portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-09 23:45:36 +00:00
Code Releases Activity

Natural place for this routine is here

Browse Source
This commit is contained in:
Peter Boyle 2022-11-15 16:24:55 -05:00
parent e2e269e03b
commit 7d302a525d
1 changed files with 21 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
Grid/tensors/Tensor_SIMT.h
View File

@ -31,6 +31,27 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
NAMESPACE_BEGIN(Grid);
////////////////////////////////////////////////
// Inside a GPU thread
////////////////////////////////////////////////
template<class vobj>
accelerator_inline void exchangeSIMT(vobj &mp0,vobj &mp1,const vobj &vp0,const vobj &vp1,Integer type)
{
typedef decltype(coalescedRead(mp0)) sobj;
unsigned int Nsimd = vobj::Nsimd();
unsigned int mask = Nsimd >> (type + 1);
int lane = acceleratorSIMTlane(Nsimd);
int j0 = lane &(~mask); // inner coor zero
int j1 = lane |(mask) ; // inner coor one
const vobj *vpa = &vp0;
const vobj *vpb = &vp1;
const vobj *vp = (lane&mask) ? (vpb) : (vpa);
auto sa = coalescedRead(vp[0],j0);
auto sb = coalescedRead(vp[0],j1);
coalescedWrite(mp0,sa);
coalescedWrite(mp1,sb);
}
#ifndef GRID_SIMT
//////////////////////////////////////////