portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-12-23 06:04:29 +00:00
Code Releases Activity

Do the force term on the accelerator too. Needed particularly because comms buffers

Browse Source 
are device memory.
This commit is contained in:
Peter Boyle
2019-07-29 22:58:35 +01:00
parent 275c1c920f
commit 1282e1067f
4 changed files with 45 additions and 63 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h
View File

@@ -320,7 +320,26 @@ void WilsonKernels<Impl>::DhopDirK( StencilView &st, DoubledGaugeFieldView &U,Si
GENERIC_DHOPDIR_LEG(Tm,spProjTm,spReconTm);
coalescedWrite(out[sF], result,lane);
}
template <class Impl>
void WilsonKernels<Impl>::DhopDirKernel( StencilImpl &st, DoubledGaugeField &U,SiteHalfSpinor *buf, int Ls,
int Nsite, const FermionField &in, FermionField &out, int dirdisp, int gamma)
{
assert(dirdisp<=7);
assert(dirdisp>=0);
auto U_v = U.View();
auto in_v = in.View();
auto out_v = out.View();
auto st_v = st.View();
accelerator_for(ss,Nsite,1,{
for(int s=0;s<Ls;s++){
int sU=ss;
int sF = s+Ls*sU;
DhopDirK(st_v,U_v,st.CommBuf(),sF,sU,in_v,out_v,dirdisp,gamma);
}
});
}
#define KERNEL_CALLNB(A) \
const uint64_t NN = Nsite*Ls; \