diff options
Diffstat (limited to 'media/libyuv/source/compare_win.cc')
-rw-r--r-- | media/libyuv/source/compare_win.cc | 131 |
1 files changed, 75 insertions, 56 deletions
diff --git a/media/libyuv/source/compare_win.cc b/media/libyuv/source/compare_win.cc index dc86fe25b1..9bb27f1dd1 100644 --- a/media/libyuv/source/compare_win.cc +++ b/media/libyuv/source/compare_win.cc @@ -13,20 +13,40 @@ #include "libyuv/compare_row.h" #include "libyuv/row.h" +#if defined(_MSC_VER) +#include <intrin.h> // For __popcnt +#endif + #ifdef __cplusplus namespace libyuv { extern "C" { #endif -// This module is for 32 bit Visual C x86 and clangcl -#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) +// This module is for 32 bit Visual C x86 +#if !defined(LIBYUV_DISABLE_X86) && defined(_MSC_VER) && \ + !defined(__clang__) && defined(_M_IX86) + +uint32_t HammingDistance_SSE42(const uint8_t* src_a, + const uint8_t* src_b, + int count) { + uint32_t diff = 0u; + + int i; + for (i = 0; i < count - 3; i += 4) { + uint32_t x = *((uint32_t*)src_a) ^ *((uint32_t*)src_b); // NOLINT + src_a += 4; + src_b += 4; + diff += __popcnt(x); + } + return diff; +} -__declspec(naked) -uint32 SumSquareError_SSE2(const uint8* src_a, const uint8* src_b, int count) { +__declspec(naked) uint32_t + SumSquareError_SSE2(const uint8_t* src_a, const uint8_t* src_b, int count) { __asm { - mov eax, [esp + 4] // src_a - mov edx, [esp + 8] // src_b - mov ecx, [esp + 12] // count + mov eax, [esp + 4] // src_a + mov edx, [esp + 8] // src_b + mov ecx, [esp + 12] // count pxor xmm0, xmm0 pxor xmm5, xmm5 @@ -58,16 +78,15 @@ uint32 SumSquareError_SSE2(const uint8* src_a, const uint8* src_b, int count) { } } -// Visual C 2012 required for AVX2. -#if _MSC_VER >= 1700 +#ifdef HAS_SUMSQUAREERROR_AVX2 // C4752: found Intel(R) Advanced Vector Extensions; consider using /arch:AVX. -#pragma warning(disable: 4752) -__declspec(naked) -uint32 SumSquareError_AVX2(const uint8* src_a, const uint8* src_b, int count) { +#pragma warning(disable : 4752) +__declspec(naked) uint32_t + SumSquareError_AVX2(const uint8_t* src_a, const uint8_t* src_b, int count) { __asm { - mov eax, [esp + 4] // src_a - mov edx, [esp + 8] // src_b - mov ecx, [esp + 12] // count + mov eax, [esp + 4] // src_a + mov edx, [esp + 8] // src_b + mov ecx, [esp + 12] // count vpxor ymm0, ymm0, ymm0 // sum vpxor ymm5, ymm5, ymm5 // constant 0 for unpck sub edx, eax @@ -99,67 +118,67 @@ uint32 SumSquareError_AVX2(const uint8* src_a, const uint8* src_b, int count) { ret } } -#endif // _MSC_VER >= 1700 +#endif // HAS_SUMSQUAREERROR_AVX2 -uvec32 kHash16x33 = { 0x92d9e201, 0, 0, 0 }; // 33 ^ 16 +uvec32 kHash16x33 = {0x92d9e201, 0, 0, 0}; // 33 ^ 16 uvec32 kHashMul0 = { - 0x0c3525e1, // 33 ^ 15 - 0xa3476dc1, // 33 ^ 14 - 0x3b4039a1, // 33 ^ 13 - 0x4f5f0981, // 33 ^ 12 + 0x0c3525e1, // 33 ^ 15 + 0xa3476dc1, // 33 ^ 14 + 0x3b4039a1, // 33 ^ 13 + 0x4f5f0981, // 33 ^ 12 }; uvec32 kHashMul1 = { - 0x30f35d61, // 33 ^ 11 - 0x855cb541, // 33 ^ 10 - 0x040a9121, // 33 ^ 9 - 0x747c7101, // 33 ^ 8 + 0x30f35d61, // 33 ^ 11 + 0x855cb541, // 33 ^ 10 + 0x040a9121, // 33 ^ 9 + 0x747c7101, // 33 ^ 8 }; uvec32 kHashMul2 = { - 0xec41d4e1, // 33 ^ 7 - 0x4cfa3cc1, // 33 ^ 6 - 0x025528a1, // 33 ^ 5 - 0x00121881, // 33 ^ 4 + 0xec41d4e1, // 33 ^ 7 + 0x4cfa3cc1, // 33 ^ 6 + 0x025528a1, // 33 ^ 5 + 0x00121881, // 33 ^ 4 }; uvec32 kHashMul3 = { - 0x00008c61, // 33 ^ 3 - 0x00000441, // 33 ^ 2 - 0x00000021, // 33 ^ 1 - 0x00000001, // 33 ^ 0 + 0x00008c61, // 33 ^ 3 + 0x00000441, // 33 ^ 2 + 0x00000021, // 33 ^ 1 + 0x00000001, // 33 ^ 0 }; -__declspec(naked) -uint32 HashDjb2_SSE41(const uint8* src, int count, uint32 seed) { +__declspec(naked) uint32_t + HashDjb2_SSE41(const uint8_t* src, int count, uint32_t seed) { __asm { - mov eax, [esp + 4] // src - mov ecx, [esp + 8] // count + mov eax, [esp + 4] // src + mov ecx, [esp + 8] // count movd xmm0, [esp + 12] // seed - pxor xmm7, xmm7 // constant 0 for unpck + pxor xmm7, xmm7 // constant 0 for unpck movdqa xmm6, xmmword ptr kHash16x33 wloop: - movdqu xmm1, [eax] // src[0-15] + movdqu xmm1, [eax] // src[0-15] lea eax, [eax + 16] - pmulld xmm0, xmm6 // hash *= 33 ^ 16 + pmulld xmm0, xmm6 // hash *= 33 ^ 16 movdqa xmm5, xmmword ptr kHashMul0 movdqa xmm2, xmm1 - punpcklbw xmm2, xmm7 // src[0-7] + punpcklbw xmm2, xmm7 // src[0-7] movdqa xmm3, xmm2 - punpcklwd xmm3, xmm7 // src[0-3] + punpcklwd xmm3, xmm7 // src[0-3] pmulld xmm3, xmm5 movdqa xmm5, xmmword ptr kHashMul1 movdqa xmm4, xmm2 - punpckhwd xmm4, xmm7 // src[4-7] + punpckhwd xmm4, xmm7 // src[4-7] pmulld xmm4, xmm5 movdqa xmm5, xmmword ptr kHashMul2 - punpckhbw xmm1, xmm7 // src[8-15] + punpckhbw xmm1, xmm7 // src[8-15] movdqa xmm2, xmm1 - punpcklwd xmm2, xmm7 // src[8-11] + punpcklwd xmm2, xmm7 // src[8-11] pmulld xmm2, xmm5 movdqa xmm5, xmmword ptr kHashMul3 - punpckhwd xmm1, xmm7 // src[12-15] + punpckhwd xmm1, xmm7 // src[12-15] pmulld xmm1, xmm5 - paddd xmm3, xmm4 // add 16 results + paddd xmm3, xmm4 // add 16 results paddd xmm1, xmm2 paddd xmm1, xmm3 @@ -171,18 +190,18 @@ uint32 HashDjb2_SSE41(const uint8* src, int count, uint32 seed) { sub ecx, 16 jg wloop - movd eax, xmm0 // return hash + movd eax, xmm0 // return hash ret } } // Visual C 2012 required for AVX2. -#if _MSC_VER >= 1700 -__declspec(naked) -uint32 HashDjb2_AVX2(const uint8* src, int count, uint32 seed) { +#ifdef HAS_HASHDJB2_AVX2 +__declspec(naked) uint32_t + HashDjb2_AVX2(const uint8_t* src, int count, uint32_t seed) { __asm { - mov eax, [esp + 4] // src - mov ecx, [esp + 8] // count + mov eax, [esp + 4] // src + mov ecx, [esp + 8] // count vmovd xmm0, [esp + 12] // seed wloop: @@ -196,7 +215,7 @@ uint32 HashDjb2_AVX2(const uint8* src, int count, uint32 seed) { vpmulld xmm2, xmm2, xmmword ptr kHashMul2 lea eax, [eax + 16] vpmulld xmm1, xmm1, xmmword ptr kHashMul3 - vpaddd xmm3, xmm3, xmm4 // add 16 results + vpaddd xmm3, xmm3, xmm4 // add 16 results vpaddd xmm1, xmm1, xmm2 vpaddd xmm1, xmm1, xmm3 vpshufd xmm2, xmm1, 0x0e // upper 2 dwords @@ -207,12 +226,12 @@ uint32 HashDjb2_AVX2(const uint8* src, int count, uint32 seed) { sub ecx, 16 jg wloop - vmovd eax, xmm0 // return hash + vmovd eax, xmm0 // return hash vzeroupper ret } } -#endif // _MSC_VER >= 1700 +#endif // HAS_HASHDJB2_AVX2 #endif // !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) |