变更摘要

本 PR 针对 dynamic_mx_quant 在 2 轴量化中尾轴很小时性能较差的问题进行性能优化，核心思路是通过多行合并计算的方式提升计算效率。改动涉及 op_kernel/arch35 下的两个内核头文件：dynamic_mx_quant_not_tail_axis_optimize_high_perf_large_tail.h 中调整了 half 转 bfloat16_t 的转换 trait 命名及调用，dynamic_mx_quant_not_tail_axis_optimize_high_perf_small_tail.h 则进行了大规模改动以实现小尾轴场景下的性能优化。

主要改动

• 重命名转换 trait: 在 dynamic_mx_quant_not_tail_axis_optimize_high_perf_large_tail.h 中，将 static constexpr Reg::CastTrait castTraitCublsHalf2Bf16 重命名为 castTraitCeilAlgHalf2Bf16，保持原有 CAST_RINT 舍入配置与 ZEROING 掩码合并模式不变。
• 更新转换调用点: 同步更新该文件中两处 Reg::Cast<bfloat16_t, xDtype, ...> 调用（对 x0AbsU16、x1AbsU16 的 half 绝对值数据做 bfloat16_t 转换），使其引用重命名后的 castTraitCeilAlgHalf2Bf16。
• 小尾轴场景大规模优化: 对 dynamic_mx_quant_not_tail_axis_optimize_high_perf_small_tail.h 进行大幅改动（新增 2001 行、删除 1156 行），依据 PR 描述通过多行合并计算方式优化尾轴很小时的量化计算性能。

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

ji-jun1, thanks for your pull request. All authors of the commits have signed the CLA. 👍

compile

您好，pr合入已准备就绪，请尽快联系committer进行检视，谢谢！

compile

您好，pr合入已准备就绪，请尽快联系committer进行检视，谢谢！

/check-pr

The following labels are not ready.

lgtm: Please wait for committers to review the code.

approved: Please wait for committers to review the code.

/lgtm
/approve

/lgtm
/approve