变更摘要

本次变更针对 impl/simt_api/math_functions_impl.h 中的浮点数学函数进行性能优化，主要消减 atan2f 和 rnorm4df 中的分支路径。atan2f 由条件分支 + 多路选择重构为无分支的共享计算流程，统一用 base + sgn * t 的 __fma 形式还原象限；rnorm4df 将特殊值（inf/NaN/零范数）的提前返回改为计算完成后的结果覆盖，减少提前退出的分支。

主要改动

• atan2f 去除零分母分支：引入 safe_hi = nz ? hi : 1.0f，使比值 r = lo / safe_hi 与多项式求值始终执行，避免 hi == 0 时的 0/0 路径发散，异常结果由后续选择丢弃。
• 无理输入的特殊值处理改为选择：原先对 inf/inf 产生的 NaN 与 |y| == |x| 对角线的特判，改为 t = (t != t) ? pi/4 : t，使仿射公式自然产出 pi/4 与 3pi/4。
• 象限还原合并为单次 fma：用 xneg = __float_as_uint(x) >> 31 提取符号，按 y_gt_x 与 xneg 选定 base 和 sgn，以 __fma(sgn, t, base) 替换原先三分支的角度拼接。
• 零幅值与符号处理重构：hi == 0 时通过 a0 选择 pi、0 或 pi/2，由 nz 决定最终取值；y 的符号改用 __float_as_uint(y) >> 31 按位判断，NaN 仍通过 ax + ay 传播。
• rnorm4df 特殊值判断后移：删除入口处的 isinf/isnan 提前返回与零范数提前返回，改为在完成倒数范数计算与一次 Newton 迭代后，依次用 ASCRT_INF_F、ASCRT_NAN_F、0.0f 覆盖 result。

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

xuyiyang0324, thanks for your pull request. All authors of the commits have signed the CLA. 👍

compile

Notification

This pull request has been changed(code update) or closed, so removes the following label(s): ci-pipeline-passed.

compile

Notification

This pull request has been changed(code update) or closed, so removes the following label(s): ci-pipeline-passed, stat/needs-squash.

compile

/lgtm

/lgtm
/approve