generic-library/webp
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

lossless*sse2: improve non-const 16-bit vector creation

Browse Source 
use _mm_set1_epi32 instead of _mm_set_epi16 with non-const values;
reduces shifts and ors.

Change-Id: Ie2cb2ab815f642855d03c6f3001223bcac4bd35c
This commit is contained in:
James Zern 2018-02-17 17:59:20 -08:00
parent 551948e45f
commit 8043504f95
2 changed files with 15 additions and 25 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
src/dsp/lossless_enc_sse2.c
View File

@ -46,16 +46,14 @@ static void SubtractGreenFromBlueAndRed_SSE2(uint32_t* argb_data,
//------------------------------------------------------------------------------
// Color Transform
#define MK_CST_16(HI, LO) \
_mm_set1_epi32((int)(((uint32_t)(HI) << 16) | ((LO) & 0xffff)))
static void TransformColor_SSE2(const VP8LMultipliers* const m,
uint32_t* argb_data, int num_pixels) {
const __m128i mults_rb = _mm_set_epi16(
CST_5b(m->green_to_red_), CST_5b(m->green_to_blue_),
CST_5b(m->green_to_red_), CST_5b(m->green_to_blue_),
CST_5b(m->green_to_red_), CST_5b(m->green_to_blue_),
CST_5b(m->green_to_red_), CST_5b(m->green_to_blue_));
const __m128i mults_b2 = _mm_set_epi16(
CST_5b(m->red_to_blue_), 0, CST_5b(m->red_to_blue_), 0,
CST_5b(m->red_to_blue_), 0, CST_5b(m->red_to_blue_), 0);
const __m128i mults_rb = MK_CST_16(CST_5b(m->green_to_red_),
CST_5b(m->green_to_blue_));
const __m128i mults_b2 = MK_CST_16(CST_5b(m->red_to_blue_), 0);
const __m128i mask_ag = _mm_set1_epi32(0xff00ff00); // alpha-green masks
const __m128i mask_rb = _mm_set1_epi32(0x00ff00ff); // red-blue masks
int i;
@ -85,12 +83,8 @@ 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[]) {
const __m128i mults_r = _mm_set_epi16(
CST_5b(red_to_blue), 0, CST_5b(red_to_blue), 0,
CST_5b(red_to_blue), 0, CST_5b(red_to_blue), 0);
const __m128i mults_g = _mm_set_epi16(
0, CST_5b(green_to_blue), 0, CST_5b(green_to_blue),
0, CST_5b(green_to_blue), 0, CST_5b(green_to_blue));
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
const __m128i mask_b = _mm_set1_epi32(0x0000ff); // blue mask
int y;
@ -135,9 +129,7 @@ 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[]) {
const __m128i mults_g = _mm_set_epi16(
0, CST_5b(green_to_red), 0, CST_5b(green_to_red),
0, CST_5b(green_to_red), 0, CST_5b(green_to_red));
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);
@ -174,6 +166,7 @@ static void CollectColorRedTransforms_SSE2(const uint32_t* argb, int stride,
}
}
#undef SPAN
#undef MK_CST_16
//------------------------------------------------------------------------------

13
src/dsp/lossless_sse2.c
View File

@ -453,14 +453,11 @@ static void TransformColorInverse_SSE2(const VP8LMultipliers* const m,
int num_pixels, uint32_t* dst) {
// sign-extended multiplying constants, pre-shifted by 5.
#define CST(X) (((int16_t)(m->X << 8)) >> 5) // sign-extend
const __m128i mults_rb = _mm_set_epi16(
CST(green_to_red_), CST(green_to_blue_),
CST(green_to_red_), CST(green_to_blue_),
CST(green_to_red_), CST(green_to_blue_),
CST(green_to_red_), CST(green_to_blue_));
const __m128i mults_b2 = _mm_set_epi16(
CST(red_to_blue_), 0, CST(red_to_blue_), 0,
CST(red_to_blue_), 0, CST(red_to_blue_), 0);
#define MK_CST_16(HI, LO) \
_mm_set1_epi32((int)(((uint32_t)(HI) << 16) | ((LO) & 0xffff)))
const __m128i mults_rb = MK_CST_16(CST(green_to_red_), CST(green_to_blue_));
const __m128i mults_b2 = MK_CST_16(CST(red_to_blue_), 0);
#undef MK_CST_16
#undef CST
const __m128i mask_ag = _mm_set1_epi32(0xff00ff00); // alpha-green masks
int i;