diff options
Diffstat (limited to 'drivers/webp/dsp/dsp.h')
| -rw-r--r-- | drivers/webp/dsp/dsp.h | 368 |
1 files changed, 329 insertions, 39 deletions
diff --git a/drivers/webp/dsp/dsp.h b/drivers/webp/dsp/dsp.h index fd686a8532..4613d9c3ff 100644 --- a/drivers/webp/dsp/dsp.h +++ b/drivers/webp/dsp/dsp.h @@ -1,8 +1,10 @@ // Copyright 2011 Google Inc. All Rights Reserved. // -// This code is licensed under the same terms as WebM: -// Software License Agreement: http://www.webmproject.org/license/software/ -// Additional IP Rights Grant: http://www.webmproject.org/license/additional/ +// 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. // ----------------------------------------------------------------------------- // // Speed-critical functions. @@ -12,44 +14,121 @@ #ifndef WEBP_DSP_DSP_H_ #define WEBP_DSP_DSP_H_ -#include "../types.h" +#ifdef HAVE_CONFIG_H +#include "webp/config.h" +#endif + +#include "webp/types.h" -#if defined(__cplusplus) || defined(c_plusplus) +#ifdef __cplusplus extern "C" { #endif +#define BPS 32 // this is the common stride for enc/dec + //------------------------------------------------------------------------------ // CPU detection -#if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86)) +#if defined(__GNUC__) +# define LOCAL_GCC_VERSION ((__GNUC__ << 8) | __GNUC_MINOR__) +# define LOCAL_GCC_PREREQ(maj, min) \ + (LOCAL_GCC_VERSION >= (((maj) << 8) | (min))) +#else +# define LOCAL_GCC_VERSION 0 +# define LOCAL_GCC_PREREQ(maj, min) 0 +#endif + +#ifndef __has_builtin +# define __has_builtin(x) 0 +#endif + +#if defined(_MSC_VER) && _MSC_VER > 1310 && \ + (defined(_M_X64) || defined(_M_IX86)) #define WEBP_MSC_SSE2 // Visual C++ SSE2 targets #endif -#if defined(__SSE2__) || defined(WEBP_MSC_SSE2) +#if defined(_MSC_VER) && _MSC_VER >= 1500 && \ + (defined(_M_X64) || defined(_M_IX86)) +#define WEBP_MSC_SSE41 // Visual C++ SSE4.1 targets +#endif + +// WEBP_HAVE_* are used to indicate the presence of the instruction set in dsp +// files without intrinsics, allowing the corresponding Init() to be called. +// Files containing intrinsics will need to be built targeting the instruction +// set so should succeed on one of the earlier tests. +#if defined(__SSE2__) || defined(WEBP_MSC_SSE2) || defined(WEBP_HAVE_SSE2) #define WEBP_USE_SSE2 #endif -#if defined(__ANDROID__) && defined(__ARM_ARCH_7A__) && defined(__ARM_NEON__) +#if defined(__SSE4_1__) || defined(WEBP_MSC_SSE41) || defined(WEBP_HAVE_SSE41) +#define WEBP_USE_SSE41 +#endif + +#if defined(__AVX2__) || defined(WEBP_HAVE_AVX2) +#define WEBP_USE_AVX2 +#endif + +#if defined(__ANDROID__) && defined(__ARM_ARCH_7A__) #define WEBP_ANDROID_NEON // Android targets that might support NEON #endif -#if ( (defined(__ARM_NEON__) && !defined(__aarch64__)) || defined(WEBP_ANDROID_NEON)) && !defined(PSP2_ENABLED) +// The intrinsics currently cause compiler errors with arm-nacl-gcc and the +// inline assembly would need to be modified for use with Native Client. +#if (defined(__ARM_NEON__) || defined(WEBP_ANDROID_NEON) || \ + defined(__aarch64__)) && !defined(__native_client__) +#define WEBP_USE_NEON +#endif + +#if defined(_MSC_VER) && _MSC_VER >= 1700 && defined(_M_ARM) #define WEBP_USE_NEON +#define WEBP_USE_INTRINSICS +#endif + +#if defined(__mips__) && !defined(__mips64) && \ + defined(__mips_isa_rev) && (__mips_isa_rev >= 1) && (__mips_isa_rev < 6) +#define WEBP_USE_MIPS32 +#if (__mips_isa_rev >= 2) +#define WEBP_USE_MIPS32_R2 +#if defined(__mips_dspr2) || (__mips_dsp_rev >= 2) +#define WEBP_USE_MIPS_DSP_R2 +#endif +#endif +#endif + +// This macro prevents thread_sanitizer from reporting known concurrent writes. +#define WEBP_TSAN_IGNORE_FUNCTION +#if defined(__has_feature) +#if __has_feature(thread_sanitizer) +#undef WEBP_TSAN_IGNORE_FUNCTION +#define WEBP_TSAN_IGNORE_FUNCTION __attribute__((no_sanitize_thread)) +#endif #endif typedef enum { kSSE2, kSSE3, - kNEON + kSSE4_1, + kAVX, + kAVX2, + kNEON, + kMIPS32, + kMIPSdspR2 } CPUFeature; // returns true if the CPU supports the feature. typedef int (*VP8CPUInfo)(CPUFeature feature); -extern VP8CPUInfo VP8GetCPUInfo; +WEBP_EXTERN(VP8CPUInfo) VP8GetCPUInfo; //------------------------------------------------------------------------------ -// Encoding +// Init stub generator -int VP8GetAlpha(const int histo[]); +// Defines an init function stub to ensure each module exposes a symbol, +// avoiding a compiler warning. +#define WEBP_DSP_INIT_STUB(func) \ + extern void func(void); \ + WEBP_TSAN_IGNORE_FUNCTION void func(void) {} + +//------------------------------------------------------------------------------ +// Encoding // Transforms // VP8Idct: Does one of two inverse transforms. If do_two is set, the transforms @@ -60,7 +139,7 @@ typedef void (*VP8Fdct)(const uint8_t* src, const uint8_t* ref, int16_t* out); typedef void (*VP8WHT)(const int16_t* in, int16_t* out); extern VP8Idct VP8ITransform; extern VP8Fdct VP8FTransform; -extern VP8WHT VP8ITransformWHT; +extern VP8Fdct VP8FTransform2; // performs two transforms at a time extern VP8WHT VP8FTransformWHT; // Predictions // *dst is the destination block. *top and *left can be NULL. @@ -79,20 +158,63 @@ extern VP8WMetric VP8TDisto4x4, VP8TDisto16x16; typedef void (*VP8BlockCopy)(const uint8_t* src, uint8_t* dst); extern VP8BlockCopy VP8Copy4x4; +extern VP8BlockCopy VP8Copy16x8; // Quantization struct VP8Matrix; // forward declaration typedef int (*VP8QuantizeBlock)(int16_t in[16], int16_t out[16], - int n, const struct VP8Matrix* const mtx); + const struct VP8Matrix* const mtx); +// Same as VP8QuantizeBlock, but quantizes two consecutive blocks. +typedef int (*VP8Quantize2Blocks)(int16_t in[32], int16_t out[32], + const struct VP8Matrix* const mtx); + extern VP8QuantizeBlock VP8EncQuantizeBlock; +extern VP8Quantize2Blocks VP8EncQuantize2Blocks; + +// specific to 2nd transform: +typedef int (*VP8QuantizeBlockWHT)(int16_t in[16], int16_t out[16], + const struct VP8Matrix* const mtx); +extern VP8QuantizeBlockWHT VP8EncQuantizeBlockWHT; -// Compute susceptibility based on DCT-coeff histograms: -// the higher, the "easier" the macroblock is to compress. -typedef int (*VP8CHisto)(const uint8_t* ref, const uint8_t* pred, - int start_block, int end_block); extern const int VP8DspScan[16 + 4 + 4]; + +// Collect histogram for susceptibility calculation. +#define MAX_COEFF_THRESH 31 // size of histogram used by CollectHistogram. +typedef struct { + // We only need to store max_value and last_non_zero, not the distribution. + int max_value; + int last_non_zero; +} VP8Histogram; +typedef void (*VP8CHisto)(const uint8_t* ref, const uint8_t* pred, + int start_block, int end_block, + VP8Histogram* const histo); extern VP8CHisto VP8CollectHistogram; +// General-purpose util function to help VP8CollectHistogram(). +void VP8SetHistogramData(const int distribution[MAX_COEFF_THRESH + 1], + VP8Histogram* const histo); + +// must be called before using any of the above +void VP8EncDspInit(void); + +//------------------------------------------------------------------------------ +// cost functions (encoding) + +extern const uint16_t VP8EntropyCost[256]; // 8bit fixed-point log(p) +// approximate cost per level: +extern const uint16_t VP8LevelFixedCosts[2047 /*MAX_LEVEL*/ + 1]; +extern const uint8_t VP8EncBands[16 + 1]; + +struct VP8Residual; +typedef void (*VP8SetResidualCoeffsFunc)(const int16_t* const coeffs, + struct VP8Residual* const res); +extern VP8SetResidualCoeffsFunc VP8SetResidualCoeffs; -void VP8EncDspInit(void); // must be called before using any of the above +// Cost calculation function. +typedef int (*VP8GetResidualCostFunc)(int ctx0, + const struct VP8Residual* const res); +extern VP8GetResidualCostFunc VP8GetResidualCost; + +// must be called before anything using the above +void VP8EncDspCostInit(void); //------------------------------------------------------------------------------ // Decoding @@ -101,17 +223,26 @@ typedef void (*VP8DecIdct)(const int16_t* coeffs, uint8_t* dst); // when doing two transforms, coeffs is actually int16_t[2][16]. typedef void (*VP8DecIdct2)(const int16_t* coeffs, uint8_t* dst, int do_two); extern VP8DecIdct2 VP8Transform; +extern VP8DecIdct VP8TransformAC3; extern VP8DecIdct VP8TransformUV; extern VP8DecIdct VP8TransformDC; extern VP8DecIdct VP8TransformDCUV; -extern void (*VP8TransformWHT)(const int16_t* in, int16_t* out); +extern VP8WHT VP8TransformWHT; // *dst is the destination block, with stride BPS. Boundary samples are // assumed accessible when needed. typedef void (*VP8PredFunc)(uint8_t* dst); -extern const VP8PredFunc VP8PredLuma16[/* NUM_B_DC_MODES */]; -extern const VP8PredFunc VP8PredChroma8[/* NUM_B_DC_MODES */]; -extern const VP8PredFunc VP8PredLuma4[/* NUM_BMODES */]; +extern VP8PredFunc VP8PredLuma16[/* NUM_B_DC_MODES */]; +extern VP8PredFunc VP8PredChroma8[/* NUM_B_DC_MODES */]; +extern VP8PredFunc VP8PredLuma4[/* NUM_BMODES */]; + +// clipping tables (for filtering) +extern const int8_t* const VP8ksclip1; // clips [-1020, 1020] to [-128, 127] +extern const int8_t* const VP8ksclip2; // clips [-112, 112] to [-16, 15] +extern const uint8_t* const VP8kclip1; // clips [-255,511] to [0,255] +extern const uint8_t* const VP8kabs0; // abs(x) for x in [-255,255] +// must be called first +void VP8InitClipTables(void); // simple filter (only for luma) typedef void (*VP8SimpleFilterFunc)(uint8_t* p, int stride, int thresh); @@ -145,6 +276,8 @@ void VP8DspInit(void); #define FANCY_UPSAMPLING // undefined to remove fancy upsampling support +// Convert a pair of y/u/v lines together to the output rgb/a colorspace. +// bottom_y can be NULL if only one line of output is needed (at top/bottom). typedef void (*WebPUpsampleLinePairFunc)( const uint8_t* top_y, const uint8_t* bottom_y, const uint8_t* top_u, const uint8_t* top_v, @@ -156,18 +289,20 @@ typedef void (*WebPUpsampleLinePairFunc)( // Fancy upsampling functions to convert YUV to RGB(A) modes extern WebPUpsampleLinePairFunc WebPUpsamplers[/* MODE_LAST */]; -// Initializes SSE2 version of the fancy upsamplers. -void WebPInitUpsamplersSSE2(void); - #endif // FANCY_UPSAMPLING -// Point-sampling methods. -typedef void (*WebPSampleLinePairFunc)( - const uint8_t* top_y, const uint8_t* bottom_y, - const uint8_t* u, const uint8_t* v, - uint8_t* top_dst, uint8_t* bottom_dst, int len); +// Per-row point-sampling methods. +typedef void (*WebPSamplerRowFunc)(const uint8_t* y, + const uint8_t* u, const uint8_t* v, + uint8_t* dst, int len); +// Generic function to apply 'WebPSamplerRowFunc' to the whole plane: +void WebPSamplerProcessPlane(const uint8_t* y, int y_stride, + const uint8_t* u, const uint8_t* v, int uv_stride, + uint8_t* dst, int dst_stride, + int width, int height, WebPSamplerRowFunc func); -extern const WebPSampleLinePairFunc WebPSamplers[/* MODE_LAST */]; +// Sampling functions to convert rows of YUV to RGB(A) +extern WebPSamplerRowFunc WebPSamplers[/* MODE_LAST */]; // General function for converting two lines of ARGB or RGBA. // 'alpha_is_last' should be true if 0xff000000 is stored in memory as @@ -179,13 +314,84 @@ typedef void (*WebPYUV444Converter)(const uint8_t* y, const uint8_t* u, const uint8_t* v, uint8_t* dst, int len); -extern const WebPYUV444Converter WebPYUV444Converters[/* MODE_LAST */]; +extern WebPYUV444Converter WebPYUV444Converters[/* MODE_LAST */]; -// Main function to be called +// Must be called before using the WebPUpsamplers[] (and for premultiplied +// colorspaces like rgbA, rgbA4444, etc) void WebPInitUpsamplers(void); +// Must be called before using WebPSamplers[] +void WebPInitSamplers(void); +// Must be called before using WebPYUV444Converters[] +void WebPInitYUV444Converters(void); //------------------------------------------------------------------------------ -// Pre-multiply planes with alpha values +// ARGB -> YUV converters + +// Convert ARGB samples to luma Y. +extern void (*WebPConvertARGBToY)(const uint32_t* argb, uint8_t* y, int width); +// Convert ARGB samples to U/V with downsampling. do_store should be '1' for +// even lines and '0' for odd ones. 'src_width' is the original width, not +// the U/V one. +extern void (*WebPConvertARGBToUV)(const uint32_t* argb, uint8_t* u, uint8_t* v, + int src_width, int do_store); + +// Convert a row of accumulated (four-values) of rgba32 toward U/V +extern void (*WebPConvertRGBA32ToUV)(const uint16_t* rgb, + uint8_t* u, uint8_t* v, int width); + +// Convert RGB or BGR to Y +extern void (*WebPConvertRGB24ToY)(const uint8_t* rgb, uint8_t* y, int width); +extern void (*WebPConvertBGR24ToY)(const uint8_t* bgr, uint8_t* y, int width); + +// used for plain-C fallback. +extern void WebPConvertARGBToUV_C(const uint32_t* argb, uint8_t* u, uint8_t* v, + int src_width, int do_store); +extern void WebPConvertRGBA32ToUV_C(const uint16_t* rgb, + uint8_t* u, uint8_t* v, int width); + +// Must be called before using the above. +void WebPInitConvertARGBToYUV(void); + +//------------------------------------------------------------------------------ +// Rescaler + +struct WebPRescaler; + +// Import a row of data and save its contribution in the rescaler. +// 'channel' denotes the channel number to be imported. 'Expand' corresponds to +// the wrk->x_expand case. Otherwise, 'Shrink' is to be used. +typedef void (*WebPRescalerImportRowFunc)(struct WebPRescaler* const wrk, + const uint8_t* src); + +extern WebPRescalerImportRowFunc WebPRescalerImportRowExpand; +extern WebPRescalerImportRowFunc WebPRescalerImportRowShrink; + +// Export one row (starting at x_out position) from rescaler. +// 'Expand' corresponds to the wrk->y_expand case. +// Otherwise 'Shrink' is to be used +typedef void (*WebPRescalerExportRowFunc)(struct WebPRescaler* const wrk); +extern WebPRescalerExportRowFunc WebPRescalerExportRowExpand; +extern WebPRescalerExportRowFunc WebPRescalerExportRowShrink; + +// Plain-C implementation, as fall-back. +extern void WebPRescalerImportRowExpandC(struct WebPRescaler* const wrk, + const uint8_t* src); +extern void WebPRescalerImportRowShrinkC(struct WebPRescaler* const wrk, + const uint8_t* src); +extern void WebPRescalerExportRowExpandC(struct WebPRescaler* const wrk); +extern void WebPRescalerExportRowShrinkC(struct WebPRescaler* const wrk); + +// Main entry calls: +extern void WebPRescalerImportRow(struct WebPRescaler* const wrk, + const uint8_t* src); +// Export one row (starting at x_out position) from rescaler. +extern void WebPRescalerExportRow(struct WebPRescaler* const wrk); + +// Must be called first before using the above. +void WebPRescalerDspInit(void); + +//------------------------------------------------------------------------------ +// Utilities for processing transparent channel. // Apply alpha pre-multiply on an rgba, bgra or argb plane of size w * h. // alpha_first should be 0 for argb, 1 for rgba or bgra (where alpha is last). @@ -196,14 +402,98 @@ extern void (*WebPApplyAlphaMultiply)( extern void (*WebPApplyAlphaMultiply4444)( uint8_t* rgba4444, int w, int h, int stride); +// Dispatch the values from alpha[] plane to the ARGB destination 'dst'. +// Returns true if alpha[] plane has non-trivial values different from 0xff. +extern int (*WebPDispatchAlpha)(const uint8_t* alpha, int alpha_stride, + int width, int height, + uint8_t* dst, int dst_stride); + +// Transfer packed 8b alpha[] values to green channel in dst[], zero'ing the +// A/R/B values. 'dst_stride' is the stride for dst[] in uint32_t units. +extern void (*WebPDispatchAlphaToGreen)(const uint8_t* alpha, int alpha_stride, + int width, int height, + uint32_t* dst, int dst_stride); + +// Extract the alpha values from 32b values in argb[] and pack them into alpha[] +// (this is the opposite of WebPDispatchAlpha). +// Returns true if there's only trivial 0xff alpha values. +extern int (*WebPExtractAlpha)(const uint8_t* argb, int argb_stride, + int width, int height, + uint8_t* alpha, int alpha_stride); + +// Pre-Multiply operation transforms x into x * A / 255 (where x=Y,R,G or B). +// Un-Multiply operation transforms x into x * 255 / A. + +// Pre-Multiply or Un-Multiply (if 'inverse' is true) argb values in a row. +extern void (*WebPMultARGBRow)(uint32_t* const ptr, int width, int inverse); + +// Same a WebPMultARGBRow(), but for several rows. +void WebPMultARGBRows(uint8_t* ptr, int stride, int width, int num_rows, + int inverse); + +// Same for a row of single values, with side alpha values. +extern void (*WebPMultRow)(uint8_t* const ptr, const uint8_t* const alpha, + int width, int inverse); + +// Same a WebPMultRow(), but for several 'num_rows' rows. +void WebPMultRows(uint8_t* ptr, int stride, + const uint8_t* alpha, int alpha_stride, + int width, int num_rows, int inverse); + +// Plain-C versions, used as fallback by some implementations. +void WebPMultRowC(uint8_t* const ptr, const uint8_t* const alpha, + int width, int inverse); +void WebPMultARGBRowC(uint32_t* const ptr, int width, int inverse); + // To be called first before using the above. -void WebPInitPremultiply(void); +void WebPInitAlphaProcessing(void); + +// ARGB packing function: a/r/g/b input is rgba or bgra order. +extern void (*VP8PackARGB)(const uint8_t* a, const uint8_t* r, + const uint8_t* g, const uint8_t* b, int len, + uint32_t* out); + +// RGB packing function. 'step' can be 3 or 4. r/g/b input is rgb or bgr order. +extern void (*VP8PackRGB)(const uint8_t* r, const uint8_t* g, const uint8_t* b, + int len, int step, uint32_t* out); -void WebPInitPremultiplySSE2(void); // should not be called directly. +// To be called first before using the above. +void VP8EncDspARGBInit(void); //------------------------------------------------------------------------------ +// Filter functions + +typedef enum { // Filter types. + WEBP_FILTER_NONE = 0, + WEBP_FILTER_HORIZONTAL, + WEBP_FILTER_VERTICAL, + WEBP_FILTER_GRADIENT, + WEBP_FILTER_LAST = WEBP_FILTER_GRADIENT + 1, // end marker + WEBP_FILTER_BEST, // meta-types + WEBP_FILTER_FAST +} WEBP_FILTER_TYPE; + +typedef void (*WebPFilterFunc)(const uint8_t* in, int width, int height, + int stride, uint8_t* out); +typedef void (*WebPUnfilterFunc)(int width, int height, int stride, + int row, int num_rows, uint8_t* data); + +// Filter the given data using the given predictor. +// 'in' corresponds to a 2-dimensional pixel array of size (stride * height) +// in raster order. +// 'stride' is number of bytes per scan line (with possible padding). +// 'out' should be pre-allocated. +extern WebPFilterFunc WebPFilters[WEBP_FILTER_LAST]; + +// In-place reconstruct the original data from the given filtered data. +// The reconstruction will be done for 'num_rows' rows starting from 'row' +// (assuming rows upto 'row - 1' are already reconstructed). +extern WebPUnfilterFunc WebPUnfilters[WEBP_FILTER_LAST]; + +// To be called first before using the above. +void VP8FiltersInit(void); -#if defined(__cplusplus) || defined(c_plusplus) +#ifdef __cplusplus } // extern "C" #endif |