portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-18 15:57:05 +01:00
Code Releases Activity

Merge pull request #25 from paboyle/develop

Browse Source 
Sync
This commit is contained in:
Christoph Lehner
2021-09-15 10:02:43 +02:00
committed by GitHub
parent c50f27e68b 402d80e197
commit e2abbf9520
152 changed files with 2669 additions and 1177 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
Grid/lattice/Lattice_transfer.h
View File

@ -97,6 +97,20 @@ accelerator_inline void convertType(ComplexF & out, const std::complex<float> &
out = in;
}
template<typename T>
accelerator_inline EnableIf<isGridFundamental<T>> convertType(T & out, const T & in) {
out = in;
}
// This would allow for conversions between GridFundamental types, but is not strictly needed as yet
/*template<typename T1, typename T2>
accelerator_inline typename std::enable_if<isGridFundamental<T1>::value && isGridFundamental<T2>::value>::type
// Or to make this very broad, conversions between anything that's not a GridTensor could be allowed
//accelerator_inline typename std::enable_if<!isGridTensor<T1>::value && !isGridTensor<T2>::value>::type
convertType(T1 & out, const T2 & in) {
out = in;
}*/
#ifdef GRID_SIMT
accelerator_inline void convertType(vComplexF & out, const ComplexF & in) {
((ComplexF*)&out)[acceleratorSIMTlane(vComplexF::Nsimd())] = in;
@ -117,23 +131,18 @@ accelerator_inline void convertType(vComplexD2 & out, const vComplexF & in) {
Optimization::PrecisionChange::StoD(in.v,out._internal[0].v,out._internal[1].v);
}
template<typename T1,typename T2,int N>
accelerator_inline void convertType(iMatrix<T1,N> & out, const iMatrix<T2,N> & in);
template<typename T1,typename T2,int N>
accelerator_inline void convertType(iVector<T1,N> & out, const iVector<T2,N> & in);
template<typename T1,typename T2, typename std::enable_if<!isGridScalar<T1>::value, T1>::type* = nullptr>
accelerator_inline void convertType(T1 & out, const iScalar<T2> & in) {
convertType(out,in._internal);
template<typename T1,typename T2>
accelerator_inline void convertType(iScalar<T1> & out, const iScalar<T2> & in) {
convertType(out._internal,in._internal);
}
template<typename T1, typename std::enable_if<!isGridScalar<T1>::value, T1>::type* = nullptr>
accelerator_inline void convertType(T1 & out, const iScalar<T1> & in) {
template<typename T1,typename T2>
accelerator_inline NotEnableIf<isGridScalar<T1>> convertType(T1 & out, const iScalar<T2> & in) {
convertType(out,in._internal);
}
template<typename T1,typename T2>
accelerator_inline void convertType(iScalar<T1> & out, const T2 & in) {
accelerator_inline NotEnableIf<isGridScalar<T2>> convertType(iScalar<T1> & out, const T2 & in) {
convertType(out._internal,in);
}
@ -150,11 +159,6 @@ accelerator_inline void convertType(iVector<T1,N> & out, const iVector<T2,N> & i
convertType(out._internal[i],in._internal[i]);
}
template<typename T, typename std::enable_if<isGridFundamental<T>::value, T>::type* = nullptr>
accelerator_inline void convertType(T & out, const T & in) {
out = in;
}
template<typename T1,typename T2>
accelerator_inline void convertType(Lattice<T1> & out, const Lattice<T2> & in) {
autoView( out_v , out,AcceleratorWrite);

37
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,ret);
GridBase *grid=iftrue._grid;
GridBase *grid=iftrue.Grid();
typedef typename vobj::scalar_object scalar_object;
typedef typename vobj::scalar_type scalar_type;
@ -52,22 +52,23 @@ inline void whereWolf(Lattice<vobj> &ret,const Lattice<iobj> &predicate,Lattice<
const int Nsimd = grid->Nsimd();
std::vector<Integer> mask(Nsimd);
std::vector<scalar_object> truevals (Nsimd);
std::vector<scalar_object> falsevals(Nsimd);
parallel_for(int ss=0;ss<iftrue._grid->oSites(); ss++){
extract(iftrue._odata[ss] ,truevals);
extract(iffalse._odata[ss] ,falsevals);
extract<vInteger,Integer>(TensorRemove(predicate._odata[ss]),mask);
for(int s=0;s<Nsimd;s++){
if (mask[s]) falsevals[s]=truevals[s];
autoView(iftrue_v,iftrue,CpuRead);
autoView(iffalse_v,iffalse,CpuRead);
autoView(predicate_v,predicate,CpuRead);
autoView(ret_v,ret,CpuWrite);
Integer NN= grid->oSites();
thread_for(ss,NN,{
Integer mask;
scalar_object trueval;
scalar_object falseval;
for(int l=0;l<Nsimd;l++){
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>
@ -76,9 +77,9 @@ inline Lattice<vobj> whereWolf(const Lattice<iobj> &predicate,Lattice<vobj> &ift
conformable(iftrue,iffalse);
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;
}