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              : #include <ios>
      12              : #include <iostream>
      13              : 
      14              : #include "log.h"
      15              : 
      16              : #include "ccu_rank_group.h"
      17              : #include "ccu_ctx_creator_registry.h"
      18              : #include "ccu_ins_group.h"
      19              : #include "ccu_context_reduce_scatter_mesh1d_2die.h"
      20              : #include "ccu_temp_reduce_scatter_mesh_1D_2die.h"
      21              : #include "alg_data_trans_wrapper.h"
      22              : 
      23              : namespace Hccl {
      24              : 
      25              : static CcuInstRegister<CcuContextReduceScatterMesh1D2Die>
      26              :     g_registrarReduceScatter(CcuInstType::CCU_REDUCE_SCATTER_MESH_1D_2DIE);
      27              : 
      28            0 : CcuTempReduceScatterMesh1D2Die::CcuTempReduceScatterMesh1D2Die(const RankId virtualRank, const u32 tempRankSize,
      29              :                                                                const std::vector<std::vector<RankId>> &tempVTopo,
      30            0 :                                                                const std::map<RankId, u32>            &tempVirtRankMap)
      31            0 :     : CcuAlgTemplateBase(virtualRank, tempRankSize, tempVTopo, tempVirtRankMap)
      32              : {
      33            0 : }
      34              : 
      35            0 : CcuTempReduceScatterMesh1D2Die::~CcuTempReduceScatterMesh1D2Die()
      36              : {
      37            0 : }
      38              : 
      39            0 : u32 CcuTempReduceScatterMesh1D2Die::CalcScratchMultiple(BufferType inBuffType, BufferType outBuffType)
      40              : {
      41              :     (void)inBuffType;
      42              :     (void)outBuffType;
      43            0 :     return 2;
      44              : }
      45              : 
      46            0 : void CcuTempReduceScatterMesh1D2Die::InitReduceInfo(const ReduceOp &reduceOp, const DataType &dataType)
      47              : {
      48            0 :     reduceOp_ = reduceOp;
      49            0 :     dataType_ = dataType;
      50            0 : }
      51              : 
      52            0 : HcclResult CcuTempReduceScatterMesh1D2Die::CalcRes(AlgTempResReq &tempResReq)
      53              : {
      54            0 :     tempResReq.queNum    = 1;
      55            0 :     tempResReq.streamNum = tempResReq.queNum + 1;
      56            0 :     HCCL_INFO("[CcuTempReduceScatterMesh1D2Die][CalcRes] tempResReq.queNum[%u]", tempResReq.queNum);
      57            0 :     CHK_RET(CalcResLinksMesh(myRank_, tempRankSize_, tempVTopo_, linkNumBtwPeers_, tempResReq));
      58            0 :     return HcclResult::HCCL_SUCCESS;
      59              : }
      60              : 
      61            0 : uint64_t CcuTempReduceScatterMesh1D2Die::GetMaxSliceSize() const
      62              : {
      63            0 :     u64 msSize     = 4 * 1024;
      64            0 :     u64 loopNum    = 64;
      65            0 :     u64 maxIterNum = 8192;
      66            0 :     return msSize * loopNum * maxIterNum; // 2G
      67              : }
      68              : 
      69            0 : HcclResult CcuTempReduceScatterMesh1D2Die::GenExtIns(const TempFuncs &tempFuncs, TemplateDataParams &tempAlgParams,
      70              :                                                      const ResLinks &tempLinks, std::vector<InsQuePtr> &tempInsQues)
      71              : {
      72              :     (void) tempFuncs;
      73            0 :     HCCL_INFO("[CcuTempReduceScatterMesh1D2Die] sliceSize[%llu], inputSliceStride[%llu], outputSliceStride[%llu], "
      74              :               "repeatNum[%llu], inputRepeatStride[%llu], outputRepeatStride[%llu], tailSize[%llu]",
      75              :               tempAlgParams.sliceSize, tempAlgParams.inputSliceStride, tempAlgParams.outputSliceStride,
      76              :               tempAlgParams.repeatNum, tempAlgParams.inputRepeatStride, tempAlgParams.outputRepeatStride,
      77              :               tempAlgParams.tailSize);
      78            0 :     CHK_PRT_RET(tempInsQues.empty(),
      79              :         HCCL_ERROR("[CcuTempReduceScatterMesh1D2Die] empty queue"), HcclResult::HCCL_E_INTERNAL);
      80            0 :     CHK_PTR_NULL(tempInsQues[0]);
      81            0 :     uint64_t inputAddr  = BufferTypeToAddr(tempAlgParams.buffInfo.inBuffType) + tempAlgParams.buffInfo.inBuffBaseOff;
      82            0 :     uint64_t outputAddr = BufferTypeToAddr(tempAlgParams.buffInfo.outBuffType) + tempAlgParams.buffInfo.outBuffBaseOff;
      83              :     uint64_t scratchAddr
      84            0 :         = BufferTypeToAddr(tempAlgParams.buffInfo.scratBuffType) + tempAlgParams.buffInfo.scratchBuffBaseOff;
      85            0 :     uint64_t inputSliceStride      = tempAlgParams.inputSliceStride;
      86            0 :     uint64_t outputSliceStride     = tempAlgParams.outputSliceStride;
      87            0 :     u64      sliceSize             = tempAlgParams.sliceSize;
      88              : 
      89              :     uint64_t token;
      90            0 :     CHK_RET(GetToken(op_, token));
      91            0 :     HCCL_INFO("[CcuTempReduceScatterMesh1D2Die] inputAddr[%llu], outputAddr[%llu], scratchAddr0[%llu], "
      92              :               "inputSliceStride[%llu], outputSliceStride[%llu]",
      93              :               inputAddr, outputAddr, scratchAddr, inputSliceStride, outputSliceStride);
      94              : 
      95            0 :     u32                                dieNum = 2;
      96            0 :     std::vector<std::vector<LinkData>> linksForDie(dieNum);
      97            0 :     std::vector<RankGroup>             rankGroupForDie(dieNum);
      98            0 :     for (auto &pair : tempLinks) {
      99            0 :         const RankId                 rank  = pair.first;
     100            0 :         const std::vector<LinkData> &links = pair.second;
     101            0 :         if (links.size() == 0) {
     102            0 :             continue;
     103              :         }
     104            0 :         const LinkData &link   = links[0];
     105            0 :         u32 dieIdx = link.GetLocalDieId();
     106            0 :         linksForDie[dieIdx].push_back(link);
     107            0 :         rankGroupForDie[dieIdx].AddRank(rank);
     108              :     }
     109            0 :     for (auto &rankGroup : rankGroupForDie) {
     110            0 :         rankGroup.AddRank(myRank_);
     111              :     }
     112            0 :     std::unique_ptr<CcuInsGroup> insGroupPtr = std::make_unique<CcuInsGroup>();
     113            0 :     for (u32 die = 0; die < dieNum; die++) {
     114            0 :         bool rmtReduceWithMyRank = linksForDie[die].size() > linksForDie[1 - die].size() ? false : true;
     115            0 :         CcuInstructionReduceScatterMesh1D2Die ccuInstruction;
     116            0 :         ccuInstruction.Init(tempVirtRankMap_[myRank_], op_, rmtReduceWithMyRank, tempVTopo_, inputAddr, outputAddr,
     117              :                             scratchAddr, sliceSize, token);
     118            0 :         ccuInstruction.SetLinks(linksForDie[die]);
     119            0 :         ccuInstruction.SetRankGroup(rankGroupForDie[die]);
     120            0 :         ccuInstruction.SetCntCkeNum(3);
     121            0 :         insGroupPtr->Append(std::move(std::make_unique<CcuInstructionReduceScatterMesh1D2Die>(ccuInstruction)));
     122            0 :     }
     123            0 :     tempInsQues[0]->Append(std::move(insGroupPtr));
     124              : 
     125            0 :     DataSlice srcSlice(tempAlgParams.buffInfo.scratBuffType, tempAlgParams.buffInfo.scratchBuffBaseOff, sliceSize);
     126            0 :     DataSlice dstSlice(tempAlgParams.buffInfo.outBuffType, tempAlgParams.buffInfo.outBuffBaseOff, sliceSize);
     127            0 :     LocalReduce(tempInsQues[0], srcSlice, dstSlice, op_.dataType, op_.reduceOp);
     128              : 
     129            0 :     HCCL_INFO("[CcuTempReduceScatterMesh1D2Die] Template Run for all steps Ends.");
     130            0 :     return HcclResult::HCCL_SUCCESS;
     131            0 : }
     132              : } // namespace Hccl