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 "hccl_aiv_utils.h"
      12              : #include "aiv_ins.h"
      13              : #include "aiv_temp_all_reduce_mesh_1D_oneshot.h"
      14              : #include "executor_utils.h"
      15              : 
      16              : namespace Hccl {
      17              : 
      18            0 : AivTempAllReduceMesh1DOneShot::AivTempAllReduceMesh1DOneShot(
      19              :     const RankId virtualRank, const u32 tempRankSize, const std::vector<std::vector<RankId>>& tempVTopo,
      20            0 :     const std::map<RankId, u32>& tempVirtRankMap)
      21            0 :     : AivAlgTemplateBase(virtualRank, tempRankSize, tempVTopo, tempVirtRankMap)
      22            0 : {}
      23              : 
      24            0 : AivTempAllReduceMesh1DOneShot::~AivTempAllReduceMesh1DOneShot() {}
      25              : 
      26            0 : HcclResult AivTempAllReduceMesh1DOneShot::CalcRes(AlgTempResReq& tempResReq)
      27              : {
      28            0 :     tempResReq.queNum = 1;
      29            0 :     tempResReq.streamNum = tempResReq.queNum;
      30            0 :     HCCL_INFO("[CalcRes] tempResReq.queNum[%u]", tempResReq.queNum);
      31            0 :     CHK_RET(CalcResLinksMesh(myRank_, tempRankSize_, tempVTopo_, linkNumBtwPeers_, tempResReq));
      32            0 :     return HcclResult::HCCL_SUCCESS;
      33              : }
      34              : 
      35            0 : u32 AivTempAllReduceMesh1DOneShot::CalcScratchMultiple(BufferType inBuffType, BufferType outBuffType)
      36              : {
      37              :     (void)inBuffType;
      38              :     (void)outBuffType;
      39              : 
      40            0 :     return tempRankSize_;
      41              : }
      42              : 
      43            0 : HcclResult AivTempAllReduceMesh1DOneShot::CalNumBlocks(u32& numBlocks, u64 dataSize, u32 numBlocksLimit)
      44              : {
      45              :     (void)dataSize;
      46            0 :     if (numBlocksLimit >= (tempRankSize_ + 1)) {
      47            0 :         numBlocks = tempRankSize_ + 1;
      48              :     } else {
      49              :         // 如果要用更少的核心可以在这里折算，比如rankSize/2个核心
      50            0 :         numBlocks = numBlocksLimit;
      51              :     }
      52            0 :     return HcclResult::HCCL_SUCCESS;
      53              : }
      54              : 
      55            0 : HcclResult AivTempAllReduceMesh1DOneShot::GenExtIns(
      56              :     const TempFuncs& tempFuncs, const TemplateDataParams& templateDataParams, const ResLinks& tempLinks,
      57              :     std::vector<InsQuePtr>& tempInsQues)
      58              : {
      59            0 :     HCCL_INFO("[AivTempAllReduceMesh1DOneShot] GenExtIns start");
      60            0 :     CHK_PRT_RET(
      61              :         tempInsQues.empty(), HCCL_ERROR("[AivTempAllReduceMesh1DOneShot] empty queue"), HcclResult::HCCL_E_INTERNAL);
      62            0 :     CHK_PTR_NULL(tempInsQues[0]);
      63            0 :     std::vector<LinkData> allLinks;
      64            0 :     for (auto iter = tempLinks.begin(); iter != tempLinks.end(); ++iter) {
      65            0 :         allLinks.emplace_back(iter->second.at(0));
      66              :     }
      67              : 
      68            0 :     IncSliceId(); // 自动增长sliceId，传入aivTag
      69              : 
      70            0 :     AivOpArgs aivScatterArgs;
      71            0 :     aivScatterArgs.cmdType = HcclCMDType::HCCL_CMD_ALLREDUCE;
      72            0 :     aivScatterArgs.input = templateDataParams.buffInfo.inBuffBaseOff;
      73            0 :     aivScatterArgs.output = templateDataParams.buffInfo.outBuffBaseOff;
      74            0 :     aivScatterArgs.rank = myRank_;
      75            0 :     aivScatterArgs.rankSize = tempRankSize_;
      76            0 :     aivScatterArgs.count = templateDataParams.sliceSize / DataTypeSizeGet(dataType_);
      77            0 :     aivScatterArgs.dataType = dataType_;
      78            0 :     aivScatterArgs.op = reduceOp_;
      79            0 :     aivScatterArgs.root = root_;
      80            0 :     aivScatterArgs.aivTag = sliceId_; // 传入aivTag，Lauch时重新组装为aivTag
      81            0 :     aivScatterArgs.isOpBase = (tempFuncs.opMode == OpMode::OPBASE);
      82            0 :     aivScatterArgs.xRankSize = tempVTopo_[0].size();
      83            0 :     aivScatterArgs.yRankSize = 0;
      84            0 :     aivScatterArgs.zRankSize = 0;
      85            0 :     CalNumBlocks(aivScatterArgs.numBlocks, templateDataParams.sliceSize, op_.numBlocksLimit);
      86            0 :     HCCL_INFO("[AivTempAllReduceMesh1DOneShot] Actually use core num[%u]", aivScatterArgs.numBlocks);
      87            0 :     for (u32 i = 0; i < tempVTopo_[0].size(); i++) {
      88            0 :         aivScatterArgs.topo_[i] = tempVTopo_[0][i];
      89              :     }
      90            0 :     if (tempVTopo_.size() > 1) {
      91            0 :         aivScatterArgs.yRankSize = tempVTopo_[1].size();
      92            0 :         for (u32 i = 0; i < tempVTopo_[1].size(); i++) {
      93            0 :             aivScatterArgs.topo_[TOPO_LEN_Y_OFFSET + i] = tempVTopo_[1][i];
      94              :         }
      95              :     }
      96            0 :     if (tempVTopo_.size() == MAX_DIM_NUM) {
      97            0 :         aivScatterArgs.zRankSize = tempVTopo_[MAX_DIM_NUM - 1].size();
      98            0 :         for (u32 i = 0; i < tempVTopo_[MAX_DIM_NUM - 1].size(); i++) {
      99            0 :             aivScatterArgs.topo_[TOPO_LEN_Z_OFFSET + i] = tempVTopo_[MAX_DIM_NUM - 1][i];
     100              :         }
     101              :     }
     102              : 
     103            0 :     aivScatterArgs.inputSliceStride = templateDataParams.inputSliceStride;
     104            0 :     aivScatterArgs.outputSliceStride = templateDataParams.outputSliceStride;
     105            0 :     aivScatterArgs.repeatNum = templateDataParams.repeatNum;
     106            0 :     aivScatterArgs.inputRepeatStride = templateDataParams.inputRepeatStride;
     107            0 :     aivScatterArgs.outputRepeatStride = templateDataParams.outputRepeatStride;
     108              : 
     109            0 :     std::unique_ptr<Instruction> aivInsScatterMesh1D = std::make_unique<AivInstruction>(allLinks, aivScatterArgs);
     110              : 
     111            0 :     tempInsQues[0]->Append(std::move(aivInsScatterMesh1D));
     112            0 :     HCCL_INFO("[AivTempAllReduceMesh1DOneShot] GenExtIns finished");
     113              : 
     114            0 :     return HcclResult::HCCL_SUCCESS;
     115            0 : }
     116              : 
     117              : } // namespace Hccl