mirror of
https://github.com/paboyle/Grid.git
synced 2025-04-07 04:35:56 +01:00
Debugged assemply split phase with interior suppression
This commit is contained in:
parent
5812eb8a8c
commit
e3d0e31525
@ -406,7 +406,7 @@ void WilsonFermion5D<Impl>::DhopInternalOverlappedComms(StencilImpl & st, Lebesg
|
||||
// Load imbalance alert. Should use dynamic schedule OMP for loop
|
||||
// Perhaps create a list of only those sites with face work, and
|
||||
// load balance process the list.
|
||||
#if 1
|
||||
#if 0
|
||||
#pragma omp parallel
|
||||
{
|
||||
int nthreads = omp_get_num_threads();
|
||||
@ -425,13 +425,15 @@ void WilsonFermion5D<Impl>::DhopInternalOverlappedComms(StencilImpl & st, Lebesg
|
||||
#else
|
||||
DhopComputeTime2-=usecond();
|
||||
if (dag == DaggerYes) {
|
||||
parallel_for (int ss = 0; ss < U._grid->oSites(); ss++) {
|
||||
#pragma omp parallel for schedule(static,4)
|
||||
for (int ss = 0; ss < U._grid->oSites(); ss++) {
|
||||
int sU = ss;
|
||||
int sF = LLs * sU;
|
||||
Kernels::DhopSiteDag(st,lo,U,st.CommBuf(),sF,sU,LLs,1,in,out,0,1);
|
||||
}
|
||||
} else {
|
||||
parallel_for (int ss = 0; ss < U._grid->oSites(); ss++) {
|
||||
#pragma omp parallel for schedule(static,1)
|
||||
for (int ss = 0; ss < U._grid->oSites(); ss++) {
|
||||
int sU = ss;
|
||||
int sF = LLs * sU;
|
||||
Kernels::DhopSite(st,lo,U,st.CommBuf(),sF,sU,LLs,1,in,out,0,1);
|
||||
|
@ -33,8 +33,8 @@ directory
|
||||
namespace Grid {
|
||||
namespace QCD {
|
||||
|
||||
int WilsonKernelsStatic::Opt = WilsonKernelsStatic::OptGeneric;
|
||||
int WilsonKernelsStatic::Comms = WilsonKernelsStatic::CommsAndCompute;
|
||||
int WilsonKernelsStatic::Opt = WilsonKernelsStatic::OptGeneric;
|
||||
int WilsonKernelsStatic::Comms = WilsonKernelsStatic::CommsAndCompute;
|
||||
|
||||
template <class Impl>
|
||||
WilsonKernels<Impl>::WilsonKernels(const ImplParams &p) : Base(p){};
|
||||
|
Loading…
x
Reference in New Issue
Block a user