portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-04 19:25:56 +01:00
Code Releases Activity

Remove lebesgue order; stick in stencil if need

Browse Source
This commit is contained in:
Peter Boyle 2018-03-22 18:13:41 -04:00
parent 334bb6792f
commit 23c880b009
1 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
lib/qcd/action/fermion/WilsonFermion5D.cc
View File

@ -445,13 +445,13 @@ void WilsonFermion5D<Impl>::DhopInternalOverlappedComms(StencilImpl & st, Lebesg
for (int ss = myblock; ss < myblock+myn; ++ss) { for (int ss = myblock; ss < myblock+myn; ++ss) {
int sU = ss; int sU = ss;
int sF = LLs * sU; int sF = LLs * sU;
Kernels::DhopSiteDag(Opt,st,lo,U_v,st.CommBuf(),sF,sU,LLs,1,in_v,out_v,1,0); Kernels::DhopSiteDag(Opt,st,U_v,st.CommBuf(),sF,sU,LLs,1,in_v,out_v,1,0);
} }
} else { } else {
for (int ss = myblock; ss < myblock+myn; ++ss) { for (int ss = myblock; ss < myblock+myn; ++ss) {
int sU = ss; int sU = ss;
int sF = LLs * sU; int sF = LLs * sU;
Kernels::DhopSite(Opt,st,lo,U_v,st.CommBuf(),sF,sU,LLs,1,in_v,out_v,1,0); Kernels::DhopSite(Opt,st,U_v,st.CommBuf(),sF,sU,LLs,1,in_v,out_v,1,0);
} }
} }
ptime = usecond() - start; ptime = usecond() - start;
@ -480,14 +480,14 @@ void WilsonFermion5D<Impl>::DhopInternalOverlappedComms(StencilImpl & st, Lebesg
thread_loop( (int ss = 0; ss < sz; ss++) ,{ thread_loop( (int ss = 0; ss < sz; ss++) ,{
int sU = st.surface_list[ss]; int sU = st.surface_list[ss];
int sF = LLs * sU; int sF = LLs * sU;
Kernels::DhopSiteDag(Opt,st,lo,U_v,st.CommBuf(),sF,sU,LLs,1,in_v,out_v,0,1); Kernels::DhopSiteDag(Opt,st,U_v,st.CommBuf(),sF,sU,LLs,1,in_v,out_v,0,1);
}); });
} else { } else {
int sz=st.surface_list.size(); int sz=st.surface_list.size();
thread_loop( (int ss = 0; ss < sz; ss++) ,{ thread_loop( (int ss = 0; ss < sz; ss++) ,{
int sU = st.surface_list[ss]; int sU = st.surface_list[ss];
int sF = LLs * sU; int sF = LLs * sU;
Kernels::DhopSite(Opt,st,lo,U_v,st.CommBuf(),sF,sU,LLs,1,in_v,out_v,0,1); Kernels::DhopSite(Opt,st,U_v,st.CommBuf(),sF,sU,LLs,1,in_v,out_v,0,1);
}); });
} }
DhopComputeTime2+=usecond(); DhopComputeTime2+=usecond();
@ -522,13 +522,13 @@ void WilsonFermion5D<Impl>::DhopInternalSerialComms(StencilImpl & st, LebesgueOr
accelerator_loop( ss, U_v, { accelerator_loop( ss, U_v, {
int sU = ss; int sU = ss;
int sF = LLs * sU; int sF = LLs * sU;
Kernels::DhopSiteDag(Opt,st,lo,U_v,st.CommBuf(),sF,sU,LLs,1,in_v,out_v); Kernels::DhopSiteDag(Opt,st,U_v,st.CommBuf(),sF,sU,LLs,1,in_v,out_v);
}); });
} else { } else {
accelerator_loop( ss, U_v , { accelerator_loop( ss, U_v , {
int sU = ss; int sU = ss;
int sF = LLs * sU; int sF = LLs * sU;
Kernels::DhopSite(Opt,st,lo,U_v,st.CommBuf(),sF,sU,LLs,1,in_v,out_v); Kernels::DhopSite(Opt,st,U_v,st.CommBuf(),sF,sU,LLs,1,in_v,out_v);
}); });
} }
DhopComputeTime+=usecond(); DhopComputeTime+=usecond();