Pull Request已成功合入, 合并人@CANN-robot
(感谢 puppy 的贡献)Hi @hanggongmao1, welcome to submitting your first PR to ops-math!
PR Merge Steps
1. CLA Signing
If the current PR label includes cann-cla/yes, it means you have signed the CLA and can proceed to the next step. If the label includes cann-cla/no, please sign the CLA first. If you have any questions, please refer to the FAQ.
2. CI Check
Please comment /compile to trigger the CI pipeline check. If the CI run is successful, the PR will be tagged with ci-pipeline-passed and you can proceed to the next step. If the CI run fails, the PR will be tagged with ci-pipeline-failed, please check the CI logs to fix the issues in the PR. If you have any questions, please refer to the FAQ.
3. Code Review
After CI passes, please refer to the PR Approval Progress and proactively @ the committers in the table to review the code. After approval, committers will comment /lgtm and /approve. Once the lgtm and approved labels are successfully added, the PR will be merged automatically.


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 |
|---|---|---|
| experimental | ✅ 顾宝成, yue-ma (2/2) | ✅ 顾宝成, yue-ma (2/1) |
💡 Tip:
- Committer can comment
/approveor/lgtm- Commenting
/approveimplies both code review (lgtm) and intent to merge (approve)
CLA Signature Pass
hanggongmao1, thanks for your pull request. All authors of the commits have signed the CLA. 👍


描述
experimental Atan2 的 epsilon 符号近似使
atan2(+0,-1)返回 0,预期为 π;原点的0/0则返回 NaN,与该算子 README 定义的原点结果 +0 不符。改为精确比较选择象限,并从分子符号位形成 ±1,保留负半轴上的 ±π 和正半轴上的带符号零;所有原点 signed-zero 组合显式选择 +0。BF16 结果为 NaN 时写回 quiet-NaN 编码,使特殊值语义不依赖 FP32 到 BF16 的 narrowing 行为。
复用现有临时缓冲,host tiling 额外预留 8 KiB Select 临时空间,tile 按 64 个 FP32 元素对齐,并初始化短 core / 尾块最后一个比较 repeat。新增最小 UB 容量检查和边界回归,保留有限值、NaN、带符号零断言;现有 FP16 随机 UT 也检查数据生成、读写及比对状态,覆盖全部 8192 个输入元素。
CMake 为随机用例显式传入原始数据目录的绝对路径,再复制数据脚本,避免依赖构建 / 测试工作目录;复制、生成、读取与精度比对失败均保持断言。
主机侧改为显式缓冲区预算的纯调度函数:三路双缓冲队列、五路 FP32 scratch、Atan 临时空间和固定 8 KiB Select 预留分别计入 UB;统一计算大小核及尾 tile,保留原调度数据格式,拒绝超出内核 uint32_t 偏移范围的输入。图上下文适配只创建一个平台对象,统一读取资源和 workspace,消除原来分散重复的平台、类型与分块处理。原版权、许可证、检查规则与精度断言均保留。
关联的Issue
Fixes #3471
https://gitcode.com/cann/ops-math/issues/3471
测试
对应当前 HEAD
b4ad3d285b49aa6fde63954b318390c9a12c8105,在 HiDevLab 的 A3 单卡 / CANN 9.0.0 环境完成:__FILE__裁剪为 basename 并核对实际预处理宏,覆盖 CI 的编译路径改写;两个目录均不存在旧版所依赖的相对数据路径。BuildSchedule生成全部分块参数;覆盖三种精度、单卡内 1 / 3 个 vector 核、四象限、零轴、原点、NaN、短 core / 尾块、完整 tile 加尾块,以及实际可用 UB 和较小 UB 预算。SDK 查询到可用 UB 为 196352 字节,设备有 48 个 AIV 核;测试仅使用逻辑设备 0。git diff --check通过。最终源码 / 测试文件均按 SHA256 核对。54 组硬件回归与最终 CPU 回归使用的全部生产源码字节一致;完整原始日志、退出状态、GoogleTest XML、运行脚本及 manifest 已归档。归档 SHA256:
a0992d5b4b9d30daf2c70c40f0d8fbd727cd36b375509191c4c89092461482bc。工作目录回归的补充归档 SHA256:
fb11e139b9169b0606d6f1d73567260a639cdffc7e3667079d615a970f656302;对应最新源码 manifest SHA256:46c9194d13a1327e53dd7411b4559e51af7cf5a478f2a52cb3bb6179e3bd6bf5。本地 OAT hook 要求的 Invalid File Type / License Header Invalid 均为 0。原始 OAT 另报 Huawei Device / Huawei Technologies 版权 policy 差异,已保留报告;未改变版权归属或扫描规则。当前提交的社区 CI/SCA 结果以对应流水线为准。
验证范围为原生 host UT、CPU 内核 UT,以及生产 kernel 和生产调度参数的真实单卡执行;未运行完整 ACLNN/C API 自定义算子包调度,未宣称全仓或其他架构硬件回归通过。
文档更新
README 明确负半轴的 ±0 对应 ±π、值域 [−π,π],以及所有原点 signed-zero 组合返回 +0。
类型标签