coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2025-12-07 09:16:48 +08:00
Code Issues Packages Projects Releases Wiki Activity

remove vpermq from ARGBToUV

Browse Source 
BUG=none
TEST=convert_test
Review URL: https://webrtc-codereview.appspot.com/1107005

git-svn-id: http://libyuv.googlecode.com/svn/trunk@575 16f28f9a-4ce2-e073-06de-1de4eb20be90
This commit is contained in:
fbarchard@google.com 2013-02-15 18:33:20 +00:00
parent d8b73cacbe
commit caf6e2470b
3 changed files with 38 additions and 35 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
README.chromium
View File

@ -1,6 +1,6 @@
Name: libyuv
URL: http://code.google.com/p/libyuv/
Version: 574
Version: 575
License: BSD
License File: LICENSE

2
include/libyuv/version.h
View File

@ -11,6 +11,6 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_ // NOLINT
#define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 574
#define LIBYUV_VERSION 575
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT

69
source/row_win.cc
View File

@ -49,11 +49,17 @@ static const lvec8 kARGBToV_AVX = {
-18, -94, 112, 0, -18, -94, 112, 0, -18, -94, 112, 0, -18, -94, 112, 0
};
// Unshuffle for vphaddw + vpackuswb vpermd.
// vpermd for vphaddw + vpackuswb vpermd.
static const lvec32 kShufARGBToY_AVX = {
0, 4, 1, 5, 2, 6, 3, 7
};
// vpshufb for vphaddw + vpackuswb packed to shorts.
static const lvec8 kShufARGBToUV_AVX = {
0, 1, 8, 9, 2, 3, 10, 11, 4, 5, 12, 13, 6, 7, 14, 15,
0, 1, 8, 9, 2, 3, 10, 11, 4, 5, 12, 13, 6, 7, 14, 15,
};
// Constants for BGRA.
static const vec8 kBGRAToY = {
0, 33, 65, 13, 0, 33, 65, 13, 0, 33, 65, 13, 0, 33, 65, 13
@ -785,11 +791,11 @@ void ARGBToYRow_AVX2(const uint8* src_argb, uint8* dst_y, int pix) {
vpmaddubsw ymm2, ymm2, ymm4
vpmaddubsw ymm3, ymm3, ymm4
lea eax, [eax + 128]
vphaddw ymm0, ymm0, ymm1
vphaddw ymm0, ymm0, ymm1 // mutates.
vphaddw ymm2, ymm2, ymm3
vpsrlw ymm0, ymm0, 7
vpsrlw ymm2, ymm2, 7
vpackuswb ymm0, ymm0, ymm2
vpackuswb ymm0, ymm0, ymm2 // mutates.
vpermd ymm0, ymm6, ymm0 // For vphaddw + vpackuswb mutation.
vpaddb ymm0, ymm0, ymm5
sub ecx, 32
@ -1125,40 +1131,37 @@ void ARGBToUVRow_AVX2(const uint8* src_argb0, int src_stride_argb,
align 16
convertloop:
/* step 1 - subsample 32x2 argb pixels to 16x1 */
vmovdqu ymm0, [eax]
vmovdqu ymm1, [eax + 32]
vmovdqu ymm2, [eax + 64]
vmovdqu ymm3, [eax + 96]
vpavgb ymm0, ymm0, [eax + esi]
vpavgb ymm1, ymm1, [eax + esi + 32]
vpavgb ymm2, ymm2, [eax + esi + 64]
vpavgb ymm3, ymm3, [eax + esi + 96]
lea eax, [eax + 128]
vshufps ymm4, ymm0, ymm1, 0x88
vshufps ymm0, ymm0, ymm1, 0xdd
vpavgb ymm0, ymm0, ymm4
vpermq ymm0, ymm0, 0xd8 // TODO(fbarchard): Remove.
vshufps ymm4, ymm2, ymm3, 0x88
vshufps ymm2, ymm2, ymm3, 0xdd
vpavgb ymm2, ymm2, ymm4
vpermq ymm2, ymm2, 0xd8 // TODO(fbarchard): Remove.
vmovdqu ymm0, [eax]
vmovdqu ymm1, [eax + 32]
vmovdqu ymm2, [eax + 64]
vmovdqu ymm3, [eax + 96]
vpavgb ymm0, ymm0, [eax + esi]
vpavgb ymm1, ymm1, [eax + esi + 32]
vpavgb ymm2, ymm2, [eax + esi + 64]
vpavgb ymm3, ymm3, [eax + esi + 96]
lea eax, [eax + 128]
vshufps ymm4, ymm0, ymm1, 0x88
vshufps ymm0, ymm0, ymm1, 0xdd
vpavgb ymm0, ymm0, ymm4 // mutated by vshufps
vshufps ymm4, ymm2, ymm3, 0x88
vshufps ymm2, ymm2, ymm3, 0xdd
vpavgb ymm2, ymm2, ymm4 // mutated by vshufps
// step 2 - convert to U and V
// from here down is very similar to Y code except
// instead of 32 different pixels, its 16 pixels of U and 16 of V
vpmaddubsw ymm1, ymm0, ymm7 // U
vpmaddubsw ymm3, ymm2, ymm7
vpmaddubsw ymm0, ymm0, ymm6 // V
vpmaddubsw ymm2, ymm2, ymm6
vphaddw ymm1, ymm1, ymm3
vpermq ymm1, ymm1, 0xd8 // TODO(fbarchard): Remove.
vphaddw ymm0, ymm0, ymm2
vpermq ymm0, ymm0, 0xd8 // TODO(fbarchard): Remove.
vpsraw ymm1, ymm1, 8
vpsraw ymm0, ymm0, 8
vpacksswb ymm0, ymm1, ymm0
vpermq ymm0, ymm0, 0xd8
vpaddb ymm0, ymm0, ymm5 // -> unsigned
vpmaddubsw ymm1, ymm0, ymm7 // U
vpmaddubsw ymm3, ymm2, ymm7
vpmaddubsw ymm0, ymm0, ymm6 // V
vpmaddubsw ymm2, ymm2, ymm6
vphaddw ymm1, ymm1, ymm3 // mutates
vphaddw ymm0, ymm0, ymm2
vpsraw ymm1, ymm1, 8
vpsraw ymm0, ymm0, 8
vpacksswb ymm0, ymm1, ymm0 // mutates
vpermq ymm0, ymm0, 0xd8 // For vpacksswb
vpshufb ymm0, ymm0, kShufARGBToUV_AVX // For vshufps + vphaddw
vpaddb ymm0, ymm0, ymm5 // -> unsigned
// step 3 - store 16 U and 16 V values
sub ecx, 32