test: zero 32 TMEM columns
This commit is contained in:
@@ -78,9 +78,9 @@ test_umma_qk_hd16(
|
||||
}
|
||||
__syncthreads();
|
||||
|
||||
// Zero TMEM
|
||||
// Zero TMEM — test with 32 columns first
|
||||
if (wid == 0) {
|
||||
for (int col = 0; col < 128; col++) {
|
||||
for (int col = 0; col < 32; col++) {
|
||||
tmem_store(tmem_base + col, 0, 0, 0, 0);
|
||||
}
|
||||
tmem_fence_store();
|
||||
|
||||
Reference in New Issue
Block a user