portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-09 21:50:45 +01:00
Code Releases Activity

fast local peek-poke

Browse Source
This commit is contained in:
Antonin Portelli 2016-10-24 19:24:21 +01:00
parent 20a091c3ed
commit 392e064513
1 changed files with 14 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
lib/lattice/Lattice_peekpoke.h
View File

@ -154,7 +154,7 @@ PARALLEL_FOR_LOOP
template<class vobj,class sobj> template<class vobj,class sobj>
void peekLocalSite(sobj &s,const Lattice<vobj> &l,std::vector<int> &site){ void peekLocalSite(sobj &s,const Lattice<vobj> &l,std::vector<int> &site){
GridBase *grid=l._grid; GridBase *grid = l._grid;
typedef typename vobj::scalar_type scalar_type; typedef typename vobj::scalar_type scalar_type;
typedef typename vobj::vector_type vector_type; typedef typename vobj::vector_type vector_type;
@ -164,16 +164,18 @@ PARALLEL_FOR_LOOP
assert( l.checkerboard== l._grid->CheckerBoard(site)); assert( l.checkerboard== l._grid->CheckerBoard(site));
assert( sizeof(sobj)*Nsimd == sizeof(vobj)); assert( sizeof(sobj)*Nsimd == sizeof(vobj));
static const int words=sizeof(vobj)/sizeof(vector_type);
int odx,idx; int odx,idx;
idx= grid->iIndex(site); idx= grid->iIndex(site);
odx= grid->oIndex(site); odx= grid->oIndex(site);
std::vector<sobj> buf(Nsimd); scalar_type * vp = (scalar_type *)&l._odata[odx];
scalar_type * pt = (scalar_type *)&s;
extract(l._odata[odx],buf);
for(int w=0;w<words;w++){
pt[w] = vp[idx+w*Nsimd];
}
s = buf[idx];
return; return;
}; };
@ -190,18 +192,17 @@ PARALLEL_FOR_LOOP
assert( l.checkerboard== l._grid->CheckerBoard(site)); assert( l.checkerboard== l._grid->CheckerBoard(site));
assert( sizeof(sobj)*Nsimd == sizeof(vobj)); assert( sizeof(sobj)*Nsimd == sizeof(vobj));
static const int words=sizeof(vobj)/sizeof(vector_type);
int odx,idx; int odx,idx;
idx= grid->iIndex(site); idx= grid->iIndex(site);
odx= grid->oIndex(site); odx= grid->oIndex(site);
std::vector<sobj> buf(Nsimd); scalar_type * vp = (scalar_type *)&l._odata[odx];
scalar_type * pt = (scalar_type *)&s;
// extract-modify-merge cycle is easiest way and this is not perf critical
extract(l._odata[odx],buf);
buf[idx] = s; for(int w=0;w<words;w++){
vp[idx+w*Nsimd] = pt[w];
merge(l._odata[odx],buf); }
return; return;
}; };