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

Faster copy region

Browse Source
This commit is contained in:
Peter Boyle 2020-04-10 11:10:52 -04:00
parent a65bc64f10
commit 6cdb09c884
1 changed files with 29 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

35
Grid/lattice/Lattice_transfer.h
View File

@ -443,9 +443,16 @@ template<class vobj>
void localCopyRegion(const Lattice<vobj> &From,Lattice<vobj> & To,Coordinate FromLowerLeft, Coordinate ToLowerLeft, Coordinate RegionSize)
{
typedef typename vobj::scalar_object sobj;
typedef typename vobj::scalar_type scalar_type;
typedef typename vobj::vector_type vector_type;
static const int words=sizeof(vobj)/sizeof(vector_type);
GridBase *Fg = From.Grid();
GridBase *Tg = To.Grid();
assert(!Fg->_isCheckerBoarded);
assert(!Tg->_isCheckerBoarded);
int Nsimd = Fg->Nsimd();
int nF = Fg->_ndimension;
int nT = Tg->_ndimension;
int nd = nF;
@ -453,17 +460,24 @@ void localCopyRegion(const Lattice<vobj> &From,Lattice<vobj> & To,Coordinate Fro
for(int d=0;d<nd;d++){
assert(Fg->_processors[d] == Tg->_processors[d]);
// assert(Fg->_ldimensions[d] == Tg->_ldimensions[d]);
}
// the above should guarantee that the operations are local
thread_for(idx,Fg->lSites(),{
Coordinate ldf = Fg->_ldimensions;
Coordinate rdf = Fg->_rdimensions;
Coordinate isf = Fg->_istride;
Coordinate osf = Fg->_ostride;
Coordinate rdt = Tg->_rdimensions;
Coordinate ist = Tg->_istride;
Coordinate ost = Tg->_ostride;
auto t_v = To.View();
auto f_v = From.View();
accelerator_for(idx,Fg->lSites(),1,{
sobj s;
Coordinate Fcoor(nd);
Coordinate Tcoor(nd);
Fg->LocalIndexToLocalCoor(idx,Fcoor);
Lexicographic::CoorFromIndex(Fcoor,idx,ldf);
int in_region=1;
// can now repromote to threaded loop
for(int d=0;d<nd;d++){
if ( (Fcoor[d] < FromLowerLeft[d]) || (Fcoor[d]>=FromLowerLeft[d]+RegionSize[d]) ){
in_region=0;
@ -471,8 +485,17 @@ void localCopyRegion(const Lattice<vobj> &From,Lattice<vobj> & To,Coordinate Fro
Tcoor[d] = ToLowerLeft[d]+ Fcoor[d]-FromLowerLeft[d];
}
if (in_region) {
peekLocalSite(s,From,Fcoor);
pokeLocalSite(s,To ,Tcoor);
Integer idx_f = 0; for(int d=0;d<nd;d++) idx_f+=isf[d]*(Fcoor[d]/rdf[d]);
Integer idx_t = 0; for(int d=0;d<nd;d++) idx_t+=ist[d]*(Tcoor[d]/rdt[d]);
Integer odx_f = 0; for(int d=0;d<nd;d++) odx_f+=osf[d]*(Fcoor[d]%rdf[d]);
Integer odx_t = 0; for(int d=0;d<nd;d++) odx_t+=ost[d]*(Tcoor[d]%rdt[d]);
scalar_type * fp = (scalar_type *)&f_v[odx_f];
scalar_type * tp = (scalar_type *)&t_v[odx_t];
for(int w=0;w<words;w++){
tp[idx_t+w*Nsimd] = fp[idx_f+w*Nsimd]; // FIXME IF RRII layout, type pun no worke
}
// peekLocalSite(s,From,Fcoor);
// pokeLocalSite(s,To ,Tcoor);
}
});
}