dany/libwebp
1
0
Fork 0
mirror of https://github.com/webmproject/libwebp.git synced 2025-07-16 05:49:51 +02:00
Code Issues Actions Packages Projects Releases Wiki Activity

Refactor histograms in predictors.

Browse Source 
Replace the 2d histograms with uint32_t 1d versions (to avoid
pointer casting and to use the optimized VP8LAddVectorEq).

Change-Id: I90b0fe98390b49e3fd03e3484289571cf7ae6eca
This commit is contained in:
Vincent Rabaud
2024-05-03 22:09:38 +02:00
parent a7aa7525b8
commit a90160e11a
6 changed files with 90 additions and 97 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
src/dsp/lossless_enc_sse2.c
View File

@ -82,7 +82,7 @@ static void TransformColor_SSE2(const VP8LMultipliers* const m,
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[]) {
uint32_t histo[]) {
const __m128i mults_r = MK_CST_16(CST_5b(red_to_blue), 0);
const __m128i mults_g = MK_CST_16(0, CST_5b(green_to_blue));
const __m128i mask_g = _mm_set1_epi32(0x00ff00); // green mask
@ -128,7 +128,7 @@ static void CollectColorBlueTransforms_SSE2(const uint32_t* argb, int stride,
static void CollectColorRedTransforms_SSE2(const uint32_t* argb, int stride,
int tile_width, int tile_height,
int green_to_red, int histo[]) {
int green_to_red, uint32_t histo[]) {
const __m128i mults_g = MK_CST_16(0, CST_5b(green_to_red));
const __m128i mask_g = _mm_set1_epi32(0x00ff00); // green mask
const __m128i mask = _mm_set1_epi32(0xff);
@ -237,10 +237,11 @@ static void AddVectorEq_SSE2(const uint32_t* a, uint32_t* out, int size) {
// 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]) {
static float CombinedShannonEntropy_SSE2(const uint32_t X[256],
const uint32_t Y[256]) {
int i;
float retval = 0.f;
int sumX = 0, sumXY = 0;
uint32_t sumX = 0, sumXY = 0;
const __m128i zero = _mm_setzero_si128();
for (i = 0; i < 256; i += 16) {
@ -260,7 +261,7 @@ static float CombinedShannonEntropy_SSE2(const int X[256], const int Y[256]) {
int32_t my = _mm_movemask_epi8(_mm_cmpgt_epi8(y4, zero)) | mx;
while (my) {
const int32_t j = BitsCtz(my);
int xy;
uint32_t xy;
if ((mx >> j) & 1) {
const int x = X[i + j];
sumXY += x;