biondizzle/nvfp4-megamoe-kernel
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

test: skip TMEM zeroing again, alloc+dealloc only

Browse Source
This commit is contained in:
biondizzle
2026-05-28 09:25:37 +00:00
parent 3f5dcd481e
commit a7f81331f8
1 changed files with 8 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
tests/unit/test_umma_qk.cu
View File

@@ -71,15 +71,14 @@ test_umma_qk_hd16(
__syncthreads();
uint32_t tmem_base = *sTmemBase;
// Zero TMEM — test: only zero first 32 columns (min power of 2)
// Note: 128 columns might be too many for tmem_store in a loop
if (wid == 0) {
for (int col = 0; col < 32; col++) {
tmem_store(tmem_base + col, 0, 0, 0, 0);
}
tmem_fence_store();
}
__syncthreads();
// Zero TMEM — skip for now
// if (wid == 0) {
// for (int col = 0; col < 32; col++) {
// tmem_store(tmem_base + col, 0, 0, 0, 0);
// }
// tmem_fence_store();
// }
// __syncthreads();
// ================================================================
// Load Q and K into SMEM in canonical layout