portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-09-20 09:15:38 +01:00
Code Releases Activity

Improvements to staggered tests timings

Browse Source
This commit is contained in:
Azusa Yamaguchi 2018-04-26 10:08:05 +01:00
parent 3e125c5b61
commit a9e8758a01
3 changed files with 111 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

50
tests/solver/Test_staggered_block_cg_prec.cc
View File

@ -74,6 +74,11 @@ int main (int argc, char ** argv)
LatticeGaugeField Umu(UGrid); SU3::HotConfiguration(pRNG,Umu);
double volume=1;
for(int mu=0;mu<Nd;mu++){
volume=volume*latt_size[mu];
}
RealD mass=0.003;
RealD c1=9.0/8.0;
RealD c2=-1.0/24.0;
@ -90,14 +95,26 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage << "****************************************************************** "<<std::endl;
std::cout << GridLogMessage << " Calling 4d CG "<<std::endl;
std::cout << GridLogMessage << "****************************************************************** "<<std::endl;
ImprovedStaggeredFermionR Ds4d(Umu,Umu,*UGrid,*UrbGrid,mass);
ImprovedStaggeredFermionR Ds4d(Umu,Umu,*UGrid,*UrbGrid,mass,c1,c2,u0);
SchurStaggeredOperator<ImprovedStaggeredFermionR,FermionField> HermOp4d(Ds4d);
FermionField src4d(UGrid); random(pRNG,src4d);
FermionField src4d_o(UrbGrid); pickCheckerboard(Odd,src4d_o,src4d);
FermionField result4d_o(UrbGrid);
double deodoe_flops=(16*(3*(6+8+8)) + 15*3*2)*volume; // == 66*16 + == 1146
result4d_o=zero;
{
double t1=usecond();
CG(HermOp4d,src4d_o,result4d_o);
double t2=usecond();
double ncall=CG.IterationsToComplete;
double flops = deodoe_flops * ncall;
std::cout<<GridLogMessage << "usec = "<< (t2-t1)<<std::endl;
std::cout<<GridLogMessage << "flops = "<< flops<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t2-t1)<<std::endl;
HermOp4d.Report();
}
Ds4d.Report();
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
@ -106,7 +123,17 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
Ds.ZeroCounters();
result_o=zero;
{
double t1=usecond();
CG(HermOp,src_o,result_o);
double t2=usecond();
double ncall=CG.IterationsToComplete*Ls;
double flops = deodoe_flops * ncall;
std::cout<<GridLogMessage << "usec = "<< (t2-t1)<<std::endl;
std::cout<<GridLogMessage << "flops = "<< flops<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t2-t1)<<std::endl;
HermOp.Report();
}
Ds.Report();
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
@ -115,7 +142,18 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
Ds.ZeroCounters();
result_o=zero;
{
double t1=usecond();
mCG(HermOp,src_o,result_o);
double t2=usecond();
double ncall=mCG.IterationsToComplete*Ls;
double flops = deodoe_flops * ncall;
std::cout<<GridLogMessage << "usec = "<< (t2-t1)<<std::endl;
std::cout<<GridLogMessage << "flops = "<< flops<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t2-t1)<<std::endl;
HermOp.Report();
}
Ds.Report();
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
@ -124,7 +162,17 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
Ds.ZeroCounters();
result_o=zero;
{
double t1=usecond();
BCGrQ(HermOp,src_o,result_o);
double t2=usecond();
double ncall=BCGrQ.IterationsToComplete*Ls;
double flops = deodoe_flops * ncall;
std::cout<<GridLogMessage << "usec = "<< (t2-t1)<<std::endl;
std::cout<<GridLogMessage << "flops = "<< flops<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t2-t1)<<std::endl;
HermOp.Report();
}
Ds.Report();
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;

45
tests/solver/Test_staggered_block_cg_unprec.cc
View File

@ -77,6 +77,12 @@ int main (int argc, char ** argv)
RealD c1=9.0/8.0;
RealD c2=-1.0/24.0;
RealD u0=1.0;
double volume=1;
for(int mu=0;mu<Nd;mu++){
volume=volume*latt_size[mu];
}
ImprovedStaggeredFermion5DR Ds(Umu,Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,c1,c2,u0);
MdagMLinearOperator<ImprovedStaggeredFermion5DR,FermionField> HermOp(Ds);
@ -93,7 +99,19 @@ int main (int argc, char ** argv)
MdagMLinearOperator<ImprovedStaggeredFermionR,FermionField> HermOp4d(Ds4d);
FermionField src4d(UGrid); random(pRNG,src4d);
FermionField result4d(UGrid); result4d=zero;
double deodoe_flops=(16*(3*(6+8+8)) + 15*3*2)*volume; // == 66*16 + == 1146
{
double t1=usecond();
CG(HermOp4d,src4d,result4d);
double t2=usecond();
double ncall=CG.IterationsToComplete;
double flops = deodoe_flops * ncall;
std::cout<<GridLogMessage << "usec = "<< (t2-t1)<<std::endl;
std::cout<<GridLogMessage << "flops = "<< flops<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t2-t1)<<std::endl;
}
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
@ -101,9 +119,18 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage << " Calling 5d CG for "<<Ls <<" right hand sides" <<std::endl;
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
result=zero;
{
Ds.ZeroCounters();
double t1=usecond();
CG(HermOp,src,result);
double t2=usecond();
double ncall=CG.IterationsToComplete;
double flops = deodoe_flops * ncall;
std::cout<<GridLogMessage << "usec = "<< (t2-t1)<<std::endl;
std::cout<<GridLogMessage << "flops = "<< flops<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t2-t1)<<std::endl;
Ds.Report();
}
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
@ -111,7 +138,16 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
result=zero;
Ds.ZeroCounters();
{
double t1=usecond();
mCG(HermOp,src,result);
double t2=usecond();
double ncall=CG.IterationsToComplete;
double flops = deodoe_flops * ncall;
std::cout<<GridLogMessage << "usec = "<< (t2-t1)<<std::endl;
std::cout<<GridLogMessage << "flops = "<< flops<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t2-t1)<<std::endl;
}
Ds.Report();
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
@ -120,7 +156,16 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;
result=zero;
Ds.ZeroCounters();
{
double t1=usecond();
BCGrQ(HermOp,src,result);
double t2=usecond();
double ncall=CG.IterationsToComplete;
double flops = deodoe_flops * ncall;
std::cout<<GridLogMessage << "usec = "<< (t2-t1)<<std::endl;
std::cout<<GridLogMessage << "flops = "<< flops<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t2-t1)<<std::endl;
}
Ds.Report();
std::cout << GridLogMessage << "************************************************************************ "<<std::endl;

12
tests/solver/Test_staggered_cg_prec.cc
View File

@ -83,7 +83,19 @@ int main (int argc, char ** argv)
SchurStaggeredOperator<ImprovedStaggeredFermionR,FermionField> HermOpEO(Ds);
ConjugateGradient<FermionField> CG(1.0e-8,10000);
double t1=usecond();
CG(HermOpEO,src_o,res_o);
double t2=usecond();
// Schur solver: uses DeoDoe => volume * 1146
double ncall=CG.IterationsToComplete;
double flops=(16*(3*(6+8+8)) + 15*3*2)*volume*ncall; // == 66*16 + == 1146
std::cout<<GridLogMessage << "usec = "<< (t2-t1)<<std::endl;
std::cout<<GridLogMessage << "flops = "<< flops<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t2-t1)<<std::endl;
FermionField tmp(&RBGrid);