mirror of
https://github.com/paboyle/Grid.git
synced 2025-06-11 03:46:55 +01:00
Merge branch 'develop' of https://github.com/paboyle/Grid into feature/staggering
This commit is contained in:
Azusa Yamaguchi
@ -42,15 +42,14 @@ int main (int argc, char ** argv)
|
||||
|
||||
int Nloop=10;
|
||||
int nmu=0;
|
||||
for(int mu=0;mu<4;mu++) if (mpi_layout[mu]>1) nmu++;
|
||||
for(int mu=0;mu<Nd;mu++) if (mpi_layout[mu]>1) nmu++;
|
||||
|
||||
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
|
||||
std::cout<<GridLogMessage << "= Benchmarking concurrent halo exchange in "<<nmu<<" dimensions"<<std::endl;
|
||||
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
|
||||
std::cout<<GridLogMessage << " L "<<"\t\t"<<" Ls "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s uni"<<"\t\t"<<"MB/s bidi"<<std::endl;
|
||||
|
||||
|
||||
|
||||
for(int lat=4;lat<=32;lat+=2){
|
||||
int maxlat=16;
|
||||
for(int lat=4;lat<=maxlat;lat+=2){
|
||||
for(int Ls=1;Ls<=16;Ls*=2){
|
||||
|
||||
std::vector<int> latt_size ({lat*mpi_layout[0],
|
||||
@ -125,7 +124,7 @@ int main (int argc, char ** argv)
|
||||
std::cout<<GridLogMessage << " L "<<"\t\t"<<" Ls "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s uni"<<"\t\t"<<"MB/s bidi"<<std::endl;
|
||||
|
||||
|
||||
for(int lat=4;lat<=32;lat+=2){
|
||||
for(int lat=4;lat<=maxlat;lat+=2){
|
||||
for(int Ls=1;Ls<=16;Ls*=2){
|
||||
|
||||
std::vector<int> latt_size ({lat,lat,lat,lat});
|
||||
@ -194,128 +193,169 @@ int main (int argc, char ** argv)
|
||||
}
|
||||
}
|
||||
|
||||
#if 0
|
||||
|
||||
Nloop=100;
|
||||
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
|
||||
std::cout<<GridLogMessage << "= Benchmarking sequential persistent halo exchange in "<<nmu<<" dimensions"<<std::endl;
|
||||
std::cout<<GridLogMessage << "= Benchmarking concurrent STENCIL halo exchange in "<<nmu<<" dimensions"<<std::endl;
|
||||
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
|
||||
std::cout<<GridLogMessage << " L "<<"\t\t"<<" Ls "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s uni"<<"\t\t"<<"MB/s bidi"<<std::endl;
|
||||
|
||||
|
||||
for(int lat=4;lat<=32;lat+=2){
|
||||
for(int lat=4;lat<=maxlat;lat+=2){
|
||||
for(int Ls=1;Ls<=16;Ls*=2){
|
||||
|
||||
std::vector<int> latt_size ({lat,lat,lat,lat});
|
||||
std::vector<int> latt_size ({lat*mpi_layout[0],
|
||||
lat*mpi_layout[1],
|
||||
lat*mpi_layout[2],
|
||||
lat*mpi_layout[3]});
|
||||
|
||||
GridCartesian Grid(latt_size,simd_layout,mpi_layout);
|
||||
|
||||
std::vector<std::vector<HalfSpinColourVectorD> > xbuf(8,std::vector<HalfSpinColourVectorD>(lat*lat*lat*Ls));
|
||||
std::vector<std::vector<HalfSpinColourVectorD> > rbuf(8,std::vector<HalfSpinColourVectorD>(lat*lat*lat*Ls));
|
||||
|
||||
std::vector<HalfSpinColourVectorD *> xbuf(8);
|
||||
std::vector<HalfSpinColourVectorD *> rbuf(8);
|
||||
Grid.ShmBufferFreeAll();
|
||||
for(int d=0;d<8;d++){
|
||||
xbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
|
||||
rbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
|
||||
}
|
||||
|
||||
int ncomm;
|
||||
int bytes=lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
|
||||
|
||||
double start=usecond();
|
||||
for(int i=0;i<Nloop;i++){
|
||||
|
||||
std::vector<CartesianCommunicator::CommsRequest_t> empty;
|
||||
std::vector<std::vector<CartesianCommunicator::CommsRequest_t> > requests_fwd(Nd,empty);
|
||||
std::vector<std::vector<CartesianCommunicator::CommsRequest_t> > requests_bwd(Nd,empty);
|
||||
std::vector<CartesianCommunicator::CommsRequest_t> requests;
|
||||
|
||||
for(int mu=0;mu<4;mu++){
|
||||
ncomm=0;
|
||||
if (mpi_layout[mu]>1 ) {
|
||||
ncomm++;
|
||||
|
||||
int comm_proc;
|
||||
int xmit_to_rank;
|
||||
int recv_from_rank;
|
||||
|
||||
comm_proc=1;
|
||||
Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
|
||||
Grid.SendToRecvFromInit(requests_fwd[mu],
|
||||
(void *)&xbuf[mu][0],
|
||||
xmit_to_rank,
|
||||
(void *)&rbuf[mu][0],
|
||||
recv_from_rank,
|
||||
bytes);
|
||||
|
||||
comm_proc = mpi_layout[mu]-1;
|
||||
Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
|
||||
Grid.SendToRecvFromInit(requests_bwd[mu],
|
||||
(void *)&xbuf[mu+4][0],
|
||||
xmit_to_rank,
|
||||
(void *)&rbuf[mu+4][0],
|
||||
recv_from_rank,
|
||||
bytes);
|
||||
|
||||
}
|
||||
}
|
||||
|
||||
{
|
||||
double start=usecond();
|
||||
for(int i=0;i<Nloop;i++){
|
||||
for(int mu=0;mu<4;mu++){
|
||||
|
||||
if (mpi_layout[mu]>1 ) {
|
||||
|
||||
for(int mu=0;mu<4;mu++){
|
||||
ncomm++;
|
||||
int comm_proc=1;
|
||||
int xmit_to_rank;
|
||||
int recv_from_rank;
|
||||
|
||||
if (mpi_layout[mu]>1 ) {
|
||||
|
||||
Grid.SendToRecvFromBegin(requests_fwd[mu]);
|
||||
Grid.SendToRecvFromComplete(requests_fwd[mu]);
|
||||
Grid.SendToRecvFromBegin(requests_bwd[mu]);
|
||||
Grid.SendToRecvFromComplete(requests_bwd[mu]);
|
||||
}
|
||||
}
|
||||
Grid.Barrier();
|
||||
}
|
||||
Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
|
||||
Grid.StencilSendToRecvFromBegin(requests,
|
||||
(void *)&xbuf[mu][0],
|
||||
xmit_to_rank,
|
||||
(void *)&rbuf[mu][0],
|
||||
recv_from_rank,
|
||||
bytes);
|
||||
|
||||
double stop=usecond();
|
||||
|
||||
double dbytes = bytes;
|
||||
double xbytes = Nloop*dbytes*2.0*ncomm;
|
||||
double rbytes = xbytes;
|
||||
double bidibytes = xbytes+rbytes;
|
||||
|
||||
double time = stop-start;
|
||||
|
||||
std::cout<<GridLogMessage << lat<<"\t\t"<<Ls<<"\t\t"<<bytes<<"\t\t"<<xbytes/time<<"\t\t"<<bidibytes/time<<std::endl;
|
||||
|
||||
}
|
||||
|
||||
|
||||
{
|
||||
double start=usecond();
|
||||
for(int i=0;i<Nloop;i++){
|
||||
comm_proc = mpi_layout[mu]-1;
|
||||
|
||||
Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
|
||||
Grid.StencilSendToRecvFromBegin(requests,
|
||||
(void *)&xbuf[mu+4][0],
|
||||
xmit_to_rank,
|
||||
(void *)&rbuf[mu+4][0],
|
||||
recv_from_rank,
|
||||
bytes);
|
||||
|
||||
for(int mu=0;mu<4;mu++){
|
||||
|
||||
if (mpi_layout[mu]>1 ) {
|
||||
|
||||
Grid.SendToRecvFromBegin(requests_fwd[mu]);
|
||||
Grid.SendToRecvFromBegin(requests_bwd[mu]);
|
||||
Grid.SendToRecvFromComplete(requests_fwd[mu]);
|
||||
Grid.SendToRecvFromComplete(requests_bwd[mu]);
|
||||
}
|
||||
}
|
||||
Grid.Barrier();
|
||||
}
|
||||
|
||||
double stop=usecond();
|
||||
|
||||
double dbytes = bytes;
|
||||
double xbytes = Nloop*dbytes*2.0*ncomm;
|
||||
double rbytes = xbytes;
|
||||
double bidibytes = xbytes+rbytes;
|
||||
|
||||
double time = stop-start;
|
||||
|
||||
std::cout<<GridLogMessage << lat<<"\t\t"<<Ls<<"\t\t"<<bytes<<"\t\t"<<xbytes/time<<"\t\t"<<bidibytes/time<<std::endl;
|
||||
Grid.StencilSendToRecvFromComplete(requests);
|
||||
Grid.Barrier();
|
||||
|
||||
}
|
||||
double stop=usecond();
|
||||
|
||||
double dbytes = bytes;
|
||||
double xbytes = Nloop*dbytes*2.0*ncomm;
|
||||
double rbytes = xbytes;
|
||||
double bidibytes = xbytes+rbytes;
|
||||
|
||||
double time = stop-start; // microseconds
|
||||
|
||||
std::cout<<GridLogMessage << lat<<"\t\t"<<Ls<<"\t\t"<<bytes<<"\t\t"<<xbytes/time<<"\t\t"<<bidibytes/time<<std::endl;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
Nloop=100;
|
||||
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
|
||||
std::cout<<GridLogMessage << "= Benchmarking sequential STENCIL halo exchange in "<<nmu<<" dimensions"<<std::endl;
|
||||
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
|
||||
std::cout<<GridLogMessage << " L "<<"\t\t"<<" Ls "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s uni"<<"\t\t"<<"MB/s bidi"<<std::endl;
|
||||
|
||||
for(int lat=4;lat<=maxlat;lat+=2){
|
||||
for(int Ls=1;Ls<=16;Ls*=2){
|
||||
|
||||
std::vector<int> latt_size ({lat*mpi_layout[0],
|
||||
lat*mpi_layout[1],
|
||||
lat*mpi_layout[2],
|
||||
lat*mpi_layout[3]});
|
||||
|
||||
GridCartesian Grid(latt_size,simd_layout,mpi_layout);
|
||||
|
||||
std::vector<HalfSpinColourVectorD *> xbuf(8);
|
||||
std::vector<HalfSpinColourVectorD *> rbuf(8);
|
||||
Grid.ShmBufferFreeAll();
|
||||
for(int d=0;d<8;d++){
|
||||
xbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
|
||||
rbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
|
||||
}
|
||||
|
||||
int ncomm;
|
||||
int bytes=lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
|
||||
|
||||
double start=usecond();
|
||||
for(int i=0;i<Nloop;i++){
|
||||
|
||||
std::vector<CartesianCommunicator::CommsRequest_t> requests;
|
||||
|
||||
ncomm=0;
|
||||
for(int mu=0;mu<4;mu++){
|
||||
|
||||
if (mpi_layout[mu]>1 ) {
|
||||
|
||||
ncomm++;
|
||||
int comm_proc=1;
|
||||
int xmit_to_rank;
|
||||
int recv_from_rank;
|
||||
|
||||
Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
|
||||
Grid.StencilSendToRecvFromBegin(requests,
|
||||
(void *)&xbuf[mu][0],
|
||||
xmit_to_rank,
|
||||
(void *)&rbuf[mu][0],
|
||||
recv_from_rank,
|
||||
bytes);
|
||||
// Grid.StencilSendToRecvFromComplete(requests);
|
||||
// requests.resize(0);
|
||||
|
||||
comm_proc = mpi_layout[mu]-1;
|
||||
|
||||
Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
|
||||
Grid.StencilSendToRecvFromBegin(requests,
|
||||
(void *)&xbuf[mu+4][0],
|
||||
xmit_to_rank,
|
||||
(void *)&rbuf[mu+4][0],
|
||||
recv_from_rank,
|
||||
bytes);
|
||||
Grid.StencilSendToRecvFromComplete(requests);
|
||||
requests.resize(0);
|
||||
|
||||
}
|
||||
}
|
||||
Grid.Barrier();
|
||||
|
||||
}
|
||||
double stop=usecond();
|
||||
|
||||
double dbytes = bytes;
|
||||
double xbytes = Nloop*dbytes*2.0*ncomm;
|
||||
double rbytes = xbytes;
|
||||
double bidibytes = xbytes+rbytes;
|
||||
|
||||
double time = stop-start; // microseconds
|
||||
|
||||
std::cout<<GridLogMessage << lat<<"\t\t"<<Ls<<"\t\t"<<bytes<<"\t\t"<<xbytes/time<<"\t\t"<<bidibytes/time<<std::endl;
|
||||
}
|
||||
}
|
||||
|
||||
Grid_finalize();
|
||||
}
|
||||
|
||||
@ -44,7 +44,6 @@ struct scal {
|
||||
Gamma::GammaT
|
||||
};
|
||||
|
||||
bool overlapComms = false;
|
||||
typedef WilsonFermion5D<DomainWallVec5dImplR> WilsonFermion5DR;
|
||||
typedef WilsonFermion5D<DomainWallVec5dImplF> WilsonFermion5DF;
|
||||
typedef WilsonFermion5D<DomainWallVec5dImplD> WilsonFermion5DD;
|
||||
@ -54,15 +53,11 @@ int main (int argc, char ** argv)
|
||||
{
|
||||
Grid_init(&argc,&argv);
|
||||
|
||||
if( GridCmdOptionExists(argv,argv+argc,"--asynch") ){
|
||||
overlapComms = true;
|
||||
}
|
||||
|
||||
int threads = GridThread::GetThreads();
|
||||
std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
|
||||
|
||||
std::vector<int> latt4 = GridDefaultLatt();
|
||||
const int Ls=16;
|
||||
const int Ls=8;
|
||||
GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
|
||||
GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
|
||||
GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
|
||||
@ -126,17 +121,24 @@ int main (int argc, char ** argv)
|
||||
|
||||
RealD NP = UGrid->_Nprocessors;
|
||||
|
||||
for(int doasm=1;doasm<2;doasm++){
|
||||
|
||||
QCD::WilsonKernelsStatic::AsmOpt=doasm;
|
||||
|
||||
DomainWallFermionR Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
|
||||
|
||||
std::cout<<GridLogMessage << "Naive wilson implementation "<<std::endl;
|
||||
std::cout << GridLogMessage<< "Calling Dw"<<std::endl;
|
||||
|
||||
std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
|
||||
std::cout << GridLogMessage<< "* Kernel options --dslash-generic, --dslash-unroll, --dslash-asm" <<std::endl;
|
||||
std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
|
||||
std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
|
||||
std::cout << GridLogMessage<< "* Benchmarking DomainWallFermionR::Dhop "<<std::endl;
|
||||
std::cout << GridLogMessage<< "* Vectorising space-time 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;
|
||||
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;
|
||||
|
||||
int ncall =100;
|
||||
if (1) {
|
||||
|
||||
FGrid->Barrier();
|
||||
Dw.ZeroCounters();
|
||||
double t0=usecond();
|
||||
for(int i=0;i<ncall;i++){
|
||||
@ -145,6 +147,7 @@ int main (int argc, char ** argv)
|
||||
__SSC_STOP;
|
||||
}
|
||||
double t1=usecond();
|
||||
FGrid->Barrier();
|
||||
|
||||
double volume=Ls; for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
|
||||
double flops=1344*volume*ncall;
|
||||
@ -156,12 +159,23 @@ int main (int argc, char ** argv)
|
||||
std::cout<<GridLogMessage << "mflop/s per rank = "<< flops/(t1-t0)/NP<<std::endl;
|
||||
err = ref-result;
|
||||
std::cout<<GridLogMessage << "norm diff "<< norm2(err)<<std::endl;
|
||||
assert (norm2(err)< 1.0e-5 );
|
||||
assert (norm2(err)< 1.0e-4 );
|
||||
Dw.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;
|
||||
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;
|
||||
|
||||
typedef WilsonFermion5D<DomainWallVec5dImplR> WilsonFermion5DR;
|
||||
LatticeFermion ssrc(sFGrid);
|
||||
LatticeFermion sref(sFGrid);
|
||||
@ -180,6 +194,7 @@ int main (int argc, char ** argv)
|
||||
pokeSite(tmp,ssrc,site);
|
||||
}}}}}
|
||||
std::cout<<GridLogMessage<< "src norms "<< norm2(src)<<" " <<norm2(ssrc)<<std::endl;
|
||||
FGrid->Barrier();
|
||||
double t0=usecond();
|
||||
sDw.ZeroCounters();
|
||||
for(int i=0;i<ncall;i++){
|
||||
@ -188,6 +203,7 @@ int main (int argc, char ** argv)
|
||||
__SSC_STOP;
|
||||
}
|
||||
double t1=usecond();
|
||||
FGrid->Barrier();
|
||||
double volume=Ls; for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
|
||||
double flops=1344*volume*ncall;
|
||||
|
||||
@ -198,12 +214,12 @@ int main (int argc, char ** argv)
|
||||
|
||||
if(0){
|
||||
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();
|
||||
sDw.Dhop(ssrc,sresult,0);
|
||||
PerformanceCounter Counter(i);
|
||||
Counter.Start();
|
||||
sDw.Dhop(ssrc,sresult,0);
|
||||
Counter.Stop();
|
||||
Counter.Report();
|
||||
}
|
||||
}
|
||||
|
||||
@ -227,7 +243,7 @@ int main (int argc, char ** argv)
|
||||
}
|
||||
}}}}}
|
||||
std::cout<<GridLogMessage<<" difference between normal and simd is "<<sum<<std::endl;
|
||||
assert (sum< 1.0e-5 );
|
||||
assert (sum< 1.0e-4 );
|
||||
|
||||
|
||||
if (1) {
|
||||
@ -248,6 +264,17 @@ int main (int argc, char ** argv)
|
||||
sr_e = zero;
|
||||
sr_o = zero;
|
||||
|
||||
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;
|
||||
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;
|
||||
|
||||
FGrid->Barrier();
|
||||
sDw.ZeroCounters();
|
||||
sDw.stat.init("DhopEO");
|
||||
double t0=usecond();
|
||||
@ -255,6 +282,7 @@ int main (int argc, char ** argv)
|
||||
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];
|
||||
@ -278,7 +306,7 @@ int main (int argc, char ** argv)
|
||||
|
||||
error+= norm2(ssrc_o);
|
||||
std::cout<<GridLogMessage << "sO norm diff "<< norm2(ssrc_o)<< " vec nrm"<<norm2(sr_o) <<std::endl;
|
||||
if(error>1.0e-5) {
|
||||
if(error>1.0e-4) {
|
||||
setCheckerboard(ssrc,ssrc_o);
|
||||
setCheckerboard(ssrc,ssrc_e);
|
||||
std::cout<< ssrc << std::endl;
|
||||
@ -308,13 +336,13 @@ int main (int argc, char ** argv)
|
||||
ref = -0.5*ref;
|
||||
}
|
||||
Dw.Dhop(src,result,1);
|
||||
std::cout << GridLogMessage << "Naive wilson implementation Dag" << std::endl;
|
||||
std::cout << GridLogMessage << "Compare to naive wilson implementation Dag to verify correctness" << std::endl;
|
||||
std::cout<<GridLogMessage << "Called DwDag"<<std::endl;
|
||||
std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
|
||||
std::cout<<GridLogMessage << "norm ref "<< norm2(ref)<<std::endl;
|
||||
err = ref-result;
|
||||
std::cout<<GridLogMessage << "norm diff "<< norm2(err)<<std::endl;
|
||||
assert(norm2(err)<1.0e-5);
|
||||
assert(norm2(err)<1.0e-4);
|
||||
LatticeFermion src_e (FrbGrid);
|
||||
LatticeFermion src_o (FrbGrid);
|
||||
LatticeFermion r_e (FrbGrid);
|
||||
@ -322,20 +350,31 @@ int main (int argc, char ** argv)
|
||||
LatticeFermion r_eo (FGrid);
|
||||
|
||||
|
||||
std::cout<<GridLogMessage << "Calling Deo and Doe"<<std::endl;
|
||||
std::cout<<GridLogMessage << "Calling Deo and Doe and assert Deo+Doe == Dunprec"<<std::endl;
|
||||
pickCheckerboard(Even,src_e,src);
|
||||
pickCheckerboard(Odd,src_o,src);
|
||||
|
||||
std::cout<<GridLogMessage << "src_e"<<norm2(src_e)<<std::endl;
|
||||
std::cout<<GridLogMessage << "src_o"<<norm2(src_o)<<std::endl;
|
||||
|
||||
std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
|
||||
std::cout << GridLogMessage<< "* Benchmarking DomainWallFermionR::DhopEO "<<std::endl;
|
||||
std::cout << GridLogMessage<< "* Vectorising space-time 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;
|
||||
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;
|
||||
{
|
||||
Dw.ZeroCounters();
|
||||
FGrid->Barrier();
|
||||
double t0=usecond();
|
||||
for(int i=0;i<ncall;i++){
|
||||
Dw.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=(1344.0*volume*ncall)/2;
|
||||
@ -357,17 +396,14 @@ int main (int argc, char ** argv)
|
||||
|
||||
err = r_eo-result;
|
||||
std::cout<<GridLogMessage << "norm diff "<< norm2(err)<<std::endl;
|
||||
assert(norm2(err)<1.0e-5);
|
||||
assert(norm2(err)<1.0e-4);
|
||||
|
||||
pickCheckerboard(Even,src_e,err);
|
||||
pickCheckerboard(Odd,src_o,err);
|
||||
std::cout<<GridLogMessage << "norm diff even "<< norm2(src_e)<<std::endl;
|
||||
std::cout<<GridLogMessage << "norm diff odd "<< norm2(src_o)<<std::endl;
|
||||
assert(norm2(src_e)<1.0e-5);
|
||||
assert(norm2(src_o)<1.0e-5);
|
||||
|
||||
|
||||
}
|
||||
assert(norm2(src_e)<1.0e-4);
|
||||
assert(norm2(src_o)<1.0e-4);
|
||||
|
||||
Grid_finalize();
|
||||
}
|
||||
|
||||
@ -1,153 +0,0 @@
|
||||
/*************************************************************************************
|
||||
|
||||
Grid physics library, www.github.com/paboyle/Grid
|
||||
|
||||
Source file: ./benchmarks/Benchmark_dwf.cc
|
||||
|
||||
Copyright (C) 2015
|
||||
|
||||
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
|
||||
Author: paboyle <paboyle@ph.ed.ac.uk>
|
||||
|
||||
This program is free software; you can redistribute it and/or modify
|
||||
it under the terms of the GNU General Public License as published by
|
||||
the Free Software Foundation; either version 2 of the License, or
|
||||
(at your option) any later version.
|
||||
|
||||
This program is distributed in the hope that it will be useful,
|
||||
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
GNU General Public License for more details.
|
||||
|
||||
You should have received a copy of the GNU General Public License along
|
||||
with this program; if not, write to the Free Software Foundation, Inc.,
|
||||
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
|
||||
|
||||
See the full license in the file "LICENSE" in the top level distribution directory
|
||||
*************************************************************************************/
|
||||
/* END LEGAL */
|
||||
#include <Grid/Grid.h>
|
||||
|
||||
using namespace std;
|
||||
using namespace Grid;
|
||||
using namespace Grid::QCD;
|
||||
|
||||
template<class d>
|
||||
struct scal {
|
||||
d internal;
|
||||
};
|
||||
|
||||
Gamma::GammaMatrix Gmu [] = {
|
||||
Gamma::GammaX,
|
||||
Gamma::GammaY,
|
||||
Gamma::GammaZ,
|
||||
Gamma::GammaT
|
||||
};
|
||||
|
||||
bool overlapComms = false;
|
||||
|
||||
|
||||
int main (int argc, char ** argv)
|
||||
{
|
||||
Grid_init(&argc,&argv);
|
||||
|
||||
if( GridCmdOptionExists(argv,argv+argc,"--asynch") ){
|
||||
overlapComms = true;
|
||||
}
|
||||
|
||||
int threads = GridThread::GetThreads();
|
||||
std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
|
||||
|
||||
std::vector<int> latt4 = GridDefaultLatt();
|
||||
const int Ls=16;
|
||||
GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
|
||||
GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
|
||||
GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
|
||||
GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
|
||||
|
||||
std::vector<int> seeds4({1,2,3,4});
|
||||
std::vector<int> seeds5({5,6,7,8});
|
||||
|
||||
GridParallelRNG RNG4(UGrid); RNG4.SeedFixedIntegers(seeds4);
|
||||
GridParallelRNG RNG5(FGrid); RNG5.SeedFixedIntegers(seeds5);
|
||||
|
||||
LatticeFermion src (FGrid); random(RNG5,src);
|
||||
LatticeFermion result(FGrid); result=zero;
|
||||
LatticeFermion ref(FGrid); ref=zero;
|
||||
LatticeFermion tmp(FGrid);
|
||||
LatticeFermion err(FGrid);
|
||||
|
||||
ColourMatrix cm = Complex(1.0,0.0);
|
||||
|
||||
LatticeGaugeField Umu(UGrid);
|
||||
random(RNG4,Umu);
|
||||
|
||||
LatticeGaugeField Umu5d(FGrid);
|
||||
|
||||
// replicate across fifth dimension
|
||||
for(int ss=0;ss<Umu._grid->oSites();ss++){
|
||||
for(int s=0;s<Ls;s++){
|
||||
Umu5d._odata[Ls*ss+s] = Umu._odata[ss];
|
||||
}
|
||||
}
|
||||
|
||||
////////////////////////////////////
|
||||
// Naive wilson implementation
|
||||
////////////////////////////////////
|
||||
std::vector<LatticeColourMatrix> U(4,FGrid);
|
||||
for(int mu=0;mu<Nd;mu++){
|
||||
U[mu] = PeekIndex<LorentzIndex>(Umu5d,mu);
|
||||
}
|
||||
|
||||
if (1)
|
||||
{
|
||||
ref = zero;
|
||||
for(int mu=0;mu<Nd;mu++){
|
||||
|
||||
tmp = U[mu]*Cshift(src,mu+1,1);
|
||||
ref=ref + tmp - Gamma(Gmu[mu])*tmp;
|
||||
|
||||
tmp =adj(U[mu])*src;
|
||||
tmp =Cshift(tmp,mu+1,-1);
|
||||
ref=ref + tmp + Gamma(Gmu[mu])*tmp;
|
||||
}
|
||||
ref = -0.5*ref;
|
||||
}
|
||||
|
||||
RealD mass=0.1;
|
||||
RealD M5 =1.8;
|
||||
|
||||
typename DomainWallFermionR::ImplParams params;
|
||||
params.overlapCommsCompute = overlapComms;
|
||||
|
||||
RealD NP = UGrid->_Nprocessors;
|
||||
|
||||
|
||||
QCD::WilsonKernelsStatic::AsmOpt=1;
|
||||
|
||||
DomainWallFermionR Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,params);
|
||||
|
||||
std::cout<<GridLogMessage << "Calling Dw"<<std::endl;
|
||||
int ncall =50;
|
||||
if (1) {
|
||||
|
||||
double t0=usecond();
|
||||
for(int i=0;i<ncall;i++){
|
||||
Dw.Dhop(src,result,0);
|
||||
}
|
||||
double t1=usecond();
|
||||
|
||||
double volume=Ls; for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
|
||||
double flops=1344*volume*ncall;
|
||||
|
||||
std::cout<<GridLogMessage << "Called Dw "<<ncall<<" times in "<<t1-t0<<" us"<<std::endl;
|
||||
std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
|
||||
std::cout<<GridLogMessage << "norm ref "<< norm2(ref)<<std::endl;
|
||||
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t1-t0)<<std::endl;
|
||||
std::cout<<GridLogMessage << "mflop/s per node = "<< flops/(t1-t0)/NP<<std::endl;
|
||||
err = ref-result;
|
||||
std::cout<<GridLogMessage << "norm diff "<< norm2(err)<<std::endl;
|
||||
// Dw.Report();
|
||||
}
|
||||
Grid_finalize();
|
||||
}
|
||||
@ -51,24 +51,26 @@ int main (int argc, char ** argv)
|
||||
{
|
||||
Grid_init(&argc,&argv);
|
||||
|
||||
std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
|
||||
std::cout << GridLogMessage<< "* Kernel options --dslash-generic, --dslash-unroll, --dslash-asm" <<std::endl;
|
||||
std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
|
||||
|
||||
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;
|
||||
|
||||
const int Ls=8;
|
||||
int threads = GridThread::GetThreads();
|
||||
std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
|
||||
|
||||
if ( getenv("ASMOPT") ) {
|
||||
QCD::WilsonKernelsStatic::AsmOpt=1;
|
||||
} else {
|
||||
QCD::WilsonKernelsStatic::AsmOpt=0;
|
||||
}
|
||||
|
||||
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 << "=========================================================================="<<std::endl;
|
||||
|
||||
int Lmax=32;
|
||||
int dmin=0;
|
||||
int Lmax=16;
|
||||
int dmin=2;
|
||||
if ( getenv("LMAX") ) Lmax=atoi(getenv("LMAX"));
|
||||
if ( getenv("DMIN") ) dmin=atoi(getenv("DMIN"));
|
||||
for (int L=8;L<=Lmax;L*=2){
|
||||
|
||||
@ -58,6 +58,19 @@ int main (int argc, char ** argv)
|
||||
std::vector<int> seeds({1,2,3,4});
|
||||
RealD mass = 0.1;
|
||||
|
||||
std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
|
||||
std::cout << GridLogMessage<< "* Kernel options --dslash-generic, --dslash-unroll, --dslash-asm" <<std::endl;
|
||||
std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
|
||||
std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
|
||||
std::cout << GridLogMessage<< "* Benchmarking WilsonFermionR::Dhop "<<std::endl;
|
||||
std::cout << GridLogMessage<< "* Vectorising space-time 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;
|
||||
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;
|
||||
|
||||
std::cout<<GridLogMessage << "============================================================================="<< std::endl;
|
||||
std::cout<<GridLogMessage << "= Benchmarking Wilson" << std::endl;
|
||||
std::cout<<GridLogMessage << "============================================================================="<< std::endl;
|
||||
|
||||
@ -1,175 +0,0 @@
|
||||
/*************************************************************************************
|
||||
|
||||
Grid physics library, www.github.com/paboyle/Grid
|
||||
|
||||
Source file: ./tests/Test_zmm.cc
|
||||
|
||||
Copyright (C) 2015
|
||||
|
||||
Author: paboyle <paboyle@ph.ed.ac.uk>
|
||||
|
||||
This program is free software; you can redistribute it and/or modify
|
||||
it under the terms of the GNU General Public License as published by
|
||||
the Free Software Foundation; either version 2 of the License, or
|
||||
(at your option) any later version.
|
||||
|
||||
This program is distributed in the hope that it will be useful,
|
||||
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
GNU General Public License for more details.
|
||||
|
||||
You should have received a copy of the GNU General Public License along
|
||||
with this program; if not, write to the Free Software Foundation, Inc.,
|
||||
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
|
||||
|
||||
See the full license in the file "LICENSE" in the top level distribution directory
|
||||
*************************************************************************************/
|
||||
/* END LEGAL */
|
||||
#include <Grid/Grid.h>
|
||||
|
||||
|
||||
using namespace Grid;
|
||||
using namespace Grid::QCD;
|
||||
|
||||
|
||||
int bench(std::ofstream &os, std::vector<int> &latt4,int Ls);
|
||||
|
||||
int main(int argc,char **argv)
|
||||
{
|
||||
Grid_init(&argc,&argv);
|
||||
std::ofstream os("zmm.dat");
|
||||
|
||||
os << "#V Ls Lxy Lzt C++ Asm OMP L1 " <<std::endl;
|
||||
std::cout<<GridLogMessage << "====================================================================="<<std::endl;
|
||||
std::cout<<GridLogMessage << "= Benchmarking ZMM"<<std::endl;
|
||||
std::cout<<GridLogMessage << "====================================================================="<<std::endl;
|
||||
std::cout<<GridLogMessage << "Volume \t\t\t\tC++DW/MFLOPs\tASM-DW/MFLOPs\tdiff"<<std::endl;
|
||||
std::cout<<GridLogMessage << "====================================================================="<<std::endl;
|
||||
for(int L=4;L<=32;L+=4){
|
||||
for(int m=1;m<=2;m++){
|
||||
for(int Ls=8;Ls<=16;Ls+=8){
|
||||
std::vector<int> grid({L,L,m*L,m*L});
|
||||
std::cout << GridLogMessage <<"\t";
|
||||
for(int i=0;i<4;i++) {
|
||||
std::cout << grid[i]<<"x";
|
||||
}
|
||||
std::cout << Ls<<"\t\t";
|
||||
bench(os,grid,Ls);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
int bench(std::ofstream &os, std::vector<int> &latt4,int Ls)
|
||||
{
|
||||
|
||||
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);
|
||||
|
||||
std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
|
||||
std::vector<int> mpi_layout = GridDefaultMpi();
|
||||
int threads = GridThread::GetThreads();
|
||||
|
||||
std::vector<int> seeds4({1,2,3,4});
|
||||
std::vector<int> seeds5({5,6,7,8});
|
||||
|
||||
GridSerialRNG sRNG; sRNG.SeedFixedIntegers(seeds4);
|
||||
|
||||
LatticeFermion src (FGrid);
|
||||
LatticeFermion tmp (FGrid);
|
||||
LatticeFermion srce(FrbGrid);
|
||||
|
||||
LatticeFermion resulto(FrbGrid); resulto=zero;
|
||||
LatticeFermion resulta(FrbGrid); resulta=zero;
|
||||
LatticeFermion junk(FrbGrid); junk=zero;
|
||||
LatticeFermion diff(FrbGrid);
|
||||
LatticeGaugeField Umu(UGrid);
|
||||
|
||||
double mfc, mfa, mfo, mfl1;
|
||||
|
||||
GridParallelRNG RNG4(UGrid); RNG4.SeedFixedIntegers(seeds4);
|
||||
GridParallelRNG RNG5(FGrid); RNG5.SeedFixedIntegers(seeds5);
|
||||
random(RNG5,src);
|
||||
#if 1
|
||||
random(RNG4,Umu);
|
||||
#else
|
||||
int mmu=2;
|
||||
std::vector<LatticeColourMatrix> U(4,UGrid);
|
||||
for(int mu=0;mu<Nd;mu++){
|
||||
U[mu] = PeekIndex<LorentzIndex>(Umu,mu);
|
||||
if ( mu!=mmu ) U[mu] = zero;
|
||||
if ( mu==mmu ) U[mu] = 1.0;
|
||||
PokeIndex<LorentzIndex>(Umu,U[mu],mu);
|
||||
}
|
||||
#endif
|
||||
pickCheckerboard(Even,srce,src);
|
||||
|
||||
RealD mass=0.1;
|
||||
RealD M5 =1.8;
|
||||
DomainWallFermionR Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
|
||||
|
||||
int ncall=50;
|
||||
double t0=usecond();
|
||||
for(int i=0;i<ncall;i++){
|
||||
Dw.DhopOE(srce,resulto,0);
|
||||
}
|
||||
double t1=usecond();
|
||||
|
||||
double volume=Ls; for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
|
||||
double flops=1344*volume/2;
|
||||
|
||||
mfc = flops*ncall/(t1-t0);
|
||||
std::cout<<mfc<<"\t\t";
|
||||
|
||||
QCD::WilsonKernelsStatic::AsmOpt=1;
|
||||
t0=usecond();
|
||||
for(int i=0;i<ncall;i++){
|
||||
Dw.DhopOE(srce,resulta,0);
|
||||
}
|
||||
t1=usecond();
|
||||
mfa = flops*ncall/(t1-t0);
|
||||
std::cout<<mfa<<"\t\t";
|
||||
/*
|
||||
int dag=DaggerNo;
|
||||
t0=usecond();
|
||||
for(int i=0;i<1;i++){
|
||||
Dw.DhopInternalOMPbench(Dw.StencilEven,Dw.LebesgueEvenOdd,Dw.UmuOdd,srce,resulta,dag);
|
||||
}
|
||||
t1=usecond();
|
||||
mfo = flops*100/(t1-t0);
|
||||
std::cout<<GridLogMessage << "Called ASM-OMP Dw"<< " mflop/s = "<< mfo<<std::endl;
|
||||
|
||||
t0=usecond();
|
||||
for(int i=0;i<1;i++){
|
||||
Dw.DhopInternalL1bench(Dw.StencilEven,Dw.LebesgueEvenOdd,Dw.UmuOdd,srce,resulta,dag);
|
||||
}
|
||||
t1=usecond();
|
||||
mfl1= flops*100/(t1-t0);
|
||||
std::cout<<GridLogMessage << "Called ASM-L1 Dw"<< " mflop/s = "<< mfl1<<std::endl;
|
||||
os << latt4[0]*latt4[1]*latt4[2]*latt4[3]<< " "<<Ls<<" "<< latt4[0] <<" " <<latt4[2]<< " "
|
||||
<< mfc<<" "
|
||||
<< mfa<<" "
|
||||
<< mfo<<" "
|
||||
<< mfl1<<std::endl;
|
||||
*/
|
||||
|
||||
#if 0
|
||||
for(int i=0;i< PerformanceCounter::NumTypes(); i++ ){
|
||||
Dw.DhopOE(srce,resulta,0);
|
||||
PerformanceCounter Counter(i);
|
||||
Counter.Start();
|
||||
Dw.DhopOE(srce,resulta,0);
|
||||
Counter.Stop();
|
||||
Counter.Report();
|
||||
}
|
||||
#endif
|
||||
//resulta = (-0.5) * resulta;
|
||||
|
||||
diff = resulto-resulta;
|
||||
std::cout<<norm2(diff)<<std::endl;
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
Reference in New Issue
Block a user