Re #2065: We still need to modify some source code to avoid build error on
Visual Studio 2005. git-svn-id: https://svn.pjsip.org/repos/pjproject/trunk@5708 74dad513-b988-da41-8d7b-12977e46ad98
This commit is contained in:
Riza Sulistyo
parent
a39e60661e
commit
126a428c1e
|
|
@ -2,3 +2,37 @@ Notes:
|
|||
|
||||
* Source code for libyuv from https://chromium.googlesource.com/libyuv/libyuv/ dated 17 November 2017.
|
||||
|
||||
* All code is compilable, except for compare_win.cc
|
||||
- Use older version (https://chromium.googlesource.com/libyuv/libyuv/+/baf6a3c1bd385e7ffe6b7634560e71fb49e4f589%5E%21/)
|
||||
Since there's a compiler error on (VS2005):
|
||||
--------------------------------------------------------------------------------------
|
||||
pmulld xmm0,xmm6
|
||||
--------------------------------------------------------------------------------------
|
||||
|
||||
- On VS2015, error C2024: 'alignas' attribute applies to variables, data members and tag types only
|
||||
--------------------------------------------------------------------------------------
|
||||
__declspec(naked) __declspec(align(16))
|
||||
|
||||
Change to :
|
||||
|
||||
__declspec(naked)
|
||||
--------------------------------------------------------------------------------------
|
||||
|
||||
* Added these lines to file include/libyuv/basic_types.h:
|
||||
--
|
||||
#if _MSC_VER==1400
|
||||
# include <stdint.h> // for uint8_t
|
||||
#endif
|
||||
...
|
||||
#if defined(_MSC_VER)
|
||||
# pragma warning(disable:4996) // This function or variable may be unsafe.
|
||||
#endif
|
||||
--
|
||||
|
||||
* Modify compare_row.h:
|
||||
- VS2005 doesn't support SSE42, resulting error on HammingDistance_SSE42().
|
||||
--------------------------------------------------------------------------------------
|
||||
diff += __popcnt(x);
|
||||
--------------------------------------------------------------------------------------
|
||||
So, we use the default method HammingDistance_C() to calculate Hamming Distance.
|
||||
|
||||
|
|
@ -14,11 +14,18 @@
|
|||
#include <stddef.h> // for NULL, size_t
|
||||
|
||||
#if defined(_MSC_VER) && (_MSC_VER < 1600)
|
||||
#if _MSC_VER==1400
|
||||
# include <stdint.h> // for uint8_t
|
||||
#endif
|
||||
#include <sys/types.h> // for uintptr_t on x86
|
||||
#else
|
||||
#include <stdint.h> // for uintptr_t
|
||||
#endif
|
||||
|
||||
#if defined(_MSC_VER)
|
||||
# pragma warning(disable:4996) // This function or variable may be unsafe.
|
||||
#endif
|
||||
|
||||
#ifndef GG_LONGLONG
|
||||
#ifndef INT_TYPES_DEFINED
|
||||
#define INT_TYPES_DEFINED
|
||||
|
|
|
|||
|
|
@ -53,7 +53,10 @@ extern "C" {
|
|||
(defined(__x86_64__) || defined(__i386__) || defined(_M_IX86))
|
||||
#define HAS_HASHDJB2_SSE41
|
||||
#define HAS_SUMSQUAREERROR_SSE2
|
||||
#define HAS_HAMMINGDISTANCE_SSE42
|
||||
# if (MSC_VER >= 1400)
|
||||
// Visual Studio 2005 doesn't support sse42
|
||||
# define HAS_HAMMINGDISTANCE_SSE42
|
||||
# endif
|
||||
#endif
|
||||
|
||||
// The following are available for Visual C and clangcl 32 bit:
|
||||
|
|
|
|||
|
|
@ -9,8 +9,6 @@
|
|||
*/
|
||||
|
||||
#include "libyuv/basic_types.h"
|
||||
|
||||
#include "libyuv/compare_row.h"
|
||||
#include "libyuv/row.h"
|
||||
|
||||
#if defined(_MSC_VER)
|
||||
|
|
@ -22,9 +20,10 @@ namespace libyuv {
|
|||
extern "C" {
|
||||
#endif
|
||||
|
||||
// This module is for 32 bit Visual C x86 and clangcl
|
||||
#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && defined(_MSC_VER)
|
||||
|
||||
/* Visual Studio 2005 doesn't support __popcnt(). */
|
||||
#if (_MSC_VER > 1400)
|
||||
uint32 HammingDistance_SSE42(const uint8* src_a,
|
||||
const uint8* src_b,
|
||||
int count) {
|
||||
|
|
@ -39,21 +38,28 @@ uint32 HammingDistance_SSE42(const uint8* src_a,
|
|||
}
|
||||
return diff;
|
||||
}
|
||||
#endif
|
||||
|
||||
__declspec(naked) uint32
|
||||
SumSquareError_SSE2(const uint8* src_a, const uint8* src_b, int count) {
|
||||
#if (_MSC_VER >= 1900)
|
||||
__declspec(naked)
|
||||
#else
|
||||
__declspec(naked) __declspec(align(16))
|
||||
#endif
|
||||
uint32 SumSquareError_SSE2(const uint8* src_a, const uint8* 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
|
||||
|
||||
align 4
|
||||
wloop:
|
||||
movdqu xmm1, [eax]
|
||||
movdqa xmm1, [eax]
|
||||
lea eax, [eax + 16]
|
||||
movdqu xmm2, [edx]
|
||||
movdqa xmm2, [edx]
|
||||
lea edx, [edx + 16]
|
||||
sub ecx, 16
|
||||
movdqa xmm3, xmm1 // abs trick
|
||||
psubusb xmm1, xmm2
|
||||
psubusb xmm2, xmm3
|
||||
|
|
@ -65,7 +71,6 @@ __declspec(naked) uint32
|
|||
pmaddwd xmm2, xmm2
|
||||
paddd xmm0, xmm1
|
||||
paddd xmm0, xmm2
|
||||
sub ecx, 16
|
||||
jg wloop
|
||||
|
||||
pshufd xmm1, xmm0, 0xee
|
||||
|
|
@ -80,21 +85,27 @@ __declspec(naked) uint32
|
|||
// Visual C 2012 required for AVX2.
|
||||
#if _MSC_VER >= 1700
|
||||
// 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)
|
||||
#if (_MSC_VER >= 1900)
|
||||
__declspec(naked)
|
||||
#else
|
||||
__declspec(naked) __declspec(align(16))
|
||||
#endif
|
||||
uint32 SumSquareError_AVX2(const uint8* src_a, const uint8* 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
|
||||
|
||||
align 4
|
||||
wloop:
|
||||
vmovdqu ymm1, [eax]
|
||||
vmovdqu ymm2, [eax + edx]
|
||||
lea eax, [eax + 32]
|
||||
sub ecx, 32
|
||||
vpsubusb ymm3, ymm1, ymm2 // abs difference trick
|
||||
vpsubusb ymm2, ymm2, ymm1
|
||||
vpor ymm1, ymm2, ymm3
|
||||
|
|
@ -104,7 +115,6 @@ __declspec(naked) uint32
|
|||
vpmaddwd ymm1, ymm1, ymm1
|
||||
vpaddd ymm0, ymm0, ymm1
|
||||
vpaddd ymm0, ymm0, ymm2
|
||||
sub ecx, 32
|
||||
jg wloop
|
||||
|
||||
vpshufd ymm1, ymm0, 0xee // 3, 2 + 1, 0 both lanes.
|
||||
|
|
@ -120,66 +130,81 @@ __declspec(naked) uint32
|
|||
}
|
||||
#endif // _MSC_VER >= 1700
|
||||
|
||||
uvec32 kHash16x33 = {0x92d9e201, 0, 0, 0}; // 33 ^ 16
|
||||
uvec32 kHashMul0 = {
|
||||
0x0c3525e1, // 33 ^ 15
|
||||
0xa3476dc1, // 33 ^ 14
|
||||
0x3b4039a1, // 33 ^ 13
|
||||
0x4f5f0981, // 33 ^ 12
|
||||
#define HAS_HASHDJB2_SSE41
|
||||
static uvec32 kHash16x33 = { 0x92d9e201, 0, 0, 0 }; // 33 ^ 16
|
||||
static uvec32 kHashMul0 = {
|
||||
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
|
||||
static uvec32 kHashMul1 = {
|
||||
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
|
||||
static uvec32 kHashMul2 = {
|
||||
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
|
||||
static uvec32 kHashMul3 = {
|
||||
0x00008c61, // 33 ^ 3
|
||||
0x00000441, // 33 ^ 2
|
||||
0x00000021, // 33 ^ 1
|
||||
0x00000001, // 33 ^ 0
|
||||
};
|
||||
|
||||
__declspec(naked) uint32
|
||||
HashDjb2_SSE41(const uint8* src, int count, uint32 seed) {
|
||||
// 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
|
||||
|
||||
#if (_MSC_VER >= 1900)
|
||||
__declspec(naked)
|
||||
#else
|
||||
__declspec(naked) __declspec(align(16))
|
||||
#endif
|
||||
uint32 HashDjb2_SSE41(const uint8* src, int count, uint32 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
|
||||
movdqa xmm6, xmmword ptr kHash16x33
|
||||
pxor xmm7, xmm7 // constant 0 for unpck
|
||||
movdqa xmm6, kHash16x33
|
||||
|
||||
align 4
|
||||
wloop:
|
||||
movdqu xmm1, [eax] // src[0-15]
|
||||
movdqu xmm1, [eax] // src[0-15]
|
||||
lea eax, [eax + 16]
|
||||
pmulld xmm0, xmm6 // hash *= 33 ^ 16
|
||||
movdqa xmm5, xmmword ptr kHashMul0
|
||||
pmulld(0xc6) // pmulld xmm0,xmm6 hash *= 33 ^ 16
|
||||
movdqa xmm5, 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]
|
||||
pmulld xmm3, xmm5
|
||||
movdqa xmm5, xmmword ptr kHashMul1
|
||||
punpcklwd xmm3, xmm7 // src[0-3]
|
||||
pmulld(0xdd) // pmulld xmm3, xmm5
|
||||
movdqa xmm5, kHashMul1
|
||||
movdqa xmm4, xmm2
|
||||
punpckhwd xmm4, xmm7 // src[4-7]
|
||||
pmulld xmm4, xmm5
|
||||
movdqa xmm5, xmmword ptr kHashMul2
|
||||
punpckhbw xmm1, xmm7 // src[8-15]
|
||||
punpckhwd xmm4, xmm7 // src[4-7]
|
||||
pmulld(0xe5) // pmulld xmm4, xmm5
|
||||
movdqa xmm5, kHashMul2
|
||||
punpckhbw xmm1, xmm7 // src[8-15]
|
||||
movdqa xmm2, xmm1
|
||||
punpcklwd xmm2, xmm7 // src[8-11]
|
||||
pmulld xmm2, xmm5
|
||||
movdqa xmm5, xmmword ptr kHashMul3
|
||||
punpckhwd xmm1, xmm7 // src[12-15]
|
||||
pmulld xmm1, xmm5
|
||||
paddd xmm3, xmm4 // add 16 results
|
||||
punpcklwd xmm2, xmm7 // src[8-11]
|
||||
pmulld(0xd5) // pmulld xmm2, xmm5
|
||||
movdqa xmm5, kHashMul3
|
||||
punpckhwd xmm1, xmm7 // src[12-15]
|
||||
pmulld(0xcd) // pmulld xmm1, xmm5
|
||||
paddd xmm3, xmm4 // add 16 results
|
||||
paddd xmm1, xmm2
|
||||
sub ecx, 16
|
||||
paddd xmm1, xmm3
|
||||
|
||||
pshufd xmm2, xmm1, 0x0e // upper 2 dwords
|
||||
|
|
@ -187,53 +212,57 @@ __declspec(naked) uint32
|
|||
pshufd xmm2, xmm1, 0x01
|
||||
paddd xmm1, xmm2
|
||||
paddd xmm0, xmm1
|
||||
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) {
|
||||
#if (_MSC_VER >= 1900)
|
||||
__declspec(naked)
|
||||
#else
|
||||
__declspec(naked) __declspec(align(16))
|
||||
#endif
|
||||
uint32 HashDjb2_AVX2(const uint8* src, int count, uint32 seed) {
|
||||
__asm {
|
||||
mov eax, [esp + 4] // src
|
||||
mov ecx, [esp + 8] // count
|
||||
vmovd xmm0, [esp + 12] // seed
|
||||
mov eax, [esp + 4] // src
|
||||
mov ecx, [esp + 8] // count
|
||||
movd xmm0, [esp + 12] // seed
|
||||
movdqa xmm6, kHash16x33
|
||||
|
||||
align 4
|
||||
wloop:
|
||||
vpmovzxbd xmm3, [eax] // src[0-3]
|
||||
vpmulld xmm0, xmm0, xmmword ptr kHash16x33 // hash *= 33 ^ 16
|
||||
vpmovzxbd xmm4, [eax + 4] // src[4-7]
|
||||
vpmulld xmm3, xmm3, xmmword ptr kHashMul0
|
||||
vpmovzxbd xmm2, [eax + 8] // src[8-11]
|
||||
vpmulld xmm4, xmm4, xmmword ptr kHashMul1
|
||||
vpmovzxbd xmm1, [eax + 12] // src[12-15]
|
||||
vpmulld xmm2, xmm2, xmmword ptr kHashMul2
|
||||
vpmovzxbd xmm3, dword ptr [eax] // src[0-3]
|
||||
pmulld xmm0, xmm6 // hash *= 33 ^ 16
|
||||
vpmovzxbd xmm4, dword ptr [eax + 4] // src[4-7]
|
||||
pmulld xmm3, kHashMul0
|
||||
vpmovzxbd xmm2, dword ptr [eax + 8] // src[8-11]
|
||||
pmulld xmm4, kHashMul1
|
||||
vpmovzxbd xmm1, dword ptr [eax + 12] // src[12-15]
|
||||
pmulld xmm2, kHashMul2
|
||||
lea eax, [eax + 16]
|
||||
vpmulld xmm1, xmm1, xmmword ptr kHashMul3
|
||||
vpaddd xmm3, xmm3, xmm4 // add 16 results
|
||||
vpaddd xmm1, xmm1, xmm2
|
||||
vpaddd xmm1, xmm1, xmm3
|
||||
vpshufd xmm2, xmm1, 0x0e // upper 2 dwords
|
||||
vpaddd xmm1, xmm1,xmm2
|
||||
vpshufd xmm2, xmm1, 0x01
|
||||
vpaddd xmm1, xmm1, xmm2
|
||||
vpaddd xmm0, xmm0, xmm1
|
||||
pmulld xmm1, kHashMul3
|
||||
paddd xmm3, xmm4 // add 16 results
|
||||
paddd xmm1, xmm2
|
||||
sub ecx, 16
|
||||
paddd xmm1, xmm3
|
||||
pshufd xmm2, xmm1, 0x0e // upper 2 dwords
|
||||
paddd xmm1, xmm2
|
||||
pshufd xmm2, xmm1, 0x01
|
||||
paddd xmm1, xmm2
|
||||
paddd xmm0, xmm1
|
||||
jg wloop
|
||||
|
||||
vmovd eax, xmm0 // return hash
|
||||
vzeroupper
|
||||
movd eax, xmm0 // return hash
|
||||
ret
|
||||
}
|
||||
}
|
||||
#endif // _MSC_VER >= 1700
|
||||
|
||||
#endif // !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86)
|
||||
#endif // !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && defined(_MSC_VER)
|
||||
|
||||
#ifdef __cplusplus
|
||||
} // extern "C"
|
||||
|
|
|
|||
Loading…
Reference in New Issue