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 "coll_reduce_scatter_semi_ring_executor.h"
      12              : 
      13              : namespace hccl {
      14              : 
      15            0 : CollReduceScatterSemiRingExecutor::CollReduceScatterSemiRingExecutor(const HcclDispatcher dispatcher,
      16            0 :     std::unique_ptr<TopoMatcher> &topoMatcher)
      17            0 :     : CollReduceScatterRingFor91093Executor(dispatcher, topoMatcher)
      18              : {
      19            0 : }
      20              : 
      21            0 : HcclResult CollReduceScatterSemiRingExecutor::CalcNotifyNum(u32 streamNum, u32 &notifyNum)
      22              : {
      23              :     // notify数量是从流的两倍 + 新增带notifyId的notify资源
      24            0 :     notifyNum = 2U * streamNum + (topoAttr_.deviceNumPerAggregation + 4U);
      25            0 :     HCCL_INFO("[CollReduceScatterSemiRingExecutor][CalcNotifyNum]tag[%s] notifyNum_ is [%u]", tag_.c_str(), notifyNum);
      26            0 :     return HCCL_SUCCESS;
      27              : }
      28              : 
      29            0 : HcclResult CollReduceScatterSemiRingExecutor::CalcStreamNum(u32& streamNum)
      30              : {
      31            0 :     u32 totalStreamNum = LEVEL0_PLANE_NUM_IN_NPRING_DOUBLE + 1U;
      32              : 
      33            0 :     streamNum = totalStreamNum - 1U;
      34            0 :     HCCL_INFO("[CollReduceScatterSemiRingExecutor][CalcStreamNum] tag[%s] streamNum_[%u]",
      35              :         tag_.c_str(), streamNum);
      36            0 :     return HCCL_SUCCESS;
      37              : }
      38              : 
      39            0 : void CollReduceScatterSemiRingExecutor::ParseParam(const OpParam& param)
      40              : {
      41            0 :     tag_ = param.tag;
      42              : 
      43              :     // 是否需要scratch memory
      44            0 :     if (isSupportSDMAReduce_ && topoAttr_.serverNum == 1) {
      45            0 :         scratchMemFlag_ = false;
      46              :     } else {
      47            0 :         scratchMemFlag_ = true;
      48              :     }
      49              : 
      50              :     // 记录图模式总数据量
      51            0 :     const HcclDataType dataType = param.GetDataType();
      52            0 :     totalSize_ = CalcTotalCount(param) * SIZE_TABLE[dataType];
      53            0 :     aicpuUnfoldMode_ = param.aicpuUnfoldMode;
      54            0 : }
      55              : 
      56            0 : HcclResult CollReduceScatterSemiRingExecutor::CalcLevel0CommInfo(TransportMemType inputType,
      57              :     TransportMemType outputType,
      58              :     std::vector<LevelNSubCommTransport>& opTransport)
      59              : {
      60            0 :     CommParaInfo commParaLevel0(COMM_LEVEL0, CommType::COMM_TAG_MESH);
      61            0 :     CHK_RET(CalcCommPlaneInfo(tag_, commParaLevel0, opTransport[COMM_LEVEL0], inputType, outputType));
      62              : 
      63            0 :     LevelNSubCommTransport &commTransportLevel0 = opTransport[COMM_LEVEL0];
      64            0 :     for (u32 subCommIndex = 0; subCommIndex < commTransportLevel0.size(); subCommIndex++) {
      65            0 :         for (auto &transportRequest : commTransportLevel0[subCommIndex].transportRequests) {
      66            0 :             transportRequest.notifyNum = topoAttr_.deviceNumPerAggregation + 4U; //只传递额外的notify个数
      67            0 :             HCCL_INFO("[CollReduceScatterSemiRingExecutor][CalcLevel0CommInfo] set extral notifyNum[%u]",
      68              :                 transportRequest.notifyNum);
      69              :         }
      70              :     }
      71            0 :     return HCCL_SUCCESS;
      72            0 : }
      73              : 
      74            0 : HcclResult CollReduceScatterSemiRingExecutor::DoubleRingMidCountReduceScatter(
      75              :     const std::string &tag, DeviceMem inputMem, DeviceMem outputMem,
      76              :     const u64 count, const HcclDataType dataType, const HcclReduceOp reductionOp,
      77              :     const std::vector<std::vector<Slice> > multRingsSliceZero, Stream stream, s32 profStage,
      78              :     const u64 baseOffset, const HcomCollOpInfo *opInfo,
      79              :     const std::vector<std::vector<Slice>> multRingsUserMemSlice, const bool retryEnable)
      80              : {
      81              :     (void) tag;
      82              :     (void) baseOffset;
      83              :     (void) opInfo;
      84              :     (void) multRingsSliceZero;
      85              :     (void) retryEnable;
      86            0 :     HCCL_CONFIG_INFO(HCCL_ALG,
      87              :         "[CollReduceScatterSemiRingExecutor] DoubleRingMidCountReduceScatter starts.");
      88              :     
      89            0 :     CHK_RET(CheckCommSize(COMM_LEVEL0, COMM_INDEX_0 + 1));
      90            0 :     SubCommInfo level0CommInfo = GetSubCommInfo(COMM_LEVEL0, COMM_INDEX_0);
      91              : 
      92              :     //此处计算reduceAttr计算outputmem使用scratchmem
      93            0 :     u64 reduceAttr = GetReduceAttr(inputMem, outputMem, dataType, reductionOp);
      94              :     // 执行
      95            0 :     std::unique_ptr<AlgTemplateBase> executor = AlgTemplateRegistry::Instance().GetAlgTemplate(
      96            0 :         TemplateType::TEMPLATE_REDUCESCATTER_UNIFIED_MARCH, dispatcher_);
      97            0 :     HCCL_CONFIG_INFO(HCCL_ALG, "[%s] Run TEMPLATE_REDUCESCATTER_UNIFIED_MARCH in COMM_LEVEL0", __func__);
      98            0 :     CHK_SMART_PTR_NULL(executor);
      99              : 
     100            0 :     CHK_RET(executor->Prepare(stream, level0CommInfo,
     101              :         algResResp_->paramInputMem, algResResp_->paramOutputMem, inputMem,
     102              :         outputMem, count, algResResp_->slaveStreams, algResResp_->notifiesMain,
     103              :         algResResp_->notifiesAux, dataType, reductionOp, multRingsUserMemSlice, reduceAttr));
     104              : 
     105            0 :     HcclResult ret = executor->RegisterProfiler(
     106              :         ((COMM_INDEX_0 + 1) << PROF_RINGINDEX_OFFSET_OF_PLANEID) +
     107            0 :         (level0CommInfo.localRankSize << PROF_RANKSIZE_OFFSET_OF_PLANEID) + level0CommInfo.localRank,
     108              :         profStage, HCCL_EXEC_STEP_NOT_SET, stream);
     109            0 :     CHK_PRT_RET(ret != HCCL_SUCCESS,
     110              :         HCCL_ERROR("[CollReduceScatterSemiRingExecutor][DoubleRingMidCountReduceScatter]"\
     111              :             "Double ring ReduceScatter failed,return[%d]", ret), ret);
     112              : 
     113            0 :     CHK_RET(executor->RunAsync());
     114              : 
     115            0 :     HCCL_INFO("[CollReduceScatterSemiRingExecutor] ReduceScatter double ring level1 run success");
     116            0 :     return ret;
     117            0 : }
     118              : 
     119            0 : HcclResult CollReduceScatterSemiRingExecutor::RunIntraSeverReduceScatter(
     120              :     const std::string &tag, DeviceMem &inputMem, DeviceMem &outputMem,
     121              :     const u64 count, const HcclDataType &dataType, const HcclReduceOp &reductionOp,
     122              :     const std::vector<std::vector<Slice>> &multRingsSliceZero, const Stream &stream, s32 profStage,
     123              :     const u64 baseOffset, const HcomCollOpInfo *opInfo,
     124              :     const std::vector<std::vector<Slice>> &multRingsUserMemSlice, const bool retryEnable)
     125              : {
     126            0 :     CHK_RET(DoubleRingMidCountReduceScatter(tag, inputMem, outputMem, count, dataType, reductionOp,
     127              :         multRingsSliceZero, stream, profStage, baseOffset, opInfo, multRingsUserMemSlice, retryEnable));
     128            0 :     return HCCL_SUCCESS;
     129              : }
     130              : 
     131              : REGISTER_EXEC("ReduceScatterSemiRingExecutor", ReduceScatterDoubleRingMidCount,
     132              :     CollReduceScatterSemiRingExecutor);
     133              : 
     134              : } // namespace hccl