变更摘要

该 PR 为 AscendC 向量接口（vf 接口）的整改与重命名重构（源分支 refactor/rename-ascendc-api），不涉及业务逻辑调整，PR 描述声明"修改后二进制未变化"。改动主要针对 arch35 下多个算子的内核实现，将旧版 API 统一替换为新版接口：DataCopy 按读写方向拆分为 LoadAlign/StoreAlign，ReduceSum/ReduceMax/ReduceMin 统一为带模板参数的 Reduce<ReduceType::...>，CompareScalar 改为 Compares，ReduceMaxWithDataBlock 改为 ReduceDataBlock<ReduceType::MAX>，MaskNot/MaskAnd 改为 Not/And，并将 __local_mem__ 指针标注统一更正为 __ubuf__。

主要改动

• 规约接口统一为 Reduce<ReduceType::...>：在 swiglu_group_base.h、swiglu_group_quant_base.h、foreach_norm_regbase.h、foreach_non_finite_check_and_unscale_regbase.h 中，将 ReduceSum、ReduceMax、ReduceMin 分别替换为 Reduce<ReduceType::SUM>、Reduce<ReduceType::MAX>、Reduce<ReduceType::MIN>，将 ReduceMaxWithDataBlock 替换为 ReduceDataBlock<ReduceType::MAX>。
• 数据搬移接口按读写拆分：在 foreach_addcmul_scalar_regbase.h、foreach_mul_scalar_list_regbase.h、foreach_non_finite_check_and_unscale_regbase.h、foreach_norm_regbase.h、sigmoid_cross_entropy_with_logits_v2_dag.h 中，将双向的 DataCopy 替换为读方向 LoadAlign、写方向 StoreAlign，并保留 PostLiteral::POST_MODE_UPDATE、StoreDist::DIST_FIRST_ELEMENT_B32、LoadDist::DIST_BRC_B32 等原有模式参数。
• 寄存器比较与掩码接口改名：在 swiglu_group_quant_base.h 中，CompareScalar<uint16_t, CMPMODE::GT/LT> 改为 Compares<uint16_t, CMPMODE::GT/LT>，MaskAnd 改为 And、MaskNot 改为 Not，函数语义与调用参数保持一致。
• UB 指针标注统一为 __ubuf__：在多个 foreach_* 与 group_norm_silu 系列的 regbase/welford 头文件（如 group_norm_silu_regbase_two_pass.h、group_norm_silu_regbase_welford_generalized.h 等）中，将指向 UB 内存的 __local_mem__ 指针声明及 GetPhyAddr() 处的强转统一改为 __ubuf__。

代码审查

✅ 未发现问题

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

💡 Tip:

• Committer can comment /approve or /lgtm
• Commenting /approve implies both code review (lgtm) and intent to merge (approve)

CLA Signature Pass

xieshengwei1024, thanks for your pull request. All authors of the commits have signed the CLA. 👍

/approve

/lgtm
/approve