diff options
| author | James Bankoski <jimbankoski@google.com> | 2016-06-29 23:05:51 +0000 |
|---|---|---|
| committer | James Bankoski <jimbankoski@google.com> | 2016-06-29 23:09:10 +0000 |
| commit | 291033032ed1abf4f43ad52165caea31a35fb33d (patch) | |
| tree | 00e8c710cf4d7f1b29f4559dcea1b8082dc2cedf /third_party/libyuv/source/row_gcc.cc | |
| parent | b8f83282f8506ad3d839440341bbe683df7d2cf6 (diff) | |
| download | libvpx-291033032ed1abf4f43ad52165caea31a35fb33d.tar libvpx-291033032ed1abf4f43ad52165caea31a35fb33d.tar.gz libvpx-291033032ed1abf4f43ad52165caea31a35fb33d.tar.bz2 libvpx-291033032ed1abf4f43ad52165caea31a35fb33d.zip | |
Revert "libyuv: update to b8ddb5a2"
This reverts commit b8f83282f8506ad3d839440341bbe683df7d2cf6.
Update was to wrong version and still has:
BUG=webm:1252
Change-Id: I80f3a7c0581ab5e2dd1a84f7840e51d7c362afac
Diffstat (limited to 'third_party/libyuv/source/row_gcc.cc')
| -rw-r--r-- | third_party/libyuv/source/row_gcc.cc | 2119 |
1 files changed, 1030 insertions, 1089 deletions
diff --git a/third_party/libyuv/source/row_gcc.cc b/third_party/libyuv/source/row_gcc.cc index 1ac7ef1aa..820de0a1c 100644 --- a/third_party/libyuv/source/row_gcc.cc +++ b/third_party/libyuv/source/row_gcc.cc @@ -17,8 +17,7 @@ extern "C" { #endif // This module is for GCC x86 and x64. -#if !defined(LIBYUV_DISABLE_X86) && \ - (defined(__x86_64__) || (defined(__i386__) && !defined(_MSC_VER))) +#if !defined(LIBYUV_DISABLE_X86) && (defined(__x86_64__) || defined(__i386__)) #if defined(HAS_ARGBTOYROW_SSSE3) || defined(HAS_ARGBGRAYROW_SSSE3) @@ -121,24 +120,6 @@ static uvec8 kShuffleMaskRAWToARGB = { 2u, 1u, 0u, 12u, 5u, 4u, 3u, 13u, 8u, 7u, 6u, 14u, 11u, 10u, 9u, 15u }; -// Shuffle table for converting RAW to RGB24. First 8. -static const uvec8 kShuffleMaskRAWToRGB24_0 = { - 2u, 1u, 0u, 5u, 4u, 3u, 8u, 7u, - 128u, 128u, 128u, 128u, 128u, 128u, 128u, 128u -}; - -// Shuffle table for converting RAW to RGB24. Middle 8. -static const uvec8 kShuffleMaskRAWToRGB24_1 = { - 2u, 7u, 6u, 5u, 10u, 9u, 8u, 13u, - 128u, 128u, 128u, 128u, 128u, 128u, 128u, 128u -}; - -// Shuffle table for converting RAW to RGB24. Last 8. -static const uvec8 kShuffleMaskRAWToRGB24_2 = { - 8u, 7u, 12u, 11u, 10u, 15u, 14u, 13u, - 128u, 128u, 128u, 128u, 128u, 128u, 128u, 128u -}; - // Shuffle table for converting ARGB to RGB24. static uvec8 kShuffleMaskARGBToRGB24 = { 0u, 1u, 2u, 4u, 5u, 6u, 8u, 9u, 10u, 12u, 13u, 14u, 128u, 128u, 128u, 128u @@ -154,39 +135,109 @@ static uvec8 kShuffleMaskARGBToRGB24_0 = { 0u, 1u, 2u, 4u, 5u, 6u, 8u, 9u, 128u, 128u, 128u, 128u, 10u, 12u, 13u, 14u }; -// YUY2 shuf 16 Y to 32 Y. -static const lvec8 kShuffleYUY2Y = { - 0, 0, 2, 2, 4, 4, 6, 6, 8, 8, 10, 10, 12, 12, 14, 14, - 0, 0, 2, 2, 4, 4, 6, 6, 8, 8, 10, 10, 12, 12, 14, 14 -}; - -// YUY2 shuf 8 UV to 16 UV. -static const lvec8 kShuffleYUY2UV = { - 1, 3, 1, 3, 5, 7, 5, 7, 9, 11, 9, 11, 13, 15, 13, 15, - 1, 3, 1, 3, 5, 7, 5, 7, 9, 11, 9, 11, 13, 15, 13, 15 -}; - -// UYVY shuf 16 Y to 32 Y. -static const lvec8 kShuffleUYVYY = { - 1, 1, 3, 3, 5, 5, 7, 7, 9, 9, 11, 11, 13, 13, 15, 15, - 1, 1, 3, 3, 5, 5, 7, 7, 9, 9, 11, 11, 13, 13, 15, 15 -}; - -// UYVY shuf 8 UV to 16 UV. -static const lvec8 kShuffleUYVYUV = { - 0, 2, 0, 2, 4, 6, 4, 6, 8, 10, 8, 10, 12, 14, 12, 14, - 0, 2, 0, 2, 4, 6, 4, 6, 8, 10, 8, 10, 12, 14, 12, 14 -}; - -// NV21 shuf 8 VU to 16 UV. -static const lvec8 kShuffleNV21 = { - 1, 0, 1, 0, 3, 2, 3, 2, 5, 4, 5, 4, 7, 6, 7, 6, - 1, 0, 1, 0, 3, 2, 3, 2, 5, 4, 5, 4, 7, 6, 7, 6, +// Shuffle table for converting ARGB to RAW. +static uvec8 kShuffleMaskARGBToRAW_0 = { + 2u, 1u, 0u, 6u, 5u, 4u, 10u, 9u, 128u, 128u, 128u, 128u, 8u, 14u, 13u, 12u }; #endif // HAS_RGB24TOARGBROW_SSSE3 +#if defined(TESTING) && defined(__x86_64__) +void TestRow_SSE2(const uint8* src_y, uint8* dst_argb, int pix) { + asm volatile ( + ".p2align 5 \n" + "mov %%eax,%%eax \n" + "mov %%ebx,%%ebx \n" + "mov %%ecx,%%ecx \n" + "mov %%edx,%%edx \n" + "mov %%esi,%%esi \n" + "mov %%edi,%%edi \n" + "mov %%ebp,%%ebp \n" + "mov %%esp,%%esp \n" + ".p2align 5 \n" + "mov %%r8d,%%r8d \n" + "mov %%r9d,%%r9d \n" + "mov %%r10d,%%r10d \n" + "mov %%r11d,%%r11d \n" + "mov %%r12d,%%r12d \n" + "mov %%r13d,%%r13d \n" + "mov %%r14d,%%r14d \n" + "mov %%r15d,%%r15d \n" + ".p2align 5 \n" + "lea (%%rax),%%eax \n" + "lea (%%rbx),%%ebx \n" + "lea (%%rcx),%%ecx \n" + "lea (%%rdx),%%edx \n" + "lea (%%rsi),%%esi \n" + "lea (%%rdi),%%edi \n" + "lea (%%rbp),%%ebp \n" + "lea (%%rsp),%%esp \n" + ".p2align 5 \n" + "lea (%%r8),%%r8d \n" + "lea (%%r9),%%r9d \n" + "lea (%%r10),%%r10d \n" + "lea (%%r11),%%r11d \n" + "lea (%%r12),%%r12d \n" + "lea (%%r13),%%r13d \n" + "lea (%%r14),%%r14d \n" + "lea (%%r15),%%r15d \n" + + ".p2align 5 \n" + "lea 0x10(%%rax),%%eax \n" + "lea 0x10(%%rbx),%%ebx \n" + "lea 0x10(%%rcx),%%ecx \n" + "lea 0x10(%%rdx),%%edx \n" + "lea 0x10(%%rsi),%%esi \n" + "lea 0x10(%%rdi),%%edi \n" + "lea 0x10(%%rbp),%%ebp \n" + "lea 0x10(%%rsp),%%esp \n" + ".p2align 5 \n" + "lea 0x10(%%r8),%%r8d \n" + "lea 0x10(%%r9),%%r9d \n" + "lea 0x10(%%r10),%%r10d \n" + "lea 0x10(%%r11),%%r11d \n" + "lea 0x10(%%r12),%%r12d \n" + "lea 0x10(%%r13),%%r13d \n" + "lea 0x10(%%r14),%%r14d \n" + "lea 0x10(%%r15),%%r15d \n" + + ".p2align 5 \n" + "add 0x10,%%eax \n" + "add 0x10,%%ebx \n" + "add 0x10,%%ecx \n" + "add 0x10,%%edx \n" + "add 0x10,%%esi \n" + "add 0x10,%%edi \n" + "add 0x10,%%ebp \n" + "add 0x10,%%esp \n" + ".p2align 5 \n" + "add 0x10,%%r8d \n" + "add 0x10,%%r9d \n" + "add 0x10,%%r10d \n" + "add 0x10,%%r11d \n" + "add 0x10,%%r12d \n" + "add 0x10,%%r13d \n" + "add 0x10,%%r14d \n" + "add 0x10,%%r15d \n" + + ".p2align 2 \n" + "1: \n" + "movq " MEMACCESS(0) ",%%xmm0 \n" + "lea " MEMLEA(0x8,0) ",%0 \n" + "movdqu %%xmm0," MEMACCESS(1) " \n" + "lea " MEMLEA(0x20,1) ",%1 \n" + "sub $0x8,%2 \n" + "jg 1b \n" + : "+r"(src_y), // %0 + "+r"(dst_argb), // %1 + "+r"(pix) // %2 + : + : "memory", "cc", "xmm0", "xmm1", "xmm5" + ); +} +#endif // TESTING + #ifdef HAS_J400TOARGBROW_SSE2 -void J400ToARGBRow_SSE2(const uint8* src_y, uint8* dst_argb, int width) { +void J400ToARGBRow_SSE2(const uint8* src_y, uint8* dst_argb, int pix) { asm volatile ( "pcmpeqb %%xmm5,%%xmm5 \n" "pslld $0x18,%%xmm5 \n" @@ -207,14 +258,14 @@ void J400ToARGBRow_SSE2(const uint8* src_y, uint8* dst_argb, int width) { "jg 1b \n" : "+r"(src_y), // %0 "+r"(dst_argb), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 :: "memory", "cc", "xmm0", "xmm1", "xmm5" ); } #endif // HAS_J400TOARGBROW_SSE2 #ifdef HAS_RGB24TOARGBROW_SSSE3 -void RGB24ToARGBRow_SSSE3(const uint8* src_rgb24, uint8* dst_argb, int width) { +void RGB24ToARGBRow_SSSE3(const uint8* src_rgb24, uint8* dst_argb, int pix) { asm volatile ( "pcmpeqb %%xmm5,%%xmm5 \n" // generate mask 0xff000000 "pslld $0x18,%%xmm5 \n" @@ -246,13 +297,13 @@ void RGB24ToARGBRow_SSSE3(const uint8* src_rgb24, uint8* dst_argb, int width) { "jg 1b \n" : "+r"(src_rgb24), // %0 "+r"(dst_argb), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 : "m"(kShuffleMaskRGB24ToARGB) // %3 : "memory", "cc" , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" ); } -void RAWToARGBRow_SSSE3(const uint8* src_raw, uint8* dst_argb, int width) { +void RAWToARGBRow_SSSE3(const uint8* src_raw, uint8* dst_argb, int pix) { asm volatile ( "pcmpeqb %%xmm5,%%xmm5 \n" // generate mask 0xff000000 "pslld $0x18,%%xmm5 \n" @@ -284,43 +335,13 @@ void RAWToARGBRow_SSSE3(const uint8* src_raw, uint8* dst_argb, int width) { "jg 1b \n" : "+r"(src_raw), // %0 "+r"(dst_argb), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 : "m"(kShuffleMaskRAWToARGB) // %3 : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" ); } -void RAWToRGB24Row_SSSE3(const uint8* src_raw, uint8* dst_rgb24, int width) { - asm volatile ( - "movdqa %3,%%xmm3 \n" - "movdqa %4,%%xmm4 \n" - "movdqa %5,%%xmm5 \n" - LABELALIGN - "1: \n" - "movdqu " MEMACCESS(0) ",%%xmm0 \n" - "movdqu " MEMACCESS2(0x4,0) ",%%xmm1 \n" - "movdqu " MEMACCESS2(0x8,0) ",%%xmm2 \n" - "lea " MEMLEA(0x18,0) ",%0 \n" - "pshufb %%xmm3,%%xmm0 \n" - "pshufb %%xmm4,%%xmm1 \n" - "pshufb %%xmm5,%%xmm2 \n" - "movq %%xmm0," MEMACCESS(1) " \n" - "movq %%xmm1," MEMACCESS2(0x8,1) " \n" - "movq %%xmm2," MEMACCESS2(0x10,1) " \n" - "lea " MEMLEA(0x18,1) ",%1 \n" - "sub $0x8,%2 \n" - "jg 1b \n" - : "+r"(src_raw), // %0 - "+r"(dst_rgb24), // %1 - "+r"(width) // %2 - : "m"(kShuffleMaskRAWToRGB24_0), // %3 - "m"(kShuffleMaskRAWToRGB24_1), // %4 - "m"(kShuffleMaskRAWToRGB24_2) // %5 - : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" - ); -} - -void RGB565ToARGBRow_SSE2(const uint8* src, uint8* dst, int width) { +void RGB565ToARGBRow_SSE2(const uint8* src, uint8* dst, int pix) { asm volatile ( "mov $0x1080108,%%eax \n" "movd %%eax,%%xmm5 \n" @@ -361,14 +382,14 @@ void RGB565ToARGBRow_SSE2(const uint8* src, uint8* dst, int width) { "jg 1b \n" : "+r"(src), // %0 "+r"(dst), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 : : "memory", "cc", "eax", NACL_R14 "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7" ); } -void ARGB1555ToARGBRow_SSE2(const uint8* src, uint8* dst, int width) { +void ARGB1555ToARGBRow_SSE2(const uint8* src, uint8* dst, int pix) { asm volatile ( "mov $0x1080108,%%eax \n" "movd %%eax,%%xmm5 \n" @@ -412,14 +433,14 @@ void ARGB1555ToARGBRow_SSE2(const uint8* src, uint8* dst, int width) { "jg 1b \n" : "+r"(src), // %0 "+r"(dst), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 : : "memory", "cc", "eax", NACL_R14 "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7" ); } -void ARGB4444ToARGBRow_SSE2(const uint8* src, uint8* dst, int width) { +void ARGB4444ToARGBRow_SSE2(const uint8* src, uint8* dst, int pix) { asm volatile ( "mov $0xf0f0f0f,%%eax \n" "movd %%eax,%%xmm4 \n" @@ -450,14 +471,14 @@ void ARGB4444ToARGBRow_SSE2(const uint8* src, uint8* dst, int width) { "jg 1b \n" : "+r"(src), // %0 "+r"(dst), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 : : "memory", "cc", "eax", NACL_R14 "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" ); } -void ARGBToRGB24Row_SSSE3(const uint8* src, uint8* dst, int width) { +void ARGBToRGB24Row_SSSE3(const uint8* src, uint8* dst, int pix) { asm volatile ( "movdqa %3,%%xmm6 \n" LABELALIGN @@ -489,13 +510,13 @@ void ARGBToRGB24Row_SSSE3(const uint8* src, uint8* dst, int width) { "jg 1b \n" : "+r"(src), // %0 "+r"(dst), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 : "m"(kShuffleMaskARGBToRGB24) // %3 : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6" ); } -void ARGBToRAWRow_SSSE3(const uint8* src, uint8* dst, int width) { +void ARGBToRAWRow_SSSE3(const uint8* src, uint8* dst, int pix) { asm volatile ( "movdqa %3,%%xmm6 \n" LABELALIGN @@ -527,13 +548,13 @@ void ARGBToRAWRow_SSSE3(const uint8* src, uint8* dst, int width) { "jg 1b \n" : "+r"(src), // %0 "+r"(dst), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 : "m"(kShuffleMaskARGBToRAW) // %3 : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6" ); } -void ARGBToRGB565Row_SSE2(const uint8* src, uint8* dst, int width) { +void ARGBToRGB565Row_SSE2(const uint8* src, uint8* dst, int pix) { asm volatile ( "pcmpeqb %%xmm3,%%xmm3 \n" "psrld $0x1b,%%xmm3 \n" @@ -564,104 +585,12 @@ void ARGBToRGB565Row_SSE2(const uint8* src, uint8* dst, int width) { "jg 1b \n" : "+r"(src), // %0 "+r"(dst), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 :: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" ); } -void ARGBToRGB565DitherRow_SSE2(const uint8* src, uint8* dst, - const uint32 dither4, int width) { - asm volatile ( - "movd %3,%%xmm6 \n" - "punpcklbw %%xmm6,%%xmm6 \n" - "movdqa %%xmm6,%%xmm7 \n" - "punpcklwd %%xmm6,%%xmm6 \n" - "punpckhwd %%xmm7,%%xmm7 \n" - "pcmpeqb %%xmm3,%%xmm3 \n" - "psrld $0x1b,%%xmm3 \n" - "pcmpeqb %%xmm4,%%xmm4 \n" - "psrld $0x1a,%%xmm4 \n" - "pslld $0x5,%%xmm4 \n" - "pcmpeqb %%xmm5,%%xmm5 \n" - "pslld $0xb,%%xmm5 \n" - - LABELALIGN - "1: \n" - "movdqu (%0),%%xmm0 \n" - "paddusb %%xmm6,%%xmm0 \n" - "movdqa %%xmm0,%%xmm1 \n" - "movdqa %%xmm0,%%xmm2 \n" - "pslld $0x8,%%xmm0 \n" - "psrld $0x3,%%xmm1 \n" - "psrld $0x5,%%xmm2 \n" - "psrad $0x10,%%xmm0 \n" - "pand %%xmm3,%%xmm1 \n" - "pand %%xmm4,%%xmm2 \n" - "pand %%xmm5,%%xmm0 \n" - "por %%xmm2,%%xmm1 \n" - "por %%xmm1,%%xmm0 \n" - "packssdw %%xmm0,%%xmm0 \n" - "lea 0x10(%0),%0 \n" - "movq %%xmm0,(%1) \n" - "lea 0x8(%1),%1 \n" - "sub $0x4,%2 \n" - "jg 1b \n" - : "+r"(src), // %0 - "+r"(dst), // %1 - "+r"(width) // %2 - : "m"(dither4) // %3 - : "memory", "cc", - "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7" - ); -} - -#ifdef HAS_ARGBTORGB565DITHERROW_AVX2 -void ARGBToRGB565DitherRow_AVX2(const uint8* src, uint8* dst, - const uint32 dither4, int width) { - asm volatile ( - "vbroadcastss %3,%%xmm6 \n" - "vpunpcklbw %%xmm6,%%xmm6,%%xmm6 \n" - "vpermq $0xd8,%%ymm6,%%ymm6 \n" - "vpunpcklwd %%ymm6,%%ymm6,%%ymm6 \n" - "vpcmpeqb %%ymm3,%%ymm3,%%ymm3 \n" - "vpsrld $0x1b,%%ymm3,%%ymm3 \n" - "vpcmpeqb %%ymm4,%%ymm4,%%ymm4 \n" - "vpsrld $0x1a,%%ymm4,%%ymm4 \n" - "vpslld $0x5,%%ymm4,%%ymm4 \n" - "vpslld $0xb,%%ymm3,%%ymm5 \n" - - LABELALIGN - "1: \n" - "vmovdqu (%0),%%ymm0 \n" - "vpaddusb %%ymm6,%%ymm0,%%ymm0 \n" - "vpsrld $0x5,%%ymm0,%%ymm2 \n" - "vpsrld $0x3,%%ymm0,%%ymm1 \n" - "vpsrld $0x8,%%ymm0,%%ymm0 \n" - "vpand %%ymm4,%%ymm2,%%ymm2 \n" - "vpand %%ymm3,%%ymm1,%%ymm1 \n" - "vpand %%ymm5,%%ymm0,%%ymm0 \n" - "vpor %%ymm2,%%ymm1,%%ymm1 \n" - "vpor %%ymm1,%%ymm0,%%ymm0 \n" - "vpackusdw %%ymm0,%%ymm0,%%ymm0 \n" - "vpermq $0xd8,%%ymm0,%%ymm0 \n" - "lea 0x20(%0),%0 \n" - "vmovdqu %%xmm0,(%1) \n" - "lea 0x10(%1),%1 \n" - "sub $0x8,%2 \n" - "jg 1b \n" - "vzeroupper \n" - : "+r"(src), // %0 - "+r"(dst), // %1 - "+r"(width) // %2 - : "m"(dither4) // %3 - : "memory", "cc", - "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7" - ); -} -#endif // HAS_ARGBTORGB565DITHERROW_AVX2 - - -void ARGBToARGB1555Row_SSE2(const uint8* src, uint8* dst, int width) { +void ARGBToARGB1555Row_SSE2(const uint8* src, uint8* dst, int pix) { asm volatile ( "pcmpeqb %%xmm4,%%xmm4 \n" "psrld $0x1b,%%xmm4 \n" @@ -696,13 +625,13 @@ void ARGBToARGB1555Row_SSE2(const uint8* src, uint8* dst, int width) { "jg 1b \n" : "+r"(src), // %0 "+r"(dst), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 :: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7" ); } -void ARGBToARGB4444Row_SSE2(const uint8* src, uint8* dst, int width) { +void ARGBToARGB4444Row_SSE2(const uint8* src, uint8* dst, int pix) { asm volatile ( "pcmpeqb %%xmm4,%%xmm4 \n" "psllw $0xc,%%xmm4 \n" @@ -725,7 +654,7 @@ void ARGBToARGB4444Row_SSE2(const uint8* src, uint8* dst, int width) { "jg 1b \n" : "+r"(src), // %0 "+r"(dst), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 :: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4" ); } @@ -733,7 +662,7 @@ void ARGBToARGB4444Row_SSE2(const uint8* src, uint8* dst, int width) { #ifdef HAS_ARGBTOYROW_SSSE3 // Convert 16 ARGB pixels (64 bytes) to 16 Y values. -void ARGBToYRow_SSSE3(const uint8* src_argb, uint8* dst_y, int width) { +void ARGBToYRow_SSSE3(const uint8* src_argb, uint8* dst_y, int pix) { asm volatile ( "movdqa %3,%%xmm4 \n" "movdqa %4,%%xmm5 \n" @@ -760,7 +689,7 @@ void ARGBToYRow_SSSE3(const uint8* src_argb, uint8* dst_y, int width) { "jg 1b \n" : "+r"(src_argb), // %0 "+r"(dst_y), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 : "m"(kARGBToY), // %3 "m"(kAddY16) // %4 : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" @@ -771,7 +700,7 @@ void ARGBToYRow_SSSE3(const uint8* src_argb, uint8* dst_y, int width) { #ifdef HAS_ARGBTOYJROW_SSSE3 // Convert 16 ARGB pixels (64 bytes) to 16 YJ values. // Same as ARGBToYRow but different coefficients, no add 16, but do rounding. -void ARGBToYJRow_SSSE3(const uint8* src_argb, uint8* dst_y, int width) { +void ARGBToYJRow_SSSE3(const uint8* src_argb, uint8* dst_y, int pix) { asm volatile ( "movdqa %3,%%xmm4 \n" "movdqa %4,%%xmm5 \n" @@ -799,7 +728,7 @@ void ARGBToYJRow_SSSE3(const uint8* src_argb, uint8* dst_y, int width) { "jg 1b \n" : "+r"(src_argb), // %0 "+r"(dst_y), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 : "m"(kARGBToYJ), // %3 "m"(kAddYJ64) // %4 : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" @@ -814,7 +743,7 @@ static const lvec32 kPermdARGBToY_AVX = { }; // Convert 32 ARGB pixels (128 bytes) to 32 Y values. -void ARGBToYRow_AVX2(const uint8* src_argb, uint8* dst_y, int width) { +void ARGBToYRow_AVX2(const uint8* src_argb, uint8* dst_y, int pix) { asm volatile ( "vbroadcastf128 %3,%%ymm4 \n" "vbroadcastf128 %4,%%ymm5 \n" @@ -844,7 +773,7 @@ void ARGBToYRow_AVX2(const uint8* src_argb, uint8* dst_y, int width) { "vzeroupper \n" : "+r"(src_argb), // %0 "+r"(dst_y), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 : "m"(kARGBToY), // %3 "m"(kAddY16), // %4 "m"(kPermdARGBToY_AVX) // %5 @@ -855,7 +784,7 @@ void ARGBToYRow_AVX2(const uint8* src_argb, uint8* dst_y, int width) { #ifdef HAS_ARGBTOYJROW_AVX2 // Convert 32 ARGB pixels (128 bytes) to 32 Y values. -void ARGBToYJRow_AVX2(const uint8* src_argb, uint8* dst_y, int width) { +void ARGBToYJRow_AVX2(const uint8* src_argb, uint8* dst_y, int pix) { asm volatile ( "vbroadcastf128 %3,%%ymm4 \n" "vbroadcastf128 %4,%%ymm5 \n" @@ -886,7 +815,7 @@ void ARGBToYJRow_AVX2(const uint8* src_argb, uint8* dst_y, int width) { "vzeroupper \n" : "+r"(src_argb), // %0 "+r"(dst_y), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 : "m"(kARGBToYJ), // %3 "m"(kAddYJ64), // %4 "m"(kPermdARGBToY_AVX) // %5 @@ -1023,67 +952,6 @@ void ARGBToUVRow_AVX2(const uint8* src_argb0, int src_stride_argb, } #endif // HAS_ARGBTOUVROW_AVX2 -#ifdef HAS_ARGBTOUVJROW_AVX2 -void ARGBToUVJRow_AVX2(const uint8* src_argb0, int src_stride_argb, - uint8* dst_u, uint8* dst_v, int width) { - asm volatile ( - "vbroadcastf128 %5,%%ymm5 \n" - "vbroadcastf128 %6,%%ymm6 \n" - "vbroadcastf128 %7,%%ymm7 \n" - "sub %1,%2 \n" - LABELALIGN - "1: \n" - "vmovdqu " MEMACCESS(0) ",%%ymm0 \n" - "vmovdqu " MEMACCESS2(0x20,0) ",%%ymm1 \n" - "vmovdqu " MEMACCESS2(0x40,0) ",%%ymm2 \n" - "vmovdqu " MEMACCESS2(0x60,0) ",%%ymm3 \n" - VMEMOPREG(vpavgb,0x00,0,4,1,ymm0,ymm0) // vpavgb (%0,%4,1),%%ymm0,%%ymm0 - VMEMOPREG(vpavgb,0x20,0,4,1,ymm1,ymm1) - VMEMOPREG(vpavgb,0x40,0,4,1,ymm2,ymm2) - VMEMOPREG(vpavgb,0x60,0,4,1,ymm3,ymm3) - "lea " MEMLEA(0x80,0) ",%0 \n" - "vshufps $0x88,%%ymm1,%%ymm0,%%ymm4 \n" - "vshufps $0xdd,%%ymm1,%%ymm0,%%ymm0 \n" - "vpavgb %%ymm4,%%ymm0,%%ymm0 \n" - "vshufps $0x88,%%ymm3,%%ymm2,%%ymm4 \n" - "vshufps $0xdd,%%ymm3,%%ymm2,%%ymm2 \n" - "vpavgb %%ymm4,%%ymm2,%%ymm2 \n" - - "vpmaddubsw %%ymm7,%%ymm0,%%ymm1 \n" - "vpmaddubsw %%ymm7,%%ymm2,%%ymm3 \n" - "vpmaddubsw %%ymm6,%%ymm0,%%ymm0 \n" - "vpmaddubsw %%ymm6,%%ymm2,%%ymm2 \n" - "vphaddw %%ymm3,%%ymm1,%%ymm1 \n" - "vphaddw %%ymm2,%%ymm0,%%ymm0 \n" - "vpaddw %%ymm5,%%ymm0,%%ymm0 \n" - "vpaddw %%ymm5,%%ymm1,%%ymm1 \n" - "vpsraw $0x8,%%ymm1,%%ymm1 \n" - "vpsraw $0x8,%%ymm0,%%ymm0 \n" - "vpacksswb %%ymm0,%%ymm1,%%ymm0 \n" - "vpermq $0xd8,%%ymm0,%%ymm0 \n" - "vpshufb %8,%%ymm0,%%ymm0 \n" - - "vextractf128 $0x0,%%ymm0," MEMACCESS(1) " \n" - VEXTOPMEM(vextractf128,1,ymm0,0x0,1,2,1) // vextractf128 $1,%%ymm0,(%1,%2,1) - "lea " MEMLEA(0x10,1) ",%1 \n" - "sub $0x20,%3 \n" - "jg 1b \n" - "vzeroupper \n" - : "+r"(src_argb0), // %0 - "+r"(dst_u), // %1 - "+r"(dst_v), // %2 - "+rm"(width) // %3 - : "r"((intptr_t)(src_stride_argb)), // %4 - "m"(kAddUVJ128), // %5 - "m"(kARGBToVJ), // %6 - "m"(kARGBToUJ), // %7 - "m"(kShufARGBToUV_AVX) // %8 - : "memory", "cc", NACL_R14 - "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7" - ); -} -#endif // HAS_ARGBTOUVJROW_AVX2 - #ifdef HAS_ARGBTOUVJROW_SSSE3 void ARGBToUVJRow_SSSE3(const uint8* src_argb0, int src_stride_argb, uint8* dst_u, uint8* dst_v, int width) { @@ -1205,7 +1073,60 @@ void ARGBToUV444Row_SSSE3(const uint8* src_argb, uint8* dst_u, uint8* dst_v, } #endif // HAS_ARGBTOUV444ROW_SSSE3 -void BGRAToYRow_SSSE3(const uint8* src_bgra, uint8* dst_y, int width) { +#ifdef HAS_ARGBTOUV422ROW_SSSE3 +void ARGBToUV422Row_SSSE3(const uint8* src_argb0, + uint8* dst_u, uint8* dst_v, int width) { + asm volatile ( + "movdqa %4,%%xmm3 \n" + "movdqa %5,%%xmm4 \n" + "movdqa %6,%%xmm5 \n" + "sub %1,%2 \n" + LABELALIGN + "1: \n" + "movdqu " MEMACCESS(0) ",%%xmm0 \n" + "movdqu " MEMACCESS2(0x10,0) ",%%xmm1 \n" + "movdqu " MEMACCESS2(0x20,0) ",%%xmm2 \n" + "movdqu " MEMACCESS2(0x30,0) ",%%xmm6 \n" + "lea " MEMLEA(0x40,0) ",%0 \n" + "movdqa %%xmm0,%%xmm7 \n" + "shufps $0x88,%%xmm1,%%xmm0 \n" + "shufps $0xdd,%%xmm1,%%xmm7 \n" + "pavgb %%xmm7,%%xmm0 \n" + "movdqa %%xmm2,%%xmm7 \n" + "shufps $0x88,%%xmm6,%%xmm2 \n" + "shufps $0xdd,%%xmm6,%%xmm7 \n" + "pavgb %%xmm7,%%xmm2 \n" + "movdqa %%xmm0,%%xmm1 \n" + "movdqa %%xmm2,%%xmm6 \n" + "pmaddubsw %%xmm4,%%xmm0 \n" + "pmaddubsw %%xmm4,%%xmm2 \n" + "pmaddubsw %%xmm3,%%xmm1 \n" + "pmaddubsw %%xmm3,%%xmm6 \n" + "phaddw %%xmm2,%%xmm0 \n" + "phaddw %%xmm6,%%xmm1 \n" + "psraw $0x8,%%xmm0 \n" + "psraw $0x8,%%xmm1 \n" + "packsswb %%xmm1,%%xmm0 \n" + "paddb %%xmm5,%%xmm0 \n" + "movlps %%xmm0," MEMACCESS(1) " \n" + MEMOPMEM(movhps,xmm0,0x00,1,2,1) // movhps %%xmm0,(%1,%2,1) + "lea " MEMLEA(0x8,1) ",%1 \n" + "sub $0x10,%3 \n" + "jg 1b \n" + : "+r"(src_argb0), // %0 + "+r"(dst_u), // %1 + "+r"(dst_v), // %2 + "+rm"(width) // %3 + : "m"(kARGBToV), // %4 + "m"(kARGBToU), // %5 + "m"(kAddUV128) // %6 + : "memory", "cc", NACL_R14 + "xmm0", "xmm1", "xmm2", "xmm6", "xmm7" + ); +} +#endif // HAS_ARGBTOUV422ROW_SSSE3 + +void BGRAToYRow_SSSE3(const uint8* src_bgra, uint8* dst_y, int pix) { asm volatile ( "movdqa %4,%%xmm5 \n" "movdqa %3,%%xmm4 \n" @@ -1232,7 +1153,7 @@ void BGRAToYRow_SSSE3(const uint8* src_bgra, uint8* dst_y, int width) { "jg 1b \n" : "+r"(src_bgra), // %0 "+r"(dst_y), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 : "m"(kBGRAToY), // %3 "m"(kAddY16) // %4 : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" @@ -1300,7 +1221,7 @@ void BGRAToUVRow_SSSE3(const uint8* src_bgra0, int src_stride_bgra, ); } -void ABGRToYRow_SSSE3(const uint8* src_abgr, uint8* dst_y, int width) { +void ABGRToYRow_SSSE3(const uint8* src_abgr, uint8* dst_y, int pix) { asm volatile ( "movdqa %4,%%xmm5 \n" "movdqa %3,%%xmm4 \n" @@ -1327,14 +1248,14 @@ void ABGRToYRow_SSSE3(const uint8* src_abgr, uint8* dst_y, int width) { "jg 1b \n" : "+r"(src_abgr), // %0 "+r"(dst_y), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 : "m"(kABGRToY), // %3 "m"(kAddY16) // %4 : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" ); } -void RGBAToYRow_SSSE3(const uint8* src_rgba, uint8* dst_y, int width) { +void RGBAToYRow_SSSE3(const uint8* src_rgba, uint8* dst_y, int pix) { asm volatile ( "movdqa %4,%%xmm5 \n" "movdqa %3,%%xmm4 \n" @@ -1361,7 +1282,7 @@ void RGBAToYRow_SSSE3(const uint8* src_rgba, uint8* dst_y, int width) { "jg 1b \n" : "+r"(src_rgba), // %0 "+r"(dst_y), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 : "m"(kRGBAToY), // %3 "m"(kAddY16) // %4 : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" @@ -1492,15 +1413,132 @@ void RGBAToUVRow_SSSE3(const uint8* src_rgba0, int src_stride_rgba, #if defined(HAS_I422TOARGBROW_SSSE3) || defined(HAS_I422TOARGBROW_AVX2) -// Read 8 UV from 444 +struct YuvConstants { + lvec8 kUVToB; // 0 + lvec8 kUVToG; // 32 + lvec8 kUVToR; // 64 + lvec16 kUVBiasB; // 96 + lvec16 kUVBiasG; // 128 + lvec16 kUVBiasR; // 160 + lvec16 kYToRgb; // 192 +}; + +// BT.601 YUV to RGB reference +// R = (Y - 16) * 1.164 - V * -1.596 +// G = (Y - 16) * 1.164 - U * 0.391 - V * 0.813 +// B = (Y - 16) * 1.164 - U * -2.018 + +// Y contribution to R,G,B. Scale and bias. +// TODO(fbarchard): Consider moving constants into a common header. +#define YG 18997 /* round(1.164 * 64 * 256 * 256 / 257) */ +#define YGB -1160 /* 1.164 * 64 * -16 + 64 / 2 */ + +// U and V contributions to R,G,B. +#define UB -128 /* max(-128, round(-2.018 * 64)) */ +#define UG 25 /* round(0.391 * 64) */ +#define VG 52 /* round(0.813 * 64) */ +#define VR -102 /* round(-1.596 * 64) */ + +// Bias values to subtract 16 from Y and 128 from U and V. +#define BB (UB * 128 + YGB) +#define BG (UG * 128 + VG * 128 + YGB) +#define BR (VR * 128 + YGB) + +// BT601 constants for YUV to RGB. +static YuvConstants SIMD_ALIGNED(kYuvConstants) = { + { UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, + UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0 }, + { UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, + UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG }, + { 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, + 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR }, + { BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB }, + { BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG }, + { BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR }, + { YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG } +}; + +// BT601 constants for NV21 where chroma plane is VU instead of UV. +static YuvConstants SIMD_ALIGNED(kYvuConstants) = { + { 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, + 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB }, + { VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, + VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG }, + { VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, + VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0 }, + { BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB }, + { BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG }, + { BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR }, + { YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG } +}; + +#undef YG +#undef YGB +#undef UB +#undef UG +#undef VG +#undef VR +#undef BB +#undef BG +#undef BR + +// JPEG YUV to RGB reference +// * R = Y - V * -1.40200 +// * G = Y - U * 0.34414 - V * 0.71414 +// * B = Y - U * -1.77200 + +// Y contribution to R,G,B. Scale and bias. +// TODO(fbarchard): Consider moving constants into a common header. +#define YGJ 16320 /* round(1.000 * 64 * 256 * 256 / 257) */ +#define YGBJ 32 /* 64 / 2 */ + +// U and V contributions to R,G,B. +#define UBJ -113 /* round(-1.77200 * 64) */ +#define UGJ 22 /* round(0.34414 * 64) */ +#define VGJ 46 /* round(0.71414 * 64) */ +#define VRJ -90 /* round(-1.40200 * 64) */ + +// Bias values to subtract 16 from Y and 128 from U and V. +#define BBJ (UBJ * 128 + YGBJ) +#define BGJ (UGJ * 128 + VGJ * 128 + YGBJ) +#define BRJ (VRJ * 128 + YGBJ) + +// JPEG constants for YUV to RGB. +YuvConstants SIMD_ALIGNED(kYuvJConstants) = { + { UBJ, 0, UBJ, 0, UBJ, 0, UBJ, 0, UBJ, 0, UBJ, 0, UBJ, 0, UBJ, 0, + UBJ, 0, UBJ, 0, UBJ, 0, UBJ, 0, UBJ, 0, UBJ, 0, UBJ, 0, UBJ, 0 }, + { UGJ, VGJ, UGJ, VGJ, UGJ, VGJ, UGJ, VGJ, + UGJ, VGJ, UGJ, VGJ, UGJ, VGJ, UGJ, VGJ, + UGJ, VGJ, UGJ, VGJ, UGJ, VGJ, UGJ, VGJ, + UGJ, VGJ, UGJ, VGJ, UGJ, VGJ, UGJ, VGJ }, + { 0, VRJ, 0, VRJ, 0, VRJ, 0, VRJ, 0, VRJ, 0, VRJ, 0, VRJ, 0, VRJ, + 0, VRJ, 0, VRJ, 0, VRJ, 0, VRJ, 0, VRJ, 0, VRJ, 0, VRJ, 0, VRJ }, + { BBJ, BBJ, BBJ, BBJ, BBJ, BBJ, BBJ, BBJ, + BBJ, BBJ, BBJ, BBJ, BBJ, BBJ, BBJ, BBJ }, + { BGJ, BGJ, BGJ, BGJ, BGJ, BGJ, BGJ, BGJ, + BGJ, BGJ, BGJ, BGJ, BGJ, BGJ, BGJ, BGJ }, + { BRJ, BRJ, BRJ, BRJ, BRJ, BRJ, BRJ, BRJ, + BRJ, BRJ, BRJ, BRJ, BRJ, BRJ, BRJ, BRJ }, + { YGJ, YGJ, YGJ, YGJ, YGJ, YGJ, YGJ, YGJ, + YGJ, YGJ, YGJ, YGJ, YGJ, YGJ, YGJ, YGJ } +}; + +#undef YGJ +#undef YGBJ +#undef UBJ +#undef UGJ +#undef VGJ +#undef VRJ +#undef BBJ +#undef BGJ +#undef BRJ + +// Read 8 UV from 411 #define READYUV444 \ "movq " MEMACCESS([u_buf]) ",%%xmm0 \n" \ MEMOPREG(movq, 0x00, [u_buf], [v_buf], 1, xmm1) \ "lea " MEMLEA(0x8, [u_buf]) ",%[u_buf] \n" \ - "punpcklbw %%xmm1,%%xmm0 \n" \ - "movq " MEMACCESS([y_buf]) ",%%xmm4 \n" \ - "punpcklbw %%xmm4,%%xmm4 \n" \ - "lea " MEMLEA(0x8, [y_buf]) ",%[y_buf] \n" + "punpcklbw %%xmm1,%%xmm0 \n" // Read 4 UV from 422, upsample to 8 UV #define READYUV422 \ @@ -1508,144 +1546,52 @@ void RGBAToUVRow_SSSE3(const uint8* src_rgba0, int src_stride_rgba, MEMOPREG(movd, 0x00, [u_buf], [v_buf], 1, xmm1) \ "lea " MEMLEA(0x4, [u_buf]) ",%[u_buf] \n" \ "punpcklbw %%xmm1,%%xmm0 \n" \ - "punpcklwd %%xmm0,%%xmm0 \n" \ - "movq " MEMACCESS([y_buf]) ",%%xmm4 \n" \ - "punpcklbw %%xmm4,%%xmm4 \n" \ - "lea " MEMLEA(0x8, [y_buf]) ",%[y_buf] \n" + "punpcklwd %%xmm0,%%xmm0 \n" -// Read 4 UV from 422, upsample to 8 UV. With 8 Alpha. -#define READYUVA422 \ +// Read 2 UV from 411, upsample to 8 UV +#define READYUV411 \ "movd " MEMACCESS([u_buf]) ",%%xmm0 \n" \ MEMOPREG(movd, 0x00, [u_buf], [v_buf], 1, xmm1) \ - "lea " MEMLEA(0x4, [u_buf]) ",%[u_buf] \n" \ - "punpcklbw %%xmm1,%%xmm0 \n" \ - "punpcklwd %%xmm0,%%xmm0 \n" \ - "movq " MEMACCESS([y_buf]) ",%%xmm4 \n" \ - "punpcklbw %%xmm4,%%xmm4 \n" \ - "lea " MEMLEA(0x8, [y_buf]) ",%[y_buf] \n" \ - "movq " MEMACCESS([a_buf]) ",%%xmm5 \n" \ - "lea " MEMLEA(0x8, [a_buf]) ",%[a_buf] \n" - -// Read 2 UV from 411, upsample to 8 UV. -// reading 4 bytes is an msan violation. -// "movd " MEMACCESS([u_buf]) ",%%xmm0 \n" -// MEMOPREG(movd, 0x00, [u_buf], [v_buf], 1, xmm1) -// pinsrw fails with drmemory -// __asm pinsrw xmm0, [esi], 0 /* U */ -// __asm pinsrw xmm1, [esi + edi], 0 /* V */ -#define READYUV411_TEMP \ - "movzwl " MEMACCESS([u_buf]) ",%[temp] \n" \ - "movd %[temp],%%xmm0 \n" \ - MEMOPARG(movzwl, 0x00, [u_buf], [v_buf], 1, [temp]) " \n" \ - "movd %[temp],%%xmm1 \n" \ "lea " MEMLEA(0x2, [u_buf]) ",%[u_buf] \n" \ "punpcklbw %%xmm1,%%xmm0 \n" \ "punpcklwd %%xmm0,%%xmm0 \n" \ - "punpckldq %%xmm0,%%xmm0 \n" \ - "movq " MEMACCESS([y_buf]) ",%%xmm4 \n" \ - "punpcklbw %%xmm4,%%xmm4 \n" \ - "lea " MEMLEA(0x8, [y_buf]) ",%[y_buf] \n" + "punpckldq %%xmm0,%%xmm0 \n" // Read 4 UV from NV12, upsample to 8 UV #define READNV12 \ "movq " MEMACCESS([uv_buf]) ",%%xmm0 \n" \ "lea " MEMLEA(0x8, [uv_buf]) ",%[uv_buf] \n" \ - "punpcklwd %%xmm0,%%xmm0 \n" \ - "movq " MEMACCESS([y_buf]) ",%%xmm4 \n" \ - "punpcklbw %%xmm4,%%xmm4 \n" \ - "lea " MEMLEA(0x8, [y_buf]) ",%[y_buf] \n" - -// Read 4 VU from NV21, upsample to 8 UV -#define READNV21 \ - "movq " MEMACCESS([vu_buf]) ",%%xmm0 \n" \ - "lea " MEMLEA(0x8, [vu_buf]) ",%[vu_buf] \n" \ - "pshufb %[kShuffleNV21], %%xmm0 \n" \ - "movq " MEMACCESS([y_buf]) ",%%xmm4 \n" \ - "punpcklbw %%xmm4,%%xmm4 \n" \ - "lea " MEMLEA(0x8, [y_buf]) ",%[y_buf] \n" - -// Read 4 YUY2 with 8 Y and update 4 UV to 8 UV. -#define READYUY2 \ - "movdqu " MEMACCESS([yuy2_buf]) ",%%xmm4 \n" \ - "pshufb %[kShuffleYUY2Y], %%xmm4 \n" \ - "movdqu " MEMACCESS([yuy2_buf]) ",%%xmm0 \n" \ - "pshufb %[kShuffleYUY2UV], %%xmm0 \n" \ - "lea " MEMLEA(0x10, [yuy2_buf]) ",%[yuy2_buf] \n" - -// Read 4 UYVY with 8 Y and update 4 UV to 8 UV. -#define READUYVY \ - "movdqu " MEMACCESS([uyvy_buf]) ",%%xmm4 \n" \ - "pshufb %[kShuffleUYVYY], %%xmm4 \n" \ - "movdqu " MEMACCESS([uyvy_buf]) ",%%xmm0 \n" \ - "pshufb %[kShuffleUYVYUV], %%xmm0 \n" \ - "lea " MEMLEA(0x10, [uyvy_buf]) ",%[uyvy_buf] \n" - -#if defined(__x86_64__) -#define YUVTORGB_SETUP(yuvconstants) \ - "movdqa " MEMACCESS([yuvconstants]) ",%%xmm8 \n" \ - "movdqa " MEMACCESS2(32, [yuvconstants]) ",%%xmm9 \n" \ - "movdqa " MEMACCESS2(64, [yuvconstants]) ",%%xmm10 \n" \ - "movdqa " MEMACCESS2(96, [yuvconstants]) ",%%xmm11 \n" \ - "movdqa " MEMACCESS2(128, [yuvconstants]) ",%%xmm12 \n" \ - "movdqa " MEMACCESS2(160, [yuvconstants]) ",%%xmm13 \n" \ - "movdqa " MEMACCESS2(192, [yuvconstants]) ",%%xmm14 \n" -// Convert 8 pixels: 8 UV and 8 Y -#define YUVTORGB(yuvconstants) \ - "movdqa %%xmm0,%%xmm1 \n" \ - "movdqa %%xmm0,%%xmm2 \n" \ - "movdqa %%xmm0,%%xmm3 \n" \ - "movdqa %%xmm11,%%xmm0 \n" \ - "pmaddubsw %%xmm8,%%xmm1 \n" \ - "psubw %%xmm1,%%xmm0 \n" \ - "movdqa %%xmm12,%%xmm1 \n" \ - "pmaddubsw %%xmm9,%%xmm2 \n" \ - "psubw %%xmm2,%%xmm1 \n" \ - "movdqa %%xmm13,%%xmm2 \n" \ - "pmaddubsw %%xmm10,%%xmm3 \n" \ - "psubw %%xmm3,%%xmm2 \n" \ - "pmulhuw %%xmm14,%%xmm4 \n" \ - "paddsw %%xmm4,%%xmm0 \n" \ - "paddsw %%xmm4,%%xmm1 \n" \ - "paddsw %%xmm4,%%xmm2 \n" \ - "psraw $0x6,%%xmm0 \n" \ - "psraw $0x6,%%xmm1 \n" \ - "psraw $0x6,%%xmm2 \n" \ - "packuswb %%xmm0,%%xmm0 \n" \ - "packuswb %%xmm1,%%xmm1 \n" \ - "packuswb %%xmm2,%%xmm2 \n" -#define YUVTORGB_REGS \ - "xmm8", "xmm9", "xmm10", "xmm11", "xmm12", "xmm13", "xmm14", + "punpcklwd %%xmm0,%%xmm0 \n" -#else -#define YUVTORGB_SETUP(yuvconstants) // Convert 8 pixels: 8 UV and 8 Y -#define YUVTORGB(yuvconstants) \ +#define YUVTORGB(YuvConstants) \ "movdqa %%xmm0,%%xmm1 \n" \ "movdqa %%xmm0,%%xmm2 \n" \ "movdqa %%xmm0,%%xmm3 \n" \ - "movdqa " MEMACCESS2(96, [yuvconstants]) ",%%xmm0 \n" \ - "pmaddubsw " MEMACCESS([yuvconstants]) ",%%xmm1 \n" \ + "movdqa " MEMACCESS2(96, [YuvConstants]) ",%%xmm0 \n" \ + "pmaddubsw " MEMACCESS([YuvConstants]) ",%%xmm1 \n" \ "psubw %%xmm1,%%xmm0 \n" \ - "movdqa " MEMACCESS2(128, [yuvconstants]) ",%%xmm1 \n" \ - "pmaddubsw " MEMACCESS2(32, [yuvconstants]) ",%%xmm2 \n" \ + "movdqa " MEMACCESS2(128, [YuvConstants]) ",%%xmm1 \n" \ + "pmaddubsw " MEMACCESS2(32, [YuvConstants]) ",%%xmm2 \n" \ "psubw %%xmm2,%%xmm1 \n" \ - "movdqa " MEMACCESS2(160, [yuvconstants]) ",%%xmm2 \n" \ - "pmaddubsw " MEMACCESS2(64, [yuvconstants]) ",%%xmm3 \n" \ + "movdqa " MEMACCESS2(160, [YuvConstants]) ",%%xmm2 \n" \ + "pmaddubsw " MEMACCESS2(64, [YuvConstants]) ",%%xmm3 \n" \ "psubw %%xmm3,%%xmm2 \n" \ - "pmulhuw " MEMACCESS2(192, [yuvconstants]) ",%%xmm4 \n" \ - "paddsw %%xmm4,%%xmm0 \n" \ - "paddsw %%xmm4,%%xmm1 \n" \ - "paddsw %%xmm4,%%xmm2 \n" \ + "movq " MEMACCESS([y_buf]) ",%%xmm3 \n" \ + "lea " MEMLEA(0x8, [y_buf]) ",%[y_buf] \n" \ + "punpcklbw %%xmm3,%%xmm3 \n" \ + "pmulhuw " MEMACCESS2(192, [YuvConstants]) ",%%xmm3 \n" \ + "paddsw %%xmm3,%%xmm0 \n" \ + "paddsw %%xmm3,%%xmm1 \n" \ + "paddsw %%xmm3,%%xmm2 \n" \ "psraw $0x6,%%xmm0 \n" \ "psraw $0x6,%%xmm1 \n" \ "psraw $0x6,%%xmm2 \n" \ "packuswb %%xmm0,%%xmm0 \n" \ "packuswb %%xmm1,%%xmm1 \n" \ "packuswb %%xmm2,%%xmm2 \n" -#define YUVTORGB_REGS -#endif -// Store 8 ARGB values. +// Store 8 ARGB values. Assumes XMM5 is zero. #define STOREARGB \ "punpcklbw %%xmm1,%%xmm0 \n" \ "punpcklbw %%xmm5,%%xmm2 \n" \ @@ -1656,7 +1602,30 @@ void RGBAToUVRow_SSSE3(const uint8* src_rgba0, int src_stride_rgba, "movdqu %%xmm1," MEMACCESS2(0x10, [dst_argb]) " \n" \ "lea " MEMLEA(0x20, [dst_argb]) ", %[dst_argb] \n" -// Store 8 RGBA values. +// Store 8 BGRA values. Assumes XMM5 is zero. +#define STOREBGRA \ + "pcmpeqb %%xmm5,%%xmm5 \n" \ + "punpcklbw %%xmm0,%%xmm1 \n" \ + "punpcklbw %%xmm2,%%xmm5 \n" \ + "movdqa %%xmm5,%%xmm0 \n" \ + "punpcklwd %%xmm1,%%xmm5 \n" \ + "punpckhwd %%xmm1,%%xmm0 \n" \ + "movdqu %%xmm5," MEMACCESS([dst_bgra]) " \n" \ + "movdqu %%xmm0," MEMACCESS2(0x10, [dst_bgra]) " \n" \ + "lea " MEMLEA(0x20, [dst_bgra]) ", %[dst_bgra] \n" + +// Store 8 ABGR values. Assumes XMM5 is zero. +#define STOREABGR \ + "punpcklbw %%xmm1,%%xmm2 \n" \ + "punpcklbw %%xmm5,%%xmm0 \n" \ + "movdqa %%xmm2,%%xmm1 \n" \ + "punpcklwd %%xmm0,%%xmm2 \n" \ + "punpckhwd %%xmm0,%%xmm1 \n" \ + "movdqu %%xmm2," MEMACCESS([dst_abgr]) " \n" \ + "movdqu %%xmm1," MEMACCESS2(0x10, [dst_abgr]) " \n" \ + "lea " MEMLEA(0x20, [dst_abgr]) ", %[dst_abgr] \n" + +// Store 8 RGBA values. Assumes XMM5 is zero. #define STORERGBA \ "pcmpeqb %%xmm5,%%xmm5 \n" \ "punpcklbw %%xmm2,%%xmm1 \n" \ @@ -1672,16 +1641,14 @@ void OMITFP I444ToARGBRow_SSSE3(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, uint8* dst_argb, - const struct YuvConstants* yuvconstants, int width) { asm volatile ( - YUVTORGB_SETUP(yuvconstants) "sub %[u_buf],%[v_buf] \n" "pcmpeqb %%xmm5,%%xmm5 \n" LABELALIGN "1: \n" READYUV444 - YUVTORGB(yuvconstants) + YUVTORGB(kYuvConstants) STOREARGB "sub $0x8,%[width] \n" "jg 1b \n" @@ -1690,27 +1657,26 @@ void OMITFP I444ToARGBRow_SSSE3(const uint8* y_buf, [v_buf]"+r"(v_buf), // %[v_buf] [dst_argb]"+r"(dst_argb), // %[dst_argb] [width]"+rm"(width) // %[width] - : [yuvconstants]"r"(yuvconstants) // %[yuvconstants] - : "memory", "cc", NACL_R14 YUVTORGB_REGS - "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" + : [kYuvConstants]"r"(&kYuvConstants.kUVToB) // %[kYuvConstants] + : "memory", "cc", NACL_R14 + "xmm0", "xmm1", "xmm2", "xmm3", "xmm5" ); } +// TODO(fbarchard): Consider putting masks into constants. void OMITFP I422ToRGB24Row_SSSE3(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, uint8* dst_rgb24, - const struct YuvConstants* yuvconstants, int width) { asm volatile ( - YUVTORGB_SETUP(yuvconstants) "movdqa %[kShuffleMaskARGBToRGB24_0],%%xmm5 \n" "movdqa %[kShuffleMaskARGBToRGB24],%%xmm6 \n" "sub %[u_buf],%[v_buf] \n" LABELALIGN "1: \n" READYUV422 - YUVTORGB(yuvconstants) + YUVTORGB(kYuvConstants) "punpcklbw %%xmm1,%%xmm0 \n" "punpcklbw %%xmm2,%%xmm2 \n" "movdqa %%xmm0,%%xmm1 \n" @@ -1728,16 +1694,61 @@ void OMITFP I422ToRGB24Row_SSSE3(const uint8* y_buf, [u_buf]"+r"(u_buf), // %[u_buf] [v_buf]"+r"(v_buf), // %[v_buf] [dst_rgb24]"+r"(dst_rgb24), // %[dst_rgb24] +// TODO(fbarchard): Make width a register for 32 bit. #if defined(__i386__) && defined(__pic__) [width]"+m"(width) // %[width] #else [width]"+rm"(width) // %[width] #endif - : [yuvconstants]"r"(yuvconstants), // %[yuvconstants] + : [kYuvConstants]"r"(&kYuvConstants.kUVToB), [kShuffleMaskARGBToRGB24_0]"m"(kShuffleMaskARGBToRGB24_0), [kShuffleMaskARGBToRGB24]"m"(kShuffleMaskARGBToRGB24) - : "memory", "cc", NACL_R14 YUVTORGB_REGS - "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6" + : "memory", "cc", NACL_R14 + "xmm0", "xmm1", "xmm2", "xmm3", "xmm5", "xmm6" + ); +} + +void OMITFP I422ToRAWRow_SSSE3(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* dst_raw, + int width) { + asm volatile ( + "movdqa %[kShuffleMaskARGBToRAW_0],%%xmm5 \n" + "movdqa %[kShuffleMaskARGBToRAW],%%xmm6 \n" + "sub %[u_buf],%[v_buf] \n" + LABELALIGN + "1: \n" + READYUV422 + YUVTORGB(kYuvConstants) + "punpcklbw %%xmm1,%%xmm0 \n" + "punpcklbw %%xmm2,%%xmm2 \n" + "movdqa %%xmm0,%%xmm1 \n" + "punpcklwd %%xmm2,%%xmm0 \n" + "punpckhwd %%xmm2,%%xmm1 \n" + "pshufb %%xmm5,%%xmm0 \n" + "pshufb %%xmm6,%%xmm1 \n" + "palignr $0xc,%%xmm0,%%xmm1 \n" + "movq %%xmm0," MEMACCESS([dst_raw]) " \n" + "movdqu %%xmm1," MEMACCESS2(0x8,[dst_raw]) "\n" + "lea " MEMLEA(0x18,[dst_raw]) ",%[dst_raw] \n" + "subl $0x8,%[width] \n" + "jg 1b \n" + : [y_buf]"+r"(y_buf), // %[y_buf] + [u_buf]"+r"(u_buf), // %[u_buf] + [v_buf]"+r"(v_buf), // %[v_buf] + [dst_raw]"+r"(dst_raw), // %[dst_raw] +// TODO(fbarchard): Make width a register for 32 bit. +#if defined(__i386__) && defined(__pic__) + [width]"+m"(width) // %[width] +#else + [width]"+rm"(width) // %[width] +#endif + : [kYuvConstants]"r"(&kYuvConstants.kUVToB), + [kShuffleMaskARGBToRAW_0]"m"(kShuffleMaskARGBToRAW_0), + [kShuffleMaskARGBToRAW]"m"(kShuffleMaskARGBToRAW) + : "memory", "cc", NACL_R14 + "xmm0", "xmm1", "xmm2", "xmm3", "xmm5", "xmm6" ); } @@ -1745,16 +1756,14 @@ void OMITFP I422ToARGBRow_SSSE3(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, uint8* dst_argb, - const struct YuvConstants* yuvconstants, int width) { asm volatile ( - YUVTORGB_SETUP(yuvconstants) "sub %[u_buf],%[v_buf] \n" "pcmpeqb %%xmm5,%%xmm5 \n" LABELALIGN "1: \n" READYUV422 - YUVTORGB(yuvconstants) + YUVTORGB(kYuvConstants) STOREARGB "sub $0x8,%[width] \n" "jg 1b \n" @@ -1763,95 +1772,74 @@ void OMITFP I422ToARGBRow_SSSE3(const uint8* y_buf, [v_buf]"+r"(v_buf), // %[v_buf] [dst_argb]"+r"(dst_argb), // %[dst_argb] [width]"+rm"(width) // %[width] - : [yuvconstants]"r"(yuvconstants) // %[yuvconstants] - : "memory", "cc", NACL_R14 YUVTORGB_REGS - "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" + : [kYuvConstants]"r"(&kYuvConstants.kUVToB) // %[kYuvConstants] + : "memory", "cc", NACL_R14 + "xmm0", "xmm1", "xmm2", "xmm3", "xmm5" ); } -#ifdef HAS_I422ALPHATOARGBROW_SSSE3 -void OMITFP I422AlphaToARGBRow_SSSE3(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - const uint8* a_buf, - uint8* dst_argb, - const struct YuvConstants* yuvconstants, - int width) { +void OMITFP J422ToARGBRow_SSSE3(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* dst_argb, + int width) { asm volatile ( - YUVTORGB_SETUP(yuvconstants) "sub %[u_buf],%[v_buf] \n" + "pcmpeqb %%xmm5,%%xmm5 \n" LABELALIGN "1: \n" - READYUVA422 - YUVTORGB(yuvconstants) + READYUV422 + YUVTORGB(kYuvConstants) STOREARGB - "subl $0x8,%[width] \n" + "sub $0x8,%[width] \n" "jg 1b \n" : [y_buf]"+r"(y_buf), // %[y_buf] [u_buf]"+r"(u_buf), // %[u_buf] [v_buf]"+r"(v_buf), // %[v_buf] - [a_buf]"+r"(a_buf), // %[a_buf] [dst_argb]"+r"(dst_argb), // %[dst_argb] -#if defined(__i386__) && defined(__pic__) - [width]"+m"(width) // %[width] -#else [width]"+rm"(width) // %[width] -#endif - : [yuvconstants]"r"(yuvconstants) // %[yuvconstants] - : "memory", "cc", NACL_R14 YUVTORGB_REGS - "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" + : [kYuvConstants]"r"(&kYuvJConstants.kUVToB) // %[kYuvConstants] + : "memory", "cc", NACL_R14 + "xmm0", "xmm1", "xmm2", "xmm3", "xmm5" ); } -#endif // HAS_I422ALPHATOARGBROW_SSSE3 -#ifdef HAS_I411TOARGBROW_SSSE3 void OMITFP I411ToARGBRow_SSSE3(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, uint8* dst_argb, - const struct YuvConstants* yuvconstants, int width) { - int temp; asm volatile ( - YUVTORGB_SETUP(yuvconstants) "sub %[u_buf],%[v_buf] \n" "pcmpeqb %%xmm5,%%xmm5 \n" LABELALIGN "1: \n" - READYUV411_TEMP - YUVTORGB(yuvconstants) + READYUV411 + YUVTORGB(kYuvConstants) STOREARGB - "subl $0x8,%[width] \n" + "sub $0x8,%[width] \n" "jg 1b \n" - : [y_buf]"+r"(y_buf), // %[y_buf] - [u_buf]"+r"(u_buf), // %[u_buf] - [v_buf]"+r"(v_buf), // %[v_buf] + : [y_buf]"+r"(y_buf), // %[y_buf] + [u_buf]"+r"(u_buf), // %[u_buf] + [v_buf]"+r"(v_buf), // %[v_buf] [dst_argb]"+r"(dst_argb), // %[dst_argb] - [temp]"=&r"(temp), // %[temp] -#if defined(__i386__) && defined(__pic__) - [width]"+m"(width) // %[width] -#else - [width]"+rm"(width) // %[width] -#endif - : [yuvconstants]"r"(yuvconstants) // %[yuvconstants] - : "memory", "cc", NACL_R14 YUVTORGB_REGS - "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" + [width]"+rm"(width) // %[width] + : [kYuvConstants]"r"(&kYuvConstants.kUVToB) // %[kYuvConstants] + : "memory", "cc", NACL_R14 + "xmm0", "xmm1", "xmm2", "xmm3", "xmm5" ); } -#endif void OMITFP NV12ToARGBRow_SSSE3(const uint8* y_buf, const uint8* uv_buf, uint8* dst_argb, - const struct YuvConstants* yuvconstants, int width) { asm volatile ( - YUVTORGB_SETUP(yuvconstants) "pcmpeqb %%xmm5,%%xmm5 \n" LABELALIGN "1: \n" READNV12 - YUVTORGB(yuvconstants) + YUVTORGB(kYuvConstants) STOREARGB "sub $0x8,%[width] \n" "jg 1b \n" @@ -1859,85 +1847,84 @@ void OMITFP NV12ToARGBRow_SSSE3(const uint8* y_buf, [uv_buf]"+r"(uv_buf), // %[uv_buf] [dst_argb]"+r"(dst_argb), // %[dst_argb] [width]"+rm"(width) // %[width] - : [yuvconstants]"r"(yuvconstants) // %[yuvconstants] - : "memory", "cc", YUVTORGB_REGS // Does not use r14. - "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" + : [kYuvConstants]"r"(&kYuvConstants.kUVToB) // %[kYuvConstants] + // Does not use r14. + : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm5" ); } void OMITFP NV21ToARGBRow_SSSE3(const uint8* y_buf, - const uint8* vu_buf, + const uint8* uv_buf, uint8* dst_argb, - const struct YuvConstants* yuvconstants, int width) { asm volatile ( - YUVTORGB_SETUP(yuvconstants) "pcmpeqb %%xmm5,%%xmm5 \n" LABELALIGN "1: \n" - READNV21 - YUVTORGB(yuvconstants) + READNV12 + YUVTORGB(kYuvConstants) STOREARGB "sub $0x8,%[width] \n" "jg 1b \n" : [y_buf]"+r"(y_buf), // %[y_buf] - [vu_buf]"+r"(vu_buf), // %[vu_buf] + [uv_buf]"+r"(uv_buf), // %[uv_buf] [dst_argb]"+r"(dst_argb), // %[dst_argb] [width]"+rm"(width) // %[width] - : [yuvconstants]"r"(yuvconstants), // %[yuvconstants] - [kShuffleNV21]"m"(kShuffleNV21) - : "memory", "cc", YUVTORGB_REGS // Does not use r14. - "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" + : [kYuvConstants]"r"(&kYvuConstants.kUVToB) // %[kYuvConstants] + // Does not use r14. + : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm5" ); } -void OMITFP YUY2ToARGBRow_SSSE3(const uint8* yuy2_buf, - uint8* dst_argb, - const struct YuvConstants* yuvconstants, +void OMITFP I422ToBGRARow_SSSE3(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* dst_bgra, int width) { asm volatile ( - YUVTORGB_SETUP(yuvconstants) + "sub %[u_buf],%[v_buf] \n" "pcmpeqb %%xmm5,%%xmm5 \n" LABELALIGN "1: \n" - READYUY2 - YUVTORGB(yuvconstants) - STOREARGB + READYUV422 + YUVTORGB(kYuvConstants) + STOREBGRA "sub $0x8,%[width] \n" "jg 1b \n" - : [yuy2_buf]"+r"(yuy2_buf), // %[yuy2_buf] - [dst_argb]"+r"(dst_argb), // %[dst_argb] + : [y_buf]"+r"(y_buf), // %[y_buf] + [u_buf]"+r"(u_buf), // %[u_buf] + [v_buf]"+r"(v_buf), // %[v_buf] + [dst_bgra]"+r"(dst_bgra), // %[dst_bgra] [width]"+rm"(width) // %[width] - : [yuvconstants]"r"(yuvconstants), // %[yuvconstants] - [kShuffleYUY2Y]"m"(kShuffleYUY2Y), - [kShuffleYUY2UV]"m"(kShuffleYUY2UV) - : "memory", "cc", YUVTORGB_REGS // Does not use r14. - "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" + : [kYuvConstants]"r"(&kYuvConstants.kUVToB) // %[kYuvConstants] + : "memory", "cc", NACL_R14 + "xmm0", "xmm1", "xmm2", "xmm3", "xmm5" ); } -void OMITFP UYVYToARGBRow_SSSE3(const uint8* uyvy_buf, - uint8* dst_argb, - const struct YuvConstants* yuvconstants, +void OMITFP I422ToABGRRow_SSSE3(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* dst_abgr, int width) { asm volatile ( - YUVTORGB_SETUP(yuvconstants) + "sub %[u_buf],%[v_buf] \n" "pcmpeqb %%xmm5,%%xmm5 \n" LABELALIGN "1: \n" - READUYVY - YUVTORGB(yuvconstants) - STOREARGB + READYUV422 + YUVTORGB(kYuvConstants) + STOREABGR "sub $0x8,%[width] \n" "jg 1b \n" - : [uyvy_buf]"+r"(uyvy_buf), // %[uyvy_buf] - [dst_argb]"+r"(dst_argb), // %[dst_argb] + : [y_buf]"+r"(y_buf), // %[y_buf] + [u_buf]"+r"(u_buf), // %[u_buf] + [v_buf]"+r"(v_buf), // %[v_buf] + [dst_abgr]"+r"(dst_abgr), // %[dst_abgr] [width]"+rm"(width) // %[width] - : [yuvconstants]"r"(yuvconstants), // %[yuvconstants] - [kShuffleUYVYY]"m"(kShuffleUYVYY), - [kShuffleUYVYUV]"m"(kShuffleUYVYUV) - : "memory", "cc", YUVTORGB_REGS // Does not use r14. - "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" + : [kYuvConstants]"r"(&kYuvConstants.kUVToB) // %[kYuvConstants] + : "memory", "cc", NACL_R14 + "xmm0", "xmm1", "xmm2", "xmm3", "xmm5" ); } @@ -1945,16 +1932,14 @@ void OMITFP I422ToRGBARow_SSSE3(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, uint8* dst_rgba, - const struct YuvConstants* yuvconstants, int width) { asm volatile ( - YUVTORGB_SETUP(yuvconstants) "sub %[u_buf],%[v_buf] \n" "pcmpeqb %%xmm5,%%xmm5 \n" LABELALIGN "1: \n" READYUV422 - YUVTORGB(yuvconstants) + YUVTORGB(kYuvConstants) STORERGBA "sub $0x8,%[width] \n" "jg 1b \n" @@ -1963,224 +1948,118 @@ void OMITFP I422ToRGBARow_SSSE3(const uint8* y_buf, [v_buf]"+r"(v_buf), // %[v_buf] [dst_rgba]"+r"(dst_rgba), // %[dst_rgba] [width]"+rm"(width) // %[width] - : [yuvconstants]"r"(yuvconstants) // %[yuvconstants] - : "memory", "cc", NACL_R14 YUVTORGB_REGS - "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" + : [kYuvConstants]"r"(&kYuvConstants.kUVToB) // %[kYuvConstants] + : "memory", "cc", NACL_R14 + "xmm0", "xmm1", "xmm2", "xmm3", "xmm5" ); } #endif // HAS_I422TOARGBROW_SSSE3 -// Read 16 UV from 444 -#define READYUV444_AVX2 \ - "vmovdqu " MEMACCESS([u_buf]) ",%%xmm0 \n" \ - MEMOPREG(vmovdqu, 0x00, [u_buf], [v_buf], 1, xmm1) \ - "lea " MEMLEA(0x10, [u_buf]) ",%[u_buf] \n" \ - "vpermq $0xd8,%%ymm0,%%ymm0 \n" \ - "vpermq $0xd8,%%ymm1,%%ymm1 \n" \ - "vpunpcklbw %%ymm1,%%ymm0,%%ymm0 \n" \ - "vmovdqu " MEMACCESS([y_buf]) ",%%xmm4 \n" \ - "vpermq $0xd8,%%ymm4,%%ymm4 \n" \ - "vpunpcklbw %%ymm4,%%ymm4,%%ymm4 \n" \ - "lea " MEMLEA(0x10, [y_buf]) ",%[y_buf] \n" - // Read 8 UV from 422, upsample to 16 UV. #define READYUV422_AVX2 \ - "vmovq " MEMACCESS([u_buf]) ",%%xmm0 \n" \ + "vmovq " MEMACCESS([u_buf]) ",%%xmm0 \n" \ MEMOPREG(vmovq, 0x00, [u_buf], [v_buf], 1, xmm1) \ "lea " MEMLEA(0x8, [u_buf]) ",%[u_buf] \n" \ "vpunpcklbw %%ymm1,%%ymm0,%%ymm0 \n" \ "vpermq $0xd8,%%ymm0,%%ymm0 \n" \ - "vpunpcklwd %%ymm0,%%ymm0,%%ymm0 \n" \ - "vmovdqu " MEMACCESS([y_buf]) ",%%xmm4 \n" \ - "vpermq $0xd8,%%ymm4,%%ymm4 \n" \ - "vpunpcklbw %%ymm4,%%ymm4,%%ymm4 \n" \ - "lea " MEMLEA(0x10, [y_buf]) ",%[y_buf] \n" - -// Read 8 UV from 422, upsample to 16 UV. With 16 Alpha. -#define READYUVA422_AVX2 \ - "vmovq " MEMACCESS([u_buf]) ",%%xmm0 \n" \ - MEMOPREG(vmovq, 0x00, [u_buf], [v_buf], 1, xmm1) \ - "lea " MEMLEA(0x8, [u_buf]) ",%[u_buf] \n" \ - "vpunpcklbw %%ymm1,%%ymm0,%%ymm0 \n" \ - "vpermq $0xd8,%%ymm0,%%ymm0 \n" \ - "vpunpcklwd %%ymm0,%%ymm0,%%ymm0 \n" \ - "vmovdqu " MEMACCESS([y_buf]) ",%%xmm4 \n" \ - "vpermq $0xd8,%%ymm4,%%ymm4 \n" \ - "vpunpcklbw %%ymm4,%%ymm4,%%ymm4 \n" \ - "lea " MEMLEA(0x10, [y_buf]) ",%[y_buf] \n" \ - "vmovdqu " MEMACCESS([a_buf]) ",%%xmm5 \n" \ - "vpermq $0xd8,%%ymm5,%%ymm5 \n" \ - "lea " MEMLEA(0x10, [a_buf]) ",%[a_buf] \n" - -// Read 4 UV from 411, upsample to 16 UV. -#define READYUV411_AVX2 \ - "vmovd " MEMACCESS([u_buf]) ",%%xmm0 \n" \ - MEMOPREG(vmovd, 0x00, [u_buf], [v_buf], 1, xmm1) \ - "lea " MEMLEA(0x4, [u_buf]) ",%[u_buf] \n" \ - "vpunpcklbw %%ymm1,%%ymm0,%%ymm0 \n" \ - "vpunpcklwd %%ymm0,%%ymm0,%%ymm0 \n" \ - "vpermq $0xd8,%%ymm0,%%ymm0 \n" \ - "vpunpckldq %%ymm0,%%ymm0,%%ymm0 \n" \ - "vmovdqu " MEMACCESS([y_buf]) ",%%xmm4 \n" \ - "vpermq $0xd8,%%ymm4,%%ymm4 \n" \ - "vpunpcklbw %%ymm4,%%ymm4,%%ymm4 \n" \ - "lea " MEMLEA(0x10, [y_buf]) ",%[y_buf] \n" - -// Read 8 UV from NV12, upsample to 16 UV. -#define READNV12_AVX2 \ - "vmovdqu " MEMACCESS([uv_buf]) ",%%xmm0 \n" \ - "lea " MEMLEA(0x10, [uv_buf]) ",%[uv_buf] \n" \ - "vpermq $0xd8,%%ymm0,%%ymm0 \n" \ - "vpunpcklwd %%ymm0,%%ymm0,%%ymm0 \n" \ - "vmovdqu " MEMACCESS([y_buf]) ",%%xmm4 \n" \ - "vpermq $0xd8,%%ymm4,%%ymm4 \n" \ - "vpunpcklbw %%ymm4,%%ymm4,%%ymm4 \n" \ - "lea " MEMLEA(0x10, [y_buf]) ",%[y_buf] \n" - -// Read 8 VU from NV21, upsample to 16 UV. -#define READNV21_AVX2 \ - "vmovdqu " MEMACCESS([vu_buf]) ",%%xmm0 \n" \ - "lea " MEMLEA(0x10, [vu_buf]) ",%[vu_buf] \n" \ - "vpermq $0xd8,%%ymm0,%%ymm0 \n" \ - "vpshufb %[kShuffleNV21], %%ymm0, %%ymm0 \n" \ - "vmovdqu " MEMACCESS([y_buf]) ",%%xmm4 \n" \ - "vpermq $0xd8,%%ymm4,%%ymm4 \n" \ - "vpunpcklbw %%ymm4,%%ymm4,%%ymm4 \n" \ - "lea " MEMLEA(0x10, [y_buf]) ",%[y_buf] \n" - -// Read 8 YUY2 with 16 Y and upsample 8 UV to 16 UV. -#define READYUY2_AVX2 \ - "vmovdqu " MEMACCESS([yuy2_buf]) ",%%ymm4 \n" \ - "vpshufb %[kShuffleYUY2Y], %%ymm4, %%ymm4 \n" \ - "vmovdqu " MEMACCESS([yuy2_buf]) ",%%ymm0 \n" \ - "vpshufb %[kShuffleYUY2UV], %%ymm0, %%ymm0 \n" \ - "lea " MEMLEA(0x20, [yuy2_buf]) ",%[yuy2_buf] \n" - -// Read 8 UYVY with 16 Y and upsample 8 UV to 16 UV. -#define READUYVY_AVX2 \ - "vmovdqu " MEMACCESS([uyvy_buf]) ",%%ymm4 \n" \ - "vpshufb %[kShuffleUYVYY], %%ymm4, %%ymm4 \n" \ - "vmovdqu " MEMACCESS([uyvy_buf]) ",%%ymm0 \n" \ - "vpshufb %[kShuffleUYVYUV], %%ymm0, %%ymm0 \n" \ - "lea " MEMLEA(0x20, [uyvy_buf]) ",%[uyvy_buf] \n" - -#if defined(__x86_64__) -#define YUVTORGB_SETUP_AVX2(yuvconstants) \ - "vmovdqa " MEMACCESS([yuvconstants]) ",%%ymm8 \n" \ - "vmovdqa " MEMACCESS2(32, [yuvconstants]) ",%%ymm9 \n" \ - "vmovdqa " MEMACCESS2(64, [yuvconstants]) ",%%ymm10 \n" \ - "vmovdqa " MEMACCESS2(96, [yuvconstants]) ",%%ymm11 \n" \ - "vmovdqa " MEMACCESS2(128, [yuvconstants]) ",%%ymm12 \n" \ - "vmovdqa " MEMACCESS2(160, [yuvconstants]) ",%%ymm13 \n" \ - "vmovdqa " MEMACCESS2(192, [yuvconstants]) ",%%ymm14 \n" -#define YUVTORGB_AVX2(yuvconstants) \ - "vpmaddubsw %%ymm10,%%ymm0,%%ymm2 \n" \ - "vpmaddubsw %%ymm9,%%ymm0,%%ymm1 \n" \ - "vpmaddubsw %%ymm8,%%ymm0,%%ymm0 \n" \ - "vpsubw %%ymm2,%%ymm13,%%ymm2 \n" \ - "vpsubw %%ymm1,%%ymm12,%%ymm1 \n" \ - "vpsubw %%ymm0,%%ymm11,%%ymm0 \n" \ - "vpmulhuw %%ymm14,%%ymm4,%%ymm4 \n" \ - "vpaddsw %%ymm4,%%ymm0,%%ymm0 \n" \ - "vpaddsw %%ymm4,%%ymm1,%%ymm1 \n" \ - "vpaddsw %%ymm4,%%ymm2,%%ymm2 \n" \ - "vpsraw $0x6,%%ymm0,%%ymm0 \n" \ - "vpsraw $0x6,%%ymm1,%%ymm1 \n" \ - "vpsraw $0x6,%%ymm2,%%ymm2 \n" \ - "vpackuswb %%ymm0,%%ymm0,%%ymm0 \n" \ - "vpackuswb %%ymm1,%%ymm1,%%ymm1 \n" \ - "vpackuswb %%ymm2,%%ymm2,%%ymm2 \n" -#define YUVTORGB_REGS_AVX2 \ - "xmm8", "xmm9", "xmm10", "xmm11", "xmm12", "xmm13", "xmm14", -#else // Convert 16 pixels: 16 UV and 16 Y. -#define YUVTORGB_SETUP_AVX2(yuvconstants) -#define YUVTORGB_AVX2(yuvconstants) \ - "vpmaddubsw " MEMACCESS2(64, [yuvconstants]) ",%%ymm0,%%ymm2 \n" \ - "vpmaddubsw " MEMACCESS2(32, [yuvconstants]) ",%%ymm0,%%ymm1 \n" \ - "vpmaddubsw " MEMACCESS([yuvconstants]) ",%%ymm0,%%ymm0 \n" \ - "vmovdqu " MEMACCESS2(160, [yuvconstants]) ",%%ymm3 \n" \ + "vpunpcklwd %%ymm0,%%ymm0,%%ymm0 \n" + +// Convert 16 pixels: 16 UV and 16 Y. +#define YUVTORGB_AVX2(YuvConstants) \ + "vpmaddubsw " MEMACCESS2(64, [YuvConstants]) ",%%ymm0,%%ymm2 \n" \ + "vpmaddubsw " MEMACCESS2(32, [YuvConstants]) ",%%ymm0,%%ymm1 \n" \ + "vpmaddubsw " MEMACCESS([YuvConstants]) ",%%ymm0,%%ymm0 \n" \ + "vmovdqu " MEMACCESS2(160, [YuvConstants]) ",%%ymm3 \n" \ "vpsubw %%ymm2,%%ymm3,%%ymm2 \n" \ - "vmovdqu " MEMACCESS2(128, [yuvconstants]) ",%%ymm3 \n" \ + "vmovdqu " MEMACCESS2(128, [YuvConstants]) ",%%ymm3 \n" \ "vpsubw %%ymm1,%%ymm3,%%ymm1 \n" \ - "vmovdqu " MEMACCESS2(96, [yuvconstants]) ",%%ymm3 \n" \ + "vmovdqu " MEMACCESS2(96, [YuvConstants]) ",%%ymm3 \n" \ "vpsubw %%ymm0,%%ymm3,%%ymm0 \n" \ - "vpmulhuw " MEMACCESS2(192, [yuvconstants]) ",%%ymm4,%%ymm4 \n" \ - "vpaddsw %%ymm4,%%ymm0,%%ymm0 \n" \ - "vpaddsw %%ymm4,%%ymm1,%%ymm1 \n" \ - "vpaddsw %%ymm4,%%ymm2,%%ymm2 \n" \ - "vpsraw $0x6,%%ymm0,%%ymm0 \n" \ - "vpsraw $0x6,%%ymm1,%%ymm1 \n" \ - "vpsraw $0x6,%%ymm2,%%ymm2 \n" \ - "vpackuswb %%ymm0,%%ymm0,%%ymm0 \n" \ - "vpackuswb %%ymm1,%%ymm1,%%ymm1 \n" \ - "vpackuswb %%ymm2,%%ymm2,%%ymm2 \n" -#define YUVTORGB_REGS_AVX2 -#endif - -// Store 16 ARGB values. -#define STOREARGB_AVX2 \ - "vpunpcklbw %%ymm1,%%ymm0,%%ymm0 \n" \ - "vpermq $0xd8,%%ymm0,%%ymm0 \n" \ - "vpunpcklbw %%ymm5,%%ymm2,%%ymm2 \n" \ - "vpermq $0xd8,%%ymm2,%%ymm2 \n" \ - "vpunpcklwd %%ymm2,%%ymm0,%%ymm1 \n" \ - "vpunpckhwd %%ymm2,%%ymm0,%%ymm0 \n" \ - "vmovdqu %%ymm1," MEMACCESS([dst_argb]) " \n" \ - "vmovdqu %%ymm0," MEMACCESS2(0x20, [dst_argb]) " \n" \ - "lea " MEMLEA(0x40, [dst_argb]) ", %[dst_argb] \n" - -#ifdef HAS_I444TOARGBROW_AVX2 + "vmovdqu " MEMACCESS([y_buf]) ",%%xmm3 \n" \ + "lea " MEMLEA(0x10, [y_buf]) ",%[y_buf] \n" \ + "vpermq $0xd8,%%ymm3,%%ymm3 \n" \ + "vpunpcklbw %%ymm3,%%ymm3,%%ymm3 \n" \ + "vpmulhuw " MEMACCESS2(192, [YuvConstants]) ",%%ymm3,%%ymm3 \n" \ + "vpaddsw %%ymm3,%%ymm0,%%ymm0 \n" \ + "vpaddsw %%ymm3,%%ymm1,%%ymm1 \n" \ + "vpaddsw %%ymm3,%%ymm2,%%ymm2 \n" \ + "vpsraw $0x6,%%ymm0,%%ymm0 \n" \ + "vpsraw $0x6,%%ymm1,%%ymm1 \n" \ + "vpsraw $0x6,%%ymm2,%%ymm2 \n" \ + "vpackuswb %%ymm0,%%ymm0,%%ymm0 \n" \ + "vpackuswb %%ymm1,%%ymm1,%%ymm1 \n" \ + "vpackuswb %%ymm2,%%ymm2,%%ymm2 \n" + +#if defined(HAS_I422TOBGRAROW_AVX2) // 16 pixels -// 16 UV values with 16 Y producing 16 ARGB (64 bytes). -void OMITFP I444ToARGBRow_AVX2(const uint8* y_buf, +// 8 UV values upsampled to 16 UV, mixed with 16 Y producing 16 BGRA (64 bytes). +void OMITFP I422ToBGRARow_AVX2(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, - uint8* dst_argb, - const struct YuvConstants* yuvconstants, + uint8* dst_bgra, int width) { asm volatile ( - YUVTORGB_SETUP_AVX2(yuvconstants) "sub %[u_buf],%[v_buf] \n" - "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" + "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" LABELALIGN "1: \n" - READYUV444_AVX2 - YUVTORGB_AVX2(yuvconstants) - STOREARGB_AVX2 + READYUV422_AVX2 + YUVTORGB_AVX2(kYuvConstants) + + // Step 3: Weave into BGRA + "vpunpcklbw %%ymm0,%%ymm1,%%ymm1 \n" // GB + "vpermq $0xd8,%%ymm1,%%ymm1 \n" + "vpunpcklbw %%ymm2,%%ymm5,%%ymm2 \n" // AR + "vpermq $0xd8,%%ymm2,%%ymm2 \n" + "vpunpcklwd %%ymm1,%%ymm2,%%ymm0 \n" // ARGB first 8 pixels + "vpunpckhwd %%ymm1,%%ymm2,%%ymm2 \n" // ARGB next 8 pixels + + "vmovdqu %%ymm0," MEMACCESS([dst_bgra]) "\n" + "vmovdqu %%ymm2," MEMACCESS2(0x20,[dst_bgra]) "\n" + "lea " MEMLEA(0x40,[dst_bgra]) ",%[dst_bgra] \n" "sub $0x10,%[width] \n" "jg 1b \n" "vzeroupper \n" : [y_buf]"+r"(y_buf), // %[y_buf] [u_buf]"+r"(u_buf), // %[u_buf] [v_buf]"+r"(v_buf), // %[v_buf] - [dst_argb]"+r"(dst_argb), // %[dst_argb] + [dst_bgra]"+r"(dst_bgra), // %[dst_bgra] [width]"+rm"(width) // %[width] - : [yuvconstants]"r"(yuvconstants) // %[yuvconstants] - : "memory", "cc", NACL_R14 YUVTORGB_REGS_AVX2 - "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" + : [kYuvConstants]"r"(&kYuvConstants.kUVToB) // %[kYuvConstants] + : "memory", "cc", NACL_R14 + "xmm0", "xmm1", "xmm2", "xmm3", "xmm5" ); } -#endif // HAS_I444TOARGBROW_AVX2 +#endif // HAS_I422TOBGRAROW_AVX2 -#ifdef HAS_I411TOARGBROW_AVX2 +#if defined(HAS_I422TOARGBROW_AVX2) // 16 pixels -// 4 UV values upsampled to 16 UV, mixed with 16 Y producing 16 ARGB (64 bytes). -void OMITFP I411ToARGBRow_AVX2(const uint8* y_buf, +// 8 UV values upsampled to 16 UV, mixed with 16 Y producing 16 ARGB (64 bytes). +void OMITFP I422ToARGBRow_AVX2(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, uint8* dst_argb, - const struct YuvConstants* yuvconstants, int width) { asm volatile ( - YUVTORGB_SETUP_AVX2(yuvconstants) "sub %[u_buf],%[v_buf] \n" - "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" + "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" LABELALIGN "1: \n" - READYUV411_AVX2 - YUVTORGB_AVX2(yuvconstants) - STOREARGB_AVX2 + READYUV422_AVX2 + YUVTORGB_AVX2(kYuvConstants) + + // Step 3: Weave into ARGB + "vpunpcklbw %%ymm1,%%ymm0,%%ymm0 \n" // BG + "vpermq $0xd8,%%ymm0,%%ymm0 \n" + "vpunpcklbw %%ymm5,%%ymm2,%%ymm2 \n" // RA + "vpermq $0xd8,%%ymm2,%%ymm2 \n" + "vpunpcklwd %%ymm2,%%ymm0,%%ymm1 \n" // BGRA first 8 pixels + "vpunpckhwd %%ymm2,%%ymm0,%%ymm0 \n" // BGRA next 8 pixels + + "vmovdqu %%ymm1," MEMACCESS([dst_argb]) "\n" + "vmovdqu %%ymm0," MEMACCESS2(0x20,[dst_argb]) "\n" + "lea " MEMLEA(0x40,[dst_argb]) ",%[dst_argb] \n" "sub $0x10,%[width] \n" "jg 1b \n" "vzeroupper \n" @@ -2189,31 +2068,40 @@ void OMITFP I411ToARGBRow_AVX2(const uint8* y_buf, [v_buf]"+r"(v_buf), // %[v_buf] [dst_argb]"+r"(dst_argb), // %[dst_argb] [width]"+rm"(width) // %[width] - : [yuvconstants]"r"(yuvconstants) // %[yuvconstants] - : "memory", "cc", NACL_R14 YUVTORGB_REGS_AVX2 - "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" + : [kYuvConstants]"r"(&kYuvConstants.kUVToB) // %[kYuvConstants] + : "memory", "cc", NACL_R14 + "xmm0", "xmm1", "xmm2", "xmm3", "xmm5" ); } -#endif // HAS_I411TOARGBROW_AVX2 +#endif // HAS_I422TOARGBROW_AVX2 -#if defined(HAS_I422TOARGBROW_AVX2) +#if defined(HAS_J422TOARGBROW_AVX2) // 16 pixels // 8 UV values upsampled to 16 UV, mixed with 16 Y producing 16 ARGB (64 bytes). -void OMITFP I422ToARGBRow_AVX2(const uint8* y_buf, +void OMITFP J422ToARGBRow_AVX2(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, uint8* dst_argb, - const struct YuvConstants* yuvconstants, int width) { asm volatile ( - YUVTORGB_SETUP_AVX2(yuvconstants) "sub %[u_buf],%[v_buf] \n" - "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" + "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" LABELALIGN "1: \n" READYUV422_AVX2 - YUVTORGB_AVX2(yuvconstants) - STOREARGB_AVX2 + YUVTORGB_AVX2(kYuvConstants) + + // Step 3: Weave into ARGB + "vpunpcklbw %%ymm1,%%ymm0,%%ymm0 \n" // BG + "vpermq $0xd8,%%ymm0,%%ymm0 \n" + "vpunpcklbw %%ymm5,%%ymm2,%%ymm2 \n" // RA + "vpermq $0xd8,%%ymm2,%%ymm2 \n" + "vpunpcklwd %%ymm2,%%ymm0,%%ymm1 \n" // BGRA first 8 pixels + "vpunpckhwd %%ymm2,%%ymm0,%%ymm0 \n" // BGRA next 8 pixels + + "vmovdqu %%ymm1," MEMACCESS([dst_argb]) "\n" + "vmovdqu %%ymm0," MEMACCESS2(0x20,[dst_argb]) "\n" + "lea " MEMLEA(0x40,[dst_argb]) ",%[dst_argb] \n" "sub $0x10,%[width] \n" "jg 1b \n" "vzeroupper \n" @@ -2222,50 +2110,53 @@ void OMITFP I422ToARGBRow_AVX2(const uint8* y_buf, [v_buf]"+r"(v_buf), // %[v_buf] [dst_argb]"+r"(dst_argb), // %[dst_argb] [width]"+rm"(width) // %[width] - : [yuvconstants]"r"(yuvconstants) // %[yuvconstants] - : "memory", "cc", NACL_R14 YUVTORGB_REGS_AVX2 - "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" + : [kYuvConstants]"r"(&kYuvJConstants.kUVToB) // %[kYuvConstants] + : "memory", "cc", NACL_R14 + "xmm0", "xmm1", "xmm2", "xmm3", "xmm5" ); } -#endif // HAS_I422TOARGBROW_AVX2 +#endif // HAS_J422TOARGBROW_AVX2 -#if defined(HAS_I422ALPHATOARGBROW_AVX2) +#if defined(HAS_I422TOABGRROW_AVX2) // 16 pixels -// 8 UV values upsampled to 16 UV, mixed with 16 Y and 16 A producing 16 ARGB. -void OMITFP I422AlphaToARGBRow_AVX2(const uint8* y_buf, +// 8 UV values upsampled to 16 UV, mixed with 16 Y producing 16 ABGR (64 bytes). +void OMITFP I422ToABGRRow_AVX2(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, - const uint8* a_buf, uint8* dst_argb, - const struct YuvConstants* yuvconstants, int width) { asm volatile ( - YUVTORGB_SETUP_AVX2(yuvconstants) "sub %[u_buf],%[v_buf] \n" + "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" LABELALIGN "1: \n" - READYUVA422_AVX2 - YUVTORGB_AVX2(yuvconstants) - STOREARGB_AVX2 - "subl $0x10,%[width] \n" + READYUV422_AVX2 + YUVTORGB_AVX2(kYuvConstants) + + // Step 3: Weave into ABGR + "vpunpcklbw %%ymm1,%%ymm2,%%ymm1 \n" // RG + "vpermq $0xd8,%%ymm1,%%ymm1 \n" + "vpunpcklbw %%ymm5,%%ymm0,%%ymm2 \n" // BA + "vpermq $0xd8,%%ymm2,%%ymm2 \n" + "vpunpcklwd %%ymm2,%%ymm1,%%ymm0 \n" // RGBA first 8 pixels + "vpunpckhwd %%ymm2,%%ymm1,%%ymm1 \n" // RGBA next 8 pixels + "vmovdqu %%ymm0," MEMACCESS([dst_argb]) "\n" + "vmovdqu %%ymm1," MEMACCESS2(0x20,[dst_argb]) "\n" + "lea " MEMLEA(0x40,[dst_argb]) ",%[dst_argb] \n" + "sub $0x10,%[width] \n" "jg 1b \n" "vzeroupper \n" : [y_buf]"+r"(y_buf), // %[y_buf] [u_buf]"+r"(u_buf), // %[u_buf] [v_buf]"+r"(v_buf), // %[v_buf] - [a_buf]"+r"(a_buf), // %[a_buf] [dst_argb]"+r"(dst_argb), // %[dst_argb] -#if defined(__i386__) && defined(__pic__) - [width]"+m"(width) // %[width] -#else [width]"+rm"(width) // %[width] -#endif - : [yuvconstants]"r"(yuvconstants) // %[yuvconstants] - : "memory", "cc", NACL_R14 YUVTORGB_REGS_AVX2 - "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" + : [kYuvConstants]"r"(&kYuvConstants.kUVToB) // %[kYuvConstants] + : "memory", "cc", NACL_R14 + "xmm0", "xmm1", "xmm2", "xmm3", "xmm5" ); } -#endif // HAS_I422ALPHATOARGBROW_AVX2 +#endif // HAS_I422TOABGRROW_AVX2 #if defined(HAS_I422TORGBAROW_AVX2) // 16 pixels @@ -2274,16 +2165,14 @@ void OMITFP I422ToRGBARow_AVX2(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, uint8* dst_argb, - const struct YuvConstants* yuvconstants, int width) { asm volatile ( - YUVTORGB_SETUP_AVX2(yuvconstants) "sub %[u_buf],%[v_buf] \n" "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" LABELALIGN "1: \n" READYUV422_AVX2 - YUVTORGB_AVX2(yuvconstants) + YUVTORGB_AVX2(kYuvConstants) // Step 3: Weave into RGBA "vpunpcklbw %%ymm2,%%ymm1,%%ymm1 \n" @@ -2303,134 +2192,13 @@ void OMITFP I422ToRGBARow_AVX2(const uint8* y_buf, [v_buf]"+r"(v_buf), // %[v_buf] [dst_argb]"+r"(dst_argb), // %[dst_argb] [width]"+rm"(width) // %[width] - : [yuvconstants]"r"(yuvconstants) // %[yuvconstants] - : "memory", "cc", NACL_R14 YUVTORGB_REGS_AVX2 - "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" + : [kYuvConstants]"r"(&kYuvConstants.kUVToB) // %[kYuvConstants] + : "memory", "cc", NACL_R14 + "xmm0", "xmm1", "xmm2", "xmm3", "xmm5" ); } #endif // HAS_I422TORGBAROW_AVX2 -#if defined(HAS_NV12TOARGBROW_AVX2) -// 16 pixels. -// 8 UV values upsampled to 16 UV, mixed with 16 Y producing 16 ARGB (64 bytes). -void OMITFP NV12ToARGBRow_AVX2(const uint8* y_buf, - const uint8* uv_buf, - uint8* dst_argb, - const struct YuvConstants* yuvconstants, - int width) { - asm volatile ( - YUVTORGB_SETUP_AVX2(yuvconstants) - "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" - LABELALIGN - "1: \n" - READNV12_AVX2 - YUVTORGB_AVX2(yuvconstants) - STOREARGB_AVX2 - "sub $0x10,%[width] \n" - "jg 1b \n" - "vzeroupper \n" - : [y_buf]"+r"(y_buf), // %[y_buf] - [uv_buf]"+r"(uv_buf), // %[uv_buf] - [dst_argb]"+r"(dst_argb), // %[dst_argb] - [width]"+rm"(width) // %[width] - : [yuvconstants]"r"(yuvconstants) // %[yuvconstants] - : "memory", "cc", YUVTORGB_REGS_AVX2 // Does not use r14. - "xmm0", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" - ); -} -#endif // HAS_NV12TOARGBROW_AVX2 - -#if defined(HAS_NV21TOARGBROW_AVX2) -// 16 pixels. -// 8 VU values upsampled to 16 UV, mixed with 16 Y producing 16 ARGB (64 bytes). -void OMITFP NV21ToARGBRow_AVX2(const uint8* y_buf, - const uint8* vu_buf, - uint8* dst_argb, - const struct YuvConstants* yuvconstants, - int width) { - asm volatile ( - YUVTORGB_SETUP_AVX2(yuvconstants) - "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" - LABELALIGN - "1: \n" - READNV21_AVX2 - YUVTORGB_AVX2(yuvconstants) - STOREARGB_AVX2 - "sub $0x10,%[width] \n" - "jg 1b \n" - "vzeroupper \n" - : [y_buf]"+r"(y_buf), // %[y_buf] - [vu_buf]"+r"(vu_buf), // %[vu_buf] - [dst_argb]"+r"(dst_argb), // %[dst_argb] - [width]"+rm"(width) // %[width] - : [yuvconstants]"r"(yuvconstants), // %[yuvconstants] - [kShuffleNV21]"m"(kShuffleNV21) - : "memory", "cc", YUVTORGB_REGS_AVX2 // Does not use r14. - "xmm0", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" - ); -} -#endif // HAS_NV21TOARGBROW_AVX2 - -#if defined(HAS_YUY2TOARGBROW_AVX2) -// 16 pixels. -// 8 YUY2 values with 16 Y and 8 UV producing 16 ARGB (64 bytes). -void OMITFP YUY2ToARGBRow_AVX2(const uint8* yuy2_buf, - uint8* dst_argb, - const struct YuvConstants* yuvconstants, - int width) { - asm volatile ( - YUVTORGB_SETUP_AVX2(yuvconstants) - "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" - LABELALIGN - "1: \n" - READYUY2_AVX2 - YUVTORGB_AVX2(yuvconstants) - STOREARGB_AVX2 - "sub $0x10,%[width] \n" - "jg 1b \n" - "vzeroupper \n" - : [yuy2_buf]"+r"(yuy2_buf), // %[yuy2_buf] - [dst_argb]"+r"(dst_argb), // %[dst_argb] - [width]"+rm"(width) // %[width] - : [yuvconstants]"r"(yuvconstants), // %[yuvconstants] - [kShuffleYUY2Y]"m"(kShuffleYUY2Y), - [kShuffleYUY2UV]"m"(kShuffleYUY2UV) - : "memory", "cc", YUVTORGB_REGS_AVX2 // Does not use r14. - "xmm0", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" - ); -} -#endif // HAS_YUY2TOARGBROW_AVX2 - -#if defined(HAS_UYVYTOARGBROW_AVX2) -// 16 pixels. -// 8 UYVY values with 16 Y and 8 UV producing 16 ARGB (64 bytes). -void OMITFP UYVYToARGBRow_AVX2(const uint8* uyvy_buf, - uint8* dst_argb, - const struct YuvConstants* yuvconstants, - int width) { - asm volatile ( - YUVTORGB_SETUP_AVX2(yuvconstants) - "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" - LABELALIGN - "1: \n" - READUYVY_AVX2 - YUVTORGB_AVX2(yuvconstants) - STOREARGB_AVX2 - "sub $0x10,%[width] \n" - "jg 1b \n" - "vzeroupper \n" - : [uyvy_buf]"+r"(uyvy_buf), // %[uyvy_buf] - [dst_argb]"+r"(dst_argb), // %[dst_argb] - [width]"+rm"(width) // %[width] - : [yuvconstants]"r"(yuvconstants), // %[yuvconstants] - [kShuffleUYVYY]"m"(kShuffleUYVYY), - [kShuffleUYVYUV]"m"(kShuffleUYVYUV) - : "memory", "cc", YUVTORGB_REGS_AVX2 // Does not use r14. - "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" - ); -} -#endif // HAS_UYVYTOARGBROW_AVX2 - #ifdef HAS_I400TOARGBROW_SSE2 void I400ToARGBRow_SSE2(const uint8* y_buf, uint8* dst_argb, int width) { asm volatile ( @@ -2576,7 +2344,35 @@ void MirrorRow_AVX2(const uint8* src, uint8* dst, int width) { } #endif // HAS_MIRRORROW_AVX2 -#ifdef HAS_MIRRORUVROW_SSSE3 +#ifdef HAS_MIRRORROW_SSE2 +void MirrorRow_SSE2(const uint8* src, uint8* dst, int width) { + intptr_t temp_width = (intptr_t)(width); + asm volatile ( + LABELALIGN + "1: \n" + MEMOPREG(movdqu,-0x10,0,2,1,xmm0) // movdqu -0x10(%0,%2),%%xmm0 + "movdqa %%xmm0,%%xmm1 \n" + "psllw $0x8,%%xmm0 \n" + "psrlw $0x8,%%xmm1 \n" + "por %%xmm1,%%xmm0 \n" + "pshuflw $0x1b,%%xmm0,%%xmm0 \n" + "pshufhw $0x1b,%%xmm0,%%xmm0 \n" + "pshufd $0x4e,%%xmm0,%%xmm0 \n" + "movdqu %%xmm0," MEMACCESS(1) " \n" + "lea " MEMLEA(0x10,1)",%1 \n" + "sub $0x10,%2 \n" + "jg 1b \n" + : "+r"(src), // %0 + "+r"(dst), // %1 + "+r"(temp_width) // %2 + : + : "memory", "cc", NACL_R14 + "xmm0", "xmm1" + ); +} +#endif // HAS_MIRRORROW_SSE2 + +#ifdef HAS_MIRRORROW_UV_SSSE3 // Shuffle table for reversing the bytes of UV channels. static uvec8 kShuffleMirrorUV = { 14u, 12u, 10u, 8u, 6u, 4u, 2u, 0u, 15u, 13u, 11u, 9u, 7u, 5u, 3u, 1u @@ -2607,7 +2403,7 @@ void MirrorUVRow_SSSE3(const uint8* src, uint8* dst_u, uint8* dst_v, "xmm0", "xmm1" ); } -#endif // HAS_MIRRORUVROW_SSSE3 +#endif // HAS_MIRRORROW_UV_SSSE3 #ifdef HAS_ARGBMIRRORROW_SSE2 @@ -2662,8 +2458,7 @@ void ARGBMirrorRow_AVX2(const uint8* src, uint8* dst, int width) { #endif // HAS_ARGBMIRRORROW_AVX2 #ifdef HAS_SPLITUVROW_AVX2 -void SplitUVRow_AVX2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, - int width) { +void SplitUVRow_AVX2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, int pix) { asm volatile ( "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" "vpsrlw $0x8,%%ymm5,%%ymm5 \n" @@ -2690,7 +2485,7 @@ void SplitUVRow_AVX2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, : "+r"(src_uv), // %0 "+r"(dst_u), // %1 "+r"(dst_v), // %2 - "+r"(width) // %3 + "+r"(pix) // %3 : : "memory", "cc", NACL_R14 "xmm0", "xmm1", "xmm2", "xmm3", "xmm5" @@ -2699,8 +2494,7 @@ void SplitUVRow_AVX2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, #endif // HAS_SPLITUVROW_AVX2 #ifdef HAS_SPLITUVROW_SSE2 -void SplitUVRow_SSE2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, - int width) { +void SplitUVRow_SSE2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, int pix) { asm volatile ( "pcmpeqb %%xmm5,%%xmm5 \n" "psrlw $0x8,%%xmm5 \n" @@ -2726,7 +2520,7 @@ void SplitUVRow_SSE2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, : "+r"(src_uv), // %0 "+r"(dst_u), // %1 "+r"(dst_v), // %2 - "+r"(width) // %3 + "+r"(pix) // %3 : : "memory", "cc", NACL_R14 "xmm0", "xmm1", "xmm2", "xmm3", "xmm5" @@ -2797,23 +2591,8 @@ void MergeUVRow_SSE2(const uint8* src_u, const uint8* src_v, uint8* dst_uv, #ifdef HAS_COPYROW_SSE2 void CopyRow_SSE2(const uint8* src, uint8* dst, int count) { asm volatile ( - "test $0xf,%0 \n" - "jne 2f \n" - "test $0xf,%1 \n" - "jne 2f \n" LABELALIGN "1: \n" - "movdqa " MEMACCESS(0) ",%%xmm0 \n" - "movdqa " MEMACCESS2(0x10,0) ",%%xmm1 \n" - "lea " MEMLEA(0x20,0) ",%0 \n" - "movdqa %%xmm0," MEMACCESS(1) " \n" - "movdqa %%xmm1," MEMACCESS2(0x10,1) " \n" - "lea " MEMLEA(0x20,1) ",%1 \n" - "sub $0x20,%2 \n" - "jg 1b \n" - "jmp 9f \n" - LABELALIGN - "2: \n" "movdqu " MEMACCESS(0) ",%%xmm0 \n" "movdqu " MEMACCESS2(0x10,0) ",%%xmm1 \n" "lea " MEMLEA(0x20,0) ",%0 \n" @@ -2821,8 +2600,7 @@ void CopyRow_SSE2(const uint8* src, uint8* dst, int count) { "movdqu %%xmm1," MEMACCESS2(0x10,1) " \n" "lea " MEMLEA(0x20,1) ",%1 \n" "sub $0x20,%2 \n" - "jg 2b \n" - "9: \n" + "jg 1b \n" : "+r"(src), // %0 "+r"(dst), // %1 "+r"(count) // %2 @@ -2936,33 +2714,6 @@ void ARGBCopyAlphaRow_AVX2(const uint8* src, uint8* dst, int width) { } #endif // HAS_ARGBCOPYALPHAROW_AVX2 -#ifdef HAS_ARGBEXTRACTALPHAROW_SSE2 -// width in pixels -void ARGBExtractAlphaRow_SSE2(const uint8* src_argb, uint8* dst_a, int width) { - asm volatile ( - LABELALIGN - "1: \n" - "movdqu " MEMACCESS(0) ", %%xmm0 \n" - "movdqu " MEMACCESS2(0x10, 0) ", %%xmm1 \n" - "lea " MEMLEA(0x20, 0) ", %0 \n" - "psrld $0x18, %%xmm0 \n" - "psrld $0x18, %%xmm1 \n" - "packssdw %%xmm1, %%xmm0 \n" - "packuswb %%xmm0, %%xmm0 \n" - "movq %%xmm0," MEMACCESS(1) " \n" - "lea " MEMLEA(0x8, 1) ", %1 \n" - "sub $0x8, %2 \n" - "jg 1b \n" - : "+r"(src_argb), // %0 - "+r"(dst_a), // %1 - "+rm"(width) // %2 - : - : "memory", "cc" - , "xmm0", "xmm1" - ); -} -#endif // HAS_ARGBEXTRACTALPHAROW_SSE2 - #ifdef HAS_ARGBCOPYYTOALPHAROW_SSE2 // width in pixels void ARGBCopyYToAlphaRow_SSE2(const uint8* src, uint8* dst, int width) { @@ -3035,7 +2786,7 @@ void ARGBCopyYToAlphaRow_AVX2(const uint8* src, uint8* dst, int width) { #ifdef HAS_SETROW_X86 void SetRow_X86(uint8* dst, uint8 v8, int width) { size_t width_tmp = (size_t)(width >> 2); - const uint32 v32 = v8 * 0x01010101u; // Duplicate byte to all bytes. + const uint32 v32 = v8 * 0x01010101; // Duplicate byte to all bytes. asm volatile ( "rep stosl " MEMSTORESTRING(eax,0) " \n" : "+D"(dst), // %0 @@ -3066,7 +2817,7 @@ void ARGBSetRow_X86(uint8* dst_argb, uint32 v32, int width) { #endif // HAS_SETROW_X86 #ifdef HAS_YUY2TOYROW_SSE2 -void YUY2ToYRow_SSE2(const uint8* src_yuy2, uint8* dst_y, int width) { +void YUY2ToYRow_SSE2(const uint8* src_yuy2, uint8* dst_y, int pix) { asm volatile ( "pcmpeqb %%xmm5,%%xmm5 \n" "psrlw $0x8,%%xmm5 \n" @@ -3084,7 +2835,7 @@ void YUY2ToYRow_SSE2(const uint8* src_yuy2, uint8* dst_y, int width) { "jg 1b \n" : "+r"(src_yuy2), // %0 "+r"(dst_y), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 : : "memory", "cc" , "xmm0", "xmm1", "xmm5" @@ -3092,7 +2843,7 @@ void YUY2ToYRow_SSE2(const uint8* src_yuy2, uint8* dst_y, int width) { } void YUY2ToUVRow_SSE2(const uint8* src_yuy2, int stride_yuy2, - uint8* dst_u, uint8* dst_v, int width) { + uint8* dst_u, uint8* dst_v, int pix) { asm volatile ( "pcmpeqb %%xmm5,%%xmm5 \n" "psrlw $0x8,%%xmm5 \n" @@ -3122,7 +2873,7 @@ void YUY2ToUVRow_SSE2(const uint8* src_yuy2, int stride_yuy2, : "+r"(src_yuy2), // %0 "+r"(dst_u), // %1 "+r"(dst_v), // %2 - "+r"(width) // %3 + "+r"(pix) // %3 : "r"((intptr_t)(stride_yuy2)) // %4 : "memory", "cc", NACL_R14 "xmm0", "xmm1", "xmm2", "xmm3", "xmm5" @@ -3130,7 +2881,7 @@ void YUY2ToUVRow_SSE2(const uint8* src_yuy2, int stride_yuy2, } void YUY2ToUV422Row_SSE2(const uint8* src_yuy2, - uint8* dst_u, uint8* dst_v, int width) { + uint8* dst_u, uint8* dst_v, int pix) { asm volatile ( "pcmpeqb %%xmm5,%%xmm5 \n" "psrlw $0x8,%%xmm5 \n" @@ -3156,14 +2907,14 @@ void YUY2ToUV422Row_SSE2(const uint8* src_yuy2, : "+r"(src_yuy2), // %0 "+r"(dst_u), // %1 "+r"(dst_v), // %2 - "+r"(width) // %3 + "+r"(pix) // %3 : : "memory", "cc", NACL_R14 "xmm0", "xmm1", "xmm5" ); } -void UYVYToYRow_SSE2(const uint8* src_uyvy, uint8* dst_y, int width) { +void UYVYToYRow_SSE2(const uint8* src_uyvy, uint8* dst_y, int pix) { asm volatile ( LABELALIGN "1: \n" @@ -3179,7 +2930,7 @@ void UYVYToYRow_SSE2(const uint8* src_uyvy, uint8* dst_y, int width) { "jg 1b \n" : "+r"(src_uyvy), // %0 "+r"(dst_y), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 : : "memory", "cc" , "xmm0", "xmm1" @@ -3187,7 +2938,7 @@ void UYVYToYRow_SSE2(const uint8* src_uyvy, uint8* dst_y, int width) { } void UYVYToUVRow_SSE2(const uint8* src_uyvy, int stride_uyvy, - uint8* dst_u, uint8* dst_v, int width) { + uint8* dst_u, uint8* dst_v, int pix) { asm volatile ( "pcmpeqb %%xmm5,%%xmm5 \n" "psrlw $0x8,%%xmm5 \n" @@ -3217,7 +2968,7 @@ void UYVYToUVRow_SSE2(const uint8* src_uyvy, int stride_uyvy, : "+r"(src_uyvy), // %0 "+r"(dst_u), // %1 "+r"(dst_v), // %2 - "+r"(width) // %3 + "+r"(pix) // %3 : "r"((intptr_t)(stride_uyvy)) // %4 : "memory", "cc", NACL_R14 "xmm0", "xmm1", "xmm2", "xmm3", "xmm5" @@ -3225,7 +2976,7 @@ void UYVYToUVRow_SSE2(const uint8* src_uyvy, int stride_uyvy, } void UYVYToUV422Row_SSE2(const uint8* src_uyvy, - uint8* dst_u, uint8* dst_v, int width) { + uint8* dst_u, uint8* dst_v, int pix) { asm volatile ( "pcmpeqb %%xmm5,%%xmm5 \n" "psrlw $0x8,%%xmm5 \n" @@ -3251,7 +3002,7 @@ void UYVYToUV422Row_SSE2(const uint8* src_uyvy, : "+r"(src_uyvy), // %0 "+r"(dst_u), // %1 "+r"(dst_v), // %2 - "+r"(width) // %3 + "+r"(pix) // %3 : : "memory", "cc", NACL_R14 "xmm0", "xmm1", "xmm5" @@ -3260,7 +3011,7 @@ void UYVYToUV422Row_SSE2(const uint8* src_uyvy, #endif // HAS_YUY2TOYROW_SSE2 #ifdef HAS_YUY2TOYROW_AVX2 -void YUY2ToYRow_AVX2(const uint8* src_yuy2, uint8* dst_y, int width) { +void YUY2ToYRow_AVX2(const uint8* src_yuy2, uint8* dst_y, int pix) { asm volatile ( "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" "vpsrlw $0x8,%%ymm5,%%ymm5 \n" @@ -3280,7 +3031,7 @@ void YUY2ToYRow_AVX2(const uint8* src_yuy2, uint8* dst_y, int width) { "vzeroupper \n" : "+r"(src_yuy2), // %0 "+r"(dst_y), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 : : "memory", "cc" , "xmm0", "xmm1", "xmm5" @@ -3288,7 +3039,7 @@ void YUY2ToYRow_AVX2(const uint8* src_yuy2, uint8* dst_y, int width) { } void YUY2ToUVRow_AVX2(const uint8* src_yuy2, int stride_yuy2, - uint8* dst_u, uint8* dst_v, int width) { + uint8* dst_u, uint8* dst_v, int pix) { asm volatile ( "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" "vpsrlw $0x8,%%ymm5,%%ymm5 \n" @@ -3319,7 +3070,7 @@ void YUY2ToUVRow_AVX2(const uint8* src_yuy2, int stride_yuy2, : "+r"(src_yuy2), // %0 "+r"(dst_u), // %1 "+r"(dst_v), // %2 - "+r"(width) // %3 + "+r"(pix) // %3 : "r"((intptr_t)(stride_yuy2)) // %4 : "memory", "cc", NACL_R14 "xmm0", "xmm1", "xmm5" @@ -3327,7 +3078,7 @@ void YUY2ToUVRow_AVX2(const uint8* src_yuy2, int stride_yuy2, } void YUY2ToUV422Row_AVX2(const uint8* src_yuy2, - uint8* dst_u, uint8* dst_v, int width) { + uint8* dst_u, uint8* dst_v, int pix) { asm volatile ( "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" "vpsrlw $0x8,%%ymm5,%%ymm5 \n" @@ -3356,14 +3107,14 @@ void YUY2ToUV422Row_AVX2(const uint8* src_yuy2, : "+r"(src_yuy2), // %0 "+r"(dst_u), // %1 "+r"(dst_v), // %2 - "+r"(width) // %3 + "+r"(pix) // %3 : : "memory", "cc", NACL_R14 "xmm0", "xmm1", "xmm5" ); } -void UYVYToYRow_AVX2(const uint8* src_uyvy, uint8* dst_y, int width) { +void UYVYToYRow_AVX2(const uint8* src_uyvy, uint8* dst_y, int pix) { asm volatile ( LABELALIGN "1: \n" @@ -3381,14 +3132,14 @@ void UYVYToYRow_AVX2(const uint8* src_uyvy, uint8* dst_y, int width) { "vzeroupper \n" : "+r"(src_uyvy), // %0 "+r"(dst_y), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 : : "memory", "cc" , "xmm0", "xmm1", "xmm5" ); } void UYVYToUVRow_AVX2(const uint8* src_uyvy, int stride_uyvy, - uint8* dst_u, uint8* dst_v, int width) { + uint8* dst_u, uint8* dst_v, int pix) { asm volatile ( "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" "vpsrlw $0x8,%%ymm5,%%ymm5 \n" @@ -3420,7 +3171,7 @@ void UYVYToUVRow_AVX2(const uint8* src_uyvy, int stride_uyvy, : "+r"(src_uyvy), // %0 "+r"(dst_u), // %1 "+r"(dst_v), // %2 - "+r"(width) // %3 + "+r"(pix) // %3 : "r"((intptr_t)(stride_uyvy)) // %4 : "memory", "cc", NACL_R14 "xmm0", "xmm1", "xmm5" @@ -3428,7 +3179,7 @@ void UYVYToUVRow_AVX2(const uint8* src_uyvy, int stride_uyvy, } void UYVYToUV422Row_AVX2(const uint8* src_uyvy, - uint8* dst_u, uint8* dst_v, int width) { + uint8* dst_u, uint8* dst_v, int pix) { asm volatile ( "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" "vpsrlw $0x8,%%ymm5,%%ymm5 \n" @@ -3457,7 +3208,7 @@ void UYVYToUV422Row_AVX2(const uint8* src_uyvy, : "+r"(src_uyvy), // %0 "+r"(dst_u), // %1 "+r"(dst_v), // %2 - "+r"(width) // %3 + "+r"(pix) // %3 : : "memory", "cc", NACL_R14 "xmm0", "xmm1", "xmm5" @@ -3465,6 +3216,92 @@ void UYVYToUV422Row_AVX2(const uint8* src_uyvy, } #endif // HAS_YUY2TOYROW_AVX2 +#ifdef HAS_ARGBBLENDROW_SSE2 +// Blend 8 pixels at a time. +void ARGBBlendRow_SSE2(const uint8* src_argb0, const uint8* src_argb1, + uint8* dst_argb, int width) { + asm volatile ( + "pcmpeqb %%xmm7,%%xmm7 \n" + "psrlw $0xf,%%xmm7 \n" + "pcmpeqb %%xmm6,%%xmm6 \n" + "psrlw $0x8,%%xmm6 \n" + "pcmpeqb %%xmm5,%%xmm5 \n" + "psllw $0x8,%%xmm5 \n" + "pcmpeqb %%xmm4,%%xmm4 \n" + "pslld $0x18,%%xmm4 \n" + "sub $0x4,%3 \n" + "jl 49f \n" + + // 4 pixel loop. + LABELALIGN + "41: \n" + "movdqu " MEMACCESS(0) ",%%xmm3 \n" + "lea " MEMLEA(0x10,0) ",%0 \n" + "movdqa %%xmm3,%%xmm0 \n" + "pxor %%xmm4,%%xmm3 \n" + "movdqu " MEMACCESS(1) ",%%xmm2 \n" + "psrlw $0x8,%%xmm3 \n" + "pshufhw $0xf5,%%xmm3,%%xmm3 \n" + "pshuflw $0xf5,%%xmm3,%%xmm3 \n" + "pand %%xmm6,%%xmm2 \n" + "paddw %%xmm7,%%xmm3 \n" + "pmullw %%xmm3,%%xmm2 \n" + "movdqu " MEMACCESS(1) ",%%xmm1 \n" + "lea " MEMLEA(0x10,1) ",%1 \n" + "psrlw $0x8,%%xmm1 \n" + "por %%xmm4,%%xmm0 \n" + "pmullw %%xmm3,%%xmm1 \n" + "psrlw $0x8,%%xmm2 \n" + "paddusb %%xmm2,%%xmm0 \n" + "pand %%xmm5,%%xmm1 \n" + "paddusb %%xmm1,%%xmm0 \n" + "movdqu %%xmm0," MEMACCESS(2) " \n" + "lea " MEMLEA(0x10,2) ",%2 \n" + "sub $0x4,%3 \n" + "jge 41b \n" + + "49: \n" + "add $0x3,%3 \n" + "jl 99f \n" + + // 1 pixel loop. + "91: \n" + "movd " MEMACCESS(0) ",%%xmm3 \n" + "lea " MEMLEA(0x4,0) ",%0 \n" + "movdqa %%xmm3,%%xmm0 \n" + "pxor %%xmm4,%%xmm3 \n" + "movd " MEMACCESS(1) ",%%xmm2 \n" + "psrlw $0x8,%%xmm3 \n" + "pshufhw $0xf5,%%xmm3,%%xmm3 \n" + "pshuflw $0xf5,%%xmm3,%%xmm3 \n" + "pand %%xmm6,%%xmm2 \n" + "paddw %%xmm7,%%xmm3 \n" + "pmullw %%xmm3,%%xmm2 \n" + "movd " MEMACCESS(1) ",%%xmm1 \n" + "lea " MEMLEA(0x4,1) ",%1 \n" + "psrlw $0x8,%%xmm1 \n" + "por %%xmm4,%%xmm0 \n" + "pmullw %%xmm3,%%xmm1 \n" + "psrlw $0x8,%%xmm2 \n" + "paddusb %%xmm2,%%xmm0 \n" + "pand %%xmm5,%%xmm1 \n" + "paddusb %%xmm1,%%xmm0 \n" + "movd %%xmm0," MEMACCESS(2) " \n" + "lea " MEMLEA(0x4,2) ",%2 \n" + "sub $0x1,%3 \n" + "jge 91b \n" + "99: \n" + : "+r"(src_argb0), // %0 + "+r"(src_argb1), // %1 + "+r"(dst_argb), // %2 + "+r"(width) // %3 + : + : "memory", "cc" + , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7" + ); +} +#endif // HAS_ARGBBLENDROW_SSE2 + #ifdef HAS_ARGBBLENDROW_SSSE3 // Shuffle table for isolating alpha. static uvec8 kShuffleAlpha = { @@ -3473,6 +3310,15 @@ static uvec8 kShuffleAlpha = { }; // Blend 8 pixels at a time +// Shuffle table for reversing the bytes. + +// Same as SSE2, but replaces +// psrlw xmm3, 8 // alpha +// pshufhw xmm3, xmm3,0F5h // 8 alpha words +// pshuflw xmm3, xmm3,0F5h +// with.. +// pshufb xmm3, kShuffleAlpha // alpha + void ARGBBlendRow_SSSE3(const uint8* src_argb0, const uint8* src_argb1, uint8* dst_argb, int width) { asm volatile ( @@ -3553,112 +3399,49 @@ void ARGBBlendRow_SSSE3(const uint8* src_argb0, const uint8* src_argb1, } #endif // HAS_ARGBBLENDROW_SSSE3 -#ifdef HAS_BLENDPLANEROW_SSSE3 -// Blend 8 pixels at a time. -// unsigned version of math -// =((A2*C2)+(B2*(255-C2))+255)/256 -// signed version of math -// =(((A2-128)*C2)+((B2-128)*(255-C2))+32768+127)/256 -void BlendPlaneRow_SSSE3(const uint8* src0, const uint8* src1, - const uint8* alpha, uint8* dst, int width) { +#ifdef HAS_ARGBATTENUATEROW_SSE2 +// Attenuate 4 pixels at a time. +void ARGBAttenuateRow_SSE2(const uint8* src_argb, uint8* dst_argb, int width) { asm volatile ( - "pcmpeqb %%xmm5,%%xmm5 \n" - "psllw $0x8,%%xmm5 \n" - "mov $0x80808080,%%eax \n" - "movd %%eax,%%xmm6 \n" - "pshufd $0x0,%%xmm6,%%xmm6 \n" - "mov $0x807f807f,%%eax \n" - "movd %%eax,%%xmm7 \n" - "pshufd $0x0,%%xmm7,%%xmm7 \n" - "sub %2,%0 \n" - "sub %2,%1 \n" - "sub %2,%3 \n" + "pcmpeqb %%xmm4,%%xmm4 \n" + "pslld $0x18,%%xmm4 \n" + "pcmpeqb %%xmm5,%%xmm5 \n" + "psrld $0x8,%%xmm5 \n" - // 8 pixel loop. - LABELALIGN - "1: \n" - "movq (%2),%%xmm0 \n" - "punpcklbw %%xmm0,%%xmm0 \n" - "pxor %%xmm5,%%xmm0 \n" - "movq (%0,%2,1),%%xmm1 \n" - "movq (%1,%2,1),%%xmm2 \n" - "punpcklbw %%xmm2,%%xmm1 \n" - "psubb %%xmm6,%%xmm1 \n" - "pmaddubsw %%xmm1,%%xmm0 \n" - "paddw %%xmm7,%%xmm0 \n" - "psrlw $0x8,%%xmm0 \n" - "packuswb %%xmm0,%%xmm0 \n" - "movq %%xmm0,(%3,%2,1) \n" - "lea 0x8(%2),%2 \n" - "sub $0x8,%4 \n" - "jg 1b \n" - : "+r"(src0), // %0 - "+r"(src1), // %1 - "+r"(alpha), // %2 - "+r"(dst), // %3 - "+rm"(width) // %4 - :: "memory", "cc", "eax", "xmm0", "xmm1", "xmm2", "xmm5", "xmm6", "xmm7" - ); -} -#endif // HAS_BLENDPLANEROW_SSSE3 - -#ifdef HAS_BLENDPLANEROW_AVX2 -// Blend 32 pixels at a time. -// unsigned version of math -// =((A2*C2)+(B2*(255-C2))+255)/256 -// signed version of math -// =(((A2-128)*C2)+((B2-128)*(255-C2))+32768+127)/256 -void BlendPlaneRow_AVX2(const uint8* src0, const uint8* src1, - const uint8* alpha, uint8* dst, int width) { - asm volatile ( - "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" - "vpsllw $0x8,%%ymm5,%%ymm5 \n" - "mov $0x80808080,%%eax \n" - "vmovd %%eax,%%xmm6 \n" - "vbroadcastss %%xmm6,%%ymm6 \n" - "mov $0x807f807f,%%eax \n" - "vmovd %%eax,%%xmm7 \n" - "vbroadcastss %%xmm7,%%ymm7 \n" - "sub %2,%0 \n" - "sub %2,%1 \n" - "sub %2,%3 \n" - - // 32 pixel loop. + // 4 pixel loop. LABELALIGN "1: \n" - "vmovdqu (%2),%%ymm0 \n" - "vpunpckhbw %%ymm0,%%ymm0,%%ymm3 \n" - "vpunpcklbw %%ymm0,%%ymm0,%%ymm0 \n" - "vpxor %%ymm5,%%ymm3,%%ymm3 \n" - "vpxor %%ymm5,%%ymm0,%%ymm0 \n" - "vmovdqu (%0,%2,1),%%ymm1 \n" - "vmovdqu (%1,%2,1),%%ymm2 \n" - "vpunpckhbw %%ymm2,%%ymm1,%%ymm4 \n" - "vpunpcklbw %%ymm2,%%ymm1,%%ymm1 \n" - "vpsubb %%ymm6,%%ymm4,%%ymm4 \n" - "vpsubb %%ymm6,%%ymm1,%%ymm1 \n" - "vpmaddubsw %%ymm4,%%ymm3,%%ymm3 \n" - "vpmaddubsw %%ymm1,%%ymm0,%%ymm0 \n" - "vpaddw %%ymm7,%%ymm3,%%ymm3 \n" - "vpaddw %%ymm7,%%ymm0,%%ymm0 \n" - "vpsrlw $0x8,%%ymm3,%%ymm3 \n" - "vpsrlw $0x8,%%ymm0,%%ymm0 \n" - "vpackuswb %%ymm3,%%ymm0,%%ymm0 \n" - "vmovdqu %%ymm0,(%3,%2,1) \n" - "lea 0x20(%2),%2 \n" - "sub $0x20,%4 \n" + "movdqu " MEMACCESS(0) ",%%xmm0 \n" + "punpcklbw %%xmm0,%%xmm0 \n" + "pshufhw $0xff,%%xmm0,%%xmm2 \n" + "pshuflw $0xff,%%xmm2,%%xmm2 \n" + "pmulhuw %%xmm2,%%xmm0 \n" + "movdqu " MEMACCESS(0) ",%%xmm1 \n" + "punpckhbw %%xmm1,%%xmm1 \n" + "pshufhw $0xff,%%xmm1,%%xmm2 \n" + "pshuflw $0xff,%%xmm2,%%xmm2 \n" + "pmulhuw %%xmm2,%%xmm1 \n" + "movdqu " MEMACCESS(0) ",%%xmm2 \n" + "lea " MEMLEA(0x10,0) ",%0 \n" + "psrlw $0x8,%%xmm0 \n" + "pand %%xmm4,%%xmm2 \n" + "psrlw $0x8,%%xmm1 \n" + "packuswb %%xmm1,%%xmm0 \n" + "pand %%xmm5,%%xmm0 \n" + "por %%xmm2,%%xmm0 \n" + "movdqu %%xmm0," MEMACCESS(1) " \n" + "lea " MEMLEA(0x10,1) ",%1 \n" + "sub $0x4,%2 \n" "jg 1b \n" - "vzeroupper \n" - : "+r"(src0), // %0 - "+r"(src1), // %1 - "+r"(alpha), // %2 - "+r"(dst), // %3 - "+rm"(width) // %4 - :: "memory", "cc", "eax", - "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7" + : "+r"(src_argb), // %0 + "+r"(dst_argb), // %1 + "+r"(width) // %2 + : + : "memory", "cc" + , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" ); } -#endif // HAS_BLENDPLANEROW_AVX2 +#endif // HAS_ARGBATTENUATEROW_SSE2 #ifdef HAS_ARGBATTENUATEROW_SSSE3 // Shuffle table duplicating alpha @@ -3759,7 +3542,7 @@ void ARGBAttenuateRow_AVX2(const uint8* src_argb, uint8* dst_argb, int width) { // Unattenuate 4 pixels at a time. void ARGBUnattenuateRow_SSE2(const uint8* src_argb, uint8* dst_argb, int width) { - uintptr_t alpha; + uintptr_t alpha = 0; asm volatile ( // 4 pixel loop. LABELALIGN @@ -3790,10 +3573,10 @@ void ARGBUnattenuateRow_SSE2(const uint8* src_argb, uint8* dst_argb, "lea " MEMLEA(0x10,1) ",%1 \n" "sub $0x4,%2 \n" "jg 1b \n" - : "+r"(src_argb), // %0 - "+r"(dst_argb), // %1 - "+r"(width), // %2 - "=&r"(alpha) // %3 + : "+r"(src_argb), // %0 + "+r"(dst_argb), // %1 + "+r"(width), // %2 + "+r"(alpha) // %3 : "r"(fixed_invtbl8) // %4 : "memory", "cc", NACL_R14 "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" @@ -3809,7 +3592,7 @@ static const uvec8 kUnattenShuffleAlpha_AVX2 = { // Unattenuate 8 pixels at a time. void ARGBUnattenuateRow_AVX2(const uint8* src_argb, uint8* dst_argb, int width) { - uintptr_t alpha; + uintptr_t alpha = 0; asm volatile ( "sub %0,%1 \n" "vbroadcastf128 %5,%%ymm5 \n" @@ -3858,10 +3641,10 @@ void ARGBUnattenuateRow_AVX2(const uint8* src_argb, uint8* dst_argb, "sub $0x8,%2 \n" "jg 1b \n" "vzeroupper \n" - : "+r"(src_argb), // %0 - "+r"(dst_argb), // %1 - "+r"(width), // %2 - "=&r"(alpha) // %3 + : "+r"(src_argb), // %0 + "+r"(dst_argb), // %1 + "+r"(width), // %2 + "+r"(alpha) // %3 : "r"(fixed_invtbl8), // %4 "m"(kUnattenShuffleAlpha_AVX2) // %5 : "memory", "cc", NACL_R14 @@ -4786,7 +4569,7 @@ LIBYUV_API void ARGBAffineRow_SSE2(const uint8* src_argb, int src_argb_stride, uint8* dst_argb, const float* src_dudv, int width) { intptr_t src_argb_stride_temp = src_argb_stride; - intptr_t temp; + intptr_t temp = 0; asm volatile ( "movq " MEMACCESS(3) ",%%xmm2 \n" "movq " MEMACCESS2(0x08,3) ",%%xmm7 \n" @@ -4858,7 +4641,7 @@ void ARGBAffineRow_SSE2(const uint8* src_argb, int src_argb_stride, "+r"(dst_argb), // %2 "+r"(src_dudv), // %3 "+rm"(width), // %4 - "=&r"(temp) // %5 + "+r"(temp) // %5 : : "memory", "cc", NACL_R14 "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7" @@ -4873,47 +4656,56 @@ void InterpolateRow_SSSE3(uint8* dst_ptr, const uint8* src_ptr, int source_y_fraction) { asm volatile ( "sub %1,%0 \n" + "shr %3 \n" "cmp $0x0,%3 \n" "je 100f \n" - "cmp $0x80,%3 \n" + "cmp $0x20,%3 \n" + "je 75f \n" + "cmp $0x40,%3 \n" "je 50f \n" + "cmp $0x60,%3 \n" + "je 25f \n" "movd %3,%%xmm0 \n" "neg %3 \n" - "add $0x100,%3 \n" + "add $0x80,%3 \n" "movd %3,%%xmm5 \n" "punpcklbw %%xmm0,%%xmm5 \n" "punpcklwd %%xmm5,%%xmm5 \n" "pshufd $0x0,%%xmm5,%%xmm5 \n" - "mov $0x80808080,%%eax \n" - "movd %%eax,%%xmm4 \n" - "pshufd $0x0,%%xmm4,%%xmm4 \n" // General purpose row blend. LABELALIGN "1: \n" "movdqu " MEMACCESS(1) ",%%xmm0 \n" MEMOPREG(movdqu,0x00,1,4,1,xmm2) - "movdqa %%xmm0,%%xmm1 \n" - "punpcklbw %%xmm2,%%xmm0 \n" - "punpckhbw %%xmm2,%%xmm1 \n" - "psubb %%xmm4,%%xmm0 \n" - "psubb %%xmm4,%%xmm1 \n" - "movdqa %%xmm5,%%xmm2 \n" - "movdqa %%xmm5,%%xmm3 \n" - "pmaddubsw %%xmm0,%%xmm2 \n" - "pmaddubsw %%xmm1,%%xmm3 \n" - "paddw %%xmm4,%%xmm2 \n" - "paddw %%xmm4,%%xmm3 \n" - "psrlw $0x8,%%xmm2 \n" - "psrlw $0x8,%%xmm3 \n" - "packuswb %%xmm3,%%xmm2 \n" - MEMOPMEM(movdqu,xmm2,0x00,1,0,1) + "movdqa %%xmm0,%%xmm1 \n" + "punpcklbw %%xmm2,%%xmm0 \n" + "punpckhbw %%xmm2,%%xmm1 \n" + "pmaddubsw %%xmm5,%%xmm0 \n" + "pmaddubsw %%xmm5,%%xmm1 \n" + "psrlw $0x7,%%xmm0 \n" + "psrlw $0x7,%%xmm1 \n" + "packuswb %%xmm1,%%xmm0 \n" + MEMOPMEM(movdqu,xmm0,0x00,1,0,1) "lea " MEMLEA(0x10,1) ",%1 \n" "sub $0x10,%2 \n" "jg 1b \n" "jmp 99f \n" + // Blend 25 / 75. + LABELALIGN + "25: \n" + "movdqu " MEMACCESS(1) ",%%xmm0 \n" + MEMOPREG(movdqu,0x00,1,4,1,xmm1) + "pavgb %%xmm1,%%xmm0 \n" + "pavgb %%xmm1,%%xmm0 \n" + MEMOPMEM(movdqu,xmm0,0x00,1,0,1) + "lea " MEMLEA(0x10,1) ",%1 \n" + "sub $0x10,%2 \n" + "jg 25b \n" + "jmp 99f \n" + // Blend 50 / 50. LABELALIGN "50: \n" @@ -4926,6 +4718,19 @@ void InterpolateRow_SSSE3(uint8* dst_ptr, const uint8* src_ptr, "jg 50b \n" "jmp 99f \n" + // Blend 75 / 25. + LABELALIGN + "75: \n" + "movdqu " MEMACCESS(1) ",%%xmm1 \n" + MEMOPREG(movdqu,0x00,1,4,1,xmm0) + "pavgb %%xmm1,%%xmm0 \n" + "pavgb %%xmm1,%%xmm0 \n" + MEMOPMEM(movdqu,xmm0,0x00,1,0,1) + "lea " MEMLEA(0x10,1) ",%1 \n" + "sub $0x10,%2 \n" + "jg 75b \n" + "jmp 99f \n" + // Blend 100 / 0 - Copy row unchanged. LABELALIGN "100: \n" @@ -4936,13 +4741,13 @@ void InterpolateRow_SSSE3(uint8* dst_ptr, const uint8* src_ptr, "jg 100b \n" "99: \n" - : "+r"(dst_ptr), // %0 - "+r"(src_ptr), // %1 - "+rm"(dst_width), // %2 + : "+r"(dst_ptr), // %0 + "+r"(src_ptr), // %1 + "+r"(dst_width), // %2 "+r"(source_y_fraction) // %3 : "r"((intptr_t)(src_stride)) // %4 - : "memory", "cc", "eax", NACL_R14 - "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" + : "memory", "cc", NACL_R14 + "xmm0", "xmm1", "xmm2", "xmm5" ); } #endif // HAS_INTERPOLATEROW_SSSE3 @@ -4953,22 +4758,25 @@ void InterpolateRow_AVX2(uint8* dst_ptr, const uint8* src_ptr, ptrdiff_t src_stride, int dst_width, int source_y_fraction) { asm volatile ( + "shr %3 \n" "cmp $0x0,%3 \n" "je 100f \n" "sub %1,%0 \n" - "cmp $0x80,%3 \n" + "cmp $0x20,%3 \n" + "je 75f \n" + "cmp $0x40,%3 \n" "je 50f \n" + "cmp $0x60,%3 \n" + "je 25f \n" "vmovd %3,%%xmm0 \n" "neg %3 \n" - "add $0x100,%3 \n" + "add $0x80,%3 \n" "vmovd %3,%%xmm5 \n" "vpunpcklbw %%xmm0,%%xmm5,%%xmm5 \n" "vpunpcklwd %%xmm5,%%xmm5,%%xmm5 \n" - "vbroadcastss %%xmm5,%%ymm5 \n" - "mov $0x80808080,%%eax \n" - "vmovd %%eax,%%xmm4 \n" - "vbroadcastss %%xmm4,%%ymm4 \n" + "vpxor %%ymm0,%%ymm0,%%ymm0 \n" + "vpermd %%ymm5,%%ymm0,%%ymm5 \n" // General purpose row blend. LABELALIGN @@ -4977,14 +4785,10 @@ void InterpolateRow_AVX2(uint8* dst_ptr, const uint8* src_ptr, MEMOPREG(vmovdqu,0x00,1,4,1,ymm2) "vpunpckhbw %%ymm2,%%ymm0,%%ymm1 \n" "vpunpcklbw %%ymm2,%%ymm0,%%ymm0 \n" - "vpsubb %%ymm4,%%ymm1,%%ymm1 \n" - "vpsubb %%ymm4,%%ymm0,%%ymm0 \n" - "vpmaddubsw %%ymm1,%%ymm5,%%ymm1 \n" - "vpmaddubsw %%ymm0,%%ymm5,%%ymm0 \n" - "vpaddw %%ymm4,%%ymm1,%%ymm1 \n" - "vpaddw %%ymm4,%%ymm0,%%ymm0 \n" - "vpsrlw $0x8,%%ymm1,%%ymm1 \n" - "vpsrlw $0x8,%%ymm0,%%ymm0 \n" + "vpmaddubsw %%ymm5,%%ymm0,%%ymm0 \n" + "vpmaddubsw %%ymm5,%%ymm1,%%ymm1 \n" + "vpsrlw $0x7,%%ymm0,%%ymm0 \n" + "vpsrlw $0x7,%%ymm1,%%ymm1 \n" "vpackuswb %%ymm1,%%ymm0,%%ymm0 \n" MEMOPMEM(vmovdqu,ymm0,0x00,1,0,1) "lea " MEMLEA(0x20,1) ",%1 \n" @@ -4992,6 +4796,19 @@ void InterpolateRow_AVX2(uint8* dst_ptr, const uint8* src_ptr, "jg 1b \n" "jmp 99f \n" + // Blend 25 / 75. + LABELALIGN + "25: \n" + "vmovdqu " MEMACCESS(1) ",%%ymm0 \n" + MEMOPREG(vmovdqu,0x00,1,4,1,ymm1) + "vpavgb %%ymm1,%%ymm0,%%ymm0 \n" + "vpavgb %%ymm1,%%ymm0,%%ymm0 \n" + MEMOPMEM(vmovdqu,ymm0,0x00,1,0,1) + "lea " MEMLEA(0x20,1) ",%1 \n" + "sub $0x20,%2 \n" + "jg 25b \n" + "jmp 99f \n" + // Blend 50 / 50. LABELALIGN "50: \n" @@ -5003,6 +4820,19 @@ void InterpolateRow_AVX2(uint8* dst_ptr, const uint8* src_ptr, "jg 50b \n" "jmp 99f \n" + // Blend 75 / 25. + LABELALIGN + "75: \n" + "vmovdqu " MEMACCESS(1) ",%%ymm1 \n" + MEMOPREG(vmovdqu,0x00,1,4,1,ymm0) + "vpavgb %%ymm1,%%ymm0,%%ymm0 \n" + "vpavgb %%ymm1,%%ymm0,%%ymm0 \n" + MEMOPMEM(vmovdqu,ymm0,0x00,1,0,1) + "lea " MEMLEA(0x20,1) ",%1 \n" + "sub $0x20,%2 \n" + "jg 75b \n" + "jmp 99f \n" + // Blend 100 / 0 - Copy row unchanged. LABELALIGN "100: \n" @@ -5014,19 +4844,130 @@ void InterpolateRow_AVX2(uint8* dst_ptr, const uint8* src_ptr, "999: \n" : "+D"(dst_ptr), // %0 "+S"(src_ptr), // %1 - "+cm"(dst_width), // %2 + "+c"(dst_width), // %2 "+r"(source_y_fraction) // %3 : "r"((intptr_t)(src_stride)) // %4 - : "memory", "cc", "eax", NACL_R14 - "xmm0", "xmm1", "xmm2", "xmm4", "xmm5" + : "memory", "cc", NACL_R14 + "xmm0", "xmm1", "xmm2", "xmm5" ); } #endif // HAS_INTERPOLATEROW_AVX2 +#ifdef HAS_INTERPOLATEROW_SSE2 +// Bilinear filter 16x2 -> 16x1 +void InterpolateRow_SSE2(uint8* dst_ptr, const uint8* src_ptr, + ptrdiff_t src_stride, int dst_width, + int source_y_fraction) { + asm volatile ( + "sub %1,%0 \n" + "shr %3 \n" + "cmp $0x0,%3 \n" + "je 100f \n" + "cmp $0x20,%3 \n" + "je 75f \n" + "cmp $0x40,%3 \n" + "je 50f \n" + "cmp $0x60,%3 \n" + "je 25f \n" + + "movd %3,%%xmm0 \n" + "neg %3 \n" + "add $0x80,%3 \n" + "movd %3,%%xmm5 \n" + "punpcklbw %%xmm0,%%xmm5 \n" + "punpcklwd %%xmm5,%%xmm5 \n" + "pshufd $0x0,%%xmm5,%%xmm5 \n" + "pxor %%xmm4,%%xmm4 \n" + + // General purpose row blend. + LABELALIGN + "1: \n" + "movdqu " MEMACCESS(1) ",%%xmm0 \n" + MEMOPREG(movdqu,0x00,1,4,1,xmm2) // movdqu (%1,%4,1),%%xmm2 + "movdqa %%xmm0,%%xmm1 \n" + "movdqa %%xmm2,%%xmm3 \n" + "punpcklbw %%xmm4,%%xmm2 \n" + "punpckhbw %%xmm4,%%xmm3 \n" + "punpcklbw %%xmm4,%%xmm0 \n" + "punpckhbw %%xmm4,%%xmm1 \n" + "psubw %%xmm0,%%xmm2 \n" + "psubw %%xmm1,%%xmm3 \n" + "paddw %%xmm2,%%xmm2 \n" + "paddw %%xmm3,%%xmm3 \n" + "pmulhw %%xmm5,%%xmm2 \n" + "pmulhw %%xmm5,%%xmm3 \n" + "paddw %%xmm2,%%xmm0 \n" + "paddw %%xmm3,%%xmm1 \n" + "packuswb %%xmm1,%%xmm0 \n" + MEMOPMEM(movdqu,xmm0,0x00,1,0,1) // movdqu %%xmm0,(%1,%0,1) + "lea " MEMLEA(0x10,1) ",%1 \n" + "sub $0x10,%2 \n" + "jg 1b \n" + "jmp 99f \n" + + // Blend 25 / 75. + LABELALIGN + "25: \n" + "movdqu " MEMACCESS(1) ",%%xmm0 \n" + MEMOPREG(movdqu,0x00,1,4,1,xmm1) // movdqu (%1,%4,1),%%xmm1 + "pavgb %%xmm1,%%xmm0 \n" + "pavgb %%xmm1,%%xmm0 \n" + MEMOPMEM(movdqu,xmm0,0x00,1,0,1) // movdqu %%xmm0,(%1,%0,1) + "lea " MEMLEA(0x10,1) ",%1 \n" + "sub $0x10,%2 \n" + "jg 25b \n" + "jmp 99f \n" + + // Blend 50 / 50. + LABELALIGN + "50: \n" + "movdqu " MEMACCESS(1) ",%%xmm0 \n" + MEMOPREG(movdqu,0x00,1,4,1,xmm1) // movdqu (%1,%4,1),%%xmm1 + "pavgb %%xmm1,%%xmm0 \n" + MEMOPMEM(movdqu,xmm0,0x00,1,0,1) // movdqu %%xmm0,(%1,%0,1) + "lea " MEMLEA(0x10,1) ",%1 \n" + "sub $0x10,%2 \n" + "jg 50b \n" + "jmp 99f \n" + + // Blend 75 / 25. + LABELALIGN + "75: \n" + "movdqu " MEMACCESS(1) ",%%xmm1 \n" + MEMOPREG(movdqu,0x00,1,4,1,xmm0) // movdqu (%1,%4,1),%%xmm0 + "pavgb %%xmm1,%%xmm0 \n" + "pavgb %%xmm1,%%xmm0 \n" + MEMOPMEM(movdqu,xmm0,0x00,1,0,1) // movdqu %%xmm0,(%1,%0,1) + "lea " MEMLEA(0x10,1) ",%1 \n" + "sub $0x10,%2 \n" + "jg 75b \n" + "jmp 99f \n" + + // Blend 100 / 0 - Copy row unchanged. + LABELALIGN + "100: \n" + "movdqu " MEMACCESS(1) ",%%xmm0 \n" + MEMOPMEM(movdqu,xmm0,0x00,1,0,1) // movdqu %%xmm0,(%1,%0,1) + "lea " MEMLEA(0x10,1) ",%1 \n" + "sub $0x10,%2 \n" + "jg 100b \n" + + "99: \n" + : "+r"(dst_ptr), // %0 + "+r"(src_ptr), // %1 + "+r"(dst_width), // %2 + "+r"(source_y_fraction) // %3 + : "r"((intptr_t)(src_stride)) // %4 + : "memory", "cc", NACL_R14 + "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" + ); +} +#endif // HAS_INTERPOLATEROW_SSE2 + #ifdef HAS_ARGBSHUFFLEROW_SSSE3 // For BGRAToARGB, ABGRToARGB, RGBAToARGB, and ARGBToRGBA. void ARGBShuffleRow_SSSE3(const uint8* src_argb, uint8* dst_argb, - const uint8* shuffler, int width) { + const uint8* shuffler, int pix) { asm volatile ( "movdqu " MEMACCESS(3) ",%%xmm5 \n" LABELALIGN @@ -5043,7 +4984,7 @@ void ARGBShuffleRow_SSSE3(const uint8* src_argb, uint8* dst_argb, "jg 1b \n" : "+r"(src_argb), // %0 "+r"(dst_argb), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 : "r"(shuffler) // %3 : "memory", "cc" , "xmm0", "xmm1", "xmm5" @@ -5054,7 +4995,7 @@ void ARGBShuffleRow_SSSE3(const uint8* src_argb, uint8* dst_argb, #ifdef HAS_ARGBSHUFFLEROW_AVX2 // For BGRAToARGB, ABGRToARGB, RGBAToARGB, and ARGBToRGBA. void ARGBShuffleRow_AVX2(const uint8* src_argb, uint8* dst_argb, - const uint8* shuffler, int width) { + const uint8* shuffler, int pix) { asm volatile ( "vbroadcastf128 " MEMACCESS(3) ",%%ymm5 \n" LABELALIGN @@ -5072,7 +5013,7 @@ void ARGBShuffleRow_AVX2(const uint8* src_argb, uint8* dst_argb, "vzeroupper \n" : "+r"(src_argb), // %0 "+r"(dst_argb), // %1 - "+r"(width) // %2 + "+r"(pix) // %2 : "r"(shuffler) // %3 : "memory", "cc" , "xmm0", "xmm1", "xmm5" @@ -5083,8 +5024,8 @@ void ARGBShuffleRow_AVX2(const uint8* src_argb, uint8* dst_argb, #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; + const uint8* shuffler, int pix) { + uintptr_t pixel_temp = 0u; asm volatile ( "pxor %%xmm5,%%xmm5 \n" "mov " MEMACCESS(4) ",%k2 \n" @@ -5189,11 +5130,11 @@ void ARGBShuffleRow_SSE2(const uint8* src_argb, uint8* dst_argb, "jg 3012b \n" "99: \n" - : "+r"(src_argb), // %0 - "+r"(dst_argb), // %1 - "=&d"(pixel_temp), // %2 - "+r"(width) // %3 - : "r"(shuffler) // %4 + : "+r"(src_argb), // %0 + "+r"(dst_argb), // %1 + "+d"(pixel_temp), // %2 + "+r"(pix) // %3 + : "r"(shuffler) // %4 : "memory", "cc", NACL_R14 "xmm0", "xmm1", "xmm5" ); @@ -5370,7 +5311,7 @@ void ARGBPolynomialRow_AVX2(const uint8* src_argb, // Tranform ARGB pixels with color table. void ARGBColorTableRow_X86(uint8* dst_argb, const uint8* table_argb, int width) { - uintptr_t pixel_temp; + uintptr_t pixel_temp = 0u; asm volatile ( // 1 pixel loop. LABELALIGN @@ -5390,10 +5331,10 @@ void ARGBColorTableRow_X86(uint8* dst_argb, const uint8* table_argb, "mov %b1," MEMACCESS2(-0x1,0) " \n" "dec %2 \n" "jg 1b \n" - : "+r"(dst_argb), // %0 - "=&d"(pixel_temp), // %1 - "+r"(width) // %2 - : "r"(table_argb) // %3 + : "+r"(dst_argb), // %0 + "+d"(pixel_temp), // %1 + "+r"(width) // %2 + : "r"(table_argb) // %3 : "memory", "cc"); } #endif // HAS_ARGBCOLORTABLEROW_X86 @@ -5401,7 +5342,7 @@ void ARGBColorTableRow_X86(uint8* dst_argb, const uint8* table_argb, #ifdef HAS_RGBCOLORTABLEROW_X86 // Tranform RGB pixels with color table. void RGBColorTableRow_X86(uint8* dst_argb, const uint8* table_argb, int width) { - uintptr_t pixel_temp; + uintptr_t pixel_temp = 0u; asm volatile ( // 1 pixel loop. LABELALIGN @@ -5418,10 +5359,10 @@ void RGBColorTableRow_X86(uint8* dst_argb, const uint8* table_argb, int width) { "mov %b1," MEMACCESS2(-0x2,0) " \n" "dec %2 \n" "jg 1b \n" - : "+r"(dst_argb), // %0 - "=&d"(pixel_temp), // %1 - "+r"(width) // %2 - : "r"(table_argb) // %3 + : "+r"(dst_argb), // %0 + "+d"(pixel_temp), // %1 + "+r"(width) // %2 + : "r"(table_argb) // %3 : "memory", "cc"); } #endif // HAS_RGBCOLORTABLEROW_X86 @@ -5431,8 +5372,8 @@ void RGBColorTableRow_X86(uint8* dst_argb, const uint8* table_argb, int width) { void ARGBLumaColorTableRow_SSSE3(const uint8* src_argb, uint8* dst_argb, int width, const uint8* luma, uint32 lumacoeff) { - uintptr_t pixel_temp; - uintptr_t table_temp; + uintptr_t pixel_temp = 0u; + uintptr_t table_temp = 0u; asm volatile ( "movd %6,%%xmm3 \n" "pshufd $0x0,%%xmm3,%%xmm3 \n" @@ -5514,13 +5455,13 @@ void ARGBLumaColorTableRow_SSSE3(const uint8* src_argb, uint8* dst_argb, "lea " MEMLEA(0x10,3) ",%3 \n" "sub $0x4,%4 \n" "jg 1b \n" - : "=&d"(pixel_temp), // %0 - "=&a"(table_temp), // %1 - "+r"(src_argb), // %2 - "+r"(dst_argb), // %3 - "+rm"(width) // %4 - : "r"(luma), // %5 - "rm"(lumacoeff) // %6 + : "+d"(pixel_temp), // %0 + "+a"(table_temp), // %1 + "+r"(src_argb), // %2 + "+r"(dst_argb), // %3 + "+rm"(width) // %4 + : "r"(luma), // %5 + "rm"(lumacoeff) // %6 : "memory", "cc", "xmm0", "xmm3", "xmm4", "xmm5" ); } |