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

Merge pull request #419 from lehner/feature/gpt

Browse Source 
Separate rankSum from sum
This commit is contained in:
Peter Boyle 2023-03-24 15:42:16 -04:00 committed by GitHub
commit 4ea48ef0c4
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 18 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

25
Grid/lattice/Lattice_reduction.h
View File

@ -156,33 +156,44 @@ inline typename vobj::scalar_objectD sumD_large(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 rankSum(const Lattice<vobj> &arg)
{ {
Integer osites = arg.Grid()->oSites(); Integer osites = arg.Grid()->oSites();
#if defined(GRID_CUDA)||defined(GRID_HIP)||defined(GRID_SYCL) #if defined(GRID_CUDA)||defined(GRID_HIP)||defined(GRID_SYCL)
typename vobj::scalar_object ssum;
autoView( arg_v, arg, AcceleratorRead); autoView( arg_v, arg, AcceleratorRead);
ssum= sum_gpu(&arg_v[0],osites); return sum_gpu(&arg_v[0],osites);
#else #else
autoView(arg_v, arg, CpuRead); autoView(arg_v, arg, CpuRead);
auto ssum= sum_cpu(&arg_v[0],osites); return sum_cpu(&arg_v[0],osites);
#endif #endif
}
template<class vobj>
inline typename vobj::scalar_object sum(const Lattice<vobj> &arg)
{
auto ssum = rankSum(arg);
arg.Grid()->GlobalSum(ssum); arg.Grid()->GlobalSum(ssum);
return ssum; return ssum;
} }
template<class vobj> template<class vobj>
inline typename vobj::scalar_object sum_large(const Lattice<vobj> &arg) inline typename vobj::scalar_object rankSumLarge(const Lattice<vobj> &arg)
{ {
#if defined(GRID_CUDA)||defined(GRID_HIP)||defined(GRID_SYCL) #if defined(GRID_CUDA)||defined(GRID_HIP)||defined(GRID_SYCL)
autoView( arg_v, arg, AcceleratorRead); autoView( arg_v, arg, AcceleratorRead);
Integer osites = arg.Grid()->oSites(); Integer osites = arg.Grid()->oSites();
auto ssum= sum_gpu_large(&arg_v[0],osites); return sum_gpu_large(&arg_v[0],osites);
#else #else
autoView(arg_v, arg, CpuRead); autoView(arg_v, arg, CpuRead);
Integer osites = arg.Grid()->oSites(); Integer osites = arg.Grid()->oSites();
auto ssum= sum_cpu(&arg_v[0],osites); return sum_cpu(&arg_v[0],osites);
#endif #endif
}
template<class vobj>
inline typename vobj::scalar_object sum_large(const Lattice<vobj> &arg)
{
auto ssum = rankSumLarge(arg);
arg.Grid()->GlobalSum(ssum); arg.Grid()->GlobalSum(ssum);
return ssum; return ssum;
} }