mirror of
https://github.com/paboyle/Grid.git
synced 2024-11-09 23:45:36 +00:00
Added a decoupled memory system benchmark to remove thread synch overhead
This commit is contained in:
Peter Boyle
parent
64d64d1ab6
commit
9f4f65cb46
76
benchmarks/Benchmark_memory_asynch.cc
Normal file
76
benchmarks/Benchmark_memory_asynch.cc
Normal file
@ -0,0 +1,76 @@
|
||||
#include <Grid.h>
|
||||
|
||||
using namespace std;
|
||||
using namespace Grid;
|
||||
using namespace Grid::QCD;
|
||||
|
||||
int main (int argc, char ** argv)
|
||||
{
|
||||
Grid_init(&argc,&argv);
|
||||
|
||||
const int Nvec=4;
|
||||
typedef Lattice< iVector< vReal,Nvec> > LatticeVec;
|
||||
typedef iVector<vReal,Nvec> Vec;
|
||||
|
||||
|
||||
std::vector<int> simd_layout = GridDefaultSimd(Nd,vReal::Nsimd());
|
||||
std::vector<int> mpi_layout = GridDefaultMpi();
|
||||
|
||||
int threads = GridThread::GetThreads();
|
||||
std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
|
||||
|
||||
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
|
||||
std::cout<<GridLogMessage << "= Benchmarking READ bandwidth"<<std::endl;
|
||||
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
|
||||
std::cout<<GridLogMessage << " L "<<"\t\t"<<"bytes"<<"\t\t"<<"bytes/thread"<<"\t\t\t"<<"GB/s"<<"\t\t\t"<<"GB/s per thread"<<std::endl;
|
||||
std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
|
||||
|
||||
for(int lat=4;lat<=16536;lat*=2){
|
||||
|
||||
int Nloop=16536*1024*4/lat;
|
||||
|
||||
std::vector<int> latt_size ({2*mpi_layout[0],2*mpi_layout[1],4*mpi_layout[2],lat*mpi_layout[3]});
|
||||
|
||||
GridCartesian Grid(latt_size,simd_layout,mpi_layout);
|
||||
|
||||
int vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3]*threads;
|
||||
|
||||
Vec tsum; tsum = zero;
|
||||
|
||||
std::vector<double> stop(threads);
|
||||
Vector<Vec> sum(threads);
|
||||
|
||||
std::vector<LatticeVec> x(threads,&Grid);
|
||||
|
||||
|
||||
double start=usecond();
|
||||
PARALLEL_FOR_LOOP
|
||||
for(int t=0;t<threads;t++){
|
||||
|
||||
sum[t] = x[t]._odata[0];
|
||||
for(int i=0;i<Nloop;i++){
|
||||
for(auto ss=x[t].begin();ss<x[t].end();ss++){
|
||||
sum[t] = sum[t]+x[t]._odata[ss];
|
||||
}
|
||||
}
|
||||
stop[t]=usecond();
|
||||
}
|
||||
|
||||
double max_stop=stop[0];
|
||||
double min_stop=stop[0];
|
||||
|
||||
for(int t=0;t<threads;t++){
|
||||
tsum+=sum[t];
|
||||
if ( stop[t]<min_stop ) min_stop=stop[t];
|
||||
if ( stop[t]>max_stop ) max_stop=stop[t];
|
||||
}
|
||||
double max_time = (max_stop-start)/Nloop*1000;
|
||||
double min_time = (min_stop-start)/Nloop*1000;
|
||||
|
||||
double bytes=vol*Nvec*sizeof(Real);
|
||||
std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"\t\t"<<bytes/threads<<"\t\t"<<bytes/max_time<<" - "<< bytes/min_time<<"\t\t"<<bytes/min_time/threads <<std::endl;
|
||||
|
||||
}
|
||||
|
||||
Grid_finalize();
|
||||
}
|
||||
@ -1,5 +1,5 @@
|
||||
|
||||
bin_PROGRAMS = Benchmark_comms Benchmark_dwf Benchmark_memory_bandwidth Benchmark_su3 Benchmark_wilson
|
||||
bin_PROGRAMS = Benchmark_comms Benchmark_dwf Benchmark_memory_asynch Benchmark_memory_bandwidth Benchmark_su3 Benchmark_wilson
|
||||
|
||||
|
||||
Benchmark_comms_SOURCES=Benchmark_comms.cc
|
||||
@ -10,6 +10,10 @@ Benchmark_dwf_SOURCES=Benchmark_dwf.cc
|
||||
Benchmark_dwf_LDADD=-lGrid
|
||||
|
||||
|
||||
Benchmark_memory_asynch_SOURCES=Benchmark_memory_asynch.cc
|
||||
Benchmark_memory_asynch_LDADD=-lGrid
|
||||
|
||||
|
||||
Benchmark_memory_bandwidth_SOURCES=Benchmark_memory_bandwidth.cc
|
||||
Benchmark_memory_bandwidth_LDADD=-lGrid
|
||||
|
||||
|
||||
@ -1,4 +1,4 @@
|
||||
|
||||
HFILES=./algorithms/approx/bigfloat.h ./algorithms/approx/bigfloat_double.h ./algorithms/approx/Chebyshev.h ./algorithms/approx/MultiShiftFunction.h ./algorithms/approx/Remez.h ./algorithms/approx/Zolotarev.h ./algorithms/CoarsenedMatrix.h ./algorithms/iterative/AdefGeneric.h ./algorithms/iterative/ConjugateGradient.h ./algorithms/iterative/ConjugateGradientMultiShift.h ./algorithms/iterative/ConjugateResidual.h ./algorithms/iterative/NormalEquations.h ./algorithms/iterative/PrecConjugateResidual.h ./algorithms/iterative/PrecGeneralisedConjugateResidual.h ./algorithms/iterative/SchurRedBlack.h ./algorithms/LinearOperator.h ./algorithms/Preconditioner.h ./algorithms/SparseMatrix.h ./Algorithms.h ./AlignedAllocator.h ./cartesian/Cartesian_base.h ./cartesian/Cartesian_full.h ./cartesian/Cartesian_red_black.h ./Cartesian.h ./communicator/Communicator_base.h ./Communicator.h ./Config.h ./cshift/Cshift_common.h ./cshift/Cshift_mpi.h ./cshift/Cshift_none.h ./Cshift.h ./Grid.h ./Init.h ./lattice/Lattice_arith.h ./lattice/Lattice_base.h ./lattice/Lattice_comparison.h ./lattice/Lattice_comparison_utils.h ./lattice/Lattice_conformable.h ./lattice/Lattice_coordinate.h ./lattice/Lattice_ET.h ./lattice/Lattice_local.h ./lattice/Lattice_overload.h ./lattice/Lattice_peekpoke.h ./lattice/Lattice_reality.h ./lattice/Lattice_reduction.h ./lattice/Lattice_rng.h ./lattice/Lattice_trace.h ./lattice/Lattice_transfer.h ./lattice/Lattice_transpose.h ./lattice/Lattice_unary.h ./lattice/Lattice_where.h ./Lattice.h ./Log.h ./Old/Tensor_peek.h ./Old/Tensor_poke.h ./parallelIO/BinaryIO.h ./parallelIO/NerscIO.h ./pugixml/pugixml.h ./qcd/action/ActionBase.h ./qcd/action/ActionParams.h ./qcd/action/Actions.h ./qcd/action/fermion/CayleyFermion5D.h ./qcd/action/fermion/ContinuedFractionFermion5D.h ./qcd/action/fermion/DomainWallFermion.h ./qcd/action/fermion/FermionOperator.h ./qcd/action/fermion/FermionOperatorImpl.h ./qcd/action/fermion/g5HermitianLinop.h ./qcd/action/fermion/MobiusFermion.h ./qcd/action/fermion/MobiusZolotarevFermion.h ./qcd/action/fermion/OverlapWilsonCayleyTanhFermion.h ./qcd/action/fermion/OverlapWilsonCayleyZolotarevFermion.h ./qcd/action/fermion/OverlapWilsonContfracTanhFermion.h ./qcd/action/fermion/OverlapWilsonContfracZolotarevFermion.h ./qcd/action/fermion/OverlapWilsonPartialFractionTanhFermion.h ./qcd/action/fermion/OverlapWilsonPartialFractionZolotarevFermion.h ./qcd/action/fermion/PartialFractionFermion5D.h ./qcd/action/fermion/ScaledShamirFermion.h ./qcd/action/fermion/ShamirZolotarevFermion.h ./qcd/action/fermion/WilsonCompressor.h ./qcd/action/fermion/WilsonFermion.h ./qcd/action/fermion/WilsonFermion5D.h ./qcd/action/fermion/WilsonKernels.h ./qcd/action/gauge/WilsonGaugeAction.h ./qcd/action/pseudofermion/EvenOddSchurDifferentiable.h ./qcd/action/pseudofermion/OneFlavourEvenOddRational.h ./qcd/action/pseudofermion/OneFlavourEvenOddRationalRatio.h ./qcd/action/pseudofermion/OneFlavourRational.h ./qcd/action/pseudofermion/OneFlavourRationalRatio.h ./qcd/action/pseudofermion/TwoFlavour.h ./qcd/action/pseudofermion/TwoFlavourEvenOdd.h ./qcd/action/pseudofermion/TwoFlavourEvenOddRatio.h ./qcd/action/pseudofermion/TwoFlavourRatio.h ./qcd/hmc/HMC.h ./qcd/hmc/integrators/Integrator.h ./qcd/hmc/integrators/Integrator_algorithm.h ./qcd/QCD.h ./qcd/spin/Dirac.h ./qcd/spin/TwoSpinor.h ./qcd/utils/CovariantCshift.h ./qcd/utils/LinalgUtils.h ./qcd/utils/SpaceTimeGrid.h ./qcd/utils/SUn.h ./qcd/utils/WilsonLoops.h ./serialisation/BaseIO.h ./serialisation/BinaryIO.h ./serialisation/MacroMagic.h ./serialisation/Serialisation.h ./serialisation/TextIO.h ./serialisation/XmlIO.h ./simd/Grid_avx.h ./simd/Grid_avx512.h ./simd/Grid_empty.h ./simd/Grid_neon.h ./simd/Grid_qpx.h ./simd/Grid_sse4.h ./simd/Grid_vector_types.h ./simd/Grid_vector_unops.h ./Simd.h ./stencil/Lebesgue.h ./Stencil.h ./tensors/Tensor_arith.h ./tensors/Tensor_arith_add.h ./tensors/Tensor_arith_mac.h ./tensors/Tensor_arith_mul.h ./tensors/Tensor_arith_scalar.h ./tensors/Tensor_arith_sub.h ./tensors/Tensor_class.h ./tensors/Tensor_determinant.h ./tensors/Tensor_exp.h ./tensors/Tensor_extract_merge.h ./tensors/Tensor_index.h ./tensors/Tensor_inner.h ./tensors/Tensor_logical.h ./tensors/Tensor_outer.h ./tensors/Tensor_reality.h ./tensors/Tensor_Ta.h ./tensors/Tensor_trace.h ./tensors/Tensor_traits.h ./tensors/Tensor_transpose.h ./tensors/Tensor_unary.h ./Tensors.h ./Threads.h ./Timer.h
|
||||
HFILES=./algorithms/approx/bigfloat.h ./algorithms/approx/bigfloat_double.h ./algorithms/approx/Chebyshev.h ./algorithms/approx/MultiShiftFunction.h ./algorithms/approx/Remez.h ./algorithms/approx/Zolotarev.h ./algorithms/CoarsenedMatrix.h ./algorithms/iterative/AdefGeneric.h ./algorithms/iterative/ConjugateGradient.h ./algorithms/iterative/ConjugateGradientMultiShift.h ./algorithms/iterative/ConjugateResidual.h ./algorithms/iterative/ImplicitlyRestartedLanczos.h ./algorithms/iterative/MatrixUtils.h ./algorithms/iterative/NormalEquations.h ./algorithms/iterative/PrecConjugateResidual.h ./algorithms/iterative/PrecGeneralisedConjugateResidual.h ./algorithms/iterative/SchurRedBlack.h ./algorithms/LinearOperator.h ./algorithms/Preconditioner.h ./algorithms/SparseMatrix.h ./Algorithms.h ./AlignedAllocator.h ./cartesian/Cartesian_base.h ./cartesian/Cartesian_full.h ./cartesian/Cartesian_red_black.h ./Cartesian.h ./communicator/Communicator_base.h ./Communicator.h ./Config.h ./cshift/Cshift_common.h ./cshift/Cshift_mpi.h ./cshift/Cshift_none.h ./Cshift.h ./Grid.h ./Init.h ./lattice/Lattice_arith.h ./lattice/Lattice_base.h ./lattice/Lattice_comparison.h ./lattice/Lattice_comparison_utils.h ./lattice/Lattice_conformable.h ./lattice/Lattice_coordinate.h ./lattice/Lattice_ET.h ./lattice/Lattice_local.h ./lattice/Lattice_overload.h ./lattice/Lattice_peekpoke.h ./lattice/Lattice_reality.h ./lattice/Lattice_reduction.h ./lattice/Lattice_rng.h ./lattice/Lattice_trace.h ./lattice/Lattice_transfer.h ./lattice/Lattice_transpose.h ./lattice/Lattice_unary.h ./lattice/Lattice_where.h ./Lattice.h ./Log.h ./Old/Tensor_peek.h ./Old/Tensor_poke.h ./parallelIO/BinaryIO.h ./parallelIO/NerscIO.h ./pugixml/pugixml.h ./qcd/action/ActionBase.h ./qcd/action/ActionParams.h ./qcd/action/Actions.h ./qcd/action/fermion/CayleyFermion5D.h ./qcd/action/fermion/ContinuedFractionFermion5D.h ./qcd/action/fermion/DomainWallFermion.h ./qcd/action/fermion/FermionOperator.h ./qcd/action/fermion/FermionOperatorImpl.h ./qcd/action/fermion/g5HermitianLinop.h ./qcd/action/fermion/MobiusFermion.h ./qcd/action/fermion/MobiusZolotarevFermion.h ./qcd/action/fermion/OverlapWilsonCayleyTanhFermion.h ./qcd/action/fermion/OverlapWilsonCayleyZolotarevFermion.h ./qcd/action/fermion/OverlapWilsonContfracTanhFermion.h ./qcd/action/fermion/OverlapWilsonContfracZolotarevFermion.h ./qcd/action/fermion/OverlapWilsonPartialFractionTanhFermion.h ./qcd/action/fermion/OverlapWilsonPartialFractionZolotarevFermion.h ./qcd/action/fermion/PartialFractionFermion5D.h ./qcd/action/fermion/ScaledShamirFermion.h ./qcd/action/fermion/ShamirZolotarevFermion.h ./qcd/action/fermion/WilsonCompressor.h ./qcd/action/fermion/WilsonFermion.h ./qcd/action/fermion/WilsonFermion5D.h ./qcd/action/fermion/WilsonKernels.h ./qcd/action/gauge/WilsonGaugeAction.h ./qcd/action/pseudofermion/EvenOddSchurDifferentiable.h ./qcd/action/pseudofermion/OneFlavourEvenOddRational.h ./qcd/action/pseudofermion/OneFlavourEvenOddRationalRatio.h ./qcd/action/pseudofermion/OneFlavourRational.h ./qcd/action/pseudofermion/OneFlavourRationalRatio.h ./qcd/action/pseudofermion/TwoFlavour.h ./qcd/action/pseudofermion/TwoFlavourEvenOdd.h ./qcd/action/pseudofermion/TwoFlavourEvenOddRatio.h ./qcd/action/pseudofermion/TwoFlavourRatio.h ./qcd/hmc/HMC.h ./qcd/hmc/integrators/Integrator.h ./qcd/hmc/integrators/Integrator_algorithm.h ./qcd/QCD.h ./qcd/spin/Dirac.h ./qcd/spin/TwoSpinor.h ./qcd/utils/CovariantCshift.h ./qcd/utils/LinalgUtils.h ./qcd/utils/SpaceTimeGrid.h ./qcd/utils/SUn.h ./qcd/utils/WilsonLoops.h ./serialisation/BaseIO.h ./serialisation/BinaryIO.h ./serialisation/MacroMagic.h ./serialisation/Serialisation.h ./serialisation/TextIO.h ./serialisation/XmlIO.h ./simd/Grid_avx.h ./simd/Grid_avx512.h ./simd/Grid_empty.h ./simd/Grid_imci.h ./simd/Grid_neon.h ./simd/Grid_qpx.h ./simd/Grid_sse4.h ./simd/Grid_vector_types.h ./simd/Grid_vector_unops.h ./Simd.h ./stencil/Lebesgue.h ./Stencil.h ./tensors/Tensor_arith.h ./tensors/Tensor_arith_add.h ./tensors/Tensor_arith_mac.h ./tensors/Tensor_arith_mul.h ./tensors/Tensor_arith_scalar.h ./tensors/Tensor_arith_sub.h ./tensors/Tensor_class.h ./tensors/Tensor_determinant.h ./tensors/Tensor_exp.h ./tensors/Tensor_extract_merge.h ./tensors/Tensor_index.h ./tensors/Tensor_inner.h ./tensors/Tensor_logical.h ./tensors/Tensor_outer.h ./tensors/Tensor_reality.h ./tensors/Tensor_Ta.h ./tensors/Tensor_trace.h ./tensors/Tensor_traits.h ./tensors/Tensor_transpose.h ./tensors/Tensor_unary.h ./Tensors.h ./Threads.h ./Timer.h
|
||||
|
||||
CCFILES=./algorithms/approx/MultiShiftFunction.cc ./algorithms/approx/Remez.cc ./algorithms/approx/Zolotarev.cc ./Init.cc ./Log.cc ./pugixml/pugixml.cc ./qcd/action/fermion/CayleyFermion5D.cc ./qcd/action/fermion/ContinuedFractionFermion5D.cc ./qcd/action/fermion/PartialFractionFermion5D.cc ./qcd/action/fermion/WilsonFermion.cc ./qcd/action/fermion/WilsonFermion5D.cc ./qcd/action/fermion/WilsonKernels.cc ./qcd/action/fermion/WilsonKernelsHand.cc ./qcd/hmc/HMC.cc ./qcd/spin/Dirac.cc ./qcd/utils/SpaceTimeGrid.cc ./stencil/Lebesgue.cc ./stencil/Stencil_common.cc
|
||||
|
||||
@ -1,5 +1,5 @@
|
||||
|
||||
bin_PROGRAMS = Test_GaugeAction Test_cayley_cg Test_cayley_coarsen_support Test_cayley_even_odd Test_cayley_ldop_cr Test_cf_coarsen_support Test_cf_cr_unprec Test_cheby Test_contfrac_cg Test_contfrac_even_odd Test_contfrac_force Test_cshift Test_cshift_red_black Test_dwf_cg_prec Test_dwf_cg_schur Test_dwf_cg_unprec Test_dwf_cr_unprec Test_dwf_even_odd Test_dwf_force Test_dwf_fpgcr Test_dwf_hdcr Test_gamma Test_hmc_EODWFRatio Test_hmc_EOWilsonFermionGauge Test_hmc_EOWilsonRatio Test_hmc_WilsonFermionGauge Test_hmc_WilsonGauge Test_hmc_WilsonRatio Test_lie_generators Test_main Test_multishift_sqrt Test_partfrac_force Test_remez Test_rhmc_EOWilson1p1 Test_rhmc_EOWilsonRatio Test_rhmc_Wilson1p1 Test_rhmc_WilsonRatio Test_rng Test_rng_fixed Test_serialisation Test_simd Test_stencil Test_wilson_cg_prec Test_wilson_cg_schur Test_wilson_cg_unprec Test_wilson_cr_unprec Test_wilson_even_odd Test_wilson_force_phiMdagMphi Test_wilson_force_phiMphi
|
||||
bin_PROGRAMS = Test_GaugeAction Test_cayley_cg Test_cayley_coarsen_support Test_cayley_even_odd Test_cayley_ldop_cr Test_cf_coarsen_support Test_cf_cr_unprec Test_cheby Test_contfrac_cg Test_contfrac_even_odd Test_contfrac_force Test_cshift Test_cshift_red_black Test_dwf_cg_prec Test_dwf_cg_schur Test_dwf_cg_unprec Test_dwf_cr_unprec Test_dwf_even_odd Test_dwf_force Test_dwf_fpgcr Test_dwf_hdcr Test_gamma Test_hmc_EODWFRatio Test_hmc_EOWilsonFermionGauge Test_hmc_EOWilsonRatio Test_hmc_WilsonFermionGauge Test_hmc_WilsonGauge Test_hmc_WilsonRatio Test_lie_generators Test_main Test_multishift_sqrt Test_nersc_io Test_partfrac_force Test_quenched_update Test_remez Test_rhmc_EOWilson1p1 Test_rhmc_EOWilsonRatio Test_rhmc_Wilson1p1 Test_rhmc_WilsonRatio Test_rng Test_rng_fixed Test_serialisation Test_simd Test_stencil Test_wilson_cg_prec Test_wilson_cg_schur Test_wilson_cg_unprec Test_wilson_cr_unprec Test_wilson_even_odd Test_wilson_force Test_wilson_force_phiMdagMphi Test_wilson_force_phiMphi
|
||||
|
||||
|
||||
Test_GaugeAction_SOURCES=Test_GaugeAction.cc
|
||||
@ -85,9 +85,11 @@ Test_dwf_fpgcr_LDADD=-lGrid
|
||||
Test_dwf_hdcr_SOURCES=Test_dwf_hdcr.cc
|
||||
Test_dwf_hdcr_LDADD=-lGrid
|
||||
|
||||
|
||||
#Test_dwf_lanczos_SOURCES=Test_dwf_lanczos.cc
|
||||
#Test_dwf_lanczos_LDADD=-lGrid
|
||||
|
||||
|
||||
Test_gamma_SOURCES=Test_gamma.cc
|
||||
Test_gamma_LDADD=-lGrid
|
||||
|
||||
@ -96,8 +98,8 @@ Test_gamma_LDADD=-lGrid
|
||||
#Test_gparity_LDADD=-lGrid
|
||||
|
||||
|
||||
#Test_gpwilson_even_odd_SOURCES=Test_gpwilson_even_odd.cc
|
||||
#Test_gpwilson_even_odd_LDADD=-lGrid
|
||||
Test_gpwilson_even_odd_SOURCES=Test_gpwilson_even_odd.cc
|
||||
Test_gpwilson_even_odd_LDADD=-lGrid
|
||||
|
||||
|
||||
Test_hmc_EODWFRatio_SOURCES=Test_hmc_EODWFRatio.cc
|
||||
|
||||
Loading…
Reference in New Issue
Block a user