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(
      18           62 :     std::vector<std::vector<u32>>& subCommPlaneVector, std::vector<bool>& isBridgeVector, u32 userRank)
      19           62 :     : CalcTransportReqBase(subCommPlaneVector, isBridgeVector, userRank)
      20           62 : {}
      21              : 
      22          124 : CalcRingTransportReq::~CalcRingTransportReq() {}
      23              : 
      24           62 : HcclResult CalcRingTransportReq::CalcTransportRequest(
      25              :     const std::string& tag, TransportMemType inputMemType, TransportMemType outputMemType,
      26              :     const CommParaInfo& commParaInfo, std::vector<SingleSubCommTransport>& commTransport, u32 subUserRankRoot)
      27              : {
      28           62 :     u32 ringSize = subCommPlaneVector_.size();
      29           62 :     commTransport.resize(ringSize);
      30              : 
      31          125 :     for (u32 ringIndex = 0; ringIndex < ringSize; ringIndex++) {
      32           55 :         if ((commParaInfo.commPlane == COMM_LEVEL1 || commParaInfo.commPlane == COMM_LEVEL1_ANYPATH_SDMA
      33           55 :              || commParaInfo.commPlane == COMM_LEVEL1_ANYPATH_RDMA)
      34          136 :             && !isBridgeVector_[ringIndex]) {
      35           13 :             continue; // 跳出本次循环
      36              :         }
      37              : 
      38           68 :         u32 rank = GetSubCollectiveRank(subCommPlaneVector_[ringIndex]);
      39           68 :         if (rank == INVALID_VALUE_RANKID) {
      40            0 :             continue;
      41              :         }
      42              : 
      43           68 :         u32 rankSize = subCommPlaneVector_[ringIndex].size();
      44           68 :         SingleSubCommTransport& subCommTransport = commTransport[ringIndex];
      45           68 :         subCommTransport.transportRequests.resize(rankSize);
      46              :         // 只有一张卡时不需要建链
      47           68 :         if (rankSize == HCCL_RANK_SIZE_EQ_ONE) {
      48           18 :             HCCL_INFO("comm base needn't to create links, rankSize_[%u].", rankSize);
      49           18 :             return HCCL_SUCCESS;
      50              :         }
      51              : 
      52          238 :         for (u32 rankIndex = 0; rankIndex < rankSize; rankIndex++) {
      53          188 :             TransportRequest& tmpTransport = subCommTransport.transportRequests[rankIndex];
      54          188 :             if (rankIndex == (rank + rankSize + HCCL_RANK_OFFSET) % rankSize
      55          138 :                 || rankIndex == (rank + rankSize - HCCL_RANK_OFFSET) % rankSize) {
      56           74 :                 tmpTransport.isValid = true;
      57           74 :                 tmpTransport.localUserRank = userRank_;
      58           74 :                 tmpTransport.remoteUserRank = subCommPlaneVector_[ringIndex][rankIndex];
      59           74 :                 tmpTransport.inputMemType = inputMemType;
      60           74 :                 tmpTransport.outputMemType = outputMemType;
      61           74 :                 HCCL_INFO(
      62              :                     "[CommFactory][CalcRingCommInfo] param_.tag[%s] ringIndex[%u], localRank[%u], "
      63              :                     "remoteRank[%u], inputMemType[%d], outputMemType[%d]",
      64              :                     tag.c_str(), ringIndex, userRank_, tmpTransport.remoteUserRank, inputMemType, outputMemType);
      65              :             } else {
      66          114 :                 tmpTransport.isValid = false;
      67              :             }
      68              :         }
      69           50 :         const u32 rdmaTaskNumRatio = 4;
      70           50 :         if (GetWorkflowMode() != HcclWorkflowMode::HCCL_WORKFLOW_MODE_OP_BASE) {
      71           26 :             subCommTransport.taskNum = rankSize * rdmaTaskNumRatio;
      72              :         }
      73              :     }
      74              : 
      75           44 :     return HCCL_SUCCESS;
      76              : }
      77              : 
      78              : HcclResult
      79            0 : CalcRingTransportReq::CalcDstRanks(const u32 rank, const std::vector<u32> commGroups, std::set<u32>& dstRanks)
      80              : {
      81            0 :     CHK_PRT_RET(
      82              :         rank >= commGroups.size(),
      83              :         HCCL_ERROR(
      84              :             "[CalcRingTransportReq][CalcDstRanks] rank [%u] exceed commGroups Size [%u] error", rank,
      85              :             commGroups.size()),
      86              :         HCCL_E_INTERNAL);
      87              : 
      88              :     // 正方向下一个节点的rank号
      89            0 :     const u32 targetRankPos = static_cast<u32>(rank + 1) % commGroups.size();
      90            0 :     dstRanks.insert(commGroups[targetRankPos]);
      91              : 
      92              :     // 反方向下一个节点的rank号
      93            0 :     const u32 targetRankNeg = static_cast<u32>(rank + commGroups.size() - 1) % commGroups.size();
      94              : 
      95            0 :     HCCL_DEBUG(
      96              :         "[CalcRingTransportReq][CalcDstRanks] local rank[%u], remote rank[%u]", commGroups[rank],
      97              :         commGroups[targetRankNeg]);
      98              : 
      99            0 :     dstRanks.insert(commGroups[targetRankNeg]);
     100              : 
     101            0 :     return HCCL_SUCCESS;
     102              : }
     103              : REGISTER_AHC_COMM_CALC_FUNC(
     104              :     AHCTemplateType::AHC_TEMPLATE_RING, CalcRingTransportReq, CalcRingTransportReq::CalcDstRanks);
     105              : } // namespace hccl