portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2026-06-04 11:14:38 +01:00
Code Releases Activity

benchmarks: add DhopEO benchmarks to Benchmark_staggered and Benchmark_staggeredF

Browse Source 
Also add NaiveStaggeredFermionF to Benchmark_staggeredF, and fix bug where
NaiveStaggered timed loop was calling Ds.Dhop instead of Dn.Dhop.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Peter Boyle
2026-05-29 13:55:27 -04:00
parent 42cd9eda71
commit 3fac61fc55
2 changed files with 72 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files
benchmarks/Benchmark_staggered.cc
+33 -25
View File
@@ -98,35 +98,43 @@ int main (int argc, char ** argv)
ImprovedStaggeredFermionD Ds(Umu,Umu,Grid,RBGrid,mass,c1,c2,u0,params); ImprovedStaggeredFermionD Ds(Umu,Umu,Grid,RBGrid,mass,c1,c2,u0,params);
NaiveStaggeredFermionD Dn(Umu,Grid,RBGrid,mass,c1,u0,params); NaiveStaggeredFermionD Dn(Umu,Grid,RBGrid,mass,c1,u0,params);
std::cout<<GridLogMessage << "Calling Ds"<<std::endl; FermionField src_e(&RBGrid); pickCheckerboard(Even, src_e, src);
int ncall=100; FermionField src_o(&RBGrid); pickCheckerboard(Odd, src_o, src);
// warm perf only FermionField res_e(&RBGrid); res_e = Zero();
for(int i=0;i<ncall;i++){
Ds.Dhop(src,result,0);
}
double t0=usecond();
for(int i=0;i<ncall;i++){
Ds.Dhop(src,result,0);
}
double t1=usecond();
double flops=(16*(3*(6+8+8)) + 15*3*2)*volume*ncall; // == 66*16 + 90 == 1146
std::cout<<GridLogMessage << "Called Ds"<<std::endl;
std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t1-t0)<<std::endl;
// Warm perf only int ncall=100;
for(int i=0;i<ncall;i++){
Dn.Dhop(src,result,0); // ImprovedStaggered Dhop
} for(int i=0;i<ncall;i++) Ds.Dhop(src,result,0);
double t0=usecond();
for(int i=0;i<ncall;i++) Ds.Dhop(src,result,0);
double t1=usecond();
double flops=(16*(3*(6+8+8)) + 15*3*2)*volume*ncall; // 1146 flops/site
std::cout<<GridLogMessage << "ImprovedStaggered Dhop mflop/s = "<< flops/(t1-t0)<<std::endl;
// ImprovedStaggered DhopEO
for(int i=0;i<ncall;i++) Ds.DhopEO(src_o,res_e,0);
t0=usecond(); t0=usecond();
for(int i=0;i<ncall;i++){ for(int i=0;i<ncall;i++) Ds.DhopEO(src_o,res_e,0);
Ds.Dhop(src,result,0); t1=usecond();
} flops=(16*(3*(6+8+8)) + 15*3*2)*(volume/2)*ncall;
std::cout<<GridLogMessage << "ImprovedStaggered DhopEO mflop/s = "<< flops/(t1-t0)<<std::endl;
// NaiveStaggered Dhop
for(int i=0;i<ncall;i++) Dn.Dhop(src,result,0);
t0=usecond();
for(int i=0;i<ncall;i++) Dn.Dhop(src,result,0);
t1=usecond(); t1=usecond();
flops=(8*(3*(6+8+8)) + 7*3*2)*volume*ncall; flops=(8*(3*(6+8+8)) + 7*3*2)*volume*ncall;
std::cout<<GridLogMessage << "Called Dn"<<std::endl; std::cout<<GridLogMessage << "NaiveStaggered Dhop mflop/s = "<< flops/(t1-t0)<<std::endl;
std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t1-t0)<<std::endl; // NaiveStaggered DhopEO
for(int i=0;i<ncall;i++) Dn.DhopEO(src_o,res_e,0);
t0=usecond();
for(int i=0;i<ncall;i++) Dn.DhopEO(src_o,res_e,0);
t1=usecond();
flops=(8*(3*(6+8+8)) + 7*3*2)*(volume/2)*ncall;
std::cout<<GridLogMessage << "NaiveStaggered DhopEO mflop/s = "<< flops/(t1-t0)<<std::endl;
Grid_finalize(); Grid_finalize();
} }
benchmarks/Benchmark_staggeredF.cc
+36 -13
View File
@@ -96,22 +96,45 @@ int main (int argc, char ** argv)
RealD c2=-1.0/24.0; RealD c2=-1.0/24.0;
RealD u0=1.0; RealD u0=1.0;
ImprovedStaggeredFermionF Ds(Umu,Umu,Grid,RBGrid,mass,c1,c2,u0,params); ImprovedStaggeredFermionF Ds(Umu,Umu,Grid,RBGrid,mass,c1,c2,u0,params);
NaiveStaggeredFermionF Dn(Umu,Grid,RBGrid,mass,c1,u0,params);
FermionField src_e(&RBGrid); pickCheckerboard(Even, src_e, src);
FermionField src_o(&RBGrid); pickCheckerboard(Odd, src_o, src);
FermionField res_e(&RBGrid); res_e = Zero();
std::cout<<GridLogMessage << "Calling Ds"<<std::endl;
int ncall=1000; int ncall=1000;
for(int i=0;i<ncall;i++){
Ds.Dhop(src,result,0);
}
double t0=usecond();
for(int i=0;i<ncall;i++){
Ds.Dhop(src,result,0);
}
double t1=usecond();
double flops=(16*(3*(6+8+8)) + 15*3*2)*volume*ncall; // == 66*16 + == 1146
std::cout<<GridLogMessage << "Called Ds"<<std::endl; // ImprovedStaggered Dhop
std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl; for(int i=0;i<ncall;i++) Ds.Dhop(src,result,0);
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t1-t0)<<std::endl; double t0=usecond();
for(int i=0;i<ncall;i++) Ds.Dhop(src,result,0);
double t1=usecond();
double flops=(16*(3*(6+8+8)) + 15*3*2)*volume*ncall; // 1146 flops/site
std::cout<<GridLogMessage << "ImprovedStaggered Dhop mflop/s = "<< flops/(t1-t0)<<std::endl;
// ImprovedStaggered DhopEO
for(int i=0;i<ncall;i++) Ds.DhopEO(src_o,res_e,0);
t0=usecond();
for(int i=0;i<ncall;i++) Ds.DhopEO(src_o,res_e,0);
t1=usecond();
flops=(16*(3*(6+8+8)) + 15*3*2)*(volume/2)*ncall;
std::cout<<GridLogMessage << "ImprovedStaggered DhopEO mflop/s = "<< flops/(t1-t0)<<std::endl;
// NaiveStaggered Dhop
for(int i=0;i<ncall;i++) Dn.Dhop(src,result,0);
t0=usecond();
for(int i=0;i<ncall;i++) Dn.Dhop(src,result,0);
t1=usecond();
flops=(8*(3*(6+8+8)) + 7*3*2)*volume*ncall;
std::cout<<GridLogMessage << "NaiveStaggered Dhop mflop/s = "<< flops/(t1-t0)<<std::endl;
// NaiveStaggered DhopEO
for(int i=0;i<ncall;i++) Dn.DhopEO(src_o,res_e,0);
t0=usecond();
for(int i=0;i<ncall;i++) Dn.DhopEO(src_o,res_e,0);
t1=usecond();
flops=(8*(3*(6+8+8)) + 7*3*2)*(volume/2)*ncall;
std::cout<<GridLogMessage << "NaiveStaggered DhopEO mflop/s = "<< flops/(t1-t0)<<std::endl;
Grid_finalize(); Grid_finalize();
} }