portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-15 14:27:06 +01:00
Code Releases Activity

Implemented mixed precision CG. Fixed filelist to exclude lib/Old directory and include Config.h.

Browse Source
This commit is contained in:
Christopher Kelly
2016-07-06 15:57:04 -04:00
parent df5c788ef2
commit 85ed8175cb
9 changed files with 311 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
lib/tensors/Tensor_extract_merge.h
View File

@ -10,6 +10,7 @@ Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: neo <cossu@post.kek.jp>
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Christopher Kelly <ckelly@phys.columbia.edu>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
@ -167,6 +168,33 @@ void extract(const vobj &vec,std::vector<typename vobj::scalar_object *> &extrac
}
}
////////////////////////////////////////////////////////////////////////
// Extract to a bunch of scalar object pointers of different scalar type, with offset. Useful for precision change
////////////////////////////////////////////////////////////////////////
template<class vobj, class sobj> inline
void extract1(const vobj &vec,std::vector<sobj*> &extracted, int offset)
{
typedef typename vobj::scalar_type vobj_scalar_type ;
typedef typename vobj::vector_type vobj_vector_type ;
typedef typename sobj::scalar_type sobj_scalar_type ;
static const int words=sizeof(vobj)/sizeof(vobj_vector_type);
static const int Nsimd=vobj_vector_type::Nsimd();
int Nextr=extracted.size();
int s = Nsimd/Nextr;
vobj_scalar_type * vp = (vobj_scalar_type *)&vec;
for(int w=0;w<words;w++){
for(int i=0;i<Nextr;i++){
sobj_scalar_type * pointer = (sobj_scalar_type *)& extracted[i][offset];
pointer[w] = vp[i*s+w*Nsimd];
}
}
}
////////////////////////////////////////////////////////////////////////
// Merge a contiguous array of scalar objects
////////////////////////////////////////////////////////////////////////