// 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. // ----------------------------------------------------------------------------- // // SSE2 version of speed-critical encoding functions. // // Author: Christian Duvivier ([email protected]) #include "src/dsp/dsp.h" #if defined(WEBP_USE_SSE2) #include <assert.h> #include <stdlib.h> // for abs() #include <emmintrin.h> #include "src/dsp/common_sse2.h" #include "src/enc/cost_enc.h" #include "src/enc/vp8i_enc.h" //------------------------------------------------------------------------------ // Transforms (Paragraph 14.4) // Does one inverse transform. static void ITransform_One_SSE2(const uint8_t* ref, const int16_t* in, uint8_t* dst) { … } // Does two inverse transforms. static void ITransform_Two_SSE2(const uint8_t* ref, const int16_t* in, uint8_t* dst) { … } // Does one or two inverse transforms. static void ITransform_SSE2(const uint8_t* ref, const int16_t* in, uint8_t* dst, int do_two) { … } static void FTransformPass1_SSE2(const __m128i* const in01, const __m128i* const in23, __m128i* const out01, __m128i* const out32) { … } static void FTransformPass2_SSE2(const __m128i* const v01, const __m128i* const v32, int16_t* out) { … } static void FTransform_SSE2(const uint8_t* src, const uint8_t* ref, int16_t* out) { … } static void FTransform2_SSE2(const uint8_t* src, const uint8_t* ref, int16_t* out) { … } static void FTransformWHTRow_SSE2(const int16_t* const in, __m128i* const out) { … } static void FTransformWHT_SSE2(const int16_t* in, int16_t* out) { … } //------------------------------------------------------------------------------ // Compute susceptibility based on DCT-coeff histograms: // the higher, the "easier" the macroblock is to compress. static void CollectHistogram_SSE2(const uint8_t* ref, const uint8_t* pred, int start_block, int end_block, VP8Histogram* const histo) { … } //------------------------------------------------------------------------------ // Intra predictions // helper for chroma-DC predictions static WEBP_INLINE void Put8x8uv_SSE2(uint8_t v, uint8_t* dst) { … } static WEBP_INLINE void Put16_SSE2(uint8_t v, uint8_t* dst) { … } static WEBP_INLINE void Fill_SSE2(uint8_t* dst, int value, int size) { … } static WEBP_INLINE void VE8uv_SSE2(uint8_t* dst, const uint8_t* top) { … } static WEBP_INLINE void VE16_SSE2(uint8_t* dst, const uint8_t* top) { … } static WEBP_INLINE void VerticalPred_SSE2(uint8_t* dst, const uint8_t* top, int size) { … } static WEBP_INLINE void HE8uv_SSE2(uint8_t* dst, const uint8_t* left) { … } static WEBP_INLINE void HE16_SSE2(uint8_t* dst, const uint8_t* left) { … } static WEBP_INLINE void HorizontalPred_SSE2(uint8_t* dst, const uint8_t* left, int size) { … } static WEBP_INLINE void TM_SSE2(uint8_t* dst, const uint8_t* left, const uint8_t* top, int size) { … } static WEBP_INLINE void TrueMotion_SSE2(uint8_t* dst, const uint8_t* left, const uint8_t* top, int size) { … } static WEBP_INLINE void DC8uv_SSE2(uint8_t* dst, const uint8_t* left, const uint8_t* top) { … } static WEBP_INLINE void DC8uvNoLeft_SSE2(uint8_t* dst, const uint8_t* top) { … } static WEBP_INLINE void DC8uvNoTop_SSE2(uint8_t* dst, const uint8_t* left) { … } static WEBP_INLINE void DC8uvNoTopLeft_SSE2(uint8_t* dst) { … } static WEBP_INLINE void DC8uvMode_SSE2(uint8_t* dst, const uint8_t* left, const uint8_t* top) { … } static WEBP_INLINE void DC16_SSE2(uint8_t* dst, const uint8_t* left, const uint8_t* top) { … } static WEBP_INLINE void DC16NoLeft_SSE2(uint8_t* dst, const uint8_t* top) { … } static WEBP_INLINE void DC16NoTop_SSE2(uint8_t* dst, const uint8_t* left) { … } static WEBP_INLINE void DC16NoTopLeft_SSE2(uint8_t* dst) { … } static WEBP_INLINE void DC16Mode_SSE2(uint8_t* dst, const uint8_t* left, const uint8_t* top) { … } //------------------------------------------------------------------------------ // 4x4 predictions #define DST … #define AVG3 … #define AVG2 … // We use the following 8b-arithmetic tricks: // (a + 2 * b + c + 2) >> 2 = (AC + b + 1) >> 1 // where: AC = (a + c) >> 1 = [(a + c + 1) >> 1] - [(a^c) & 1] // and: // (a + 2 * b + c + 2) >> 2 = (AB + BC + 1) >> 1 - (ab|bc)&lsb // where: AC = (a + b + 1) >> 1, BC = (b + c + 1) >> 1 // and ab = a ^ b, bc = b ^ c, lsb = (AC^BC)&1 static WEBP_INLINE void VE4_SSE2(uint8_t* dst, const uint8_t* top) { … } static WEBP_INLINE void HE4_SSE2(uint8_t* dst, const uint8_t* top) { … } static WEBP_INLINE void DC4_SSE2(uint8_t* dst, const uint8_t* top) { … } static WEBP_INLINE void LD4_SSE2(uint8_t* dst, const uint8_t* top) { … } static WEBP_INLINE void VR4_SSE2(uint8_t* dst, const uint8_t* top) { … } static WEBP_INLINE void VL4_SSE2(uint8_t* dst, const uint8_t* top) { … } static WEBP_INLINE void RD4_SSE2(uint8_t* dst, const uint8_t* top) { … } static WEBP_INLINE void HU4_SSE2(uint8_t* dst, const uint8_t* top) { … } static WEBP_INLINE void HD4_SSE2(uint8_t* dst, const uint8_t* top) { … } static WEBP_INLINE void TM4_SSE2(uint8_t* dst, const uint8_t* top) { … } #undef DST #undef AVG3 #undef AVG2 //------------------------------------------------------------------------------ // luma 4x4 prediction // Left samples are top[-5 .. -2], top_left is top[-1], top are // located at top[0..3], and top right is top[4..7] static void Intra4Preds_SSE2(uint8_t* dst, const uint8_t* top) { … } //------------------------------------------------------------------------------ // Chroma 8x8 prediction (paragraph 12.2) static void IntraChromaPreds_SSE2(uint8_t* dst, const uint8_t* left, const uint8_t* top) { … } //------------------------------------------------------------------------------ // luma 16x16 prediction (paragraph 12.3) static void Intra16Preds_SSE2(uint8_t* dst, const uint8_t* left, const uint8_t* top) { … } //------------------------------------------------------------------------------ // Metric static WEBP_INLINE void SubtractAndAccumulate_SSE2(const __m128i a, const __m128i b, __m128i* const sum) { … } static WEBP_INLINE int SSE_16xN_SSE2(const uint8_t* a, const uint8_t* b, int num_pairs) { … } static int SSE16x16_SSE2(const uint8_t* a, const uint8_t* b) { … } static int SSE16x8_SSE2(const uint8_t* a, const uint8_t* b) { … } #define LOAD_8x16b … static int SSE8x8_SSE2(const uint8_t* a, const uint8_t* b) { … } #undef LOAD_8x16b static int SSE4x4_SSE2(const uint8_t* a, const uint8_t* b) { … } //------------------------------------------------------------------------------ static void Mean16x4_SSE2(const uint8_t* ref, uint32_t dc[4]) { … } //------------------------------------------------------------------------------ // Texture distortion // // We try to match the spectral content (weighted) between source and // reconstructed samples. // Hadamard transform // Returns the weighted sum of the absolute value of transformed coefficients. // w[] contains a row-major 4 by 4 symmetric matrix. static int TTransform_SSE2(const uint8_t* inA, const uint8_t* inB, const uint16_t* const w) { … } static int Disto4x4_SSE2(const uint8_t* const a, const uint8_t* const b, const uint16_t* const w) { … } static int Disto16x16_SSE2(const uint8_t* const a, const uint8_t* const b, const uint16_t* const w) { … } //------------------------------------------------------------------------------ // Quantization // static WEBP_INLINE int DoQuantizeBlock_SSE2(int16_t in[16], int16_t out[16], const uint16_t* const sharpen, const VP8Matrix* const mtx) { … } static int QuantizeBlock_SSE2(int16_t in[16], int16_t out[16], const VP8Matrix* const mtx) { … } static int QuantizeBlockWHT_SSE2(int16_t in[16], int16_t out[16], const VP8Matrix* const mtx) { … } static int Quantize2Blocks_SSE2(int16_t in[32], int16_t out[32], const VP8Matrix* const mtx) { … } //------------------------------------------------------------------------------ // Entry point extern void VP8EncDspInitSSE2(void); WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspInitSSE2(void) { … } #else // !WEBP_USE_SSE2 WEBP_DSP_INIT_STUB(VP8EncDspInitSSE2) #endif // WEBP_USE_SSE2
Codebase Browser
godot