UPSAMPLING +static int EmitFancyRGB(const VP8Io* const io, WebPDecParams* const p) { + int num_lines_out = io->mb_h; // a priori guess + const WebPRGBABuffer* const buf = &p->output->u.RGBA; + uint8_t* dst = buf->rgba + io->mb_y * buf->stride; + WebPUpsampleLinePairFunc upsample = WebPUpsamplers[p->output->colorspace]; + const uint8_t* cur_y = io->y; + const uint8_t* cur_u = io->u; + const uint8_t* cur_v = io->v; + const uint8_t* top_u = p->tmp_u; + const uint8_t* top_v = p->tmp_v; + int y = io->mb_y; + const int y_end = io->mb_y + io->mb_h; + const int mb_w = io->mb_w; + const int uv_w = (mb_w + 1) / 2; + + if (y == 0) { + // First line is special cased. We mirror the u/v samples at boundary. + upsample(cur_y, NULL, cur_u, cur_v, cur_u, cur_v, dst, NULL, mb_w); + } else { + // We can finish the left-over line from previous call. + upsample(p->tmp_y, cur_y, top_u, top_v, cur_u, cur_v, + dst - buf->stride, dst, mb_w); + ++num_lines_out; + } + // Loop over each output pairs of row. + for (; y + 2 < y_end; y += 2) { + top_u = cur_u; + top_v = cur_v; + cur_u += io->uv_stride; + cur_v += io->uv_stride; + dst += 2 * buf->stride; + cur_y += 2 * io->y_stride; + upsample(cur_y - io->y_stride, cur_y, + top_u, top_v, cur_u, cur_v, + dst - buf->stride, dst, mb_w); + } + // move to last row + cur_y += io->y_stride; + if (io->crop_top + y_end < io->crop_bottom) { + // Save the unfinished samples for next call (as we're not done yet). + memcpy(p->tmp_y, cur_y, mb_w * sizeof(*p->tmp_y)); + memcpy(p->tmp_u, cur_u, uv_w * sizeof(*p->tmp_u)); + memcpy(p->tmp_v, cur_v, uv_w * sizeof(*p->tmp_v)); + // The fancy upsampler leaves a row unfinished behind + // (except for the very last row) + num_lines_out--; + } else { + // Process the very last row of even-sized picture + if (!(y_end & 1)) { + upsample(cur_y, NULL, cur_u, cur_v, cur_u, cur_v, + dst + buf->stride, NULL, mb_w); + } + } + return num_lines_out; +} + +#endif /* FANCY_UPSAMPLING */ + +//------------------------------------------------------------------------------ + +static void FillAlphaPlane(uint8_t* dst, int w, int h, int stride) { + int j; + for (j = 0; j < h; ++j) { + memset(dst, 0xff, w * sizeof(*dst)); + dst += stride; + } +} + +static int EmitAlphaYUV(const VP8Io* const io, WebPDecParams* const p, + int expected_num_lines_out) { + const uint8_t* alpha = io->a; + const WebPYUVABuffer* const buf = &p->output->u.YUVA; + const int mb_w = io->mb_w; + const int mb_h = io->mb_h; + uint8_t* dst = buf->a + io->mb_y * buf->a_stride; + int j; + (void)expected_num_lines_out; + assert(expected_num_lines_out == mb_h); + if (alpha != NULL) { + for (j = 0; j < mb_h; ++j) { + memcpy(dst, alpha, mb_w * sizeof(*dst)); + alpha += io->width; + dst += buf->a_stride; + } + } else if (buf->a != NULL) { + // the user requested alpha, but there is none, set it to opaque. + FillAlphaPlane(dst, mb_w, mb_h, buf->a_stride); + } + return 0; +} + +static int GetAlphaSourceRow(const VP8Io* const io, + const uint8_t** alpha, int* const num_rows) { + int start_y = io->mb_y; + *num_rows = io->mb_h; + + // Compensate for the 1-line delay of the fancy upscaler. + // This is similar to EmitFancyRGB(). + if (io->fancy_upsampling) { + if (start_y == 0) { + // We don't process the last row yet. It'll be done during the next call. + --*num_rows; + } else { + --start_y; + // Fortunately, *alpha data is persistent, so we can go back + // one row and finish alpha blending, now that the fancy upscaler + // completed the YUV->RGB interpolation. + *alpha -= io->width; + } + if (io->crop_top + io->mb_y + io->mb_h == io->crop_bottom) { + // If it's the very last call, we process all the remaining rows! + *num_rows = io->crop_bottom - io->crop_top - start_y; + } + } + return start_y; +} + +static int EmitAlphaRGB(const VP8Io* const io, WebPDecParams* const p, + int expected_num_lines_out) { + const uint8_t* alpha = io->a; + if (alpha != NULL) { + const int mb_w = io->mb_w; + const WEBP_CSP_MODE colorspace = p->output->colorspace; + const int alpha_first = + (colorspace == MODE_ARGB || colorspace == MODE_Argb); + const WebPRGBABuffer* const buf = &p->output->u.RGBA; + int num_rows; + const int start_y = GetAlphaSourceRow(io, &alpha, &num_rows); + uint8_t* const base_rgba = buf->rgba + start_y * buf->stride; + uint8_t* const dst = base_rgba + (alpha_first ? 0 : 3); + const int has_alpha = WebPDispatchAlpha(alpha, io->width, mb_w, + num_rows, dst, buf->stride); + (void)expected_num_lines_out; + assert(expected_num_lines_out == num_rows); + // has_alpha is true if there's non-trivial alpha to premultiply with. + if (has_alpha && WebPIsPremultipliedMode(colorspace)) { + WebPApplyAlphaMultiply(base_rgba, alpha_first, + mb_w, num_rows, buf->stride); + } + } + return 0; +} + +static int EmitAlphaRGBA4444(const VP8Io* const io, WebPDecParams* const p, + int expected_num_lines_out) { + const uint8_t* alpha = io->a; + if (alpha != NULL) { + const int mb_w = io->mb_w; + const WEBP_CSP_MODE colorspace = p->output->colorspace; + const WebPRGBABuffer* const buf = &p->output->u.RGBA; + int num_rows; + const int start_y = GetAlphaSourceRow(io, &alpha, &num_rows); + uint8_t* const base_rgba = buf->rgba + start_y * buf->stride; +#if (WEBP_SWAP_16BIT_CSP == 1) + uint8_t* alpha_dst = base_rgba; +#else + uint8_t* alpha_dst = base_rgba + 1; +#endif + uint32_t alpha_mask = 0x0f; + int i, j; + for (j = 0; j < num_rows; ++j) { + for (i = 0; i < mb_w; ++i) { + // Fill in the alpha value (converted to 4 bits). + const uint32_t alpha_value = alpha[i] >> 4; + alpha_dst[2 * i] = (alpha_dst[2 * i] & 0xf0) | alpha_value; + alpha_mask &= alpha_value; + } + alpha += io->width; + alpha_dst += buf->stride; + } + (void)expected_num_lines_out; + assert(expected_num_lines_out == num_rows); + if (alpha_mask != 0x0f && WebPIsPremultipliedMode(colorspace)) { + WebPApplyAlphaMultiply4444(base_rgba, mb_w, num_rows, buf->stride); + } + } + return 0; +} + +//------------------------------------------------------------------------------ +// YUV rescaling (no final RGB conversion needed) + +#if !defined(WEBP_REDUCE_SIZE) +static int Rescale(const uint8_t* src, int src_stride, + int new_lines, WebPRescaler* const wrk) { + int num_lines_out = 0; + while (new_lines > 0) { // import new contributions of source rows. + const int lines_in = WebPRescalerImport(wrk, new_lines, src, src_stride); + src += lines_in * src_stride; + new_lines -= lines_in; + num_lines_out += WebPRescalerExport(wrk); // emit output row(s) + } + return num_lines_out; +} + +static int EmitRescaledYUV(const VP8Io* const io, WebPDecParams* const p) { + const int mb_h = io->mb_h; + const int uv_mb_h = (mb_h + 1) >> 1; + WebPRescaler* const scaler = p->scaler_y; + int num_lines_out = 0; + if (WebPIsAlphaMode(p->output->colorspace) && io->a != NULL) { + // Before rescaling, we premultiply the luma directly into the io->y + // internal buffer. This is OK since these samples are not used for + // intra-prediction (the top samples are saved in cache_y_/u_/v_). + // But we need to cast the const away, though. + WebPMultRows((uint8_t*)io->y, io->y_stride, + io->a, io->width, io->mb_w, mb_h, 0); + } + num_lines_out = Rescale(io->y, io->y_stride, mb_h, scaler); + Rescale(io->u, io->uv_stride, uv_mb_h, p->scaler_u); + Rescale(io->v, io->uv_stride, uv_mb_h, p->scaler_v); + return num_lines_out; +} + +static int EmitRescaledAlphaYUV(const VP8Io* const io, WebPDecParams* const p, + int expected_num_lines_out) { + const WebPYUVABuffer* const buf = &p->output->u.YUVA; + uint8_t* const dst_a = buf->a + p->last_y * buf->a_stride; + if (io->a != NULL) { + uint8_t* const dst_y = buf->y + p->last_y * buf->y_stride; + const int num_lines_out = Rescale(io->a, io->width, io->mb_h, p->scaler_a); + assert(expected_num_lines_out == num_lines_out); + if (num_lines_out > 0) { // unmultiply the Y + WebPMultRows(dst_y, buf->y_stride, dst_a, buf->a_stride, + p->scaler_a->dst_width, num_lines_out, 1); + } + } else if (buf->a != NULL) { + // the user requested alpha, but there is none, set it to opaque. + assert(p->last_y + expected_num_lines_out <= io->scaled_height); + FillAlphaPlane(dst_a, io->scaled_width, expected_num_lines_out, + buf->a_stride); + } + return 0; +} + +static int InitYUVRescaler(const VP8Io* const io, WebPDecParams* const p) { + const int has_alpha = WebPIsAlphaMode(p->output->colorspace); + const WebPYUVABuffer* const buf = &p->output->u.YUVA; + const int out_width = io->scaled_width; + const int out_height = io->scaled_height; + const int uv_out_width = (out_width + 1) >> 1; + const int uv_out_height = (out_height + 1) >> 1; + const int uv_in_width = (io->mb_w + 1) >> 1; + const int uv_in_height = (io->mb_h + 1) >> 1; + const size_t work_size = 2 * out_width; // scratch memory for luma rescaler + const size_t uv_work_size = 2 * uv_out_width; // and for each u/v ones + size_t tmp_size, rescaler_size; + rescaler_t* work; + WebPRescaler* scalers; + const int num_rescalers = has_alpha ? 4 : 3; + + tmp_size = (work_size + 2 * uv_work_size) * sizeof(*work); + if (has_alpha) { + tmp_size += work_size * sizeof(*work); + } + rescaler_size = num_rescalers * sizeof(*p->scaler_y) + WEBP_ALIGN_CST; + + p->memory = WebPSafeMalloc(1ULL, tmp_size + rescaler_size); + if (p->memory == NULL) { + return 0; // memory error + } + work = (rescaler_t*)p->memory; + + scalers = (WebPRescaler*)WEBP_ALIGN((const uint8_t*)work + tmp_size); + p->scaler_y = &scalers[0]; + p->scaler_u = &scalers[1]; + p->scaler_v = &scalers[2]; + p->scaler_a = has_alpha ? &scalers[3] : NULL; + + WebPRescalerInit(p->scaler_y, io->mb_w, io->mb_h, + buf->y, out_width, out_height, buf->y_stride, 1, + work); + WebPRescalerInit(p->scaler_u, uv_in_width, uv_in_height, + buf->u, uv_out_width, uv_out_height, buf->u_stride, 1, + work + work_size); + WebPRescalerInit(p->scaler_v, uv_in_width, uv_in_height, + buf->v, uv_out_width, uv_out_height, buf->v_stride, 1, + work + work_size + uv_work_size); + p->emit = EmitRescaledYUV; + + if (has_alpha) { + WebPRescalerInit(p->scaler_a, io->mb_w, io->mb_h, + buf->a, out_width, out_height, buf->a_stride, 1, + work + work_size + 2 * uv_work_size); + p->emit_alpha = EmitRescaledAlphaYUV; + WebPInitAlphaProcessing(); + } + return 1; +} + +//------------------------------------------------------------------------------ +// RGBA rescaling + +static int ExportRGB(WebPDecParams* const p, int y_pos) { + const WebPYUV444Converter convert = + WebPYUV444Converters[p->output->colorspace]; + const WebPRGBABuffer* const buf = &p->output->u.RGBA; + uint8_t* dst = buf->rgba + y_pos * buf->stride; + int num_lines_out = 0; + // For RGB rescaling, because of the YUV420, current scan position + // U/V can be +1/-1 line from the Y one. Hence the double test. + while (WebPRescalerHasPendingOutput(p->scaler_y) && + WebPRescalerHasPendingOutput(p->scaler_u)) { + assert(y_pos + num_lines_out < p->output->height); + assert(p->scaler_u->y_accum == p->scaler_v->y_accum); + WebPRescalerExportRow(p->scaler_y); + WebPRescalerExportRow(p->scaler_u); + WebPRescalerExportRow(p->scaler_v); + convert(p->scaler_y->dst, p->scaler_u->dst, p->scaler_v->dst, + dst, p->scaler_y->dst_width); + dst += buf->stride; + ++num_lines_out; + } + return num_lines_out; +} + +static int EmitRescaledRGB(const VP8Io* const io, WebPDecParams* const p) { + const int mb_h = io->mb_h; + const int uv_mb_h = (mb_h + 1) >> 1; + int j = 0, uv_j = 0; + int num_lines_out = 0; + while (j < mb_h) { + const int y_lines_in = + WebPRescalerImport(p->scaler_y, mb_h - j, + io->y + j * io->y_stride, io->y_stride); + j += y_lines_in; + if (WebPRescaleNeededLines(p->scaler_u, uv_mb_h - uv_j)) { + const int u_lines_in = + WebPRescalerImport(p->scaler_u, uv_mb_h - uv_j, + io->u + uv_j * io->uv_stride, io->uv_stride); + const int v_lines_in = + WebPRescalerImport(p->scaler_v, uv_mb_h - uv_j, + io->v + uv_j * io->uv_stride, io->uv_stride); + (void)v_lines_in; // remove a gcc warning + assert(u_lines_in == v_lines_in); + uv_j += u_lines_in; + } + num_lines_out += ExportRGB(p, p->last_y + num_lines_out); + } + return num_lines_out; +} + +static int ExportAlpha(WebPDecParams* const p, int y_pos, int max_lines_out) { + const WebPRGBABuffer* const buf = &p->output->u.RGBA; + uint8_t* const base_rgba = buf->rgba + y_pos * buf->stride; + const WEBP_CSP_MODE colorspace = p->output->colorspace; + const int alpha_first = + (colorspace == MODE_ARGB || colorspace == MODE_Argb); + uint8_t* dst = base_rgba + (alpha_first ? 0 : 3); + int num_lines_out = 0; + const int is_premult_alpha = WebPIsPremultipliedMode(colorspace); + uint32_t non_opaque = 0; + const int width = p->scaler_a->dst_width; + + while (WebPRescalerHasPendingOutput(p->scaler_a) && + num_lines_out < max_lines_out) { + assert(y_pos + num_lines_out < p->output->height); + WebPRescalerExportRow(p->scaler_a); + non_opaque |= WebPDispatchAlpha(p->scaler_a->dst, 0, width, 1, dst, 0); + dst += buf->stride; + ++num_lines_out; + } + if (is_premult_alpha && non_opaque) { + WebPApplyAlphaMultiply(base_rgba, alpha_first, + width, num_lines_out, buf->stride); + } + return num_lines_out; +} + +static int ExportAlphaRGBA4444(WebPDecParams* const p, int y_pos, + int max_lines_out) { + const WebPRGBABuffer* const buf = &p->output->u.RGBA; + uint8_t* const base_rgba = buf->rgba + y_pos * buf->stride; +#if (WEBP_SWAP_16BIT_CSP == 1) + uint8_t* alpha_dst = base_rgba; +#else + uint8_t* alpha_dst = base_rgba + 1; +#endif + int num_lines_out = 0; + const WEBP_CSP_MODE colorspace = p->output->colorspace; + const int width = p->scaler_a->dst_width; + const int is_premult_alpha = WebPIsPremultipliedMode(colorspace); + uint32_t alpha_mask = 0x0f; + + while (WebPRescalerHasPendingOutput(p->scaler_a) && + num_lines_out < max_lines_out) { + int i; + assert(y_pos + num_lines_out < p->output->height); + WebPRescalerExportRow(p->scaler_a); + for (i = 0; i < width; ++i) { + // Fill in the alpha value (converted to 4 bits). + const uint32_t alpha_value = p->scaler_a->dst[i] >> 4; + alpha_dst[2 * i] = (alpha_dst[2 * i] & 0xf0) | alpha_value; + alpha_mask &= alpha_value; + } + alpha_dst += buf->stride; + ++num_lines_out; + } + if (is_premult_alpha && alpha_mask != 0x0f) { + WebPApplyAlphaMultiply4444(base_rgba, width, num_lines_out, buf->stride); + } + return num_lines_out; +} + +static int EmitRescaledAlphaRGB(const VP8Io* const io, WebPDecParams* const p, + int expected_num_out_lines) { + if (io->a != NULL) { + WebPRescaler* const scaler = p->scaler_a; + int lines_left = expected_num_out_lines; + const int y_end = p->last_y + lines_left; + while (lines_left > 0) { + const int row_offset = scaler->src_y - io->mb_y; + WebPRescalerImport(scaler, io->mb_h + io->mb_y - scaler->src_y, + io->a + row_offset * io->width, io->width); + lines_left -= p->emit_alpha_row(p, y_end - lines_left, lines_left); + } + } + return 0; +} + +static int InitRGBRescaler(const VP8Io* const io, WebPDecParams* const p) { + const int has_alpha = WebPIsAlphaMode(p->output->colorspace); + const int out_width = io->scaled_width; + const int out_height = io->scaled_height; + const int uv_in_width = (io->mb_w + 1) >> 1; + const int uv_in_height = (io->mb_h + 1) >> 1; + const size_t work_size = 2 * out_width; // scratch memory for one rescaler + rescaler_t* work; // rescalers work area + uint8_t* tmp; // tmp storage for scaled YUV444 samples before RGB conversion + size_t tmp_size1, tmp_size2, total_size, rescaler_size; + WebPRescaler* scalers; + const int num_rescalers = has_alpha ? 4 : 3; + + tmp_size1 = 3 * work_size; + tmp_size2 = 3 * out_width; + if (has_alpha) { + tmp_size1 += work_size; + tmp_size2 += out_width; + } + total_size = tmp_size1 * sizeof(*work) + tmp_size2 * sizeof(*tmp); + rescaler_size = num_rescalers * sizeof(*p->scaler_y) + WEBP_ALIGN_CST; + + p->memory = WebPSafeMalloc(1ULL, total_size + rescaler_size); + if (p->memory == NULL) { + return 0; // memory error + } + work = (rescaler_t*)p->memory; + tmp = (uint8_t*)(work + tmp_size1); + + scalers = (WebPRescaler*)WEBP_ALIGN((const uint8_t*)work + total_size); + p->scaler_y = &scalers[0]; + p->scaler_u = &scalers[1]; + p->scaler_v = &scalers[2]; + p->scaler_a = has_alpha ? &scalers[3] : NULL; + + WebPRescalerInit(p->scaler_y, io->mb_w, io->mb_h, + tmp + 0 * out_width, out_width, out_height, 0, 1, + work + 0 * work_size); + WebPRescalerInit(p->scaler_u, uv_in_width, uv_in_height, + tmp + 1 * out_width, out_width, out_height, 0, 1, + work + 1 * work_size); + WebPRescalerInit(p->scaler_v, uv_in_width, uv_in_height, + tmp + 2 * out_width, out_width, out_height, 0, 1, + work + 2 * work_size); + p->emit = EmitRescaledRGB; + WebPInitYUV444Converters(); + + if (has_alpha) { + WebPRescalerInit(p->scaler_a, io->mb_w, io->mb_h, + tmp + 3 * out_width, out_width, out_height, 0, 1, + work + 3 * work_size); + p->emit_alpha = EmitRescaledAlphaRGB; + if (p->output->colorspace == MODE_RGBA_4444 || + p->output->colorspace == MODE_rgbA_4444) { + p->emit_alpha_row = ExportAlphaRGBA4444; + } else { + p->emit_alpha_row = ExportAlpha; + } + WebPInitAlphaProcessing(); + } + return 1; +} + +#endif // WEBP_REDUCE_SIZE + +//------------------------------------------------------------------------------ +// Default custom functions + +static int CustomSetup(VP8Io* io) { + WebPDecParams* const p = (WebPDecParams*)io->opaque; + const WEBP_CSP_MODE colorspace = p->output->colorspace; + const int is_rgb = WebPIsRGBMode(colorspace); + const int is_alpha = WebPIsAlphaMode(colorspace); + + p->memory = NULL; + p->emit = NULL; + p->emit_alpha = NULL; + p->emit_alpha_row = NULL; + if (!WebPIoInitFromOptions(p->options, io, is_alpha ? MODE_YUV : MODE_YUVA)) { + return 0; + } + if (is_alpha && WebPIsPremultipliedMode(colorspace)) { + WebPInitUpsamplers(); + } + if (io->use_scaling) { +#if !defined(WEBP_REDUCE_SIZE) + const int ok = is_rgb ? InitRGBRescaler(io, p) : InitYUVRescaler(io, p); + if (!ok) { + return 0; // memory error + } +#else + return 0; // rescaling support not compiled +#endif + } else { + if (is_rgb) { + WebPInitSamplers(); + p->emit = EmitSampledRGB; // default + if (io->fancy_upsampling) { +#ifdef FANCY_UPSAMPLING + const int uv_width = (io->mb_w + 1) >> 1; + p->memory = WebPSafeMalloc(1ULL, (size_t)(io->mb_w + 2 * uv_width)); + if (p->memory == NULL) { + return 0; // memory error. + } + p->tmp_y = (uint8_t*)p->memory; + p->tmp_u = p->tmp_y + io->mb_w; + p->tmp_v = p->tmp_u + uv_width; + p->emit = EmitFancyRGB; + WebPInitUpsamplers(); +#endif + } + } else { + p->emit = EmitYUV; + } + if (is_alpha) { // need transparency output + p->emit_alpha = + (colorspace == MODE_RGBA_4444 || colorspace == MODE_rgbA_4444) ? + EmitAlphaRGBA4444 + : is_rgb ? EmitAlphaRGB + : EmitAlphaYUV; + if (is_rgb) { + WebPInitAlphaProcessing(); + } + } + } + + return 1; +} + +//------------------------------------------------------------------------------ + +static int CustomPut(const VP8Io* io) { + WebPDecParams* const p = (WebPDecParams*)io->opaque; + const int mb_w = io->mb_w; + const int mb_h = io->mb_h; + int num_lines_out; + assert(!(io->mb_y & 1)); + + if (mb_w <= 0 || mb_h <= 0) { + return 0; + } + num_lines_out = p->emit(io, p); + if (p->emit_alpha != NULL) { + p->emit_alpha(io, p, num_lines_out); + } + p->last_y += num_lines_out; + return 1; +} + +//------------------------------------------------------------------------------ + +static void CustomTeardown(const VP8Io* io) { + WebPDecParams* const p = (WebPDecParams*)io->opaque; + WebPSafeFree(p->memory); + p->memory = NULL; +} + +//------------------------------------------------------------------------------ +// Main entry point + +void WebPInitCustomIo(WebPDecParams* const params, VP8Io* const io) { + io->put = CustomPut; + io->setup = CustomSetup; + io->teardown = CustomTeardown; + io->opaque = params; +} + +//------------------------------------------------------------------------------ diff --git a/thirdparty/libwebp/src/dec/quant_dec.c b/thirdparty/libwebp/src/dec/quant_dec.c new file mode 100644 index 0000000000..f07212ad73 --- /dev/null +++ b/thirdparty/libwebp/src/dec/quant_dec.c @@ -0,0 +1,110 @@ +// Copyright 2010 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. +// ----------------------------------------------------------------------------- +// +// Quantizer initialization +// +// Author: Skal (pascal.massimino@gmail.com) + +#include "src/dec/vp8i_dec.h" + +static WEBP_INLINE int clip(int v, int M) { + return v < 0 ? 0 : v > M ? M : v; +} + +// Paragraph 14.1 +static const uint8_t kDcTable[128] = { + 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 uint16_t kAcTable[128] = { + 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 +}; + +//------------------------------------------------------------------------------ +// Paragraph 9.6 + +void VP8ParseQuant(VP8Decoder* const dec) { + VP8BitReader* const br = &dec->br_; + const int base_q0 = VP8GetValue(br, 7); + const int dqy1_dc = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0; + const int dqy2_dc = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0; + const int dqy2_ac = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0; + const int dquv_dc = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0; + const int dquv_ac = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0; + + const VP8SegmentHeader* const hdr = &dec->segment_hdr_; + int i; + + for (i = 0; i < NUM_MB_SEGMENTS; ++i) { + int q; + if (hdr->use_segment_) { + q = hdr->quantizer_[i]; + if (!hdr->absolute_delta_) { + q += base_q0; + } + } else { + if (i > 0) { + dec->dqm_[i] = dec->dqm_[0]; + continue; + } else { + q = base_q0; + } + } + { + VP8QuantMatrix* const m = &dec->dqm_[i]; + m->y1_mat_[0] = kDcTable[clip(q + dqy1_dc, 127)]; + m->y1_mat_[1] = kAcTable[clip(q + 0, 127)]; + + m->y2_mat_[0] = kDcTable[clip(q + dqy2_dc, 127)] * 2; + // 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. + m->y2_mat_[1] = (kAcTable[clip(q + dqy2_ac, 127)] * 101581) >> 16; + if (m->y2_mat_[1] < 8) m->y2_mat_[1] = 8; + + m->uv_mat_[0] = kDcTable[clip(q + dquv_dc, 117)]; + m->uv_mat_[1] = kAcTable[clip(q + dquv_ac, 127)]; + + m->uv_quant_ = q + dquv_ac; // for dithering strength evaluation + } + } +} + +//------------------------------------------------------------------------------ + diff --git a/thirdparty/libwebp/src/dec/tree_dec.c b/thirdparty/libwebp/src/dec/tree_dec.c new file mode 100644 index 0000000000..3f5a957d32 --- /dev/null +++ b/thirdparty/libwebp/src/dec/tree_dec.c @@ -0,0 +1,532 @@ +// Copyright 2010 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. +// ----------------------------------------------------------------------------- +// +// Coding trees and probas +// +// Author: Skal (pascal.massimino@gmail.com) + +#include "src/dec/vp8i_dec.h" +#include "src/utils/bit_reader_inl_utils.h" + +#if !defined(USE_GENERIC_TREE) +#if !defined(__arm__) && !defined(_M_ARM) && !defined(__aarch64__) +// using a table is ~1-2% slower on ARM. Prefer the coded-tree approach then. +#define USE_GENERIC_TREE 1 // ALTERNATE_CODE +#else +#define USE_GENERIC_TREE 0 +#endif +#endif // USE_GENERIC_TREE + +#if (USE_GENERIC_TREE == 1) +static const int8_t kYModesIntra4[18] = { + -B_DC_PRED, 1, + -B_TM_PRED, 2, + -B_VE_PRED, 3, + 4, 6, + -B_HE_PRED, 5, + -B_RD_PRED, -B_VR_PRED, + -B_LD_PRED, 7, + -B_VL_PRED, 8, + -B_HD_PRED, -B_HU_PRED +}; +#endif + +//------------------------------------------------------------------------------ +// Default probabilities + +// Paragraph 13.5 +static const uint8_t + CoeffsProba0[NUM_TYPES][NUM_BANDS][NUM_CTX][NUM_PROBAS] = { + { { { 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 } + }, + { { 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 } + }, + { { 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 }, + }, + { { 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 }, + }, + { { 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 } + }, + { { 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 } + }, + { { 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 } + }, + { { 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 } + } + }, + { { { 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 } + }, + { { 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 } + }, + { { 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 } + }, + { { 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 } + }, + { { 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 } + }, + { { 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 } + }, + { { 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 } + }, + { { 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 } + } + }, + { { { 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 } + }, + { { 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 } + }, + { { 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 } + }, + { { 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 } + }, + { { 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 } + }, + { { 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 } + }, + { { 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 } + }, + { { 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 } + } + }, + { { { 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 } + }, + { { 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 } + }, + { { 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 } + }, + { { 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 } + }, + { { 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 } + }, + { { 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 } + }, + { { 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 } + }, + { { 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 } + } + } +}; + +// Paragraph 11.5 +static const uint8_t kBModesProba[NUM_BMODES][NUM_BMODES][NUM_BMODES - 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 }, + { 114, 26, 17, 163, 44, 195, 21, 10, 173 }, + { 121, 24, 80, 195, 26, 62, 44, 64, 85 }, + { 144, 71, 10, 38, 171, 213, 144, 34, 26 }, + { 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 }, + { 74, 43, 26, 146, 73, 166, 49, 23, 157 }, + { 65, 38, 105, 160, 51, 52, 31, 115, 128 }, + { 104, 79, 12, 27, 217, 255, 87, 17, 7 }, + { 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 }, + { 39, 28, 85, 171, 58, 165, 90, 98, 64 }, + { 34, 22, 116, 206, 23, 34, 43, 166, 73 }, + { 107, 54, 32, 26, 51, 1, 81, 43, 31 }, + { 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 }, + { 88, 43, 29, 140, 166, 213, 37, 43, 154 }, + { 61, 63, 30, 155, 67, 45, 68, 1, 209 }, + { 100, 80, 8, 43, 154, 1, 51, 26, 71 }, + { 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 } }, + { { 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 }, + { 47, 15, 16, 183, 34, 223, 49, 45, 183 }, + { 46, 17, 33, 183, 6, 98, 15, 32, 183 }, + { 57, 46, 22, 24, 128, 1, 54, 17, 37 }, + { 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 }, + { 39, 19, 53, 221, 26, 114, 32, 73, 255 }, + { 31, 9, 65, 234, 2, 15, 1, 118, 73 }, + { 75, 32, 12, 51, 192, 255, 160, 43, 51 }, + { 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 } }, + { { 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 }, + { 38, 33, 13, 121, 57, 73, 26, 1, 85 }, + { 41, 10, 67, 138, 77, 110, 90, 47, 114 }, + { 115, 21, 2, 10, 102, 255, 166, 23, 6 }, + { 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 } }, + { { 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 }, + { 37, 43, 37, 154, 100, 163, 85, 160, 1 }, + { 63, 9, 92, 136, 28, 64, 32, 201, 85 }, + { 75, 15, 9, 9, 64, 255, 184, 119, 16 }, + { 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 }, + { 45, 16, 21, 91, 64, 222, 7, 1, 197 }, + { 56, 21, 39, 155, 60, 138, 23, 102, 213 }, + { 83, 12, 13, 54, 192, 255, 68, 47, 28 }, + { 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 }, + { 71, 30, 17, 119, 118, 255, 17, 18, 138 }, + { 101, 38, 60, 138, 55, 70, 43, 26, 142 }, + { 146, 36, 19, 30, 171, 255, 97, 27, 20 }, + { 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 } } +}; + +void VP8ResetProba(VP8Proba* const proba) { + memset(proba->segments_, 255u, sizeof(proba->segments_)); + // proba->bands_[][] is initialized later +} + +static void ParseIntraMode(VP8BitReader* const br, + VP8Decoder* const dec, int mb_x) { + uint8_t* const top = dec->intra_t_ + 4 * mb_x; + uint8_t* const left = dec->intra_l_; + VP8MBData* const block = dec->mb_data_ + mb_x; + + // Note: we don't save segment map (yet), as we don't expect + // to decode more than 1 keyframe. + if (dec->segment_hdr_.update_map_) { + // Hardcoded tree parsing + block->segment_ = !VP8GetBit(br, dec->proba_.segments_[0]) + ? VP8GetBit(br, dec->proba_.segments_[1]) + : 2 + VP8GetBit(br, dec->proba_.segments_[2]); + } else { + block->segment_ = 0; // default for intra + } + if (dec->use_skip_proba_) block->skip_ = VP8GetBit(br, dec->skip_p_); + + block->is_i4x4_ = !VP8GetBit(br, 145); // decide for B_PRED first + if (!block->is_i4x4_) { + // Hardcoded 16x16 intra-mode decision tree. + const int ymode = + VP8GetBit(br, 156) ? (VP8GetBit(br, 128) ? TM_PRED : H_PRED) + : (VP8GetBit(br, 163) ? V_PRED : DC_PRED); + block->imodes_[0] = ymode; + memset(top, ymode, 4 * sizeof(*top)); + memset(left, ymode, 4 * sizeof(*left)); + } else { + uint8_t* modes = block->imodes_; + int y; + for (y = 0; y < 4; ++y) { + int ymode = left[y]; + int x; + for (x = 0; x < 4; ++x) { + const uint8_t* const prob = kBModesProba[top[x]][ymode]; +#if (USE_GENERIC_TREE == 1) + // Generic tree-parsing + int i = kYModesIntra4[VP8GetBit(br, prob[0])]; + while (i > 0) { + i = kYModesIntra4[2 * i + VP8GetBit(br, prob[i])]; + } + ymode = -i; +#else + // Hardcoded tree parsing + ymode = !VP8GetBit(br, prob[0]) ? B_DC_PRED : + !VP8GetBit(br, prob[1]) ? B_TM_PRED : + !VP8GetBit(br, prob[2]) ? B_VE_PRED : + !VP8GetBit(br, prob[3]) ? + (!VP8GetBit(br, prob[4]) ? B_HE_PRED : + (!VP8GetBit(br, prob[5]) ? B_RD_PRED : B_VR_PRED)) : + (!VP8GetBit(br, prob[6]) ? B_LD_PRED : + (!VP8GetBit(br, prob[7]) ? B_VL_PRED : + (!VP8GetBit(br, prob[8]) ? B_HD_PRED : B_HU_PRED))); +#endif // USE_GENERIC_TREE + top[x] = ymode; + } + memcpy(modes, top, 4 * sizeof(*top)); + modes += 4; + left[y] = ymode; + } + } + // Hardcoded UVMode decision tree + block->uvmode_ = !VP8GetBit(br, 142) ? DC_PRED + : !VP8GetBit(br, 114) ? V_PRED + : VP8GetBit(br, 183) ? TM_PRED : H_PRED; +} + +int VP8ParseIntraModeRow(VP8BitReader* const br, VP8Decoder* const dec) { + int mb_x; + for (mb_x = 0; mb_x < dec->mb_w_; ++mb_x) { + ParseIntraMode(br, dec, mb_x); + } + return !dec->br_.eof_; +} + +//------------------------------------------------------------------------------ +// Paragraph 13 + +static const uint8_t + CoeffsUpdateProba[NUM_TYPES][NUM_BANDS][NUM_CTX][NUM_PROBAS] = { + { { { 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 } + } + } +}; + +// Paragraph 9.9 + +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 +}; + +void VP8ParseProba(VP8BitReader* const br, VP8Decoder* const dec) { + VP8Proba* const proba = &dec->proba_; + int t, b, c, p; + for (t = 0; t < NUM_TYPES; ++t) { + for (b = 0; b < NUM_BANDS; ++b) { + for (c = 0; c < NUM_CTX; ++c) { + for (p = 0; p < NUM_PROBAS; ++p) { + const int v = VP8GetBit(br, CoeffsUpdateProba[t][b][c][p]) ? + VP8GetValue(br, 8) : CoeffsProba0[t][b][c][p]; + proba->bands_[t][b].probas_[c][p] = v; + } + } + } + for (b = 0; b < 16 + 1; ++b) { + proba->bands_ptr_[t][b] = &proba->bands_[t][kBands[b]]; + } + } + dec->use_skip_proba_ = VP8Get(br); + if (dec->use_skip_proba_) { + dec->skip_p_ = VP8GetValue(br, 8); + } +} + diff --git a/thirdparty/libwebp/src/dec/vp8_dec.c b/thirdparty/libwebp/src/dec/vp8_dec.c new file mode 100644 index 0000000000..6212efd179 --- /dev/null +++ b/thirdparty/libwebp/src/dec/vp8_dec.c @@ -0,0 +1,721 @@ +// Copyright 2010 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. +// ----------------------------------------------------------------------------- +// +// main entry for the decoder +// +// Author: Skal (pascal.massimino@gmail.com) + +#include <stdlib.h> + +#include "src/dec/alphai_dec.h" +#include "src/dec/vp8i_dec.h" +#include "src/dec/vp8li_dec.h" +#include "src/dec/webpi_dec.h" +#include "src/utils/bit_reader_inl_utils.h" +#include "src/utils/utils.h" + +//------------------------------------------------------------------------------ + +int WebPGetDecoderVersion(void) { + return (DEC_MAJ_VERSION << 16) | (DEC_MIN_VERSION << 8) | DEC_REV_VERSION; +} + +//------------------------------------------------------------------------------ +// Signature and pointer-to-function for GetCoeffs() variants below. + +typedef int (*GetCoeffsFunc)(VP8BitReader* const br, + const VP8BandProbas* const prob[], + int ctx, const quant_t dq, int n, int16_t* out); +static volatile GetCoeffsFunc GetCoeffs = NULL; + +static void InitGetCoeffs(void); + +//------------------------------------------------------------------------------ +// VP8Decoder + +static void SetOk(VP8Decoder* const dec) { + dec->status_ = VP8_STATUS_OK; + dec->error_msg_ = "OK"; +} + +int VP8InitIoInternal(VP8Io* const io, int version) { + if (WEBP_ABI_IS_INCOMPATIBLE(version, WEBP_DECODER_ABI_VERSION)) { + return 0; // mismatch error + } + if (io != NULL) { + memset(io, 0, sizeof(*io)); + } + return 1; +} + +VP8Decoder* VP8New(void) { + VP8Decoder* const dec = (VP8Decoder*)WebPSafeCalloc(1ULL, sizeof(*dec)); + if (dec != NULL) { + SetOk(dec); + WebPGetWorkerInterface()->Init(&dec->worker_); + dec->ready_ = 0; + dec->num_parts_minus_one_ = 0; + InitGetCoeffs(); + } + return dec; +} + +VP8StatusCode VP8Status(VP8Decoder* const dec) { + if (!dec) return VP8_STATUS_INVALID_PARAM; + return dec->status_; +} + +const char* VP8StatusMessage(VP8Decoder* const dec) { + if (dec == NULL) return "no object"; + if (!dec->error_msg_) return "OK"; + return dec->error_msg_; +} + +void VP8Delete(VP8Decoder* const dec) { + if (dec != NULL) { + VP8Clear(dec); + WebPSafeFree(dec); + } +} + +int VP8SetError(VP8Decoder* const dec, + VP8StatusCode error, const char* const msg) { + // The oldest error reported takes precedence over the new one. + if (dec->status_ == VP8_STATUS_OK) { + dec->status_ = error; + dec->error_msg_ = msg; + dec->ready_ = 0; + } + return 0; +} + +//------------------------------------------------------------------------------ + +int VP8CheckSignature(const uint8_t* const data, size_t data_size) { + return (data_size >= 3 && + data[0] == 0x9d && data[1] == 0x01 && data[2] == 0x2a); +} + +int VP8GetInfo(const uint8_t* data, size_t data_size, size_t chunk_size, + int* const width, int* const height) { + if (data == NULL || data_size < VP8_FRAME_HEADER_SIZE) { + return 0; // not enough data + } + // check signature + if (!VP8CheckSignature(data + 3, data_size - 3)) { + return 0; // Wrong signature. + } else { + const uint32_t bits = data[0] | (data[1] << 8) | (data[2] << 16); + const int key_frame = !(bits & 1); + const int w = ((data[7] << 8) | data[6]) & 0x3fff; + const int h = ((data[9] << 8) | data[8]) & 0x3fff; + + if (!key_frame) { // Not a keyframe. + return 0; + } + + if (((bits >> 1) & 7) > 3) { + return 0; // unknown profile + } + if (!((bits >> 4) & 1)) { + return 0; // first frame is invisible! + } + if (((bits >> 5)) >= chunk_size) { // partition_length + return 0; // inconsistent size information. + } + if (w == 0 || h == 0) { + return 0; // We don't support both width and height to be zero. + } + + if (width) { + *width = w; + } + if (height) { + *height = h; + } + + return 1; + } +} + +//------------------------------------------------------------------------------ +// Header parsing + +static void ResetSegmentHeader(VP8SegmentHeader* const hdr) { + assert(hdr != NULL); + hdr->use_segment_ = 0; + hdr->update_map_ = 0; + hdr->absolute_delta_ = 1; + memset(hdr->quantizer_, 0, sizeof(hdr->quantizer_)); + memset(hdr->filter_strength_, 0, sizeof(hdr->filter_strength_)); +} + +// Paragraph 9.3 +static int ParseSegmentHeader(VP8BitReader* br, + VP8SegmentHeader* hdr, VP8Proba* proba) { + assert(br != NULL); + assert(hdr != NULL); + hdr->use_segment_ = VP8Get(br); + if (hdr->use_segment_) { + hdr->update_map_ = VP8Get(br); + if (VP8Get(br)) { // update data + int s; + hdr->absolute_delta_ = VP8Get(br); + for (s = 0; s < NUM_MB_SEGMENTS; ++s) { + hdr->quantizer_[s] = VP8Get(br) ? VP8GetSignedValue(br, 7) : 0; + } + for (s = 0; s < NUM_MB_SEGMENTS; ++s) { + hdr->filter_strength_[s] = VP8Get(br) ? VP8GetSignedValue(br, 6) : 0; + } + } + if (hdr->update_map_) { + int s; + for (s = 0; s < MB_FEATURE_TREE_PROBS; ++s) { + proba->segments_[s] = VP8Get(br) ? VP8GetValue(br, 8) : 255u; + } + } + } else { + hdr->update_map_ = 0; + } + return !br->eof_; +} + +// Paragraph 9.5 +// This function returns VP8_STATUS_SUSPENDED if we don't have all the +// necessary data in 'buf'. +// This case is not necessarily an error (for incremental decoding). +// Still, no bitreader is ever initialized to make it possible to read +// unavailable memory. +// If we don't even have the partitions' sizes, than VP8_STATUS_NOT_ENOUGH_DATA +// is returned, and this is an unrecoverable error. +// If the partitions were positioned ok, VP8_STATUS_OK is returned. +static VP8StatusCode ParsePartitions(VP8Decoder* const dec, + const uint8_t* buf, size_t size) { + VP8BitReader* const br = &dec->br_; + const uint8_t* sz = buf; + const uint8_t* buf_end = buf + size; + const uint8_t* part_start; + size_t size_left = size; + size_t last_part; + size_t p; + + dec->num_parts_minus_one_ = (1 << VP8GetValue(br, 2)) - 1; + last_part = dec->num_parts_minus_one_; + if (size < 3 * last_part) { + // we can't even read the sizes with sz[]! That's a failure. + return VP8_STATUS_NOT_ENOUGH_DATA; + } + part_start = buf + last_part * 3; + size_left -= last_part * 3; + for (p = 0; p < last_part; ++p) { + size_t psize = sz[0] | (sz[1] << 8) | (sz[2] << 16); + if (psize > size_left) psize = size_left; + VP8InitBitReader(dec->parts_ + p, part_start, psize); + part_start += psize; + size_left -= psize; + sz += 3; + } + VP8InitBitReader(dec->parts_ + last_part, part_start, size_left); + return (part_start < buf_end) ? VP8_STATUS_OK : + VP8_STATUS_SUSPENDED; // Init is ok, but there's not enough data +} + +// Paragraph 9.4 +static int ParseFilterHeader(VP8BitReader* br, VP8Decoder* const dec) { + VP8FilterHeader* const hdr = &dec->filter_hdr_; + hdr->simple_ = VP8Get(br); + hdr->level_ = VP8GetValue(br, 6); + hdr->sharpness_ = VP8GetValue(br, 3); + hdr->use_lf_delta_ = VP8Get(br); + if (hdr->use_lf_delta_) { + if (VP8Get(br)) { // update lf-delta? + int i; + for (i = 0; i < NUM_REF_LF_DELTAS; ++i) { + if (VP8Get(br)) { + hdr->ref_lf_delta_[i] = VP8GetSignedValue(br, 6); + } + } + for (i = 0; i < NUM_MODE_LF_DELTAS; ++i) { + if (VP8Get(br)) { + hdr->mode_lf_delta_[i] = VP8GetSignedValue(br, 6); + } + } + } + } + dec->filter_type_ = (hdr->level_ == 0) ? 0 : hdr->simple_ ? 1 : 2; + return !br->eof_; +} + +// Topmost call +int VP8GetHeaders(VP8Decoder* const dec, VP8Io* const io) { + const uint8_t* buf; + size_t buf_size; + VP8FrameHeader* frm_hdr; + VP8PictureHeader* pic_hdr; + VP8BitReader* br; + VP8StatusCode status; + + if (dec == NULL) { + return 0; + } + SetOk(dec); + if (io == NULL) { + return VP8SetError(dec, VP8_STATUS_INVALID_PARAM, + "null VP8Io passed to VP8GetHeaders()"); + } + buf = io->data; + buf_size = io->data_size; + if (buf_size < 4) { + return VP8SetError(dec, VP8_STATUS_NOT_ENOUGH_DATA, + "Truncated header."); + } + + // Paragraph 9.1 + { + const uint32_t bits = buf[0] | (buf[1] << 8) | (buf[2] << 16); + frm_hdr = &dec->frm_hdr_; + frm_hdr->key_frame_ = !(bits & 1); + frm_hdr->profile_ = (bits >> 1) & 7; + frm_hdr->show_ = (bits >> 4) & 1; + frm_hdr->partition_length_ = (bits >> 5); + if (frm_hdr->profile_ > 3) { + return VP8SetError(dec, VP8_STATUS_BITSTREAM_ERROR, + "Incorrect keyframe parameters."); + } + if (!frm_hdr->show_) { + return VP8SetError(dec, VP8_STATUS_UNSUPPORTED_FEATURE, + "Frame not displayable."); + } + buf += 3; + buf_size -= 3; + } + + pic_hdr = &dec->pic_hdr_; + if (frm_hdr->key_frame_) { + // Paragraph 9.2 + if (buf_size < 7) { + return VP8SetError(dec, VP8_STATUS_NOT_ENOUGH_DATA, + "cannot parse picture header"); + } + if (!VP8CheckSignature(buf, buf_size)) { + return VP8SetError(dec, VP8_STATUS_BITSTREAM_ERROR, + "Bad code word"); + } + pic_hdr->width_ = ((buf[4] << 8) | buf[3]) & 0x3fff; + pic_hdr->xscale_ = buf[4] >> 6; // ratio: 1, 5/4 5/3 or 2 + pic_hdr->height_ = ((buf[6] << 8) | buf[5]) & 0x3fff; + pic_hdr->yscale_ = buf[6] >> 6; + buf += 7; + buf_size -= 7; + + dec->mb_w_ = (pic_hdr->width_ + 15) >> 4; + dec->mb_h_ = (pic_hdr->height_ + 15) >> 4; + + // Setup default output area (can be later modified during io->setup()) + io->width = pic_hdr->width_; + io->height = pic_hdr->height_; + // IMPORTANT! use some sane dimensions in crop_* and scaled_* fields. + // So they can be used interchangeably without always testing for + // 'use_cropping'. + io->use_cropping = 0; + io->crop_top = 0; + io->crop_left = 0; + io->crop_right = io->width; + io->crop_bottom = io->height; + io->use_scaling = 0; + io->scaled_width = io->width; + io->scaled_height = io->height; + + io->mb_w = io->width; // sanity check + io->mb_h = io->height; // ditto + + VP8ResetProba(&dec->proba_); + ResetSegmentHeader(&dec->segment_hdr_); + } + + // Check if we have all the partition #0 available, and initialize dec->br_ + // to read this partition (and this partition only). + if (frm_hdr->partition_length_ > buf_size) { + return VP8SetError(dec, VP8_STATUS_NOT_ENOUGH_DATA, + "bad partition length"); + } + + br = &dec->br_; + VP8InitBitReader(br, buf, frm_hdr->partition_length_); + buf += frm_hdr->partition_length_; + buf_size -= frm_hdr->partition_length_; + + if (frm_hdr->key_frame_) { + pic_hdr->colorspace_ = VP8Get(br); + pic_hdr->clamp_type_ = VP8Get(br); + } + if (!ParseSegmentHeader(br, &dec->segment_hdr_, &dec->proba_)) { + return VP8SetError(dec, VP8_STATUS_BITSTREAM_ERROR, + "cannot parse segment header"); + } + // Filter specs + if (!ParseFilterHeader(br, dec)) { + return VP8SetError(dec, VP8_STATUS_BITSTREAM_ERROR, + "cannot parse filter header"); + } + status = ParsePartitions(dec, buf, buf_size); + if (status != VP8_STATUS_OK) { + return VP8SetError(dec, status, "cannot parse partitions"); + } + + // quantizer change + VP8ParseQuant(dec); + + // Frame buffer marking + if (!frm_hdr->key_frame_) { + return VP8SetError(dec, VP8_STATUS_UNSUPPORTED_FEATURE, + "Not a key frame."); + } + + VP8Get(br); // ignore the value of update_proba_ + + VP8ParseProba(br, dec); + + // sanitized state + dec->ready_ = 1; + return 1; +} + +//------------------------------------------------------------------------------ +// Residual decoding (Paragraph 13.2 / 13.3) + +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 +}; + +// See section 13-2: http://tools.ietf.org/html/rfc6386#section-13.2 +static int GetLargeValue(VP8BitReader* const br, const uint8_t* const p) { + int v; + 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); + } + } + return v; +} + +// Returns the position of the last non-zero coeff plus one +static int GetCoeffsFast(VP8BitReader* const br, + const VP8BandProbas* const prob[], + int ctx, const quant_t dq, int n, int16_t* out) { + const uint8_t* p = prob[n]->probas_[ctx]; + for (; n < 16; ++n) { + if (!VP8GetBit(br, p[0])) { + return n; // previous coeff was last non-zero coeff + } + while (!VP8GetBit(br, p[1])) { // sequence of zero coeffs + p = prob[++n]->probas_[0]; + if (n == 16) return 16; + } + { // non zero coeff + const VP8ProbaArray* const p_ctx = &prob[n + 1]->probas_[0]; + int v; + if (!VP8GetBit(br, p[2])) { + v = 1; + p = p_ctx[1]; + } else { + v = GetLargeValue(br, p); + p = p_ctx[2]; + } + out[kZigzag[n]] = VP8GetSigned(br, v) * dq[n > 0]; + } + } + return 16; +} + +// This version of GetCoeffs() uses VP8GetBitAlt() which is an alternate version +// of VP8GetBitAlt() targeting specific platforms. +static int GetCoeffsAlt(VP8BitReader* const br, + const VP8BandProbas* const prob[], + int ctx, const quant_t dq, int n, int16_t* out) { + const uint8_t* p = prob[n]->probas_[ctx]; + for (; n < 16; ++n) { + if (!VP8GetBitAlt(br, p[0])) { + return n; // previous coeff was last non-zero coeff + } + while (!VP8GetBitAlt(br, p[1])) { // sequence of zero coeffs + p = prob[++n]->probas_[0]; + if (n == 16) return 16; + } + { // non zero coeff + const VP8ProbaArray* const p_ctx = &prob[n + 1]->probas_[0]; + int v; + if (!VP8GetBitAlt(br, p[2])) { + v = 1; + p = p_ctx[1]; + } else { + v = GetLargeValue(br, p); + p = p_ctx[2]; + } + out[kZigzag[n]] = VP8GetSigned(br, v) * dq[n > 0]; + } + } + return 16; +} + +WEBP_TSAN_IGNORE_FUNCTION static void InitGetCoeffs(void) { + if (GetCoeffs == NULL) { + if (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kSlowSSSE3)) { + GetCoeffs = GetCoeffsAlt; + } else { + GetCoeffs = GetCoeffsFast; + } + } +} + +static WEBP_INLINE uint32_t NzCodeBits(uint32_t nz_coeffs, int nz, int dc_nz) { + nz_coeffs <<= 2; + nz_coeffs |= (nz > 3) ? 3 : (nz > 1) ? 2 : dc_nz; + return nz_coeffs; +} + +static int ParseResiduals(VP8Decoder* const dec, + VP8MB* const mb, VP8BitReader* const token_br) { + const VP8BandProbas* (* const bands)[16 + 1] = dec->proba_.bands_ptr_; + const VP8BandProbas* const * ac_proba; + VP8MBData* const block = dec->mb_data_ + dec->mb_x_; + const VP8QuantMatrix* const q = &dec->dqm_[block->segment_]; + int16_t* dst = block->coeffs_; + VP8MB* const left_mb = dec->mb_info_ - 1; + uint8_t tnz, lnz; + uint32_t non_zero_y = 0; + uint32_t non_zero_uv = 0; + int x, y, ch; + uint32_t out_t_nz, out_l_nz; + int first; + + memset(dst, 0, 384 * sizeof(*dst)); + if (!block->is_i4x4_) { // parse DC + int16_t dc[16] = { 0 }; + const int ctx = mb->nz_dc_ + left_mb->nz_dc_; + const int nz = GetCoeffs(token_br, bands[1], ctx, q->y2_mat_, 0, dc); + mb->nz_dc_ = left_mb->nz_dc_ = (nz > 0); + if (nz > 1) { // more than just the DC -> perform the full transform + VP8TransformWHT(dc, dst); + } else { // only DC is non-zero -> inlined simplified transform + int i; + const int dc0 = (dc[0] + 3) >> 3; + for (i = 0; i < 16 * 16; i += 16) dst[i] = dc0; + } + first = 1; + ac_proba = bands[0]; + } else { + first = 0; + ac_proba = bands[3]; + } + + tnz = mb->nz_ & 0x0f; + lnz = left_mb->nz_ & 0x0f; + for (y = 0; y < 4; ++y) { + int l = lnz & 1; + uint32_t nz_coeffs = 0; + for (x = 0; x < 4; ++x) { + const int ctx = l + (tnz & 1); + const int nz = GetCoeffs(token_br, ac_proba, ctx, q->y1_mat_, first, dst); + l = (nz > first); + tnz = (tnz >> 1) | (l << 7); + nz_coeffs = NzCodeBits(nz_coeffs, nz, dst[0] != 0); + dst += 16; + } + tnz >>= 4; + lnz = (lnz >> 1) | (l << 7); + non_zero_y = (non_zero_y << 8) | nz_coeffs; + } + out_t_nz = tnz; + out_l_nz = lnz >> 4; + + for (ch = 0; ch < 4; ch += 2) { + uint32_t nz_coeffs = 0; + tnz = mb->nz_ >> (4 + ch); + lnz = left_mb->nz_ >> (4 + ch); + for (y = 0; y < 2; ++y) { + int l = lnz & 1; + for (x = 0; x < 2; ++x) { + const int ctx = l + (tnz & 1); + const int nz = GetCoeffs(token_br, bands[2], ctx, q->uv_mat_, 0, dst); + l = (nz > 0); + tnz = (tnz >> 1) | (l << 3); + nz_coeffs = NzCodeBits(nz_coeffs, nz, dst[0] != 0); + dst += 16; + } + tnz >>= 2; + lnz = (lnz >> 1) | (l << 5); + } + // Note: we don't really need the per-4x4 details for U/V blocks. + non_zero_uv |= nz_coeffs << (4 * ch); + out_t_nz |= (tnz << 4) << ch; + out_l_nz |= (lnz & 0xf0) << ch; + } + mb->nz_ = out_t_nz; + left_mb->nz_ = out_l_nz; + + block->non_zero_y_ = non_zero_y; + block->non_zero_uv_ = non_zero_uv; + + // We look at the mode-code of each block and check if some blocks have less + // than three non-zero coeffs (code < 2). This is to avoid dithering flat and + // empty blocks. + block->dither_ = (non_zero_uv & 0xaaaa) ? 0 : q->dither_; + + return !(non_zero_y | non_zero_uv); // will be used for further optimization +} + +//------------------------------------------------------------------------------ +// Main loop + +int VP8DecodeMB(VP8Decoder* const dec, VP8BitReader* const token_br) { + VP8MB* const left = dec->mb_info_ - 1; + VP8MB* const mb = dec->mb_info_ + dec->mb_x_; + VP8MBData* const block = dec->mb_data_ + dec->mb_x_; + int skip = dec->use_skip_proba_ ? block->skip_ : 0; + + if (!skip) { + skip = ParseResiduals(dec, mb, token_br); + } else { + left->nz_ = mb->nz_ = 0; + if (!block->is_i4x4_) { + left->nz_dc_ = mb->nz_dc_ = 0; + } + block->non_zero_y_ = 0; + block->non_zero_uv_ = 0; + block->dither_ = 0; + } + + if (dec->filter_type_ > 0) { // store filter info + VP8FInfo* const finfo = dec->f_info_ + dec->mb_x_; + *finfo = dec->fstrengths_[block->segment_][block->is_i4x4_]; + finfo->f_inner_ |= !skip; + } + + return !token_br->eof_; +} + +void VP8InitScanline(VP8Decoder* const dec) { + VP8MB* const left = dec->mb_info_ - 1; + left->nz_ = 0; + left->nz_dc_ = 0; + memset(dec->intra_l_, B_DC_PRED, sizeof(dec->intra_l_)); + dec->mb_x_ = 0; +} + +static int ParseFrame(VP8Decoder* const dec, VP8Io* io) { + for (dec->mb_y_ = 0; dec->mb_y_ < dec->br_mb_y_; ++dec->mb_y_) { + // Parse bitstream for this row. + VP8BitReader* const token_br = + &dec->parts_[dec->mb_y_ & dec->num_parts_minus_one_]; + if (!VP8ParseIntraModeRow(&dec->br_, dec)) { + return VP8SetError(dec, VP8_STATUS_NOT_ENOUGH_DATA, + "Premature end-of-partition0 encountered."); + } + for (; dec->mb_x_ < dec->mb_w_; ++dec->mb_x_) { + if (!VP8DecodeMB(dec, token_br)) { + return VP8SetError(dec, VP8_STATUS_NOT_ENOUGH_DATA, + "Premature end-of-file encountered."); + } + } + VP8InitScanline(dec); // Prepare for next scanline + + // Reconstruct, filter and emit the row. + if (!VP8ProcessRow(dec, io)) { + return VP8SetError(dec, VP8_STATUS_USER_ABORT, "Output aborted."); + } + } + if (dec->mt_method_ > 0) { + if (!WebPGetWorkerInterface()->Sync(&dec->worker_)) return 0; + } + + return 1; +} + +// Main entry point +int VP8Decode(VP8Decoder* const dec, VP8Io* const io) { + int ok = 0; + if (dec == NULL) { + return 0; + } + if (io == NULL) { + return VP8SetError(dec, VP8_STATUS_INVALID_PARAM, + "NULL VP8Io parameter in VP8Decode()."); + } + + if (!dec->ready_) { + if (!VP8GetHeaders(dec, io)) { + return 0; + } + } + assert(dec->ready_); + + // Finish setting up the decoding parameter. Will call io->setup(). + ok = (VP8EnterCritical(dec, io) == VP8_STATUS_OK); + if (ok) { // good to go. + // Will allocate memory and prepare everything. + if (ok) ok = VP8InitFrame(dec, io); + + // Main decoding loop + if (ok) ok = ParseFrame(dec, io); + + // Exit. + ok &= VP8ExitCritical(dec, io); + } + + if (!ok) { + VP8Clear(dec); + return 0; + } + + dec->ready_ = 0; + return ok; +} + +void VP8Clear(VP8Decoder* const dec) { + if (dec == NULL) { + return; + } + WebPGetWorkerInterface()->End(&dec->worker_); + WebPDeallocateAlphaMemory(dec); + WebPSafeFree(dec->mem_); + dec->mem_ = NULL; + dec->mem_size_ = 0; + memset(&dec->br_, 0, sizeof(dec->br_)); + dec->ready_ = 0; +} + +//------------------------------------------------------------------------------ diff --git a/thirdparty/libwebp/src/dec/vp8_dec.h b/thirdparty/libwebp/src/dec/vp8_dec.h new file mode 100644 index 0000000000..ca85b340cf --- /dev/null +++ b/thirdparty/libwebp/src/dec/vp8_dec.h @@ -0,0 +1,185 @@ +// Copyright 2010 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. +// ----------------------------------------------------------------------------- +// +// Low-level API for VP8 decoder +// +// Author: Skal (pascal.massimino@gmail.com) + +#ifndef WEBP_DEC_VP8_DEC_H_ +#define WEBP_DEC_VP8_DEC_H_ + +#include "src/webp/decode.h" + +#ifdef __cplusplus +extern "C" { +#endif + +//------------------------------------------------------------------------------ +// Lower-level API +// +// These functions provide fine-grained control of the decoding process. +// The call flow should resemble: +// +// VP8Io io; +// VP8InitIo(&io); +// io.data = data; +// io.data_size = size; +// /* customize io's functions (setup()/put()/teardown()) if needed. */ +// +// VP8Decoder* dec = VP8New(); +// int ok = VP8Decode(dec, &io); +// if (!ok) printf("Error: %s\n", VP8StatusMessage(dec)); +// VP8Delete(dec); +// return ok; + +// Input / Output +typedef struct VP8Io VP8Io; +typedef int (*VP8IoPutHook)(const VP8Io* io); +typedef int (*VP8IoSetupHook)(VP8Io* io); +typedef void (*VP8IoTeardownHook)(const VP8Io* io); + +struct VP8Io { + // set by VP8GetHeaders() + int width, height; // picture dimensions, in pixels (invariable). + // These are the original, uncropped dimensions. + // The actual area passed to put() is stored + // in mb_w / mb_h fields. + + // set before calling put() + int mb_y; // position of the current rows (in pixels) + int mb_w; // number of columns in the sample + int mb_h; // number of rows in the sample + const uint8_t* y, *u, *v; // rows to copy (in yuv420 format) + int y_stride; // row stride for luma + int uv_stride; // row stride for chroma + + void* opaque; // user data + + // called when fresh samples are available. Currently, samples are in + // YUV420 format, and can be up to width x 24 in size (depending on the + // in-loop filtering level, e.g.). Should return false in case of error + // or abort request. The actual size of the area to update is mb_w x mb_h + // in size, taking cropping into account. + VP8IoPutHook put; + + // called just before starting to decode the blocks. + // Must return false in case of setup error, true otherwise. If false is + // returned, teardown() will NOT be called. But if the setup succeeded + // and true is returned, then teardown() will always be called afterward. + VP8IoSetupHook setup; + + // Called just after block decoding is finished (or when an error occurred + // during put()). Is NOT called if setup() failed. + VP8IoTeardownHook teardown; + + // this is a recommendation for the user-side yuv->rgb converter. This flag + // is set when calling setup() hook and can be overwritten by it. It then + // can be taken into consideration during the put() method. + int fancy_upsampling; + + // Input buffer. + size_t data_size; + const uint8_t* data; + + // If true, in-loop filtering will not be performed even if present in the + // bitstream. Switching off filtering may speed up decoding at the expense + // of more visible blocking. Note that output will also be non-compliant + // with the VP8 specifications. + int bypass_filtering; + + // Cropping parameters. + int use_cropping; + int crop_left, crop_right, crop_top, crop_bottom; + + // Scaling parameters. + int use_scaling; + int scaled_width, scaled_height; + + // If non NULL, pointer to the alpha data (if present) corresponding to the + // start of the current row (That is: it is pre-offset by mb_y and takes + // cropping into account). + const uint8_t* a; +}; + +// Internal, version-checked, entry point +int VP8InitIoInternal(VP8Io* const, int); + +// Set the custom IO function pointers and user-data. The setter for IO hooks +// should be called before initiating incremental decoding. Returns true if +// WebPIDecoder object is successfully modified, false otherwise. +int WebPISetIOHooks(WebPIDecoder* const idec, + VP8IoPutHook put, + VP8IoSetupHook setup, + VP8IoTeardownHook teardown, + void* user_data); + +// Main decoding object. This is an opaque structure. +typedef struct VP8Decoder VP8Decoder; + +// Create a new decoder object. +VP8Decoder* VP8New(void); + +// Must be called to make sure 'io' is initialized properly. +// Returns false in case of version mismatch. Upon such failure, no other +// decoding function should be called (VP8Decode, VP8GetHeaders, ...) +static WEBP_INLINE int VP8InitIo(VP8Io* const io) { + return VP8InitIoInternal(io, WEBP_DECODER_ABI_VERSION); +} + +// Decode the VP8 frame header. Returns true if ok. +// Note: 'io->data' must be pointing to the start of the VP8 frame header. +int VP8GetHeaders(VP8Decoder* const dec, VP8Io* const io); + +// Decode a picture. Will call VP8GetHeaders() if it wasn't done already. +// Returns false in case of error. +int VP8Decode(VP8Decoder* const dec, VP8Io* const io); + +// Return current status of the decoder: +VP8StatusCode VP8Status(VP8Decoder* const dec); + +// return readable string corresponding to the last status. +const char* VP8StatusMessage(VP8Decoder* const dec); + +// Resets the decoder in its initial state, reclaiming memory. +// Not a mandatory call between calls to VP8Decode(). +void VP8Clear(VP8Decoder* const dec); + +// Destroy the decoder object. +void VP8Delete(VP8Decoder* const dec); + +//------------------------------------------------------------------------------ +// Miscellaneous VP8/VP8L bitstream probing functions. + +// Returns true if the next 3 bytes in data contain the VP8 signature. +WEBP_EXTERN int VP8CheckSignature(const uint8_t* const data, size_t data_size); + +// Validates the VP8 data-header and retrieves basic header information viz +// width and height. Returns 0 in case of formatting error. *width/*height +// can be passed NULL. +WEBP_EXTERN int VP8GetInfo( + const uint8_t* data, + size_t data_size, // data available so far + size_t chunk_size, // total data size expected in the chunk + int* const width, int* const height); + +// Returns true if the next byte(s) in data is a VP8L signature. +WEBP_EXTERN int VP8LCheckSignature(const uint8_t* const data, size_t size); + +// Validates the VP8L data-header and retrieves basic header information viz +// width, height and alpha. Returns 0 in case of formatting error. +// width/height/has_alpha can be passed NULL. +WEBP_EXTERN int VP8LGetInfo( + const uint8_t* data, size_t data_size, // data available so far + int* const width, int* const height, int* const has_alpha); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif /* WEBP_DEC_VP8_DEC_H_ */ diff --git a/thirdparty/libwebp/src/dec/vp8i_dec.h b/thirdparty/libwebp/src/dec/vp8i_dec.h new file mode 100644 index 0000000000..28244d9d7a --- /dev/null +++ b/thirdparty/libwebp/src/dec/vp8i_dec.h @@ -0,0 +1,319 @@ +// Copyright 2010 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. +// ----------------------------------------------------------------------------- +// +// VP8 decoder: internal header. +// +// Author: Skal (pascal.massimino@gmail.com) + +#ifndef WEBP_DEC_VP8I_DEC_H_ +#define WEBP_DEC_VP8I_DEC_H_ + +#include <string.h> // for memcpy() +#include "src/dec/common_dec.h" +#include "src/dec/vp8li_dec.h" +#include "src/utils/bit_reader_utils.h" +#include "src/utils/random_utils.h" +#include "src/utils/thread_utils.h" +#include "src/dsp/dsp.h" + +#ifdef __cplusplus +extern "C" { +#endif + +//------------------------------------------------------------------------------ +// Various defines and enums + +// version numbers +#define DEC_MAJ_VERSION 0 +#define DEC_MIN_VERSION 6 +#define DEC_REV_VERSION 1 + +// YUV-cache parameters. Cache is 32-bytes wide (= one cacheline). +// Constraints are: We need to store one 16x16 block of luma samples (y), +// and two 8x8 chroma blocks (u/v). These are better be 16-bytes aligned, +// in order to be SIMD-friendly. We also need to store the top, left and +// top-left samples (from previously decoded blocks), along with four +// extra top-right samples for luma (intra4x4 prediction only). +// One possible layout is, using 32 * (17 + 9) bytes: +// +// .+------ <- only 1 pixel high +// .|yyyyt. +// .|yyyyt. +// .|yyyyt. +// .|yyyy.. +// .+--.+-- <- only 1 pixel high +// .|uu.|vv +// .|uu.|vv +// +// Every character is a 4x4 block, with legend: +// '.' = unused +// 'y' = y-samples 'u' = u-samples 'v' = u-samples +// '|' = left sample, '-' = top sample, '+' = top-left sample +// 't' = extra top-right sample for 4x4 modes +#define YUV_SIZE (BPS * 17 + BPS * 9) +#define Y_OFF (BPS * 1 + 8) +#define U_OFF (Y_OFF + BPS * 16 + BPS) +#define V_OFF (U_OFF + 16) + +// minimal width under which lossy multi-threading is always disabled +#define MIN_WIDTH_FOR_THREADS 512 + +//------------------------------------------------------------------------------ +// Headers + +typedef struct { + uint8_t key_frame_; + uint8_t profile_; + uint8_t show_; + uint32_t partition_length_; +} VP8FrameHeader; + +typedef struct { + uint16_t width_; + uint16_t height_; + uint8_t xscale_; + uint8_t yscale_; + uint8_t colorspace_; // 0 = YCbCr + uint8_t clamp_type_; +} VP8PictureHeader; + +// segment features +typedef struct { + int use_segment_; + int update_map_; // whether to update the segment map or not + int absolute_delta_; // absolute or delta values for quantizer and filter + int8_t quantizer_[NUM_MB_SEGMENTS]; // quantization changes + int8_t filter_strength_[NUM_MB_SEGMENTS]; // filter strength for segments +} VP8SegmentHeader; + +// probas associated to one of the contexts +typedef uint8_t VP8ProbaArray[NUM_PROBAS]; + +typedef struct { // all the probas associated to one band + VP8ProbaArray probas_[NUM_CTX]; +} VP8BandProbas; + +// Struct collecting all frame-persistent probabilities. +typedef struct { + uint8_t segments_[MB_FEATURE_TREE_PROBS]; + // Type: 0:Intra16-AC 1:Intra16-DC 2:Chroma 3:Intra4 + VP8BandProbas bands_[NUM_TYPES][NUM_BANDS]; + const VP8BandProbas* bands_ptr_[NUM_TYPES][16 + 1]; +} VP8Proba; + +// Filter parameters +typedef struct { + int simple_; // 0=complex, 1=simple + int level_; // [0..63] + int sharpness_; // [0..7] + int use_lf_delta_; + int ref_lf_delta_[NUM_REF_LF_DELTAS]; + int mode_lf_delta_[NUM_MODE_LF_DELTAS]; +} VP8FilterHeader; + +//------------------------------------------------------------------------------ +// Informations about the macroblocks. + +typedef struct { // filter specs + uint8_t f_limit_; // filter limit in [3..189], or 0 if no filtering + uint8_t f_ilevel_; // inner limit in [1..63] + uint8_t f_inner_; // do inner filtering? + uint8_t hev_thresh_; // high edge variance threshold in [0..2] +} VP8FInfo; + +typedef struct { // Top/Left Contexts used for syntax-parsing + uint8_t nz_; // non-zero AC/DC coeffs (4bit for luma + 4bit for chroma) + uint8_t nz_dc_; // non-zero DC coeff (1bit) +} VP8MB; + +// Dequantization matrices +typedef int quant_t[2]; // [DC / AC]. Can be 'uint16_t[2]' too (~slower). +typedef struct { + quant_t y1_mat_, y2_mat_, uv_mat_; + + int uv_quant_; // U/V quantizer value + int dither_; // dithering amplitude (0 = off, max=255) +} VP8QuantMatrix; + +// Data needed to reconstruct a macroblock +typedef struct { + int16_t coeffs_[384]; // 384 coeffs = (16+4+4) * 4*4 + uint8_t is_i4x4_; // true if intra4x4 + uint8_t imodes_[16]; // one 16x16 mode (#0) or sixteen 4x4 modes + uint8_t uvmode_; // chroma prediction mode + // bit-wise info about the content of each sub-4x4 blocks (in decoding order). + // Each of the 4x4 blocks for y/u/v is associated with a 2b code according to: + // code=0 -> no coefficient + // code=1 -> only DC + // code=2 -> first three coefficients are non-zero + // code=3 -> more than three coefficients are non-zero + // This allows to call specialized transform functions. + uint32_t non_zero_y_; + uint32_t non_zero_uv_; + uint8_t dither_; // local dithering strength (deduced from non_zero_*) + uint8_t skip_; + uint8_t segment_; +} VP8MBData; + +// Persistent information needed by the parallel processing +typedef struct { + int id_; // cache row to process (in [0..2]) + int mb_y_; // macroblock position of the row + int filter_row_; // true if row-filtering is needed + VP8FInfo* f_info_; // filter strengths (swapped with dec->f_info_) + VP8MBData* mb_data_; // reconstruction data (swapped with dec->mb_data_) + VP8Io io_; // copy of the VP8Io to pass to put() +} VP8ThreadContext; + +// Saved top samples, per macroblock. Fits into a cache-line. +typedef struct { + uint8_t y[16], u[8], v[8]; +} VP8TopSamples; + +//------------------------------------------------------------------------------ +// VP8Decoder: the main opaque structure handed over to user + +struct VP8Decoder { + VP8StatusCode status_; + int ready_; // true if ready to decode a picture with VP8Decode() + const char* error_msg_; // set when status_ is not OK. + + // Main data source + VP8BitReader br_; + + // headers + VP8FrameHeader frm_hdr_; + VP8PictureHeader pic_hdr_; + VP8FilterHeader filter_hdr_; + VP8SegmentHeader segment_hdr_; + + // Worker + WebPWorker worker_; + int mt_method_; // multi-thread method: 0=off, 1=[parse+recon][filter] + // 2=[parse][recon+filter] + int cache_id_; // current cache row + int num_caches_; // number of cached rows of 16 pixels (1, 2 or 3) + VP8ThreadContext thread_ctx_; // Thread context + + // dimension, in macroblock units. + int mb_w_, mb_h_; + + // Macroblock to process/filter, depending on cropping and filter_type. + int tl_mb_x_, tl_mb_y_; // top-left MB that must be in-loop filtered + int br_mb_x_, br_mb_y_; // last bottom-right MB that must be decoded + + // number of partitions minus one. + uint32_t num_parts_minus_one_; + // per-partition boolean decoders. + VP8BitReader parts_[MAX_NUM_PARTITIONS]; + + // Dithering strength, deduced from decoding options + int dither_; // whether to use dithering or not + VP8Random dithering_rg_; // random generator for dithering + + // dequantization (one set of DC/AC dequant factor per segment) + VP8QuantMatrix dqm_[NUM_MB_SEGMENTS]; + + // probabilities + VP8Proba proba_; + int use_skip_proba_; + uint8_t skip_p_; + + // Boundary data cache and persistent buffers. + uint8_t* intra_t_; // top intra modes values: 4 * mb_w_ + uint8_t intra_l_[4]; // left intra modes values + + VP8TopSamples* yuv_t_; // top y/u/v samples + + VP8MB* mb_info_; // contextual macroblock info (mb_w_ + 1) + VP8FInfo* f_info_; // filter strength info + uint8_t* yuv_b_; // main block for Y/U/V (size = YUV_SIZE) + + uint8_t* cache_y_; // macroblock row for storing unfiltered samples + uint8_t* cache_u_; + uint8_t* cache_v_; + int cache_y_stride_; + int cache_uv_stride_; + + // main memory chunk for the above data. Persistent. + void* mem_; + size_t mem_size_; + + // Per macroblock non-persistent infos. + int mb_x_, mb_y_; // current position, in macroblock units + VP8MBData* mb_data_; // parsed reconstruction data + + // Filtering side-info + int filter_type_; // 0=off, 1=simple, 2=complex + VP8FInfo fstrengths_[NUM_MB_SEGMENTS][2]; // precalculated per-segment/type + + // Alpha + struct ALPHDecoder* alph_dec_; // alpha-plane decoder object + const uint8_t* alpha_data_; // compressed alpha data (if present) + size_t alpha_data_size_; + int is_alpha_decoded_; // true if alpha_data_ is decoded in alpha_plane_ + uint8_t* alpha_plane_mem_; // memory allocated for alpha_plane_ + uint8_t* alpha_plane_; // output. Persistent, contains the whole data. + const uint8_t* alpha_prev_line_; // last decoded alpha row (or NULL) + int alpha_dithering_; // derived from decoding options (0=off, 100=full) +}; + +//------------------------------------------------------------------------------ +// internal functions. Not public. + +// in vp8.c +int VP8SetError(VP8Decoder* const dec, + VP8StatusCode error, const char* const msg); + +// in tree.c +void VP8ResetProba(VP8Proba* const proba); +void VP8ParseProba(VP8BitReader* const br, VP8Decoder* const dec); +// parses one row of intra mode data in partition 0, returns !eof +int VP8ParseIntraModeRow(VP8BitReader* const br, VP8Decoder* const dec); + +// in quant.c +void VP8ParseQuant(VP8Decoder* const dec); + +// in frame.c +int VP8InitFrame(VP8Decoder* const dec, VP8Io* const io); +// Call io->setup() and finish setting up scan parameters. +// After this call returns, one must always call VP8ExitCritical() with the +// same parameters. Both functions should be used in pair. Returns VP8_STATUS_OK +// if ok, otherwise sets and returns the error status on *dec. +VP8StatusCode VP8EnterCritical(VP8Decoder* const dec, VP8Io* const io); +// Must always be called in pair with VP8EnterCritical(). +// Returns false in case of error. +int VP8ExitCritical(VP8Decoder* const dec, VP8Io* const io); +// Return the multi-threading method to use (0=off), depending +// on options and bitstream size. Only for lossy decoding. +int VP8GetThreadMethod(const WebPDecoderOptions* const options, + const WebPHeaderStructure* const headers, + int width, int height); +// Initialize dithering post-process if needed. +void VP8InitDithering(const WebPDecoderOptions* const options, + VP8Decoder* const dec); +// Process the last decoded row (filtering + output). +int VP8ProcessRow(VP8Decoder* const dec, VP8Io* const io); +// To be called at the start of a new scanline, to initialize predictors. +void VP8InitScanline(VP8Decoder* const dec); +// Decode one macroblock. Returns false if there is not enough data. +int VP8DecodeMB(VP8Decoder* const dec, VP8BitReader* const token_br); + +// in alpha.c +const uint8_t* VP8DecompressAlphaRows(VP8Decoder* const dec, + const VP8Io* const io, + int row, int num_rows); + +//------------------------------------------------------------------------------ + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif /* WEBP_DEC_VP8I_DEC_H_ */ diff --git a/thirdparty/libwebp/src/dec/vp8l_dec.c b/thirdparty/libwebp/src/dec/vp8l_dec.c new file mode 100644 index 0000000000..42ea3b5e4c --- /dev/null +++ b/thirdparty/libwebp/src/dec/vp8l_dec.c @@ -0,0 +1,1689 @@ +// Copyright 2012 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. +// ----------------------------------------------------------------------------- +// +// main entry for the decoder +// +// Authors: Vikas Arora (vikaas.arora@gmail.com) +// Jyrki Alakuijala (jyrki@google.com) + +#include <stdlib.h> + +#include "src/dec/alphai_dec.h" +#include "src/dec/vp8li_dec.h" +#include "src/dsp/dsp.h" +#include "src/dsp/lossless.h" +#include "src/dsp/lossless_common.h" +#include "src/dsp/yuv.h" +#include "src/utils/endian_inl_utils.h" +#include "src/utils/huffman_utils.h" +#include "src/utils/utils.h" + +#define NUM_ARGB_CACHE_ROWS 16 + +static const int kCodeLengthLiterals = 16; +static const int kCodeLengthRepeatCode = 16; +static const uint8_t kCodeLengthExtraBits[3] = { 2, 3, 7 }; +static const uint8_t kCodeLengthRepeatOffsets[3] = { 3, 3, 11 }; + +// ----------------------------------------------------------------------------- +// Five Huffman codes are used at each meta code: +// 1. green + length prefix codes + color cache codes, +// 2. alpha, +// 3. red, +// 4. blue, and, +// 5. distance prefix codes. +typedef enum { + GREEN = 0, + RED = 1, + BLUE = 2, + ALPHA = 3, + DIST = 4 +} HuffIndex; + +static const uint16_t kAlphabetSize[HUFFMAN_CODES_PER_META_CODE] = { + NUM_LITERAL_CODES + NUM_LENGTH_CODES, + NUM_LITERAL_CODES, NUM_LITERAL_CODES, NUM_LITERAL_CODES, + NUM_DISTANCE_CODES +}; + +static const uint8_t kLiteralMap[HUFFMAN_CODES_PER_META_CODE] = { + 0, 1, 1, 1, 0 +}; + +#define NUM_CODE_LENGTH_CODES 19 +static const uint8_t kCodeLengthCodeOrder[NUM_CODE_LENGTH_CODES] = { + 17, 18, 0, 1, 2, 3, 4, 5, 16, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 +}; + +#define CODE_TO_PLANE_CODES 120 +static const uint8_t kCodeToPlane[CODE_TO_PLANE_CODES] = { + 0x18, 0x07, 0x17, 0x19, 0x28, 0x06, 0x27, 0x29, 0x16, 0x1a, + 0x26, 0x2a, 0x38, 0x05, 0x37, 0x39, 0x15, 0x1b, 0x36, 0x3a, + 0x25, 0x2b, 0x48, 0x04, 0x47, 0x49, 0x14, 0x1c, 0x35, 0x3b, + 0x46, 0x4a, 0x24, 0x2c, 0x58, 0x45, 0x4b, 0x34, 0x3c, 0x03, + 0x57, 0x59, 0x13, 0x1d, 0x56, 0x5a, 0x23, 0x2d, 0x44, 0x4c, + 0x55, 0x5b, 0x33, 0x3d, 0x68, 0x02, 0x67, 0x69, 0x12, 0x1e, + 0x66, 0x6a, 0x22, 0x2e, 0x54, 0x5c, 0x43, 0x4d, 0x65, 0x6b, + 0x32, 0x3e, 0x78, 0x01, 0x77, 0x79, 0x53, 0x5d, 0x11, 0x1f, + 0x64, 0x6c, 0x42, 0x4e, 0x76, 0x7a, 0x21, 0x2f, 0x75, 0x7b, + 0x31, 0x3f, 0x63, 0x6d, 0x52, 0x5e, 0x00, 0x74, 0x7c, 0x41, + 0x4f, 0x10, 0x20, 0x62, 0x6e, 0x30, 0x73, 0x7d, 0x51, 0x5f, + 0x40, 0x72, 0x7e, 0x61, 0x6f, 0x50, 0x71, 0x7f, 0x60, 0x70 +}; + +// Memory needed for lookup tables of one Huffman tree group. Red, blue, alpha +// and distance alphabets are constant (256 for red, blue and alpha, 40 for +// distance) and lookup table sizes for them in worst case are 630 and 410 +// respectively. Size of green alphabet depends on color cache size and is equal +// to 256 (green component values) + 24 (length prefix values) +// + color_cache_size (between 0 and 2048). +// All values computed for 8-bit first level lookup with Mark Adler's tool: +// http://www.hdfgroup.org/ftp/lib-external/zlib/zlib-1.2.5/examples/enough.c +#define FIXED_TABLE_SIZE (630 * 3 + 410) +static const uint16_t kTableSize[12] = { + FIXED_TABLE_SIZE + 654, + FIXED_TABLE_SIZE + 656, + FIXED_TABLE_SIZE + 658, + FIXED_TABLE_SIZE + 662, + FIXED_TABLE_SIZE + 670, + FIXED_TABLE_SIZE + 686, + FIXED_TABLE_SIZE + 718, + FIXED_TABLE_SIZE + 782, + FIXED_TABLE_SIZE + 912, + FIXED_TABLE_SIZE + 1168, + FIXED_TABLE_SIZE + 1680, + FIXED_TABLE_SIZE + 2704 +}; + +static int DecodeImageStream(int xsize, int ysize, + int is_level0, + VP8LDecoder* const dec, + uint32_t** const decoded_data); + +//------------------------------------------------------------------------------ + +int VP8LCheckSignature(const uint8_t* const data, size_t size) { + return (size >= VP8L_FRAME_HEADER_SIZE && + data[0] == VP8L_MAGIC_BYTE && + (data[4] >> 5) == 0); // version +} + +static int ReadImageInfo(VP8LBitReader* const br, + int* const width, int* const height, + int* const has_alpha) { + if (VP8LReadBits(br, 8) != VP8L_MAGIC_BYTE) return 0; + *width = VP8LReadBits(br, VP8L_IMAGE_SIZE_BITS) + 1; + *height = VP8LReadBits(br, VP8L_IMAGE_SIZE_BITS) + 1; + *has_alpha = VP8LReadBits(br, 1); + if (VP8LReadBits(br, VP8L_VERSION_BITS) != 0) return 0; + return !br->eos_; +} + +int VP8LGetInfo(const uint8_t* data, size_t data_size, + int* const width, int* const height, int* const has_alpha) { + if (data == NULL || data_size < VP8L_FRAME_HEADER_SIZE) { + return 0; // not enough data + } else if (!VP8LCheckSignature(data, data_size)) { + return 0; // bad signature + } else { + int w, h, a; + VP8LBitReader br; + VP8LInitBitReader(&br, data, data_size); + if (!ReadImageInfo(&br, &w, &h, &a)) { + return 0; + } + if (width != NULL) *width = w; + if (height != NULL) *height = h; + if (has_alpha != NULL) *has_alpha = a; + return 1; + } +} + +//------------------------------------------------------------------------------ + +static WEBP_INLINE int GetCopyDistance(int distance_symbol, + VP8LBitReader* const br) { + int extra_bits, offset; + if (distance_symbol < 4) { + return distance_symbol + 1; + } + extra_bits = (distance_symbol - 2) >> 1; + offset = (2 + (distance_symbol & 1)) << extra_bits; + return offset + VP8LReadBits(br, extra_bits) + 1; +} + +static WEBP_INLINE int GetCopyLength(int length_symbol, + VP8LBitReader* const br) { + // Length and distance prefixes are encoded the same way. + return GetCopyDistance(length_symbol, br); +} + +static WEBP_INLINE int PlaneCodeToDistance(int xsize, int plane_code) { + if (plane_code > CODE_TO_PLANE_CODES) { + return plane_code - CODE_TO_PLANE_CODES; + } else { + const int dist_code = kCodeToPlane[plane_code - 1]; + const int yoffset = dist_code >> 4; + const int xoffset = 8 - (dist_code & 0xf); + const int dist = yoffset * xsize + xoffset; + return (dist >= 1) ? dist : 1; // dist<1 can happen if xsize is very small + } +} + +//------------------------------------------------------------------------------ +// Decodes the next Huffman code from bit-stream. +// FillBitWindow(br) needs to be called at minimum every second call +// to ReadSymbol, in order to pre-fetch enough bits. +static WEBP_INLINE int ReadSymbol(const HuffmanCode* table, + VP8LBitReader* const br) { + int nbits; + uint32_t val = VP8LPrefetchBits(br); + table += val & HUFFMAN_TABLE_MASK; + nbits = table->bits - HUFFMAN_TABLE_BITS; + if (nbits > 0) { + VP8LSetBitPos(br, br->bit_pos_ + HUFFMAN_TABLE_BITS); + val = VP8LPrefetchBits(br); + table += table->value; + table += val & ((1 << nbits) - 1); + } + VP8LSetBitPos(br, br->bit_pos_ + table->bits); + return table->value; +} + +// Reads packed symbol depending on GREEN channel +#define BITS_SPECIAL_MARKER 0x100 // something large enough (and a bit-mask) +#define PACKED_NON_LITERAL_CODE 0 // must be < NUM_LITERAL_CODES +static WEBP_INLINE int ReadPackedSymbols(const HTreeGroup* group, + VP8LBitReader* const br, + uint32_t* const dst) { + const uint32_t val = VP8LPrefetchBits(br) & (HUFFMAN_PACKED_TABLE_SIZE - 1); + const HuffmanCode32 code = group->packed_table[val]; + assert(group->use_packed_table); + if (code.bits < BITS_SPECIAL_MARKER) { + VP8LSetBitPos(br, br->bit_pos_ + code.bits); + *dst = code.value; + return PACKED_NON_LITERAL_CODE; + } else { + VP8LSetBitPos(br, br->bit_pos_ + code.bits - BITS_SPECIAL_MARKER); + assert(code.value >= NUM_LITERAL_CODES); + return code.value; + } +} + +static int AccumulateHCode(HuffmanCode hcode, int shift, + HuffmanCode32* const huff) { + huff->bits += hcode.bits; + huff->value |= (uint32_t)hcode.value << shift; + assert(huff->bits <= HUFFMAN_TABLE_BITS); + return hcode.bits; +} + +static void BuildPackedTable(HTreeGroup* const htree_group) { + uint32_t code; + for (code = 0; code < HUFFMAN_PACKED_TABLE_SIZE; ++code) { + uint32_t bits = code; + HuffmanCode32* const huff = &htree_group->packed_table[bits]; + HuffmanCode hcode = htree_group->htrees[GREEN][bits]; + if (hcode.value >= NUM_LITERAL_CODES) { + huff->bits = hcode.bits + BITS_SPECIAL_MARKER; + huff->value = hcode.value; + } else { + huff->bits = 0; + huff->value = 0; + bits >>= AccumulateHCode(hcode, 8, huff); + bits >>= AccumulateHCode(htree_group->htrees[RED][bits], 16, huff); + bits >>= AccumulateHCode(htree_group->htrees[BLUE][bits], 0, huff); + bits >>= AccumulateHCode(htree_group->htrees[ALPHA][bits], 24, huff); + (void)bits; + } + } +} + +static int ReadHuffmanCodeLengths( + VP8LDecoder* const dec, const int* const code_length_code_lengths, + int num_symbols, int* const code_lengths) { + int ok = 0; + VP8LBitReader* const br = &dec->br_; + int symbol; + int max_symbol; + int prev_code_len = DEFAULT_CODE_LENGTH; + HuffmanCode table[1 << LENGTHS_TABLE_BITS]; + + if (!VP8LBuildHuffmanTable(table, LENGTHS_TABLE_BITS, + code_length_code_lengths, + NUM_CODE_LENGTH_CODES)) { + goto End; + } + + if (VP8LReadBits(br, 1)) { // use length + const int length_nbits = 2 + 2 * VP8LReadBits(br, 3); + max_symbol = 2 + VP8LReadBits(br, length_nbits); + if (max_symbol > num_symbols) { + goto End; + } + } else { + max_symbol = num_symbols; + } + + symbol = 0; + while (symbol < num_symbols) { + const HuffmanCode* p; + int code_len; + if (max_symbol-- == 0) break; + VP8LFillBitWindow(br); + p = &table[VP8LPrefetchBits(br) & LENGTHS_TABLE_MASK]; + VP8LSetBitPos(br, br->bit_pos_ + p->bits); + code_len = p->value; + if (code_len < kCodeLengthLiterals) { + code_lengths[symbol++] = code_len; + if (code_len != 0) prev_code_len = code_len; + } else { + const int use_prev = (code_len == kCodeLengthRepeatCode); + const int slot = code_len - kCodeLengthLiterals; + const int extra_bits = kCodeLengthExtraBits[slot]; + const int repeat_offset = kCodeLengthRepeatOffsets[slot]; + int repeat = VP8LReadBits(br, extra_bits) + repeat_offset; + if (symbol + repeat > num_symbols) { + goto End; + } else { + const int length = use_prev ? prev_code_len : 0; + while (repeat-- > 0) code_lengths[symbol++] = length; + } + } + } + ok = 1; + + End: + if (!ok) dec->status_ = VP8_STATUS_BITSTREAM_ERROR; + return ok; +} + +// 'code_lengths' is pre-allocated temporary buffer, used for creating Huffman +// tree. +static int ReadHuffmanCode(int alphabet_size, VP8LDecoder* const dec, + int* const code_lengths, HuffmanCode* const table) { + int ok = 0; + int size = 0; + VP8LBitReader* const br = &dec->br_; + const int simple_code = VP8LReadBits(br, 1); + + memset(code_lengths, 0, alphabet_size * sizeof(*code_lengths)); + + if (simple_code) { // Read symbols, codes & code lengths directly. + const int num_symbols = VP8LReadBits(br, 1) + 1; + const int first_symbol_len_code = VP8LReadBits(br, 1); + // The first code is either 1 bit or 8 bit code. + int symbol = VP8LReadBits(br, (first_symbol_len_code == 0) ? 1 : 8); + code_lengths[symbol] = 1; + // The second code (if present), is always 8 bit long. + if (num_symbols == 2) { + symbol = VP8LReadBits(br, 8); + code_lengths[symbol] = 1; + } + ok = 1; + } else { // Decode Huffman-coded code lengths. + int i; + int code_length_code_lengths[NUM_CODE_LENGTH_CODES] = { 0 }; + const int num_codes = VP8LReadBits(br, 4) + 4; + if (num_codes > NUM_CODE_LENGTH_CODES) { + dec->status_ = VP8_STATUS_BITSTREAM_ERROR; + return 0; + } + + for (i = 0; i < num_codes; ++i) { + code_length_code_lengths[kCodeLengthCodeOrder[i]] = VP8LReadBits(br, 3); + } + ok = ReadHuffmanCodeLengths(dec, code_length_code_lengths, alphabet_size, + code_lengths); + } + + ok = ok && !br->eos_; + if (ok) { + size = VP8LBuildHuffmanTable(table, HUFFMAN_TABLE_BITS, + code_lengths, alphabet_size); + } + if (!ok || size == 0) { + dec->status_ = VP8_STATUS_BITSTREAM_ERROR; + return 0; + } + return size; +} + +static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize, + int color_cache_bits, int allow_recursion) { + int i, j; + VP8LBitReader* const br = &dec->br_; + VP8LMetadata* const hdr = &dec->hdr_; + uint32_t* huffman_image = NULL; + HTreeGroup* htree_groups = NULL; + HuffmanCode* huffman_tables = NULL; + HuffmanCode* next = NULL; + int num_htree_groups = 1; + int max_alphabet_size = 0; + int* code_lengths = NULL; + const int table_size = kTableSize[color_cache_bits]; + + if (allow_recursion && VP8LReadBits(br, 1)) { + // use meta Huffman codes. + const int huffman_precision = VP8LReadBits(br, 3) + 2; + const int huffman_xsize = VP8LSubSampleSize(xsize, huffman_precision); + const int huffman_ysize = VP8LSubSampleSize(ysize, huffman_precision); + const int huffman_pixs = huffman_xsize * huffman_ysize; + if (!DecodeImageStream(huffman_xsize, huffman_ysize, 0, dec, + &huffman_image)) { + goto Error; + } + hdr->huffman_subsample_bits_ = huffman_precision; + for (i = 0; i < huffman_pixs; ++i) { + // The huffman data is stored in red and green bytes. + const int group = (huffman_image[i] >> 8) & 0xffff; + huffman_image[i] = group; + if (group >= num_htree_groups) { + num_htree_groups = group + 1; + } + } + } + + if (br->eos_) goto Error; + + // Find maximum alphabet size for the htree group. + for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; ++j) { + int alphabet_size = kAlphabetSize[j]; + if (j == 0 && color_cache_bits > 0) { + alphabet_size += 1 << color_cache_bits; + } + if (max_alphabet_size < alphabet_size) { + max_alphabet_size = alphabet_size; + } + } + + huffman_tables = (HuffmanCode*)WebPSafeMalloc(num_htree_groups * table_size, + sizeof(*huffman_tables)); + htree_groups = VP8LHtreeGroupsNew(num_htree_groups); + code_lengths = (int*)WebPSafeCalloc((uint64_t)max_alphabet_size, + sizeof(*code_lengths)); + + if (htree_groups == NULL || code_lengths == NULL || huffman_tables == NULL) { + dec->status_ = VP8_STATUS_OUT_OF_MEMORY; + goto Error; + } + + next = huffman_tables; + for (i = 0; i < num_htree_groups; ++i) { + HTreeGroup* const htree_group = &htree_groups[i]; + HuffmanCode** const htrees = htree_group->htrees; + int size; + int total_size = 0; + int is_trivial_literal = 1; + int max_bits = 0; + for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; ++j) { + int alphabet_size = kAlphabetSize[j]; + htrees[j] = next; + if (j == 0 && color_cache_bits > 0) { + alphabet_size += 1 << color_cache_bits; + } + size = ReadHuffmanCode(alphabet_size, dec, code_lengths, next); + if (size == 0) { + goto Error; + } + if (is_trivial_literal && kLiteralMap[j] == 1) { + is_trivial_literal = (next->bits == 0); + } + total_size += next->bits; + next += size; + if (j <= ALPHA) { + int local_max_bits = code_lengths[0]; + int k; + for (k = 1; k < alphabet_size; ++k) { + if (code_lengths[k] > local_max_bits) { + local_max_bits = code_lengths[k]; + } + } + max_bits += local_max_bits; + } + } + htree_group->is_trivial_literal = is_trivial_literal; + htree_group->is_trivial_code = 0; + if (is_trivial_literal) { + const int red = htrees[RED][0].value; + const int blue = htrees[BLUE][0].value; + const int alpha = htrees[ALPHA][0].value; + htree_group->literal_arb = + ((uint32_t)alpha << 24) | (red << 16) | blue; + if (total_size == 0 && htrees[GREEN][0].value < NUM_LITERAL_CODES) { + htree_group->is_trivial_code = 1; + htree_group->literal_arb |= htrees[GREEN][0].value << 8; + } + } + htree_group->use_packed_table = !htree_group->is_trivial_code && + (max_bits < HUFFMAN_PACKED_BITS); + if (htree_group->use_packed_table) BuildPackedTable(htree_group); + } + WebPSafeFree(code_lengths); + + // All OK. Finalize pointers and return. + hdr->huffman_image_ = huffman_image; + hdr->num_htree_groups_ = num_htree_groups; + hdr->htree_groups_ = htree_groups; + hdr->huffman_tables_ = huffman_tables; + return 1; + + Error: + WebPSafeFree(code_lengths); + WebPSafeFree(huffman_image); + WebPSafeFree(huffman_tables); + VP8LHtreeGroupsFree(htree_groups); + return 0; +} + +//------------------------------------------------------------------------------ +// Scaling. + +#if !defined(WEBP_REDUCE_SIZE) +static int AllocateAndInitRescaler(VP8LDecoder* const dec, VP8Io* const io) { + const int num_channels = 4; + const int in_width = io->mb_w; + const int out_width = io->scaled_width; + const int in_height = io->mb_h; + const int out_height = io->scaled_height; + const uint64_t work_size = 2 * num_channels * (uint64_t)out_width; + rescaler_t* work; // Rescaler work area. + const uint64_t scaled_data_size = (uint64_t)out_width; + uint32_t* scaled_data; // Temporary storage for scaled BGRA data. + const uint64_t memory_size = sizeof(*dec->rescaler) + + work_size * sizeof(*work) + + scaled_data_size * sizeof(*scaled_data); + uint8_t* memory = (uint8_t*)WebPSafeMalloc(memory_size, sizeof(*memory)); + if (memory == NULL) { + dec->status_ = VP8_STATUS_OUT_OF_MEMORY; + return 0; + } + assert(dec->rescaler_memory == NULL); + dec->rescaler_memory = memory; + + dec->rescaler = (WebPRescaler*)memory; + memory += sizeof(*dec->rescaler); + work = (rescaler_t*)memory; + memory += work_size * sizeof(*work); + scaled_data = (uint32_t*)memory; + + WebPRescalerInit(dec->rescaler, in_width, in_height, (uint8_t*)scaled_data, + out_width, out_height, 0, num_channels, work); + return 1; +} +#endif // WEBP_REDUCE_SIZE + +//------------------------------------------------------------------------------ +// Export to ARGB + +#if !defined(WEBP_REDUCE_SIZE) + +// We have special "export" function since we need to convert from BGRA +static int Export(WebPRescaler* const rescaler, WEBP_CSP_MODE colorspace, + int rgba_stride, uint8_t* const rgba) { + uint32_t* const src = (uint32_t*)rescaler->dst; + const int dst_width = rescaler->dst_width; + int num_lines_out = 0; + while (WebPRescalerHasPendingOutput(rescaler)) { + uint8_t* const dst = rgba + num_lines_out * rgba_stride; + WebPRescalerExportRow(rescaler); + WebPMultARGBRow(src, dst_width, 1); + VP8LConvertFromBGRA(src, dst_width, colorspace, dst); + ++num_lines_out; + } + return num_lines_out; +} + +// Emit scaled rows. +static int EmitRescaledRowsRGBA(const VP8LDecoder* const dec, + uint8_t* in, int in_stride, int mb_h, + uint8_t* const out, int out_stride) { + const WEBP_CSP_MODE colorspace = dec->output_->colorspace; + int num_lines_in = 0; + int num_lines_out = 0; + while (num_lines_in < mb_h) { + uint8_t* const row_in = in + num_lines_in * in_stride; + uint8_t* const row_out = out + num_lines_out * out_stride; + const int lines_left = mb_h - num_lines_in; + const int needed_lines = WebPRescaleNeededLines(dec->rescaler, lines_left); + int lines_imported; + assert(needed_lines > 0 && needed_lines <= lines_left); + WebPMultARGBRows(row_in, in_stride, + dec->rescaler->src_width, needed_lines, 0); + lines_imported = + WebPRescalerImport(dec->rescaler, lines_left, row_in, in_stride); + assert(lines_imported == needed_lines); + num_lines_in += lines_imported; + num_lines_out += Export(dec->rescaler, colorspace, out_stride, row_out); + } + return num_lines_out; +} + +#endif // WEBP_REDUCE_SIZE + +// Emit rows without any scaling. +static int EmitRows(WEBP_CSP_MODE colorspace, + const uint8_t* row_in, int in_stride, + int mb_w, int mb_h, + uint8_t* const out, int out_stride) { + int lines = mb_h; + uint8_t* row_out = out; + while (lines-- > 0) { + VP8LConvertFromBGRA((const uint32_t*)row_in, mb_w, colorspace, row_out); + row_in += in_stride; + row_out += out_stride; + } + return mb_h; // Num rows out == num rows in. +} + +//------------------------------------------------------------------------------ +// Export to YUVA + +static void ConvertToYUVA(const uint32_t* const src, int width, int y_pos, + const WebPDecBuffer* const output) { + const WebPYUVABuffer* const buf = &output->u.YUVA; + + // first, the luma plane + WebPConvertARGBToY(src, buf->y + y_pos * buf->y_stride, width); + + // then U/V planes + { + uint8_t* const u = buf->u + (y_pos >> 1) * buf->u_stride; + uint8_t* const v = buf->v + (y_pos >> 1) * buf->v_stride; + // even lines: store values + // odd lines: average with previous values + WebPConvertARGBToUV(src, u, v, width, !(y_pos & 1)); + } + // Lastly, store alpha if needed. + if (buf->a != NULL) { + uint8_t* const a = buf->a + y_pos * buf->a_stride; +#if defined(WORDS_BIGENDIAN) + WebPExtractAlpha((uint8_t*)src + 0, 0, width, 1, a, 0); +#else + WebPExtractAlpha((uint8_t*)src + 3, 0, width, 1, a, 0); +#endif + } +} + +static int ExportYUVA(const VP8LDecoder* const dec, int y_pos) { + WebPRescaler* const rescaler = dec->rescaler; + uint32_t* const src = (uint32_t*)rescaler->dst; + const int dst_width = rescaler->dst_width; + int num_lines_out = 0; + while (WebPRescalerHasPendingOutput(rescaler)) { + WebPRescalerExportRow(rescaler); + WebPMultARGBRow(src, dst_width, 1); + ConvertToYUVA(src, dst_width, y_pos, dec->output_); + ++y_pos; + ++num_lines_out; + } + return num_lines_out; +} + +static int EmitRescaledRowsYUVA(const VP8LDecoder* const dec, + uint8_t* in, int in_stride, int mb_h) { + int num_lines_in = 0; + int y_pos = dec->last_out_row_; + while (num_lines_in < mb_h) { + const int lines_left = mb_h - num_lines_in; + const int needed_lines = WebPRescaleNeededLines(dec->rescaler, lines_left); + int lines_imported; + WebPMultARGBRows(in, in_stride, dec->rescaler->src_width, needed_lines, 0); + lines_imported = + WebPRescalerImport(dec->rescaler, lines_left, in, in_stride); + assert(lines_imported == needed_lines); + num_lines_in += lines_imported; + in += needed_lines * in_stride; + y_pos += ExportYUVA(dec, y_pos); + } + return y_pos; +} + +static int EmitRowsYUVA(const VP8LDecoder* const dec, + const uint8_t* in, int in_stride, + int mb_w, int num_rows) { + int y_pos = dec->last_out_row_; + while (num_rows-- > 0) { + ConvertToYUVA((const uint32_t*)in, mb_w, y_pos, dec->output_); + in += in_stride; + ++y_pos; + } + return y_pos; +} + +//------------------------------------------------------------------------------ +// Cropping. + +// Sets io->mb_y, io->mb_h & io->mb_w according to start row, end row and +// crop options. Also updates the input data pointer, so that it points to the +// start of the cropped window. Note that pixels are in ARGB format even if +// 'in_data' is uint8_t*. +// Returns true if the crop window is not empty. +static int SetCropWindow(VP8Io* const io, int y_start, int y_end, + uint8_t** const in_data, int pixel_stride) { + assert(y_start < y_end); + assert(io->crop_left < io->crop_right); + if (y_end > io->crop_bottom) { + y_end = io->crop_bottom; // make sure we don't overflow on last row. + } + if (y_start < io->crop_top) { + const int delta = io->crop_top - y_start; + y_start = io->crop_top; + *in_data += delta * pixel_stride; + } + if (y_start >= y_end) return 0; // Crop window is empty. + + *in_data += io->crop_left * sizeof(uint32_t); + + io->mb_y = y_start - io->crop_top; + io->mb_w = io->crop_right - io->crop_left; + io->mb_h = y_end - y_start; + return 1; // Non-empty crop window. +} + +//------------------------------------------------------------------------------ + +static WEBP_INLINE int GetMetaIndex( + const uint32_t* const image, int xsize, int bits, int x, int y) { + if (bits == 0) return 0; + return image[xsize * (y >> bits) + (x >> bits)]; +} + +static WEBP_INLINE HTreeGroup* GetHtreeGroupForPos(VP8LMetadata* const hdr, + int x, int y) { + const int meta_index = GetMetaIndex(hdr->huffman_image_, hdr->huffman_xsize_, + hdr->huffman_subsample_bits_, x, y); + assert(meta_index < hdr->num_htree_groups_); + return hdr->htree_groups_ + meta_index; +} + +//------------------------------------------------------------------------------ +// Main loop, with custom row-processing function + +typedef void (*ProcessRowsFunc)(VP8LDecoder* const dec, int row); + +static void ApplyInverseTransforms(VP8LDecoder* const dec, int num_rows, + const uint32_t* const rows) { + int n = dec->next_transform_; + const int cache_pixs = dec->width_ * num_rows; + const int start_row = dec->last_row_; + const int end_row = start_row + num_rows; + const uint32_t* rows_in = rows; + uint32_t* const rows_out = dec->argb_cache_; + + // Inverse transforms. + while (n-- > 0) { + VP8LTransform* const transform = &dec->transforms_[n]; + VP8LInverseTransform(transform, start_row, end_row, rows_in, rows_out); + rows_in = rows_out; + } + if (rows_in != rows_out) { + // No transform called, hence just copy. + memcpy(rows_out, rows_in, cache_pixs * sizeof(*rows_out)); + } +} + +// Processes (transforms, scales & color-converts) the rows decoded after the +// last call. +static void ProcessRows(VP8LDecoder* const dec, int row) { + const uint32_t* const rows = dec->pixels_ + dec->width_ * dec->last_row_; + const int num_rows = row - dec->last_row_; + + assert(row <= dec->io_->crop_bottom); + // We can't process more than NUM_ARGB_CACHE_ROWS at a time (that's the size + // of argb_cache_), but we currently don't need more than that. + assert(num_rows <= NUM_ARGB_CACHE_ROWS); + if (num_rows > 0) { // Emit output. + VP8Io* const io = dec->io_; + uint8_t* rows_data = (uint8_t*)dec->argb_cache_; + const int in_stride = io->width * sizeof(uint32_t); // in unit of RGBA + + ApplyInverseTransforms(dec, num_rows, rows); + if (!SetCropWindow(io, dec->last_row_, row, &rows_data, in_stride)) { + // Nothing to output (this time). + } else { + const WebPDecBuffer* const output = dec->output_; + if (WebPIsRGBMode(output->colorspace)) { // convert to RGBA + const WebPRGBABuffer* const buf = &output->u.RGBA; + uint8_t* const rgba = buf->rgba + dec->last_out_row_ * buf->stride; + const int num_rows_out = +#if !defined(WEBP_REDUCE_SIZE) + io->use_scaling ? + EmitRescaledRowsRGBA(dec, rows_data, in_stride, io->mb_h, + rgba, buf->stride) : +#endif // WEBP_REDUCE_SIZE + EmitRows(output->colorspace, rows_data, in_stride, + io->mb_w, io->mb_h, rgba, buf->stride); + // Update 'last_out_row_'. + dec->last_out_row_ += num_rows_out; + } else { // convert to YUVA + dec->last_out_row_ = io->use_scaling ? + EmitRescaledRowsYUVA(dec, rows_data, in_stride, io->mb_h) : + EmitRowsYUVA(dec, rows_data, in_stride, io->mb_w, io->mb_h); + } + assert(dec->last_out_row_ <= output->height); + } + } + + // Update 'last_row_'. + dec->last_row_ = row; + assert(dec->last_row_ <= dec->height_); +} + +// Row-processing for the special case when alpha data contains only one +// transform (color indexing), and trivial non-green literals. +static int Is8bOptimizable(const VP8LMetadata* const hdr) { + int i; + if (hdr->color_cache_size_ > 0) return 0; + // When the Huffman tree contains only one symbol, we can skip the + // call to ReadSymbol() for red/blue/alpha channels. + for (i = 0; i < hdr->num_htree_groups_; ++i) { + HuffmanCode** const htrees = hdr->htree_groups_[i].htrees; + if (htrees[RED][0].bits > 0) return 0; + if (htrees[BLUE][0].bits > 0) return 0; + if (htrees[ALPHA][0].bits > 0) return 0; + } + return 1; +} + +static void AlphaApplyFilter(ALPHDecoder* const alph_dec, + int first_row, int last_row, + uint8_t* out, int stride) { + if (alph_dec->filter_ != WEBP_FILTER_NONE) { + int y; + const uint8_t* prev_line = alph_dec->prev_line_; + assert(WebPUnfilters[alph_dec->filter_] != NULL); + for (y = first_row; y < last_row; ++y) { + WebPUnfilters[alph_dec->filter_](prev_line, out, out, stride); + prev_line = out; + out += stride; + } + alph_dec->prev_line_ = prev_line; + } +} + +static void ExtractPalettedAlphaRows(VP8LDecoder* const dec, int last_row) { + // For vertical and gradient filtering, we need to decode the part above the + // crop_top row, in order to have the correct spatial predictors. + ALPHDecoder* const alph_dec = (ALPHDecoder*)dec->io_->opaque; + const int top_row = + (alph_dec->filter_ == WEBP_FILTER_NONE || + alph_dec->filter_ == WEBP_FILTER_HORIZONTAL) ? dec->io_->crop_top + : dec->last_row_; + const int first_row = (dec->last_row_ < top_row) ? top_row : dec->last_row_; + assert(last_row <= dec->io_->crop_bottom); + if (last_row > first_row) { + // Special method for paletted alpha data. We only process the cropped area. + const int width = dec->io_->width; + uint8_t* out = alph_dec->output_ + width * first_row; + const uint8_t* const in = + (uint8_t*)dec->pixels_ + dec->width_ * first_row; + VP8LTransform* const transform = &dec->transforms_[0]; + assert(dec->next_transform_ == 1); + assert(transform->type_ == COLOR_INDEXING_TRANSFORM); + VP8LColorIndexInverseTransformAlpha(transform, first_row, last_row, + in, out); + AlphaApplyFilter(alph_dec, first_row, last_row, out, width); + } + dec->last_row_ = dec->last_out_row_ = last_row; +} + +//------------------------------------------------------------------------------ +// Helper functions for fast pattern copy (8b and 32b) + +// cyclic rotation of pattern word +static WEBP_INLINE uint32_t Rotate8b(uint32_t V) { +#if defined(WORDS_BIGENDIAN) + return ((V & 0xff000000u) >> 24) | (V << 8); +#else + return ((V & 0xffu) << 24) | (V >> 8); +#endif +} + +// copy 1, 2 or 4-bytes pattern +static WEBP_INLINE void CopySmallPattern8b(const uint8_t* src, uint8_t* dst, + int length, uint32_t pattern) { + int i; + // align 'dst' to 4-bytes boundary. Adjust the pattern along the way. + while ((uintptr_t)dst & 3) { + *dst++ = *src++; + pattern = Rotate8b(pattern); + --length; + } + // Copy the pattern 4 bytes at a time. + for (i = 0; i < (length >> 2); ++i) { + ((uint32_t*)dst)[i] = pattern; + } + // Finish with left-overs. 'pattern' is still correctly positioned, + // so no Rotate8b() call is needed. + for (i <<= 2; i < length; ++i) { + dst[i] = src[i]; + } +} + +static WEBP_INLINE void CopyBlock8b(uint8_t* const dst, int dist, int length) { + const uint8_t* src = dst - dist; + if (length >= 8) { + uint32_t pattern = 0; + switch (dist) { + case 1: + pattern = src[0]; +#if defined(__arm__) || defined(_M_ARM) // arm doesn't like multiply that much + pattern |= pattern << 8; + pattern |= pattern << 16; +#elif defined(WEBP_USE_MIPS_DSP_R2) + __asm__ volatile ("replv.qb %0, %0" : "+r"(pattern)); +#else + pattern = 0x01010101u * pattern; +#endif + break; + case 2: + memcpy(&pattern, src, sizeof(uint16_t)); +#if defined(__arm__) || defined(_M_ARM) + pattern |= pattern << 16; +#elif defined(WEBP_USE_MIPS_DSP_R2) + __asm__ volatile ("replv.ph %0, %0" : "+r"(pattern)); +#else + pattern = 0x00010001u * pattern; +#endif + break; + case 4: + memcpy(&pattern, src, sizeof(uint32_t)); + break; + default: + goto Copy; + break; + } + CopySmallPattern8b(src, dst, length, pattern); + return; + } + Copy: + if (dist >= length) { // no overlap -> use memcpy() + memcpy(dst, src, length * sizeof(*dst)); + } else { + int i; + for (i = 0; i < length; ++i) dst[i] = src[i]; + } +} + +// copy pattern of 1 or 2 uint32_t's +static WEBP_INLINE void CopySmallPattern32b(const uint32_t* src, + uint32_t* dst, + int length, uint64_t pattern) { + int i; + if ((uintptr_t)dst & 4) { // Align 'dst' to 8-bytes boundary. + *dst++ = *src++; + pattern = (pattern >> 32) | (pattern << 32); + --length; + } + assert(0 == ((uintptr_t)dst & 7)); + for (i = 0; i < (length >> 1); ++i) { + ((uint64_t*)dst)[i] = pattern; // Copy the pattern 8 bytes at a time. + } + if (length & 1) { // Finish with left-over. + dst[i << 1] = src[i << 1]; + } +} + +static WEBP_INLINE void CopyBlock32b(uint32_t* const dst, + int dist, int length) { + const uint32_t* const src = dst - dist; + if (dist <= 2 && length >= 4 && ((uintptr_t)dst & 3) == 0) { + uint64_t pattern; + if (dist == 1) { + pattern = (uint64_t)src[0]; + pattern |= pattern << 32; + } else { + memcpy(&pattern, src, sizeof(pattern)); + } + CopySmallPattern32b(src, dst, length, pattern); + } else if (dist >= length) { // no overlap + memcpy(dst, src, length * sizeof(*dst)); + } else { + int i; + for (i = 0; i < length; ++i) dst[i] = src[i]; + } +} + +//------------------------------------------------------------------------------ + +static int DecodeAlphaData(VP8LDecoder* const dec, uint8_t* const data, + int width, int height, int last_row) { + int ok = 1; + int row = dec->last_pixel_ / width; + int col = dec->last_pixel_ % width; + VP8LBitReader* const br = &dec->br_; + VP8LMetadata* const hdr = &dec->hdr_; + int pos = dec->last_pixel_; // current position + const int end = width * height; // End of data + const int last = width * last_row; // Last pixel to decode + const int len_code_limit = NUM_LITERAL_CODES + NUM_LENGTH_CODES; + const int mask = hdr->huffman_mask_; + const HTreeGroup* htree_group = + (pos < last) ? GetHtreeGroupForPos(hdr, col, row) : NULL; + assert(pos <= end); + assert(last_row <= height); + assert(Is8bOptimizable(hdr)); + + while (!br->eos_ && pos < last) { + int code; + // Only update when changing tile. + if ((col & mask) == 0) { + htree_group = GetHtreeGroupForPos(hdr, col, row); + } + assert(htree_group != NULL); + VP8LFillBitWindow(br); + code = ReadSymbol(htree_group->htrees[GREEN], br); + if (code < NUM_LITERAL_CODES) { // Literal + data[pos] = code; + ++pos; + ++col; + if (col >= width) { + col = 0; + ++row; + if (row <= last_row && (row % NUM_ARGB_CACHE_ROWS == 0)) { + ExtractPalettedAlphaRows(dec, row); + } + } + } else if (code < len_code_limit) { // Backward reference + int dist_code, dist; + const int length_sym = code - NUM_LITERAL_CODES; + const int length = GetCopyLength(length_sym, br); + const int dist_symbol = ReadSymbol(htree_group->htrees[DIST], br); + VP8LFillBitWindow(br); + dist_code = GetCopyDistance(dist_symbol, br); + dist = PlaneCodeToDistance(width, dist_code); + if (pos >= dist && end - pos >= length) { + CopyBlock8b(data + pos, dist, length); + } else { + ok = 0; + goto End; + } + pos += length; + col += length; + while (col >= width) { + col -= width; + ++row; + if (row <= last_row && (row % NUM_ARGB_CACHE_ROWS == 0)) { + ExtractPalettedAlphaRows(dec, row); + } + } + if (pos < last && (col & mask)) { + htree_group = GetHtreeGroupForPos(hdr, col, row); + } + } else { // Not reached + ok = 0; + goto End; + } + br->eos_ = VP8LIsEndOfStream(br); + } + // Process the remaining rows corresponding to last row-block. + ExtractPalettedAlphaRows(dec, row > last_row ? last_row : row); + + End: + br->eos_ = VP8LIsEndOfStream(br); + if (!ok || (br->eos_ && pos < end)) { + ok = 0; + dec->status_ = br->eos_ ? VP8_STATUS_SUSPENDED + : VP8_STATUS_BITSTREAM_ERROR; + } else { + dec->last_pixel_ = pos; + } + return ok; +} + +static void SaveState(VP8LDecoder* const dec, int last_pixel) { + assert(dec->incremental_); + dec->saved_br_ = dec->br_; + dec->saved_last_pixel_ = last_pixel; + if (dec->hdr_.color_cache_size_ > 0) { + VP8LColorCacheCopy(&dec->hdr_.color_cache_, &dec->hdr_.saved_color_cache_); + } +} + +static void RestoreState(VP8LDecoder* const dec) { + assert(dec->br_.eos_); + dec->status_ = VP8_STATUS_SUSPENDED; + dec->br_ = dec->saved_br_; + dec->last_pixel_ = dec->saved_last_pixel_; + if (dec->hdr_.color_cache_size_ > 0) { + VP8LColorCacheCopy(&dec->hdr_.saved_color_cache_, &dec->hdr_.color_cache_); + } +} + +#define SYNC_EVERY_N_ROWS 8 // minimum number of rows between check-points +static int DecodeImageData(VP8LDecoder* const dec, uint32_t* const data, + int width, int height, int last_row, + ProcessRowsFunc process_func) { + int row = dec->last_pixel_ / width; + int col = dec->last_pixel_ % width; + VP8LBitReader* const br = &dec->br_; + VP8LMetadata* const hdr = &dec->hdr_; + uint32_t* src = data + dec->last_pixel_; + uint32_t* last_cached = src; + uint32_t* const src_end = data + width * height; // End of data + uint32_t* const src_last = data + width * last_row; // Last pixel to decode + const int len_code_limit = NUM_LITERAL_CODES + NUM_LENGTH_CODES; + const int color_cache_limit = len_code_limit + hdr->color_cache_size_; + int next_sync_row = dec->incremental_ ? row : 1 << 24; + VP8LColorCache* const color_cache = + (hdr->color_cache_size_ > 0) ? &hdr->color_cache_ : NULL; + const int mask = hdr->huffman_mask_; + const HTreeGroup* htree_group = + (src < src_last) ? GetHtreeGroupForPos(hdr, col, row) : NULL; + assert(dec->last_row_ < last_row); + assert(src_last <= src_end); + + while (src < src_last) { + int code; + if (row >= next_sync_row) { + SaveState(dec, (int)(src - data)); + next_sync_row = row + SYNC_EVERY_N_ROWS; + } + // Only update when changing tile. Note we could use this test: + // if "((((prev_col ^ col) | prev_row ^ row)) > mask)" -> tile changed + // but that's actually slower and needs storing the previous col/row. + if ((col & mask) == 0) { + htree_group = GetHtreeGroupForPos(hdr, col, row); + } + assert(htree_group != NULL); + if (htree_group->is_trivial_code) { + *src = htree_group->literal_arb; + goto AdvanceByOne; + } + VP8LFillBitWindow(br); + if (htree_group->use_packed_table) { + code = ReadPackedSymbols(htree_group, br, src); + if (VP8LIsEndOfStream(br)) break; + if (code == PACKED_NON_LITERAL_CODE) goto AdvanceByOne; + } else { + code = ReadSymbol(htree_group->htrees[GREEN], br); + } + if (VP8LIsEndOfStream(br)) break; + if (code < NUM_LITERAL_CODES) { // Literal + if (htree_group->is_trivial_literal) { + *src = htree_group->literal_arb | (code << 8); + } else { + int red, blue, alpha; + red = ReadSymbol(htree_group->htrees[RED], br); + VP8LFillBitWindow(br); + blue = ReadSymbol(htree_group->htrees[BLUE], br); + alpha = ReadSymbol(htree_group->htrees[ALPHA], br); + if (VP8LIsEndOfStream(br)) break; + *src = ((uint32_t)alpha << 24) | (red << 16) | (code << 8) | blue; + } + AdvanceByOne: + ++src; + ++col; + if (col >= width) { + col = 0; + ++row; + if (process_func != NULL) { + if (row <= last_row && (row % NUM_ARGB_CACHE_ROWS == 0)) { + process_func(dec, row); + } + } + if (color_cache != NULL) { + while (last_cached < src) { + VP8LColorCacheInsert(color_cache, *last_cached++); + } + } + } + } else if (code < len_code_limit) { // Backward reference + int dist_code, dist; + const int length_sym = code - NUM_LITERAL_CODES; + const int length = GetCopyLength(length_sym, br); + const int dist_symbol = ReadSymbol(htree_group->htrees[DIST], br); + VP8LFillBitWindow(br); + dist_code = GetCopyDistance(dist_symbol, br); + dist = PlaneCodeToDistance(width, dist_code); + if (VP8LIsEndOfStream(br)) break; + if (src - data < (ptrdiff_t)dist || src_end - src < (ptrdiff_t)length) { + goto Error; + } else { + CopyBlock32b(src, dist, length); + } + src += length; + col += length; + while (col >= width) { + col -= width; + ++row; + if (process_func != NULL) { + if (row <= last_row && (row % NUM_ARGB_CACHE_ROWS == 0)) { + process_func(dec, row); + } + } + } + // Because of the check done above (before 'src' was incremented by + // 'length'), the following holds true. + assert(src <= src_end); + if (col & mask) htree_group = GetHtreeGroupForPos(hdr, col, row); + if (color_cache != NULL) { + while (last_cached < src) { + VP8LColorCacheInsert(color_cache, *last_cached++); + } + } + } else if (code < color_cache_limit) { // Color cache + const int key = code - len_code_limit; + assert(color_cache != NULL); + while (last_cached < src) { + VP8LColorCacheInsert(color_cache, *last_cached++); + } + *src = VP8LColorCacheLookup(color_cache, key); + goto AdvanceByOne; + } else { // Not reached + goto Error; + } + } + + br->eos_ = VP8LIsEndOfStream(br); + if (dec->incremental_ && br->eos_ && src < src_end) { + RestoreState(dec); + } else if (!br->eos_) { + // Process the remaining rows corresponding to last row-block. + if (process_func != NULL) { + process_func(dec, row > last_row ? last_row : row); + } + dec->status_ = VP8_STATUS_OK; + dec->last_pixel_ = (int)(src - data); // end-of-scan marker + } else { + // if not incremental, and we are past the end of buffer (eos_=1), then this + // is a real bitstream error. + goto Error; + } + return 1; + + Error: + dec->status_ = VP8_STATUS_BITSTREAM_ERROR; + return 0; +} + +// ----------------------------------------------------------------------------- +// VP8LTransform + +static void ClearTransform(VP8LTransform* const transform) { + WebPSafeFree(transform->data_); + transform->data_ = NULL; +} + +// For security reason, we need to remap the color map to span +// the total possible bundled values, and not just the num_colors. +static int ExpandColorMap(int num_colors, VP8LTransform* const transform) { + int i; + const int final_num_colors = 1 << (8 >> transform->bits_); + uint32_t* const new_color_map = + (uint32_t*)WebPSafeMalloc((uint64_t)final_num_colors, + sizeof(*new_color_map)); + if (new_color_map == NULL) { + return 0; + } else { + uint8_t* const data = (uint8_t*)transform->data_; + uint8_t* const new_data = (uint8_t*)new_color_map; + new_color_map[0] = transform->data_[0]; + for (i = 4; i < 4 * num_colors; ++i) { + // Equivalent to AddPixelEq(), on a byte-basis. + new_data[i] = (data[i] + new_data[i - 4]) & 0xff; + } + for (; i < 4 * final_num_colors; ++i) { + new_data[i] = 0; // black tail. + } + WebPSafeFree(transform->data_); + transform->data_ = new_color_map; + } + return 1; +} + +static int ReadTransform(int* const xsize, int const* ysize, + VP8LDecoder* const dec) { + int ok = 1; + VP8LBitReader* const br = &dec->br_; + VP8LTransform* transform = &dec->transforms_[dec->next_transform_]; + const VP8LImageTransformType type = + (VP8LImageTransformType)VP8LReadBits(br, 2); + + // Each transform type can only be present once in the stream. + if (dec->transforms_seen_ & (1U << type)) { + return 0; // Already there, let's not accept the second same transform. + } + dec->transforms_seen_ |= (1U << type); + + transform->type_ = type; + transform->xsize_ = *xsize; + transform->ysize_ = *ysize; + transform->data_ = NULL; + ++dec->next_transform_; + assert(dec->next_transform_ <= NUM_TRANSFORMS); + + switch (type) { + case PREDICTOR_TRANSFORM: + case CROSS_COLOR_TRANSFORM: + transform->bits_ = VP8LReadBits(br, 3) + 2; + ok = DecodeImageStream(VP8LSubSampleSize(transform->xsize_, + transform->bits_), + VP8LSubSampleSize(transform->ysize_, + transform->bits_), + 0, dec, &transform->data_); + break; + case COLOR_INDEXING_TRANSFORM: { + const int num_colors = VP8LReadBits(br, 8) + 1; + const int bits = (num_colors > 16) ? 0 + : (num_colors > 4) ? 1 + : (num_colors > 2) ? 2 + : 3; + *xsize = VP8LSubSampleSize(transform->xsize_, bits); + transform->bits_ = bits; + ok = DecodeImageStream(num_colors, 1, 0, dec, &transform->data_); + ok = ok && ExpandColorMap(num_colors, transform); + break; + } + case SUBTRACT_GREEN: + break; + default: + assert(0); // can't happen + break; + } + + return ok; +} + +// ----------------------------------------------------------------------------- +// VP8LMetadata + +static void InitMetadata(VP8LMetadata* const hdr) { + assert(hdr != NULL); + memset(hdr, 0, sizeof(*hdr)); +} + +static void ClearMetadata(VP8LMetadata* const hdr) { + assert(hdr != NULL); + + WebPSafeFree(hdr->huffman_image_); + WebPSafeFree(hdr->huffman_tables_); + VP8LHtreeGroupsFree(hdr->htree_groups_); + VP8LColorCacheClear(&hdr->color_cache_); + VP8LColorCacheClear(&hdr->saved_color_cache_); + InitMetadata(hdr); +} + +// ----------------------------------------------------------------------------- +// VP8LDecoder + +VP8LDecoder* VP8LNew(void) { + VP8LDecoder* const dec = (VP8LDecoder*)WebPSafeCalloc(1ULL, sizeof(*dec)); + if (dec == NULL) return NULL; + dec->status_ = VP8_STATUS_OK; + dec->state_ = READ_DIM; + + VP8LDspInit(); // Init critical function pointers. + + return dec; +} + +void VP8LClear(VP8LDecoder* const dec) { + int i; + if (dec == NULL) return; + ClearMetadata(&dec->hdr_); + + WebPSafeFree(dec->pixels_); + dec->pixels_ = NULL; + for (i = 0; i < dec->next_transform_; ++i) { + ClearTransform(&dec->transforms_[i]); + } + dec->next_transform_ = 0; + dec->transforms_seen_ = 0; + + WebPSafeFree(dec->rescaler_memory); + dec->rescaler_memory = NULL; + + dec->output_ = NULL; // leave no trace behind +} + +void VP8LDelete(VP8LDecoder* const dec) { + if (dec != NULL) { + VP8LClear(dec); + WebPSafeFree(dec); + } +} + +static void UpdateDecoder(VP8LDecoder* const dec, int width, int height) { + VP8LMetadata* const hdr = &dec->hdr_; + const int num_bits = hdr->huffman_subsample_bits_; + dec->width_ = width; + dec->height_ = height; + + hdr->huffman_xsize_ = VP8LSubSampleSize(width, num_bits); + hdr->huffman_mask_ = (num_bits == 0) ? ~0 : (1 << num_bits) - 1; +} + +static int DecodeImageStream(int xsize, int ysize, + int is_level0, + VP8LDecoder* const dec, + uint32_t** const decoded_data) { + int ok = 1; + int transform_xsize = xsize; + int transform_ysize = ysize; + VP8LBitReader* const br = &dec->br_; + VP8LMetadata* const hdr = &dec->hdr_; + uint32_t* data = NULL; + int color_cache_bits = 0; + + // Read the transforms (may recurse). + if (is_level0) { + while (ok && VP8LReadBits(br, 1)) { + ok = ReadTransform(&transform_xsize, &transform_ysize, dec); + } + } + + // Color cache + if (ok && VP8LReadBits(br, 1)) { + color_cache_bits = VP8LReadBits(br, 4); + ok = (color_cache_bits >= 1 && color_cache_bits <= MAX_CACHE_BITS); + if (!ok) { + dec->status_ = VP8_STATUS_BITSTREAM_ERROR; + goto End; + } + } + + // Read the Huffman codes (may recurse). + ok = ok && ReadHuffmanCodes(dec, transform_xsize, transform_ysize, + color_cache_bits, is_level0); + if (!ok) { + dec->status_ = VP8_STATUS_BITSTREAM_ERROR; + goto End; + } + + // Finish setting up the color-cache + if (color_cache_bits > 0) { + hdr->color_cache_size_ = 1 << color_cache_bits; + if (!VP8LColorCacheInit(&hdr->color_cache_, color_cache_bits)) { + dec->status_ = VP8_STATUS_OUT_OF_MEMORY; + ok = 0; + goto End; + } + } else { + hdr->color_cache_size_ = 0; + } + UpdateDecoder(dec, transform_xsize, transform_ysize); + + if (is_level0) { // level 0 complete + dec->state_ = READ_HDR; + goto End; + } + + { + const uint64_t total_size = (uint64_t)transform_xsize * transform_ysize; + data = (uint32_t*)WebPSafeMalloc(total_size, sizeof(*data)); + if (data == NULL) { + dec->status_ = VP8_STATUS_OUT_OF_MEMORY; + ok = 0; + goto End; + } + } + + // Use the Huffman trees to decode the LZ77 encoded data. + ok = DecodeImageData(dec, data, transform_xsize, transform_ysize, + transform_ysize, NULL); + ok = ok && !br->eos_; + + End: + if (!ok) { + WebPSafeFree(data); + ClearMetadata(hdr); + } else { + if (decoded_data != NULL) { + *decoded_data = data; + } else { + // We allocate image data in this function only for transforms. At level 0 + // (that is: not the transforms), we shouldn't have allocated anything. + assert(data == NULL); + assert(is_level0); + } + dec->last_pixel_ = 0; // Reset for future DECODE_DATA_FUNC() calls. + if (!is_level0) ClearMetadata(hdr); // Clean up temporary data behind. + } + return ok; +} + +//------------------------------------------------------------------------------ +// Allocate internal buffers dec->pixels_ and dec->argb_cache_. +static int AllocateInternalBuffers32b(VP8LDecoder* const dec, int final_width) { + const uint64_t num_pixels = (uint64_t)dec->width_ * dec->height_; + // Scratch buffer corresponding to top-prediction row for transforming the + // first row in the row-blocks. Not needed for paletted alpha. + const uint64_t cache_top_pixels = (uint16_t)final_width; + // Scratch buffer for temporary BGRA storage. Not needed for paletted alpha. + const uint64_t cache_pixels = (uint64_t)final_width * NUM_ARGB_CACHE_ROWS; + const uint64_t total_num_pixels = + num_pixels + cache_top_pixels + cache_pixels; + + assert(dec->width_ <= final_width); + dec->pixels_ = (uint32_t*)WebPSafeMalloc(total_num_pixels, sizeof(uint32_t)); + if (dec->pixels_ == NULL) { + dec->argb_cache_ = NULL; // for sanity check + dec->status_ = VP8_STATUS_OUT_OF_MEMORY; + return 0; + } + dec->argb_cache_ = dec->pixels_ + num_pixels + cache_top_pixels; + return 1; +} + +static int AllocateInternalBuffers8b(VP8LDecoder* const dec) { + const uint64_t total_num_pixels = (uint64_t)dec->width_ * dec->height_; + dec->argb_cache_ = NULL; // for sanity check + dec->pixels_ = (uint32_t*)WebPSafeMalloc(total_num_pixels, sizeof(uint8_t)); + if (dec->pixels_ == NULL) { + dec->status_ = VP8_STATUS_OUT_OF_MEMORY; + return 0; + } + return 1; +} + +//------------------------------------------------------------------------------ + +// Special row-processing that only stores the alpha data. +static void ExtractAlphaRows(VP8LDecoder* const dec, int last_row) { + int cur_row = dec->last_row_; + int num_rows = last_row - cur_row; + const uint32_t* in = dec->pixels_ + dec->width_ * cur_row; + + assert(last_row <= dec->io_->crop_bottom); + while (num_rows > 0) { + const int num_rows_to_process = + (num_rows > NUM_ARGB_CACHE_ROWS) ? NUM_ARGB_CACHE_ROWS : num_rows; + // Extract alpha (which is stored in the green plane). + ALPHDecoder* const alph_dec = (ALPHDecoder*)dec->io_->opaque; + uint8_t* const output = alph_dec->output_; + const int width = dec->io_->width; // the final width (!= dec->width_) + const int cache_pixs = width * num_rows_to_process; + uint8_t* const dst = output + width * cur_row; + const uint32_t* const src = dec->argb_cache_; + ApplyInverseTransforms(dec, num_rows_to_process, in); + WebPExtractGreen(src, dst, cache_pixs); + AlphaApplyFilter(alph_dec, + cur_row, cur_row + num_rows_to_process, dst, width); + num_rows -= num_rows_to_process; + in += num_rows_to_process * dec->width_; + cur_row += num_rows_to_process; + } + assert(cur_row == last_row); + dec->last_row_ = dec->last_out_row_ = last_row; +} + +int VP8LDecodeAlphaHeader(ALPHDecoder* const alph_dec, + const uint8_t* const data, size_t data_size) { + int ok = 0; + VP8LDecoder* dec = VP8LNew(); + + if (dec == NULL) return 0; + + assert(alph_dec != NULL); + alph_dec->vp8l_dec_ = dec; + + dec->width_ = alph_dec->width_; + dec->height_ = alph_dec->height_; + dec->io_ = &alph_dec->io_; + dec->io_->opaque = alph_dec; + dec->io_->width = alph_dec->width_; + dec->io_->height = alph_dec->height_; + + dec->status_ = VP8_STATUS_OK; + VP8LInitBitReader(&dec->br_, data, data_size); + + if (!DecodeImageStream(alph_dec->width_, alph_dec->height_, 1, dec, NULL)) { + goto Err; + } + + // Special case: if alpha data uses only the color indexing transform and + // doesn't use color cache (a frequent case), we will use DecodeAlphaData() + // method that only needs allocation of 1 byte per pixel (alpha channel). + if (dec->next_transform_ == 1 && + dec->transforms_[0].type_ == COLOR_INDEXING_TRANSFORM && + Is8bOptimizable(&dec->hdr_)) { + alph_dec->use_8b_decode_ = 1; + ok = AllocateInternalBuffers8b(dec); + } else { + // Allocate internal buffers (note that dec->width_ may have changed here). + alph_dec->use_8b_decode_ = 0; + ok = AllocateInternalBuffers32b(dec, alph_dec->width_); + } + + if (!ok) goto Err; + + return 1; + + Err: + VP8LDelete(alph_dec->vp8l_dec_); + alph_dec->vp8l_dec_ = NULL; + return 0; +} + +int VP8LDecodeAlphaImageStream(ALPHDecoder* const alph_dec, int last_row) { + VP8LDecoder* const dec = alph_dec->vp8l_dec_; + assert(dec != NULL); + assert(last_row <= dec->height_); + + if (dec->last_row_ >= last_row) { + return 1; // done + } + + if (!alph_dec->use_8b_decode_) WebPInitAlphaProcessing(); + + // Decode (with special row processing). + return alph_dec->use_8b_decode_ ? + DecodeAlphaData(dec, (uint8_t*)dec->pixels_, dec->width_, dec->height_, + last_row) : + DecodeImageData(dec, dec->pixels_, dec->width_, dec->height_, + last_row, ExtractAlphaRows); +} + +//------------------------------------------------------------------------------ + +int VP8LDecodeHeader(VP8LDecoder* const dec, VP8Io* const io) { + int width, height, has_alpha; + + if (dec == NULL) return 0; + if (io == NULL) { + dec->status_ = VP8_STATUS_INVALID_PARAM; + return 0; + } + + dec->io_ = io; + dec->status_ = VP8_STATUS_OK; + VP8LInitBitReader(&dec->br_, io->data, io->data_size); + if (!ReadImageInfo(&dec->br_, &width, &height, &has_alpha)) { + dec->status_ = VP8_STATUS_BITSTREAM_ERROR; + goto Error; + } + dec->state_ = READ_DIM; + io->width = width; + io->height = height; + + if (!DecodeImageStream(width, height, 1, dec, NULL)) goto Error; + return 1; + + Error: + VP8LClear(dec); + assert(dec->status_ != VP8_STATUS_OK); + return 0; +} + +int VP8LDecodeImage(VP8LDecoder* const dec) { + VP8Io* io = NULL; + WebPDecParams* params = NULL; + + // Sanity checks. + if (dec == NULL) return 0; + + assert(dec->hdr_.huffman_tables_ != NULL); + assert(dec->hdr_.htree_groups_ != NULL); + assert(dec->hdr_.num_htree_groups_ > 0); + + io = dec->io_; + assert(io != NULL); + params = (WebPDecParams*)io->opaque; + assert(params != NULL); + + // Initialization. + if (dec->state_ != READ_DATA) { + dec->output_ = params->output; + assert(dec->output_ != NULL); + + if (!WebPIoInitFromOptions(params->options, io, MODE_BGRA)) { + dec->status_ = VP8_STATUS_INVALID_PARAM; + goto Err; + } + + if (!AllocateInternalBuffers32b(dec, io->width)) goto Err; + +#if !defined(WEBP_REDUCE_SIZE) + if (io->use_scaling && !AllocateAndInitRescaler(dec, io)) goto Err; + + if (io->use_scaling || WebPIsPremultipliedMode(dec->output_->colorspace)) { + // need the alpha-multiply functions for premultiplied output or rescaling + WebPInitAlphaProcessing(); + } +#else + if (io->use_scaling) { + dec->status_ = VP8_STATUS_INVALID_PARAM; + goto Err; + } +#endif + if (!WebPIsRGBMode(dec->output_->colorspace)) { + WebPInitConvertARGBToYUV(); + if (dec->output_->u.YUVA.a != NULL) WebPInitAlphaProcessing(); + } + if (dec->incremental_) { + if (dec->hdr_.color_cache_size_ > 0 && + dec->hdr_.saved_color_cache_.colors_ == NULL) { + if (!VP8LColorCacheInit(&dec->hdr_.saved_color_cache_, + dec->hdr_.color_cache_.hash_bits_)) { + dec->status_ = VP8_STATUS_OUT_OF_MEMORY; + goto Err; + } + } + } + dec->state_ = READ_DATA; + } + + // Decode. + if (!DecodeImageData(dec, dec->pixels_, dec->width_, dec->height_, + io->crop_bottom, ProcessRows)) { + goto Err; + } + + params->last_y = dec->last_out_row_; + return 1; + + Err: + VP8LClear(dec); + assert(dec->status_ != VP8_STATUS_OK); + return 0; +} + +//------------------------------------------------------------------------------ diff --git a/thirdparty/libwebp/src/dec/vp8li_dec.h b/thirdparty/libwebp/src/dec/vp8li_dec.h new file mode 100644 index 0000000000..8e500cf9ff --- /dev/null +++ b/thirdparty/libwebp/src/dec/vp8li_dec.h @@ -0,0 +1,135 @@ +// Copyright 2012 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. +// ----------------------------------------------------------------------------- +// +// Lossless decoder: internal header. +// +// Author: Skal (pascal.massimino@gmail.com) +// Vikas Arora(vikaas.arora@gmail.com) + +#ifndef WEBP_DEC_VP8LI_DEC_H_ +#define WEBP_DEC_VP8LI_DEC_H_ + +#include <string.h> // for memcpy() +#include "src/dec/webpi_dec.h" +#include "src/utils/bit_reader_utils.h" +#include "src/utils/color_cache_utils.h" +#include "src/utils/huffman_utils.h" + +#ifdef __cplusplus +extern "C" { +#endif + +typedef enum { + READ_DATA = 0, + READ_HDR = 1, + READ_DIM = 2 +} VP8LDecodeState; + +typedef struct VP8LTransform VP8LTransform; +struct VP8LTransform { + VP8LImageTransformType type_; // transform type. + int bits_; // subsampling bits defining transform window. + int xsize_; // transform window X index. + int ysize_; // transform window Y index. + uint32_t *data_; // transform data. +}; + +typedef struct { + int color_cache_size_; + VP8LColorCache color_cache_; + VP8LColorCache saved_color_cache_; // for incremental + + int huffman_mask_; + int huffman_subsample_bits_; + int huffman_xsize_; + uint32_t *huffman_image_; + int num_htree_groups_; + HTreeGroup *htree_groups_; + HuffmanCode *huffman_tables_; +} VP8LMetadata; + +typedef struct VP8LDecoder VP8LDecoder; +struct VP8LDecoder { + VP8StatusCode status_; + VP8LDecodeState state_; + VP8Io *io_; + + const WebPDecBuffer *output_; // shortcut to io->opaque->output + + uint32_t *pixels_; // Internal data: either uint8_t* for alpha + // or uint32_t* for BGRA. + uint32_t *argb_cache_; // Scratch buffer for temporary BGRA storage. + + VP8LBitReader br_; + int incremental_; // if true, incremental decoding is expected + VP8LBitReader saved_br_; // note: could be local variables too + int saved_last_pixel_; + + int width_; + int height_; + int last_row_; // last input row decoded so far. + int last_pixel_; // last pixel decoded so far. However, it may + // not be transformed, scaled and + // color-converted yet. + int last_out_row_; // last row output so far. + + VP8LMetadata hdr_; + + int next_transform_; + VP8LTransform transforms_[NUM_TRANSFORMS]; + // or'd bitset storing the transforms types. + uint32_t transforms_seen_; + + uint8_t *rescaler_memory; // Working memory for rescaling work. + WebPRescaler *rescaler; // Common rescaler for all channels. +}; + +//------------------------------------------------------------------------------ +// internal functions. Not public. + +struct ALPHDecoder; // Defined in dec/alphai.h. + +// in vp8l.c + +// Decodes image header for alpha data stored using lossless compression. +// Returns false in case of error. +int VP8LDecodeAlphaHeader(struct ALPHDecoder* const alph_dec, + const uint8_t* const data, size_t data_size); + +// Decodes *at least* 'last_row' rows of alpha. If some of the initial rows are +// already decoded in previous call(s), it will resume decoding from where it +// was paused. +// Returns false in case of bitstream error. +int VP8LDecodeAlphaImageStream(struct ALPHDecoder* const alph_dec, + int last_row); + +// Allocates and initialize a new lossless decoder instance. +VP8LDecoder* VP8LNew(void); + +// Decodes the image header. Returns false in case of error. +int VP8LDecodeHeader(VP8LDecoder* const dec, VP8Io* const io); + +// Decodes an image. It's required to decode the lossless header before calling +// this function. Returns false in case of error, with updated dec->status_. +int VP8LDecodeImage(VP8LDecoder* const dec); + +// Resets the decoder in its initial state, reclaiming memory. +// Preserves the dec->status_ value. +void VP8LClear(VP8LDecoder* const dec); + +// Clears and deallocate a lossless decoder instance. +void VP8LDelete(VP8LDecoder* const dec); + +//------------------------------------------------------------------------------ + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif /* WEBP_DEC_VP8LI_DEC_H_ */ diff --git a/thirdparty/libwebp/src/dec/webp_dec.c b/thirdparty/libwebp/src/dec/webp_dec.c new file mode 100644 index 0000000000..42d098874d --- /dev/null +++ b/thirdparty/libwebp/src/dec/webp_dec.c @@ -0,0 +1,845 @@ +// Copyright 2010 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. +// ----------------------------------------------------------------------------- +// +// Main decoding functions for WEBP images. +// +// Author: Skal (pascal.massimino@gmail.com) + +#include <stdlib.h> + +#include "src/dec/vp8i_dec.h" +#include "src/dec/vp8li_dec.h" +#include "src/dec/webpi_dec.h" +#include "src/utils/utils.h" +#include "src/webp/mux_types.h" // ALPHA_FLAG + +//------------------------------------------------------------------------------ +// RIFF layout is: +// Offset tag +// 0...3 "RIFF" 4-byte tag +// 4...7 size of image data (including metadata) starting at offset 8 +// 8...11 "WEBP" our form-type signature +// The RIFF container (12 bytes) is followed by appropriate chunks: +// 12..15 "VP8 ": 4-bytes tags, signaling the use of VP8 video format +// 16..19 size of the raw VP8 image data, starting at offset 20 +// 20.... the VP8 bytes +// Or, +// 12..15 "VP8L": 4-bytes tags, signaling the use of VP8L lossless format +// 16..19 size of the raw VP8L image data, starting at offset 20 +// 20.... the VP8L bytes +// Or, +// 12..15 "VP8X": 4-bytes tags, describing the extended-VP8 chunk. +// 16..19 size of the VP8X chunk starting at offset 20. +// 20..23 VP8X flags bit-map corresponding to the chunk-types present. +// 24..26 Width of the Canvas Image. +// 27..29 Height of the Canvas Image. +// There can be extra chunks after the "VP8X" chunk (ICCP, ANMF, VP8, VP8L, +// XMP, EXIF ...) +// All sizes are in little-endian order. +// Note: chunk data size must be padded to multiple of 2 when written. + +// Validates the RIFF container (if detected) and skips over it. +// If a RIFF container is detected, returns: +// VP8_STATUS_BITSTREAM_ERROR for invalid header, +// VP8_STATUS_NOT_ENOUGH_DATA for truncated data if have_all_data is true, +// and VP8_STATUS_OK otherwise. +// In case there are not enough bytes (partial RIFF container), return 0 for +// *riff_size. Else return the RIFF size extracted from the header. +static VP8StatusCode ParseRIFF(const uint8_t** const data, + size_t* const data_size, int have_all_data, + size_t* const riff_size) { + assert(data != NULL); + assert(data_size != NULL); + assert(riff_size != NULL); + + *riff_size = 0; // Default: no RIFF present. + if (*data_size >= RIFF_HEADER_SIZE && !memcmp(*data, "RIFF", TAG_SIZE)) { + if (memcmp(*data + 8, "WEBP", TAG_SIZE)) { + return VP8_STATUS_BITSTREAM_ERROR; // Wrong image file signature. + } else { + const uint32_t size = GetLE32(*data + TAG_SIZE); + // Check that we have at least one chunk (i.e "WEBP" + "VP8?nnnn"). + if (size < TAG_SIZE + CHUNK_HEADER_SIZE) { + return VP8_STATUS_BITSTREAM_ERROR; + } + if (size > MAX_CHUNK_PAYLOAD) { + return VP8_STATUS_BITSTREAM_ERROR; + } + if (have_all_data && (size > *data_size - CHUNK_HEADER_SIZE)) { + return VP8_STATUS_NOT_ENOUGH_DATA; // Truncated bitstream. + } + // We have a RIFF container. Skip it. + *riff_size = size; + *data += RIFF_HEADER_SIZE; + *data_size -= RIFF_HEADER_SIZE; + } + } + return VP8_STATUS_OK; +} + +// Validates the VP8X header and skips over it. +// Returns VP8_STATUS_BITSTREAM_ERROR for invalid VP8X header, +// VP8_STATUS_NOT_ENOUGH_DATA in case of insufficient data, and +// VP8_STATUS_OK otherwise. +// If a VP8X chunk is found, found_vp8x is set to true and *width_ptr, +// *height_ptr and *flags_ptr are set to the corresponding values extracted +// from the VP8X chunk. +static VP8StatusCode ParseVP8X(const uint8_t** const data, + size_t* const data_size, + int* const found_vp8x, + int* const width_ptr, int* const height_ptr, + uint32_t* const flags_ptr) { + const uint32_t vp8x_size = CHUNK_HEADER_SIZE + VP8X_CHUNK_SIZE; + assert(data != NULL); + assert(data_size != NULL); + assert(found_vp8x != NULL); + + *found_vp8x = 0; + + if (*data_size < CHUNK_HEADER_SIZE) { + return VP8_STATUS_NOT_ENOUGH_DATA; // Insufficient data. + } + + if (!memcmp(*data, "VP8X", TAG_SIZE)) { + int width, height; + uint32_t flags; + const uint32_t chunk_size = GetLE32(*data + TAG_SIZE); + if (chunk_size != VP8X_CHUNK_SIZE) { + return VP8_STATUS_BITSTREAM_ERROR; // Wrong chunk size. + } + + // Verify if enough data is available to validate the VP8X chunk. + if (*data_size < vp8x_size) { + return VP8_STATUS_NOT_ENOUGH_DATA; // Insufficient data. + } + flags = GetLE32(*data + 8); + width = 1 + GetLE24(*data + 12); + height = 1 + GetLE24(*data + 15); + if (width * (uint64_t)height >= MAX_IMAGE_AREA) { + return VP8_STATUS_BITSTREAM_ERROR; // image is too large + } + + if (flags_ptr != NULL) *flags_ptr = flags; + if (width_ptr != NULL) *width_ptr = width; + if (height_ptr != NULL) *height_ptr = height; + // Skip over VP8X header bytes. + *data += vp8x_size; + *data_size -= vp8x_size; + *found_vp8x = 1; + } + return VP8_STATUS_OK; +} + +// Skips to the next VP8/VP8L chunk header in the data given the size of the +// RIFF chunk 'riff_size'. +// Returns VP8_STATUS_BITSTREAM_ERROR if any invalid chunk size is encountered, +// VP8_STATUS_NOT_ENOUGH_DATA in case of insufficient data, and +// VP8_STATUS_OK otherwise. +// If an alpha chunk is found, *alpha_data and *alpha_size are set +// appropriately. +static VP8StatusCode ParseOptionalChunks(const uint8_t** const data, + size_t* const data_size, + size_t const riff_size, + const uint8_t** const alpha_data, + size_t* const alpha_size) { + const uint8_t* buf; + size_t buf_size; + uint32_t total_size = TAG_SIZE + // "WEBP". + CHUNK_HEADER_SIZE + // "VP8Xnnnn". + VP8X_CHUNK_SIZE; // data. + assert(data != NULL); + assert(data_size != NULL); + buf = *data; + buf_size = *data_size; + + assert(alpha_data != NULL); + assert(alpha_size != NULL); + *alpha_data = NULL; + *alpha_size = 0; + + while (1) { + uint32_t chunk_size; + uint32_t disk_chunk_size; // chunk_size with padding + + *data = buf; + *data_size = buf_size; + + if (buf_size < CHUNK_HEADER_SIZE) { // Insufficient data. + return VP8_STATUS_NOT_ENOUGH_DATA; + } + + chunk_size = GetLE32(buf + TAG_SIZE); + if (chunk_size > MAX_CHUNK_PAYLOAD) { + return VP8_STATUS_BITSTREAM_ERROR; // Not a valid chunk size. + } + // For odd-sized chunk-payload, there's one byte padding at the end. + disk_chunk_size = (CHUNK_HEADER_SIZE + chunk_size + 1) & ~1; + total_size += disk_chunk_size; + + // Check that total bytes skipped so far does not exceed riff_size. + if (riff_size > 0 && (total_size > riff_size)) { + return VP8_STATUS_BITSTREAM_ERROR; // Not a valid chunk size. + } + + // Start of a (possibly incomplete) VP8/VP8L chunk implies that we have + // parsed all the optional chunks. + // Note: This check must occur before the check 'buf_size < disk_chunk_size' + // below to allow incomplete VP8/VP8L chunks. + if (!memcmp(buf, "VP8 ", TAG_SIZE) || + !memcmp(buf, "VP8L", TAG_SIZE)) { + return VP8_STATUS_OK; + } + + if (buf_size < disk_chunk_size) { // Insufficient data. + return VP8_STATUS_NOT_ENOUGH_DATA; + } + + if (!memcmp(buf, "ALPH", TAG_SIZE)) { // A valid ALPH header. + *alpha_data = buf + CHUNK_HEADER_SIZE; + *alpha_size = chunk_size; + } + + // We have a full and valid chunk; skip it. + buf += disk_chunk_size; + buf_size -= disk_chunk_size; + } +} + +// Validates the VP8/VP8L Header ("VP8 nnnn" or "VP8L nnnn") and skips over it. +// Returns VP8_STATUS_BITSTREAM_ERROR for invalid (chunk larger than +// riff_size) VP8/VP8L header, +// VP8_STATUS_NOT_ENOUGH_DATA in case of insufficient data, and +// VP8_STATUS_OK otherwise. +// If a VP8/VP8L chunk is found, *chunk_size is set to the total number of bytes +// extracted from the VP8/VP8L chunk header. +// The flag '*is_lossless' is set to 1 in case of VP8L chunk / raw VP8L data. +static VP8StatusCode ParseVP8Header(const uint8_t** const data_ptr, + size_t* const data_size, int have_all_data, + size_t riff_size, size_t* const chunk_size, + int* const is_lossless) { + const uint8_t* const data = *data_ptr; + const int is_vp8 = !memcmp(data, "VP8 ", TAG_SIZE); + const int is_vp8l = !memcmp(data, "VP8L", TAG_SIZE); + const uint32_t minimal_size = + TAG_SIZE + CHUNK_HEADER_SIZE; // "WEBP" + "VP8 nnnn" OR + // "WEBP" + "VP8Lnnnn" + assert(data != NULL); + assert(data_size != NULL); + assert(chunk_size != NULL); + assert(is_lossless != NULL); + + if (*data_size < CHUNK_HEADER_SIZE) { + return VP8_STATUS_NOT_ENOUGH_DATA; // Insufficient data. + } + + if (is_vp8 || is_vp8l) { + // Bitstream contains VP8/VP8L header. + const uint32_t size = GetLE32(data + TAG_SIZE); + if ((riff_size >= minimal_size) && (size > riff_size - minimal_size)) { + return VP8_STATUS_BITSTREAM_ERROR; // Inconsistent size information. + } + if (have_all_data && (size > *data_size - CHUNK_HEADER_SIZE)) { + return VP8_STATUS_NOT_ENOUGH_DATA; // Truncated bitstream. + } + // Skip over CHUNK_HEADER_SIZE bytes from VP8/VP8L Header. + *chunk_size = size; + *data_ptr += CHUNK_HEADER_SIZE; + *data_size -= CHUNK_HEADER_SIZE; + *is_lossless = is_vp8l; + } else { + // Raw VP8/VP8L bitstream (no header). + *is_lossless = VP8LCheckSignature(data, *data_size); + *chunk_size = *data_size; + } + + return VP8_STATUS_OK; +} + +//------------------------------------------------------------------------------ + +// Fetch '*width', '*height', '*has_alpha' and fill out 'headers' based on +// 'data'. All the output parameters may be NULL. If 'headers' is NULL only the +// minimal amount will be read to fetch the remaining parameters. +// If 'headers' is non-NULL this function will attempt to locate both alpha +// data (with or without a VP8X chunk) and the bitstream chunk (VP8/VP8L). +// Note: The following chunk sequences (before the raw VP8/VP8L data) are +// considered valid by this function: +// RIFF + VP8(L) +// RIFF + VP8X + (optional chunks) + VP8(L) +// ALPH + VP8 <-- Not a valid WebP format: only allowed for internal purpose. +// VP8(L) <-- Not a valid WebP format: only allowed for internal purpose. +static VP8StatusCode ParseHeadersInternal(const uint8_t* data, + size_t data_size, + int* const width, + int* const height, + int* const has_alpha, + int* const has_animation, + int* const format, + WebPHeaderStructure* const headers) { + int canvas_width = 0; + int canvas_height = 0; + int image_width = 0; + int image_height = 0; + int found_riff = 0; + int found_vp8x = 0; + int animation_present = 0; + const int have_all_data = (headers != NULL) ? headers->have_all_data : 0; + + VP8StatusCode status; + WebPHeaderStructure hdrs; + + if (data == NULL || data_size < RIFF_HEADER_SIZE) { + return VP8_STATUS_NOT_ENOUGH_DATA; + } + memset(&hdrs, 0, sizeof(hdrs)); + hdrs.data = data; + hdrs.data_size = data_size; + + // Skip over RIFF header. + status = ParseRIFF(&data, &data_size, have_all_data, &hdrs.riff_size); + if (status != VP8_STATUS_OK) { + return status; // Wrong RIFF header / insufficient data. + } + found_riff = (hdrs.riff_size > 0); + + // Skip over VP8X. + { + uint32_t flags = 0; + status = ParseVP8X(&data, &data_size, &found_vp8x, + &canvas_width, &canvas_height, &flags); + if (status != VP8_STATUS_OK) { + return status; // Wrong VP8X / insufficient data. + } + animation_present = !!(flags & ANIMATION_FLAG); + if (!found_riff && found_vp8x) { + // Note: This restriction may be removed in the future, if it becomes + // necessary to send VP8X chunk to the decoder. + return VP8_STATUS_BITSTREAM_ERROR; + } + if (has_alpha != NULL) *has_alpha = !!(flags & ALPHA_FLAG); + if (has_animation != NULL) *has_animation = animation_present; + if (format != NULL) *format = 0; // default = undefined + + image_width = canvas_width; + image_height = canvas_height; + if (found_vp8x && animation_present && headers == NULL) { + status = VP8_STATUS_OK; + goto ReturnWidthHeight; // Just return features from VP8X header. + } + } + + if (data_size < TAG_SIZE) { + status = VP8_STATUS_NOT_ENOUGH_DATA; + goto ReturnWidthHeight; + } + + // Skip over optional chunks if data started with "RIFF + VP8X" or "ALPH". + if ((found_riff && found_vp8x) || + (!found_riff && !found_vp8x && !memcmp(data, "ALPH", TAG_SIZE))) { + status = ParseOptionalChunks(&data, &data_size, hdrs.riff_size, + &hdrs.alpha_data, &hdrs.alpha_data_size); + if (status != VP8_STATUS_OK) { + goto ReturnWidthHeight; // Invalid chunk size / insufficient data. + } + } + + // Skip over VP8/VP8L header. + status = ParseVP8Header(&data, &data_size, have_all_data, hdrs.riff_size, + &hdrs.compressed_size, &hdrs.is_lossless); + if (status != VP8_STATUS_OK) { + goto ReturnWidthHeight; // Wrong VP8/VP8L chunk-header / insufficient data. + } + if (hdrs.compressed_size > MAX_CHUNK_PAYLOAD) { + return VP8_STATUS_BITSTREAM_ERROR; + } + + if (format != NULL && !animation_present) { + *format = hdrs.is_lossless ? 2 : 1; + } + + if (!hdrs.is_lossless) { + if (data_size < VP8_FRAME_HEADER_SIZE) { + status = VP8_STATUS_NOT_ENOUGH_DATA; + goto ReturnWidthHeight; + } + // Validates raw VP8 data. + if (!VP8GetInfo(data, data_size, (uint32_t)hdrs.compressed_size, + &image_width, &image_height)) { + return VP8_STATUS_BITSTREAM_ERROR; + } + } else { + if (data_size < VP8L_FRAME_HEADER_SIZE) { + status = VP8_STATUS_NOT_ENOUGH_DATA; + goto ReturnWidthHeight; + } + // Validates raw VP8L data. + if (!VP8LGetInfo(data, data_size, &image_width, &image_height, has_alpha)) { + return VP8_STATUS_BITSTREAM_ERROR; + } + } + // Validates image size coherency. + if (found_vp8x) { + if (canvas_width != image_width || canvas_height != image_height) { + return VP8_STATUS_BITSTREAM_ERROR; + } + } + if (headers != NULL) { + *headers = hdrs; + headers->offset = data - headers->data; + assert((uint64_t)(data - headers->data) < MAX_CHUNK_PAYLOAD); + assert(headers->offset == headers->data_size - data_size); + } + ReturnWidthHeight: + if (status == VP8_STATUS_OK || + (status == VP8_STATUS_NOT_ENOUGH_DATA && found_vp8x && headers == NULL)) { + if (has_alpha != NULL) { + // If the data did not contain a VP8X/VP8L chunk the only definitive way + // to set this is by looking for alpha data (from an ALPH chunk). + *has_alpha |= (hdrs.alpha_data != NULL); + } + if (width != NULL) *width = image_width; + if (height != NULL) *height = image_height; + return VP8_STATUS_OK; + } else { + return status; + } +} + +VP8StatusCode WebPParseHeaders(WebPHeaderStructure* const headers) { + // status is marked volatile as a workaround for a clang-3.8 (aarch64) bug + volatile VP8StatusCode status; + int has_animation = 0; + assert(headers != NULL); + // fill out headers, ignore width/height/has_alpha. + status = ParseHeadersInternal(headers->data, headers->data_size, + NULL, NULL, NULL, &has_animation, + NULL, headers); + if (status == VP8_STATUS_OK || status == VP8_STATUS_NOT_ENOUGH_DATA) { + // The WebPDemux API + libwebp can be used to decode individual + // uncomposited frames or the WebPAnimDecoder can be used to fully + // reconstruct them (see webp/demux.h). + if (has_animation) { + status = VP8_STATUS_UNSUPPORTED_FEATURE; + } + } + return status; +} + +//------------------------------------------------------------------------------ +// WebPDecParams + +void WebPResetDecParams(WebPDecParams* const params) { + if (params != NULL) { + memset(params, 0, sizeof(*params)); + } +} + +//------------------------------------------------------------------------------ +// "Into" decoding variants + +// Main flow +static VP8StatusCode DecodeInto(const uint8_t* const data, size_t data_size, + WebPDecParams* const params) { + VP8StatusCode status; + VP8Io io; + WebPHeaderStructure headers; + + headers.data = data; + headers.data_size = data_size; + headers.have_all_data = 1; + status = WebPParseHeaders(&headers); // Process Pre-VP8 chunks. + if (status != VP8_STATUS_OK) { + return status; + } + + assert(params != NULL); + VP8InitIo(&io); + io.data = headers.data + headers.offset; + io.data_size = headers.data_size - headers.offset; + WebPInitCustomIo(params, &io); // Plug the I/O functions. + + if (!headers.is_lossless) { + VP8Decoder* const dec = VP8New(); + if (dec == NULL) { + return VP8_STATUS_OUT_OF_MEMORY; + } + dec->alpha_data_ = headers.alpha_data; + dec->alpha_data_size_ = headers.alpha_data_size; + + // Decode bitstream header, update io->width/io->height. + if (!VP8GetHeaders(dec, &io)) { + status = dec->status_; // An error occurred. Grab error status. + } else { + // Allocate/check output buffers. + status = WebPAllocateDecBuffer(io.width, io.height, params->options, + params->output); + if (status == VP8_STATUS_OK) { // Decode + // This change must be done before calling VP8Decode() + dec->mt_method_ = VP8GetThreadMethod(params->options, &headers, + io.width, io.height); + VP8InitDithering(params->options, dec); + if (!VP8Decode(dec, &io)) { + status = dec->status_; + } + } + } + VP8Delete(dec); + } else { + VP8LDecoder* const dec = VP8LNew(); + if (dec == NULL) { + return VP8_STATUS_OUT_OF_MEMORY; + } + if (!VP8LDecodeHeader(dec, &io)) { + status = dec->status_; // An error occurred. Grab error status. + } else { + // Allocate/check output buffers. + status = WebPAllocateDecBuffer(io.width, io.height, params->options, + params->output); + if (status == VP8_STATUS_OK) { // Decode + if (!VP8LDecodeImage(dec)) { + status = dec->status_; + } + } + } + VP8LDelete(dec); + } + + if (status != VP8_STATUS_OK) { + WebPFreeDecBuffer(params->output); + } else { + if (params->options != NULL && params->options->flip) { + // This restores the original stride values if options->flip was used + // during the call to WebPAllocateDecBuffer above. + status = WebPFlipBuffer(params->output); + } + } + return status; +} + +// Helpers +static uint8_t* DecodeIntoRGBABuffer(WEBP_CSP_MODE colorspace, + const uint8_t* const data, + size_t data_size, + uint8_t* const rgba, + int stride, size_t size) { + WebPDecParams params; + WebPDecBuffer buf; + if (rgba == NULL) { + return NULL; + } + WebPInitDecBuffer(&buf); + WebPResetDecParams(¶ms); + params.output = &buf; + buf.colorspace = colorspace; + buf.u.RGBA.rgba = rgba; + buf.u.RGBA.stride = stride; + buf.u.RGBA.size = size; + buf.is_external_memory = 1; + if (DecodeInto(data, data_size, ¶ms) != VP8_STATUS_OK) { + return NULL; + } + return rgba; +} + +uint8_t* WebPDecodeRGBInto(const uint8_t* data, size_t data_size, + uint8_t* output, size_t size, int stride) { + return DecodeIntoRGBABuffer(MODE_RGB, data, data_size, output, stride, size); +} + +uint8_t* WebPDecodeRGBAInto(const uint8_t* data, size_t data_size, + uint8_t* output, size_t size, int stride) { + return DecodeIntoRGBABuffer(MODE_RGBA, data, data_size, output, stride, size); +} + +uint8_t* WebPDecodeARGBInto(const uint8_t* data, size_t data_size, + uint8_t* output, size_t size, int stride) { + return DecodeIntoRGBABuffer(MODE_ARGB, data, data_size, output, stride, size); +} + +uint8_t* WebPDecodeBGRInto(const uint8_t* data, size_t data_size, + uint8_t* output, size_t size, int stride) { + return DecodeIntoRGBABuffer(MODE_BGR, data, data_size, output, stride, size); +} + +uint8_t* WebPDecodeBGRAInto(const uint8_t* data, size_t data_size, + uint8_t* output, size_t size, int stride) { + return DecodeIntoRGBABuffer(MODE_BGRA, data, data_size, output, stride, size); +} + +uint8_t* WebPDecodeYUVInto(const uint8_t* data, size_t data_size, + uint8_t* luma, size_t luma_size, int luma_stride, + uint8_t* u, size_t u_size, int u_stride, + uint8_t* v, size_t v_size, int v_stride) { + WebPDecParams params; + WebPDecBuffer output; + if (luma == NULL) return NULL; + WebPInitDecBuffer(&output); + WebPResetDecParams(¶ms); + params.output = &output; + output.colorspace = MODE_YUV; + output.u.YUVA.y = luma; + output.u.YUVA.y_stride = luma_stride; + output.u.YUVA.y_size = luma_size; + output.u.YUVA.u = u; + output.u.YUVA.u_stride = u_stride; + output.u.YUVA.u_size = u_size; + output.u.YUVA.v = v; + output.u.YUVA.v_stride = v_stride; + output.u.YUVA.v_size = v_size; + output.is_external_memory = 1; + if (DecodeInto(data, data_size, ¶ms) != VP8_STATUS_OK) { + return NULL; + } + return luma; +} + +//------------------------------------------------------------------------------ + +static uint8_t* Decode(WEBP_CSP_MODE mode, const uint8_t* const data, + size_t data_size, int* const width, int* const height, + WebPDecBuffer* const keep_info) { + WebPDecParams params; + WebPDecBuffer output; + + WebPInitDecBuffer(&output); + WebPResetDecParams(¶ms); + params.output = &output; + output.colorspace = mode; + + // Retrieve (and report back) the required dimensions from bitstream. + if (!WebPGetInfo(data, data_size, &output.width, &output.height)) { + return NULL; + } + if (width != NULL) *width = output.width; + if (height != NULL) *height = output.height; + + // Decode + if (DecodeInto(data, data_size, ¶ms) != VP8_STATUS_OK) { + return NULL; + } + if (keep_info != NULL) { // keep track of the side-info + WebPCopyDecBuffer(&output, keep_info); + } + // return decoded samples (don't clear 'output'!) + return WebPIsRGBMode(mode) ? output.u.RGBA.rgba : output.u.YUVA.y; +} + +uint8_t* WebPDecodeRGB(const uint8_t* data, size_t data_size, + int* width, int* height) { + return Decode(MODE_RGB, data, data_size, width, height, NULL); +} + +uint8_t* WebPDecodeRGBA(const uint8_t* data, size_t data_size, + int* width, int* height) { + return Decode(MODE_RGBA, data, data_size, width, height, NULL); +} + +uint8_t* WebPDecodeARGB(const uint8_t* data, size_t data_size, + int* width, int* height) { + return Decode(MODE_ARGB, data, data_size, width, height, NULL); +} + +uint8_t* WebPDecodeBGR(const uint8_t* data, size_t data_size, + int* width, int* height) { + return Decode(MODE_BGR, data, data_size, width, height, NULL); +} + +uint8_t* WebPDecodeBGRA(const uint8_t* data, size_t data_size, + int* width, int* height) { + return Decode(MODE_BGRA, data, data_size, width, height, NULL); +} + +uint8_t* WebPDecodeYUV(const uint8_t* data, size_t data_size, + int* width, int* height, uint8_t** u, uint8_t** v, + int* stride, int* uv_stride) { + WebPDecBuffer output; // only to preserve the side-infos + uint8_t* const out = Decode(MODE_YUV, data, data_size, + width, height, &output); + + if (out != NULL) { + const WebPYUVABuffer* const buf = &output.u.YUVA; + *u = buf->u; + *v = buf->v; + *stride = buf->y_stride; + *uv_stride = buf->u_stride; + assert(buf->u_stride == buf->v_stride); + } + return out; +} + +static void DefaultFeatures(WebPBitstreamFeatures* const features) { + assert(features != NULL); + memset(features, 0, sizeof(*features)); +} + +static VP8StatusCode GetFeatures(const uint8_t* const data, size_t data_size, + WebPBitstreamFeatures* const features) { + if (features == NULL || data == NULL) { + return VP8_STATUS_INVALID_PARAM; + } + DefaultFeatures(features); + + // Only parse enough of the data to retrieve the features. + return ParseHeadersInternal(data, data_size, + &features->width, &features->height, + &features->has_alpha, &features->has_animation, + &features->format, NULL); +} + +//------------------------------------------------------------------------------ +// WebPGetInfo() + +int WebPGetInfo(const uint8_t* data, size_t data_size, + int* width, int* height) { + WebPBitstreamFeatures features; + + if (GetFeatures(data, data_size, &features) != VP8_STATUS_OK) { + return 0; + } + + if (width != NULL) { + *width = features.width; + } + if (height != NULL) { + *height = features.height; + } + + return 1; +} + +//------------------------------------------------------------------------------ +// Advance decoding API + +int WebPInitDecoderConfigInternal(WebPDecoderConfig* config, + int version) { + if (WEBP_ABI_IS_INCOMPATIBLE(version, WEBP_DECODER_ABI_VERSION)) { + return 0; // version mismatch + } + if (config == NULL) { + return 0; + } + memset(config, 0, sizeof(*config)); + DefaultFeatures(&config->input); + WebPInitDecBuffer(&config->output); + return 1; +} + +VP8StatusCode WebPGetFeaturesInternal(const uint8_t* data, size_t data_size, + WebPBitstreamFeatures* features, + int version) { + if (WEBP_ABI_IS_INCOMPATIBLE(version, WEBP_DECODER_ABI_VERSION)) { + return VP8_STATUS_INVALID_PARAM; // version mismatch + } + if (features == NULL) { + return VP8_STATUS_INVALID_PARAM; + } + return GetFeatures(data, data_size, features); +} + +VP8StatusCode WebPDecode(const uint8_t* data, size_t data_size, + WebPDecoderConfig* config) { + WebPDecParams params; + VP8StatusCode status; + + if (config == NULL) { + return VP8_STATUS_INVALID_PARAM; + } + + status = GetFeatures(data, data_size, &config->input); + if (status != VP8_STATUS_OK) { + if (status == VP8_STATUS_NOT_ENOUGH_DATA) { + return VP8_STATUS_BITSTREAM_ERROR; // Not-enough-data treated as error. + } + return status; + } + + WebPResetDecParams(¶ms); + params.options = &config->options; + params.output = &config->output; + if (WebPAvoidSlowMemory(params.output, &config->input)) { + // decoding to slow memory: use a temporary in-mem buffer to decode into. + WebPDecBuffer in_mem_buffer; + WebPInitDecBuffer(&in_mem_buffer); + in_mem_buffer.colorspace = config->output.colorspace; + in_mem_buffer.width = config->input.width; + in_mem_buffer.height = config->input.height; + params.output = &in_mem_buffer; + status = DecodeInto(data, data_size, ¶ms); + if (status == VP8_STATUS_OK) { // do the slow-copy + status = WebPCopyDecBufferPixels(&in_mem_buffer, &config->output); + } + WebPFreeDecBuffer(&in_mem_buffer); + } else { + status = DecodeInto(data, data_size, ¶ms); + } + + return status; +} + +//------------------------------------------------------------------------------ +// Cropping and rescaling. + +int WebPIoInitFromOptions(const WebPDecoderOptions* const options, + VP8Io* const io, WEBP_CSP_MODE src_colorspace) { + const int W = io->width; + const int H = io->height; + int x = 0, y = 0, w = W, h = H; + + // Cropping + io->use_cropping = (options != NULL) && (options->use_cropping > 0); + if (io->use_cropping) { + w = options->crop_width; + h = options->crop_height; + x = options->crop_left; + y = options->crop_top; + if (!WebPIsRGBMode(src_colorspace)) { // only snap for YUV420 + x &= ~1; + y &= ~1; + } + if (x < 0 || y < 0 || w <= 0 || h <= 0 || x + w > W || y + h > H) { + return 0; // out of frame boundary error + } + } + io->crop_left = x; + io->crop_top = y; + io->crop_right = x + w; + io->crop_bottom = y + h; + io->mb_w = w; + io->mb_h = h; + + // Scaling + io->use_scaling = (options != NULL) && (options->use_scaling > 0); + if (io->use_scaling) { + int scaled_width = options->scaled_width; + int scaled_height = options->scaled_height; + if (!WebPRescalerGetScaledDimensions(w, h, &scaled_width, &scaled_height)) { + return 0; + } + io->scaled_width = scaled_width; + io->scaled_height = scaled_height; + } + + // Filter + io->bypass_filtering = (options != NULL) && options->bypass_filtering; + + // Fancy upsampler +#ifdef FANCY_UPSAMPLING + io->fancy_upsampling = (options == NULL) || (!options->no_fancy_upsampling); +#endif + + if (io->use_scaling) { + // disable filter (only for large downscaling ratio). + io->bypass_filtering = (io->scaled_width < W * 3 / 4) && + (io->scaled_height < H * 3 / 4); + io->fancy_upsampling = 0; + } + return 1; +} + +//------------------------------------------------------------------------------ diff --git a/thirdparty/libwebp/src/dec/webpi_dec.h b/thirdparty/libwebp/src/dec/webpi_dec.h new file mode 100644 index 0000000000..c378ba6fc3 --- /dev/null +++ b/thirdparty/libwebp/src/dec/webpi_dec.h @@ -0,0 +1,133 @@ +// Copyright 2011 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. +// ----------------------------------------------------------------------------- +// +// Internal header: WebP decoding parameters and custom IO on buffer +// +// Author: somnath@google.com (Somnath Banerjee) + +#ifndef WEBP_DEC_WEBPI_DEC_H_ +#define WEBP_DEC_WEBPI_DEC_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +#include "src/utils/rescaler_utils.h" +#include "src/dec/vp8_dec.h" + +//------------------------------------------------------------------------------ +// WebPDecParams: Decoding output parameters. Transient internal object. + +typedef struct WebPDecParams WebPDecParams; +typedef int (*OutputFunc)(const VP8Io* const io, WebPDecParams* const p); +typedef int (*OutputAlphaFunc)(const VP8Io* const io, WebPDecParams* const p, + int expected_num_out_lines); +typedef int (*OutputRowFunc)(WebPDecParams* const p, int y_pos, + int max_out_lines); + +struct WebPDecParams { + WebPDecBuffer* output; // output buffer. + uint8_t* tmp_y, *tmp_u, *tmp_v; // cache for the fancy upsampler + // or used for tmp rescaling + + int last_y; // coordinate of the line that was last output + const WebPDecoderOptions* options; // if not NULL, use alt decoding features + + WebPRescaler* scaler_y, *scaler_u, *scaler_v, *scaler_a; // rescalers + void* memory; // overall scratch memory for the output work. + + OutputFunc emit; // output RGB or YUV samples + OutputAlphaFunc emit_alpha; // output alpha channel + OutputRowFunc emit_alpha_row; // output one line of rescaled alpha values +}; + +// Should be called first, before any use of the WebPDecParams object. +void WebPResetDecParams(WebPDecParams* const params); + +//------------------------------------------------------------------------------ +// Header parsing helpers + +// Structure storing a description of the RIFF headers. +typedef struct { + const uint8_t* data; // input buffer + size_t data_size; // input buffer size + int have_all_data; // true if all data is known to be available + size_t offset; // offset to main data chunk (VP8 or VP8L) + const uint8_t* alpha_data; // points to alpha chunk (if present) + size_t alpha_data_size; // alpha chunk size + size_t compressed_size; // VP8/VP8L compressed data size + size_t riff_size; // size of the riff payload (or 0 if absent) + int is_lossless; // true if a VP8L chunk is present +} WebPHeaderStructure; + +// Skips over all valid chunks prior to the first VP8/VP8L frame header. +// Returns: VP8_STATUS_OK, VP8_STATUS_BITSTREAM_ERROR (invalid header/chunk), +// VP8_STATUS_NOT_ENOUGH_DATA (partial input) or VP8_STATUS_UNSUPPORTED_FEATURE +// in the case of non-decodable features (animation for instance). +// In 'headers', compressed_size, offset, alpha_data, alpha_size, and lossless +// fields are updated appropriately upon success. +VP8StatusCode WebPParseHeaders(WebPHeaderStructure* const headers); + +//------------------------------------------------------------------------------ +// Misc utils + +// Initializes VP8Io with custom setup, io and teardown functions. The default +// hooks will use the supplied 'params' as io->opaque handle. +void WebPInitCustomIo(WebPDecParams* const params, VP8Io* const io); + +// Setup crop_xxx fields, mb_w and mb_h in io. 'src_colorspace' refers +// to the *compressed* format, not the output one. +int WebPIoInitFromOptions(const WebPDecoderOptions* const options, + VP8Io* const io, WEBP_CSP_MODE src_colorspace); + +//------------------------------------------------------------------------------ +// Internal functions regarding WebPDecBuffer memory (in buffer.c). +// Don't really need to be externally visible for now. + +// Prepare 'buffer' with the requested initial dimensions width/height. +// If no external storage is supplied, initializes buffer by allocating output +// memory and setting up the stride information. Validate the parameters. Return +// an error code in case of problem (no memory, or invalid stride / size / +// dimension / etc.). If *options is not NULL, also verify that the options' +// parameters are valid and apply them to the width/height dimensions of the +// output buffer. This takes cropping / scaling / rotation into account. +// Also incorporates the options->flip flag to flip the buffer parameters if +// needed. +VP8StatusCode WebPAllocateDecBuffer(int width, int height, + const WebPDecoderOptions* const options, + WebPDecBuffer* const buffer); + +// Flip buffer vertically by negating the various strides. +VP8StatusCode WebPFlipBuffer(WebPDecBuffer* const buffer); + +// Copy 'src' into 'dst' buffer, making sure 'dst' is not marked as owner of the +// memory (still held by 'src'). No pixels are copied. +void WebPCopyDecBuffer(const WebPDecBuffer* const src, + WebPDecBuffer* const dst); + +// Copy and transfer ownership from src to dst (beware of parameter order!) +void WebPGrabDecBuffer(WebPDecBuffer* const src, WebPDecBuffer* const dst); + +// Copy pixels from 'src' into a *preallocated* 'dst' buffer. Returns +// VP8_STATUS_INVALID_PARAM if the 'dst' is not set up correctly for the copy. +VP8StatusCode WebPCopyDecBufferPixels(const WebPDecBuffer* const src, + WebPDecBuffer* const dst); + +// Returns true if decoding will be slow with the current configuration +// and bitstream features. +int WebPAvoidSlowMemory(const WebPDecBuffer* const output, + const WebPBitstreamFeatures* const features); + +//------------------------------------------------------------------------------ + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif /* WEBP_DEC_WEBPI_DEC_H_ */ |