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

Updating preparing for solvers etc..

Browse Source
This commit is contained in:
Peter Boyle
2015-05-16 23:35:08 +01:00
parent 9b0aae665f
commit dc6b6bdc96
8 changed files with 1170 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
benchmarks/Grid_memory_bandwidth.cc
View File

@@ -19,7 +19,7 @@ int main (int argc, char ** argv)
std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking fused AXPY bandwidth"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t"<<"GB/s"<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<std::endl;
std::cout << "----------------------------------------------------------"<<std::endl;
for(int lat=4;lat<=32;lat+=4){
@@ -44,15 +44,16 @@ int main (int argc, char ** argv)
double stop=usecond();
double time = (stop-start)/Nloop*1000;
double flops=lat*lat*lat*lat*Nvec*2;// mul,add
double bytes=3*lat*lat*lat*lat*Nvec*sizeof(Real);
std::cout<<std::setprecision(2) << lat<<"\t\t"<<bytes<<"\t\t"<<bytes/time<<std::endl;
std::cout<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t"<<flops/time<<std::endl;
}
std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking a*x + y bandwidth"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t"<<"GB/s"<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<std::endl;
std::cout << "----------------------------------------------------------"<<std::endl;
for(int lat=4;lat<=32;lat+=4){
@@ -75,15 +76,16 @@ int main (int argc, char ** argv)
double stop=usecond();
double time = (stop-start)/Nloop*1000;
double flops=lat*lat*lat*lat*Nvec*2;// mul,add
double bytes=3*lat*lat*lat*lat*Nvec*sizeof(Real);
std::cout<<std::setprecision(2) << lat<<"\t\t"<<bytes<<"\t\t"<<bytes/time<<std::endl;
std::cout<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t"<<flops/time<<std::endl;
}
std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking SCALE bandwidth"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t"<<"GB/s"<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<std::endl;
for(int lat=4;lat<=32;lat+=4){
@@ -107,14 +109,15 @@ int main (int argc, char ** argv)
double time = (stop-start)/Nloop*1000;
double bytes=2*lat*lat*lat*lat*Nvec*sizeof(Real);
std::cout <<std::setprecision(2) << lat<<"\t\t"<<bytes<<"\t\t"<<bytes/time<<std::endl;
double flops=lat*lat*lat*lat*Nvec*1;// mul
std::cout <<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t"<<flops/time<<std::endl;
}
std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking READ bandwidth"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t"<<"GB/s"<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<std::endl;
std::cout << "----------------------------------------------------------"<<std::endl;
for(int lat=4;lat<=32;lat+=4){
@@ -139,7 +142,8 @@ int main (int argc, char ** argv)
double time = (stop-start)/Nloop*1000;
double bytes=lat*lat*lat*lat*Nvec*sizeof(Real);
std::cout<<std::setprecision(2) << lat<<"\t\t"<<bytes<<"\t\t"<<bytes/time<<std::endl;
double flops=lat*lat*lat*lat*Nvec*2;// mul,add
std::cout<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t"<<flops/time<<std::endl;
}

16
benchmarks/Grid_su3.cc
View File

@@ -16,7 +16,7 @@ int main (int argc, char ** argv)
std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking SU3xSU3 x= x*y"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
std::cout << "----------------------------------------------------------"<<std::endl;
for(int lat=2;lat<=24;lat+=2){
@@ -40,7 +40,7 @@ int main (int argc, char ** argv)
double bytes=3.0*lat*lat*lat*lat*Nc*Nc*sizeof(Complex);
double footprint=2.0*lat*lat*lat*lat*Nc*Nc*sizeof(Complex);
double flops=Nc*Nc*(6.0+8.0+8.0)*lat*lat*lat*lat;
std::cout<<std::setprecision(3) << lat<<"\t\t"<<footprint<<"\t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
std::cout<<std::setprecision(3) << lat<<"\t\t"<<footprint<<" \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
}
@@ -48,7 +48,7 @@ int main (int argc, char ** argv)
std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking SU3xSU3 z= x*y"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
std::cout << "----------------------------------------------------------"<<std::endl;
for(int lat=2;lat<=24;lat+=2){
@@ -71,14 +71,14 @@ int main (int argc, char ** argv)
double bytes=3*lat*lat*lat*lat*Nc*Nc*sizeof(Complex);
double flops=Nc*Nc*(6+8+8)*lat*lat*lat*lat;
std::cout<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"\t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
std::cout<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
}
std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking SU3xSU3 mult(z,x,y)"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
std::cout << "----------------------------------------------------------"<<std::endl;
for(int lat=2;lat<=24;lat+=2){
@@ -101,14 +101,14 @@ int main (int argc, char ** argv)
double bytes=3*lat*lat*lat*lat*Nc*Nc*sizeof(Complex);
double flops=Nc*Nc*(6+8+8)*lat*lat*lat*lat;
std::cout<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"\t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
std::cout<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
}
std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking SU3xSU3 mac(z,x,y)"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
std::cout << "----------------------------------------------------------"<<std::endl;
for(int lat=2;lat<=24;lat+=2){
@@ -131,7 +131,7 @@ int main (int argc, char ** argv)
double bytes=3*lat*lat*lat*lat*Nc*Nc*sizeof(Complex);
double flops=Nc*Nc*(8+8+8)*lat*lat*lat*lat;
std::cout<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"\t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
std::cout<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
}