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_ALLGATHER_RING_FOR_910_93_EXECUTOR_H
      12              : #define COLL_ALLGATHER_RING_FOR_910_93_EXECUTOR_H
      13              : #include "coll_all_gather_executor.h"
      14              : namespace hccl {
      15              : class CollAllGatherRingFor91093Executor : public CollAllGatherExecutor {
      16              : public:
      17              :     explicit CollAllGatherRingFor91093Executor(const HcclDispatcher dispatcher, std::unique_ptr<TopoMatcher> &topoMatcher);
      18            2 :     ~CollAllGatherRingFor91093Executor() override = default;
      19              : 
      20              : private:
      21              :     /* *************** 资源计算 *************** */
      22              :     HcclResult CalcStreamNum(u32& streamNum) override;
      23              :     HcclResult CalcCommInfo(std::vector<LevelNSubCommTransport>& opTransport) override;
      24              :     HcclResult CalcLevel0CommInfo(TransportMemType inputType, TransportMemType outputType,
      25              :         std::vector<LevelNSubCommTransport>& opTransport) override;
      26              :     HcclResult CalcLevel2CommInfo(TransportMemType inputType, TransportMemType outputType,
      27              :         std::vector<LevelNSubCommTransport>& opTransport) override;
      28              :     HcclResult CalcTransportMemType(TransportMemType &inputType, TransportMemType &outputType);
      29              : 
      30              :     /* *************** 算法编排 *************** */
      31              :     u64 CalcLoopMaxCount(const u64 cclBuffSize, const u32 unitSize) override;
      32              : 
      33              :     virtual u64 CalcDstMemOffset(const OpParam &param, u32 perDataSize, u64 inputMemSize) const;
      34              :     virtual HcomCollOpInfo GetHcomCollOpInfo(const OpParam &param, const ExecMem &execMem) const;
      35              :     virtual std::vector<Slice> PrepareSlicesL2(const OpParam &param, const SubCommInfo &level2CommInfo,
      36              :         const SubCommInfo &level1CommInfo, const SubCommInfo &level0CommInfo, u32 perDataSize, u64 inputMemSize) const;
      37              :     virtual std::vector<Slice> PrepareSlicesL1(const OpParam &param, const SubCommInfo &level2CommInfo,
      38              :         const SubCommInfo &level1CommInfo, const SubCommInfo &level0CommInfo, u32 perDataSize, u64 inputMemSize) const;
      39              :     virtual HcclResult PrepareSlicesL0(std::vector<std::vector<Slice>> &multRingsSlice, const OpParam &param,
      40              :         const SubCommInfo &level2CommInfo, const SubCommInfo &level1CommInfo, const SubCommInfo &level0CommInfo,
      41              :         u32 perDataSize, u64 inputMemSize);
      42              :     virtual HcclResult PrepareUserMemSlices(std::vector<std::vector<Slice>> &userMemSlices,
      43              :         const std::vector<std::vector<Slice>> &multRingsSlice, const OpParam &param, const SubCommInfo &level2CommInfo,
      44              :         const SubCommInfo &level1CommInfo, const SubCommInfo &level0CommInfo, u32 perDataSize, u64 inputMemSize);
      45              :     virtual HcclResult GetLevelCommInfo();
      46              :     
      47              :     virtual HcclResult RunIntraSeverAllGather(const std::string &tag, DeviceMem &inputMem, DeviceMem &outputMem,
      48              :         const u64 count, const HcclDataType &dataType,
      49              :         const std::vector<std::vector<Slice>> &multRingsSliceZero, const Stream &stream,
      50              :         s32 profStage, const u64 baseOffset = 0, const HcomCollOpInfo *opInfo = nullptr,
      51              :         const std::vector<std::vector<Slice>> &multRingsUserMemSlice = std::vector<std::vector<Slice>> (0));
      52              :     HcclResult KernelRun(const OpParam &param, ExecMem &execMem) override;
      53              :     HcclResult Getlevel1CommRank(SubCommInfo& level1CommInfo) override;
      54              :     HcclResult SelectTempAlg(std::unique_ptr<AlgTemplateBase> &level1TempAlg, u32 level1RankSize) override;
      55              : protected:
      56              :     SubCommInfo logicalLevel0CommInfo_ = {0, 0, {}, {}};
      57              :     SubCommInfo logicalLevel1CommInfo_ = {0, 0, {}, {}};
      58              :     CommPlane logicalLevel0plane_ = COMM_LEVEL_RESERVED;
      59              :     CommPlane logicalLevel1plane_ = COMM_LEVEL_RESERVED;
      60              : };
      61              : 
      62              : } // namespace hccl
      63              : 
      64              : #endif