92 lines
2.9 KiB
Python
92 lines
2.9 KiB
Python
#
|
|
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
#
|
|
|
|
from __future__ import annotations
|
|
|
|
import threading
|
|
|
|
import torch
|
|
|
|
_lock = threading.RLock()
|
|
_attention_compute_start_gate: AttentionComputeStartGate | None = None
|
|
|
|
|
|
class AttentionComputeStartGate:
|
|
"""Gate that opens when the compute stream reaches attention.
|
|
|
|
The attention worker records an NPU event immediately before submitting the
|
|
attention op. MemCache worker threads wait for that event to complete before
|
|
submitting H2D/L2G work, so transfer starts when the compute stream is
|
|
actually at the attention boundary rather than merely after the Python call
|
|
site was reached.
|
|
"""
|
|
|
|
def __init__(self) -> None:
|
|
self._condition = threading.Condition()
|
|
self._event: torch.npu.Event | None = None
|
|
|
|
def record(
|
|
self,
|
|
stream: torch.npu.Stream | None = None,
|
|
) -> None:
|
|
stream = stream or torch.npu.current_stream()
|
|
event = torch.npu.Event()
|
|
event.record(stream)
|
|
with self._condition:
|
|
if self._event is None:
|
|
self._event = event
|
|
self._condition.notify_all()
|
|
|
|
def wait(self, timeout: float = 10.0) -> bool:
|
|
with self._condition:
|
|
while self._event is None:
|
|
if not self._condition.wait(timeout=timeout):
|
|
return False
|
|
event = self._event
|
|
|
|
event.synchronize()
|
|
return True
|
|
|
|
|
|
def reset_attention_compute_start_gate() -> AttentionComputeStartGate:
|
|
"""Create a new per-layer gate for MemCache work.
|
|
|
|
Layerwise prefetch tasks keep a reference to the gate that was current when
|
|
they were submitted. The attention path opens that same gate when attention
|
|
compute is about to be launched.
|
|
"""
|
|
global _attention_compute_start_gate
|
|
gate = AttentionComputeStartGate()
|
|
with _lock:
|
|
_attention_compute_start_gate = gate
|
|
return gate
|
|
|
|
|
|
def get_attention_compute_start_gate() -> AttentionComputeStartGate:
|
|
with _lock:
|
|
gate = _attention_compute_start_gate
|
|
if gate is None:
|
|
gate = reset_attention_compute_start_gate()
|
|
return gate
|
|
|
|
|
|
def record_attention_compute_start() -> None:
|
|
"""Record the compute-stream boundary immediately before attention."""
|
|
with _lock:
|
|
gate = _attention_compute_start_gate
|
|
if gate is not None:
|
|
gate.record()
|