portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-13 01:05:36 +00:00
Code Releases Activity

Tracing replaces self timing

Browse Source
This commit is contained in:
Peter Boyle 2022-08-31 17:11:48 -04:00
parent 24182ca8bf
commit 111b30ca1d
1 changed files with 0 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
Grid/qcd/action/fermion/implementation/CayleyFermion5Dcache.h
View File

@ -63,9 +63,6 @@ CayleyFermion5D<Impl>::M5D(const FermionField &psi_i,
// 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++;
M5Dtime-=usecond();
uint64_t nloop = grid->oSites(); uint64_t nloop = grid->oSites();
accelerator_for(sss,nloop,Simd::Nsimd(),{ accelerator_for(sss,nloop,Simd::Nsimd(),{
uint64_t s = sss%Ls; uint64_t s = sss%Ls;
@ -78,7 +75,6 @@ CayleyFermion5D<Impl>::M5D(const FermionField &psi_i,
spProj5p(tmp2,psi(idx_l)); spProj5p(tmp2,psi(idx_l));
coalescedWrite(chi[ss+s],pdiag[s]*phi(ss+s)+pupper[s]*tmp1+plower[s]*tmp2); coalescedWrite(chi[ss+s],pdiag[s]*phi(ss+s)+pupper[s]*tmp1+plower[s]*tmp2);
}); });
M5Dtime+=usecond();
} }
template<class Impl> template<class Impl>
@ -104,9 +100,6 @@ CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi_i,
int Ls=this->Ls; int Ls=this->Ls;
// Flops = 6.0*(Nc*Ns) *Ls*vol // Flops = 6.0*(Nc*Ns) *Ls*vol
M5Dcalls++;
M5Dtime-=usecond();
uint64_t nloop = grid->oSites(); uint64_t nloop = grid->oSites();
accelerator_for(sss,nloop,Simd::Nsimd(),{ accelerator_for(sss,nloop,Simd::Nsimd(),{
uint64_t s = sss%Ls; uint64_t s = sss%Ls;
@ -119,7 +112,6 @@ CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi_i,
spProj5m(tmp2,psi(idx_l)); spProj5m(tmp2,psi(idx_l));
coalescedWrite(chi[ss+s],pdiag[s]*phi(ss+s)+pupper[s]*tmp1+plower[s]*tmp2); coalescedWrite(chi[ss+s],pdiag[s]*phi(ss+s)+pupper[s]*tmp1+plower[s]*tmp2);
}); });
M5Dtime+=usecond();
} }
template<class Impl> template<class Impl>
@ -140,8 +132,6 @@ CayleyFermion5D<Impl>::MooeeInv (const FermionField &psi_i, FermionField &chi
auto pleem = & leem[0]; auto pleem = & leem[0];
auto pueem = & ueem[0]; auto pueem = & ueem[0];
MooeeInvCalls++;
MooeeInvTime-=usecond();
uint64_t nloop = grid->oSites()/Ls; uint64_t nloop = grid->oSites()/Ls;
accelerator_for(sss,nloop,Simd::Nsimd(),{ accelerator_for(sss,nloop,Simd::Nsimd(),{
uint64_t ss=sss*Ls; uint64_t ss=sss*Ls;
@ -178,8 +168,6 @@ CayleyFermion5D<Impl>::MooeeInv (const FermionField &psi_i, FermionField &chi
coalescedWrite(chi[ss+s],res); coalescedWrite(chi[ss+s],res);
} }
}); });
MooeeInvTime+=usecond();
} }
@ -202,10 +190,6 @@ CayleyFermion5D<Impl>::MooeeInvDag (const FermionField &psi_i, FermionField &chi
assert(psi.Checkerboard() == psi.Checkerboard()); assert(psi.Checkerboard() == psi.Checkerboard());
MooeeInvCalls++;
MooeeInvTime-=usecond();
uint64_t nloop = grid->oSites()/Ls; uint64_t nloop = grid->oSites()/Ls;
accelerator_for(sss,nloop,Simd::Nsimd(),{ accelerator_for(sss,nloop,Simd::Nsimd(),{
uint64_t ss=sss*Ls; uint64_t ss=sss*Ls;
@ -242,7 +226,6 @@ CayleyFermion5D<Impl>::MooeeInvDag (const FermionField &psi_i, FermionField &chi
coalescedWrite(chi[ss+s],res); coalescedWrite(chi[ss+s],res);
} }
}); });
MooeeInvTime+=usecond();
} }