portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-11 14:40:46 +01:00
Code Releases Activity

Debugged assemply split phase with interior suppression

Browse Source
This commit is contained in:
Peter Boyle 2017-04-23 19:29:27 -04:00
parent 5812eb8a8c
commit e3d0e31525
2 changed files with 7 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
lib/qcd/action/fermion/WilsonFermion5D.cc
View File

@ -406,7 +406,7 @@ void WilsonFermion5D<Impl>::DhopInternalOverlappedComms(StencilImpl & st, Lebesg
// Load imbalance alert. Should use dynamic schedule OMP for loop // Load imbalance alert. Should use dynamic schedule OMP for loop
// Perhaps create a list of only those sites with face work, and // Perhaps create a list of only those sites with face work, and
// load balance process the list. // load balance process the list.
#if 1 #if 0
#pragma omp parallel #pragma omp parallel
{ {
int nthreads = omp_get_num_threads(); int nthreads = omp_get_num_threads();
@ -425,13 +425,15 @@ void WilsonFermion5D<Impl>::DhopInternalOverlappedComms(StencilImpl & st, Lebesg
#else #else
DhopComputeTime2-=usecond(); DhopComputeTime2-=usecond();
if (dag == DaggerYes) { 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 sU = ss;
int sF = LLs * sU; int sF = LLs * sU;
Kernels::DhopSiteDag(st,lo,U,st.CommBuf(),sF,sU,LLs,1,in,out,0,1); Kernels::DhopSiteDag(st,lo,U,st.CommBuf(),sF,sU,LLs,1,in,out,0,1);
} }
} else { } 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 sU = ss;
int sF = LLs * sU; int sF = LLs * sU;
Kernels::DhopSite(st,lo,U,st.CommBuf(),sF,sU,LLs,1,in,out,0,1); Kernels::DhopSite(st,lo,U,st.CommBuf(),sF,sU,LLs,1,in,out,0,1);

4
lib/qcd/action/fermion/WilsonKernels.cc
View File

@ -33,8 +33,8 @@ directory
namespace Grid { namespace Grid {
namespace QCD { namespace QCD {
int WilsonKernelsStatic::Opt = WilsonKernelsStatic::OptGeneric; int WilsonKernelsStatic::Opt = WilsonKernelsStatic::OptGeneric;
int WilsonKernelsStatic::Comms = WilsonKernelsStatic::CommsAndCompute; int WilsonKernelsStatic::Comms = WilsonKernelsStatic::CommsAndCompute;
template <class Impl> template <class Impl>
WilsonKernels<Impl>::WilsonKernels(const ImplParams &p) : Base(p){}; WilsonKernels<Impl>::WilsonKernels(const ImplParams &p) : Base(p){};