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 __AICPU_ALLREDUCE_H__
      12              : #define __AICPU_ALLREDUCE_H__
      13              : 
      14              : #include "aicpu_algorithm.h"
      15              : 
      16              : class AicpuAllreduce : public AicpuAlgorithm {
      17              : public:
      18           40 :     explicit AicpuAllreduce(AicpuComContext *ctx) : AicpuAlgorithm(ctx) {}
      19           39 :     ~AicpuAllreduce() override = default;
      20              : 
      21              :     HcclResult RunAlgorithm(HcclReduceOp opType, void *sendBuffer, void *recvBuffer, u64 dataCount,
      22              :         HcclDataType dataType, u64 strideLen = 0, AivAicpuOpParam *nextTask = nullptr) override;
      23              : 
      24              : private:
      25              :     int64_t RoundUpWithDivisor(u64 value, u64 divisor) const;
      26              :     HcclResult PrepareSlice(u64 dataCount, HcclDataType dataType, u32 sliceNum, std::vector<Slice>& dataSlice) const;
      27              :     void GetDataSizes16K(std::vector<u64>& dataSizes, u64 allDataSize) const;
      28              :     void RunAllReduceSlice(u8 *curOutputPtr, u8 *curInputPtr, u64 *sliceSize, u64 *dataSlice,
      29              :         HcclReduceOp opType, HcclDataType dataType) const;
      30              :     void RunAllReduceSliceWin2Win(u8 *curOutputPtr, u64 *sliceSize, u64 *dataSlice,
      31              :         HcclReduceOp opType, HcclDataType dataType) const;
      32              :     HcclResult RunAllReduceAlign(HcclReduceOp opType, void *sendBuffer, void *recvBuffer, u64 dataCount,
      33              :         HcclDataType dataType) const;
      34              :     HcclResult RunAllReduceAlignWin2Win(HcclReduceOp opType, void *recvBuffer,
      35              :         u64 dataCount, HcclDataType dataType) const;
      36              :     HcclResult RunAllReduce(HcclReduceOp opType, void *sendBuffer, void *recvBuffer, u64 dataCount,
      37              :         HcclDataType dataType) const;
      38              :     HcclResult RunAllReduceOneShot4Stream(HcclReduceOp opType, void *sendBuffer, void *recvBuffer, u64 dataSize,
      39              :         HcclDataType dataType) const;
      40              :     HcclResult RunAllReduceReduceBcast(HcclReduceOp opType, void *sendBuffer, void *recvBuffer, u64 dataSize,
      41              :         HcclDataType dataType) const;
      42              :     HcclResult RunReduceBcastOnMainSq(u32 mainRankId, u32 maxStreamNum, u32 /* startRank */, u32 /* end_rank */,
      43              :         void *sendBuffer, void *recvBuffer, u64 dataSize, HcclDataType dataType) const;
      44              :     HcclResult RunReduceBcastOnOtherSq(HcclReduceOp opType, u32 mainRankId, u32 maxStreamNum,
      45              :         void *sendBuffer, void *recvBuffer, u64 dataSize, HcclDataType dataType) const;
      46              :     HcclResult RunAllReduceOneShot1Stream(HcclReduceOp opType, void *sendBuffer, void *recvBuffer,
      47              :         u64 dataSize, HcclDataType dataType) const;
      48              :     HcclResult RunAllReduceTwoShot1Stream(HcclReduceOp opType, void *sendBuffer, void *recvBuffer,
      49              :         u64 dataCount, HcclDataType dataType) const;
      50              :     u32 GetHdPeer(const u32 hdRound, const u32 curRank) const;
      51              :     HcclResult RunAllReduceOneshotHD(HcclReduceOp opType, void *sendBuffer, void *recvBuffer,
      52              :         u64 dataSize, HcclDataType dataType) const;
      53              :     HcclResult RunAllReduceRing(HcclReduceOp opType, void *sendBuffer, void *recvBuffer,
      54              :         u64 dataCount, HcclDataType dataType) const;
      55              :     u64 AlignWith(u64 oriValue, u64 alignValue) const;
      56              :     HcclResult GetBurstDataCounts(u64 windowSize, u64 dataCount, std::vector<u64>& burstDataCounts) const;
      57              :     std::vector<std::vector<u32>> GetRingOrders() const;
      58              :     HcclResult reorderRingSlice(const std::vector<u32> &ringOrder, const std::vector<Slice> &ringSlices,
      59              :         std::vector<Slice> &orderedRingSlices) const;
      60              :     HcclResult PrepareRingSlice(
      61              :         const std::vector<std::vector<u32>> &ringOrders, u64 dataCount,
      62              :         HcclDataType dataType,
      63              :         std::vector<std::vector<Slice>> &orderedAllRingSlice) const;
      64              :     HcclResult RingIPCPreSync(const u32 stream, const u32 prevRank, const u32 nextRank) const;
      65              :     HcclResult RingIPCPostSync(const u32 stream, const u32 prevRank, const u32 nextRank) const;
      66              :     HcclResult GetPrevRankList(const std::vector<u32> &ringOrder, std::vector<u32> &previousRankList) const;
      67              :     size_t FindNextRank(const std::vector<u32> &previousRankList, const u32 localRank) const;
      68              :     Slice* GetNextRingSlice(const std::vector<u32> &previousRankList,
      69              :         std::vector<Slice> &orderedRingSlices, u32 &curSliceIdx) const;
      70              :     HcclResult RunAllReduceRingSingleBurst(
      71              :         HcclReduceOp opType, void *sendBuffer, void *recvBuffer, u64 dataCount,
      72              :         HcclDataType dataType, std::vector<std::vector<u32>> &ringOrders) const;
      73              :     HcclResult RunAllReduceRingAlg(HcclReduceOp opType, void *sendBuffer, void *recvBuffer,
      74              :         std::vector<Slice> &orderedRingSlices, std::vector<u32> &ringOrder, HcclDataType dataType) const;
      75              : };
      76              : 
      77              : #endif