变更摘要

本 PR（标题 "ubx perf bugfix"）旨在修复 UBX 场景下爬坡（性能提升/吞吐增长）性能差的问题。核心改动集中在 all_reduce、reduce、reduce_scatter 的算法选择器及 ReduceNHR 的通道计算逻辑：将小数据量的判断方式由 IsSmallData(dataSize) 统一替换为与 SMALL_COUNT_512KB 的显式比较，并为 ReduceAutoSelector 增加基于 dataSize 的算法分支，同时针对特定拓扑优化 ReduceNHR::CalcRes 中通道的获取方式，以改善小数据量下的性能表现。

主要改动

• AllReduceAutoSelector::SelectMeshAlgoAicpuUBX 小数据判断调整：将 isClosNumMultipleOfMeshNum && !IsSmallData(dataSize) 改为 isClosNumMultipleOfMeshNum && dataSize > SMALL_COUNT_512KB，统一小数据量判定口径，影响是否选择 AicpuAllReducePipeLineUBX 等算法。
• ReduceScatterAutoSelector::SelectMeshAlgoAicpuForMesh1DClos 小数据判断调整：同样的条件由 !IsSmallData(dataSize) 改为 dataSize > SMALL_COUNT_512KB，影响 AicpuReduceScatterPipeLineUBX 等算法的选择。
• ReduceAutoSelector::SelectMeshAlgoAicpu 增加数据量分支：在非 64 位数据类型且非 HCCL_REDUCE_PROD 的分支中，按 dataSize 区分算法——大于 SMALL_COUNT_512KB 时选择 ReduceParallelMesh1DNHRUBX，否则选择 AicpuReduceSoleNHR。
• ReduceNHR::CalcRes 通道计算优化：当 topoInfo->level0Topo == Level0Shape::MESH_1D_CLOS 且 !topoInfo->level0PcieMix 时，改用 CalcChannelRequestNhrMultiJetty 计算通道并仅保留协议为 COMM_PROTOCOL_UBC_CTP 的通道加入 level1Channels，其余场景仍走原 CalcChannelRequestNhr 路径。

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

yjzz1007, thanks for your pull request. All authors of the commits have signed the CLA. 👍

complie