portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-03 02:35:55 +01:00
Code Releases Activity

Merge branch 'develop' of https://github.com/paboyle/Grid into develop

Browse Source
This commit is contained in:
Guido Cossu 2018-04-26 11:31:57 +09:00
commit c897878776
3 changed files with 66 additions and 34 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

63
benchmarks/Benchmark_su3.cc
View File

@ -35,24 +35,25 @@ using namespace Grid::QCD;
int main (int argc, char ** argv) int main (int argc, char ** argv)
{ {
Grid_init(&argc,&argv); Grid_init(&argc,&argv);
#define LMAX (40) #define LMAX (16)
#define LMIN (16)
#define LINC (4) #define LINC (4)
int64_t Nloop=20; int64_t Nloop=2000;
std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
std::vector<int> mpi_layout = GridDefaultMpi(); std::vector<int> mpi_layout = GridDefaultMpi();
int64_t threads = GridThread::GetThreads(); int64_t threads = GridThread::GetThreads();
std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl; std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
#if 1
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl; std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << "= Benchmarking SU3xSU3 x= x*y"<<std::endl; std::cout<<GridLogMessage << "= Benchmarking SU3xSU3 x= x*y"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl; std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl; std::cout<<GridLogMessage << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
std::cout<<GridLogMessage << "----------------------------------------------------------"<<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]}); 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]; int64_t vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
@ -147,30 +148,31 @@ int main (int argc, char ** argv)
std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl; std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
for(int lat=2;lat<=LMAX;lat+=LINC){ 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]}); 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]; int64_t vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
GridCartesian Grid(latt_size,simd_layout,mpi_layout); 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 z(&Grid); random(pRNG,z);
LatticeColourMatrix x(&Grid); random(pRNG,x); LatticeColourMatrix x(&Grid); random(pRNG,x);
LatticeColourMatrix y(&Grid); random(pRNG,y); LatticeColourMatrix y(&Grid); random(pRNG,y);
double start=usecond(); double start=usecond();
for(int64_t i=0;i<Nloop;i++){ for(int64_t i=0;i<Nloop;i++){
mac(z,x,y); 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;
} }
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;
}
#endif
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl; std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << "= Benchmarking SU3xSU3 CovShiftForward(z,x,y)"<<std::endl; std::cout<<GridLogMessage << "= Benchmarking SU3xSU3 CovShiftForward(z,x,y)"<<std::endl;
@ -178,7 +180,7 @@ 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 << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl; std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
for(int lat=2;lat<=LMAX;lat+=LINC){ for(int lat=LMIN;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]}); 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]; int64_t vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
@ -190,7 +192,7 @@ int main (int argc, char ** argv)
LatticeColourMatrix x(&Grid); random(pRNG,x); LatticeColourMatrix x(&Grid); random(pRNG,x);
LatticeColourMatrix y(&Grid); random(pRNG,y); LatticeColourMatrix y(&Grid); random(pRNG,y);
for(int mu=0;mu<=4;mu++){ for(int mu=0;mu<4;mu++){
double start=usecond(); double start=usecond();
for(int64_t i=0;i<Nloop;i++){ for(int64_t i=0;i<Nloop;i++){
z = PeriodicBC::CovShiftForward(x,mu,y); z = PeriodicBC::CovShiftForward(x,mu,y);
@ -198,11 +200,12 @@ int main (int argc, char ** argv)
double stop=usecond(); double stop=usecond();
double time = (stop-start)/Nloop*1000.0; double time = (stop-start)/Nloop*1000.0;
double bytes=3*vol*Nc*Nc*sizeof(Complex); double bytes=3*vol*Nc*Nc*sizeof(Complex);
double flops=Nc*Nc*(6+8+8)*vol; 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::setprecision(3) << lat<<"\t\t"<<bytes<<" \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
} }
} }
Grid_finalize(); Grid_finalize();
} }

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 ) { if ( !comm_dim ) {
// std::cout << "Cshift_local" <<std::endl; //std::cout << "CSHIFT: Cshift_local" <<std::endl;
Cshift_local(ret,rhs,dimension,shift); // Handles checkerboarding Cshift_local(ret,rhs,dimension,shift); // Handles checkerboarding
} else if ( splice_dim ) { } 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); Cshift_comms_simd(ret,rhs,dimension,shift);
} else { } else {
// std::cout << "Cshift_comms" <<std::endl; //std::cout << "CSHIFT: Cshift_comms" <<std::endl;
Cshift_comms(ret,rhs,dimension,shift); Cshift_comms(ret,rhs,dimension,shift);
} }
return ret; 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[0] = rhs._grid->CheckerBoardShiftForCB(rhs.checkerboard,dimension,shift,Even);
sshift[1] = rhs._grid->CheckerBoardShiftForCB(rhs.checkerboard,dimension,shift,Odd); 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] ) { if ( sshift[0] == sshift[1] ) {
//std::cout << "Single pass Cshift_comms" <<std::endl;
Cshift_comms_simd(ret,rhs,dimension,shift,0x3); Cshift_comms_simd(ret,rhs,dimension,shift,0x3);
} else { } 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,0x1);// if checkerboard is unfavourable take two passes
Cshift_comms_simd(ret,rhs,dimension,shift,0x2);// both with block stride loop iteration 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 simd_layout = grid->_simd_layout[dimension];
int comm_dim = grid->_processors[dimension] >1 ; 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(comm_dim==1);
assert(simd_layout==2); assert(simd_layout==2);
assert(shift>=0); assert(shift>=0);

24
lib/lattice/Lattice_base.h
View File

@ -257,7 +257,11 @@ public:
} }
} }
Lattice(Lattice&& r){ // move constructor
_grid = r._grid;
checkerboard = r.checkerboard;
_odata=std::move(r._odata);
}
virtual ~Lattice(void) = default; virtual ~Lattice(void) = default;
@ -286,6 +290,24 @@ public:
} }
return *this; return *this;
} }
strong_inline Lattice<vobj> & operator = (const Lattice<vobj> & r){
_grid = r._grid;
checkerboard = r.checkerboard;
_odata.resize(_grid->oSites());// essential
parallel_for(int ss=0;ss<_grid->oSites();ss++){
_odata[ss]=r._odata[ss];
}
return *this;
}
strong_inline Lattice<vobj> & operator = (Lattice<vobj> && r)
{
_grid = r._grid;
checkerboard = r.checkerboard;
_odata =std::move(r._odata);
return *this;
}
// *=,+=,-= operators inherit behvour from correspond */+/- operation // *=,+=,-= operators inherit behvour from correspond */+/- operation
template<class T> strong_inline Lattice<vobj> &operator *=(const T &r) { template<class T> strong_inline Lattice<vobj> &operator *=(const T &r) {