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_REDUCESCATTER_RING_ZEROCOPY_EXECUTOR_H
      12              : #define COLL_REDUCESCATTER_RING_ZEROCOPY_EXECUTOR_H
      13              : #include "coll_reduce_scatter_executor.h"
      14              : 
      15              : namespace hccl {
      16              : class CollReduceScatterRingZerocopyExecutor : public CollReduceScatterExecutor {
      17              : public:
      18              :     explicit CollReduceScatterRingZerocopyExecutor(const HcclDispatcher dispatcher,
      19              :     std::unique_ptr<TopoMatcher> &topoMatcher);
      20            0 :     ~CollReduceScatterRingZerocopyExecutor() override = default;
      21              : 
      22              : protected:
      23              :     HcclResult CalcCommInfo(std::vector<LevelNSubCommTransport>& opTransport) override;
      24              :     u32 level0Rank_ = INVALID_VALUE_RANKID;
      25              :     u32 level1Rank_ = INVALID_VALUE_RANKID;
      26              :     u32 level2Rank_ = INVALID_VALUE_RANKID;
      27              :     u32 level0RankSize_ = INVALID_VALUE_RANKSIZE;
      28              :     u32 level1RankSize_ = INVALID_VALUE_RANKSIZE;
      29              :     u32 level2RankSize_ = INVALID_VALUE_RANKSIZE;
      30              : 
      31              : private:
      32              :     void ParseParam(const OpParam& param) override;
      33              :     /* *************** 资源计算 *************** */
      34              :     std::set<u32> commTargetUserRankSet_;
      35              :     HcclResult CalcStreamNum(u32& streamNum) override;
      36              :     HcclResult CalcLevel0CommInfo(TransportMemType inputType,
      37              :         TransportMemType outputType, std::vector<LevelNSubCommTransport>& opTransport) override;
      38              :     HcclResult CalcTransportMemType(TransportMemType &inputType, TransportMemType &outputType);
      39              :     u64 CalcLoopMaxCount(const u32 unitSize) override;
      40              : 
      41              :     /* *************** 算法编排 *************** */
      42              :     HcclResult SemiRingReduceScatter(
      43              :         const std::string &tag, DeviceMem inputMem, DeviceMem outputMem,
      44              :         const u64 count, const HcclDataType dataType, const HcclReduceOp reductionOp,
      45              :         const std::vector<std::vector<Slice> > multRingsSliceZero, Stream stream, s32 profStage,
      46              :         const u64 baseOffset, const HcomCollOpInfo *opInfo,
      47              :         const std::vector<std::vector<Slice>> multRingsUserMemSlice);
      48              :     HcclResult RunIntraSeverReduceScatter(const std::string &tag, DeviceMem &inputMem, DeviceMem &outputMem,
      49              :         const u64 count, const HcclDataType &dataType, const HcclReduceOp &reductionOp,
      50              :         const std::vector<std::vector<Slice>> &multRingsSliceZero, const Stream &stream, s32 profStage,
      51              :         const u64 baseOffset = 0, const HcomCollOpInfo *opInfo = nullptr,
      52              :         const std::vector<std::vector<Slice>> &multRingsUserMemSlice = std::vector<std::vector<Slice>>(0),
      53              :         const bool disableDMAReduce = false);
      54              :     HcclResult KernelRunIntraServerPre(const OpParam &param, ExecMem &execMem) override;
      55              :     HcclResult KernelRunInterServer(const OpParam &param, ExecMem &execMem) override;
      56              :     virtual HcclResult KernelRunInterServerPostProcess(const OpParam &param, const ExecMem &execMem);
      57              : 
      58              :     virtual HcclResult CalcLevel0DataSlices(const OpParam &param, const ExecMem &execMem, std::vector<Slice> &dataSegsSlice);
      59              :     virtual HcclResult KernelRunInterServerPreProcess(const OpParam &param, const ExecMem &execMem);
      60              : };
      61              : 
      62              : } // namespace hccl
      63              : 
      64              : #endif