Skip to content

L
libyuv
Commit 212a1a50 authored by fbarchard@google.com's avatar fbarchard@google.com
Browse files
Options

ARGBShuffle_SSE2 for lower end CPUs 

BUG=271
TESTED=out\release\libyuv_unittest --gtest_filter=**R*ToARGB*
R=johannkoenig@google.com, ryanpetrie@google.com

Review URL: https://webrtc-codereview.appspot.com/2361004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@807 16f28f9a-4ce2-e073-06de-1de4eb20be90
parent 98fb959e
Show whitespace changes
Inline Side-by-side
Showing with 143 additions and 5 deletions
README.chromium
View file @ 212a1a50
Name: libyuv
URL: http://code.google.com/p/libyuv/
Version: 806
Version: 807
License: BSD
License File: LICENSE
......
include/libyuv/row.h
View file @ 212a1a50
......@@ -163,6 +163,8 @@ extern "C" {
// The following are Windows only:
// TODO(fbarchard): Port to gcc.
#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && defined(_MSC_VER)
#define HAS_ARGBSHUFFLEROW_SSE2
// Effects:
// TODO(fbarchard): Optimize and enable
// #define HAS_ARGBLUMACOLORTABLEROW_SSSE3
......@@ -709,6 +711,8 @@ void ARGBSetRows_C(uint8* dst, uint32 v32, int width, int dst_stride,
// ARGBShufflers for BGRAToARGB etc.
void ARGBShuffleRow_C(const uint8* src_argb, uint8* dst_argb,
const uint8* shuffler, int pix);
void ARGBShuffleRow_SSE2(const uint8* src_argb, uint8* dst_argb,
const uint8* shuffler, int pix);
void ARGBShuffleRow_SSSE3(const uint8* src_argb, uint8* dst_argb,
const uint8* shuffler, int pix);
void ARGBShuffleRow_AVX2(const uint8* src_argb, uint8* dst_argb,
......@@ -717,6 +721,8 @@ void ARGBShuffleRow_NEON(const uint8* src_argb, uint8* dst_argb,
const uint8* shuffler, int pix);
void ARGBShuffleRow_Unaligned_SSSE3(const uint8* src_argb, uint8* dst_argb,
const uint8* shuffler, int pix);
void ARGBShuffleRow_Any_SSE2(const uint8* src_argb, uint8* dst_argb,
const uint8* shuffler, int pix);
void ARGBShuffleRow_Any_SSSE3(const uint8* src_argb, uint8* dst_argb,
const uint8* shuffler, int pix);
void ARGBShuffleRow_Any_AVX2(const uint8* src_argb, uint8* dst_argb,
......
include/libyuv/version.h
View file @ 212a1a50
......@@ -11,6 +11,6 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_ // NOLINT
#define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 806
#define LIBYUV_VERSION 807
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT
source/planar_functions.cc
View file @ 212a1a50
......@@ -1815,6 +1815,14 @@ int ARGBShuffle(const uint8* src_bgra, int src_stride_bgra,
}
void (*ARGBShuffleRow)(const uint8* src_bgra, uint8* dst_argb,
const uint8* shuffler, int pix) = ARGBShuffleRow_C;
#if defined(HAS_ARGBSHUFFLEROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2) && width >= 4) {
ARGBShuffleRow = ARGBShuffleRow_Any_SSE2;
if (IS_ALIGNED(width, 4)) {
ARGBShuffleRow = ARGBShuffleRow_SSE2;
}
}
#endif
#if defined(HAS_ARGBSHUFFLEROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3) && width >= 8) {
ARGBShuffleRow = ARGBShuffleRow_Any_SSSE3;
......
source/row_any.cc
View file @ 212a1a50
......@@ -474,6 +474,10 @@ MATHROW_ANY(ARGBSubtractRow_Any_NEON, ARGBSubtractRow_NEON, ARGBSubtractRow_C,
dst_argb + n * BPP, shuffler, width & MASK); \
}
#ifdef HAS_ARGBSHUFFLEROW_SSE2
YANY(ARGBShuffleRow_Any_SSE2, ARGBShuffleRow_SSE2,
ARGBShuffleRow_C, 4, 4, 3)
#endif
#ifdef HAS_ARGBSHUFFLEROW_SSSE3
YANY(ARGBShuffleRow_Any_SSSE3, ARGBShuffleRow_Unaligned_SSSE3,
ARGBShuffleRow_C, 4, 4, 7)
......
source/row_win.cc
View file @ 212a1a50
......@@ -6552,7 +6552,7 @@ void ARGBShuffleRow_SSSE3(const uint8* src_argb, uint8* dst_argb,
const uint8* shuffler, int pix) {
__asm {
mov eax, [esp + 4] // src_argb
mov edx, [esp + 8] // dst_bayer
mov edx, [esp + 8] // dst_argb
mov ecx, [esp + 12] // shuffler
movdqa xmm5, [ecx]
mov ecx, [esp + 16] // pix
......@@ -6578,7 +6578,7 @@ void ARGBShuffleRow_Unaligned_SSSE3(const uint8* src_argb, uint8* dst_argb,
const uint8* shuffler, int pix) {
__asm {
mov eax, [esp + 4] // src_argb
mov edx, [esp + 8] // dst_bayer
mov edx, [esp + 8] // dst_argb
mov ecx, [esp + 12] // shuffler
movdqa xmm5, [ecx]
mov ecx, [esp + 16] // pix
......@@ -6605,7 +6605,7 @@ void ARGBShuffleRow_AVX2(const uint8* src_argb, uint8* dst_argb,
const uint8* shuffler, int pix) {
__asm {
mov eax, [esp + 4] // src_argb
mov edx, [esp + 8] // dst_bayer
mov edx, [esp + 8] // dst_argb
mov ecx, [esp + 12] // shuffler
vbroadcastf128 ymm5, [ecx] // same shuffle in high as low.
mov ecx, [esp + 16] // pix
......@@ -6629,6 +6629,126 @@ void ARGBShuffleRow_AVX2(const uint8* src_argb, uint8* dst_argb,
}
#endif
__declspec(naked) __declspec(align(16))
void ARGBShuffleRow_SSE2(const uint8* src_argb, uint8* dst_argb,
const uint8* shuffler, int pix) {
__asm {
push ebx
push esi
mov eax, [esp + 8 + 4] // src_argb
mov edx, [esp + 8 + 8] // dst_argb
mov esi, [esp + 8 + 12] // shuffler
mov ecx, [esp + 8 + 16] // pix
pxor xmm7, xmm7
mov ebx, [esi] // shuffler
cmp ebx, 0x03000102
je shuf_3012
cmp ebx, 0x00010203
je shuf_0123
cmp ebx, 0x00030201
je shuf_0321
cmp ebx, 0x02010003
je shuf_2103
// TODO(fbarchard): Use one source pointer and 3 offsets.
shuf_any1:
movzx ebx, byte ptr [esi]
movzx ebx, byte ptr [eax + ebx]
mov [edx], bl
movzx ebx, byte ptr [esi + 1]
movzx ebx, byte ptr [eax + ebx]
mov [edx + 1], bl
movzx ebx, byte ptr [esi + 2]
movzx ebx, byte ptr [eax + ebx]
mov [edx + 2], bl
movzx ebx, byte ptr [esi + 3]
movzx ebx, byte ptr [eax + ebx]
mov [edx + 3], bl
lea eax, [eax + 4]
lea edx, [edx + 4]
sub ecx, 1
jg shuf_any1
jmp shuf99
align 16
shuf_0123:
movdqu xmm0, [eax]
lea eax, [eax + 16]
movdqa xmm1, xmm0
punpcklbw xmm0, xmm7
punpckhbw xmm1, xmm7
pshufhw xmm0, xmm0, 01Bh // 1B = 00011011 = 0x0123 = BGRAToARGB
pshuflw xmm0, xmm0, 01Bh
pshufhw xmm1, xmm1, 01Bh
pshuflw xmm1, xmm1, 01Bh
packuswb xmm0, xmm1
sub ecx, 4
movdqu [edx], xmm0
lea edx, [edx + 16]
jg shuf_0123
jmp shuf99
align 16
shuf_0321:
movdqu xmm0, [eax]
lea eax, [eax + 16]
movdqa xmm1, xmm0
punpcklbw xmm0, xmm7
punpckhbw xmm1, xmm7
pshufhw xmm0, xmm0, 039h // 39 = 00111001 = 0x0321 = RGBAToARGB
pshuflw xmm0, xmm0, 039h
pshufhw xmm1, xmm1, 039h
pshuflw xmm1, xmm1, 039h
packuswb xmm0, xmm1
sub ecx, 4
movdqu [edx], xmm0
lea edx, [edx + 16]
jg shuf_0321
jmp shuf99
align 16
shuf_2103:
movdqu xmm0, [eax]
lea eax, [eax + 16]
movdqa xmm1, xmm0
punpcklbw xmm0, xmm7
punpckhbw xmm1, xmm7
pshufhw xmm0, xmm0, 093h // 93 = 10010011 = 0x2103 = ARGBToRGBA
pshuflw xmm0, xmm0, 093h
pshufhw xmm1, xmm1, 093h
pshuflw xmm1, xmm1, 093h
packuswb xmm0, xmm1
sub ecx, 4
movdqu [edx], xmm0
lea edx, [edx + 16]
jg shuf_2103
jmp shuf99
align 16
shuf_3012:
movdqu xmm0, [eax]
lea eax, [eax + 16]
movdqa xmm1, xmm0
punpcklbw xmm0, xmm7
punpckhbw xmm1, xmm7
pshufhw xmm0, xmm0, 0C6h // C6 = 11000110 = 0x3012 = ABGRToARGB
pshuflw xmm0, xmm0, 0C6h
pshufhw xmm1, xmm1, 0C6h
pshuflw xmm1, xmm1, 0C6h
packuswb xmm0, xmm1
sub ecx, 4
movdqu [edx], xmm0
lea edx, [edx + 16]
jg shuf_3012
shuf99:
pop esi
pop ebx
ret
}
}
// YUY2 - Macro-pixel = 2 image pixels
// Y0U0Y1V0....Y2U2Y3V2...Y4U4Y5V4....
......
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