portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2026-08-23 19:09:35 +01:00
Code Releases Activity

Merge pull request #440 from giltirn/feature/paddedcellgauge

Browse Source 
Feature/paddedcellgauge
This commit is contained in:
Peter Boyle
2023-10-02 10:00:42 -04:00
committed by GitHub
parent b77bccfac2 1dfaa08afb
commit 018e6da872
13 changed files with 1612 additions and 111 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Grid/tensors/Tensor_SIMT.h
+19 -1
View File
@@ -73,6 +73,16 @@ vobj coalescedReadPermute(const vobj & __restrict__ vec,int ptype,int doperm,int
return vec;
}
}
//'perm_mask' acts as a bitmask
template<class vobj> accelerator_inline
vobj coalescedReadGeneralPermute(const vobj & __restrict__ vec,int perm_mask,int nd,int lane=0)
{
auto obj = vec, tmp = vec;
for (int d=0;d<nd;d++)
if (perm_mask & (0x1 << d)) { permute(obj,tmp,d); tmp=obj;}
return obj;
}
template<class vobj> accelerator_inline
void coalescedWrite(vobj & __restrict__ vec,const vobj & __restrict__ extracted,int lane=0)
{
@@ -83,7 +93,7 @@ void coalescedWriteNonTemporal(vobj & __restrict__ vec,const vobj & __restrict__
{
vstream(vec, extracted);
}
#else
#else //==GRID_SIMT
//#ifndef GRID_SYCL
@@ -166,6 +176,14 @@ typename vobj::scalar_object coalescedReadPermute(const vobj & __restrict__ vec,
return extractLane(plane,vec);
}
template<class vobj> accelerator_inline
typename vobj::scalar_object coalescedReadGeneralPermute(const vobj & __restrict__ vec,int perm_mask,int nd,int lane=acceleratorSIMTlane(vobj::Nsimd()))
{
int plane = lane;
for (int d=0;d<nd;d++)
plane = (perm_mask & (0x1 << d)) ? plane ^ (vobj::Nsimd() >> (d + 1)) : plane;
return extractLane(plane,vec);
}
template<class vobj> accelerator_inline
void coalescedWrite(vobj & __restrict__ vec,const typename vobj::scalar_object & __restrict__ extracted,int lane=acceleratorSIMTlane(vobj::Nsimd()))
{
insertLane(lane,vec,extracted);