Line data    Source code

       1              : /**
       2              :  * Copyright (c) 2025 Huawei Technologies Co., Ltd.
       3              :  * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
       4              :  * CANN Open Software License Agreement Version 2.0 (the "License").
       5              :  * Please refer to the License for details. You may not use this file except in compliance with the License.
       6              :  * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
       7              :  * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
       8              :  * See LICENSE in the root of the software repository for the full text of the License.
       9              :  */
      10              : 
      11              : #ifndef COLL_RUN_ALLTOALLV_DIRECT_FULLMESH_EXECUTOR_H
      12              : #define COLL_RUN_ALLTOALLV_DIRECT_FULLMESH_EXECUTOR_H
      13              : #include "coll_all_to_all_executor.h"
      14              : namespace hccl {
      15              : class CollRunAlltoAllDirectFullmesh : public CollAlltoAllExecutor {
      16              : public:
      17              :     CollRunAlltoAllDirectFullmesh(const HcclDispatcher dispatcher, std::unique_ptr<TopoMatcher> &topoMatcher);
      18            0 :     ~CollRunAlltoAllDirectFullmesh() override = default;
      19              :     HcclResult Orchestrate(OpParam& param, AlgResourceResponse& algRes) override;
      20              :     HcclResult GetAdjInfo(AlgResourceResponse& algRes, AdjInfo& adjInfo) override;
      21              : 
      22              :     // 用于alltoallv算子的aicpu展开cache
      23              :     HcclResult MarkNeedAlltoallvCache() override;
      24              :     HcclResult GetHcclOffsetDstRanksMap(std::unordered_map<uint64_t, std::vector<uint32_t>>& hcclOffsetDstRanksMap) const override;
      25              : private:
      26              :     HcclOpMetaInfo GetOpMeta(HcclCMDType opType, const u64 size) override;
      27              :     HcclResult CalcStreamNum(u32& streamNum) override;
      28              :     HcclResult CalcLevel0CommInfo(TransportMemType inputType, TransportMemType outputType,
      29              :         std::vector<LevelNSubCommTransport>& opTransport) override;
      30              :     HcclResult CalcCommInfo(std::vector<LevelNSubCommTransport>& opTransport) override;
      31              :     HcclResult KernelRun(const OpParam &param, ExecMem &execMem) override;
      32              :     HcclResult Getlevel1CommRank(SubCommInfo& level1CommInfo) override;
      33              :     HcclResult SelectTempAlg(std::unique_ptr<AlgTemplateBase> &level1TempAlg, u32 level1RankSize) override;
      34              :     HcclResult GetDevNumInlocalPod(u32& devNumInlocalPod) override;
      35              :     HcclResult GetAlltoAllvTmpRankSendRecvInfo(const OpParam &param);
      36              :     HcclResult GetLocalSendRecvInfoforAlltoall(const OpParam &param);
      37              :     HcclResult GetLocalSendRecvInfoforAlltoallV(const OpParam &param);
      38              :     HcclResult GetLocalSendRecvInfoforAlltoallVC(const OpParam &param);
      39              :     HcclResult CalcTransportMemType(TransportMemType &inputType, TransportMemType &outputType);
      40              :     HcclResult GetLocalSDMAGroupInfo(const u32 userRank, u32& devNumInlocalPod, u32& rankIdxInPod);
      41              : 
      42              :     // 用于alltoallv算子的aicpu展开cache
      43              :     bool needAlltoallvCache_ = false; // 是否需要对当前alltoallv算子做aicpu cache
      44              :     std::unordered_map<uint64_t, std::vector<uint32_t>> hcclOffsetDstRanksMap_; // Local hccl input buffer中的local hccl offset到remote dst ranks的映射 (用于PrepareIntraData)
      45              : };
      46              : } // namespace hccl
      47              : #endif