mirror of
https://github.com/paboyle/Grid.git
synced 2024-11-10 07:55:35 +00:00
Remove Ls Vec cases from benchmarks
This commit is contained in:
Peter Boyle
parent
7e27a5213a
commit
47c063f984
@ -30,9 +30,6 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
|
||||
|
||||
using namespace Grid;
|
||||
|
||||
typedef WilsonFermion5D<DomainWallVec5dImplR> WilsonFermion5DR;
|
||||
typedef WilsonFermion5D<DomainWallVec5dImplF> WilsonFermion5DF;
|
||||
typedef WilsonFermion5D<DomainWallVec5dImplD> WilsonFermion5DD;
|
||||
|
||||
std::vector<int> L_list;
|
||||
std::vector<int> Ls_list;
|
||||
@ -273,6 +270,7 @@ public:
|
||||
}
|
||||
};
|
||||
|
||||
#if 0
|
||||
static double DWF5(int Ls,int L)
|
||||
{
|
||||
// RealD mass=0.1;
|
||||
@ -454,6 +452,7 @@ public:
|
||||
}
|
||||
return mflops_best;
|
||||
}
|
||||
#endif
|
||||
|
||||
static double DWF(int Ls,int L, double & robust)
|
||||
{
|
||||
@ -742,24 +741,14 @@ int main (int argc, char ** argv)
|
||||
}
|
||||
}
|
||||
|
||||
if ( do_dwf ) {
|
||||
std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
|
||||
std::cout<<GridLogMessage << " Domain wall dslash 4D vectorised" <<std::endl;
|
||||
std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
|
||||
for(int l=0;l<L_list.size();l++){
|
||||
dwf5.push_back(Benchmark::DWF5(Ls,L_list[l]));
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
if ( do_dwf ) {
|
||||
|
||||
std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
|
||||
std::cout<<GridLogMessage << " Summary table Ls="<<Ls <<std::endl;
|
||||
std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
|
||||
std::cout<<GridLogMessage << "L \t\t Wilson \t DWF4 \t DWF5 " <<std::endl;
|
||||
std::cout<<GridLogMessage << "L \t\t Wilson \t DWF4 " <<std::endl;
|
||||
for(int l=0;l<L_list.size();l++){
|
||||
std::cout<<GridLogMessage << L_list[l] <<" \t\t "<< wilson[l]<<" \t "<<dwf4[l]<<" \t "<<dwf5[l] <<std::endl;
|
||||
std::cout<<GridLogMessage << L_list[l] <<" \t\t "<< wilson[l]<<" \t "<<dwf4[l] <<std::endl;
|
||||
}
|
||||
std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
|
||||
}
|
||||
@ -783,9 +772,9 @@ int main (int argc, char ** argv)
|
||||
std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
|
||||
std::cout<<GridLogMessage << " Per Node Summary table Ls="<<Ls <<std::endl;
|
||||
std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
|
||||
std::cout<<GridLogMessage << " L \t\t Wilson\t\t DWF4 \t\t DWF5 " <<std::endl;
|
||||
std::cout<<GridLogMessage << " L \t\t Wilson\t\t DWF4 " <<std::endl;
|
||||
for(int l=0;l<L_list.size();l++){
|
||||
std::cout<<GridLogMessage << L_list[l] <<" \t\t "<< wilson[l]/NN<<" \t "<<dwf4[l]/NN<<" \t "<<dwf5[l] /NN<<std::endl;
|
||||
std::cout<<GridLogMessage << L_list[l] <<" \t\t "<< wilson[l]/NN<<" \t "<<dwf4[l]/NN<<std::endl;
|
||||
}
|
||||
std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
|
||||
|
||||
@ -798,6 +787,5 @@ int main (int argc, char ** argv)
|
||||
|
||||
}
|
||||
|
||||
|
||||
Grid_finalize();
|
||||
}
|
||||
|
||||
@ -38,9 +38,6 @@ struct scal {
|
||||
Gamma::Algebra::GammaT
|
||||
};
|
||||
|
||||
typedef WilsonFermion5D<DomainWallVec5dImplR> WilsonFermion5DR;
|
||||
typedef WilsonFermion5D<DomainWallVec5dImplF> WilsonFermion5DF;
|
||||
typedef WilsonFermion5D<DomainWallVec5dImplD> WilsonFermion5DD;
|
||||
|
||||
int main (int argc, char ** argv)
|
||||
{
|
||||
@ -247,171 +244,6 @@ int main (int argc, char ** argv)
|
||||
DwH.Report();
|
||||
}
|
||||
|
||||
if (1)
|
||||
{
|
||||
|
||||
std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
|
||||
std::cout << GridLogMessage<< "* Benchmarking WilsonFermion5D<DomainWallVec5dImplR>::Dhop "<<std::endl;
|
||||
std::cout << GridLogMessage<< "* Vectorising fifth dimension by "<<vComplex::Nsimd()<<std::endl;
|
||||
if ( sizeof(Real)==4 ) std::cout << GridLogMessage<< "* SINGLE precision "<<std::endl;
|
||||
if ( sizeof(Real)==8 ) std::cout << GridLogMessage<< "* DOUBLE 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::OptGpu ) std::cout << GridLogMessage<< "* Using Gpu 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;
|
||||
|
||||
typedef WilsonFermion5D<DomainWallVec5dImplR> WilsonFermion5DR;
|
||||
LatticeFermion ssrc(sFGrid);
|
||||
LatticeFermion sref(sFGrid);
|
||||
LatticeFermion sresult(sFGrid);
|
||||
|
||||
WilsonFermion5DR sDw(Umu,*sFGrid,*sFrbGrid,*sUGrid,*sUrbGrid,M5);
|
||||
|
||||
localConvert(src,ssrc);
|
||||
std::cout<<GridLogMessage<< "src norms "<< norm2(src)<<" " <<norm2(ssrc)<<std::endl;
|
||||
FGrid->Barrier();
|
||||
sDw.Dhop(ssrc,sresult,0);
|
||||
sDw.ZeroCounters();
|
||||
double t0=usecond();
|
||||
for(int i=0;i<ncall;i++){
|
||||
__SSC_START;
|
||||
sDw.Dhop(ssrc,sresult,0);
|
||||
__SSC_STOP;
|
||||
}
|
||||
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;
|
||||
|
||||
std::cout<<GridLogMessage << "Called Dw s_inner "<<ncall<<" times in "<<t1-t0<<" us"<<std::endl;
|
||||
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t1-t0)<<std::endl;
|
||||
std::cout<<GridLogMessage << "mflop/s per rank = "<< flops/(t1-t0)/NP<<std::endl;
|
||||
std::cout<<GridLogMessage << "mflop/s per node = "<< flops/(t1-t0)/NN<<std::endl;
|
||||
// std::cout<<GridLogMessage<< "res norms "<< norm2(result)<<" " <<norm2(sresult)<<std::endl;
|
||||
sDw.Report();
|
||||
RealD sum=0;
|
||||
|
||||
err=Zero();
|
||||
localConvert(sresult,err);
|
||||
err = err - ref;
|
||||
sum = norm2(err);
|
||||
std::cout<<GridLogMessage<<" difference between normal ref and simd is "<<sum<<std::endl;
|
||||
if(sum > 1.0e-4 ){
|
||||
std::cout<< "sD REF\n " <<ref << std::endl;
|
||||
std::cout<< "sD ERR \n " <<err <<std::endl;
|
||||
}
|
||||
// assert(sum < 1.0e-4);
|
||||
|
||||
err=Zero();
|
||||
localConvert(sresult,err);
|
||||
err = err - result;
|
||||
sum = norm2(err);
|
||||
std::cout<<GridLogMessage<<" difference between normal result and simd is "<<sum<<std::endl;
|
||||
if(sum > 1.0e-4 ){
|
||||
std::cout<< "sD REF\n " <<result << std::endl;
|
||||
std::cout<< "sD ERR \n " << err <<std::endl;
|
||||
}
|
||||
assert(sum < 1.0e-4);
|
||||
|
||||
|
||||
if(1){
|
||||
std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
|
||||
std::cout << GridLogMessage<< "* Benchmarking WilsonFermion5D<DomainWallVec5dImplR>::DhopEO "<<std::endl;
|
||||
std::cout << GridLogMessage<< "* Vectorising fifth dimension by "<<vComplex::Nsimd()<<std::endl;
|
||||
if ( sizeof(Real)==4 ) std::cout << GridLogMessage<< "* SINGLE precision "<<std::endl;
|
||||
if ( sizeof(Real)==8 ) std::cout << GridLogMessage<< "* DOUBLE 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::OptGpu ) std::cout << GridLogMessage<< "* Using Gpu 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;
|
||||
|
||||
LatticeFermion sr_eo(sFGrid);
|
||||
LatticeFermion ssrc_e (sFrbGrid);
|
||||
LatticeFermion ssrc_o (sFrbGrid);
|
||||
LatticeFermion sr_e (sFrbGrid);
|
||||
LatticeFermion sr_o (sFrbGrid);
|
||||
|
||||
pickCheckerboard(Even,ssrc_e,ssrc);
|
||||
pickCheckerboard(Odd,ssrc_o,ssrc);
|
||||
// setCheckerboard(sr_eo,ssrc_o);
|
||||
// setCheckerboard(sr_eo,ssrc_e);
|
||||
|
||||
sr_e = Zero();
|
||||
sr_o = Zero();
|
||||
|
||||
FGrid->Barrier();
|
||||
sDw.DhopEO(ssrc_o, sr_e, DaggerNo);
|
||||
sDw.ZeroCounters();
|
||||
// sDw.stat.init("DhopEO");
|
||||
double t0=usecond();
|
||||
for (int i = 0; i < ncall; i++) {
|
||||
sDw.DhopEO(ssrc_o, sr_e, DaggerNo);
|
||||
}
|
||||
double t1=usecond();
|
||||
FGrid->Barrier();
|
||||
// sDw.stat.print();
|
||||
|
||||
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 << "sDeo mflop/s = "<< flops/(t1-t0)<<std::endl;
|
||||
std::cout<<GridLogMessage << "sDeo mflop/s per rank "<< flops/(t1-t0)/NP<<std::endl;
|
||||
std::cout<<GridLogMessage << "sDeo mflop/s per node "<< flops/(t1-t0)/NN<<std::endl;
|
||||
sDw.Report();
|
||||
|
||||
sDw.DhopEO(ssrc_o,sr_e,DaggerNo);
|
||||
sDw.DhopOE(ssrc_e,sr_o,DaggerNo);
|
||||
sDw.Dhop (ssrc ,sresult,DaggerNo);
|
||||
|
||||
pickCheckerboard(Even,ssrc_e,sresult);
|
||||
pickCheckerboard(Odd ,ssrc_o,sresult);
|
||||
|
||||
ssrc_e = ssrc_e - sr_e;
|
||||
RealD error = norm2(ssrc_e);
|
||||
std::cout<<GridLogMessage << "sE norm diff "<< norm2(ssrc_e)<< " vec nrm"<<norm2(sr_e) <<std::endl;
|
||||
|
||||
ssrc_o = ssrc_o - sr_o;
|
||||
error+= norm2(ssrc_o);
|
||||
std::cout<<GridLogMessage << "sO norm diff "<< norm2(ssrc_o)<< " vec nrm"<<norm2(sr_o) <<std::endl;
|
||||
|
||||
if(( error>1.0e-4) ) {
|
||||
setCheckerboard(ssrc,ssrc_o);
|
||||
setCheckerboard(ssrc,ssrc_e);
|
||||
std::cout<< "DIFF\n " <<ssrc << std::endl;
|
||||
setCheckerboard(ssrc,sr_o);
|
||||
setCheckerboard(ssrc,sr_e);
|
||||
std::cout<< "CBRESULT\n " <<ssrc << std::endl;
|
||||
std::cout<< "RESULT\n " <<sresult<< std::endl;
|
||||
}
|
||||
assert(error<1.0e-4);
|
||||
}
|
||||
|
||||
if(0){
|
||||
std::cout << "Single cache warm call to sDw.Dhop " <<std::endl;
|
||||
for(int i=0;i< PerformanceCounter::NumTypes(); i++ ){
|
||||
sDw.Dhop(ssrc,sresult,0);
|
||||
PerformanceCounter Counter(i);
|
||||
Counter.Start();
|
||||
sDw.Dhop(ssrc,sresult,0);
|
||||
Counter.Stop();
|
||||
Counter.Report();
|
||||
}
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
|
||||
|
||||
if (1)
|
||||
{ // Naive wilson dag implementation
|
||||
ref = Zero();
|
||||
|
||||
@ -44,7 +44,6 @@ struct scal {
|
||||
};
|
||||
|
||||
void benchDw(std::vector<int> & L, int Ls, int threads, int report =0 );
|
||||
void benchsDw(std::vector<int> & L, int Ls, int threads, int report=0 );
|
||||
|
||||
int main (int argc, char ** argv)
|
||||
{
|
||||
@ -66,7 +65,7 @@ int main (int argc, char ** argv)
|
||||
std::cout<<GridLogMessage << "=========================================================================="<<std::endl;
|
||||
std::cout<<GridLogMessage << "= Benchmarking DWF"<<std::endl;
|
||||
std::cout<<GridLogMessage << "=========================================================================="<<std::endl;
|
||||
std::cout<<GridLogMessage << "Volume \t\t\tProcs \t Dw \t eoDw \t sDw \t eosDw (Mflop/s) "<<std::endl;
|
||||
std::cout<<GridLogMessage << "Volume \t\t\tProcs \t Dw \t eoDw "<<std::endl;
|
||||
std::cout<<GridLogMessage << "=========================================================================="<<std::endl;
|
||||
|
||||
int Lmax=16;
|
||||
@ -83,7 +82,6 @@ int main (int argc, char ** argv)
|
||||
}
|
||||
std::cout <<Ls<<"\t" ;
|
||||
benchDw (latt4,Ls,threads,0);
|
||||
benchsDw(latt4,Ls,threads,0);
|
||||
std::cout<<std::endl;
|
||||
}
|
||||
}
|
||||
@ -93,8 +91,6 @@ int main (int argc, char ** argv)
|
||||
std::vector<int> latt4(4,16);
|
||||
std::cout<<GridLogMessage << "16^4 Dw miss rate"<<std::endl;
|
||||
benchDw (latt4,Ls,threads,1);
|
||||
std::cout<<GridLogMessage << "16^4 sDw miss rate"<<std::endl;
|
||||
benchsDw(latt4,Ls,threads,1);
|
||||
}
|
||||
#endif
|
||||
Grid_finalize();
|
||||
@ -240,147 +236,5 @@ void benchDw(std::vector<int> & latt4, int Ls, int threads,int report )
|
||||
}
|
||||
}
|
||||
|
||||
#define CHECK_SDW
|
||||
void benchsDw(std::vector<int> & latt4, int Ls, int threads, int report )
|
||||
{
|
||||
long unsigned int single_site_flops = 8*Nc*(7+16*Nc);
|
||||
|
||||
GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(latt4, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
|
||||
GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
|
||||
GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
|
||||
GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
|
||||
|
||||
GridCartesian * sUGrid = SpaceTimeGrid::makeFourDimDWFGrid(latt4,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> seeds5({5,6,7,8});
|
||||
|
||||
#ifdef CHECK_SDW
|
||||
GridParallelRNG RNG4(UGrid); RNG4.SeedFixedIntegers(seeds4);
|
||||
GridParallelRNG RNG5(FGrid); RNG5.SeedFixedIntegers(seeds5);
|
||||
LatticeFermion src (FGrid); random(RNG5,src);
|
||||
LatticeGaugeField Umu(UGrid);
|
||||
random(RNG4,Umu);
|
||||
#else
|
||||
LatticeFermion src (FGrid); src=Zero();
|
||||
LatticeGaugeField Umu(UGrid); Umu=Zero();
|
||||
#endif
|
||||
|
||||
LatticeFermion result(FGrid); result=Zero();
|
||||
LatticeFermion ref(FGrid); ref=Zero();
|
||||
LatticeFermion tmp(FGrid);
|
||||
LatticeFermion err(FGrid);
|
||||
|
||||
ColourMatrix cm = Complex(1.0,0.0);
|
||||
|
||||
LatticeGaugeField Umu5d(FGrid);
|
||||
|
||||
// replicate across fifth dimension
|
||||
auto Umu5d_v = Umu5d.View();
|
||||
auto Umu_v = Umu.View();
|
||||
for(int ss=0;ss<Umu.Grid()->oSites();ss++){
|
||||
for(int s=0;s<Ls;s++){
|
||||
Umu5d_v[Ls*ss+s] = Umu_v[ss];
|
||||
}
|
||||
}
|
||||
|
||||
// RealD mass=0.1;
|
||||
RealD M5 =1.8;
|
||||
|
||||
typedef WilsonFermion5D<DomainWallVec5dImplR> WilsonFermion5DR;
|
||||
LatticeFermion ssrc(sFGrid);
|
||||
LatticeFermion sref(sFGrid);
|
||||
LatticeFermion sresult(sFGrid);
|
||||
WilsonFermion5DR sDw(Umu,*sFGrid,*sFrbGrid,*sUGrid,*sUrbGrid,M5);
|
||||
|
||||
for(int x=0;x<latt4[0];x++){
|
||||
for(int y=0;y<latt4[1];y++){
|
||||
for(int z=0;z<latt4[2];z++){
|
||||
for(int t=0;t<latt4[3];t++){
|
||||
for(int s=0;s<Ls;s++){
|
||||
std::vector<int> site({s,x,y,z,t});
|
||||
SpinColourVector tmp;
|
||||
peekSite(tmp,src,site);
|
||||
pokeSite(tmp,ssrc,site);
|
||||
}}}}}
|
||||
|
||||
double t0=usecond();
|
||||
sDw.Dhop(ssrc,sresult,0);
|
||||
double t1=usecond();
|
||||
|
||||
#ifdef TIMERS_OFF
|
||||
int ncall =10;
|
||||
#else
|
||||
int ncall =1+(int) ((5.0*1000*1000)/(t1-t0));
|
||||
#endif
|
||||
|
||||
#ifdef PERFCOUNT
|
||||
PerformanceCounter Counter(8);
|
||||
Counter.Start();
|
||||
#endif
|
||||
t0=usecond();
|
||||
for(int i=0;i<ncall;i++){
|
||||
sDw.Dhop(ssrc,sresult,0);
|
||||
}
|
||||
t1=usecond();
|
||||
|
||||
#ifdef PERFCOUNT
|
||||
Counter.Stop();
|
||||
if ( report ) {
|
||||
Counter.Report();
|
||||
}
|
||||
#endif
|
||||
|
||||
if ( !report){
|
||||
double volume=Ls; for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
|
||||
double flops=single_site_flops*volume*ncall;
|
||||
std::cout<<"\t"<< flops/(t1-t0);
|
||||
}
|
||||
|
||||
LatticeFermion sr_eo(sFGrid);
|
||||
LatticeFermion serr(sFGrid);
|
||||
|
||||
LatticeFermion ssrc_e (sFrbGrid);
|
||||
LatticeFermion ssrc_o (sFrbGrid);
|
||||
LatticeFermion sr_e (sFrbGrid);
|
||||
LatticeFermion sr_o (sFrbGrid);
|
||||
|
||||
pickCheckerboard(Even,ssrc_e,ssrc);
|
||||
pickCheckerboard(Odd,ssrc_o,ssrc);
|
||||
|
||||
setCheckerboard(sr_eo,ssrc_o);
|
||||
setCheckerboard(sr_eo,ssrc_e);
|
||||
|
||||
sr_e = Zero();
|
||||
sr_o = Zero();
|
||||
|
||||
sDw.DhopEO(ssrc_o,sr_e,DaggerNo);
|
||||
#ifdef PERFCOUNT
|
||||
PerformanceCounter CounterSdw(8);
|
||||
CounterSdw.Start();
|
||||
#endif
|
||||
t0=usecond();
|
||||
for(int i=0;i<ncall;i++){
|
||||
__SSC_START;
|
||||
sDw.DhopEO(ssrc_o,sr_e,DaggerNo);
|
||||
__SSC_STOP;
|
||||
}
|
||||
t1=usecond();
|
||||
#ifdef PERFCOUNT
|
||||
CounterSdw.Stop();
|
||||
if ( report ) {
|
||||
CounterSdw.Report();
|
||||
}
|
||||
#endif
|
||||
|
||||
if ( ! report ) {
|
||||
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<<"\t"<< flops/(t1-t0);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
|
||||
@ -47,12 +47,6 @@ int main (int argc, char ** argv)
|
||||
GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
|
||||
GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
|
||||
|
||||
std::cout << GridLogMessage << "Making Vec5d 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> seeds5({5,6,7,8});
|
||||
|
||||
@ -159,58 +153,5 @@ int main (int argc, char ** argv)
|
||||
|
||||
}
|
||||
|
||||
if (1)
|
||||
{
|
||||
const int ncall=1000;
|
||||
|
||||
std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
|
||||
std::cout << GridLogMessage<< "* Benchmarking DomainWallFermionVec5dR::Dhop "<<std::endl;
|
||||
std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
|
||||
|
||||
GridParallelRNG RNG5(sFGrid); RNG5.SeedFixedIntegers(seeds5);
|
||||
LatticeFermion src(sFGrid); random(RNG5,src);
|
||||
LatticeFermion sref(sFGrid);
|
||||
LatticeFermion result(sFGrid);
|
||||
|
||||
|
||||
std::cout<<GridLogMessage << "Constructing Vec5D Dw "<<std::endl;
|
||||
DomainWallFermionVec5dR Dw(Umu,*sFGrid,*sFrbGrid,*sUGrid,*sUrbGrid,mass,M5);
|
||||
|
||||
RealD b=1.5;// Scale factor b+c=2, b-c=1
|
||||
RealD c=0.5;
|
||||
Vector<ComplexD> gamma(Ls,std::complex<double>(1.0,0.0));
|
||||
ZMobiusFermionVec5dR zDw(Umu,*sFGrid,*sFrbGrid,*sUGrid,*sUrbGrid,mass,M5,gamma,b,c);
|
||||
|
||||
std::cout<<GridLogMessage << "Calling Dhop "<<std::endl;
|
||||
FGrid->Barrier();
|
||||
|
||||
double t0,t1;
|
||||
|
||||
LatticeFermion r_eo(sFGrid);
|
||||
LatticeFermion src_e (sFrbGrid);
|
||||
LatticeFermion src_o (sFrbGrid);
|
||||
LatticeFermion r_e (sFrbGrid);
|
||||
LatticeFermion r_o (sFrbGrid);
|
||||
|
||||
pickCheckerboard(Even,src_e,src);
|
||||
pickCheckerboard(Odd,src_o,src);
|
||||
|
||||
setCheckerboard(r_eo,src_o);
|
||||
setCheckerboard(r_eo,src_e);
|
||||
|
||||
r_e = Zero();
|
||||
r_o = Zero();
|
||||
|
||||
BENCH_DW(Dhop ,src,result,0);
|
||||
BENCH_DW(DhopEO ,src_o,r_e,0);
|
||||
BENCH_DW_SSC(Meooe ,src_o,r_e);
|
||||
BENCH_DW(Mooee ,src_o,r_o);
|
||||
BENCH_DW(MooeeInv,src_o,r_o);
|
||||
|
||||
BENCH_ZDW(Mooee ,src_o,r_o);
|
||||
BENCH_ZDW(MooeeInv,src_o,r_o);
|
||||
|
||||
}
|
||||
|
||||
Grid_finalize();
|
||||
}
|
||||
|
||||
Loading…
Reference in New Issue
Block a user