Repository navigation
fix: 修复了 Ascend NPU 上下文导致的崩溃问题 - #3512
Conversation
dosubot
Bot
added
the
size:S
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request updates several Ascend-based offline model implementations, including Paraformer, SenseVoice, Whisper, and Zipformer CTC, to explicitly set the current ACL context using aclrtSetCurrentContext before processing. These additions are placed within existing mutex-protected blocks and include error handling via the SHERPA_ONNX_ASCEND_CHECK macro. I have no feedback to provide as there were no review comments to evaluate.
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (4)
📝 WalkthroughWalkthroughAdded explicit Ascend context-setting calls to four offline model implementations. Each Changes
Estimated code review effort🎯 2 (Simple) | ⏱️ ~10 minutes Possibly related PRs
Suggested labels
Poem
🚥 Pre-merge checks | ✅ 2 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
csukuangfj
left a comment
csukuangfj
left a comment
There was a problem hiding this comment.
Thank you for your contribution!
2 participants
鲲鹏aarch64 CPU，Ascend 910A NPU卡，CANN 8.2.RC1华为官方容器。使用sherpa-onnx-offline工作正常。sherpa-onnx-offline-websocket-server启动服务正常，使用python客户程序连接时服务崩溃。
root@localhost:/opt/sherpa-onnx/build# ./bin/sherpa-onnx-offline-websocket-server --port=6006 --num-work-threads=5 --provider=ascend --paraformer="/data/paraformer/sherpa-onnx-ascend-910B-cann-8.2-paraformer-zh-2025-10-07/encoder.om,/data/paraformer/sherpa-onnx-ascend-910B-cann-8.2-paraformer-zh-2025-10-07/predictor.om,/data/paraformer/sherpa-onnx-ascend-910B-cann-8.2-paraformer-zh-2025-10-07/decoder.om" --tokens=/data/paraformer/sherpa-onnx-ascend-910B-cann-8.2-paraformer-zh-2025-10-07/tokens.txt --log-file=./log.txt --max-batch-size=5
onnxruntime cpuid_info warning: Unknown CPU vendor. cpuinfo_vendor value: 15
/opt/sherpa-onnx/sherpa-onnx/csrc/parse-options.cc:Read:373 ./bin/sherpa-onnx-offline-websocket-server --port=6006 --num-work-threads=5 --provider=ascend --paraformer=/data/paraformer/sherpa-onnx-ascend-910B-cann-8.2-paraformer-zh-2025-10-07/encoder.om,/data/paraformer/sherpa-onnx-ascend-910B-cann-8.2-paraformer-zh-2025-10-07/predictor.om,/data/paraformer/sherpa-onnx-ascend-910B-cann-8.2-paraformer-zh-2025-10-07/decoder.om --tokens=/data/paraformer/sherpa-onnx-ascend-910B-cann-8.2-paraformer-zh-2025-10-07/tokens.txt --log-file=./log.txt --max-batch-size=5
/opt/sherpa-onnx/sherpa-onnx/csrc/offline-websocket-server.cc:main:93 Started!
/opt/sherpa-onnx/sherpa-onnx/csrc/offline-websocket-server.cc:main:94 Listening on: 6006
/opt/sherpa-onnx/sherpa-onnx/csrc/offline-websocket-server.cc:main:95 Number of work threads: 5
[2026-04-13 11:32:21] [connect] WebSocket Connection 192.168.1.41:51694 v13 "Python/3.9 websockets/15.0.1" / 101
/opt/sherpa-onnx/sherpa-onnx/csrc/offline-websocket-server-impl.cc:OnOpen:170 Number of active connections: 1
/opt/sherpa-onnx/sherpa-onnx/csrc/offline-websocket-server-impl.cc:Decode:67 size: 1
/opt/sherpa-onnx/sherpa-onnx/csrc/ascend/offline-paraformer-model-ascend.cc:RunEncoder:149 Return code is: 107002
/opt/sherpa-onnx/sherpa-onnx/csrc/ascend/offline-paraformer-model-ascend.cc:RunEncoder:149 Error message: EE1001: [PID: 118] 2026-04-13-11:32:21.618.180 The argument is invalid.Reason: rtMemcpy execute failed, reason=[context pointer null]
Solution: 1.Check the input parameter range of the function. 2.Check the function invocation relationship.
TraceBack (most recent call last):
ctx is NULL![FUNC:MemCopySync][FILE:api_impl.cc][LINE:2555]
The argument is invalid.Reason: rtMemcpy execute failed, reason=[context pointer null]
synchronized memcpy failed, kind = 1, runtime result = 107002[FUNC:ReportCallError][FILE:log_inner.cpp][LINE:161]
ctx is NULL![FUNC:GetDevErrMsg][FILE:api_impl.cc][LINE:6147]
The argument is invalid.Reason: rtGetDevMsg execute failed, reason=[context pointer null]
/opt/sherpa-onnx/sherpa-onnx/csrc/ascend/offline-paraformer-model-ascend.cc:RunEncoder:149 Failed to call aclrtMemcpy
错误原因：Ascend NPU 的上下文（aclrtContext）与线程绑定。当多工作线程并发调用模型时，aclrtMemcpy 和 aclmdlExecute调用时当前线程没有有效的上下文。
解决方案：在sherpa-onnx/sherpa-onnx/csrc/ascend/offline-*-model-ascend.cc 4个文件的 std::lock_guardstd::mutex lock(mutex_); 行之后加上：
aclError ret_set_ctx = aclrtSetCurrentContext(*context_);
SHERPA_ONNX_ASCEND_CHECK(ret_set_ctx, "Failed to call aclrtSetCurrentContext");
经测试在昇腾910A NPU 上正常工作。
Summary by CodeRabbit