diag: print bSG shapes for TMA store indexing
This commit is contained in:
@@ -140,6 +140,9 @@ class FmhaV3StageCMulti:
|
||||
cute.group_modes(sC, 0, 2),
|
||||
cute.group_modes(gC_epi, 0, 2),
|
||||
)
|
||||
# Print TMA partition shapes for debugging
|
||||
print(f'DEBUG bSG_sC shape: {bSG_sC.shape} rank: {cute.rank(bSG_sC)}', flush=True)
|
||||
print(f'DEBUG bSG_gC shape: {bSG_gC.shape} rank: {cute.rank(bSG_gC)}', flush=True)
|
||||
|
||||
n_kv_tiles = cute.size(gK, mode=[3])
|
||||
|
||||
|
||||
Reference in New Issue
Block a user