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

Zero changes, acceleartor on kernels and some thread loop changes

Browse Source
This commit is contained in:
paboyle
2018-01-27 23:47:38 +00:00
parent 45df59720e
commit 2d0bcc2606
40 changed files with 174 additions and 175 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
lib/qcd/spin/TwoSpinor.h
View File

@ -159,15 +159,15 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> strong_inline void spProj
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
rfspin(0)=fspin(0);
rfspin(1)=fspin(1);
rfspin(2)=zero;
rfspin(3)=zero;
rfspin(2)=Zero();
rfspin(3)=Zero();
}
// template<class vtype> strong_inline void fspProj5m (iVector<vtype,Ns> &rfspin,const iVector<vtype,Ns> &fspin)
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> strong_inline void spProj5m (iVector<vtype,Ns> &rfspin,const iVector<vtype,Ns> &fspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
rfspin(0)=zero;
rfspin(1)=zero;
rfspin(0)=Zero();
rfspin(1)=Zero();
rfspin(2)=fspin(2);
rfspin(3)=fspin(3);
}
@ -339,14 +339,14 @@ template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> strong_inline void spReco
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
fspin(0)=hspin(0)+hspin(0); // add is lower latency than mul
fspin(1)=hspin(1)+hspin(1); // probably no measurable diffence though
fspin(2)=zero;
fspin(3)=zero;
fspin(2)=Zero();
fspin(3)=Zero();
}
template<class vtype,IfSpinor<iVector<vtype,Ns> > = 0> strong_inline void spRecon5m (iVector<vtype,Ns> &fspin,const iVector<vtype,Nhs> &hspin)
{
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;
fspin(0)=zero;
fspin(1)=zero;
fspin(0)=Zero();
fspin(1)=Zero();
fspin(2)=hspin(0)+hspin(0);
fspin(3)=hspin(1)+hspin(1);
}