coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2025-12-07 01:06:46 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix for InterpolateRow_AVX2

Browse Source 
port scaledownby4_avx2 to gcc

TBR=harryjin@google.com
BUG=libyuv:492

Review URL: https://codereview.chromium.org/1546763002 .
This commit is contained in:
Frank Barchard 2015-12-22 12:29:54 -08:00
parent 71deb7ba3a
commit 36615d62a0
4 changed files with 89 additions and 7 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 Name: libyuv
URL: http://code.google.com/p/libyuv/ URL: http://code.google.com/p/libyuv/
Version: 1560 Version: 1561
License: BSD License: BSD
License File: LICENSE License File: LICENSE

5
include/libyuv/scale_row.h
View File

@ -70,11 +70,6 @@ extern "C" {
defined(CLANG_HAS_AVX2) || defined(GCC_HAS_AVX2)) defined(CLANG_HAS_AVX2) || defined(GCC_HAS_AVX2))
#define HAS_SCALEADDROW_AVX2 #define HAS_SCALEADDROW_AVX2
#define HAS_SCALEROWDOWN2_AVX2 #define HAS_SCALEROWDOWN2_AVX2
#endif
// The following are available for Visual C and clangcl 32 bit:
#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && \
(defined(VISUALC_HAS_AVX2) || defined(CLANG_HAS_AVX2))
#define HAS_SCALEROWDOWN4_AVX2 #define HAS_SCALEROWDOWN4_AVX2
#endif #endif

2
include/libyuv/version.h
View File

@ -11,6 +11,6 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_ // NOLINT #ifndef INCLUDE_LIBYUV_VERSION_H_ // NOLINT
#define INCLUDE_LIBYUV_VERSION_H_ #define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 1560 #define LIBYUV_VERSION 1561
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT #endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT

87
source/scale_gcc.cc
View File

@ -368,6 +368,93 @@ void ScaleRowDown4Box_SSSE3(const uint8* src_ptr, ptrdiff_t src_stride,
); );
} }
#ifdef HAS_SCALEROWDOWN4_AVX2
void ScaleRowDown4_AVX2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
asm volatile (
"vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n"
"vpsrld $0x18,%%ymm5,%%ymm5 \n"
"vpslld $0x10,%%ymm5,%%ymm5 \n"
LABELALIGN
"1: \n"
"vmovdqu " MEMACCESS(0) ",%%ymm0 \n"
"vmovdqu " MEMACCESS2(0x20,0) ",%%ymm1 \n"
"lea " MEMLEA(0x40,0) ",%0 \n"
"vpand %%ymm5,%%ymm0,%%ymm0 \n"
"vpand %%ymm5,%%ymm1,%%ymm1 \n"
"vpackuswb %%ymm1,%%ymm0,%%ymm0 \n"
"vpermq $0xd8,%%ymm0,%%ymm0 \n"
"vpsrlw $0x8,%%ymm0,%%ymm0 \n"
"vpackuswb %%ymm0,%%ymm0,%%ymm0 \n"
"vpermq $0xd8,%%ymm0,%%ymm0 \n"
"vmovdqu %%xmm0," MEMACCESS(1) " \n"
"lea " MEMLEA(0x10,1) ",%1 \n"
"sub $0x10,%2 \n"
"jg 1b \n"
"vzeroupper \n"
: "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1
"+r"(dst_width) // %2
:: "memory", "cc", "xmm0", "xmm1", "xmm5"
);
}
void ScaleRowDown4Box_AVX2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
asm volatile (
"vpcmpeqb %%ymm4,%%ymm4,%%ymm4 \n"
"vpsrlw $0xf,%%ymm4,%%ymm4 \n"
"vpsllw $0x3,%%ymm4,%%ymm5 \n"
"vpackuswb %%ymm4,%%ymm4,%%ymm4 \n"
LABELALIGN
"1: \n"
"vmovdqu " MEMACCESS(0) ",%%ymm0 \n"
"vmovdqu " MEMACCESS2(0x20,0) ",%%ymm1 \n"
MEMOPREG(vmovdqu,0x00,0,3,1,ymm2) // vmovdqu (%0,%3,1),%%ymm2
MEMOPREG(vmovdqu,0x20,0,3,1,ymm3) // vmovdqu 0x20(%0,%3,1),%%ymm3
"vpmaddubsw %%ymm4,%%ymm0,%%ymm0 \n"
"vpmaddubsw %%ymm4,%%ymm1,%%ymm1 \n"
"vpmaddubsw %%ymm4,%%ymm2,%%ymm2 \n"
"vpmaddubsw %%ymm4,%%ymm3,%%ymm3 \n"
"vpaddw %%ymm2,%%ymm0,%%ymm0 \n"
"vpaddw %%ymm3,%%ymm1,%%ymm1 \n"
MEMOPREG(vmovdqu,0x00,0,3,2,ymm2) // vmovdqu (%0,%3,2),%%ymm2
MEMOPREG(vmovdqu,0x20,0,3,2,ymm3) // vmovdqu 0x20(%0,%3,2),%%ymm3
"vpmaddubsw %%ymm4,%%ymm2,%%ymm2 \n"
"vpmaddubsw %%ymm4,%%ymm3,%%ymm3 \n"
"vpaddw %%ymm2,%%ymm0,%%ymm0 \n"
"vpaddw %%ymm3,%%ymm1,%%ymm1 \n"
MEMOPREG(vmovdqu,0x00,0,4,1,ymm2) // vmovdqu (%0,%4,1),%%ymm2
MEMOPREG(vmovdqu,0x20,0,4,1,ymm3) // vmovdqu 0x20(%0,%4,1),%%ymm3
"lea " MEMLEA(0x40,0) ",%0 \n"
"vpmaddubsw %%ymm4,%%ymm2,%%ymm2 \n"
"vpmaddubsw %%ymm4,%%ymm3,%%ymm3 \n"
"vpaddw %%ymm2,%%ymm0,%%ymm0 \n"
"vpaddw %%ymm3,%%ymm1,%%ymm1 \n"
"vphaddw %%ymm1,%%ymm0,%%ymm0 \n"
"vpermq $0xd8,%%ymm0,%%ymm0 \n"
"vpaddw %%ymm5,%%ymm0,%%ymm0 \n"
"vpsrlw $0x4,%%ymm0,%%ymm0 \n"
"vpackuswb %%ymm0,%%ymm0,%%ymm0 \n"
"vpermq $0xd8,%%ymm0,%%ymm0 \n"
"vmovdqu %%xmm0," MEMACCESS(1) " \n"
"lea " MEMLEA(0x10,1) ",%1 \n"
"sub $0x10,%2 \n"
"jg 1b \n"
"vzeroupper \n"
: "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1
"+r"(dst_width) // %2
: "r"((intptr_t)(src_stride)), // %3
"r"((intptr_t)(src_stride * 3)) // %4
: "memory", "cc", NACL_R14
"xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
);
}
#endif // HAS_SCALEROWDOWN4_AVX2
void ScaleRowDown34_SSSE3(const uint8* src_ptr, ptrdiff_t src_stride, void ScaleRowDown34_SSSE3(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) { uint8* dst_ptr, int dst_width) {
asm volatile ( asm volatile (