portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-17 23:37:06 +01:00
Code Releases Activity

sliceSumReduction_cub_large now also faster than CPU on Frontier

Browse Source
This commit is contained in:
Dennis Bollweg
2024-02-16 13:55:21 -05:00
parent e0d5e3c6c7
commit 15878f7613
2 changed files with 4 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
Grid/lattice/Lattice_slicesum_core.h
View File

@ -119,7 +119,7 @@ template<class vobj> inline void sliceSumReduction_cub_large(const vobj *Data, V
typedef typename vobj::vector_type vector;
const int words = sizeof(vobj)/sizeof(vector);
const int osites = rd*e1*e2;
Vector<vector>buffer(osites);
commVector<vector>buffer(osites);
vector *dat = (vector *)Data;
vector *buf = &buffer[0];
Vector<vector> lvSum_small(rd);