portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-09 21:50:45 +01:00
Code Releases Activity

Accelerator loop use

Browse Source
This commit is contained in:
Peter Boyle 2020-06-03 22:50:09 -04:00
parent 6bf7f839ff
commit f67830587f
1 changed files with 3 additions and 33 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

36
Grid/lattice/Lattice_reduction.h
View File

@ -78,7 +78,7 @@ inline typename vobj::scalar_object sum(const vobj *arg, Integer osites)
template<class vobj> template<class vobj>
inline typename vobj::scalar_object sum(const Lattice<vobj> &arg) inline typename vobj::scalar_object sum(const Lattice<vobj> &arg)
{ {
#if defined(GRID_CUDA) #if defined(GRID_CUDA)||defined(GRID_HIP)
auto arg_v = arg.View(AcceleratorRead); auto arg_v = arg.View(AcceleratorRead);
Integer osites = arg.Grid()->oSites(); Integer osites = arg.Grid()->oSites();
auto ssum= sum_gpu(&arg_v[0],osites); auto ssum= sum_gpu(&arg_v[0],osites);
@ -112,7 +112,6 @@ inline ComplexD innerProduct(const Lattice<vobj> &left,const Lattice<vobj> &righ
const uint64_t nsimd = grid->Nsimd(); const uint64_t nsimd = grid->Nsimd();
const uint64_t sites = grid->oSites(); const uint64_t sites = grid->oSites();
#if defined(GRID_CUDA)
// Might make all code paths go this way. // Might make all code paths go this way.
auto left_v = left.View(AcceleratorRead); auto left_v = left.View(AcceleratorRead);
auto right_v=right.View(AcceleratorRead); auto right_v=right.View(AcceleratorRead);
@ -121,7 +120,6 @@ inline ComplexD innerProduct(const Lattice<vobj> &left,const Lattice<vobj> &righ
typedef decltype(innerProduct(left_v[0],right_v[0])) inner_t; typedef decltype(innerProduct(left_v[0],right_v[0])) inner_t;
Vector<inner_t> inner_tmp(sites); Vector<inner_t> inner_tmp(sites);
auto inner_tmp_v = &inner_tmp[0]; auto inner_tmp_v = &inner_tmp[0];
accelerator_for( ss, sites, nsimd,{ accelerator_for( ss, sites, nsimd,{
auto x_l = left_v(ss); auto x_l = left_v(ss);
@ -131,22 +129,9 @@ inline ComplexD innerProduct(const Lattice<vobj> &left,const Lattice<vobj> &righ
// This is in single precision and fails some tests // This is in single precision and fails some tests
// Need a sumD that sums in double // Need a sumD that sums in double
#if defined(GRID_CUDA)||defined(GRID_HIP)
nrm = TensorRemove(sumD_gpu(inner_tmp_v,sites)); nrm = TensorRemove(sumD_gpu(inner_tmp_v,sites));
#else #else
// Might make all code paths go this way.
auto left_v = left.View(CpuRead);
auto right_v=right.View(CpuRead);
// CPU
typedef decltype(innerProductD(left_v[0],right_v[0])) inner_t;
Vector<inner_t> inner_tmp(sites);
auto inner_tmp_v = &inner_tmp[0];
thread_for( ss, sites,{
auto x_l = left_v[ss];
auto y_l = right_v[ss];
inner_tmp_v[ss]=innerProductD(x_l,y_l);
})
nrm = TensorRemove(sum_cpu(inner_tmp_v,sites)); nrm = TensorRemove(sum_cpu(inner_tmp_v,sites));
#endif #endif
grid->GlobalSum(nrm); grid->GlobalSum(nrm);
@ -182,7 +167,6 @@ axpby_norm_fast(Lattice<vobj> &z,sobj a,sobj b,const Lattice<vobj> &x,const Latt
const uint64_t nsimd = grid->Nsimd(); const uint64_t nsimd = grid->Nsimd();
const uint64_t sites = grid->oSites(); const uint64_t sites = grid->oSites();
#if defined(GRID_CUDA)||defined(GRID_HIP)
// GPU // GPU
auto x_v=x.View(AcceleratorRead); auto x_v=x.View(AcceleratorRead);
auto y_v=y.View(AcceleratorRead); auto y_v=y.View(AcceleratorRead);
@ -197,23 +181,9 @@ axpby_norm_fast(Lattice<vobj> &z,sobj a,sobj b,const Lattice<vobj> &x,const Latt
coalescedWrite(inner_tmp_v[ss],innerProduct(tmp,tmp)); coalescedWrite(inner_tmp_v[ss],innerProduct(tmp,tmp));
coalescedWrite(z_v[ss],tmp); coalescedWrite(z_v[ss],tmp);
}); });
#if defined(GRID_CUDA)||defined(GRID_HIP)
nrm = real(TensorRemove(sumD_gpu(inner_tmp_v,sites))); nrm = real(TensorRemove(sumD_gpu(inner_tmp_v,sites)));
#else #else
auto x_v=x.View(AcceleratorRead);
auto y_v=y.View(AcceleratorRead);
auto z_v=z.View(AcceleratorWrite);
// CPU
typedef decltype(innerProductD(x_v[0],y_v[0])) inner_t;
Vector<inner_t> inner_tmp(sites);
auto inner_tmp_v = &inner_tmp[0];
accelerator_for( ss, sites, nsimd,{
auto tmp = a*x_v(ss)+b*y_v(ss);
inner_tmp_v[ss]=innerProductD(tmp,tmp);
z_v[ss]=tmp;
});
// Already promoted to double // Already promoted to double
nrm = real(TensorRemove(sum(inner_tmp_v,sites))); nrm = real(TensorRemove(sum(inner_tmp_v,sites)));
#endif #endif