coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2026-01-01 03:12:16 +08:00
Code Issues Packages Projects Releases Wiki Activity

ShortToHalfFloat_AVX2 function

Browse Source 
BUG=libyuv:560
TEST=local compile for windows
R=wangcheng@google.com

Review URL: https://codereview.chromium.org/2364293002 .
This commit is contained in:
Frank Barchard 2016-09-27 14:18:32 -07:00
parent bcd823805c
commit 6732bcbde9
2 changed files with 35 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
include/libyuv/row.h
View File

@ -252,6 +252,7 @@ extern "C" {
#define HAS_ARGBTORGB565ROW_AVX2
#define HAS_J400TOARGBROW_AVX2
#define HAS_RGB565TOARGBROW_AVX2
#define HAS_SHORTTOF16ROW_AVX2
#endif
// The following are also available on x64 Visual C.
@ -1932,6 +1933,10 @@ void ARGBPolynomialRow_AVX2(const uint8* src_argb,
uint8* dst_argb, const float* poly,
int width);
// Scale and convert to half float.
void ShortToF16Row_C(const uint16* src, int16* dst, float scale, int width);
void ShortToF16Row_AVX2(const uint16* src, int16* dst, float scale, int width);
void ARGBLumaColorTableRow_C(const uint8* src_argb, uint8* dst_argb, int width,
const uint8* luma, uint32 lumacoeff);
void ARGBLumaColorTableRow_SSSE3(const uint8* src_argb, uint8* dst_argb,

30
source/row_win.cc
View File

@ -6095,6 +6095,36 @@ void ARGBPolynomialRow_AVX2(const uint8* src_argb,
}
#endif // HAS_ARGBPOLYNOMIALROW_AVX2
// Samples assumed to be unsigned in low 9, 10 or 12 bits. Scale factor
// adjust the sample range to 0 to 1 using a float multiply.
// e.g. 9 bit scale is 1.0f / 512.0f
// e.g. 10 bit scale is 1.0f / 1024.0f
#ifdef HAS_SHORTTOHALFFLOAT_AVX2
__declspec(naked)
void ShortToF16Row_AVX2(const uint16* src, int16* dst, float scale, int width) {
__asm {
mov eax, [esp + 4] /* src */
mov edx, [esp + 8] /* dst */
vbroadcastss ymm4, [esp + 12] /* scale */
mov ecx, [esp + 16] /* width */
// 8 pixel loop.
convertloop:
vpmovzxwd ymm0, xmmword ptr [eax] // 8 shorts -> 8 ints
lea eax, [eax + 16]
vcvtdq2ps ymm0, ymm0 // convert 8 ints to floats
vmulps ymm0, ymm0, ymm4 // scale to normalized range 0 to 1
vcvtps2ph xmm0, ymm0, 0 // float conver to 8 half floats round even
vmovdqu [edx], xmm0
lea edx, [edx + 16]
sub ecx, 8
jg convertloop
vzeroupper
ret
}
}
#endif // HAS_SHORTTOHALFFLOAT_AVX2
#ifdef HAS_ARGBCOLORTABLEROW_X86
// Tranform ARGB pixels with color table.
__declspec(naked)