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 "hccl_ccu_res.h"
      12              : 
      13              : #include <mutex>
      14              : 
      15              : #include "hccl_comm_pub.h"
      16              : 
      17              : #include "exception_handler.h"
      18              : #include "ccu_instance_mgr.h"
      19              : 
      20              : /**
      21              :  * @note 职责：集合通信的通信域CCU管理的C接口的C到C++适配
      22              :  */
      23            0 : HcclResult HcclCommQueryCcuIns(HcclComm comm,
      24              :     CcuInsHandle *insHandles, uint32_t *insNum)
      25              : {
      26              :     EXCEPTION_HANDLE_BEGIN
      27              : 
      28            0 :     HcclUs startut = TIME_NOW();
      29              : 
      30            0 :     CHK_PTR_NULL(comm);
      31            0 :     auto *hcclComm = static_cast<hccl::hcclComm *>(comm);
      32            0 :     const auto &commId = hcclComm->GetIdentifier();
      33            0 :     HCCL_INFO("[%s] CommId[%s] query ccu instance.", __func__, commId.c_str());
      34              : 
      35            0 :     CHK_PTR_NULL(insHandles);
      36            0 :     CHK_PTR_NULL(insNum);
      37              : 
      38              :     // CCU不支持A5之前代际
      39            0 :     if (!hcclComm->IsCommunicatorV2()) {
      40            0 :         HCCL_WARNING("[%s] is not supported.", __func__);
      41            0 :         return HcclResult::HCCL_E_NOT_SUPPORT;
      42              :     }
      43              : 
      44            0 :     auto *collComm = hcclComm->GetCollComm();
      45            0 :     CHK_PTR_NULL(collComm);
      46            0 :     auto *myRank = collComm->GetMyRank();
      47            0 :     CHK_PTR_NULL(myRank);
      48              : 
      49              :     // 非CCU通信域允许查询，不认为是错误
      50            0 :     auto ccuInsHandle = myRank->GetCcuInstance();
      51            0 :     if (ccuInsHandle == 0) {
      52            0 :         auto opExpansionMode = myRank->GetOpExpansionMode();
      53            0 :         HCCL_WARNING("[%s] failed to get ccu instance, commId[%s] op expansion mode[%u].",
      54              :             __func__, commId.c_str(), opExpansionMode);
      55            0 :         return HcclResult::HCCL_E_UNAVAIL;
      56              :     }
      57              : 
      58            0 :     insHandles[0] = ccuInsHandle;
      59            0 :     *insNum = 1;
      60            0 :     HCCL_INFO("[%s] success, take time [%lld]us.",
      61              :         __func__, DURATION_US(TIME_NOW() - startut).count());
      62              :     
      63            0 :     EXCEPTION_HANDLE_END
      64            0 :     return HcclResult::HCCL_SUCCESS;
      65              : }
      66              : 
      67           13 : HcclResult HcclCommAssignCcuIns(HcclComm comm, CcuInsHandle insHandle)
      68              : {
      69              :     EXCEPTION_HANDLE_BEGIN
      70              : 
      71           13 :     HcclUs startut = TIME_NOW();
      72              : 
      73           21 :     CHK_PTR_NULL(comm);
      74           12 :     auto *hcclComm = static_cast<hccl::hcclComm *>(comm);
      75           12 :     const auto &commId = hcclComm->GetIdentifier();
      76           12 :     HCCL_INFO("[%s] CommId[%s] assign ccu instance[%llu].",
      77              :         __func__, commId.c_str(), static_cast<unsigned long long>(insHandle));
      78              : 
      79           12 :     if (insHandle == 0) {
      80            1 :         HCCL_ERROR("[%s] failed, commId[%s] insHandle[%llu] is invalid.",
      81              :             __func__, commId.c_str(), static_cast<unsigned long long>(insHandle));
      82            1 :         return HcclResult::HCCL_E_PARA;
      83              :     }
      84              : 
      85           11 :     if (!hcclComm->IsCommunicatorV2()) {
      86            1 :         HCCL_WARNING("[%s] is not supported.", __func__);
      87            1 :         return HcclResult::HCCL_E_NOT_SUPPORT;
      88              :     }
      89              : 
      90           10 :     auto *collComm = hcclComm->GetCollComm();
      91           10 :     CHK_PTR_NULL(collComm);
      92            9 :     auto *myRank = collComm->GetMyRank();
      93            9 :     CHK_PTR_NULL(myRank);
      94              : 
      95              :     {
      96              :         // 仅保证多个 Assign 调用之间并发安全，Query 和通信域销毁由调用方保证不与 Assign 并发。
      97              :         static std::mutex assignCcuInsMutex;
      98            8 :         std::lock_guard<std::mutex> lock(assignCcuInsMutex);
      99              : 
     100            8 :         auto oldInsHandle = myRank->GetCcuInstance();
     101            8 :         if (oldInsHandle != 0) {
     102            3 :             HCCL_ERROR("[%s] failed, commId[%s] already has ccu instance[%llu], "
     103              :                 "new instance[%llu] will not be assigned.",
     104              :                 __func__, commId.c_str(), static_cast<unsigned long long>(oldInsHandle),
     105              :                 static_cast<unsigned long long>(insHandle));
     106            3 :             return HcclResult::HCCL_E_PARA;
     107              :         }
     108              : 
     109            5 :         const auto devLogicId = collComm->GetDeviceLogicId();
     110            5 :         auto *ccuIns = hcomm::CcuInstanceMgr::GetInstance(devLogicId).Get(insHandle);
     111            4 :         if (ccuIns == nullptr) {
     112            1 :             HCCL_ERROR("[%s] failed, commId[%s] ccu instance[%llu] is not found.",
     113              :                 __func__, commId.c_str(), static_cast<unsigned long long>(insHandle));
     114            1 :             return HcclResult::HCCL_E_NOT_FOUND;
     115              :         }
     116              : 
     117            3 :         myRank->SetCcuInstance(insHandle);
     118            8 :     }
     119              : 
     120            3 :     HCCL_INFO("[%s] success, commId[%s] ccu instance[%llu], take time [%lld]us.",
     121              :         __func__, commId.c_str(), static_cast<unsigned long long>(insHandle),
     122              :         DURATION_US(TIME_NOW() - startut));
     123              : 
     124           13 :     EXCEPTION_HANDLE_END
     125            3 :     return HcclResult::HCCL_SUCCESS;
     126              : }