godot/thirdparty/libwebp/src/dsp/lossless_enc_sse2.c

// 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.
// -----------------------------------------------------------------------------
//
// SSE2 variant of methods for lossless encoder
//
// Author: Skal ([email protected])

#include "src/dsp/dsp.h"

#if defined(WEBP_USE_SSE2)
#include <assert.h>
#include <emmintrin.h>
#include "src/dsp/lossless.h"
#include "src/dsp/common_sse2.h"
#include "src/dsp/lossless_common.h"

// For sign-extended multiplying constants, pre-shifted by 5:
#define CST_5b(X) …

//------------------------------------------------------------------------------
// Subtract-Green Transform

static void SubtractGreenFromBlueAndRed_SSE2(uint32_t* argb_data,
                                             int num_pixels) { … }

//------------------------------------------------------------------------------
// Color Transform

#define MK_CST_16 …

static void TransformColor_SSE2(const VP8LMultipliers* const m,
                                uint32_t* argb_data, int num_pixels) { … }

//------------------------------------------------------------------------------
#define SPAN …
static void CollectColorBlueTransforms_SSE2(const uint32_t* argb, int stride,
                                            int tile_width, int tile_height,
                                            int green_to_blue, int red_to_blue,
                                            int histo[]) { … }

static void CollectColorRedTransforms_SSE2(const uint32_t* argb, int stride,
                                           int tile_width, int tile_height,
                                           int green_to_red, int histo[]) { … }
#undef SPAN
#undef MK_CST_16

//------------------------------------------------------------------------------

// Note we are adding uint32_t's as *signed* int32's (using _mm_add_epi32). But
// that's ok since the histogram values are less than 1<<28 (max picture size).
#define LINE_SIZE …
static void AddVector_SSE2(const uint32_t* a, const uint32_t* b, uint32_t* out,
                           int size) { … }

static void AddVectorEq_SSE2(const uint32_t* a, uint32_t* out, int size) { … }
#undef LINE_SIZE

//------------------------------------------------------------------------------
// Entropy

// TODO(https://crbug.com/webp/499): this function produces different results
// from the C code due to use of double/float resulting in output differences
// when compared to -noasm.
#if !(defined(WEBP_HAVE_SLOW_CLZ_CTZ) || defined(__i386__) || defined(_M_IX86))

static float CombinedShannonEntropy_SSE2(const int X[256], const int Y[256]) { … }

#else

#define DONT_USE_COMBINED_SHANNON_ENTROPY_SSE2_FUNC …

#endif

//------------------------------------------------------------------------------

static int VectorMismatch_SSE2(const uint32_t* const array1,
                               const uint32_t* const array2, int length) { … }

// Bundles multiple (1, 2, 4 or 8) pixels into a single pixel.
static void BundleColorMap_SSE2(const uint8_t* const row, int width, int xbits,
                                uint32_t* dst) { … }

//------------------------------------------------------------------------------
// Batch version of Predictor Transform subtraction

static WEBP_INLINE void Average2_m128i(const __m128i* const a0,
                                       const __m128i* const a1,
                                       __m128i* const avg) { … }

// Predictor0: ARGB_BLACK.
static void PredictorSub0_SSE2(const uint32_t* in, const uint32_t* upper,
                               int num_pixels, uint32_t* out) { … }

#define GENERATE_PREDICTOR_1 …

GENERATE_PREDICTOR_1       // Predictor1: L
GENERATE_PREDICTOR_1        // Predictor2: T
GENERATE_PREDICTOR_1    // Predictor3: TR
GENERATE_PREDICTOR_1    // Predictor4: TL
#undef GENERATE_PREDICTOR_1

// Predictor5: avg2(avg2(L, TR), T)
static void PredictorSub5_SSE2(const uint32_t* in, const uint32_t* upper,
                               int num_pixels, uint32_t* out) { … }

#define GENERATE_PREDICTOR_2 …

GENERATE_PREDICTOR_2   // Predictor6: avg(L, TL)
GENERATE_PREDICTOR_2       // Predictor7: avg(L, T)
GENERATE_PREDICTOR_2    // Predictor8: avg(TL, T)
GENERATE_PREDICTOR_2    // Predictor9: average(T, TR)
#undef GENERATE_PREDICTOR_2

// Predictor10: avg(avg(L,TL), avg(T, TR)).
static void PredictorSub10_SSE2(const uint32_t* in, const uint32_t* upper,
                                int num_pixels, uint32_t* out) { … }

// Predictor11: select.
static void GetSumAbsDiff32_SSE2(const __m128i* const A, const __m128i* const B,
                                 __m128i* const out) { … }

static void PredictorSub11_SSE2(const uint32_t* in, const uint32_t* upper,
                                int num_pixels, uint32_t* out) { … }

// Predictor12: ClampedSubSubtractFull.
static void PredictorSub12_SSE2(const uint32_t* in, const uint32_t* upper,
                                int num_pixels, uint32_t* out) { … }

// Predictors13: ClampedAddSubtractHalf
static void PredictorSub13_SSE2(const uint32_t* in, const uint32_t* upper,
                                int num_pixels, uint32_t* out) { … }

//------------------------------------------------------------------------------
// Entry point

extern void VP8LEncDspInitSSE2(void);

WEBP_TSAN_IGNORE_FUNCTION void VP8LEncDspInitSSE2(void) { … }

#else  // !WEBP_USE_SSE2

WEBP_DSP_INIT_STUB(VP8LEncDspInitSSE2)

#endif  // WEBP_USE_SSE2