TensorRT/parsers/caffe/blobNameToTensor.h源碼研讀

TensorRT/parsers/caffe/blobNameToTensor.h

/*
 * 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.
 */

#ifndef TRT_CAFFE_PARSER_BLOB_NAME_TO_TENSOR_H
#define TRT_CAFFE_PARSER_BLOB_NAME_TO_TENSOR_H

#include <map>
#include <string>

#include "NvCaffeParser.h"
#include "NvInfer.h"

//對TensorRT/include/NvCaffeParser.h裡定義的nvcaffeparser1命名空間的延伸
namespace nvcaffeparser1
{
/*
IBlobNameToTensor
定義於TensorRT/include/NvCaffeParser.h
為一抽象類別
Object used to store and query Tensors after they have been extracted from a Caffe model using the ICaffeParser.
*/
/*
class BlobNameToTensor
其核心為mMap這個字典，用於把blob name對應到ITensor*，
圍繞著mMap有add，find，[]等函數
另外還有mError用於指出目前是否出錯
*/
class BlobNameToTensor : public IBlobNameToTensor
{
public:
    //新增一對（name，tensor）到字典裡
    void add(const std::string& name, nvinfer1::ITensor* tensor)
    {
        mMap[name] = tensor;
    }

    //在字典裡查找名為name的ITensor，如果找不到，則回傳nullptr
    nvinfer1::ITensor* find(const char* name) const override
    {
        auto p = mMap.find(name);
        if (p == mMap.end())
        {
            return nullptr;
        }
        return p->second;
    }

    //override []這個operator，這代表我們可以使用obj_of_ BlobNameToTensor[name]
    nvinfer1::ITensor*& operator[](const std::string& name)
    {
        return mMap[name];
    }

    //將字典裡的ITensor的名字都設為它所對應的key
    void setTensorNames()
    {
        for (auto& p : mMap)
        {
            /*
            ITensor::setName
            定義於TensorRT/include/NvInfer.h
            virtual void setName(const char* name) TRTNOEXCEPT = 0;
            */
            p.second->setName(p.first.c_str());
        }
    }

    ~BlobNameToTensor() override = default;

    bool isOK()
    {
        return !mError;
    }

private:
	/*
	nvinfer1::ITensor
	定義於TensorRT/include/NvInfer.h
	A tensor in a network definition.
	*/
    std::map<std::string, nvinfer1::ITensor*> mMap;
    bool mError{false};
};
} // namespace nvcaffeparser1
#endif // TRT_CAFFE_PARSER_BLOB_NAME_TO_TENSOR_H

延伸命名空間

TensorRT/include/NvCaffeParser.h已經定義過了nvcaffeparser1這個命名空間，這裡又定義了一次，這是為什麼呢？詳見C++ namespace，extending namespace。

const成員函數

在定義函數BlobNameToTensor::find時：，詳見C++ const member function。

override

在定義函數BlobNameToTensor::find時：

nvinfer1::ITensor* find(const char* name) const override

用到了一個關鍵字override，詳見C++ （pure）virutal function & abstract class。

NULL v.s. nullptr

函數BlobNameToTensor::find回傳了nullptr。nullptr與NULL有何區別呢？詳見C++ NULL v.s. nullptr。

operator[]的回傳型別

operator[]的回傳型別為nvinfer1::ITensor*&，看著有些奇怪，相關內容詳見C++ Overload []。

= default

在~BlobNameToTensor()函數時用到了= default指示詞（specifier）。詳見C++ Explicitly defaulted function。

參考連結

C++ namespace，extending namespace

C++ const member function

C++ （pure）virutal function & abstract class

C++ NULL v.s. nullptr

C++ Overload []

C++ Explicitly defaulted function