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              : #include "log.h"
      12              : 
      13              : #include "alg_data_trans_wrapper.h"
      14              : #include "ins_temp_reduce_aicpu_reduce.h"
      15              : 
      16              : namespace Hccl {
      17            0 : InsTempReduceAicpuReduce::InsTempReduceAicpuReduce(
      18              :     const RankId virtualRank, const u32 tempRankSize, const std::vector<std::vector<RankId>>& tempVTopo,
      19            0 :     const std::map<RankId, u32>& tempVirtRankMap)
      20            0 :     : InsAlgTemplateBase(virtualRank, tempRankSize, tempVTopo, tempVirtRankMap)
      21            0 : {}
      22              : 
      23            0 : InsTempReduceAicpuReduce::~InsTempReduceAicpuReduce() {}
      24              : 
      25            0 : HcclResult InsTempReduceAicpuReduce::CalcRes(AlgTempResReq& tempResReq)
      26              : {
      27            0 :     tempResReq.queNum = tempVTopo_[0].size();
      28            0 :     tempResReq.streamNum = tempResReq.queNum;
      29            0 :     tempResReq.queNotifys = CreateMasterSlaveQueNotifiesRequest(tempResReq.queNum);
      30            0 :     QId centerQ = 0;
      31            0 :     tempResReq.localWaitGroupCntNotify.emplace_back(centerQ, 0);
      32            0 :     tempResReq.localBcastPostCntNotify.emplace_back(centerQ, 0);
      33            0 :     CHK_RET(CalcResLinksMesh(myRank_, tempRankSize_, tempVTopo_, linkNumBtwPeers_, tempResReq));
      34            0 :     HCCL_DEBUG(
      35              :         "[InsTempReduceAicpuReduce]CalcRes: queNum[%u], myRank[%d], tempRankSize[%u]", tempResReq.queNum, myRank_,
      36              :         tempRankSize_);
      37            0 :     return HcclResult::HCCL_SUCCESS;
      38              : }
      39              : 
      40            0 : u32 InsTempReduceAicpuReduce::CalcScratchMultiple(BufferType inBuffType, BufferType outBuffType) const
      41              : {
      42              :     (void)inBuffType;
      43              :     (void)outBuffType;
      44            0 :     return tempRankSize_;
      45              : }
      46              : 
      47            0 : HcclResult InsTempReduceAicpuReduce::RunGatherMesh(
      48              :     const TempFuncs& tempFuncs, const TemplateDataParams& templateDataParams, const ResLinks& tempLinks,
      49              :     std::vector<InsQuePtr>& tempInsQues)
      50              : {
      51              :     (void)tempFuncs;
      52            0 :     if (u32(myRank_) == root_) {
      53              :         // 本端rank数据从本端input -> 本端scratch
      54            0 :         CHK_RET(PreSyncInterQueues(tempInsQues));
      55            0 :         u64 srcOffset = templateDataParams.buffInfo.inBuffBaseOff;
      56            0 :         u64 srcSize = templateDataParams.sliceSize;
      57            0 :         u64 dstOffset = templateDataParams.sliceSize * root_;
      58            0 :         DataSlice srcSlice = DataSlice(BufferType::INPUT, srcOffset, srcSize);
      59            0 :         DataSlice dstSlice = DataSlice(BufferType::SCRATCH, dstOffset, srcSize);
      60            0 :         std::unique_ptr<Instruction> insLocalCopy = std::make_unique<InsLocalCopy>(srcSlice, dstSlice);
      61            0 :         tempInsQues[0]->Append(std::move(insLocalCopy));
      62              :         // recv from other rank
      63              :         u32 myAlgRank;
      64            0 :         CHK_RET(GetAlgRank(myRank_, tempVTopo_[0], myAlgRank));
      65            0 :         for (u32 queIdx = 1; queIdx < tempVTopo_[0].size(); queIdx++) {
      66            0 :             RankId neighborRank = tempVTopo_[0][(myAlgRank + queIdx) % tempRankSize_];
      67            0 :             LinkData neighborLinkData = tempLinks.at(neighborRank)[0];
      68            0 :             std::vector<DataSlice> srcSlices;
      69            0 :             std::vector<DataSlice> dstSlices;
      70            0 :             srcSlices.emplace_back(BufferType::INPUT, srcOffset, srcSize);
      71            0 :             dstSlices.emplace_back(BufferType::SCRATCH, srcSize * neighborRank, srcSize);
      72            0 :             SlicesList rxSlicesList(srcSlices, dstSlices);
      73            0 :             DataInfo recvData(neighborLinkData, rxSlicesList);
      74            0 :             CHK_PRT_RET(
      75              :                 Recv(recvData, tempInsQues[queIdx], 0, true, dmaMode_),
      76              :                 HCCL_ERROR("[InsTempReduceAicpuReduce] BatchSend failed"), HcclResult::HCCL_E_INTERNAL);
      77            0 :         }
      78            0 :         CHK_RET(PostSyncInterQueues(tempInsQues));
      79            0 :     } else {
      80              :         // send to root rank
      81            0 :         u64 srcOffset = templateDataParams.buffInfo.inBuffBaseOff;
      82            0 :         u64 srcSize = templateDataParams.sliceSize;
      83            0 :         LinkData linkSend = tempLinks.at(root_)[0];
      84            0 :         std::vector<DataSlice> srcSlices;
      85            0 :         std::vector<DataSlice> dstSlices;
      86            0 :         srcSlices.emplace_back(BufferType::INPUT, srcOffset, srcSize);
      87            0 :         dstSlices.emplace_back(BufferType::SCRATCH, srcSize * u32(myRank_), srcSize);
      88            0 :         SlicesList txSlicesList(srcSlices, dstSlices);
      89            0 :         DataInfo sendData(linkSend, txSlicesList);
      90            0 :         CHK_PRT_RET(
      91              :             Send(sendData, tempInsQues[0], 0, true, dmaMode_),
      92              :             HCCL_ERROR("[InsTempReduceAicpuReduce] BatchSend failed"), HcclResult::HCCL_E_INTERNAL);
      93            0 :     }
      94            0 :     return HCCL_SUCCESS;
      95              : }
      96              : 
      97            0 : HcclResult InsTempReduceAicpuReduce::RunAicpuLocalReduce(
      98              :     const TemplateDataParams& templateDataParams, std::vector<InsQuePtr>& tempInsQues)
      99              : {
     100            0 :     DataSlice dataSlice = DataSlice(BufferType::SCRATCH, 0, templateDataParams.sliceSize);
     101            0 :     for (u32 rankId = 1; rankId < tempRankSize_; rankId++) {
     102              :         DataSlice reduceSlice
     103            0 :             = DataSlice(BufferType::SCRATCH, templateDataParams.sliceSize * rankId, templateDataParams.sliceSize);
     104            0 :         AicpuReduce(tempInsQues[0], reduceSlice, dataSlice, dataType_, redOp_);
     105              :     }
     106              :     DataSlice outputSlice
     107            0 :         = DataSlice(BufferType::OUTPUT, templateDataParams.buffInfo.inBuffBaseOff, templateDataParams.sliceSize);
     108            0 :     LocalCopy(tempInsQues[0], dataSlice, outputSlice);
     109            0 :     return HCCL_SUCCESS;
     110              : }
     111              : 
     112            0 : HcclResult InsTempReduceAicpuReduce::GenExtIns(
     113              :     const TempFuncs& tempFuncs, const TemplateDataParams& templateDataParams, const ResLinks& tempLinks,
     114              :     std::vector<InsQuePtr>& tempInsQues)
     115              : {
     116            0 :     HCCL_INFO("[InsTempReduceAicpuReduce] Run start");
     117              : 
     118            0 :     dmaMode_ = DmaMode::PUT;
     119            0 :     if (IsPcieLink(tempLinks)) {
     120            0 :         dmaMode_ = DmaMode::GET;
     121              :     }
     122            0 :     if (tempVTopo_[0].size() == 1) {
     123            0 :         return HcclResult::HCCL_SUCCESS;
     124              :     }
     125            0 :     opMode_ = tempFuncs.opMode;
     126            0 :     queNum_ = tempVTopo_[0].size();
     127            0 :     CHK_PRT_RET(
     128              :         queNum_ != tempInsQues.size(),
     129              :         HCCL_ERROR("[CollAlgFactory] [InsTempReduceAicpuReduce] Rank [%d], requiredQue Error.", myRank_),
     130              :         HcclResult::HCCL_E_INTERNAL);
     131              : 
     132            0 :     RunGatherMesh(tempFuncs, templateDataParams, tempLinks, tempInsQues);
     133            0 :     StreamSync(tempInsQues);
     134            0 :     RunAicpuLocalReduce(templateDataParams, tempInsQues);
     135            0 :     HCCL_INFO("[InsTempReduceAicpuReduce] Run finished");
     136            0 :     return HCCL_SUCCESS;
     137              : }
     138              : 
     139              : } // namespace Hccl