portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-17 07:17:06 +01:00
Code Releases Activity

GPU reductions first cut; use thrust, non-reproducible. Inclusive scan can fix this if desired.

Browse Source 
Local reduction to LatticeComplex and then further reduction.
This commit is contained in:
Peter Boyle
2019-01-01 13:53:37 +00:00
parent 3eae9a9e3f
commit 715babeac8
4 changed files with 82 additions and 171 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
Grid/lattice/Lattice_transfer.h
View File

@ -640,7 +640,7 @@ unvectorizeToRevLexOrdArray(std::vector<sobj> &out, const Lattice<vobj> &in)
in_grid->iCoorFromIindex(in_icoor[lane], lane);
}
parallel_for(int in_oidx = 0; in_oidx < in_grid->oSites(); in_oidx++){ //loop over outer index
thread_loop( (int in_oidx = 0; in_oidx < in_grid->oSites(); in_oidx++),{ //loop over outer index
//Assemble vector of pointers to output elements
std::vector<sobj*> out_ptrs(in_nsimd);
@ -661,7 +661,7 @@ unvectorizeToRevLexOrdArray(std::vector<sobj> &out, const Lattice<vobj> &in)
//Unpack into those ptrs
const vobj & in_vobj = in._odata[in_oidx];
extract1(in_vobj, out_ptrs, 0);
}
});
}
//Copy SIMD-vectorized lattice to array of scalar objects in lexicographic order
@ -733,7 +733,7 @@ vectorizeFromRevLexOrdArray( std::vector<sobj> &in, Lattice<vobj> &out)
grid->iCoorFromIindex(icoor[lane],lane);
}
parallel_for(uint64_t oidx = 0; oidx < grid->oSites(); oidx++){ //loop over outer index
thread_loop( (uint64_t oidx = 0; oidx < grid->oSites(); oidx++),{ //loop over outer index
//Assemble vector of pointers to output elements
std::vector<sobj*> ptrs(nsimd);
@ -757,7 +757,7 @@ vectorizeFromRevLexOrdArray( std::vector<sobj> &in, Lattice<vobj> &out)
vobj vecobj;
merge1(vecobj, ptrs, 0);
out._odata[oidx] = vecobj;
}
});
}
//Convert a Lattice from one precision to another