portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-03 18:55:56 +01:00
Code Releases Activity

acceleration compiles and doesn't break scalar mode

Browse Source
This commit is contained in:
david clarke 2024-02-06 18:40:13 -07:00
parent a38fb0e04a
commit 2da09ae99b
1 changed files with 7 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
Grid/qcd/smearing/HISQSmearing.h
View File

@ -153,18 +153,18 @@ public:
// This loop handles 3-, 5-, and 7-link constructs, minus Lepage and Naik.
for(int mu=0;mu<Nd;mu++) {
// Create the accessors
autoView(U_v , Ughost , AcceleratorRead);
autoView(U_fat_v , Ughost_fat , CpuWrite);
autoView(U_3link_v , Ughost_3link , CpuWrite);
autoView(U_5linkA_v, Ughost_5linkA, CpuWrite);
autoView(U_5linkB_v, Ughost_5linkB, CpuWrite);
// TODO: This approach is slightly memory inefficient. It uses 25% extra memory
Ughost_3link =Zero();
Ughost_5linkA=Zero();
Ughost_5linkB=Zero();
// Create the accessors
autoView(U_v , Ughost , AcceleratorRead);
autoView(U_fat_v , Ughost_fat , AcceleratorWrite);
autoView(U_3link_v , Ughost_3link , AcceleratorWrite);
autoView(U_5linkA_v, Ughost_5linkA, AcceleratorWrite);
autoView(U_5linkB_v, Ughost_5linkB, AcceleratorWrite);
// We infer some types that will be needed in the calculation.
typedef decltype(gStencil.GetEntry(0,0)) stencilElement;
typedef decltype(coalescedReadGeneralPermute(U_v[0](0),gStencil.GetEntry(0,0)->_permute,Nd)) U3matrix;