portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-04 19:25:56 +01:00
Code Releases Activity

Merge branch 'develop' into gpu/baryons

Browse Source
This commit is contained in:
Felix Erben 2021-01-27 21:17:37 +00:00
commit 1bf2e4d187
3 changed files with 9 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
Grid/qcd/action/fermion/WilsonImpl.h
View File

@ -106,11 +106,15 @@ public:
const _SpinorField & phi, const _SpinorField & phi,
int mu) int mu)
{ {
const int Nsimd = SiteHalfSpinor::Nsimd();
autoView( out_v, out, AcceleratorWrite); autoView( out_v, out, AcceleratorWrite);
autoView( phi_v, phi, AcceleratorRead); autoView( phi_v, phi, AcceleratorRead);
autoView( Umu_v, Umu, AcceleratorRead); autoView( Umu_v, Umu, AcceleratorRead);
accelerator_for(sss,out.Grid()->oSites(),1,{ typedef decltype(coalescedRead(out_v[0])) calcSpinor;
multLink(out_v[sss],Umu_v[sss],phi_v[sss],mu); accelerator_for(sss,out.Grid()->oSites(),Nsimd,{
calcSpinor tmp;
multLink(tmp,Umu_v[sss],phi_v(sss),mu);
coalescedWrite(out_v[sss],tmp);
}); });
} }

2
benchmarks/Benchmark_ITT.cc
View File

@ -445,7 +445,7 @@ public:
// 1344= 3*(2*8+6)*2*8 + 8*3*2*2 + 3*4*2*8 // 1344= 3*(2*8+6)*2*8 + 8*3*2*2 + 3*4*2*8
// 1344 = Nc* (6+(Nc-1)*8)*2*Nd + Nd*Nc*2*2 + Nd*Nc*Ns*2 // 1344 = Nc* (6+(Nc-1)*8)*2*Nd + Nd*Nc*2*2 + Nd*Nc*Ns*2
// double flops=(1344.0*volume)/2; // double flops=(1344.0*volume)/2;
#if 1 #if 0
double fps = Nc* (6+(Nc-1)*8)*Ns*Nd + Nd*Nc*Ns + Nd*Nc*Ns*2; double fps = Nc* (6+(Nc-1)*8)*Ns*Nd + Nd*Nc*Ns + Nd*Nc*Ns*2;
#else #else
double fps = Nc* (6+(Nc-1)*8)*Ns*Nd + 2*Nd*Nc*Ns + 2*Nd*Nc*Ns*2; double fps = Nc* (6+(Nc-1)*8)*Ns*Nd + 2*Nd*Nc*Ns + 2*Nd*Nc*Ns*2;

4
tests/debug/Test_cayley_mres.cc
View File

@ -117,8 +117,8 @@ int main (int argc, char ** argv)
else else
{ {
std::cout<<GridLogMessage <<"Using cold configuration"<<std::endl; std::cout<<GridLogMessage <<"Using cold configuration"<<std::endl;
SU<Nc>::ColdConfiguration(Umu); //SU<Nc>::ColdConfiguration(Umu);
// SU<Nc>::HotConfiguration(RNG4,Umu); SU<Nc>::HotConfiguration(RNG4,Umu);
} }
RealD mass=0.3; RealD mass=0.3;