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_ALLTOALL_COMM_EXECUTOR_H
      12              : #define COLL_ALLTOALL_COMM_EXECUTOR_H
      13              : #include "coll_comm_executor.h"
      14              : namespace hccl {
      15              : constexpr u32 MINORS_NUM_TWO = 2;
      16              : constexpr u64 SMALL_SIZE_FULLMESH = 262144;
      17              : 
      18              : class CollAlltoAllExecutor : public CollNativeExecutorBase {
      19              : public:
      20              :     CollAlltoAllExecutor(const HcclDispatcher dispatcher, std::unique_ptr<TopoMatcher> &topoMatcher);
      21            5 :     ~CollAlltoAllExecutor() override = default;
      22              : 
      23              :     HcclResult Orchestrate(OpParam& param, AlgResourceResponse& algRes) override;
      24              :     HcclResult GetAdjInfo(AlgResourceResponse& algRes, AdjInfo& adjInfo) override;
      25              :     virtual HcclResult SetExcutorExtraInfo(const std::vector<SendRecvInfo> &allMeshAggregationSendRecvInfo, u64 cclbufferSize);
      26              :     HcclResult CalcResRequest(const OpParam& param, AlgResourceRequest &resourceRequest) override;
      27              :     virtual HcclResult CheckNeedCreateVirtualLinks(AlgResourceRequest &resourceRequest);
      28              : 
      29              :     HcclResult SetVirtualDispatcher(const HcclDispatcher virtualDispatcher);
      30              : 
      31              :     HcclResult SetParallelTaskLoader(ParallelTaskLoader *parallelTaskLoader);
      32              : 
      33              :     virtual HcclResult CheckNeedRecreateComm(u64 lastScratchMemSize, bool& needRecreateAlltoallComm);
      34              :     static HcclResult RunAlltoAllTemplate(const std::unique_ptr<AlgTemplateBase> &executor,
      35              :         const SubCommInfo &commInfo);
      36              :     static HcclResult RunAlltoAllVTemplateStaged(const std::unique_ptr<AlgTemplateBase> &executor,
      37              :         const SubCommInfo &commInfo);
      38              :     static HcclResult RunTemplateWithVirtualLink(const std::unique_ptr<AlgTemplateBase> &executor,
      39              :         const SubCommInfo &commInfo);
      40              : 
      41              : protected:
      42              :     /* *************** 算法编排 *************** */
      43              :     // 公共接口
      44              :     virtual HcclOpMetaInfo GetOpMeta(HcclCMDType opType, const u64 size);
      45              :     void UpdateAlltoAllZCopyMode(std::vector<SendRecvInfo> &allMeshAggregationSendRecvInfo, u64 cclbufferSize);
      46              :     void CalcIntraMeshAggregationAlltoAllMemInfo(const AlltoAllUserRankInfo &userRankInfo,
      47              :         const std::vector<SendRecvInfo> &allSendRecvInfo,
      48              :         std::map<u32, std::list<OneSendRecvAddrInfo>> &sendAddrInfosIntra,
      49              :         std::map<u32, std::list<OneSendRecvAddrInfo>> &recvAddrInfosIntra, u32 meshAggregationRankSize,
      50              :         const bool &isSingleMesh);
      51              :     void CalcIntraMeshAggregationSendInfo(const AlltoAllUserRankInfo &userRankInfo,
      52              :         const SendRecvInfo &mySendRecvInfo, const std::vector<SendRecvInfo> &myMeshAggregationSendRecvInfo,
      53              :         u32 rankInMeshAggregation, u32 infoIndex, OneSendRecvAddrInfo &curSendInfo, u32 meshAggregationRankSize,
      54              :         const bool &isSingleMesh);
      55              :     void CalcIntraMeshAggregationRecvInfo(const AlltoAllUserRankInfo &userRankInfo,
      56              :         const std::vector<SendRecvInfo> &myMeshAggregationSendRecvInfo, u32 infoIndex, OneSendRecvAddrInfo &curRecvInfo,
      57              :         u32 meshAggregationRankSize, const bool &isSingleMesh);
      58              :     void CalcIntraMeshAggregationRecvInfoInMeshAggregation(u32 rankIndex, u32 infoIndex,
      59              :         const std::vector<SendRecvInfo> &myMeshAggregationSendRecvInfo, u64 &localOffset, u32 &offsetCounter,
      60              :         u64 &localLength, u64 &remoteOffset, u32 meshAggregationRankSize);
      61              :     u64 CalAlltoAllVScratchMemSize(u64 &workSpaceMemSize);
      62              :     bool HasMassTasks(std::vector<SendRecvInfo> &allMeshAggregationSendRecvInfo);
      63              : 
      64              :     OpParam AlltoAllVParam_;
      65              :     std::vector<SendRecvInfo> allMeshAggregationSendRecvInfo_;
      66              :     SendRecvInfo localSendRecvInfo_;
      67              :     bool isAlltoAllZCopyMode_ = false;
      68              : 
      69              :     HcclDispatcher vDispatcher_;
      70              : 
      71              :     ParallelTaskLoader* parallelTaskLoader_; // 并行下发taskloader管理
      72              : };
      73              : 
      74              : } // namespace hccl
      75              : 
      76              : #endif