portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-09 23:45:36 +00:00
Code Releases Activity

correct throughput in wilson and dwf

Browse Source
This commit is contained in:
nmeyer-ur 2020-05-29 22:15:59 +02:00
parent 1732f9319e
commit 936071773e
2 changed files with 15 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
benchmarks/Benchmark_dwf.cc
View File

@ -201,11 +201,14 @@ int main (int argc, char ** argv)
double volume=Ls; for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
double flops=single_site_flops*volume*ncall;
// RF/L1: 4d Wilson
double data_L1 = (volume * 180 * 64 / 4 * ncall) / (1024.*1024.*1024.);
auto nsimd = vComplex::Nsimd();
auto simdwidth = sizeof(vComplex);
// L2 throughput
double data_L2 = (volume * 9 * 12 * 64 / 4 * ncall + (volume/Ls) * 8*9 * 64/4) / (1024.*1024.*1024.);
// RF: Nd Wilson * Ls, Nc colors
double data_L1 = (volume * ((2*Nd+1)*Nd*Nc + (2*Nd)*Nc*Nc) * simdwidth / nsimd * ncall) / (1024.*1024.*1024.);
// mem: Nd+1 fermion, Nd gauge, Nc colors
double data_L2 = (volume * (2*(Nd+1)+1)*Nd*Nc * simdwidth / nsimd * ncall + (volume/Ls) *(2*Nd)*Nc*Nc * simdwidth / nsimd) / (1024.*1024.*1024.);
std::cout<<GridLogMessage << "Called Dw "<<ncall<<" times in "<<t1-t0<<" us"<<std::endl;
// std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
@ -213,8 +216,8 @@ int main (int argc, char ** argv)
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t1-t0)<<std::endl;
std::cout<<GridLogMessage << "mflop/s per rank = "<< flops/(t1-t0)/NP<<std::endl;
std::cout<<GridLogMessage << "mflop/s per node = "<< flops/(t1-t0)/NN<<std::endl;
std::cout<<GridLogMessage << "RF/L1 GiB/s (base 2) = "<< 1000000. * data_L1/((t1-t0))<<std::endl;
std::cout<<GridLogMessage << "L2 GiB/s (base 2) = "<< 1000000. * data_L2/((t1-t0))<<std::endl;
std::cout<<GridLogMessage << "RF GiB/s (base 2) = "<< 1000000. * data_L1/((t1-t0))<<std::endl;
std::cout<<GridLogMessage << "mem GiB/s (base 2) = "<< 1000000. * data_L2/((t1-t0))<<std::endl;
err = ref-result;
std::cout<<GridLogMessage << "norm diff "<< norm2(err)<<std::endl;
//exit(0);

8
benchmarks/Benchmark_wilson.cc
View File

@ -179,14 +179,18 @@ int main (int argc, char ** argv)
}
double data = (volume * 180 * 64 / 4 * ncall) / (1024.*1024.*1024.);
auto nsimd = vComplex::Nsimd();
auto simdwidth = sizeof(vComplex);
// RF: Nd Wilson, Nc colors
double data = (volume * ((2*Nd+1)*Nd*Nc + (2*Nd)*Nc*Nc) * simdwidth / nsimd * ncall) / (1024.*1024.*1024.);
std::cout<<GridLogMessage << "Called Dw"<<std::endl;
std::cout<<GridLogMessage << "flops per site " << single_site_flops << std::endl;
std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
std::cout<<GridLogMessage << "norm ref "<< norm2(ref)<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t1-t0)<<std::endl;
//std::cout<<GridLogMessage << "GiB/s (base 2) = "<< 1000000. * data/((t1-t0))<<std::endl;
std::cout<<GridLogMessage << "RF GiB/s (base 2) = "<< 1000000. * data/(t1-t0)<<std::endl;
err = ref-result;
std::cout<<GridLogMessage << "norm diff "<< norm2(err)<<std::endl;