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_BATCH_SEND_RECV_EXECUTOR_H
      12              : #define COLL_BATCH_SEND_RECV_EXECUTOR_H
      13              : 
      14              : #include "coll_comm_executor.h"
      15              : 
      16              : namespace hccl {
      17              : class CollBatchSendRecvExecutor : public CollCommExecutor {
      18              : public:
      19              :     CollBatchSendRecvExecutor(const HcclDispatcher dispatcher, std::unique_ptr<TopoMatcher> &topoMatcher);
      20           64 :     ~CollBatchSendRecvExecutor() override = default;
      21              :     HcclResult Orchestrate(OpParam& param, AlgResourceResponse& algResource) override;
      22              :     HcclResult GetAdjInfo(AlgResourceResponse& algRes, AdjInfo& adjInfo) override;
      23              :     // 增量建链资源计算接口
      24              :     HcclResult CalcIncreLinkRequest(const OpParam& param, std::set<u32>& ranksLinked, AlgResourceRequest& resourceRequest, bool& needIncreLink)
      25              :         override;
      26              :     HcclResult GetSendTargetLink(u32 remoteUserRank, LINK& targetLink);
      27              :     HcclResult GetRecvTargetLink(u32 remoteUserRank, LINK& targetLink);
      28              : protected:
      29              :     /* *************** 资源计算 *************** */
      30              :     void ParseParam(const OpParam& param) override;
      31              :     HcclResult CalcCommInfo(std::vector<LevelNSubCommTransport>& opTransport) override;
      32              : 
      33              :     /* *************** 算法编排 *************** */
      34              : 
      35              :     u64 CalcSendLoopMaxCount(DeviceMem& inCCLBuffer, const u32 unitSize);
      36              :     u64 CalcRecvLoopMaxCount(DeviceMem& outCCLBuffer, const u32 unitSize);
      37              :     HcclResult ProcessSendDataSlice(Stream& stream, bool needStreamSync, bool retryEnable);
      38              :     HcclResult ProcessRecvDataSlice(Stream& stream, bool retryEnable);
      39              :     HcclResult CalcSendSlices(AlgResourceResponse& algRes);
      40              :     HcclResult CalcRecvSlices(AlgResourceResponse& algRes);
      41              :     HcclResult GetPairWiseList(HcclSendRecvItem *sendRecvInfo, u32 itemNum);
      42              :     HcclResult ProcessSelfSendRecvTasks(Stream& stream);
      43              :     HcclResult SendKernelRun(Stream& stream, ExecMem &execMem, u32 remoteUserRank, bool retryEnable);
      44              :     HcclResult RecvKernelRun(Stream& stream, ExecMem &execMem, u32 remoteUserRank, bool retryEnable);
      45              :     HcclResult GetTransport(u32 commIndex, u32 remoteUserRank, LINK &targetLink);
      46              :     struct SendRecvSlice {
      47              :         u8* addr;
      48              :         u64 size;
      49              :         u32 remoteRank;
      50            0 :         SendRecvSlice(u8* addr, u64 size, u32 remoteRank) : addr(addr), size(size), remoteRank(remoteRank) {}
      51              :     };
      52              : 
      53              :     u32 remoteUserRank_ = 0;
      54              :     const u32 MAX_LOOP_IN_ONCE_LAUNCH = 200;
      55              :     std::deque<SendRecvSlice> sendDataSilces_;
      56              :     std::deque<SendRecvSlice> recvDataSilces_;
      57              : private:
      58              :     HcclResult RunLoopInHostUnfoldMode(OpParam& param);
      59              :     HcclResult RunLoopInAicpuUnfoldMode(OpParam& param);
      60              :     HcclResult CalcStreamNum(u32& streamNum) override;
      61              : 
      62              :     HcclResult MainPostSubWait(Stream& mainStream, Stream& subStream);
      63              :     HcclResult SubPostMainWait(Stream& mainStream, Stream& subStream);
      64              : 
      65              : protected:
      66              : 
      67              :     std::set<u32> commTargetUserRankSet_;
      68              :     std::deque<HcclSendRecvItem*> sendToSelfDeque_;
      69              :     std::deque<HcclSendRecvItem*> recvFromSelfDeque_;
      70              :     std::deque<HcclSendRecvItem*> sendDeque_;
      71              :     std::deque<HcclSendRecvItem*> recvDeque_;
      72              : };
      73              : } // namespace hccl
      74              : 
      75              : #endif