diff --git a/BMM算子优化分析_Release/MergeBatch_vs_IterBatch分析_v1.0.html b/BMM算子优化分析_Release/MergeBatch_vs_IterBatch分析_v1.0.html index 75d6c33..61f4d00 100644 --- a/BMM算子优化分析_Release/MergeBatch_vs_IterBatch分析_v1.0.html +++ b/BMM算子优化分析_Release/MergeBatch_vs_IterBatch分析_v1.0.html @@ -242,7 +242,7 @@ if (!l0CanLoadBatch_ && balanceRate < 0.8) return false; // 负载

参考文献

  1. [昇腾 950 NPU 架构白皮书](https://public-download.obs.cn-east-2.myhuaweicloud.com/ascend/%E6%98%87%E8%85%BE950%20NPU%E6%9E%B6%E6%9E%84%E7%99%BD%E7%9A%AE%E4%B9%A6.pdf)，华为技术有限公司，2026
    2. -
  2. [CANN AscendC 算子开发——Matmul 高阶 API 使能 UnitFlag](https://www.hiascend.com/document/detail/zh/canncommercial/850/opdevg/Ascendcopdevg/atlas_ascendc_best_practices_10_10003.html)，CANN 8.5.0
    3. +
  3. [CANN 9.2.0-beta.1 Ascend C API 参考](https://www.hiascend.com/document/detail/zh/CANNCommunityEdition/920beta1/API/ascendcopapi/atlasascendc_api_07_0003.html)——UnitFlag 机制详见"矩阵计算（ISASI）/Mmad 计算关键特性说明/UnitFlag"章节
  4. [cann-ops-nn 源码仓](https://gitcode.com/cann/ops-nn/tree/master/matmul/batch_mat_mul_v3)