portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-29 15:15:56 +01:00
Code Releases Activity

HIP runs sensible

Browse Source
This commit is contained in:
Peter Boyle 2020-09-16 03:35:03 +01:00
parent ecd3f890f5
commit 81441e98f4
4 changed files with 44 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
Grid/lattice/Lattice_arith.h
View File

@ -60,9 +60,9 @@ void mac(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
autoView( lhs_v , lhs, AcceleratorRead);
autoView( rhs_v , rhs, AcceleratorRead);
accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
decltype(coalescedRead(obj1())) tmp;
auto lhs_t=lhs_v(ss);
auto rhs_t=rhs_v(ss);
auto tmp =ret_v(ss);
mac(&tmp,&lhs_t,&rhs_t);
coalescedWrite(ret_v[ss],tmp);
});
@ -124,7 +124,7 @@ void mac(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
autoView( ret_v , ret, AcceleratorWrite);
autoView( lhs_v , lhs, AcceleratorRead);
accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
decltype(coalescedRead(obj1())) tmp;
auto tmp =ret_v(ss);
auto lhs_t=lhs_v(ss);
mac(&tmp,&lhs_t,&rhs);
coalescedWrite(ret_v[ss],tmp);
@ -182,7 +182,7 @@ void mac(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
autoView( ret_v , ret, AcceleratorWrite);
autoView( rhs_v , lhs, AcceleratorRead);
accelerator_for(ss,rhs_v.size(),obj1::Nsimd(),{
decltype(coalescedRead(obj1())) tmp;
auto tmp =ret_v(ss);
auto rhs_t=rhs_v(ss);
mac(&tmp,&lhs,&rhs_t);
coalescedWrite(ret_v[ss],tmp);

2
Grid/threads/Accelerator.cc
View File

@ -127,7 +127,7 @@ void acceleratorInit(void)
printf("AcceleratorHipInit: ========================\n");
printf("AcceleratorHipInit: Device identifier: %s\n", prop.name);
GPU_PROP_FMT(totalGlobalMem,"%lld");
GPU_PROP_FMT(totalGlobalMem,"%lu");
// GPU_PROP(managedMemory);
GPU_PROP(isMultiGpuBoard);
GPU_PROP(warpSize);

42
benchmarks/Benchmark_su3.cc
View File

@ -36,12 +36,12 @@ int main (int argc, char ** argv)
{
Grid_init(&argc,&argv);
#define LMAX (48)
#define LMAX (40)
#define LMIN (8)
#define LADD (8)
int64_t Nwarm=50;
int64_t Nloop=500;
int64_t Nwarm=10;
int64_t Nloop=100;
Coordinate simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
Coordinate mpi_layout = GridDefaultMpi();
@ -118,6 +118,41 @@ int main (int argc, char ** argv)
}
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << "= Benchmarking SU3xSU3 z=z+ x*y"<<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 << "----------------------------------------------------------"<<std::endl;
for(int lat=LMIN;lat<=LMAX;lat+=LADD){
Coordinate 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);
for(int64_t i=0;i<Nwarm;i++){
z=z+x*y;
}
double start=usecond();
for(int64_t i=0;i<Nloop;i++){
z=z+x*y;
}
double stop=usecond();
double time = (stop-start)/Nloop*1000.0;
double bytes=4*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 mult(z,x,y)"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
@ -143,7 +178,6 @@ int main (int argc, char ** argv)
double start=usecond();
for(int64_t i=0;i<Nloop;i++){
mult(z,x,y);
// mac(z,x,y);
}
double stop=usecond();
double time = (stop-start)/Nloop*1000.0;

3
benchmarks/Benchmark_su3_gpu.cc
View File

@ -187,7 +187,8 @@ int main (int argc, char ** argv)
auto xx = coalescedRead(x_v[ss]);
auto yy = coalescedRead(y_v[ss]);
auto zz = coalescedRead(z_v[ss]);
zz = zz+xx*yy;
//zz = zz+xx*yy;
mac(&zz,&xx,&yy);
coalescedWrite(z_v[ss],zz);
});
}