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

Split sum in rankSum and GlobalSum

Browse Source
This commit is contained in:
Christoph Lehner 2022-12-26 10:01:32 +01:00
parent 317bdcf158
commit 88015b0858
1 changed files with 9 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
Grid/lattice/Lattice_reduction.h
View File

@ -144,17 +144,23 @@ inline typename vobj::scalar_objectD sumD(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)
{ {
#if defined(GRID_CUDA)||defined(GRID_HIP) #if defined(GRID_CUDA)||defined(GRID_HIP)
autoView( arg_v, arg, AcceleratorRead); autoView( arg_v, arg, AcceleratorRead);
Integer osites = arg.Grid()->oSites(); Integer osites = arg.Grid()->oSites();
auto 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);
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(const Lattice<vobj> &arg)
{
auto ssum = rankSum(arg);
arg.Grid()->GlobalSum(ssum); arg.Grid()->GlobalSum(ssum);
return ssum; return ssum;
} }