portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-10 07:55:35 +00:00
Code Releases Activity

Correction of a minor bug in the su3 benchmark

Browse Source
This commit is contained in:
Guido Cossu 2018-04-24 08:03:57 -07:00
parent a1be533329
commit c5b9147b53
2 changed files with 36 additions and 29 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

52
benchmarks/Benchmark_su3.cc
View File

@ -52,7 +52,7 @@ int main (int argc, char ** argv)
std::cout<<GridLogMessage << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
for(int lat=2;lat<=LMAX;lat+=LINC){
for(int lat=4;lat<=LMAX;lat+=LINC){
std::vector<int> latt_size ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
int64_t vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
@ -147,30 +147,30 @@ int main (int argc, char ** argv)
std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
for(int lat=2;lat<=LMAX;lat+=LINC){
std::vector<int> latt_size ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[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}));
LatticeColourMatrix z(&Grid); random(pRNG,z);
LatticeColourMatrix x(&Grid); random(pRNG,x);
LatticeColourMatrix y(&Grid); random(pRNG,y);
double start=usecond();
for(int64_t i=0;i<Nloop;i++){
mac(z,x,y);
}
double stop=usecond();
double time = (stop-start)/Nloop*1000.0;
double bytes=3*vol*Nc*Nc*sizeof(Complex);
double flops=Nc*Nc*(6+8+8)*vol;
std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
std::vector<int> latt_size ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[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}));
LatticeColourMatrix z(&Grid); random(pRNG,z);
LatticeColourMatrix x(&Grid); random(pRNG,x);
LatticeColourMatrix y(&Grid); random(pRNG,y);
double start=usecond();
for(int64_t i=0;i<Nloop;i++){
mac(z,x,y);
}
double stop=usecond();
double time = (stop-start)/Nloop*1000.0;
double bytes=3*vol*Nc*Nc*sizeof(Complex);
double flops=Nc*Nc*(6+8+8)*vol;
std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
}
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << "= Benchmarking SU3xSU3 CovShiftForward(z,x,y)"<<std::endl;
@ -179,7 +179,6 @@ int main (int argc, char ** argv)
std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
for(int lat=2;lat<=LMAX;lat+=LINC){
std::vector<int> latt_size ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
int64_t vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
@ -190,7 +189,7 @@ int main (int argc, char ** argv)
LatticeColourMatrix x(&Grid); random(pRNG,x);
LatticeColourMatrix y(&Grid); random(pRNG,y);
for(int mu=0;mu<=4;mu++){
for(int mu=0;mu<4;mu++){
double start=usecond();
for(int64_t i=0;i<Nloop;i++){
z = PeriodicBC::CovShiftForward(x,mu,y);
@ -198,6 +197,7 @@ int main (int argc, char ** argv)
double stop=usecond();
double time = (stop-start)/Nloop*1000.0;
double bytes=3*vol*Nc*Nc*sizeof(Complex);
double flops=Nc*Nc*(6+8+8)*vol;
std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;

13
lib/cshift/Cshift_mpi.h
View File

@ -54,13 +54,13 @@ template<class vobj> Lattice<vobj> Cshift(const Lattice<vobj> &rhs,int dimension
if ( !comm_dim ) {
// std::cout << "Cshift_local" <<std::endl;
//std::cout << "CSHIFT: Cshift_local" <<std::endl;
Cshift_local(ret,rhs,dimension,shift); // Handles checkerboarding
} else if ( splice_dim ) {
// std::cout << "Cshift_comms_simd" <<std::endl;
//std::cout << "CSHIFT: Cshift_comms_simd call - splice_dim = " << splice_dim << " shift " << shift << " dimension = " << dimension << std::endl;
Cshift_comms_simd(ret,rhs,dimension,shift);
} else {
// std::cout << "Cshift_comms" <<std::endl;
//std::cout << "CSHIFT: Cshift_comms" <<std::endl;
Cshift_comms(ret,rhs,dimension,shift);
}
return ret;
@ -91,9 +91,12 @@ template<class vobj> void Cshift_comms_simd(Lattice<vobj>& ret,const Lattice<vob
sshift[0] = rhs._grid->CheckerBoardShiftForCB(rhs.checkerboard,dimension,shift,Even);
sshift[1] = rhs._grid->CheckerBoardShiftForCB(rhs.checkerboard,dimension,shift,Odd);
//std::cout << "Cshift_comms_simd dim "<<dimension<<"cb "<<rhs.checkerboard<<"shift "<<shift<<" sshift " << sshift[0]<<" "<<sshift[1]<<std::endl;
if ( sshift[0] == sshift[1] ) {
//std::cout << "Single pass Cshift_comms" <<std::endl;
Cshift_comms_simd(ret,rhs,dimension,shift,0x3);
} else {
//std::cout << "Two pass Cshift_comms" <<std::endl;
Cshift_comms_simd(ret,rhs,dimension,shift,0x1);// if checkerboard is unfavourable take two passes
Cshift_comms_simd(ret,rhs,dimension,shift,0x2);// both with block stride loop iteration
}
@ -175,6 +178,10 @@ template<class vobj> void Cshift_comms_simd(Lattice<vobj> &ret,const Lattice<vo
int simd_layout = grid->_simd_layout[dimension];
int comm_dim = grid->_processors[dimension] >1 ;
//std::cout << "Cshift_comms_simd dim "<< dimension << " fd "<<fd<<" rd "<<rd
// << " ld "<<ld<<" pd " << pd<<" simd_layout "<<simd_layout
// << " comm_dim " << comm_dim << " cbmask " << cbmask <<std::endl;
assert(comm_dim==1);
assert(simd_layout==2);
assert(shift>=0);