portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-11-04 05:54:32 +00:00
Code Releases Activity

Timing reports in benchmarks now reflect the asynch comms thread statistics

Browse Source
This commit is contained in:
paboyle
2016-01-04 14:42:16 +00:00
parent 02452afd36
commit c99d748da6
3 changed files with 10 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -102,6 +102,7 @@ WilsonFermion5D<Impl>::WilsonFermion5D(GaugeField &_Umu,
ImportGauge(_Umu);
commtime=0;
jointime=0;
dslashtime=0;
}
template<class Impl>
@@ -237,6 +238,7 @@ void WilsonFermion5D<Impl>::Report(void)
std::cout<<GridLogMessage << "********************"<<std::endl;
std::cout<<GridLogMessage << "Halo time "<<commtime <<" us"<<std::endl;
std::cout<<GridLogMessage << "Dslash time "<<dslashtime<<" us"<<std::endl;
std::cout<<GridLogMessage << "join time "<<jointime<<" us"<<std::endl;
std::cout<<GridLogMessage << "Stencil All time "<<Stencil.halotime<<" us"<<std::endl;
std::cout<<GridLogMessage << "********************"<<std::endl;
std::cout<<GridLogMessage << "Stencil nosplice time "<<Stencil.nosplicetime<<" us"<<std::endl;
@@ -297,8 +299,12 @@ void WilsonFermion5D<Impl>::DhopInternalCommsThenCompute(StencilImpl & st, Lebes
int nwork = U._grid->oSites();
commtime -=usecond();
st.HaloExchange(in,comm_buf,compressor);
std::thread thr = st.HaloExchangeBegin(in,comm_buf,compressor);
commtime +=usecond();
jointime -=usecond();
thr.join();
jointime +=usecond();
// Dhop takes the 4d grid from U, and makes a 5d index for fermion
// Not loop ordering and data layout.
@@ -483,9 +489,9 @@ PARALLEL_FOR_LOOP
}
dslashtime +=usecond();
commtime -=usecond();
jointime -=usecond();
thr.join();
commtime +=usecond();
jointime +=usecond();
local = false;
nonlocal = true;