变更摘要

本 PR 修复 foreach_add_list（AddListV2）在 int16/int8/uint8 数据类型下的整数溢出语义，将其从原先经 AddListFloatAdapter 的浮点中间态计算改为二进制补码回绕（wrap-around）语义，以对齐 PyTorch CPU/GPU 行为。核心改动是新增内核模板类 ForeachAddListWrap<T>（位于新增文件 foreach_add_list_wrap.h），并在 foreach_add_list.cpp 中将 TILING_KEY_IS(5)、(7)、(8) 三个分支的算子从 ForeachOneScalarTernary 切换为 ForeachAddListWrap。

主要改动

• 新增 ForeachAddListWrap<T> 内核类：在 foreach_add_list_wrap.h 中新增按数据块处理的 int16/int8/uint8 加法内核，包含 Init/Process/CopyIn/Compute/CopyOut 流水线，双缓冲队列及余量（remainder）数据处理逻辑。
• int16 路径直接计算：T=int16_t 时在 int16 域内直接执行 Muls(inLocal2, inLocal2, alphaVal, dataCount) 与 Add，利用硬件二进制补码回绕语义，无需中间转换。
• int8/uint8 提升计算后掩码回绕：T=int8_t/uint8_t 时先经 half 提升为 int16 精确计算 a + alpha * b，再回绕到低 8 位：uint8_t 通过 & 0x00FF 保留低 8 位，int8_t 通过 (v & 0x7F) - (v & 0x80) 映射到 [-128, 127]。
• 算子入口替换：foreach_add_list.cpp 中 TILING_KEY_IS(5)/(7)/(8) 分支由 ForeachOneScalarTernary<int16_t, float, AddListFloatAdapter<float>>、ForeachOneScalarTernary<int8_t, half, ...>、ForeachOneScalarTernary<uint8_t, half, ...> 分别替换为 ForeachAddListWrap<int16_t>、ForeachAddListWrap<int8_t>、ForeachAddListWrap<uint8_t>，并新增对 foreach_add_list_wrap.h 头文件的引用。

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

Tian_1122, thanks for your pull request. All authors of the commits have signed the CLA. 👍

/approve