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 <ios>
      12              : #include <iostream>
      13              : 
      14              : #include "log.h"
      15              : 
      16              : #include "ccu_instruction_scatter_mesh1d.h"
      17              : #include "ccu_rank_group.h"
      18              : #include "ccu_ctx_creator_registry.h"
      19              : #include "ccu_context_scatter_mesh1d.h"
      20              : #include "ccu_temp_scatter_mesh_1D.h"
      21              : #include "dev_mode.h"
      22              : 
      23              : namespace Hccl {
      24              : 
      25              : static CcuInstRegister<CcuContextScatterMesh1D> g_registrarScatter(CcuInstType::CCU_SCATTER_MESH_1D_DIRECT);
      26              : 
      27            0 : CcuTempScatterMesh1D::CcuTempScatterMesh1D(const RankId virtualRank, const u32 tempRankSize,
      28              :                                            const std::vector<std::vector<RankId>> &tempVTopo,
      29            0 :                                            const std::map<RankId, u32>            &tempVirtRankMap)
      30            0 :     : CcuAlgTemplateBase(virtualRank, tempRankSize, tempVTopo, tempVirtRankMap)
      31              : {
      32            0 : }
      33              : 
      34            0 : CcuTempScatterMesh1D::~CcuTempScatterMesh1D()
      35              : {
      36            0 : }
      37              : 
      38            0 : uint64_t CcuTempScatterMesh1D::GetExpandedMode() const
      39              : {
      40            0 :     return DeviceMode::CCU;
      41              : }
      42              : 
      43            0 : uint64_t CcuTempScatterMesh1D::GetMaxSliceSize() const
      44              : {
      45            0 :     return UB_MAX_DATA_SIZE;
      46              : }
      47              : 
      48            0 : HcclResult CcuTempScatterMesh1D::CalcRes(AlgTempResReq &tempResReq)
      49              : {
      50            0 :     tempResReq.queNum    = 1;
      51            0 :     tempResReq.streamNum = tempResReq.queNum;
      52            0 :     HCCL_INFO("[CalcRes] tempResReq.queNum[%u]", tempResReq.queNum);
      53            0 :     CHK_RET(CalcResLinksMesh(myRank_, tempRankSize_, tempVTopo_, linkNumBtwPeers_, tempResReq));
      54            0 :     return HcclResult::HCCL_SUCCESS;
      55              : }
      56              : 
      57            0 : HcclResult CcuTempScatterMesh1D::GenExtIns(const TempFuncs &tempFuncs, const TemplateDataParams &templateDataParams,
      58              :                                            const ResLinks &tempLinks, std::vector<InsQuePtr> &tempInsQues)
      59              : {
      60            0 :     CHK_PRT_RET(tempInsQues.empty(),
      61              :         HCCL_ERROR("[CcuTempScatterMesh1D] empty queue"), HcclResult::HCCL_E_INTERNAL);
      62            0 :     CHK_PTR_NULL(tempInsQues[0]);
      63            0 :     HCCL_INFO("[CcuTempScatterMesh1D] Run.");
      64            0 :     opMode_   = tempFuncs.opMode;
      65            0 :     buffInfo_ = templateDataParams.buffInfo;
      66            0 :     CcuInstructionScatterMesh1D ccuIns;
      67            0 :     uint32_t                    virtRankId = tempVirtRankMap_[myRank_];
      68              : 
      69            0 :     const CollAlgOperator                  &op        = op_;
      70            0 :     const std::vector<std::vector<RankId>> &tempVTopo = tempVTopo_;
      71            0 :     uint64_t                                rootId    = tempVirtRankMap_[rootId_];
      72              :     uint64_t                                inputAddr
      73            0 :         = BufferTypeToAddr(templateDataParams.buffInfo.inBuffType) + templateDataParams.buffInfo.inBuffBaseOff;
      74              :     uint64_t outputAddr
      75            0 :         = BufferTypeToAddr(templateDataParams.buffInfo.outBuffType) + templateDataParams.buffInfo.outBuffBaseOff;
      76              :     uint64_t token;
      77            0 :     CHK_RET(GetToken(op_, token));
      78            0 :     uint64_t inputSliceStride   = templateDataParams.inputSliceStride;
      79            0 :     uint64_t outputSliceStride  = templateDataParams.outputSliceStride;
      80            0 :     uint64_t inputRepeatStride  = templateDataParams.inputRepeatStride;
      81            0 :     uint64_t outputRepeatStride = templateDataParams.outputRepeatStride;
      82            0 :     uint64_t normalSliceSize    = templateDataParams.sliceSize;
      83            0 :     uint64_t lastSliceSize      = templateDataParams.tailSize;
      84            0 :     uint64_t repeatNum          = templateDataParams.repeatNum;
      85            0 :     uint64_t repeatNumVar = UINT64_MAX - repeatNum; // 在context中让repeatNumVar累加到UINT64_MAX结束ccu while循环
      86              : 
      87            0 :     ccuIns.Init(virtRankId, rootId, op, tempVTopo, inputAddr, outputAddr, token, inputSliceStride, outputSliceStride,
      88              :                 inputRepeatStride, outputRepeatStride, normalSliceSize, lastSliceSize, repeatNumVar);
      89            0 :     HCCL_INFO("[CcuTempScatterMesh1D] Run Init: virtRankId[%u], rankId[%d], inputAddr[%llu], "
      90              :                "outputAddr[%llu], inputSliceStride[%llu], outputSliceStride[%llu], "
      91              :                "inputRepeatStride[%llu], outputRepeatStride[%llu], normalSliceSize[%llu], lastSliceSize[%llu], "
      92              :                "repeatNumVar[%llu]",
      93              :                virtRankId, myRank_, inputAddr, outputAddr, inputSliceStride, outputSliceStride, inputRepeatStride,
      94              :                outputRepeatStride, normalSliceSize, lastSliceSize, repeatNumVar);
      95              : 
      96            0 :     if (normalSliceSize == 0) {
      97            0 :         HCCL_INFO("[CcuTempScatterMesh1D] DataCount == 0, Template Run Ends.");
      98            0 :         return HCCL_SUCCESS;
      99              :     }
     100              : 
     101            0 :     std::vector<LinkData> links;
     102            0 :     for (auto &pair : tempLinks) {
     103            0 :         if (pair.second.empty()) {
     104            0 :             continue;
     105              :         }
     106            0 :         links.push_back(pair.second[0]);
     107              :     }
     108            0 :     HCCL_INFO("[CcuTempScatterMesh1D] links.size[%zu]", links.size());
     109            0 :     ccuIns.SetLinks(links);
     110              : 
     111            0 :     RankGroup rankGroup;
     112            0 :     for (auto &peer : tempVTopo_[0]) {
     113            0 :         rankGroup.AddRank(peer);
     114              :     }
     115            0 :     u32 cntCkeNum = 3;
     116            0 :     ccuIns.SetCntCkeNum(cntCkeNum);
     117            0 :     ccuIns.SetRankGroup(rankGroup);
     118            0 :     HCCL_INFO("CcuInstructionScatterMesh1D is [%s]", ccuIns.Describe().c_str());
     119            0 :     tempInsQues[0]->Append(std::move(std::make_unique<CcuInstructionScatterMesh1D>(ccuIns)));
     120              : 
     121            0 :     return HcclResult::HCCL_SUCCESS;
     122            0 : }
     123              : 
     124            0 : HcclResult CcuTempScatterMesh1D::GenExtIns(const RankGraph *rankGraph, const TemplateInfo &tmpInfo,
     125              :                                            const std::vector<InsQuePtr> &tempInsQues) const
     126              : {
     127              :     (void)rankGraph;
     128              :     (void)tmpInfo;
     129              :     (void)tempInsQues;
     130            0 :     return HcclResult::HCCL_SUCCESS;
     131              : }
     132              : 
     133              : } // namespace Hccl