portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-12-23 19:35:26 +00:00
Code Releases Activity

AVXFMA target fixed

Browse Source
This commit is contained in:
Antonin Portelli 2016-11-21 17:47:18 +01:00
parent bafbac6ac4
commit a2cffb0304
3 changed files with 12 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
configure.ac
View File

@ -206,8 +206,8 @@ case ${ax_cv_cxx_compiler_vendor} in
AC_DEFINE([AVX1],[1],[AVX intrinsics])
SIMD_FLAGS='-mavx -xavx';;
AVXFMA)
AC_DEFINE([AVXFMA],[1],[AVX intrinsics with FMA4])
SIMD_FLAGS='-mavx -mfma';;
AC_DEFINE([AVXFMA],[1],[AVX intrinsics with FMA3])
SIMD_FLAGS='-mavx -fma';;
AVX2)
AC_DEFINE([AVX2],[1],[AVX2 intrinsics])
SIMD_FLAGS='-march=core-avx2 -xcore-avx2';;

18
lib/simd/Grid_avx.h
View File

@ -167,7 +167,7 @@ namespace Optimization {
}
//Integer
inline __m256i operator()(__m256i a, __m256i b){
#if defined (AVX1) || defined (AVXFMA4)
#if defined (AVX1) || defined (AVXFMA) || defined (AVXFMA4)
__m128i a0,a1;
__m128i b0,b1;
a0 = _mm256_extractf128_si256(a,0);
@ -195,7 +195,7 @@ namespace Optimization {
}
//Integer
inline __m256i operator()(__m256i a, __m256i b){
#if defined (AVX1) || defined (AVXFMA4)
#if defined (AVX1) || defined (AVXFMA) || defined (AVXFMA4)
__m128i a0,a1;
__m128i b0,b1;
a0 = _mm256_extractf128_si256(a,0);
@ -216,7 +216,7 @@ namespace Optimization {
struct MultComplex{
// Complex float
inline __m256 operator()(__m256 a, __m256 b){
#if defined (AVX1)
#if defined (AVX1)
__m256 ymm0,ymm1,ymm2;
ymm0 = _mm256_shuffle_ps(a,a,_MM_SELECT_FOUR_FOUR(2,2,0,0)); // ymm0 <- ar ar,
ymm0 = _mm256_mul_ps(ymm0,b); // ymm0 <- ar bi, ar br
@ -233,7 +233,7 @@ namespace Optimization {
a_imag = _mm256_mul_ps( a_imag,tmp ); // (Ai, Ai) * (Bi, Br) = Ai Bi, Ai Br
return _mm256_maddsub_ps( a_real, b, a_imag ); // Ar Br , Ar Bi +- Ai Bi = ArBr-AiBi , ArBi+AiBr
#endif
#if defined (AVX2)
#if defined (AVX2) || defined (AVXFMA)
__m256 a_real = _mm256_moveldup_ps( a ); // Ar Ar
__m256 a_imag = _mm256_movehdup_ps( a ); // Ai Ai
a_imag = _mm256_mul_ps( a_imag, _mm256_shuffle_ps( b,b, _MM_SELECT_FOUR_FOUR(2,3,0,1) )); // (Ai, Ai) * (Bi, Br) = Ai Bi, Ai Br
@ -264,7 +264,7 @@ namespace Optimization {
IF IMM0[3] = 0
THEN DEST[255:192]=SRC2[191:128] ELSE DEST[255:192]=SRC2[255:192] FI; // Ox5 r<->i ; 0xC unchanged
*/
#if defined (AVX1)
#if defined (AVX1)
__m256d ymm0,ymm1,ymm2;
ymm0 = _mm256_shuffle_pd(a,a,0x0); // ymm0 <- ar ar, ar,ar b'00,00
ymm0 = _mm256_mul_pd(ymm0,b); // ymm0 <- ar bi, ar br
@ -279,7 +279,7 @@ namespace Optimization {
a_imag = _mm256_mul_pd( a_imag, _mm256_permute_pd( b, 0x5 ) ); // (Ai, Ai) * (Bi, Br) = Ai Bi, Ai Br
return _mm256_maddsub_pd( a_real, b, a_imag ); // Ar Br , Ar Bi +- Ai Bi = ArBr-AiBi , ArBi+AiBr
#endif
#if defined (AVX2)
#if defined (AVX2) || defined (AVXFMA)
__m256d a_real = _mm256_movedup_pd( a ); // Ar Ar
__m256d a_imag = _mm256_shuffle_pd(a,a,0xF);//aiai
a_imag = _mm256_mul_pd( a_imag, _mm256_permute_pd( b, 0x5 ) ); // (Ai, Ai) * (Bi, Br) = Ai Bi, Ai Br
@ -320,7 +320,7 @@ namespace Optimization {
#if defined (AVXFMA4)
a= _mm256_macc_ps(b,c,a);
#endif
#if defined (AVX2)
#if defined (AVX2) || defined (AVXFMA)
a= _mm256_fmadd_ps( b, c, a);
#endif
}
@ -332,7 +332,7 @@ namespace Optimization {
#if defined (AVXFMA4)
a= _mm256_macc_pd(b,c,a);
#endif
#if defined (AVX2)
#if defined (AVX2) || defined (AVXFMA)
a= _mm256_fmadd_pd( b, c, a);
#endif
}
@ -347,7 +347,7 @@ namespace Optimization {
}
// Integer
inline __m256i operator()(__m256i a, __m256i b){
#if defined (AVX1)
#if defined (AVX1) || defined (AVXFMA)
__m128i a0,a1;
__m128i b0,b1;
a0 = _mm256_extractf128_si256(a,0);

2
lib/simd/Grid_vector_types.h
View File

@ -44,7 +44,7 @@ directory
#ifdef SSE4
#include "Grid_sse4.h"
#endif
#if defined(AVX1) || defined(AVX2) || defined(AVXFMA4)
#if defined(AVX1) || defined (AVXFMA) || defined(AVX2) || defined(AVXFMA4)
#include "Grid_avx.h"
#endif
#if defined AVX512