1
0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-04 11:15:55 +01:00

reverse direction lexico mapping

This commit is contained in:
paboyle 2017-05-30 23:38:30 +01:00
parent 62cf9cf638
commit 58e8d0a10d

View File

@ -551,7 +551,10 @@ void Replicate(Lattice<vobj> &coarse,Lattice<vobj> & fine)
//Copy SIMD-vectorized lattice to array of scalar objects in lexicographic order
template<typename vobj, typename sobj>
typename std::enable_if<isSIMDvectorized<vobj>::value && !isSIMDvectorized<sobj>::value, void>::type unvectorizeToLexOrdArray(std::vector<sobj> &out, const Lattice<vobj> &in){
typename std::enable_if<isSIMDvectorized<vobj>::value && !isSIMDvectorized<sobj>::value, void>::type
unvectorizeToLexOrdArray(std::vector<sobj> &out, const Lattice<vobj> &in)
{
typedef typename vobj::vector_type vtype;
GridBase* in_grid = in._grid;
@ -590,6 +593,51 @@ typename std::enable_if<isSIMDvectorized<vobj>::value && !isSIMDvectorized<sobj>
extract1(in_vobj, out_ptrs, 0);
}
}
//Copy SIMD-vectorized lattice to array of scalar objects in lexicographic order
template<typename vobj, typename sobj>
typename std::enable_if<isSIMDvectorized<vobj>::value && !isSIMDvectorized<sobj>::value, void>::type
vectorizeFromLexOrdArray(std::vector<sobj> &in, const Lattice<vobj> &out)
{
typedef typename vobj::vector_type vtype;
GridBase* grid = out._grid;
assert(in.size()==grid->lSites());
int ndim = grid->Nd();
int nsimd = vtype::Nsimd();
std::vector<std::vector<int> > icoor(nsimd);
for(int lane=0; lane < nsimd; lane++){
icoor[lane].resize(ndim);
grid->iCoorFromIindex(icoor[lane],lane);
}
parallel_for(int oidx = 0; oidx < grid->oSites(); oidx++){ //loop over outer index
//Assemble vector of pointers to output elements
std::vector<sobj*> ptrs(nsimd);
std::vector<int> ocoor(ndim);
grid->oCoorFromOindex(ocoor, oidx);
std::vector<int> lcoor(grid->Nd());
for(int lane=0; lane < nsimd; lane++){
for(int mu=0;mu<ndim;mu++)
lcoor[mu] = ocoor[mu] + grid->_rdimensions[mu]*icoor[lane][mu];
int lex;
Lexicographic::IndexFromCoor(lcoor, lex, grid->_ldimensions);
ptrs[lane] = &in[lex];
}
//pack from those ptrs
vobj vecobj;
merge1(vecobj, ptrs, 0);
out._odata[oidx] = vecobj;
}
}
//Convert a Lattice from one precision to another
template<class VobjOut, class VobjIn>