Skip to content

L
libyuv
Commit 9a0e12f5 authored by Frank Barchard's avatar Frank Barchard
Browse files
Options

AVX2 1 step I422AlphaToARGB for gcc and win. 

C     I420AlphaToARGB_Opt (5169 ms)
SSSE3 I420AlphaToARGB_Opt (432 ms)
AVX2  I420AlphaToARGB_Opt (358 ms)

and with premultiplication as 2 step process:
I420AlphaToARGB_Premult (7029 ms)
I420AlphaToARGB_Premult (757 ms)
I420AlphaToARGB_Premult (508 ms)

R=harryjin@google.com
BUG=libyuv:496,libyuv:473

Review URL: https://codereview.chromium.org/1372653003 .
parent e365cdde
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 143 additions and 14 deletions
README.chromium
View file @ 9a0e12f5
Name: libyuv
URL: http://code.google.com/p/libyuv/
Version: 1493
Version: 1494
License: BSD
License File: LICENSE
......
include/libyuv/row.h
View file @ 9a0e12f5
......@@ -147,6 +147,8 @@ extern "C" {
#define HAS_YUY2TOYROW_SSE2
#define HAS_I444TOARGBROW_SSSE3
#define HAS_I444TOABGRROW_SSSE3
#define HAS_I422ALPHATOARGBROW_SSSE3
#define HAS_I422ALPHATOABGRROW_SSSE3
// Effects:
#define HAS_ARGBADDROW_SSE2
......@@ -187,14 +189,6 @@ extern "C" {
#define HAS_I422TOABGRROW_SSSE3
#endif
// The following are available on 32 bit x86 Visual C and clangcl.
// TODO(fbarchard): Port to gcc.
#if !defined(LIBYUV_DISABLE_X86) && defined (_M_IX86)
#define HAS_I422ALPHATOARGBROW_SSSE3
#define HAS_I422ALPHATOABGRROW_SSSE3
#endif
// The following are available for AVX2 Visual C and clangcl 32 bit:
// TODO(fbarchard): Port to gcc.
#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && \
......@@ -255,6 +249,8 @@ extern "C" {
#define HAS_YUY2TOUVROW_AVX2
#define HAS_YUY2TOYROW_AVX2
#define HAS_NV12TOARGBROW_AVX2
#define HAS_I422ALPHATOARGBROW_AVX2
#define HAS_I422ALPHATOABGRROW_AVX2
// Effects:
#define HAS_ARGBADDROW_AVX2
......@@ -1250,6 +1246,20 @@ void I422AlphaToABGRRow_SSSE3(const uint8* y_buf,
uint8* dst_argb,
struct YuvConstants* yuvconstants,
int width);
void I422AlphaToARGBRow_AVX2(const uint8* y_buf,
const uint8* u_buf,
const uint8* v_buf,
const uint8* a_buf,
uint8* dst_argb,
struct YuvConstants* yuvconstants,
int width);
void I422AlphaToABGRRow_AVX2(const uint8* y_buf,
const uint8* u_buf,
const uint8* v_buf,
const uint8* a_buf,
uint8* dst_argb,
struct YuvConstants* yuvconstants,
int width);
void I422ToARGBRow_SSSE3(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
......@@ -1456,6 +1466,20 @@ void I422AlphaToABGRRow_Any_SSSE3(const uint8* y_buf,
uint8* dst_abgr,
struct YuvConstants* yuvconstants,
int width);
void I422AlphaToARGBRow_Any_AVX2(const uint8* y_buf,
const uint8* u_buf,
const uint8* v_buf,
const uint8* a_buf,
uint8* dst_argb,
struct YuvConstants* yuvconstants,
int width);
void I422AlphaToABGRRow_Any_AVX2(const uint8* y_buf,
const uint8* u_buf,
const uint8* v_buf,
const uint8* a_buf,
uint8* dst_abgr,
struct YuvConstants* yuvconstants,
int width);
void I411ToARGBRow_Any_SSSE3(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
......
include/libyuv/version.h
View file @ 9a0e12f5
......@@ -11,6 +11,6 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_ // NOLINT
#define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 1493
#define LIBYUV_VERSION 1494
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT
source/row_any.cc
View file @ 9a0e12f5
......@@ -48,6 +48,10 @@ extern "C" {
ANY41C(I422AlphaToARGBRow_Any_SSSE3, I422AlphaToARGBRow_SSSE3, 1, 0, 4, 7)
ANY41C(I422AlphaToABGRRow_Any_SSSE3, I422AlphaToABGRRow_SSSE3, 1, 0, 4, 7)
#endif
#ifdef HAS_I422ALPHATOARGBROW_AVX2
ANY41C(I422AlphaToARGBRow_Any_AVX2, I422AlphaToARGBRow_AVX2, 1, 0, 4, 7)
ANY41C(I422AlphaToABGRRow_Any_AVX2, I422AlphaToABGRRow_AVX2, 1, 0, 4, 7)
#endif
#undef ANY41C
// Any 3 planes to 1.
......
source/row_gcc.cc
View file @ 9a0e12f5
This diff is collapsed.
source/row_win.cc
View file @ 9a0e12f5
......@@ -1896,6 +1896,23 @@ void RGBAToUVRow_SSSE3(const uint8* src_argb0, int src_stride_argb,
__asm lea eax, [eax + 16] \
}
// Read 8 UV from 422, upsample to 16 UV. With 16 Alpha.
#define READYUVA422_AVX2 __asm { \
__asm vmovq xmm0, qword ptr [esi] /* U */ \
__asm vmovq xmm1, qword ptr [esi + edi] /* V */ \
__asm lea esi, [esi + 8] \
__asm vpunpcklbw ymm0, ymm0, ymm1 /* UV */ \
__asm vpermq ymm0, ymm0, 0xd8 \
__asm vpunpcklwd ymm0, ymm0, ymm0 /* UVUV (upsample) */ \
__asm vmovdqu xmm4, [eax] /* Y */ \
__asm vpermq ymm4, ymm4, 0xd8 \
__asm vpunpcklbw ymm4, ymm4, ymm4 \
__asm lea eax, [eax + 16] \
__asm vmovdqu xmm5, [ebp] /* A */ \
__asm vpermq ymm5, ymm5, 0xd8 \
__asm lea ebp, [ebp + 16] \
}
// Read 4 UV from 411, upsample to 16 UV.
#define READYUV411_AVX2 __asm { \
__asm vmovd xmm0, dword ptr [esi] /* U */ \
......@@ -2057,6 +2074,92 @@ void I422ToARGBRow_AVX2(const uint8* y_buf,
}
#endif // HAS_I422TOARGBROW_AVX2
#ifdef HAS_I422ALPHATOARGBROW_AVX2
// 16 pixels
// 8 UV values upsampled to 16 UV, mixed with 16 Y and 16 A producing 16 ARGB.
__declspec(naked)
void I422AlphaToARGBRow_AVX2(const uint8* y_buf,
const uint8* u_buf,
const uint8* v_buf,
const uint8* a_buf,
uint8* dst_argb,
struct YuvConstants* yuvconstants,
int width) {
__asm {
push esi
push edi
push ebx
push ebp
mov eax, [esp + 16 + 4] // Y
mov esi, [esp + 16 + 8] // U
mov edi, [esp + 16 + 12] // V
mov ebp, [esp + 16 + 16] // A
mov edx, [esp + 16 + 20] // argb
mov ebx, [esp + 16 + 24] // yuvconstants
mov ecx, [esp + 16 + 28] // width
sub edi, esi
convertloop:
READYUVA422_AVX2
YUVTORGB_AVX2(ebx)
STOREARGB_AVX2
sub ecx, 16
jg convertloop
pop ebp
pop ebx
pop edi
pop esi
vzeroupper
ret
}
}
#endif // HAS_I422ALPHATOARGBROW_AVX2
#ifdef HAS_I422ALPHATOABGRROW_AVX2
// 16 pixels
// 8 UV values upsampled to 16 UV, mixed with 16 Y and 16 A producing 16 ABGR.
__declspec(naked)
void I422AlphaToABGRRow_AVX2(const uint8* y_buf,
const uint8* u_buf,
const uint8* v_buf,
const uint8* a_buf,
uint8* dst_abgr,
struct YuvConstants* yuvconstants,
int width) {
__asm {
push esi
push edi
push ebx
push ebp
mov eax, [esp + 16 + 4] // Y
mov esi, [esp + 16 + 8] // U
mov edi, [esp + 16 + 12] // V
mov ebp, [esp + 16 + 16] // A
mov edx, [esp + 16 + 20] // abgr
mov ebx, [esp + 16 + 24] // yuvconstants
mov ecx, [esp + 16 + 28] // width
sub edi, esi
convertloop:
READYUVA422_AVX2
YUVTORGB_AVX2(ebx)
STOREABGR_AVX2
sub ecx, 16
jg convertloop
pop ebp
pop ebx
pop edi
pop esi
vzeroupper
ret
}
}
#endif // HAS_I422ALPHATOABGRROW_AVX2
#ifdef HAS_I444TOARGBROW_AVX2
// 16 pixels
// 16 UV values with 16 Y producing 16 ARGB (64 bytes).
......@@ -2848,7 +2951,7 @@ void I422ToARGBRow_SSSE3(const uint8* y_buf,
}
// 8 pixels.
// 4 UV values upsampled to 8 UV, mixed with 8 Y and 8 A producing 8 ARGB (32 bytes).
// 4 UV values upsampled to 8 UV, mixed with 8 Y and 8 A producing 8 ARGB.
__declspec(naked)
void I422AlphaToARGBRow_SSSE3(const uint8* y_buf,
const uint8* u_buf,
......@@ -2870,7 +2973,6 @@ void I422AlphaToARGBRow_SSSE3(const uint8* y_buf,
mov ebx, [esp + 16 + 24] // yuvconstants
mov ecx, [esp + 16 + 28] // width
sub edi, esi
pcmpeqb xmm5, xmm5 // generate 0xffffffff for alpha
convertloop:
READYUVA422
......@@ -2889,7 +2991,7 @@ void I422AlphaToARGBRow_SSSE3(const uint8* y_buf,
}
// 8 pixels.
// 4 UV values upsampled to 8 UV, mixed with 8 Y and 8 A producing 8 ABGR (32 bytes).
// 4 UV values upsampled to 8 UV, mixed with 8 Y and 8 A producing 8 ABGR.
__declspec(naked)
void I422AlphaToABGRRow_SSSE3(const uint8* y_buf,
const uint8* u_buf,
......@@ -2911,7 +3013,6 @@ void I422AlphaToABGRRow_SSSE3(const uint8* y_buf,
mov ebx, [esp + 16 + 24] // yuvconstants
mov ecx, [esp + 16 + 28] // width
sub edi, esi
pcmpeqb xmm5, xmm5 // generate 0xffffffff for alpha
convertloop:
READYUVA422
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment