portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-09 21:50:45 +01:00
Code Releases Activity

Commet

Browse Source
This commit is contained in:
Peter Boyle 2023-03-28 08:34:24 -07:00
parent 900e01f49b
commit 71c6960eea
1 changed files with 3 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
Grid/lattice/Lattice_reduction_gpu.h
View File

@ -211,12 +211,9 @@ inline typename vobj::scalar_objectD sumD_gpu_small(const vobj *lat, Integer osi
assert(ok); assert(ok);
Integer smemSize = numThreads * sizeof(sobj); Integer smemSize = numThreads * sizeof(sobj);
// UVM seems to be buggy under later CUDA drivers // Move out of UVM
// This fails on A100 and driver 5.30.02 / CUDA 12.1 // Turns out I had messed up the synchronise after move to compute stream
// Fails with multiple NVCC versions back to 11.4, // as running this on the default stream fools the synchronise
// which worked with earlier drivers.
// Not sure which driver had first fail and this bears checking
// Is awkward as must install multiple driver versions
#undef UVM_BLOCK_BUFFER #undef UVM_BLOCK_BUFFER
#ifndef UVM_BLOCK_BUFFER #ifndef UVM_BLOCK_BUFFER
commVector<sobj> buffer(numBlocks); commVector<sobj> buffer(numBlocks);