portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2026-05-27 20:44:16 +01:00
Code Releases Activity

Modified repack

Browse Source
This commit is contained in:
Peter Boyle
2026-05-19 08:53:13 -04:00
parent 796c6cae4e
commit d6b1388741
1 changed files with 3 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Grid/lattice/Lattice_reduction_gpu.h
+3 -5
View File
@@ -261,11 +261,9 @@ inline void sumD_gpu_reduce_words(const vobj *lat, Integer osites,
#ifdef GRID_REDUCTION_TIMING #ifdef GRID_REDUCTION_TIMING
RealD t_pack = -usecond(); RealD t_pack = -usecond();
#endif #endif
accelerator_for(ss, osites, 1, { constexpr int Nsimd = vobj::Nsimd();
Bundle b; accelerator_for2d(k, R, ss, osites, Nsimd, {
for (int k = 0; k < R; k++) coalescedWrite(buf_p[ss]._internal[k], coalescedRead(idat[ss * words + base + k]));
b._internal[k] = idat[ss * words + base + k];
buf_p[ss] = b;
}); });
#ifdef GRID_REDUCTION_TIMING #ifdef GRID_REDUCTION_TIMING
t_pack += usecond(); t_pack += usecond();