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

Possible fallback plan for Fionn's compiler bbug in nvcc

Browse Source
This commit is contained in:
Peter Boyle 2021-03-11 22:20:53 +01:00
parent f786ff8d69
commit ce1fc1f48a
1 changed files with 19 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

36
Grid/lattice/Lattice_where.h
View File

@ -43,7 +43,7 @@ inline void whereWolf(Lattice<vobj> &ret,const Lattice<iobj> &predicate,Lattice<
conformable(iftrue,predicate); conformable(iftrue,predicate);
conformable(iftrue,ret); conformable(iftrue,ret);
GridBase *grid=iftrue._grid; GridBase *grid=iftrue.Grid();
typedef typename vobj::scalar_object scalar_object; typedef typename vobj::scalar_object scalar_object;
typedef typename vobj::scalar_type scalar_type; typedef typename vobj::scalar_type scalar_type;
@ -52,22 +52,24 @@ inline void whereWolf(Lattice<vobj> &ret,const Lattice<iobj> &predicate,Lattice<
const int Nsimd = grid->Nsimd(); const int Nsimd = grid->Nsimd();
std::vector<Integer> mask(Nsimd); Integer mask;
std::vector<scalar_object> truevals (Nsimd); scalar_object trueval;
std::vector<scalar_object> falsevals(Nsimd); scalar_object falseval;
parallel_for(int ss=0;ss<iftrue._grid->oSites(); ss++){ autoView(iftrue_v,iftrue,CpuRead);
autoView(iffalse_v,iffalse,CpuRead);
extract(iftrue._odata[ss] ,truevals); autoView(predicate_v,predicate,CpuRead);
extract(iffalse._odata[ss] ,falsevals); autoView(ret_v,ret,CpuWrite);
extract<vInteger,Integer>(TensorRemove(predicate._odata[ss]),mask); Integer NN= grid->oSites();
thread_for(ss,NN,{
for(int s=0;s<Nsimd;s++){ for(int l=0;l<Nsimd;l++){
if (mask[s]) falsevals[s]=truevals[s]; trueval =extractLane(l,iftrue_v[ss]);
falseval=extractLane(l,iffalse_v[ss]);
mask =extractLane(l,predicate_v[ss]);
if (mask) falseval=trueval;
insertLane(l,ret_v[ss],falseval);
} }
});
merge(ret._odata[ss],falsevals);
}
} }
template<class vobj,class iobj> template<class vobj,class iobj>
@ -76,9 +78,9 @@ inline Lattice<vobj> whereWolf(const Lattice<iobj> &predicate,Lattice<vobj> &ift
conformable(iftrue,iffalse); conformable(iftrue,iffalse);
conformable(iftrue,predicate); conformable(iftrue,predicate);
Lattice<vobj> ret(iftrue._grid); Lattice<vobj> ret(iftrue.Grid());
where(ret,predicate,iftrue,iffalse); whereWolf(ret,predicate,iftrue,iffalse);
return ret; return ret;
} }