portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-17 07:17:06 +01:00
Code Releases Activity

Merge branch 'develop' into feature/hmc_generalise

Browse Source
This commit is contained in:
Guido Cossu
2017-04-05 14:41:04 +01:00
parent 4c1ea8677e 5592f7b8c1
commit 8c540333d5
205 changed files with 27899 additions and 3601 deletions
Download Patch File Download Diff File Expand all files Collapse all files

55
lib/simd/Grid_avx.h
View File

@ -460,9 +460,62 @@ namespace Optimization {
static inline __m256d Permute3(__m256d in){
return in;
};
};
struct Exchange{
// 3210 ordering
static inline void Exchange0(__m256 &out1,__m256 &out2,__m256 in1,__m256 in2){
//Invertible
//AB CD -> AC BD
//AC BD -> AB CD
out1= _mm256_permute2f128_ps(in1,in2,0x20);
out2= _mm256_permute2f128_ps(in1,in2,0x31);
};
static inline void Exchange1(__m256 &out1,__m256 &out2,__m256 in1,__m256 in2){
//Invertible
// ABCD EFGH ->ABEF CDGH
// ABEF CDGH ->ABCD EFGH
out1= _mm256_shuffle_ps(in1,in2,_MM_SELECT_FOUR_FOUR(1,0,1,0));
out2= _mm256_shuffle_ps(in1,in2,_MM_SELECT_FOUR_FOUR(3,2,3,2));
};
static inline void Exchange2(__m256 &out1,__m256 &out2,__m256 in1,__m256 in2){
// Invertible ?
// ABCD EFGH -> ACEG BDFH
// ACEG BDFH -> AEBF CGDH
// out1= _mm256_shuffle_ps(in1,in2,_MM_SELECT_FOUR_FOUR(2,0,2,0));
// out2= _mm256_shuffle_ps(in1,in2,_MM_SELECT_FOUR_FOUR(3,1,3,1));
// Bollocks; need
// AECG BFDH -> ABCD EFGH
out1= _mm256_shuffle_ps(in1,in2,_MM_SELECT_FOUR_FOUR(2,0,2,0)); /*ACEG*/
out2= _mm256_shuffle_ps(in1,in2,_MM_SELECT_FOUR_FOUR(3,1,3,1)); /*BDFH*/
out1= _mm256_shuffle_ps(out1,out1,_MM_SELECT_FOUR_FOUR(3,1,2,0)); /*AECG*/
out2= _mm256_shuffle_ps(out2,out2,_MM_SELECT_FOUR_FOUR(3,1,2,0)); /*AECG*/
};
static inline void Exchange3(__m256 &out1,__m256 &out2,__m256 in1,__m256 in2){
assert(0);
return;
};
static inline void Exchange0(__m256d &out1,__m256d &out2,__m256d in1,__m256d in2){
out1= _mm256_permute2f128_pd(in1,in2,0x20);
out2= _mm256_permute2f128_pd(in1,in2,0x31);
return;
};
static inline void Exchange1(__m256d &out1,__m256d &out2,__m256d in1,__m256d in2){
out1= _mm256_shuffle_pd(in1,in2,0x0);
out2= _mm256_shuffle_pd(in1,in2,0xF);
};
static inline void Exchange2(__m256d &out1,__m256d &out2,__m256d in1,__m256d in2){
assert(0);
return;
};
static inline void Exchange3(__m256d &out1,__m256d &out2,__m256d in1,__m256d in2){
assert(0);
return;
};
};
#if defined (AVX2)
#define _mm256_alignr_epi32_grid(ret,a,b,n) ret=(__m256) _mm256_alignr_epi8((__m256i)a,(__m256i)b,(n*4)%16)
#define _mm256_alignr_epi64_grid(ret,a,b,n) ret=(__m256d) _mm256_alignr_epi8((__m256i)a,(__m256i)b,(n*8)%16)