变更摘要

本 PR 仅调整 CI 编译脚本 scripts/ci/gen_ops_soc.py 中的 group_op_dict 配置，为 ascend950（A5）与 ascend960dt 两档 SoC 增加或调整多算子 obj 编译的拆分组数，以匹配相关算子当前的编译规模；拆分规则本身不变，仍按 op,N-i 生成编译任务并轮询分配到各编译组，不涉及算子 kernel / host 代码。

主要改动

• ascend950 组数上调: mega_moe 由 20 提升至 40、flash_attention_score_grad 由 5 提升至 10，增加这两类算子的编译任务拆分组数。
• ascend950 新增算子分组: 新增 matmul_all_reduce=40、allto_all_matmul=20、matmul_reduce_scatter_v2=20 三项通信类算子配置。
• ascend960dt 新增分组配置: 新增 flash_attention_score=5、mega_moe=40，以及 matmul_all_reduce=40、allto_all_matmul=20、matmul_reduce_scatter_v2=20，并使 flash_attention_score_grad 由 5 提升至 10，与 A5 档配置形成镜像一致。
• 其他 SoC 保持原状: ascend350 的 flash_attention_score_grad=5 等其余 SoC 配置未作修改。

🧩 本次 Review 已使用自定义 Skill：

• ascendc-code-review

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

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

compile

/lgtm

/approve
/lgtm