portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-10 03:17:07 +01:00
Code Releases Activity

Lots of changes required to compile for MIC under ICPC

Browse Source
This commit is contained in:
Peter Boyle
2015-05-10 23:29:21 +01:00
parent 48b9692845
commit 5555a852be
20 changed files with 3438 additions and 2858 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
lib/simd/Grid_vComplexD.h
View File

@ -143,10 +143,10 @@ namespace Grid {
* }
*/
zvec vzero,ymm0,ymm1,real,imag;
vzero = _mm512_setzero();
vzero =(zvec)_mm512_setzero();
ymm0 = _mm512_swizzle_pd(a.v, _MM_SWIZ_REG_CDAB); //
real = _mm512_mask_or_epi64(a.v, 0xAAAA,vzero, ymm0);
imag = _mm512_mask_sub_pd(a.v, 0x5555,vzero, ymm0);
real =(zvec)_mm512_mask_or_epi64((__m512i)a.v, 0xAA,(__m512i)vzero,(__m512i) ymm0);
imag = _mm512_mask_sub_pd(a.v, 0x55,vzero, ymm0);
ymm1 = _mm512_mul_pd(real, b.v);
ymm0 = _mm512_swizzle_pd(b.v, _MM_SWIZ_REG_CDAB); // OK
ret.v= _mm512_fmadd_pd(ymm0,imag,ymm1);
@ -250,7 +250,8 @@ friend inline void vstore(const vComplexD &ret, ComplexD *a){
_mm_stream_pd((double *)&out.v,in.v);
#endif
#ifdef AVX512
_mm512_stream_pd((double *)&out.v,in.v);
_mm512_storenrngo_pd((double *)&out.v,in.v);
// _mm512_stream_pd((double *)&out.v,in.v);
//Note v has a3 a2 a1 a0
#endif
#ifdef QPX
@ -277,7 +278,7 @@ friend inline void vstore(const vComplexD &ret, ComplexD *a){
ret.v = _mm_shuffle_pd(tmp,tmp,0x1);
#endif
#ifdef AVX512
ret.v = _mm512_mask_sub_pd(in.v, 0xaaaa,ret.v, in.v);
ret.v = _mm512_mask_sub_pd(in.v, 0xaa,ret.v, in.v);
#endif
#ifdef QPX
assert(0);
@ -297,7 +298,7 @@ friend inline void vstore(const vComplexD &ret, ComplexD *a){
ret.v =_mm_shuffle_pd(tmp.v,tmp.v,0x1);
#endif
#ifdef AVX512
ret.v = _mm512_mask_sub_pd(in.v,0xaaaa,ret.v,in.v); // real -imag
ret.v = _mm512_mask_sub_pd(in.v,0xaa,ret.v,in.v); // real -imag
ret.v = _mm512_swizzle_pd(ret.v, _MM_SWIZ_REG_CDAB);// OK
#endif
#ifdef QPX
@ -319,7 +320,7 @@ friend inline void vstore(const vComplexD &ret, ComplexD *a){
#endif
#ifdef AVX512
tmp.v = _mm512_swizzle_pd(in.v, _MM_SWIZ_REG_CDAB);// OK
ret.v = _mm512_mask_sub_pd(tmp.v,0xaaaa,ret.v,tmp.v); // real -imag
ret.v = _mm512_mask_sub_pd(tmp.v,0xaa,ret.v,tmp.v); // real -imag
#endif
#ifdef QPX
assert(0);
@ -337,7 +338,7 @@ friend inline void vstore(const vComplexD &ret, ComplexD *a){
return ComplexD(c_[0]+c_[2],c_[1]+c_[3]);
#endif
#ifdef AVX512
return ComplexD(_mm512_mask_reduce_add_pd(0x5555, in.v),_mm512_mask_reduce_add_pd(0xAAAA, in.v));
return ComplexD(_mm512_mask_reduce_add_pd(0x55, in.v),_mm512_mask_reduce_add_pd(0xAA, in.v));
#endif
#ifdef QPX
#endif

5
lib/simd/Grid_vComplexF.h
View File

@ -133,7 +133,7 @@ namespace Grid {
cvec vzero,ymm0,ymm1,real, imag;
vzero = _mm512_setzero();
ymm0 = _mm512_swizzle_ps(a.v, _MM_SWIZ_REG_CDAB); //
real = _mm512_mask_or_epi32(a.v, 0xAAAA,vzero, ymm0);
real = (__m512)_mm512_mask_or_epi32((__m512i)a.v, 0xAAAA,(__m512i)vzero,(__m512i)ymm0);
imag = _mm512_mask_sub_ps(a.v, 0x5555,vzero, ymm0);
ymm1 = _mm512_mul_ps(real, b.v);
ymm0 = _mm512_swizzle_ps(b.v, _MM_SWIZ_REG_CDAB); // OK
@ -199,7 +199,8 @@ namespace Grid {
_mm_stream_ps((float *)&out.v,in.v);
#endif
#ifdef AVX512
_mm512_stream_ps((float *)&out.v,in.v);
_mm512_storenrngo_ps((float *)&out.v,in.v);
// _mm512_stream_ps((float *)&out.v,in.v);
//Note v has a3 a2 a1 a0
#endif
#ifdef QPX

7
lib/simd/Grid_vInteger.h
View File

@ -3,8 +3,10 @@
namespace Grid {
#define _mm256_set_m128i(hi,lo) _mm256_insertf128_si256(_mm256_castsi128_si256(lo),(hi),1)
// _mm256_set_m128i(hi,lo); // not defined in all versions of immintrin.h
#ifndef _mm256_set_m128i
#define _mm256_set_m128i(hi,lo) _mm256_insertf128_si256(_mm256_castsi128_si256(lo),(hi),1)
#endif
typedef uint32_t Integer;
@ -110,7 +112,8 @@ namespace Grid {
ret.v = _mm_mul_epi32(a.v,b.v);
#endif
#ifdef AVX512
ret.v = _mm512_mul_epi32(a.v,b.v);
// ret.v = _mm512_mul_epi32(a.v,b.v);
ret.v = _mm512_mullo_epi32(a.v,b.v);
#endif
#ifdef QPX
// Implement as array of ints is only option

2
lib/simd/Grid_vRealD.h
View File

@ -184,7 +184,7 @@ namespace Grid {
_mm_stream_pd((double *)&out.v,in.v);
#endif
#ifdef AVX512
_mm512_stream_pd((double *)&out.v,in.v);
_mm512_storenrngo_pd((double *)&out.v,in.v);
//Note v has a3 a2 a1 a0
#endif
#ifdef QPX

3
lib/simd/Grid_vRealF.h
View File

@ -216,7 +216,8 @@ friend inline void vstore(const vRealF &ret, float *a){
_mm_stream_ps((float *)&out.v,in.v);
#endif
#ifdef AVX512
_mm512_stream_ps((float *)&out.v,in.v);
_mm512_storenrngo_ps((float *)&out.v,in.v);
// _mm512_stream_ps((float *)&out.v,in.v);
//Note v has a3 a2 a1 a0
#endif
#ifdef QPX