portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-15 06:17:05 +01:00
Code Releases Activity

Merge branch 'develop' into feature/gpu-port

Browse Source
This commit is contained in:
Peter Boyle
2018-12-13 05:11:34 +00:00
parent adbdc4e65b c509bd3fe2
commit b57a4d32aa
647 changed files with 49155 additions and 11160 deletions
Download Patch File Download Diff File Expand all files Collapse all files

78
benchmarks/Benchmark_su3.cc
View File

@ -124,6 +124,7 @@ int main (int argc, char ** argv)
std::cout<<GridLogMessage << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
for(int lat=LMIN;lat<=LMAX;lat+=LADD){
Coordinate latt_size ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
@ -187,5 +188,82 @@ int main (int argc, char ** argv)
}
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << "= Benchmarking SU3xSU3 CovShiftForward(z,x,y)"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
for(int lat=LMIN;lat<=LMAX;lat+=LADD){
std::vector<int> latt_size ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
int64_t vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
GridCartesian Grid(latt_size,simd_layout,mpi_layout);
GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9}));
LatticeColourMatrix z(&Grid); random(pRNG,z);
LatticeColourMatrix x(&Grid); random(pRNG,x);
LatticeColourMatrix y(&Grid); random(pRNG,y);
for(int mu=0;mu<4;mu++){
double start=usecond();
for(int64_t i=0;i<Nloop;i++){
z = PeriodicBC::CovShiftForward(x,mu,y);
}
double stop=usecond();
double time = (stop-start)/Nloop*1000.0;
double bytes=3*vol*Nc*Nc*sizeof(Complex);
double flops=Nc*Nc*(6+8+8)*vol;
std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
}
}
#if 1
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << "= Benchmarking SU3xSU3 z= x * Cshift(y)"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
for(int lat=LMIN;lat<=LMAX;lat+=LADD){
std::vector<int> latt_size ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
int64_t vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
GridCartesian Grid(latt_size,simd_layout,mpi_layout);
GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9}));
LatticeColourMatrix z(&Grid); random(pRNG,z);
LatticeColourMatrix x(&Grid); random(pRNG,x);
LatticeColourMatrix y(&Grid); random(pRNG,y);
LatticeColourMatrix tmp(&Grid);
for(int mu=0;mu<4;mu++){
double tshift=0;
double tmult =0;
double start=usecond();
for(int64_t i=0;i<Nloop;i++){
tshift-=usecond();
tmp = Cshift(y,mu,-1);
tshift+=usecond();
tmult-=usecond();
z = x*tmp;
tmult+=usecond();
}
double stop=usecond();
double time = (stop-start)/Nloop;
tshift = tshift/Nloop;
tmult = tmult /Nloop;
double bytes=3*vol*Nc*Nc*sizeof(Complex);
double flops=Nc*Nc*(6+8+8)*vol;
std::cout<<GridLogMessage<<std::setprecision(3) << "total us "<<time<<" shift "<<tshift <<" mult "<<tmult<<std::endl;
time = time * 1000; // convert to NS for GB/s
std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
}
}
#endif
Grid_finalize();
}