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_ring_transport_req.h"
      12              : #include "calc_ahc_template_register.h"
      13              : 
      14              : constexpr u32 HCCL_RANK_OFFSET = 1;
      15              : 
      16              : namespace hccl {
      17           62 : CalcRingTransportReq::CalcRingTransportReq(std::vector<std::vector<u32>> &subCommPlaneVector,
      18           62 :     std::vector<bool> &isBridgeVector, u32 userRank)
      19           62 :     : CalcTransportReqBase(subCommPlaneVector, isBridgeVector, userRank)
      20              : {
      21           62 : }
      22              : 
      23          124 : CalcRingTransportReq::~CalcRingTransportReq()
      24              : {
      25          124 : }
      26              : 
      27           61 : HcclResult CalcRingTransportReq::CalcTransportRequest(const std::string &tag, TransportMemType inputMemType,
      28              :     TransportMemType outputMemType, const CommParaInfo &commParaInfo,
      29              :     std::vector<SingleSubCommTransport> &commTransport, u32 subUserRankRoot)
      30              : {
      31           61 :     u32 ringSize = subCommPlaneVector_.size();
      32           61 :     commTransport.resize(ringSize);
      33              : 
      34          122 :     for (u32 ringIndex = 0; ringIndex < ringSize; ringIndex++) {
      35           55 :         if ((commParaInfo.commPlane == COMM_LEVEL1 || commParaInfo.commPlane == COMM_LEVEL1_ANYPATH_SDMA ||
      36          158 :             commParaInfo.commPlane == COMM_LEVEL1_ANYPATH_RDMA)&&
      37           22 :             !isBridgeVector_[ringIndex]) {
      38           13 :             continue; // 跳出本次循环
      39              :         }
      40              : 
      41           68 :         u32 rank = GetSubCollectiveRank(subCommPlaneVector_[ringIndex]);
      42           68 :         if (rank == INVALID_VALUE_RANKID) {
      43            0 :             continue;
      44              :         }
      45              : 
      46           68 :         u32 rankSize = subCommPlaneVector_[ringIndex].size();
      47           68 :         SingleSubCommTransport &subCommTransport = commTransport[ringIndex];
      48           68 :         subCommTransport.transportRequests.resize(rankSize);
      49              :         // 只有一张卡时不需要建链
      50           68 :         if (rankSize == HCCL_RANK_SIZE_EQ_ONE) {
      51           18 :             HCCL_INFO("comm base needn't to create links, rankSize_[%u].", rankSize);
      52           18 :             return HCCL_SUCCESS;
      53              :         }
      54              : 
      55          238 :         for (u32 rankIndex = 0; rankIndex < rankSize; rankIndex++) {
      56          188 :             TransportRequest &tmpTransport = subCommTransport.transportRequests[rankIndex];
      57          188 :             if (rankIndex == (rank + rankSize + HCCL_RANK_OFFSET) % rankSize ||
      58          138 :                 rankIndex == (rank + rankSize - HCCL_RANK_OFFSET) % rankSize) {
      59           74 :                 tmpTransport.isValid = true;
      60           74 :                 tmpTransport.localUserRank  = userRank_;
      61           74 :                 tmpTransport.remoteUserRank = subCommPlaneVector_[ringIndex][rankIndex];
      62           74 :                 tmpTransport.inputMemType = inputMemType;
      63           74 :                 tmpTransport.outputMemType = outputMemType;
      64           74 :                 HCCL_INFO("[CommFactory][CalcRingCommInfo] param_.tag[%s] ringIndex[%u], localRank[%u], "\
      65              :                     "remoteRank[%u], inputMemType[%d], outputMemType[%d]", tag.c_str(), ringIndex, userRank_,
      66              :                     tmpTransport.remoteUserRank, inputMemType, outputMemType);
      67              :             } else {
      68          114 :                 tmpTransport.isValid = false;
      69              :             }
      70              :         }
      71           50 :         const u32 rdmaTaskNumRatio = 4;
      72           50 :         if (GetWorkflowMode() != HcclWorkflowMode::HCCL_WORKFLOW_MODE_OP_BASE) {
      73           26 :             subCommTransport.taskNum = rankSize * rdmaTaskNumRatio;
      74              :         }
      75              :     }
      76              : 
      77           44 :     return HCCL_SUCCESS;
      78              : }
      79              : 
      80            0 : HcclResult CalcRingTransportReq::CalcDstRanks(const u32 rank, const std::vector<u32> commGroups, std::set<u32> &dstRanks)
      81              : {
      82            0 :     CHK_PRT_RET(rank >= commGroups.size(),
      83              :         HCCL_ERROR("[CalcRingTransportReq][CalcDstRanks] rank [%u] exceed commGroups Size [%u]  error", 
      84              :         rank, commGroups.size() ), HCCL_E_INTERNAL);
      85              :     
      86              :     // 正方向下一个节点的rank号
      87            0 :     const u32 targetRankPos = static_cast<u32>(rank + 1) % commGroups.size();
      88            0 :     dstRanks.insert(commGroups[targetRankPos]);
      89              :  
      90              :     // 反方向下一个节点的rank号
      91            0 :     const u32 targetRankNeg = static_cast<u32>(rank + commGroups.size() - 1) % commGroups.size();
      92              :     
      93            0 :     HCCL_DEBUG("[CalcRingTransportReq][CalcDstRanks] local rank[%u], remote rank[%u]", commGroups[rank], commGroups[targetRankNeg]);
      94              : 
      95            0 :     dstRanks.insert(commGroups[targetRankNeg]);
      96              :  
      97            0 :     return HCCL_SUCCESS;
      98              : }
      99              : REGISTER_AHC_COMM_CALC_FUNC(AHCTemplateType::AHC_TEMPLATE_RING, CalcRingTransportReq, CalcRingTransportReq::CalcDstRanks);
     100              : }  // namespace hccl