Merge blocked
Thanks for your pull-request.
The full list of commands accepted by me can be found at here.
You can get sig-info at here.
You can self-configure the PR merge rules for this repository. For more details, please refer to Here.
For more, you also can visit HICANN.
PR Approval Progress
⚠️ This PR does not yet meet the following requirements:lgtm (requires ≥ 2 person(s) per module)、approve (requires ≥ 1 person(s) per module)
Module Approval Details
| module | lgtm status | approve status |
|---|---|---|
| repo-cann/ops-tilelang | ❌ ChaoyangJi (1/2)(You can also ask: 俞郑中, 蒋礼锐, kevin365, 马兵, 刘丹) | ❌ (0/1)(You can also ask: 黄俊健, 马兵, 蒋礼锐, 刘丹, jiawei_zhuang) |
💡 Tip:
- Committer can comment
/approveor/lgtm- Commenting
/approveimplies both code review (lgtm) and intent to merge (approve)
CLA Signature Pass
hanxu88, thanks for your pull request. All authors of the commits have signed the CLA. 👍


/compile


| 🚀 CI 流水线已启动 |
|---|
| 📋 执行详情: 点击查看流水线 |


流水线分析报告 — ops-tilelang run#42
流水线: https://gitcode.com/cann/ops-tilelang/actions/runs/f890a771051145549c0c6b95a18a38ef | PR#60 | 2026-11-03 15:00:59 ~ 2026-11-03 18:28:41 | 失败 Job 1 个
定界结论: 环境问题 / Runner机器异常 — ST_test 排队超时(CP.COMP1105)
- 失败类型: 环境问题 / Runner机器异常
- 根因:
ST_test自 15:03 进入排队,至 18:28 仍未获调度,平台延迟调度重试耗尽,报任务排队超时(Queued timeout);该 Job 未真正开始执行。 - 处置建议: 直接重新触发流水线即可(执行机资源/调度瞬时不足,可重试恢复)。
- 排除项: 非代码问题——其余 11 个 Job(Test_get-pr / Test_pre_pipeline / Antipoison / CodeCheck* / SCA / Test_finish 等)全部 COMPLETED;
ST_test的 Checkout、download、run_st_test、后处理步骤日志均为空,说明失败发生在调度阶段而非用例执行或编译阶段。
失败根因分析
根因 #1: 环境问题 / Runner机器异常 — ST_test 排队超时(CP.COMP1105)
- 判据:
queued state timeout | 任务排队超时→ 平台调度问题,可重试 - 定位步骤:
ST_test/0_初始化步骤.log - 影响 Job: ST_test
- 根因证据(日志原文):
[2026-11-03 18:28:20.346+08:00] [ERROR] [server] - Exceed maximum retry after delay schedule.
[2026-11-03 18:28:20.497+08:00] [ERROR] [server] - [CP.COMP1105][Description] 任务排队超时
[Description_en] Queued timeout.(addition_en) Exceed maximum retry after delay schedule.
[RootCause] 可能是执行机配置或者网络问题。[Solution] 请检查执行机配置或者网络。
- 补充证据:
ST_test的1_Checkout.log/2_download.log/3_run_st_test.log/4_后处理步骤.log均为空文件,印证任务自 15:03 起一直处于排队态、始终未落到执行机上运行。失败与 PR 代码变更无关。
结构化定界摘要
[{"repo": "ops-tilelang", "run_number": 42, "Failed_Type": "环境问题", "Failed_Sub_Type": "Runner机器异常", "Failed_Reason": "ST_test 任务排队超时(CP.COMP1105), 延迟调度重试耗尽, 任务未进入执行阶段, 与 PR 代码变更无关", "evidence": "[CP.COMP1105][Description] 任务排队超时 Queued timeout. Exceed maximum retry after delay schedule.", "suggestion": "直接重新触发流水线重试", "jobs": "ST_test"}]
由门禁失败定界分析自动生成 | repo: ops-tilelang | run: f890a771051145549c0c6b95a18a38ef | PR#60


/compile


| 🚀 CI 流水线已启动 |
|---|
| 📋 执行详情: 点击查看流水线 |


/compile


| 🚀 CI 流水线已启动 |
|---|
| 📋 执行详情: 点击查看流水线 |


流水线分析报告 — ops-tilelang run#52
流水线: https://gitcode.com/CANN/ops-tilelang/actions/runs/be6dd7c68a584c829b5b6eb59380455f | PR#60 | 2026-11-04 15:57:13 ~ 2026-11-04 21:36:04 | 失败 Job 1 个
定界结论: 环境问题 / 设备/NPU异常 — NPU 设备启动失败(devId=2/3)
- 失败类型: 环境问题(非 PR 代码问题)
- 根因:
ST_test运行时torch_npu在LazySetDevice阶段报 error code 507033,底层Failed to start the device:TSDaemon process initialization is abnormal (E39005),TsdOpen failed. devId=2/3,rtSetDevice execution failed, reason=device retain error。分配到的 2、3 号卡均无法打开,NPU 用例全部在设备初始化阶段失败。 - 处置建议: ① 直接重试流水线(设备/驱动侧瞬时异常,重试大概率恢复);② 重试仍复现 → 转运维检查该 runner 的 NPU 卡 2/3(设备状态/驱动/TSD 进程),按日志提示获取 host/device 日志并联系技术支持。
- 排除项: 排除 PR 代码问题——失败点在各用例的
generate_big_fuse_test_data首个torch.randn(..., device=device)设备初始化,而非断言/精度/算子执行;本次 run 其余 Job(编译、CodeCheck、SCA、PreSmoke 等)均通过;同一轮 8 个不依赖 NPU 的用例通过。
失败根因分析
根因 #1: 环境问题 / 设备/NPU异常 — NPU 设备启动失败(507033, devId=2/3)
- 判据:
RuntimeError: LazySetDevice ... error code is 507033+Failed to start the device.+TSDaemon 初始化异常(E39005)+TsdOpen failed. devId=2/3 - 定位步骤:
ST_test / 3_run_st_test.log - 影响 Job: ST_test
- 影响范围: 跑在 NPU 2、3 上的 38 个用例(target=auto 全失败;target=pto 30 failed / 8 passed),失败均发生在首次设备访问处
- 根因证据(日志原文):
E RuntimeError: LazySetDevice:../torch_npu/csrc/core/npu/NPUFunctions.cpp:342 NPU function error: err, error code is 507033
E [Error]: Failed to start the device.
E [PID: 18835] ... Inner_Error_TSDaemon_Process_Abnormal(E39005): The TSDaemon process initialization is abnormal.
E TsdOpen failed. devId=3, tdt error=1.[FUNC:PrintfTsdError][FILE:runtime.cc][LINE:2194]
E rtSetDevice execution failed, reason=device retain error[FUNC:FuncErrorReason][FILE:error_message_manage.cc][LINE:76]
E Context validation failed. Reason: No valid context is bound to the current thread.
第二轮 target=pto 同样失败,TsdOpen failed. devId=2,说明是设备侧问题而非单个用例偶发:
============ 30 failed, 8 passed, 104 warnings in 109.77s (0:01:49) ============
>>> UT FAILED for target: pto
结构化定界摘要
[{"repo": "ops-tilelang", "run_number": 52, "Failed_Type": "环境问题", "Failed_Sub_Type": "设备/NPU异常", "Failed_Reason": "ST_test 全部 NPU 用例在 torch_npu LazySetDevice 阶段失败: error code 507033, Failed to start the device, TSDaemon 初始化异常(E39005), TsdOpen failed devId=2/3, rtSetDevice device retain error", "evidence": "RuntimeError: LazySetDevice:../torch_npu/csrc/core/npu/NPUFunctions.cpp:342 NPU function error: err, error code is 507033 / [Error]: Failed to start the device. / Inner_Error_TSDaemon_Process_Abnormal(E39005) / TsdOpen failed. devId=3, tdt error=1", "suggestion": "重试流水线; 若复现转运维检查 runner NPU 2/3 设备状态/驱动/TSD 进程, 获取 host/device 日志联系技术支持; 非 PR 代码问题", "jobs": "ST_test"}]
由门禁失败定界分析自动生成 | repo: ops-tilelang | run: be6dd7c68a584c829b5b6eb59380455f | PR#60


/compile


| 🚀 CI 流水线已启动 |
|---|
| 📋 执行详情: 点击查看流水线 |


/lgtm


/compile


| 🚀 CI 流水线已启动 |
|---|
| 📋 执行详情: 点击查看流水线 |


流水线分析报告 — ops-tilelang run#67
流水线: https://gitcode.com/CANN/ops-tilelang/actions/runs/96cb07daf5b641baa058b04b307ceb66 | PR#60 | 2026-11-06 14:13 ~ 20:16 | 失败 Job 1 个
定界结论: 环境问题 / Runner机器异常(任务排队超时,可重试)
- 失败类型: 环境问题(非代码问题,PR 代码无需修改)
- 根因:
ST_test于 14:16 进入 Queued 状态后长达 6 小时未分配到执行机,20:16 平台判定排队超时(CP.COMP1105),Job 未启动任何步骤即失败 - 影响范围: 仅
ST_test1 个 Job;同 run 其余 14 个 Job(CodeCheck、SCA、Antipoison、CodeCheck_precommit、Test_PreBuild 等)全部 COMPLETED - 处置建议: ① 直接重新触发流水线(平台调度/执行机资源问题,大概率恢复);② 重试仍持续排队超时 → 联系平台侧核查执行机配置与资源池可用性(ST 任务所需标签机器是否被占满/离线)
- 排除项: 非代码问题、非编译/测试用例失败、非 PR 冲突(Checkout 与 PreBuild 均已完成)
失败根因分析
根因 #1: 环境问题 / Runner机器异常 — ST_test 排队超时(CP.COMP1105)
- 判据:
(?i)queued state timeout|任务排队超时→ 任务排队超时(平台调度,可重试) - 定位步骤:
0_初始化步骤.log(1_Checkout/2_download/3_run_st_test/4_后处理步骤均为空,Job 从未开始执行) - 影响 Job: ST_test
- 时间线: 2026-11-06 14:16:19 进入排队 → 20:16:19 触发排队超时,全程未进入运行态
- 根因证据(日志原文):
[2026-11-06 20:16:19.994+08:00] [ERROR] [server] - Queued state timeout, job_run_id: c879a947d9bb42cfaaf4bf1d0dcad1f7_1568267113326063653, state: Queued, timeout_seconds: 300
[2026-11-06 20:16:20.142+08:00] [ERROR] [server] - [CP.COMP1105][Description] 任务排队超时[Description_en] Queued timeout.(addition_en) Queued state timeout, job_run_id: c879a947d9bb42cfaaf4bf1d0dcad1f7_1568267113326063653, state: Queued, timeout_seconds: 300[RootCause] 可能是执行机配置或者网络问题。[Solution] 请检查执行机配置或者网络。
结构化定界摘要
[{"repo": "ops-tilelang", "run_number": 67, "Failed_Type": "环境问题", "Failed_Sub_Type": "Runner机器异常", "Failed_Reason": "ST_test 任务排队 6 小时未分配执行机,平台判定排队超时 CP.COMP1105,Job 未启动任何步骤", "evidence": "[CP.COMP1105][Description] 任务排队超时[Description_en] Queued timeout. state: Queued, timeout_seconds: 300", "suggestion": "重新触发流水线;持续超时则联系平台核查执行机配置与资源池可用性", "jobs": "ST_test"}]
由门禁失败定界分析自动生成 | repo: ops-tilelang | run: 96cb07daf5b641baa058b04b307ceb66 | PR#60


当前PR是否有AI参与:
[x] 否
[ ] 是
__1. AI Agent 平台:
__2. AI 模型:
__3. Prompt上下文 :
PR功能描述 / 为什么需要这个合入**:
去除了
mhc_sinkhorn_backward_kernel算子硬编码的target="pto"该PR关联的issue
(格式为fixes #<issue号>, 或者resolves #<issue号>): fixes #
希望检视人员了解:
改动类型 / Change Type
测试信息 / Testing
检查清单 / Checklist