portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-11-04 05:54:32 +00:00
Code Releases Activity

Comms and memory benchmarks added

Browse Source
This commit is contained in:
Peter Boyle
2015-05-03 09:44:47 +01:00
parent 253362f978
commit 9d93d1e6d4
14 changed files with 300 additions and 59 deletions
Download Patch File Download Diff File Expand all files Collapse all files

38
benchmarks/Grid_comms.cc
View File

@@ -25,17 +25,19 @@ int main (int argc, char ** argv)
for(int lat=4;lat<=16;lat+=4){
for(int Ls=1;Ls<=16;Ls*=2){
std::vector<int> latt_size ({lat,lat,lat,lat});
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));
int ncomm;
int bytes=lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
double start=usecond();
int ncomm=0;
for(int i=0;i<Nloop;i++){
std::vector<int> latt_size ({lat,lat,lat,lat});
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<CartesianCommunicator::CommsRequest_t> requests;
ncomm=0;
@@ -68,11 +70,10 @@ int main (int argc, char ** argv)
}
}
Grid.SendToRecvFromComplete(requests);
Grid.Barrier();
}
}
double stop=usecond();
double xbytes = Nloop*bytes*2*ncomm;
@@ -96,18 +97,20 @@ int main (int argc, char ** argv)
for(int lat=4;lat<=16;lat+=4){
for(int Ls=1;Ls<=16;Ls*=2){
std::vector<int> latt_size ({lat,lat,lat,lat});
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));
int ncomm;
int bytes=lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
double start=usecond();
int ncomm=0;
for(int i=0;i<Nloop;i++){
std::vector<int> latt_size ({lat,lat,lat,lat});
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));
ncomm=0;
for(int mu=0;mu<4;mu++){
@@ -131,7 +134,6 @@ int main (int argc, char ** argv)
}
comm_proc = mpi_layout[mu]-1;
{
std::vector<CartesianCommunicator::CommsRequest_t> requests;
Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);

150
benchmarks/Grid_memory_bandwidth.cc Normal file
View File

@@ -0,0 +1,150 @@
#include <Grid.h>
using namespace std;
using namespace Grid;
using namespace Grid::QCD;
int main (int argc, char ** argv)
{
Grid_init(&argc,&argv);
std::vector<int> simd_layout({1,2,2,2});
std::vector<int> mpi_layout ({1,1,1,1});
const int Nvec=8;
typedef Lattice< iVector< vReal,Nvec> > LatticeVec;
int Nloop=100;
std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking AXPY bandwidth"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s"<<std::endl;
for(int lat=4;lat<=32;lat+=4){
std::vector<int> latt_size ({lat,lat,lat,lat});
GridCartesian Grid(latt_size,simd_layout,mpi_layout);
GridParallelRNG pRNG(&Grid); pRNG.SeedRandomDevice();
LatticeVec z(&Grid); random(pRNG,z);
LatticeVec x(&Grid); random(pRNG,x);
LatticeVec y(&Grid); random(pRNG,y);
double a=1.0;
double start=usecond();
for(int i=0;i<Nloop;i++){
// z=a*x+y;
// inline void axpy(Lattice<vobj> &ret,double a,const Lattice<vobj> &lhs,const Lattice<vobj> &rhs){
axpy(z,a,x,y);
}
double stop=usecond();
double time = stop-start;
double bytes=3*lat*lat*lat*lat*Nvec*sizeof(Real)*Nloop;
std::cout << lat<<"\t\t"<<bytes<<"\t\t"<<bytes/time<<std::endl;
}
std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking a*x + y bandwidth"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s"<<std::endl;
for(int lat=4;lat<=32;lat+=4){
std::vector<int> latt_size ({lat,lat,lat,lat});
GridCartesian Grid(latt_size,simd_layout,mpi_layout);
GridParallelRNG pRNG(&Grid); pRNG.SeedRandomDevice();
LatticeVec z(&Grid); random(pRNG,z);
LatticeVec x(&Grid); random(pRNG,x);
LatticeVec y(&Grid); random(pRNG,y);
double a=1.0;
double start=usecond();
for(int i=0;i<Nloop;i++){
z=a*x+y;
}
double stop=usecond();
double time = stop-start;
double bytes=3*lat*lat*lat*lat*Nvec*sizeof(Real)*Nloop;
std::cout << lat<<"\t\t"<<bytes<<"\t\t"<<bytes/time<<std::endl;
}
std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking COPY bandwidth"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s"<<std::endl;
for(int lat=4;lat<=32;lat+=4){
std::vector<int> latt_size ({lat,lat,lat,lat});
GridCartesian Grid(latt_size,simd_layout,mpi_layout);
GridParallelRNG pRNG(&Grid); pRNG.SeedRandomDevice();
LatticeVec z(&Grid); random(pRNG,z);
LatticeVec x(&Grid); random(pRNG,x);
LatticeVec y(&Grid); random(pRNG,y);
RealD a=1.0;
double start=usecond();
for(int i=0;i<Nloop;i++){
x=z;
}
double stop=usecond();
double time = stop-start;
double bytes=2*lat*lat*lat*lat*Nvec*sizeof(Real)*Nloop;
std::cout << lat<<"\t\t"<<bytes<<"\t\t"<<bytes/time<<std::endl;
}
std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking READ bandwidth"<<std::endl;
std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s"<<std::endl;
for(int lat=4;lat<=32;lat+=4){
std::vector<int> latt_size ({lat,lat,lat,lat});
GridCartesian Grid(latt_size,simd_layout,mpi_layout);
GridParallelRNG pRNG(&Grid); pRNG.SeedRandomDevice();
LatticeVec z(&Grid); random(pRNG,z);
LatticeVec x(&Grid); random(pRNG,x);
LatticeVec y(&Grid); random(pRNG,y);
RealD a=1.0;
ComplexD nn;
double start=usecond();
for(int i=0;i<Nloop;i++){
nn=norm2(x);
}
double stop=usecond();
double time = stop-start;
double bytes=lat*lat*lat*lat*Nvec*sizeof(Real)*Nloop;
std::cout << lat<<"\t\t"<<bytes<<"\t\t"<<bytes/time<<std::endl;
}
Grid_finalize();
}