coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2025-12-07 01:06:46 +08:00
Code Issues Packages Projects Releases Wiki Activity

Using the visual C source allows clangcl to fallback seamlessly to visual c, and supports SSE41 and AVX2 versions.

Browse Source 
R=harryjin@google.com
BUG=libyuv:469

Review URL: https://webrtc-codereview.appspot.com/58469004.
This commit is contained in:
Frank Barchard 2015-08-17 10:47:43 -07:00
parent 278d88f872
commit baf6a3c1bd
6 changed files with 25 additions and 37 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
README.chromium
View File

@ -1,6 +1,6 @@
Name: libyuv Name: libyuv
URL: http://code.google.com/p/libyuv/ URL: http://code.google.com/p/libyuv/
Version: 1464 Version: 1465
License: BSD License: BSD
License File: LICENSE License File: LICENSE

4
include/libyuv/row.h
View File

@ -365,7 +365,7 @@ extern "C" {
#endif #endif
#endif #endif
#if defined(_MSC_VER) && !defined(__CLR_VER) #if defined(_MSC_VER) && !defined(__CLR_VER) && !defined(__clang__)
#define SIMD_ALIGNED(var) __declspec(align(16)) var #define SIMD_ALIGNED(var) __declspec(align(16)) var
#define SIMD_ALIGNED32(var) __declspec(align(64)) var #define SIMD_ALIGNED32(var) __declspec(align(64)) var
typedef __declspec(align(16)) int16 vec16[8]; typedef __declspec(align(16)) int16 vec16[8];
@ -380,7 +380,7 @@ typedef __declspec(align(32)) int8 lvec8[32];
typedef __declspec(align(32)) uint16 ulvec16[16]; typedef __declspec(align(32)) uint16 ulvec16[16];
typedef __declspec(align(32)) uint32 ulvec32[8]; typedef __declspec(align(32)) uint32 ulvec32[8];
typedef __declspec(align(32)) uint8 ulvec8[32]; typedef __declspec(align(32)) uint8 ulvec8[32];
#elif defined(__GNUC__) #elif defined(__GNUC__) || defined(__clang__)
// Caveat GCC 4.2 to 4.7 have a known issue using vectors with const. // Caveat GCC 4.2 to 4.7 have a known issue using vectors with const.
#define SIMD_ALIGNED(var) var __attribute__((aligned(16))) #define SIMD_ALIGNED(var) var __attribute__((aligned(16)))
#define SIMD_ALIGNED32(var) var __attribute__((aligned(64))) #define SIMD_ALIGNED32(var) var __attribute__((aligned(64)))

2
include/libyuv/version.h
View File

@ -11,6 +11,6 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_ // NOLINT #ifndef INCLUDE_LIBYUV_VERSION_H_ // NOLINT
#define INCLUDE_LIBYUV_VERSION_H_ #define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 1464 #define LIBYUV_VERSION 1465
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT #endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT

2
source/compare.cc
View File

@ -37,7 +37,7 @@ uint32 HashDjb2_C(const uint8* src, int count, uint32 seed);
#define HAS_HASHDJB2_SSE41 #define HAS_HASHDJB2_SSE41
uint32 HashDjb2_SSE41(const uint8* src, int count, uint32 seed); uint32 HashDjb2_SSE41(const uint8* src, int count, uint32 seed);
#ifdef VISUALC_HAS_AVX2 #if defined(_M_IX86) && (defined(VISUALC_HAS_AVX2) || defined(CLANG_HAS_AVX2))
#define HAS_HASHDJB2_AVX2 #define HAS_HASHDJB2_AVX2
uint32 HashDjb2_AVX2(const uint8* src, int count, uint32 seed); uint32 HashDjb2_AVX2(const uint8* src, int count, uint32 seed);
#endif #endif

9
source/compare_gcc.cc
View File

@ -16,7 +16,9 @@ namespace libyuv {
extern "C" { extern "C" {
#endif #endif
#if !defined(LIBYUV_DISABLE_X86) && (defined(__x86_64__) || defined(__i386__)) // This module is for GCC x86 and x64.
#if !defined(LIBYUV_DISABLE_X86) && \
(defined(__x86_64__) || (defined(__i386__) && !defined(_MSC_VER)))
uint32 SumSquareError_SSE2(const uint8* src_a, const uint8* src_b, int count) { uint32 SumSquareError_SSE2(const uint8* src_a, const uint8* src_b, int count) {
uint32 sse; uint32 sse;
@ -58,11 +60,6 @@ uint32 SumSquareError_SSE2(const uint8* src_a, const uint8* src_b, int count) {
return sse; return sse;
} }
#endif // defined(__x86_64__) || defined(__i386__)
#if !defined(LIBYUV_DISABLE_X86) && \
(defined(__x86_64__) || (defined(__i386__) && !defined(__pic__)))
#define HAS_HASHDJB2_SSE41
static uvec32 kHash16x33 = { 0x92d9e201, 0, 0, 0 }; // 33 ^ 16 static uvec32 kHash16x33 = { 0x92d9e201, 0, 0, 0 }; // 33 ^ 16
static uvec32 kHashMul0 = { static uvec32 kHashMul0 = {
0x0c3525e1, // 33 ^ 15 0x0c3525e1, // 33 ^ 15

43
source/compare_win.cc
View File

@ -16,9 +16,8 @@ namespace libyuv {
extern "C" { extern "C" {
#endif #endif
// This module is for Visual C x86. // This module is for 32 bit Visual C x86 and clangcl
#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && \ #if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86)
defined(_MSC_VER) && !defined(__clang__)
__declspec(naked) __declspec(naked)
uint32 SumSquareError_SSE2(const uint8* src_a, const uint8* src_b, int count) { uint32 SumSquareError_SSE2(const uint8* src_a, const uint8* src_b, int count) {
@ -100,41 +99,32 @@ uint32 SumSquareError_AVX2(const uint8* src_a, const uint8* src_b, int count) {
} }
#endif // _MSC_VER >= 1700 #endif // _MSC_VER >= 1700
#define HAS_HASHDJB2_SSE41 uvec32 kHash16x33 = { 0x92d9e201, 0, 0, 0 }; // 33 ^ 16
static uvec32 kHash16x33 = { 0x92d9e201, 0, 0, 0 }; // 33 ^ 16 uvec32 kHashMul0 = {
static uvec32 kHashMul0 = {
0x0c3525e1, // 33 ^ 15 0x0c3525e1, // 33 ^ 15
0xa3476dc1, // 33 ^ 14 0xa3476dc1, // 33 ^ 14
0x3b4039a1, // 33 ^ 13 0x3b4039a1, // 33 ^ 13
0x4f5f0981, // 33 ^ 12 0x4f5f0981, // 33 ^ 12
}; };
static uvec32 kHashMul1 = { uvec32 kHashMul1 = {
0x30f35d61, // 33 ^ 11 0x30f35d61, // 33 ^ 11
0x855cb541, // 33 ^ 10 0x855cb541, // 33 ^ 10
0x040a9121, // 33 ^ 9 0x040a9121, // 33 ^ 9
0x747c7101, // 33 ^ 8 0x747c7101, // 33 ^ 8
}; };
static uvec32 kHashMul2 = { uvec32 kHashMul2 = {
0xec41d4e1, // 33 ^ 7 0xec41d4e1, // 33 ^ 7
0x4cfa3cc1, // 33 ^ 6 0x4cfa3cc1, // 33 ^ 6
0x025528a1, // 33 ^ 5 0x025528a1, // 33 ^ 5
0x00121881, // 33 ^ 4 0x00121881, // 33 ^ 4
}; };
static uvec32 kHashMul3 = { uvec32 kHashMul3 = {
0x00008c61, // 33 ^ 3 0x00008c61, // 33 ^ 3
0x00000441, // 33 ^ 2 0x00000441, // 33 ^ 2
0x00000021, // 33 ^ 1 0x00000021, // 33 ^ 1
0x00000001, // 33 ^ 0 0x00000001, // 33 ^ 0
}; };
// 27: 66 0F 38 40 C6 pmulld xmm0,xmm6
// 44: 66 0F 38 40 DD pmulld xmm3,xmm5
// 59: 66 0F 38 40 E5 pmulld xmm4,xmm5
// 72: 66 0F 38 40 D5 pmulld xmm2,xmm5
// 83: 66 0F 38 40 CD pmulld xmm1,xmm5
#define pmulld(reg) _asm _emit 0x66 _asm _emit 0x0F _asm _emit 0x38 \
_asm _emit 0x40 _asm _emit reg
__declspec(naked) __declspec(naked)
uint32 HashDjb2_SSE41(const uint8* src, int count, uint32 seed) { uint32 HashDjb2_SSE41(const uint8* src, int count, uint32 seed) {
__asm { __asm {
@ -148,25 +138,25 @@ uint32 HashDjb2_SSE41(const uint8* src, int count, uint32 seed) {
wloop: wloop:
movdqu xmm1, [eax] // src[0-15] movdqu xmm1, [eax] // src[0-15]
lea eax, [eax + 16] lea eax, [eax + 16]
pmulld(0xc6) // pmulld xmm0,xmm6 hash *= 33 ^ 16 pmulld xmm0, xmm6 // hash *= 33 ^ 16
movdqa xmm5, kHashMul0 movdqa xmm5, kHashMul0
movdqa xmm2, xmm1 movdqa xmm2, xmm1
punpcklbw xmm2, xmm7 // src[0-7] punpcklbw xmm2, xmm7 // src[0-7]
movdqa xmm3, xmm2 movdqa xmm3, xmm2
punpcklwd xmm3, xmm7 // src[0-3] punpcklwd xmm3, xmm7 // src[0-3]
pmulld(0xdd) // pmulld xmm3, xmm5 pmulld xmm3, xmm5
movdqa xmm5, kHashMul1 movdqa xmm5, kHashMul1
movdqa xmm4, xmm2 movdqa xmm4, xmm2
punpckhwd xmm4, xmm7 // src[4-7] punpckhwd xmm4, xmm7 // src[4-7]
pmulld(0xe5) // pmulld xmm4, xmm5 pmulld xmm4, xmm5
movdqa xmm5, kHashMul2 movdqa xmm5, kHashMul2
punpckhbw xmm1, xmm7 // src[8-15] punpckhbw xmm1, xmm7 // src[8-15]
movdqa xmm2, xmm1 movdqa xmm2, xmm1
punpcklwd xmm2, xmm7 // src[8-11] punpcklwd xmm2, xmm7 // src[8-11]
pmulld(0xd5) // pmulld xmm2, xmm5 pmulld xmm2, xmm5
movdqa xmm5, kHashMul3 movdqa xmm5, kHashMul3
punpckhwd xmm1, xmm7 // src[12-15] punpckhwd xmm1, xmm7 // src[12-15]
pmulld(0xcd) // pmulld xmm1, xmm5 pmulld xmm1, xmm5
paddd xmm3, xmm4 // add 16 results paddd xmm3, xmm4 // add 16 results
paddd xmm1, xmm2 paddd xmm1, xmm2
paddd xmm1, xmm3 paddd xmm1, xmm3
@ -195,13 +185,13 @@ uint32 HashDjb2_AVX2(const uint8* src, int count, uint32 seed) {
movdqa xmm6, kHash16x33 movdqa xmm6, kHash16x33
wloop: wloop:
vpmovzxbd xmm3, dword ptr [eax] // src[0-3] vpmovzxbd xmm3, [eax] // src[0-3]
pmulld xmm0, xmm6 // hash *= 33 ^ 16 pmulld xmm0, xmm6 // hash *= 33 ^ 16
vpmovzxbd xmm4, dword ptr [eax + 4] // src[4-7] vpmovzxbd xmm4, [eax + 4] // src[4-7]
pmulld xmm3, kHashMul0 pmulld xmm3, kHashMul0
vpmovzxbd xmm2, dword ptr [eax + 8] // src[8-11] vpmovzxbd xmm2, [eax + 8] // src[8-11]
pmulld xmm4, kHashMul1 pmulld xmm4, kHashMul1
vpmovzxbd xmm1, dword ptr [eax + 12] // src[12-15] vpmovzxbd xmm1, [eax + 12] // src[12-15]
pmulld xmm2, kHashMul2 pmulld xmm2, kHashMul2
lea eax, [eax + 16] lea eax, [eax + 16]
pmulld xmm1, kHashMul3 pmulld xmm1, kHashMul3
@ -221,6 +211,7 @@ uint32 HashDjb2_AVX2(const uint8* src, int count, uint32 seed) {
} }
} }
#endif // _MSC_VER >= 1700 #endif // _MSC_VER >= 1700
#endif // !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) #endif // !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86)
#ifdef __cplusplus #ifdef __cplusplus