本文共 5545 字，大约阅读时间需要 18 分钟。

前言

由於篇幅有限，筆者將 TensorRT/parsers/caffe/caffeParser/caffeParser.cpp 中的內容拆分成 ， ， 等幾篇，本篇為第一篇。

TensorRT/parsers/caffe/caffeParser/caffeParser.cpp

/* * Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at * *     http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. */#include 
   
    #include "caffeMacros.h"#include "caffeParser.h"#include "opParsers.h"#include "parserUtils.h"#include "readProto.h"#include "binaryProtoBlob.h"#include "google/protobuf/text_format.h"#include "half.h"#include "NvInferPluginUtils.h"using namespace nvinfer1; //定義於多處，此處用的是include/NvInferPluginUtils.husing namespace nvcaffeparser1; //定義於include/NvCaffeParser.hCaffeParser::~CaffeParser(){       /*    於TensorRT/parsers/caffe/caffeParser/caffeParser.h    std::vector
    
      mTmpAllocs;    */    for (auto v : mTmpAllocs)    {           free(v);    }    /*    於TensorRT/parsers/caffe/caffeParser/caffeParser.h    std::vector
     
       mNewPlugins;    */    for (auto p : mNewPlugins)    {           if (p)        {               /*            nvinfer1::IPluginV2::destroy            宣告於TensorRT/include/NvInferRuntimeCommon.h            virtual void destroy() TRTNOEXCEPT = 0;            Destroy the plugin object. This will be called when the network, builder or engine is destroyed.            */            p->destroy();        }    }    /*    在parse(INetworkDefinition&, DataType, bool)中：    mBlobNameToTensor = new (BlobNameToTensor);    */    delete mBlobNameToTensor;}/*讀取buffer中的數據，設定好mModel及mDeploy後，填滿network並回傳mBlobNameToTensor這個字典，如果建構過程都正常，最後會回傳mBlobNameToTensor，否則回傳nullptr*/const IBlobNameToTensor* CaffeParser::parseBuffers(const char* deployBuffer,                                                   std::size_t deployLength,                                                   const char* modelBuffer,                                                   std::size_t modelLength,                                                   INetworkDefinition& network,                                                   DataType weightType){       //設定CaffeParser成員變數std::shared_ptr
      
        mDeploy    //宣告時是shared_ptr，到了這裡變成unique_ptr?    mDeploy = std::unique_ptr
       
        (new trtcaffe::NetParameter);    google::protobuf::io::ArrayInputStream deployStream(deployBuffer, deployLength);    //從給定的ZeroCopyInputStream裡讀取並解析文字格式的protocol message，存到給定的Message物件當中    if (!google::protobuf::TextFormat::Parse(&deployStream, mDeploy.get()))    {           RETURN_AND_LOG_ERROR(nullptr, "Could not parse deploy file");    }    //解析modelBuffer，設定CaffeParser類別的成員變數std::shared_ptr
        
          mModel if (modelBuffer) { //宣告時是shared_ptr，到了這裡變成unique_ptr? mModel = std::unique_ptr
         
          (new trtcaffe::NetParameter); google::protobuf::io::ArrayInputStream modelStream(modelBuffer, modelLength); google::protobuf::io::CodedInputStream codedModelStream(&modelStream); codedModelStream.SetTotalBytesLimit(modelLength, -1); if (!mModel->ParseFromCodedStream(&codedModelStream)) { RETURN_AND_LOG_ERROR(nullptr, "Could not parse model file"); } } /* 設定好mModel及mDeploy後呼叫parse， 用於填滿network並回傳mBlobNameToTensor這個字典， 如果建構過程都正常，最後會回傳mBlobNameToTensor，否則回傳nullptr */ return parse(network, weightType, modelBuffer != nullptr);}/*讀取deploy檔及model檔中的數據，設定好mModel及mDeploy後，填滿network並回傳mBlobNameToTensor這個字典，如果建構過程都正常，最後會回傳mBlobNameToTensor，否則回傳nullptr*/const IBlobNameToTensor* CaffeParser::parse(const char* deployFile, const char* modelFile, INetworkDefinition& network, DataType weightType){ //因為是macro函數，所以結尾不加分號 //檢查deployFile是否為nullptr，如果是，則回傳nullptr CHECK_NULL_RET_NULL(deployFile) // this is used to deal with dropout layers which have different input and output /* trtcaffe::NetParameter來自： TensorRT/parsers/caffe/proto/trtcaffe.proto裡的 trtcaffe package下的NetParameter message */ /* CaffeParser的成員變數 std::shared_ptr
          
            mModel; 為何到這裡變成unique_ptr？ */ mModel = std::unique_ptr
           
            (new trtcaffe::NetParameter); /* readBinaryProto 定義於TensorRT/parsers/caffe/caffeParser/readProto.h */ //將modelFile裡的權重讀到mModel.get()指標所指向的trtcaffe::NetParameter裡 if (modelFile && !readBinaryProto(mModel.get(), modelFile, mProtobufBufferSize)) { RETURN_AND_LOG_ERROR(nullptr, "Could not parse model file"); } /* CaffeParser的成員變數 std::shared_ptr
            
              mDeploy; 為何到這裡變成unique_ptr？ */ /* readTextProto 定義於TensorRT/parsers/caffe/caffeParser/readProto.h */ //將deployFile裡的模型架構讀到mDeploy.get()指標所指向的trtcaffe::NetParameter裡 mDeploy = std::unique_ptr
             
              (new trtcaffe::NetParameter); if (!readTextProto(mDeploy.get(), deployFile)) { RETURN_AND_LOG_ERROR(nullptr, "Could not parse deploy file"); } /* 設定好mModel及mDeploy後呼叫parse， 用於填滿network並回傳mBlobNameToTensor這個字典， 如果建構過程都正常，最後會回傳mBlobNameToTensor，否則回傳nullptr */ return parse(network, weightType, modelFile != nullptr);}
             
            
           
          
         
        
       
      
     
    
   

google::protobuf

用到了 google::protobuf::io::ArrayInputStream， google::protobuf::io::CodedInputStream 等類別及 google::protobuf::TextFormat::Parse， MessageLite::ParseFromCodedStream 等函數，詳見。

RETURN_AND_LOG_ERROR，CHECK_NULL_RET_NULL

用到了RETURN_AND_LOG_ERROR及CHECK_NULL_RET_NULL兩個函數，詳見。

參考連結