portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-10 07:55:35 +00:00
Code Releases Activity

M5D offloaded to GPU

Browse Source
This commit is contained in:
gfilaci 2019-05-09 11:17:55 +01:00
parent 2b3c22f03d
commit f1744b3f01
1 changed files with 26 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

48
Grid/qcd/action/fermion/CayleyFermion5Dgpu.cc
View File

@ -50,35 +50,39 @@ void CayleyFermion5D<Impl>::M5D(const FermionField &psi_i,
auto psi = psi_i.View(); auto psi = psi_i.View();
auto phi = phi_i.View(); auto phi = phi_i.View();
auto chi = chi_i.View(); auto chi = chi_i.View();
Coeff_t *lower_v = &lower[0];
Coeff_t *diag_v = &diag[0];
Coeff_t *upper_v = &upper[0];
int Ls =this->Ls; int Ls =this->Ls;
assert(phi.Checkerboard() == psi.Checkerboard()); assert(phi.Checkerboard() == psi.Checkerboard());
const uint64_t nsimd = grid->Nsimd();
const uint64_t sites4d = nsimd * grid->oSites() / Ls;
// 10 = 3 complex mult + 2 complex add // 10 = 3 complex mult + 2 complex add
// Flops = 10.0*(Nc*Ns) *Ls*vol (/2 for red black counting) // Flops = 10.0*(Nc*Ns) *Ls*vol (/2 for red black counting)
M5Dcalls++; M5Dcalls++;
M5Dtime-=usecond(); M5Dtime-=usecond();
thread_loop( (int ss=0;ss<grid->oSites();ss+=Ls),{ // adds Ls typedef typename SiteSpinor::scalar_object ScalarSiteSpinor;
accelerator_loopN( sss, sites4d ,{
uint64_t lane = sss % nsimd;
uint64_t ss = Ls * (sss / nsimd);
for(int s=0;s<Ls;s++){ for(int s=0;s<Ls;s++){
auto tmp = psi[0]; auto res = extractLane(lane,phi[ss+s]);
if ( s==0 ) { res = diag_v[s]*res;
spProj5m(tmp,psi[ss+s+1]);
chi[ss+s]=diag[s]*phi[ss+s]+upper[s]*tmp; auto tmp = extractLane(lane,psi[ss+(s+1)%Ls]);
spProj5m(tmp,tmp);
spProj5p(tmp,psi[ss+Ls-1]); res += upper_v[s]*tmp;
chi[ss+s]=chi[ss+s]+lower[s]*tmp;
} else if ( s==(Ls-1)) { tmp = extractLane(lane,psi[ss+(s+Ls-1)%Ls]);
spProj5m(tmp,psi[ss+0]); spProj5p(tmp,tmp);
chi[ss+s]=diag[s]*phi[ss+s]+upper[s]*tmp; res += lower_v[s]*tmp;
spProj5p(tmp,psi[ss+s-1]); insertLane(lane,chi[ss+s],res);
chi[ss+s]=chi[ss+s]+lower[s]*tmp;
} else {
spProj5m(tmp,psi[ss+s+1]);
chi[ss+s]=diag[s]*phi[ss+s]+upper[s]*tmp;
spProj5p(tmp,psi[ss+s-1]);
chi[ss+s]=chi[ss+s]+lower[s]*tmp;
}
} }
}); });
M5Dtime+=usecond(); M5Dtime+=usecond();