portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-10 07:55:35 +00:00
Code Releases Activity

Partial optimisation of the extraction/merger of simd vecs.

Browse Source
This commit is contained in:
Peter Boyle 2015-11-06 05:26:20 -06:00
parent f9b2fce93b
commit 955b482aaf
1 changed files with 16 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
lib/tensors/Tensor_extract_merge.h
View File

@ -115,23 +115,21 @@ template<class vobj> inline void extract(const vobj &vec,std::vector<typename vo
template<class vobj> inline
void extract(const vobj &vec,std::vector<typename vobj::scalar_object *> &extracted, int offset)
{
typedef typename vobj::scalar_type scalar_type ;
typedef typename vobj::vector_type vector_type ;
const int words=sizeof(vobj)/sizeof(vector_type);
const int Nsimd=vobj::vector_type::Nsimd();
int Nextr=extracted.size();
int s = Nsimd/Nextr;
scalar_type * vp = (scalar_type *)&vec;
std::vector<scalar_type *> pointers(Nsimd);
for(int i=0;i<Nextr;i++) {
pointers[i] =(scalar_type *)& extracted[i][offset];
}
vector_type *vp = (vector_type *)&vec;
for(int w=0;w<words;w++){
extract<vector_type,scalar_type>(&vp[w],pointers,w);
for(int i=0;i<Nextr;i++){
scalar_type * pointer = (scalar_type *)& extracted[i][offset];
pointer[w] = vp[i*s+w*Nsimd];
}
}
}
@ -173,16 +171,19 @@ void merge(vobj &vec,std::vector<typename vobj::scalar_object *> &extracted,int
const int words=sizeof(vobj)/sizeof(vector_type);
int Nextr=extracted.size();
int s=Nsimd/Nextr;
std::vector<scalar_type *> pointers(Nextr);
for(int i=0;i<Nextr;i++)
pointers[i] =(scalar_type *)& extracted[i][offset];
vector_type *vp = (vector_type *)&vec;
scalar_type *pointer;
scalar_type *vp = (scalar_type *)&vec;
for(int w=0;w<words;w++){
merge<vector_type,scalar_type>(&vp[w],pointers,w);
for(int i=0;i<Nextr;i++){
for(int ii=0;ii<s;ii++){
pointer=(scalar_type *)&extracted[i][offset];
vp[w*Nsimd+i*s+ii] = pointer[w];
}
}
}
}
}
}
#endif