变更摘要

本 PR 为 2026 年 8 月社区任务 aclblasCgemmBatched 新增 Ascend 950PR 算子的设计文档，属于纯文档更新（1 个文件，新增 448 行，无代码改动）。文档位于 04_tasks/01_community-task-2026/tasklist/08-26-aclblasCgemmBatched-950/qq_51183872/docs/design.md，系统阐述了该算子的接口语义、实现方案与验证方案，并附有基于 Ascend 950PR 真机与 CANN 9.1.0 环境的测试结论。

主要改动

• 功能与接口对标: 新增 COMPLEX64 batched GEMM 的功能、接口定义及与 cuBLAS 的对标语义说明，明确该社区任务的目标行为。
• Host 侧实现设计: 补充参数校验、快速返回路径、workspace 规划及 tiling 设计的说明，覆盖算子调度与资源分配策略。
• 计算路径设计: 说明对 N/T/C 转置、复数 alpha/beta、leading dimension 及设备指针数组的支持，并描述小输出直接复数累加、通用 FP32 4M 分解及 realification 三条性能路径。
• 验证方案: 给出 Ascend 950PR 上正确性、精度与性能的验证方案，并说明原任务包 1,200 条 CSV 测试输入的覆盖情况，测试后端为真实 NPU（CANN 9.1.0），性能门禁与 verify_performance.py 的 GPU / 0.4 公式均通过。

Hi @qq_51183872, welcome to submitting your first PR to cann-ops-competitions!

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.

/approve

The MR can not be merged, because of CodeReview discussion not resolved

If you want to solve this problem, you can click here to do it in the FAQs.