Skip to content

L
libyuv
Commit 140fc0a2 authored by Frank Barchard's avatar Frank Barchard Committed by Frank Barchard
Browse files
Options

Remove LIBYUV_SSSE3_ONLY and ARGBSHUFFLEROW_SSE2 

LIBYUV_SSSE3_ONLY was for functions that have SSE2 and SSSE3 but are compiling for SSSE3, so SSE2 will never be used.
Remove the SSE2 implementation of ARGBSHUFFLEROW_SSE2 and rely on SSSE3.

Bug: libyuv: 769
Test: ~/intelsde/sde -p4 -- out/Release/libyuv_unittest --gtest_filter=LibYUVConvertTest.ARGBToABGR_Opt
Change-Id: I7443f4d8ee3c6f47edd2cf1d5a1eb0f8d7a1eeeb
Reviewed-on: https://chromium-review.googlesource.com/846541Reviewed-by: 's avatarWeiyong Yao <braveyao@chromium.org>
Reviewed-by: 's avatarFrank Barchard <fbarchard@chromium.org>
parent 768f103b
Hide whitespace changes
Inline Side-by-side
Showing with 0 additions and 264 deletions
include/libyuv/row.h
View file @ 140fc0a2
......@@ -40,11 +40,6 @@ extern "C" {
// define LIBYUV_DISABLE_X86
#endif
#endif
// True if compiling for SSSE3 as a requirement.
#if defined(__SSSE3__) || (defined(_M_IX86_FP) && (_M_IX86_FP >= 3))
#define LIBYUV_SSSE3_ONLY
#endif
#if defined(__native_client__)
#define LIBYUV_DISABLE_NEON
#endif
......@@ -85,7 +80,6 @@ extern "C" {
#define HAS_ARGB4444TOARGBROW_SSE2
#define HAS_ARGBEXTRACTALPHAROW_SSE2
#define HAS_ARGBSETROW_X86
#define HAS_ARGBSHUFFLEROW_SSE2
#define HAS_ARGBSHUFFLEROW_SSSE3
#define HAS_ARGBTOARGB1555ROW_SSE2
#define HAS_ARGBTOARGB4444ROW_SSE2
......@@ -1536,10 +1530,6 @@ void ARGBShuffleRow_C(const uint8* src_argb,
uint8* dst_argb,
const uint8* shuffler,
int width);
void ARGBShuffleRow_SSE2(const uint8* src_argb,
uint8* dst_argb,
const uint8* shuffler,
int width);
void ARGBShuffleRow_SSSE3(const uint8* src_argb,
uint8* dst_argb,
const uint8* shuffler,
......@@ -1556,10 +1546,6 @@ void ARGBShuffleRow_MSA(const uint8* src_argb,
uint8* dst_argb,
const uint8* shuffler,
int width);
void ARGBShuffleRow_Any_SSE2(const uint8* src_argb,
uint8* dst_argb,
const uint8* shuffler,
int width);
void ARGBShuffleRow_Any_SSSE3(const uint8* src_argb,
uint8* dst_argb,
const uint8* shuffler,
......
source/planar_functions.cc
View file @ 140fc0a2
......@@ -2647,14 +2647,6 @@ int ARGBShuffle(const uint8* src_bgra,
height = 1;
src_stride_bgra = dst_stride_argb = 0;
}
#if defined(HAS_ARGBSHUFFLEROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2)) {
ARGBShuffleRow = ARGBShuffleRow_Any_SSE2;
if (IS_ALIGNED(width, 4)) {
ARGBShuffleRow = ARGBShuffleRow_SSE2;
}
}
#endif
#if defined(HAS_ARGBSHUFFLEROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3)) {
ARGBShuffleRow = ARGBShuffleRow_Any_SSSE3;
......
source/row_any.cc
View file @ 140fc0a2
......@@ -681,9 +681,6 @@ ANY11P(ARGBToRGB565DitherRow_Any_MSA,
2,
7)
#endif
#ifdef HAS_ARGBSHUFFLEROW_SSE2
ANY11P(ARGBShuffleRow_Any_SSE2, ARGBShuffleRow_SSE2, const uint8*, 4, 4, 3)
#endif
#ifdef HAS_ARGBSHUFFLEROW_SSSE3
ANY11P(ARGBShuffleRow_Any_SSSE3, ARGBShuffleRow_SSSE3, const uint8*, 4, 4, 7)
#endif
......
source/row_gcc.cc
View file @ 140fc0a2
......@@ -5687,128 +5687,6 @@ void ARGBShuffleRow_AVX2(const uint8* src_argb,
}
#endif // HAS_ARGBSHUFFLEROW_AVX2
#ifdef HAS_ARGBSHUFFLEROW_SSE2
// For BGRAToARGB, ABGRToARGB, RGBAToARGB, and ARGBToRGBA.
void ARGBShuffleRow_SSE2(const uint8* src_argb,
uint8* dst_argb,
const uint8* shuffler,
int width) {
uintptr_t pixel_temp;
asm volatile (
"pxor %%xmm5,%%xmm5 \n"
"mov " MEMACCESS(4) ",%k2 \n"
"cmp $0x3000102,%k2 \n"
"je 3012f \n"
"cmp $0x10203,%k2 \n"
"je 123f \n"
"cmp $0x30201,%k2 \n"
"je 321f \n"
"cmp $0x2010003,%k2 \n"
"je 2103f \n"
LABELALIGN
"1: \n"
"movzb " MEMACCESS(4) ",%2 \n"
MEMOPARG(movzb,0x00,0,2,1,2) " \n" // movzb (%0,%2,1),%2
"mov %b2," MEMACCESS(1) " \n"
"movzb " MEMACCESS2(0x1,4) ",%2 \n"
MEMOPARG(movzb,0x00,0,2,1,2) " \n" // movzb (%0,%2,1),%2
"mov %b2," MEMACCESS2(0x1,1) " \n"
"movzb " MEMACCESS2(0x2,4) ",%2 \n"
MEMOPARG(movzb,0x00,0,2,1,2) " \n" // movzb (%0,%2,1),%2
"mov %b2," MEMACCESS2(0x2,1) " \n"
"movzb " MEMACCESS2(0x3,4) ",%2 \n"
MEMOPARG(movzb,0x00,0,2,1,2) " \n" // movzb (%0,%2,1),%2
"mov %b2," MEMACCESS2(0x3,1) " \n"
"lea " MEMLEA(0x4,0) ",%0 \n"
"lea " MEMLEA(0x4,1) ",%1 \n"
"sub $0x1,%3 \n"
"jg 1b \n"
"jmp 99f \n"
LABELALIGN
"123: \n"
"movdqu " MEMACCESS(0) ",%%xmm0 \n"
"lea " MEMLEA(0x10,0) ",%0 \n"
"movdqa %%xmm0,%%xmm1 \n"
"punpcklbw %%xmm5,%%xmm0 \n"
"punpckhbw %%xmm5,%%xmm1 \n"
"pshufhw $0x1b,%%xmm0,%%xmm0 \n"
"pshuflw $0x1b,%%xmm0,%%xmm0 \n"
"pshufhw $0x1b,%%xmm1,%%xmm1 \n"
"pshuflw $0x1b,%%xmm1,%%xmm1 \n"
"packuswb %%xmm1,%%xmm0 \n"
"movdqu %%xmm0," MEMACCESS(1) " \n"
"lea " MEMLEA(0x10,1) ",%1 \n"
"sub $0x4,%3 \n"
"jg 123b \n"
"jmp 99f \n"
LABELALIGN
"321: \n"
"movdqu " MEMACCESS(0) ",%%xmm0 \n"
"lea " MEMLEA(0x10,0) ",%0 \n"
"movdqa %%xmm0,%%xmm1 \n"
"punpcklbw %%xmm5,%%xmm0 \n"
"punpckhbw %%xmm5,%%xmm1 \n"
"pshufhw $0x39,%%xmm0,%%xmm0 \n"
"pshuflw $0x39,%%xmm0,%%xmm0 \n"
"pshufhw $0x39,%%xmm1,%%xmm1 \n"
"pshuflw $0x39,%%xmm1,%%xmm1 \n"
"packuswb %%xmm1,%%xmm0 \n"
"movdqu %%xmm0," MEMACCESS(1) " \n"
"lea " MEMLEA(0x10,1) ",%1 \n"
"sub $0x4,%3 \n"
"jg 321b \n"
"jmp 99f \n"
LABELALIGN
"2103: \n"
"movdqu " MEMACCESS(0) ",%%xmm0 \n"
"lea " MEMLEA(0x10,0) ",%0 \n"
"movdqa %%xmm0,%%xmm1 \n"
"punpcklbw %%xmm5,%%xmm0 \n"
"punpckhbw %%xmm5,%%xmm1 \n"
"pshufhw $0x93,%%xmm0,%%xmm0 \n"
"pshuflw $0x93,%%xmm0,%%xmm0 \n"
"pshufhw $0x93,%%xmm1,%%xmm1 \n"
"pshuflw $0x93,%%xmm1,%%xmm1 \n"
"packuswb %%xmm1,%%xmm0 \n"
"movdqu %%xmm0," MEMACCESS(1) " \n"
"lea " MEMLEA(0x10,1) ",%1 \n"
"sub $0x4,%3 \n"
"jg 2103b \n"
"jmp 99f \n"
LABELALIGN
"3012: \n"
"movdqu " MEMACCESS(0) ",%%xmm0 \n"
"lea " MEMLEA(0x10,0) ",%0 \n"
"movdqa %%xmm0,%%xmm1 \n"
"punpcklbw %%xmm5,%%xmm0 \n"
"punpckhbw %%xmm5,%%xmm1 \n"
"pshufhw $0xc6,%%xmm0,%%xmm0 \n"
"pshuflw $0xc6,%%xmm0,%%xmm0 \n"
"pshufhw $0xc6,%%xmm1,%%xmm1 \n"
"pshuflw $0xc6,%%xmm1,%%xmm1 \n"
"packuswb %%xmm1,%%xmm0 \n"
"movdqu %%xmm0," MEMACCESS(1) " \n"
"lea " MEMLEA(0x10,1) ",%1 \n"
"sub $0x4,%3 \n"
"jg 3012b \n"
"99: \n"
: "+r"(src_argb), // %0
"+r"(dst_argb), // %1
"=&d"(pixel_temp), // %2
"+r"(width) // %3
: "r"(shuffler) // %4
: "memory", "cc", NACL_R14
"xmm0", "xmm1", "xmm5"
);
}
#endif // HAS_ARGBSHUFFLEROW_SSE2
#ifdef HAS_I422TOYUY2ROW_SSE2
void I422ToYUY2Row_SSE2(const uint8* src_y,
const uint8* src_u,
......
source/row_win.cc
View file @ 140fc0a2
......@@ -5761,123 +5761,6 @@ __declspec(naked) void ARGBShuffleRow_AVX2(const uint8* src_argb,
}
#endif // HAS_ARGBSHUFFLEROW_AVX2
__declspec(naked) void ARGBShuffleRow_SSE2(const uint8* src_argb,
uint8* dst_argb,
const uint8* shuffler,
int width) {
__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] // width
pxor xmm5, xmm5
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
shuf_0123:
movdqu xmm0, [eax]
lea eax, [eax + 16]
movdqa xmm1, xmm0
punpcklbw xmm0, xmm5
punpckhbw xmm1, xmm5
pshufhw xmm0, xmm0, 01Bh // 1B = 00011011 = 0x0123 = BGRAToARGB
pshuflw xmm0, xmm0, 01Bh
pshufhw xmm1, xmm1, 01Bh
pshuflw xmm1, xmm1, 01Bh
packuswb xmm0, xmm1
movdqu [edx], xmm0
lea edx, [edx + 16]
sub ecx, 4
jg shuf_0123
jmp shuf99
shuf_0321:
movdqu xmm0, [eax]
lea eax, [eax + 16]
movdqa xmm1, xmm0
punpcklbw xmm0, xmm5
punpckhbw xmm1, xmm5
pshufhw xmm0, xmm0, 039h // 39 = 00111001 = 0x0321 = RGBAToARGB
pshuflw xmm0, xmm0, 039h
pshufhw xmm1, xmm1, 039h
pshuflw xmm1, xmm1, 039h
packuswb xmm0, xmm1
movdqu [edx], xmm0
lea edx, [edx + 16]
sub ecx, 4
jg shuf_0321
jmp shuf99
shuf_2103:
movdqu xmm0, [eax]
lea eax, [eax + 16]
movdqa xmm1, xmm0
punpcklbw xmm0, xmm5
punpckhbw xmm1, xmm5
pshufhw xmm0, xmm0, 093h // 93 = 10010011 = 0x2103 = ARGBToRGBA
pshuflw xmm0, xmm0, 093h
pshufhw xmm1, xmm1, 093h
pshuflw xmm1, xmm1, 093h
packuswb xmm0, xmm1
movdqu [edx], xmm0
lea edx, [edx + 16]
sub ecx, 4
jg shuf_2103
jmp shuf99
shuf_3012:
movdqu xmm0, [eax]
lea eax, [eax + 16]
movdqa xmm1, xmm0
punpcklbw xmm0, xmm5
punpckhbw xmm1, xmm5
pshufhw xmm0, xmm0, 0C6h // C6 = 11000110 = 0x3012 = ABGRToARGB
pshuflw xmm0, xmm0, 0C6h
pshufhw xmm1, xmm1, 0C6h
pshuflw xmm1, xmm1, 0C6h
packuswb xmm0, xmm1
movdqu [edx], xmm0
lea edx, [edx + 16]
sub ecx, 4
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