portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-13 04:37:05 +01:00
Code Releases Activity

Merge branch 'develop' into feature/multi-communicator

Browse Source
This commit is contained in:
Peter Boyle
2017-08-19 13:03:35 -04:00
parent 2f619482b8 c7036f6717
commit 7d88198387
27 changed files with 875 additions and 240 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
benchmarks/Benchmark_dwf.cc
View File

@ -165,7 +165,7 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage<< "*****************************************************************" <<std::endl;
DomainWallFermionR Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
int ncall =1000;
int ncall =500;
if (1) {
FGrid->Barrier();
Dw.ZeroCounters();
@ -302,6 +302,7 @@ int main (int argc, char ** argv)
std::cout<< "sD ERR \n " << err <<std::endl;
}
assert(sum < 1.0e-4);
if(1){
std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
@ -381,8 +382,23 @@ int main (int argc, char ** argv)
}
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;

60
benchmarks/Benchmark_memory_bandwidth.cc
View File

@ -55,21 +55,21 @@ int main (int argc, char ** argv)
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<"\t\t seconds"<<std::endl;
std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
uint64_t lmax=64;
#define NLOOP (100*lmax*lmax*lmax*lmax/vol)
for(int lat=4;lat<=lmax;lat+=4){
uint64_t lmax=96;
#define NLOOP (10*lmax*lmax*lmax*lmax/vol)
for(int lat=8;lat<=lmax;lat+=8){
std::vector<int> latt_size ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
int vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
int64_t vol= latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
GridCartesian Grid(latt_size,simd_layout,mpi_layout);
uint64_t Nloop=NLOOP;
// GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9});
// GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9}));
LatticeVec z(&Grid); //random(pRNG,z);
LatticeVec x(&Grid); //random(pRNG,x);
LatticeVec y(&Grid); //random(pRNG,y);
LatticeVec z(&Grid);// random(pRNG,z);
LatticeVec x(&Grid);// random(pRNG,x);
LatticeVec y(&Grid);// random(pRNG,y);
double a=2.0;
@ -83,7 +83,7 @@ int main (int argc, char ** argv)
double time = (stop-start)/Nloop*1000;
double flops=vol*Nvec*2;// mul,add
double bytes=3*vol*Nvec*sizeof(Real);
double bytes=3.0*vol*Nvec*sizeof(Real);
std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t"<<flops/time<<"\t\t"<<(stop-start)/1000./1000.<<std::endl;
}
@ -94,17 +94,17 @@ int main (int argc, char ** argv)
std::cout<<GridLogMessage << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<"\t\t seconds"<<std::endl;
std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
for(int lat=4;lat<=lmax;lat+=4){
for(int lat=8;lat<=lmax;lat+=8){
std::vector<int> latt_size ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
int vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
int64_t vol= latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
GridCartesian Grid(latt_size,simd_layout,mpi_layout);
// GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9});
// GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9}));
LatticeVec z(&Grid); //random(pRNG,z);
LatticeVec x(&Grid); //random(pRNG,x);
LatticeVec y(&Grid); //random(pRNG,y);
LatticeVec z(&Grid);// random(pRNG,z);
LatticeVec x(&Grid);// random(pRNG,x);
LatticeVec y(&Grid);// random(pRNG,y);
double a=2.0;
uint64_t Nloop=NLOOP;
@ -119,7 +119,7 @@ int main (int argc, char ** argv)
double time = (stop-start)/Nloop*1000;
double flops=vol*Nvec*2;// mul,add
double bytes=3*vol*Nvec*sizeof(Real);
double bytes=3.0*vol*Nvec*sizeof(Real);
std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t"<<flops/time<<"\t\t"<<(stop-start)/1000./1000.<<std::endl;
}
@ -129,20 +129,20 @@ int main (int argc, char ** argv)
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<"\t\t seconds"<<std::endl;
for(int lat=4;lat<=lmax;lat+=4){
for(int lat=8;lat<=lmax;lat+=8){
std::vector<int> latt_size ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
int vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
int64_t vol= latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
uint64_t Nloop=NLOOP;
GridCartesian Grid(latt_size,simd_layout,mpi_layout);
// GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9});
// GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9}));
LatticeVec z(&Grid); //random(pRNG,z);
LatticeVec x(&Grid); //random(pRNG,x);
LatticeVec y(&Grid); //random(pRNG,y);
LatticeVec z(&Grid);// random(pRNG,z);
LatticeVec x(&Grid);// random(pRNG,x);
LatticeVec y(&Grid);// random(pRNG,y);
RealD a=2.0;
@ -154,7 +154,7 @@ int main (int argc, char ** argv)
double stop=usecond();
double time = (stop-start)/Nloop*1000;
double bytes=2*vol*Nvec*sizeof(Real);
double bytes=2.0*vol*Nvec*sizeof(Real);
double flops=vol*Nvec*1;// mul
std::cout<<GridLogMessage <<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t"<<flops/time<<"\t\t"<<(stop-start)/1000./1000.<<std::endl;
@ -166,17 +166,17 @@ int main (int argc, char ** argv)
std::cout<<GridLogMessage << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<"\t\t seconds"<<std::endl;
std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
for(int lat=4;lat<=lmax;lat+=4){
for(int lat=8;lat<=lmax;lat+=8){
std::vector<int> latt_size ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
int vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
int64_t vol= latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
uint64_t Nloop=NLOOP;
GridCartesian Grid(latt_size,simd_layout,mpi_layout);
// GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9});
LatticeVec z(&Grid); //random(pRNG,z);
LatticeVec x(&Grid); //random(pRNG,x);
LatticeVec y(&Grid); //random(pRNG,y);
// GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9}));
LatticeVec z(&Grid);// random(pRNG,z);
LatticeVec x(&Grid);// random(pRNG,x);
LatticeVec y(&Grid);// random(pRNG,y);
RealD a=2.0;
Real nn;
double start=usecond();
@ -187,7 +187,7 @@ int main (int argc, char ** argv)
double stop=usecond();
double time = (stop-start)/Nloop*1000;
double bytes=vol*Nvec*sizeof(Real);
double bytes=1.0*vol*Nvec*sizeof(Real);
double flops=vol*Nvec*2;// mul,add
std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t"<<flops/time<< "\t\t"<<(stop-start)/1000./1000.<< "\t\t " <<std::endl;

52
benchmarks/Benchmark_su3.cc
View File

@ -37,12 +37,12 @@ int main (int argc, char ** argv)
Grid_init(&argc,&argv);
#define LMAX (64)
int Nloop=20;
int64_t Nloop=20;
std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
std::vector<int> mpi_layout = GridDefaultMpi();
int threads = GridThread::GetThreads();
int64_t threads = GridThread::GetThreads();
std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
@ -54,16 +54,16 @@ int main (int argc, char ** argv)
for(int lat=2;lat<=LMAX;lat+=2){
std::vector<int> latt_size ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
int vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
int64_t vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
GridCartesian Grid(latt_size,simd_layout,mpi_layout);
// GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9});
GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9}));
LatticeColourMatrix z(&Grid);// random(pRNG,z);
LatticeColourMatrix x(&Grid);// random(pRNG,x);
LatticeColourMatrix y(&Grid);// random(pRNG,y);
LatticeColourMatrix z(&Grid); random(pRNG,z);
LatticeColourMatrix x(&Grid); random(pRNG,x);
LatticeColourMatrix y(&Grid); random(pRNG,y);
double start=usecond();
for(int i=0;i<Nloop;i++){
for(int64_t i=0;i<Nloop;i++){
x=x*y;
}
double stop=usecond();
@ -86,17 +86,17 @@ int main (int argc, char ** argv)
for(int lat=2;lat<=LMAX;lat+=2){
std::vector<int> latt_size ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
int vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
int64_t vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
GridCartesian Grid(latt_size,simd_layout,mpi_layout);
// GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9});
GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9}));
LatticeColourMatrix z(&Grid); //random(pRNG,z);
LatticeColourMatrix x(&Grid); //random(pRNG,x);
LatticeColourMatrix y(&Grid); //random(pRNG,y);
LatticeColourMatrix z(&Grid); random(pRNG,z);
LatticeColourMatrix x(&Grid); random(pRNG,x);
LatticeColourMatrix y(&Grid); random(pRNG,y);
double start=usecond();
for(int i=0;i<Nloop;i++){
for(int64_t i=0;i<Nloop;i++){
z=x*y;
}
double stop=usecond();
@ -117,17 +117,17 @@ int main (int argc, char ** argv)
for(int lat=2;lat<=LMAX;lat+=2){
std::vector<int> latt_size ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
int vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
int64_t vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
GridCartesian Grid(latt_size,simd_layout,mpi_layout);
// GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9});
GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9}));
LatticeColourMatrix z(&Grid); //random(pRNG,z);
LatticeColourMatrix x(&Grid); //random(pRNG,x);
LatticeColourMatrix y(&Grid); //random(pRNG,y);
LatticeColourMatrix z(&Grid); random(pRNG,z);
LatticeColourMatrix x(&Grid); random(pRNG,x);
LatticeColourMatrix y(&Grid); random(pRNG,y);
double start=usecond();
for(int i=0;i<Nloop;i++){
for(int64_t i=0;i<Nloop;i++){
mult(z,x,y);
}
double stop=usecond();
@ -148,17 +148,17 @@ int main (int argc, char ** argv)
for(int lat=2;lat<=LMAX;lat+=2){
std::vector<int> latt_size ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
int vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
int64_t vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
GridCartesian Grid(latt_size,simd_layout,mpi_layout);
// GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9});
GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9}));
LatticeColourMatrix z(&Grid); //random(pRNG,z);
LatticeColourMatrix x(&Grid); //random(pRNG,x);
LatticeColourMatrix y(&Grid); //random(pRNG,y);
LatticeColourMatrix z(&Grid); random(pRNG,z);
LatticeColourMatrix x(&Grid); random(pRNG,x);
LatticeColourMatrix y(&Grid); random(pRNG,y);
double start=usecond();
for(int i=0;i<Nloop;i++){
for(int64_t i=0;i<Nloop;i++){
mac(z,x,y);
}
double stop=usecond();