Line data    Source code

       1              : /**
       2              :  * Copyright (c) 2026 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_PIPELINE_FOR_910_93_EXECUTOR_H
      12              : #define COLL_ALLGATHER_PIPELINE_FOR_910_93_EXECUTOR_H
      13              : 
      14              : #include "coll_all_gather_executor.h"
      15              : 
      16              : namespace hccl {
      17              : class CollAllGatherPipelineFor91093Executor
      18              :     : public CollAllGatherExecutor {
      19              : public:
      20              :     explicit CollAllGatherPipelineFor91093Executor(
      21              :         const HcclDispatcher dispatcher,
      22              :         std::unique_ptr<TopoMatcher> &topoMatcher);
      23            0 :     ~CollAllGatherPipelineFor91093Executor() override = default;
      24              : 
      25              :     // 流水线编排
      26              :     HcclResult Orchestrate(OpParam& param, AlgResourceResponse& algRes) override;
      27              : 
      28              : private:
      29              :     // 资源计算
      30              :     HcclResult CalcStreamNum(u32& streamNum) override;
      31              :     HcclResult CalcCommInfo(std::vector<LevelNSubCommTransport>& opTransport) override;
      32              :     HcclResult CalcLevel0CommInfo(TransportMemType inputType, TransportMemType outputType,
      33              :         std::vector<LevelNSubCommTransport>& opTransport) override;
      34              :     HcclResult CalcLevel2CommInfo(TransportMemType inputType, TransportMemType outputType,
      35              :         std::vector<LevelNSubCommTransport>& opTransport) override;
      36              :     HcclResult CalcTransportMemType(TransportMemType &inputType, TransportMemType &outputType);
      37              : 
      38              :     // 算法编排
      39              :     u64 CalcLoopMaxCount(const u64 cclBuffSize, const u32 unitSize) override;
      40              :     HcclResult RunLoop(OpParam &param);
      41              :     HcclResult RunL2Stage(const OpParam &param, ExecMem &execMem, u64 loopIdx, u64 memIdx, u64 bufferSliceNum);
      42              :     HcclResult RunL1L0Stage(const OpParam &param, ExecMem &lastExecMem, u64 loopIdx, u64 memIdx,
      43              :         u64 bufferSliceNum);
      44              : 
      45              :     // 层级算法调用方法（从模板移动到执行器）
      46              :     HcclResult KernelRunInterSuperPod(const OpParam &param, ExecMem &execMem, u64 baseOffset); // 跨超
      47              :     HcclResult KernelRunIntraServer(const OpParam &param, ExecMem &execMem, u64 baseOffset); // server内
      48              :     HcclResult KernelRunInterServer(const OpParam &param, ExecMem &execMem, u64 baseOffset); // 跨server
      49              : 
      50              :     virtual std::vector<Slice> PrepareSlicesL2(const OpParam &param, const SubCommInfo &level2CommInfo,
      51              :         const SubCommInfo &level1CommInfo, const SubCommInfo &level0CommInfo, u32 perDataSize, u64 inputMemSize) const;
      52              :     virtual std::vector<Slice> PrepareSlicesL1(const OpParam &param, const SubCommInfo &level2CommInfo,
      53              :         const SubCommInfo &level1CommInfo, const SubCommInfo &level0CommInfo, u32 perDataSize, u64 inputMemSize) const;
      54              :     virtual HcclResult PrepareSlicesL0(std::vector<std::vector<Slice>> &multRingsSlice, const OpParam &param,
      55              :         const SubCommInfo &level2CommInfo, const SubCommInfo &level1CommInfo, const SubCommInfo &level0CommInfo,
      56              :         u32 perDataSize, u64 inputMemSize);
      57              :     virtual HcclResult PrepareUserMemSlices(std::vector<std::vector<Slice>> &userMemSlices,
      58              :         const std::vector<std::vector<Slice>> &multRingsSlice, const OpParam &param, const SubCommInfo &level2CommInfo,
      59              :         const SubCommInfo &level1CommInfo, const SubCommInfo &level0CommInfo, u32 perDataSize, u64 inputMemSize);
      60              : 
      61              :     HcclResult DoubleRingAllGather(const std::string &tag, DeviceMem inputMem, DeviceMem outputMem, const u64 count,
      62              :         const HcclDataType dataType, const std::vector<std::vector<Slice>> multRingsSliceZero, Stream stream,
      63              :         s32 profStage, const u64 baseOffset, const HcomCollOpInfo *opInfo,
      64              :         const std::vector<std::vector<Slice>> multRingsUserMemSlice);
      65              : 
      66              :     HcclResult GetSubStreamInfoOnOneRing(const u32 ringIndex,
      67              :                                          std::vector<Stream>                       &subStreamsInOneRing,
      68              :                                          std::vector<std::shared_ptr<LocalNotify>> &mainSignalsInOneRing,
      69              :                                          std::vector<std::shared_ptr<LocalNotify>> &subSignalsInOneRing) override;
      70              :     // 成员变量
      71              :     u32 unitSize_ = 0;
      72              : 
      73              :     Stream mainStreamL2_; // Reserved slave stream for L2.
      74              :     std::vector<Stream> subStreams_;
      75              :     Stream mainStreamL1L0_; // Main stream for L1/L0.
      76              :     std::vector<Stream> ringSubStreams_;
      77              :     std::shared_ptr<LocalNotify> notifyL2ToL1L0A_{nullptr};
      78              :     std::shared_ptr<LocalNotify> notifyL1L0ToL2A_{nullptr};
      79              :     std::shared_ptr<LocalNotify> notifyL2ToL1L0B_{nullptr};
      80              :     std::shared_ptr<LocalNotify> notifyL1L0ToL2B_{nullptr};
      81              :     std::vector<std::shared_ptr<LocalNotify>> notifyRingMain_;
      82              :     std::vector<std::shared_ptr<LocalNotify>> notifyRingSub_;
      83              :     SubCommInfo level0CommInfo_{};           // L0层通信域信息
      84              :     SubCommInfo level1CommInfo_{};           // L1层通信域信息
      85              :     SubCommInfo level2CommInfo_{};           // L2层通信域信息
      86              : 
      87              :     DeviceMem cclInputAMem_;
      88              :     DeviceMem cclOutputAMem_;
      89              :     DeviceMem cclInputBMem_;
      90              :     DeviceMem cclOutputBMem_;
      91              :     u64 cclInputSizeHalved_{0};
      92              :     u64 cclOutputSizeHalved_{0};
      93              : 
      94              :     HcomCollOpInfo l0OpInfo_{};
      95              : };
      96              : 
      97              : } // namespace hccl
      98              : 
      99              : #endif