coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2025-12-06 16:56:55 +08:00
Code Issues Packages Projects Releases Wiki Activity

InterpolateRow_AVX2 use AVX2 instead of ERMS for 100%

Browse Source 
Bug: b/210066781
Change-Id: I709e403f03bd6b9f8fe693b165b242b784076fe0
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/3329072
Reviewed-by: richard winterton <rrwinterton@gmail.com>
Reviewed-by: Frank Barchard <fbarchard@chromium.org>
Commit-Queue: Frank Barchard <fbarchard@chromium.org>
This commit is contained in:
Frank Barchard 2021-12-14 17:50:17 -08:00 committed by libyuv LUCI CQ
parent fdc71956bd
commit 78625492cb
3 changed files with 14 additions and 13 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: 1806
Version: 1807
License: BSD
License File: LICENSE

2
include/libyuv/version.h
View File

@ -11,6 +11,6 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_
#define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 1806
#define LIBYUV_VERSION 1807
#endif // INCLUDE_LIBYUV_VERSION_H_

23
source/row_gcc.cc
View File

@ -8124,7 +8124,7 @@ void ARGBAffineRow_SSE2(const uint8_t* src_argb,
void InterpolateRow_SSSE3(uint8_t* dst_ptr,
const uint8_t* src_ptr,
ptrdiff_t src_stride,
int dst_width,
int width,
int source_y_fraction) {
asm volatile(
"sub %1,%0 \n"
@ -8193,7 +8193,7 @@ void InterpolateRow_SSSE3(uint8_t* dst_ptr,
"99: \n"
: "+r"(dst_ptr), // %0
"+r"(src_ptr), // %1
"+rm"(dst_width), // %2
"+rm"(width), // %2
"+r"(source_y_fraction) // %3
: "r"((intptr_t)(src_stride)) // %4
: "memory", "cc", "eax", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5");
@ -8205,13 +8205,12 @@ void InterpolateRow_SSSE3(uint8_t* dst_ptr,
void InterpolateRow_AVX2(uint8_t* dst_ptr,
const uint8_t* src_ptr,
ptrdiff_t src_stride,
int dst_width,
int width,
int source_y_fraction) {
size_t width_tmp = (size_t)(dst_width);
asm volatile(
"sub %1,%0 \n"
"cmp $0x0,%3 \n"
"je 100f \n"
"sub %1,%0 \n"
"cmp $0x80,%3 \n"
"je 50f \n"
@ -8262,15 +8261,17 @@ void InterpolateRow_AVX2(uint8_t* dst_ptr,
// Blend 100 / 0 - Copy row unchanged.
LABELALIGN
"100: \n"
"rep movsb \n"
"jmp 999f \n"
"vmovdqu (%1),%%ymm0 \n"
"vmovdqu %%ymm0,0x00(%1,%0,1) \n"
"lea 0x20(%1),%1 \n"
"sub $0x20,%2 \n"
"jg 100b \n"
"99: \n"
"vzeroupper \n"
"999: \n"
: "+D"(dst_ptr), // %0
"+S"(src_ptr), // %1
"+c"(width_tmp), // %2
: "+r"(dst_ptr), // %0
"+r"(src_ptr), // %1
"+r"(width), // %2
"+r"(source_y_fraction) // %3
: "r"((intptr_t)(src_stride)) // %4
: "memory", "cc", "eax", "xmm0", "xmm1", "xmm2", "xmm4", "xmm5");