portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-11-04 05:54:32 +00:00
Code Releases Activity

Merge pull request #28 from paboyle/develop

Browse Source 
Sync with Upstream
This commit is contained in:
Christoph Lehner
2022-03-08 09:58:51 +01:00
committed by GitHub
parent 9616811c3d 694306f202
commit e2fc3a0f04
133 changed files with 6483 additions and 970 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
benchmarks/Benchmark_IO.cc
View File

@@ -137,7 +137,7 @@ int main (int argc, char ** argv)
Eigen::MatrixXd mean(nVol, 4), stdDev(nVol, 4), rob(nVol, 4);
Eigen::VectorXd avMean(4), avStdDev(4), avRob(4);
double n = BENCH_IO_NPASS;
// double n = BENCH_IO_NPASS;
stats(mean, stdDev, perf);
stats(avMean, avStdDev, avPerf);
@@ -164,7 +164,7 @@ int main (int argc, char ** argv)
mean(volInd(l), gWrite), stdDev(volInd(l), gWrite));
}
MSG << std::endl;
MSG << "Robustness of individual results, in \%. (rob = 100\% - std dev / mean)" << std::endl;
MSG << "Robustness of individual results, in %. (rob = 100% - std dev / mean)" << std::endl;
MSG << std::endl;
grid_printf("%4s %12s %12s %12s %12s\n",
"L", "std read", "std write", "Grid read", "Grid write");
@@ -185,7 +185,7 @@ int main (int argc, char ** argv)
avMean(sRead), avStdDev(sRead), avMean(sWrite), avStdDev(sWrite),
avMean(gRead), avStdDev(gRead), avMean(gWrite), avStdDev(gWrite));
MSG << std::endl;
MSG << "Robustness of volume-averaged results, in \%. (rob = 100\% - std dev / mean)" << std::endl;
MSG << "Robustness of volume-averaged results, in %. (rob = 100% - std dev / mean)" << std::endl;
MSG << std::endl;
grid_printf("%12s %12s %12s %12s\n",
"std read", "std write", "Grid read", "Grid write");

4
benchmarks/Benchmark_ITT.cc
View File

@@ -142,7 +142,7 @@ public:
// bzero((void *)rbuf[d],lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
}
int ncomm;
// int ncomm;
double dbytes;
for(int dir=0;dir<8;dir++) {
@@ -290,7 +290,7 @@ public:
LatticeSU4 z(&Grid); z=Zero();
LatticeSU4 x(&Grid); x=Zero();
LatticeSU4 y(&Grid); y=Zero();
double a=2.0;
// double a=2.0;
uint64_t Nloop=NLOOP;

4
benchmarks/Benchmark_comms_host_device.cc
View File

@@ -53,7 +53,7 @@ struct time_statistics{
void header(){
std::cout <<GridLogMessage << " L "<<"\t"<<" Ls "<<"\t"
<<std::setw(11)<<"bytes\t\t"<<"MB/s uni (err/min/max)"<<"\t\t"<<"MB/s bidi (err/min/max)"<<std::endl;
<<std::setw(11)<<"bytes\t\t"<<"MB/s uni"<<"\t"<<"MB/s bidi"<<std::endl;
};
int main (int argc, char ** argv)
@@ -72,7 +72,7 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage << "Number of iterations to average: "<< Nloop << std::endl;
std::vector<double> t_time(Nloop);
time_statistics timestat;
// time_statistics timestat;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << "= Benchmarking sequential halo exchange from host memory "<<std::endl;

63
benchmarks/Benchmark_dwf_fp32.cc
View File

@@ -126,19 +126,10 @@ int main (int argc, char ** argv)
// Naive wilson implementation
////////////////////////////////////
// replicate across fifth dimension
LatticeGaugeFieldF Umu5d(FGrid);
std::vector<LatticeColourMatrixF> U(4,FGrid);
{
autoView( Umu5d_v, Umu5d, CpuWrite);
autoView( Umu_v , Umu , CpuRead);
for(int ss=0;ss<Umu.Grid()->oSites();ss++){
for(int s=0;s<Ls;s++){
Umu5d_v[Ls*ss+s] = Umu_v[ss];
}
}
}
// LatticeGaugeFieldF Umu5d(FGrid);
std::vector<LatticeColourMatrixF> U(4,UGrid);
for(int mu=0;mu<Nd;mu++){
U[mu] = PeekIndex<LorentzIndex>(Umu5d,mu);
U[mu] = PeekIndex<LorentzIndex>(Umu,mu);
}
std::cout << GridLogMessage << "Setting up Cshift based reference " << std::endl;
@@ -147,10 +138,28 @@ int main (int argc, char ** argv)
ref = Zero();
for(int mu=0;mu<Nd;mu++){
tmp = U[mu]*Cshift(src,mu+1,1);
tmp = Cshift(src,mu+1,1);
{
autoView( tmp_v , tmp , CpuWrite);
autoView( U_v , U[mu] , CpuRead);
for(int ss=0;ss<U[mu].Grid()->oSites();ss++){
for(int s=0;s<Ls;s++){
tmp_v[Ls*ss+s] = U_v[ss]*tmp_v[Ls*ss+s];
}
}
}
ref=ref + tmp - Gamma(Gmu[mu])*tmp;
tmp =adj(U[mu])*src;
{
autoView( tmp_v , tmp , CpuWrite);
autoView( U_v , U[mu] , CpuRead);
autoView( src_v, src , CpuRead);
for(int ss=0;ss<U[mu].Grid()->oSites();ss++){
for(int s=0;s<Ls;s++){
tmp_v[Ls*ss+s] = adj(U_v[ss])*src_v[Ls*ss+s];
}
}
}
tmp =Cshift(tmp,mu+1,-1);
ref=ref + tmp + Gamma(Gmu[mu])*tmp;
}
@@ -182,7 +191,7 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
DomainWallFermionF Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
int ncall =3000;
int ncall =300;
if (1) {
FGrid->Barrier();
@@ -242,16 +251,30 @@ int main (int argc, char ** argv)
for(int mu=0;mu<Nd;mu++){
// ref = src - Gamma(Gamma::Algebra::GammaX)* src ; // 1+gamma_x
tmp = U[mu]*Cshift(src,mu+1,1);
tmp = Cshift(src,mu+1,1);
{
autoView( ref_v, ref, CpuWrite);
autoView( tmp_v, tmp, CpuRead);
for(int i=0;i<ref_v.size();i++){
ref_v[i]+= tmp_v[i] + Gamma(Gmu[mu])*tmp_v[i]; ;
autoView( U_v , U[mu] , CpuRead);
for(int ss=0;ss<U[mu].Grid()->oSites();ss++){
for(int s=0;s<Ls;s++){
int i=s+Ls*ss;
ref_v[i]+= U_v[ss]*(tmp_v[i] + Gamma(Gmu[mu])*tmp_v[i]); ;
}
}
}
tmp =adj(U[mu])*src;
{
autoView( tmp_v , tmp , CpuWrite);
autoView( U_v , U[mu] , CpuRead);
autoView( src_v, src , CpuRead);
for(int ss=0;ss<U[mu].Grid()->oSites();ss++){
for(int s=0;s<Ls;s++){
tmp_v[Ls*ss+s] = adj(U_v[ss])*src_v[Ls*ss+s];
}
}
}
// tmp =adj(U[mu])*src;
tmp =Cshift(tmp,mu+1,-1);
{
autoView( ref_v, ref, CpuWrite);

6
benchmarks/Benchmark_memory_bandwidth.cc
View File

@@ -184,8 +184,10 @@ int main (int argc, char ** argv)
double bytes=1.0*vol*Nvec*sizeof(Real);
double flops=vol*Nvec*2;// mul,add
std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t"<<flops/time<< "\t\t"<<(stop-start)/1000./1000.<< "\t\t " <<std::endl;
std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"
<<bytes<<" \t\t"<<bytes/time<<"\t\t"<<flops/time<< "\t\t"
<<(stop-start)/1000./1000.<< "\t\t " <<std::endl;
assert(nn==nn);
}
Grid_finalize();