portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-24 12:45:56 +01:00
Code Releases Activity

Domain decomposed benchmark

Browse Source
This commit is contained in:
Quadro 2021-06-09 16:35:24 -04:00
parent 87ec14c353
commit dcd48a0435
1 changed files with 62 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

74
benchmarks/Benchmark_dwf_fp32.cc
View File

@ -28,14 +28,14 @@
#ifdef CUDA_PROFILE #ifdef CUDA_PROFILE
#include <cuda_profiler_api.h> #include <cuda_profiler_api.h>
#endif #endif
#include <Grid/qcd/action/momentum/DirichletFilter.h>
#include <Grid/qcd/action/momentum/DDHMCfilter.h>
#include <Grid/qcd/action/fermion/DirichletFermionOperator.h>
using namespace std; using namespace std;
using namespace Grid; using namespace Grid;
template<class d> typedef DirichletFermionOperator<WilsonImplF> DirichletFermionF;
struct scal {
d internal;
};
Gamma::Algebra Gmu [] = { Gamma::Algebra Gmu [] = {
Gamma::Algebra::GammaX, Gamma::Algebra::GammaX,
@ -44,7 +44,6 @@ struct scal {
Gamma::Algebra::GammaT Gamma::Algebra::GammaT
}; };
int main (int argc, char ** argv) int main (int argc, char ** argv)
{ {
Grid_init(&argc,&argv); Grid_init(&argc,&argv);
@ -61,20 +60,17 @@ int main (int argc, char ** argv)
GridLogLayout(); GridLogLayout();
Coordinate latt = GridDefaultLatt();
Coordinate mpi = GridDefaultMpi();
Coordinate simd = GridDefaultSimd(Nd,vComplexF::Nsimd());
long unsigned int single_site_flops = 8*Nc*(7+16*Nc); long unsigned int single_site_flops = 8*Nc*(7+16*Nc);
GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(latt,simd,mpi);
GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplexF::Nsimd()),GridDefaultMpi());
GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid); GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid); GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid); GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
std::cout << GridLogMessage << "Making s innermost grids"<<std::endl;
GridCartesian * sUGrid = SpaceTimeGrid::makeFourDimDWFGrid(GridDefaultLatt(),GridDefaultMpi());
GridRedBlackCartesian * sUrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(sUGrid);
GridCartesian * sFGrid = SpaceTimeGrid::makeFiveDimDWFGrid(Ls,UGrid);
GridRedBlackCartesian * sFrbGrid = SpaceTimeGrid::makeFiveDimDWFRedBlackGrid(Ls,UGrid);
std::vector<int> seeds4({1,2,3,4}); std::vector<int> seeds4({1,2,3,4});
std::vector<int> seeds5({5,6,7,8}); std::vector<int> seeds5({5,6,7,8});
@ -292,7 +288,6 @@ int main (int argc, char ** argv)
std::cout<<GridLogMessage << "src_o"<<norm2(src_o)<<std::endl; std::cout<<GridLogMessage << "src_o"<<norm2(src_o)<<std::endl;
// S-direction is INNERMOST and takes no part in the parity.
std::cout << GridLogMessage<< "*********************************************************" <<std::endl; std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
std::cout << GridLogMessage<< "* Benchmarking DomainWallFermionF::DhopEO "<<std::endl; std::cout << GridLogMessage<< "* Benchmarking DomainWallFermionF::DhopEO "<<std::endl;
std::cout << GridLogMessage<< "* Vectorising space-time by "<<vComplexF::Nsimd()<<std::endl; std::cout << GridLogMessage<< "* Vectorising space-time by "<<vComplexF::Nsimd()<<std::endl;
@ -359,6 +354,57 @@ int main (int argc, char ** argv)
assert(norm2(src_e)<1.0e-4); assert(norm2(src_e)<1.0e-4);
assert(norm2(src_o)<1.0e-4); assert(norm2(src_o)<1.0e-4);
std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
std::cout << GridLogMessage<< "* Benchmarking DirichletFermionF::DhopEO "<<std::endl;
std::cout << GridLogMessage<< "* SINGLE precision "<<std::endl;
#ifdef GRID_OMP
if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsAndCompute ) std::cout << GridLogMessage<< "* Using Overlapped Comms/Compute" <<std::endl;
if ( WilsonKernelsStatic::Comms == WilsonKernelsStatic::CommsThenCompute) std::cout << GridLogMessage<< "* Using sequential comms compute" <<std::endl;
#endif
if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptGeneric ) std::cout << GridLogMessage<< "* Using GENERIC Nc WilsonKernels" <<std::endl;
if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptHandUnroll) std::cout << GridLogMessage<< "* Using Nc=3 WilsonKernels" <<std::endl;
if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptInlineAsm ) std::cout << GridLogMessage<< "* Using Asm Nc=3 WilsonKernels" <<std::endl;
std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
// Dirichlet benchmark
Coordinate local(Nd);
Coordinate nil(Nd,0);
for(int d=0;d<Nd;d++){
local[d] = latt[d]/mpi[d];
}
std::vector<Complex> boundary = {1,1,1,-1};
DomainWallFermionF::ImplParams DirichletParams(boundary);
DirichletParams.locally_periodic=true;
DomainWallFermionF DwDirichlet(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,DirichletParams);
DirichletFermionF Dirichlet(DwDirichlet,local,nil);
{
FGrid->Barrier();
Dirichlet.DhopEO(src_o,r_e,DaggerNo);
DwDirichlet.ZeroCounters();
double t0=usecond();
for(int i=0;i<ncall;i++){
Dirichlet.DhopEO(src_o,r_e,DaggerNo);
}
double t1=usecond();
FGrid->Barrier();
double volume=Ls; for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
double flops=(single_site_flops*volume*ncall)/2.0;
std::cout<<GridLogMessage << "DirichletDeo flop "<< flops<<" usec " <<(t1-t0)<<std::endl;
std::cout<<GridLogMessage << "DirichletDeo mflop/s = "<< flops/(t1-t0)<<std::endl;
std::cout<<GridLogMessage << "DirichletDeo mflop/s per rank "<< flops/(t1-t0)/NP<<std::endl;
std::cout<<GridLogMessage << "DirichletDeo mflop/s per node "<< flops/(t1-t0)/NN<<std::endl;
DwDirichlet.Report();
}
Grid_finalize(); Grid_finalize();
exit(0); exit(0);
} }