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 "calc_nhr_v1_transport_req.h"
      12              : #include "nonuniform_hierarchical_ring_v1_base_pub.h"
      13              : 
      14              : namespace hccl {
      15            0 : CalcNHRV1TransportReq::CalcNHRV1TransportReq(std::vector<std::vector<u32>> &subCommPlaneVector,
      16            0 :     std::vector<bool> &isBridgeVector, u32 userRank)
      17            0 :     : CalcTransportReqBase(subCommPlaneVector, isBridgeVector, userRank)
      18              : {
      19            0 : }
      20              : 
      21            0 : CalcNHRV1TransportReq::~CalcNHRV1TransportReq()
      22              : {
      23            0 : }
      24              : 
      25            0 : HcclResult CalcNHRV1TransportReq::CalcTransportRequest(const std::string &tag, TransportMemType inputMemType,
      26              :     TransportMemType outputMemType, const CommParaInfo &commParaInfo,
      27              :     std::vector<SingleSubCommTransport> &commTransport, u32 subUserRankRoot)
      28              : {
      29              :     (void)subUserRankRoot;
      30            0 :     u32 ringSize = subCommPlaneVector_.size();
      31            0 :     commTransport.resize(ringSize);
      32              : 
      33            0 :     for (u32 ringIndex = 0; ringIndex < ringSize; ringIndex++) {
      34            0 :         if (commParaInfo.commPlane == COMM_LEVEL1 && !isBridgeVector_[ringIndex]) {
      35            0 :             continue; // 跳出本次循环
      36              :         }
      37              : 
      38            0 :         u32 rank = GetSubCollectiveRank(subCommPlaneVector_[ringIndex]);
      39            0 :         if (rank == INVALID_VALUE_RANKID) {
      40            0 :             continue;
      41              :         }
      42              : 
      43            0 :         u32 rankSize = subCommPlaneVector_[ringIndex].size();
      44            0 :         SingleSubCommTransport &subCommTransport = commTransport[ringIndex];
      45            0 :         subCommTransport.transportRequests.resize(rankSize);
      46              :         // 只有一张卡时不需要建链
      47            0 :         if (rankSize == HCCL_RANK_SIZE_EQ_ONE) {
      48            0 :             HCCL_INFO("comm base needn't to create links, rankSize_[%u].", rankSize);
      49            0 :             return HCCL_SUCCESS;
      50              :         }
      51              : 
      52            0 :         RingInfo info = NHRV1Base::GetRingInfo(subCommPlaneVector_[ringIndex].size());
      53              : 
      54            0 :         u32 hIndex = info.GetHIndex(rank);
      55            0 :         u32 vIndex = info.GetVIndex(rank);
      56            0 :         u32 hSize = info.GetHSizeByVIndex(vIndex);
      57            0 :         u32 vSize = info.GetVSizeByHIndex(hIndex);
      58              : 
      59              :         // 收集邻居
      60            0 :         std::set<u32> links;
      61              :         // -- 水平方向：左邻居
      62            0 :         links.insert(info.GetRank(vIndex, (hIndex + hSize - 1) % hSize));
      63              :         // -- 水平方向：右邻居
      64            0 :         links.insert(info.GetRank(vIndex, (hIndex + 1) % hSize));
      65              :         // -- 垂直方向：上邻居
      66            0 :         links.insert(info.GetRank((vIndex + vSize - 1) % vSize, hIndex));
      67              :         // -- 垂直方向：下邻居
      68            0 :         links.insert(info.GetRank((vIndex + 1) % vSize, hIndex));
      69              : 
      70            0 :         u32 rankOffset = info.GetRankOffset();
      71            0 :         if (rankOffset != rankSize) {
      72              :             // -- 水平方向：建立(x,0)和(x, sqrt-1)之间的链接
      73            0 :             if (hIndex == info.GetSqrtRankSize() - 1) {
      74            0 :                 links.insert(info.GetRank(vIndex, 0));
      75            0 :             } else if (hIndex == 0) {
      76            0 :                 links.insert(info.GetRank(vIndex, info.GetSqrtRankSize() - 1));
      77              :             }
      78              :             // -- 垂直方向：建立(0,y)和(y,sqrt)之间、(sqrt-1,y)和(y,sqrt)之间的链接
      79            0 :             if (hIndex < info.GetSqrtRankSize()) {
      80            0 :                 if (info.GetHSizeByVIndex(hIndex) > info.GetSqrtRankSize() \
      81            0 :                         && (vIndex == info.GetVSizeByHIndex(hIndex) - 1 || vIndex == 0)) {
      82            0 :                     links.insert(info.GetRank(hIndex, info.GetSqrtRankSize()));
      83              :                 }
      84              :             } else {
      85            0 :                 links.insert(info.GetRank(0, vIndex));
      86            0 :                 links.insert(info.GetRank(info.GetVSizeByHIndex(vIndex) - 1, vIndex));
      87              :             }
      88              :         }
      89            0 :         for (u32 dstRank : links) {
      90            0 :             if (dstRank != rank) {
      91            0 :                 TransportRequest &tmpTransport = subCommTransport.transportRequests[dstRank];
      92            0 :                 tmpTransport.isValid = true;
      93            0 :                 tmpTransport.localUserRank  = userRank_;
      94            0 :                 tmpTransport.remoteUserRank = subCommPlaneVector_[ringIndex][dstRank];
      95            0 :                 tmpTransport.inputMemType = inputMemType;
      96            0 :                 tmpTransport.outputMemType = outputMemType;
      97            0 :                 HCCL_INFO("[CommFactory][CalcNHRV1CommInfo] param_.tag[%s] ringIndex[%u], localRank[%u], \
      98              :                     remoteRank[%u], inputMemType[%d], outputMemType[%d]", tag.c_str(), ringIndex, userRank_,
      99              :                     tmpTransport.remoteUserRank, inputMemType, outputMemType);
     100              :             }
     101              :         }
     102            0 :     }
     103            0 :     return HCCL_SUCCESS;
     104              : }
     105              : 
     106              : }  // namespace hccl