portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-22 17:52:02 +01:00
Code Releases Activity

Compiles GPU and CPU, still gives good performance on CPU

Browse Source
This commit is contained in:
Peter Boyle
2019-06-05 13:28:16 +01:00
parent 18d3cde29a
commit 0ee6e77cbc
71 changed files with 1512 additions and 33769 deletions
Download Patch File Download Diff File Expand all files Collapse all files

93
Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h
View File

@ -375,78 +375,47 @@ void WilsonFermion<Impl>::DhopInternalOverlappedComms(StencilImpl &st, LebesgueO
const FermionField &in,
FermionField &out, int dag) {
assert((dag == DaggerNo) || (dag == DaggerYes));
#ifdef GRID_OMP
Compressor compressor(dag);
int len = U.Grid()->oSites();
const int LLs = 1;
/////////////////////////////
// Start comms // Gather intranode and extra node differentiated??
/////////////////////////////
std::vector<std::vector<CommsRequest_t> > requests;
st.Prepare();
st.HaloGather(in,compressor);
st.CommunicateBegin(requests);
/////////////////////////////
// Overlap with comms
/////////////////////////////
st.CommsMergeSHM(compressor);
#pragma omp parallel
{
int tid = omp_get_thread_num();
int nthreads = omp_get_num_threads();
int ncomms = CartesianCommunicator::nCommThreads;
if (ncomms == -1) ncomms = 1;
assert(nthreads > ncomms);
if (tid >= ncomms) {
nthreads -= ncomms;
int ttid = tid - ncomms;
int n = len;
int chunk = n / nthreads;
int rem = n % nthreads;
int myblock, myn;
if (ttid < rem) {
myblock = ttid * chunk + ttid;
myn = chunk+1;
} else {
myblock = ttid*chunk + rem;
myn = chunk;
}
// do the compute
auto U_v = U.View();
auto in_v = in.View();
auto out_v = out.View();
auto st_v = st.View();
int Opt = WilsonKernelsStatic::Opt;
if (dag == DaggerYes) {
for (int sss = myblock; sss < myblock+myn; ++sss) {
Kernels::DhopSiteDag(Opt,st_v,U_v,st.CommBuf(),sss,sss,1,1,in_v,out_v,1,0);
// Kernels::DhopSiteDag(st_v, lo, U_v, st.CommBuf(), sss, sss, 1, 1, in_v, out_v);
}
} else {
for (int sss = myblock; sss < myblock+myn; ++sss) {
Kernels::DhopSite(Opt,st_v,U_v,st.CommBuf(),sss,sss,1,1,in_v,out_v,1,0);
// Kernels::DhopSite(st_v, lo, U_v, st.CommBuf(), sss, sss, 1, 1, in_v, out_v);
}
}
/////////////////////////////
// do the compute interior
/////////////////////////////
int Opt = WilsonKernelsStatic::Opt;
if (dag == DaggerYes) {
Kernels::DhopDagKernel(Opt,st,U,st.CommBuf(),1,U.oSites(),in,out,1,0);
} else {
Kernels::DhopKernel(Opt,st,U,st.CommBuf(),1,U.oSites(),in,out,1,0);
}
} else {
st.CommunicateThreaded();
}
} //pragma
/////////////////////////////
// Complete comms
/////////////////////////////
st.CommunicateComplete(requests);
st.CommsMerge(compressor);
{
auto U_v = U.View();
auto in_v = in.View();
auto out_v = out.View();
auto st_v = st.View();
int Opt = WilsonKernelsStatic::Opt;
if (dag == DaggerYes) {
thread_loop( (int sss = 0; sss < in.Grid()->oSites(); sss++) ,{
Kernels::DhopSiteDag(Opt,st_v,U_v,st.CommBuf(),sss,sss,1,1,in_v,out_v,0,1);
});
} else {
thread_loop( (int sss = 0; sss < in.Grid()->oSites(); sss++) ,{
Kernels::DhopSite(Opt,st_v,U_v,st.CommBuf(),sss,sss,1,1,in_v,out_v,0,1);
});
}
/////////////////////////////
// do the compute exterior
/////////////////////////////
if (dag == DaggerYes) {
Kernels::DhopDagKernel(Opt,st,U,st.CommBuf(),1,U.oSites(),in,out,0,1);
} else {
Kernels::DhopKernel(Opt,st,U,st.CommBuf(),1,U.oSites(),in,out,0,1);
}
#else
assert(0);
#endif
};