portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-04 11:15:55 +01:00
Code Releases Activity

Added streaming stores

Browse Source
This commit is contained in:
Peter Boyle 2015-05-05 18:09:28 +01:00
parent bf60764e4b
commit 5ebc7a1756
5 changed files with 71 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
lib/simd/Grid_vComplexD.h
View File

@ -242,6 +242,21 @@ friend inline void vstore(const vComplexD &ret, ComplexD *a){
assert(0);
#endif
}
friend inline void vstream(vComplexD &out,const vComplexD &in){
#if defined (AVX1)|| defined (AVX2)
_mm256_stream_pd((double *)&out.v,in.v);
#endif
#ifdef SSE4
_mm_stream_pd((double *)&out.v,in.v);
#endif
#ifdef AVX512
_mm512_stream_pd((double *)&out.v,in.v);
//Note v has a3 a2 a1 a0
#endif
#ifdef QPX
assert(0);
#endif
}
friend inline void vprefetch(const vComplexD &v)
{
_mm_prefetch((const char*)&v.v,_MM_HINT_T0);

29
lib/simd/Grid_vComplexF.h
View File

@ -176,21 +176,36 @@ namespace Grid {
vsplat(ret,a,b);
}
friend inline void vstore(const vComplexF &ret, ComplexF *a){
friend inline void vstore(const vComplexF &ret, ComplexF *a){
#if defined (AVX1)|| defined (AVX2)
_mm256_store_ps((float *)a,ret.v);
_mm256_store_ps((float *)a,ret.v);
#endif
#ifdef SSE4
_mm_store_ps((float *)a,ret.v);
_mm_store_ps((float *)a,ret.v);
#endif
#ifdef AVX512
_mm512_store_ps((float *)a,ret.v);
//Note v has a3 a2 a1 a0
_mm512_store_ps((float *)a,ret.v);
//Note v has a3 a2 a1 a0
#endif
#ifdef QPX
assert(0);
assert(0);
#endif
}
}
friend inline void vstream(vComplexF &out,const vComplexF &in){
#if defined (AVX1)|| defined (AVX2)
_mm256_stream_ps((float *)&out.v,in.v);
#endif
#ifdef SSE4
_mm_stream_ps((float *)&out.v,in.v);
#endif
#ifdef AVX512
_mm512_stream_ps((float *)&out.v,in.v);
//Note v has a3 a2 a1 a0
#endif
#ifdef QPX
assert(0);
#endif
}
friend inline void vprefetch(const vComplexF &v)
{
_mm_prefetch((const char*)&v.v,_MM_HINT_T0);

4
lib/simd/Grid_vInteger.h
View File

@ -186,6 +186,10 @@ namespace Grid {
#endif
}
friend inline void vstream(vInteger & out,const vInteger &in){
out=in;
}
friend inline void vprefetch(const vInteger &v)
{
_mm_prefetch((const char*)&v.v,_MM_HINT_T0);

15
lib/simd/Grid_vRealD.h
View File

@ -174,6 +174,21 @@ namespace Grid {
#endif
#ifdef QPX
assert(0);
#endif
}
friend inline void vstream(vRealD &out,const vRealD &in){
#if defined (AVX1)|| defined (AVX2)
_mm256_stream_pd((double *)&out.v,in.v);
#endif
#ifdef SSE4
_mm_stream_pd((double *)&out.v,in.v);
#endif
#ifdef AVX512
_mm512_stream_pd((double *)&out.v,in.v);
//Note v has a3 a2 a1 a0
#endif
#ifdef QPX
assert(0);
#endif
}
friend inline void vprefetch(const vRealD &v)

15
lib/simd/Grid_vRealF.h
View File

@ -208,6 +208,21 @@ friend inline void vstore(const vRealF &ret, float *a){
assert(0);
#endif
}
friend inline void vstream(vRealF &out,const vRealF &in){
#if defined (AVX1)|| defined (AVX2)
_mm256_stream_ps((float *)&out.v,in.v);
#endif
#ifdef SSE4
_mm_stream_ps((float *)&out.v,in.v);
#endif
#ifdef AVX512
_mm512_stream_ps((float *)&out.v,in.v);
//Note v has a3 a2 a1 a0
#endif
#ifdef QPX
assert(0);
#endif
}
friend inline void vprefetch(const vRealF &v)