Test: Python range() instead of cutlass.range() for TMA loop

This commit is contained in:
2026-05-22 20:32:44 +00:00
parent ae6197fc78
commit b8d4ac0ecb

View File

@@ -225,7 +225,7 @@ class FmhaV3StageCMulti:
qp.tail()
kvp.reset()
kv_coord = n_kv_tiles - n_kv_tiles # SSA runtime zero
for kt in cutlass.range(0, n_kv_tiles, 1, unroll=1):
for kt in range(n_kv_tiles):
kvh = kvp.acquire_and_advance()
cute.copy(tma_k, tBgK[(None, kv_coord)], tBsK[(None, kvh.index)], tma_bar_ptr=kvh.barrier)
cute.copy(tma_v, tVgV[(None, kv_coord)], tVsV[(None, kvh.index)], tma_bar_ptr=kvh.barrier)