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

port scale box filter sse2 to gcc

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

Review URL: https://codereview.chromium.org/1393653002 .
This commit is contained in:
Frank Barchard 2015-10-06 16:54:26 -07:00
parent f4c1ac10f0
commit d70293993f
4 changed files with 16 additions and 36 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: 1499
Version: 1500
License: BSD
License File: LICENSE

6
include/libyuv/scale_row.h
View File

@ -53,6 +53,7 @@ extern "C" {
#define HAS_SCALEROWDOWN34_SSSE3
#define HAS_SCALEROWDOWN38_SSSE3
#define HAS_SCALEROWDOWN4_SSE2
#define HAS_SCALEADDROW_SSE2
#endif
// The following are available for Visual C and clangcl 32 bit:
@ -63,11 +64,6 @@ extern "C" {
#define HAS_SCALEROWDOWN4_AVX2
#endif
// The following are available on Visual C:
#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86)
#define HAS_SCALEADDROW_SSE2
#endif
// The following are available on Neon platforms:
#if !defined(LIBYUV_DISABLE_NEON) && !defined(__native_client__) && \
(defined(__ARM_NEON__) || defined(LIBYUV_NEON) || defined(__aarch64__))

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 1499
#define LIBYUV_VERSION 1500
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT

42
source/scale_gcc.cc
View File

@ -575,47 +575,31 @@ void ScaleRowDown38_3_Box_SSSE3(const uint8* src_ptr,
}
// Reads 16xN bytes and produces 16 shorts at a time.
void ScaleAddRows_SSE2(const uint8* src_ptr, ptrdiff_t src_stride,
uint16* dst_ptr, int src_width, int src_height) {
int tmp_height = 0;
intptr_t tmp_src = 0;
void ScaleAddRow_SSE2(const uint8* src_ptr, uint16* dst_ptr, int src_width) {
asm volatile (
"mov %0,%3 \n" // row pointer
"mov %5,%2 \n" // height
"pxor %%xmm0,%%xmm0 \n" // clear accumulators
"pxor %%xmm1,%%xmm1 \n"
"pxor %%xmm4,%%xmm4 \n"
"pxor %%xmm5,%%xmm5 \n"
LABELALIGN
"1: \n"
"movdqu " MEMACCESS(3) ",%%xmm2 \n"
"add %6,%3 \n"
"movdqa %%xmm2,%%xmm3 \n"
"punpcklbw %%xmm4,%%xmm2 \n"
"punpckhbw %%xmm4,%%xmm3 \n"
"movdqu " MEMACCESS(0) ",%%xmm3 \n"
"lea " MEMLEA(0x10,0) ",%0 \n" // src_ptr += 16
"movdqu " MEMACCESS(1) ",%%xmm0 \n"
"movdqu " MEMACCESS2(0x10,1) ",%%xmm1 \n"
"movdqa %%xmm3,%%xmm2 \n"
"punpcklbw %%xmm5,%%xmm2 \n"
"punpckhbw %%xmm5,%%xmm3 \n"
"paddusw %%xmm2,%%xmm0 \n"
"paddusw %%xmm3,%%xmm1 \n"
"sub $0x1,%2 \n"
"jg 1b \n"
"movdqu %%xmm0," MEMACCESS(1) " \n"
"movdqu %%xmm1," MEMACCESS2(0x10,1) " \n"
"lea " MEMLEA(0x20,1) ",%1 \n"
"lea " MEMLEA(0x10,0) ",%0 \n" // src_ptr += 16
"mov %0,%3 \n" // row pointer
"mov %5,%2 \n" // height
"pxor %%xmm0,%%xmm0 \n" // clear accumulators
"pxor %%xmm1,%%xmm1 \n"
"sub $0x10,%4 \n"
"sub $0x10,%2 \n"
"jg 1b \n"
: "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1
"+r"(tmp_height), // %2
"+r"(tmp_src), // %3
"+r"(src_width), // %4
"+rm"(src_height) // %5
: "rm"((intptr_t)(src_stride)) // %6
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4"
"+r"(src_width) // %2
:
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm5"
);
}