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

Debug hack; raises from 62GF/s to 72 GF/s per node on BG/Q

Browse Source
This commit is contained in:
Peter Boyle 2016-12-22 17:50:14 +00:00
parent 5241245534
commit b8cdb3e90a
1 changed files with 13 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
lib/qcd/action/fermion/WilsonFermion5D.cc
View File

@ -437,12 +437,25 @@ void WilsonFermion5D<Impl>::DhopInternal(StencilImpl & st, LebesgueOrder &lo,
stat.accum(nthreads); stat.accum(nthreads);
#endif #endif
} else { } else {
#if 0
PARALLEL_FOR_LOOP PARALLEL_FOR_LOOP
for (int ss = 0; ss < U._grid->oSites(); ss++) { for (int ss = 0; ss < U._grid->oSites(); ss++) {
int sU = ss; int sU = ss;
int sF = LLs * sU; int sF = LLs * sU;
Kernels::DiracOptDhopSite(st,lo,U,st.CommBuf(),sF,sU,LLs,1,in,out); Kernels::DiracOptDhopSite(st,lo,U,st.CommBuf(),sF,sU,LLs,1,in,out);
} }
#else
#pragma omp parallel
{
for(int i=0;i<10;i++){
int me, myoff,mywork;
int len = U._grid->oSites();
GridThread::GetWorkBarrier(len,me, mywork,myoff);
int sF = LLs * myoff;
Kernels::DiracOptDhopSite(st,lo,U,st.CommBuf(),sF,myoff,LLs,mywork,in,out);
}
}
#endif
} }
DhopComputeTime+=usecond(); DhopComputeTime+=usecond();
} }