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 "aicpusd_profiler.h"
      11              : #include "hiperf_marker.h"
      12              : #include "hiperf_exception.h"
      13              : #include "aicpusd_common.h"
      14              : 
      15              : namespace AicpuSchedule {
      16              : namespace {
      17              : constexpr float64_t SECOND_TO_NANO = 1000000000.0;
      18              : constexpr float64_t ONE_MS = 1000000.0;
      19              : constexpr uint32_t HIGH_POISTION_OFFSET_64_BIT = 32U;
      20              : constexpr uint32_t HIGH_POISTION_OFFSET_32_BIT = 16U;
      21              : constexpr uint32_t LOW_POISTION_MASK_32_BIT = 0x0000ffffU;
      22              : const std::string HIPERF_LIB_PATH = "/usr/lib64/libhiperf_executor.so";
      23              : } // namespace
      24              : 
      25              : /**
      26              :  * @ingroup AicpuProfiler
      27              :  * @brief it is used to construct a object of AicpuScheduleCore.
      28              :  */
      29            3 : AicpuProfiler::AicpuProfiler() noexcept
      30            3 :     : pid_(0),
      31            3 :       tid_(0),
      32            3 :       kernelTrack_({0UL}),
      33            3 :       frequence_(0.0),
      34            3 :       hiperfExisted_(false),
      35            3 :       accessHiperfSo_(false),
      36            3 :       oneMsForTick_(0UL),
      37            3 :       tenMsForTick_(0UL)
      38            3 : {}
      39              : 
      40              : /**
      41              :  * @ingroup AicpuProfiler
      42              :  * @brief it is used to destructor a object of AicpuProfiler.
      43              :  */
      44            3 : void AicpuProfiler::Uninit() { ::FiniMarker(); }
      45              : 
      46            3 : void AicpuProfiler::InitProfiler(const pid_t processId, const pid_t threadId)
      47              : {
      48            3 :     if (!accessHiperfSo_) {
      49            3 :         if (access(HIPERF_LIB_PATH.c_str(), F_OK) == 0) {
      50            2 :             hiperfExisted_ = true;
      51              :         }
      52            3 :         accessHiperfSo_ = true;
      53              :     } else {
      54            0 :         if (!hiperfExisted_) {
      55            0 :             return;
      56              :         }
      57              :     }
      58            3 :     kernelTrack_ = {0UL};
      59            3 :     pid_ = processId;
      60            3 :     tid_ = threadId;
      61              : 
      62            3 :     if (tenMsForTick_ == 0UL) {
      63            3 :         frequence_ = static_cast<float64_t>(GetAicpuSysFreq());
      64            3 :         const float64_t oneMsForTickTemp = ONE_MS / (SECOND_TO_NANO / frequence_);
      65            3 :         oneMsForTick_ = static_cast<uint64_t>(oneMsForTickTemp);
      66            3 :         tenMsForTick_ = oneMsForTick_ * 10UL;
      67              :     }
      68              : }
      69              : 
      70            1 : void AicpuProfiler::ProfilerAgentInit() { ::InitMarker(); }
      71              : 
      72            3 : void AicpuProfiler::Profiler()
      73              : {
      74            3 :     if (!hiperfExisted_) {
      75            1 :         return;
      76              :     }
      77            2 :     kernelTrack_.pid = pid_;
      78            2 :     kernelTrack_.tid = static_cast<int32_t>(GetTid());
      79            2 :     (void)Hiva::MarkerAicpuScheduler(kernelTrack_);
      80            2 :     const uint64_t prefKey = BuildKey();
      81            2 :     if (kernelTrack_.activeStream != 0UL) {
      82            1 :         (void)Hiva::PerfDurationBegin(prefKey, kernelTrack_.activeStream);
      83            1 :     } else if (kernelTrack_.endGraph != 0UL) {
      84            1 :         (void)Hiva::PerfDurationEnd(prefKey, kernelTrack_.endGraph, tenMsForTick_, kernelTrack_);
      85              :     } else {
      86            0 :         return;
      87              :     }
      88              : }
      89              : 
      90            2 : uint64_t AicpuProfiler::BuildKey()
      91              : {
      92            2 :     const uint64_t key = (static_cast<uint64_t>(kernelTrack_.modelId) << HIGH_POISTION_OFFSET_64_BIT) |
      93            2 :                          ((static_cast<uint64_t>(kernelTrack_.rawStamp.tv_sec) & LOW_POISTION_MASK_32_BIT)
      94            2 :                           << HIGH_POISTION_OFFSET_32_BIT) |
      95            2 :                          (static_cast<uint64_t>(kernelTrack_.rawStamp.tv_nsec) & LOW_POISTION_MASK_32_BIT);
      96            2 :     return key;
      97              : }
      98              : 
      99            1 : uint64_t AicpuProfiler::SchedGetCurCpuTick(void) const
     100              : {
     101            1 :     uint64_t cnt = 0UL;
     102              : #ifndef RUN_TEST
     103              :     asm volatile("mrs %0, CNTVCT_EL0" : "=r"(cnt) :);
     104              : #endif
     105            1 :     return cnt;
     106              : }
     107              : 
     108            4 : uint64_t AicpuProfiler::GetAicpuSysFreq() const
     109              : {
     110            4 :     uint64_t freq = 1UL;
     111              : #ifndef RUN_TEST
     112              :     asm volatile("mrs %0, CNTFRQ_EL0" : "=r"(freq) :);
     113              : #endif
     114            4 :     return freq;
     115              : }
     116              : 
     117              : thread_local AicpuProfiler g_aicpuProfiler;
     118              : } // namespace AicpuSchedule