Pull Request has been successfully merged by CANN-robot.
(Thank you for 佐助大王's contribution.)变更摘要
本 PR 主要针对算子报错信息进行规范化改造,将原先通用的 OP_LOGE 日志替换为带结构化错误码的宏(如 OP_LOGE_FOR_INVALID_DTYPE、OP_LOGE_FOR_INVALID_SHAPE_WITH_REASON、OP_LOGE_FOR_INVALID_TENSORNUM 等),使各类校验失败时能够携带输入/输出名称、实际取值以及失败原因,提升错误定位能力。改动覆盖 cross_v2、ctc_loss_v3、ctc_loss_v3_grad、mse_loss_grad_v2 以及 fused_adam、fused_adamw、fused_sgd 等算子的 tiling 与 infershape 逻辑。
主要改动
- ctc_loss_v3 系列校验报错结构化: 在
ctc_loss_v3_tiling.cpp与ctc_loss_v3_grad_tiling.cpp中,将 dtype、shape dim、batch/time/symbol 一致性等校验由OP_LOGE改为OP_LOGE_FOR_INVALID_DTYPE、OP_LOGE_FOR_INVALID_DTYPES_WITH_REASON、OP_LOGE_FOR_INVALID_SHAPEDIM、OP_LOGE_FOR_INVALID_VALUES_WITH_REASON等宏,并传入具体张量名、维度数与取值说明。 - cross_v2 tiling 校验报错结构化:
cross_v2_tiling.cpp中空张量、dim越界、batchSize/stepSize非法、inputBytes为 0 等场景改用OP_LOGE_FOR_INVALID_SHAPESIZE_WITH_REASON、OP_LOGE_FOR_INVALID_VALUE_WITH_REASON、OP_LOGE_FOR_INVALID_SHAPE_WITH_REASON、OP_LOGE_FOR_INVALID_DTYPE_WITH_REASON,并补充实际取值与约束原因。 - mse_loss_grad_v2 infershape 报错补充:
mse_loss_grad_v2_infershape.cpp中原先仅return ge::GRAPH_FAILED的分支新增OP_LOGE_FOR_INVALID_VALUE_WITH_REASON、OP_LOGE_FOR_INVALID_SHAPES_WITH_REASON、OP_LOGE_FOR_INVALID_SHAPEDIM_WITH_REASON、OP_LOGE_FOR_INVALID_SHAPE_WITH_REASON输出,覆盖reduction、广播失败及y声明形状冲突场景。 - fused_adam/adamw infershape 与 tiling 报错结构化:
fused_adam_infershape.cpp、fused_adamw_infershape.cpp将输入张量数为 0 与各张量列表形状不一致的OP_LOGE改为OP_LOGE_FOR_INVALID_TENSORNUM、OP_LOGE_FOR_INVALID_SHAPES_WITH_REASON;fused_adam_tiling.cpp的state_steps数量不匹配改用OP_LOGE_FOR_INVALID_TENSORNUMS_WITH_REASON,fused_adamw_tiling.cpp张量数为 0 改用OP_LOGE_FOR_INVALID_TENSORNUM。 - fused_sgd infershape 与 tiling 报错结构化:
fused_sgd_infershape.cpp将张量数为 0 及params/grads/momentum_buffer_list形状不一致的报错改为OP_LOGE_FOR_INVALID_SHAPESIZE_WITH_REASON、OP_LOGE_FOR_INVALID_SHAPES_WITH_REASON,并按是否存在momentum分别组织错误信息;fused_sgd_tiling.cpp张量数为 0 同样改用OP_LOGE_FOR_INVALID_SHAPESIZE_WITH_REASON。
🧩 本次 Review 已使用自定义 Skill:




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
✅ Congratulations! All modules have met the lgtm and approve requirements.
Module Approval Details
| module | lgtm status | approve status |
|---|---|---|
| loss | ✅ 陈风, 陈展熹 (2/2) | ✅ 陈风, 陈展熹 (2/1) |
| optim | ✅ 陈展熹, 陈风 (2/2) | ✅ 陈风, 陈展熹 (2/1) |
💡 Tip:
- Committer can comment
/approveor/lgtm- Commenting
/approveimplies both code review (lgtm) and intent to merge (approve)
CLA Signature Pass
ji-songyuan, thanks for your pull request. All authors of the commits have signed the CLA. 👍


Pull Request 已合并或已关闭。
If you want to solve this problem, you can click here to do it in the FAQs.


Pull Request 已合并或已关闭。
If you want to solve this problem, you can click here to do it in the FAQs.


描述
本 PR 对 loss/optim 目录下 8 个算子的错误码上报路径进行整改,消除失败路径上无错误码上报的裸
OP_LOGE,统一错误码首报规范,并补齐对应负向 UT。涉及算子:cross_v2、ctc_loss_v3、ctc_loss_v3_grad、mse_loss_grad_v2、apply_fused_ema_adam、fused_adam、fused_adamw、fused_sgd
主要改动:
OP_CHECK_IF验空后再使用,失败时通过OP_LOGE_FOR_INVALID_ARGUMENT_WITH_REASON上报 EZ0037(含参数名、op 名、失败原因)OP_LOGE_FOR_INVALID_TENSORNUM(EZ0022)ReportTilingFailure(REPORT_PREDEFINED_ERR_MSG上报 EZ1008 +OP_LOGE_WITHOUT_REPORT),覆盖 coreNum/ubSize/sysWorkspaceSize/memset_s 等约 20 处失败出口关联的Issue
#7497
测试
build.sh -u --ophost,覆盖全部 8 个受影响算子)文档更新
无
类型标签
AI/Agent生成声明
关联issue
Closes #7497