portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-04 19:25:56 +01:00
Code Releases Activity

sum_gpu_large and sum_gpu templates added.

Browse Source
This commit is contained in:
Fabian Joswig 2022-03-02 15:40:18 +00:00
parent 9aac1e6d64
commit d4ae71b880
2 changed files with 24 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
Grid/lattice/Lattice_reduction.h
View File

@ -168,6 +168,22 @@ inline typename vobj::scalar_object sum(const Lattice<vobj> &arg)
return ssum;
}
template<class vobj>
inline typename vobj::scalar_object sum_large(const Lattice<vobj> &arg)
{
#if defined(GRID_CUDA)||defined(GRID_HIP)
autoView( arg_v, arg, AcceleratorRead);
Integer osites = arg.Grid()->oSites();
auto ssum= sum_gpu_large(&arg_v[0],osites);
#else
autoView(arg_v, arg, CpuRead);
Integer osites = arg.Grid()->oSites();
auto ssum= sum_cpu(&arg_v[0],osites);
#endif
arg.Grid()->GlobalSum(ssum);
return ssum;
}
////////////////////////////////////////////////////////////////////////////////////////////////////
// Deterministic Reduction operations
////////////////////////////////////////////////////////////////////////////////////////////////////

8
Grid/lattice/Lattice_reduction_gpu.h
View File

@ -288,6 +288,14 @@ inline typename vobj::scalar_object sum_gpu(const vobj *lat, Integer osites)
return result;
}
template <class vobj>
inline typename vobj::scalar_object sum_gpu_large(const vobj *lat, Integer osites)
{
typedef typename vobj::scalar_object sobj;
sobj result;
result = sumD_gpu_large(lat,osites);
return result;
}
NAMESPACE_END(Grid);