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              : #include "conn_local_notify_manager.h"
      11              : 
      12              : #include "rdma_handle_manager.h"
      13              : #include "communicator_impl.h"
      14              : #include "invalid_params_exception.h"
      15              : 
      16              : #include "ipc_local_notify.h"
      17              : #include "rdma_local_notify.h"
      18              : #include "ub_local_notify.h"
      19              : 
      20              : namespace Hccl {
      21              : 
      22          488 : ConnLocalNotifyManager::ConnLocalNotifyManager(CommunicatorImpl* communicator) : comm(communicator) {}
      23              : 
      24          488 : ConnLocalNotifyManager::~ConnLocalNotifyManager() { DECTOR_TRY_CATCH("ConnLocalNotifyManager", Destroy()); }
      25              : 
      26           12 : bool ConnLocalNotifyManager::IsExist(RankId remoteRankId, const LinkData& linkData)
      27              : {
      28           12 :     return notifyPool.count(remoteRankId) != 0 && notifyPool[remoteRankId].count(linkData) != 0;
      29              : }
      30              : 
      31            6 : void ConnLocalNotifyManager::ApplyFor(RankId remoteRankId, const LinkData& linkData)
      32              : {
      33           18 :     HCCL_INFO("Local notify for remoteRankId[%d] and linkData[%s] alloc.", remoteRankId, linkData.Describe().c_str());
      34            6 :     if (IsExist(remoteRankId, linkData)) {
      35            3 :         HCCL_WARNING(
      36              :             "Local notify for remoteRankId[%d] and linkData[%s] already exists, no need to alloc.", remoteRankId,
      37              :             linkData.Describe().c_str());
      38            1 :         return;
      39              :     }
      40              : 
      41            5 :     u32 count = 3; // 待修改: 需要定义GetCount()
      42            5 :     notifyPool[remoteRankId][linkData].resize(count);
      43              : 
      44           20 :     for (u32 i = 0; i < count; ++i) {
      45           15 :         if (linkData.GetType() == PortDeploymentType::P2P) {
      46           12 :             notifyPool[remoteRankId][linkData][i]
      47           24 :                 = make_unique<IpcLocalNotify>(comm->GetOpAiCpuTSFeatureFlag()); // 算子粒度
      48           12 :             continue;
      49            3 :         } else if (linkData.GetType() == PortDeploymentType::DEV_NET) {
      50            3 :             auto linkProtocol = linkData.GetLinkProtocol();
      51            3 :             if (linkProtocol == LinkProtocol::ROCE) {
      52            0 :                 RdmaHandle rdmaHandle = RdmaHandleManager::GetInstance().Get(
      53            0 :                     comm->GetDevicePhyId(), linkData.GetLocalPort(), linkProtocol);
      54            0 :                 notifyPool[remoteRankId][linkData][i]
      55            0 :                     = make_unique<RdmaLocalNotify>(rdmaHandle, comm->GetOpAiCpuTSFeatureFlag()); // 算子粒度
      56            0 :                 continue;
      57            0 :             } else if (
      58            3 :                 linkProtocol == LinkProtocol::UB_CTP || linkProtocol == LinkProtocol::UB_TP
      59            3 :                 || linkProtocol == LinkProtocol::UBOE || linkProtocol == LinkProtocol::UBG) {
      60            6 :                 RdmaHandle rdmaHandle = RdmaHandleManager::GetInstance().Get(
      61            3 :                     comm->GetDevicePhyId(), linkData.GetLocalPort(), linkProtocol);
      62            3 :                 notifyPool[remoteRankId][linkData][i]
      63            6 :                     = make_unique<UbLocalNotify>(rdmaHandle, comm->GetOpAiCpuTSFeatureFlag()); // 算子粒度
      64              :             } else {
      65              :                 // 待修改: 仅支持 P2P 和 RDMA 申请 notify
      66              :                 string msg = StringFormat(
      67            0 :                     "Unsupported %s of link %s", linkProtocol.Describe().c_str(), linkData.Describe().c_str());
      68            0 :                 THROW<InvalidParamsException>(msg);
      69            0 :             }
      70              :         } else {
      71              :             // 待修改: 仅支持 P2P 和 RDMA 申请 notify
      72              :             string msg = StringFormat(
      73            0 :                 "Unsupported %s of link %s", linkData.GetType().Describe().c_str(), linkData.Describe().c_str());
      74            0 :             THROW<InvalidParamsException>(msg);
      75            0 :         }
      76              :     };
      77              : }
      78              : 
      79            1 : bool ConnLocalNotifyManager::Release(RankId remoteRankId, const LinkData& linkData)
      80              : {
      81            1 :     if (!IsExist(remoteRankId, linkData)) {
      82            0 :         HCCL_WARNING(
      83              :             "Notify for remoteRankId[%d] and linkData[%p] does not exist, no need to release.", remoteRankId,
      84              :             &linkData);
      85            0 :         return true;
      86              :     }
      87              : 
      88            1 :     notifyPool[remoteRankId].erase(linkData);
      89            1 :     if (notifyPool[remoteRankId].empty()) {
      90            1 :         notifyPool.erase(remoteRankId);
      91              :     }
      92            1 :     return true;
      93              : }
      94              : 
      95            5 : vector<BaseLocalNotify*> ConnLocalNotifyManager::Get(RankId remoteRankId, const LinkData& linkData)
      96              : {
      97            5 :     vector<BaseLocalNotify*> v;
      98              : 
      99            5 :     if (!IsExist(remoteRankId, linkData)) {
     100           15 :         HCCL_WARNING(
     101              :             "Get Local notify for remoteRankId[%d] and linkData[%s] does not exist", remoteRankId,
     102              :             linkData.Describe().c_str());
     103            5 :         return v;
     104              :     }
     105              : 
     106            0 :     for (const auto& i : notifyPool[remoteRankId][linkData]) {
     107            0 :         v.emplace_back(i.get());
     108              :     }
     109              : 
     110            0 :     return v;
     111            0 : }
     112              : 
     113          489 : bool ConnLocalNotifyManager::Destroy()
     114              : {
     115          489 :     notifyPool.clear();
     116          489 :     return true;
     117              : }
     118              : 
     119              : } // namespace Hccl