inference-engine/src/vpu/graph_transformer/src/utils/perf_report.cpp

   1 // Copyright (C) 2018-2019 Intel Corporation
   2 // SPDX-License-Identifier: Apache-2.0
   3 //
   4
   5 #include <vpu/utils/perf_report.hpp>
   6
   7 #include <vector>
   8 #include <string>
   9 #include <map>
  10
  11 namespace vpu {
  12
  13 std::map<std::string, ie::InferenceEngineProfileInfo> parsePerformanceReport(
  14         const std::vector<StageMetaInfo>& stagesMeta,
  15         const float* deviceTimings,
  16         int deviceTimingsCount,
  17         PerfReport perfReport,
  18         bool printReceiveTensorTime) {
  19     IE_ASSERT(deviceTimings != nullptr);
  20     IE_ASSERT(deviceTimingsCount > 0);
  21
  22     std::map<std::string, ie::InferenceEngineProfileInfo> outPerfMap;
  23
  24     int timeIndex = 0;
  25     int execIndex = 1;
  26
  27     for (const auto& stageMeta : stagesMeta) {
  28         float timeMS = 0;
  29         if (stageMeta.status == ie::InferenceEngineProfileInfo::EXECUTED &&
  30             timeIndex < deviceTimingsCount) {
  31             timeMS = deviceTimings[timeIndex];
  32             timeIndex++;
  33         }
  34
  35         if (stageMeta.stageType == "<Receive-Tensor>" &&
  36             !printReceiveTensorTime) {
  37             continue;
  38         }
  39
  40         ie::InferenceEngineProfileInfo profInfo = {};
  41
  42         profInfo.status = stageMeta.status;
  43
  44         profInfo.cpu_uSec = 0;
  45         profInfo.realTime_uSec = static_cast<long long int>(timeMS * 1000);
  46
  47         stageMeta.layerType.copy(profInfo.layer_type, sizeof(profInfo.layer_type) / sizeof(profInfo.layer_type[0]), 0);
  48         stageMeta.stageType.copy(profInfo.exec_type, sizeof(profInfo.exec_type) / sizeof(profInfo.exec_type[0]), 0);
  49
  50         if (stageMeta.stageType == "<Receive-Tensor>") {
  51             profInfo.execution_index = 0;
  52         } else if (stageMeta.status == ie::InferenceEngineProfileInfo::EXECUTED) {
  53             profInfo.execution_index = execIndex;
  54             execIndex++;
  55         }
  56
  57         if (perfReport == PerfReport::PerStage) {
  58             outPerfMap[stageMeta.stageName] = profInfo;
  59         } else if (perfReport == PerfReport::PerLayer) {
  60             auto it = outPerfMap.find(stageMeta.layerName);
  61             if (it == outPerfMap.end()) {
  62                 outPerfMap[stageMeta.layerName] = profInfo;
  63             } else {
  64                 auto& prevProfInfo = it->second;
  65
  66                 if (profInfo.status == ie::InferenceEngineProfileInfo::EXECUTED) {
  67                     prevProfInfo.status = ie::InferenceEngineProfileInfo::EXECUTED;
  68                 }
  69
  70                 prevProfInfo.cpu_uSec += profInfo.cpu_uSec;
  71                 prevProfInfo.realTime_uSec += profInfo.realTime_uSec;
  72             }
  73         }
  74     }
  75
  76     return outPerfMap;
  77 }
  78
  79 }  // namespace vpu