add missing condition for sync (#2889)
This commit is contained in:
@@ -787,11 +787,13 @@ class DenseGemmKernel:
|
|||||||
# Async arrive AB buffer empty
|
# Async arrive AB buffer empty
|
||||||
consumer_handle.release()
|
consumer_handle.release()
|
||||||
|
|
||||||
# Peek (try_wait) AB buffer empty for k_tile = prefetch_k_tile_cnt + k_tile + 1
|
if k_tile_idx + 1 < k_tile_cnt - prefetch_k_tile_cnt:
|
||||||
peek_ab_empty_status = ab_producer.try_acquire()
|
# Peek (try_wait) AB buffer empty for k_tile = prefetch_k_tile_cnt + k_tile + 1
|
||||||
|
peek_ab_empty_status = ab_producer.try_acquire()
|
||||||
|
|
||||||
# Peek (try_wait) AB buffer full for k_tile = k_tile + 1
|
if k_tile_idx + 1 < k_tile_cnt and is_leader_cta:
|
||||||
peek_ab_full_status = ab_consumer.try_wait()
|
# Peek (try_wait) AB buffer full for k_tile = k_tile + 1
|
||||||
|
peek_ab_full_status = ab_consumer.try_wait()
|
||||||
|
|
||||||
# Async arrive accumulator buffer full
|
# Async arrive accumulator buffer full
|
||||||
if is_leader_cta:
|
if is_leader_cta:
|
||||||
|
|||||||
Reference in New Issue
Block a user