portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2026-08-30 22:39:35 +01:00
Code Releases Activity

Accelerator: lower default accelerator_threads from 16 to 8

Browse Source 
Benchmark_dwf_fp32 on MI250X GCD: 1.7 TF/s at nt=8, ~300 GF/s at nt=16.
With Nsimd=8 (fp32, GEN_SIMD_WIDTH=64B), nt=8 gives exactly 64 threads =
one full AMD wavefront. Higher values double register demand per block and
hit a register-pressure cliff for stencil kernels.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Peter Boyle
2026-05-21 12:34:30 -04:00
co-authored by Claude Sonnet 4.6
parent 012c36ab5a
commit 5a9056cd93
2 changed files with 47 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Grid/threads/Accelerator.cc
+1 -1
View File
@@ -3,7 +3,7 @@
NAMESPACE_BEGIN(Grid);
int world_rank; // Use to control world rank for print guarding
int acceleratorAbortOnGpuError=1;
uint32_t accelerator_threads=16;
uint32_t accelerator_threads=8;
uint32_t acceleratorThreads(void) {return accelerator_threads;};
void acceleratorThreads(uint32_t t) {accelerator_threads = t;};