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 #388 from fjosw/feature/sumd-npr

Browse Source 
Feature/sumd npr
This commit is contained in:
Peter Boyle 2022-03-15 09:05:57 -04:00 committed by GitHub
commit b615fa0f35
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 24 additions and 9 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; 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 // Deterministic Reduction operations
//////////////////////////////////////////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////////////////////////////////////////

17
Grid/lattice/Lattice_reduction_gpu.h
View File

@ -232,10 +232,6 @@ inline typename vobj::scalar_objectD sumD_gpu_large(const vobj *lat, Integer osi
const int words = sizeof(vobj)/sizeof(vector); const int words = sizeof(vobj)/sizeof(vector);
Integer nsimd= vobj::Nsimd();
Integer size = osites*nsimd;
Integer numThreads, numBlocks;
Vector<vector> buffer(osites); Vector<vector> buffer(osites);
vector *dat = (vector *)lat; vector *dat = (vector *)lat;
vector *buf = &buffer[0]; vector *buf = &buffer[0];
@ -258,10 +254,7 @@ inline typename vobj::scalar_objectD sumD_gpu(const vobj *lat, Integer osites)
typedef typename vobj::scalar_typeD scalarD; typedef typename vobj::scalar_typeD scalarD;
typedef typename vobj::scalar_objectD sobj; typedef typename vobj::scalar_objectD sobj;
sobj ret; sobj ret;
scalarD *ret_p = (scalarD *)&ret;
const int words = sizeof(vobj)/sizeof(vector);
Integer nsimd= vobj::Nsimd(); Integer nsimd= vobj::Nsimd();
Integer size = osites*nsimd; Integer size = osites*nsimd;
Integer numThreads, numBlocks; Integer numThreads, numBlocks;
@ -275,7 +268,6 @@ inline typename vobj::scalar_objectD sumD_gpu(const vobj *lat, Integer osites)
return ret; return ret;
} }
///////////////////////////////////////////////////////////////////////////////////////////////////////// /////////////////////////////////////////////////////////////////////////////////////////////////////////
// Return as same precision as input performing reduction in double precision though // Return as same precision as input performing reduction in double precision though
///////////////////////////////////////////////////////////////////////////////////////////////////////// /////////////////////////////////////////////////////////////////////////////////////////////////////////
@ -288,6 +280,13 @@ inline typename vobj::scalar_object sum_gpu(const vobj *lat, Integer osites)
return result; 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); NAMESPACE_END(Grid);