portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-17 23:37:06 +01:00
Code Releases Activity

Regressing to not overlap comms and compute becasue bluewaters, edison, and cori are so rubbish at it.

Browse Source
This commit is contained in:
Peter Boyle
2016-02-16 13:56:44 -06:00
parent 340a29b735
commit 81395e85d1
3 changed files with 290 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@ -280,11 +280,11 @@ void WilsonFermion5D<Impl>::DhopInternal(StencilImpl & st, LebesgueOrder &lo,
DoubledGaugeField & U,
const FermionField &in, FermionField &out,int dag)
{
if ( Impl::overlapCommsCompute () ) {
DhopInternalCommsOverlapCompute(st,lo,U,in,out,dag);
} else {
// if ( Impl::overlapCommsCompute () ) {
// DhopInternalCommsOverlapCompute(st,lo,U,in,out,dag);
// } else {
DhopInternalCommsThenCompute(st,lo,U,in,out,dag);
}
// }
}
template<class Impl>
@ -319,7 +319,7 @@ void WilsonFermion5D<Impl>::DhopInternalCommsThenCompute(StencilImpl & st, Lebes
dslashtime -=usecond();
if ( dag == DaggerYes ) {
if( this->HandOptDslash ) {
#pragma omp parallel for schedule(static)
PARALLEL_FOR_LOOP
for(int ss=0;ss<U._grid->oSites();ss++){
int sU=ss;
for(int s=0;s<Ls;s++){
@ -398,7 +398,7 @@ PARALLEL_FOR_LOOP
}
*/
#pragma omp parallel for schedule(static)
PARALLEL_FOR_LOOP
for(int ss=0;ss<U._grid->oSites();ss++){
int sU=ss;
for(int s=0;s<Ls;s++){
@ -426,6 +426,7 @@ void WilsonFermion5D<Impl>::DhopInternalCommsOverlapCompute(StencilImpl & st, Le
DoubledGaugeField & U,
const FermionField &in, FermionField &out,int dag)
{
assert(0);
// assert((dag==DaggerNo) ||(dag==DaggerYes));
alltime-=usecond();