TensorRT/samples/common/argsParser.h源碼研讀
发布日期:2021-05-06 19:47:45 浏览次数:27 分类:精选文章

本文共 5657 字,大约阅读时间需要 18 分钟。

TensorRT/samples/common/argsParser.h源碼研讀

argsParser.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 TENSORRT_ARGS_PARSER_H#define TENSORRT_ARGS_PARSER_H#include 
#include
#ifdef _MSC_VER#include "..\common\windows\getopt.h"#else#include
#endif#include
namespace samplesCommon{ //!//! \brief The SampleParams structure groups the basic parameters required by//! all sample networks.//!struct SampleParams{ int batchSize{ 1}; //!< Number of inputs in a batch int dlaCore{ -1}; //!< Specify the DLA core to run network on. bool int8{ false}; //!< Allow runnning the network in Int8 mode. bool fp16{ false}; //!< Allow running the network in FP16 mode. std::vector
dataDirs; //!< Directory paths where sample data files are stored std::vector
inputTensorNames; std::vector
outputTensorNames;};//!//! \brief The CaffeSampleParams structure groups the additional parameters required by//! networks that use caffe//!// 在SampleParams的基礎上加上caffe特有的參數struct CaffeSampleParams : public SampleParams{ std::string prototxtFileName; //!< Filename of prototxt design file of a network std::string weightsFileName; //!< Filename of trained weights file of a network std::string meanFileName; //!< Filename of mean file of a network};//!//! \brief The OnnxSampleParams structure groups the additional parameters required by//! networks that use ONNX//!struct OnnxSampleParams : public SampleParams{ std::string onnxFileName; //!< Filename of ONNX file of a network};//!//! \brief The UffSampleParams structure groups the additional parameters required by//! networks that use Uff//!struct UffSampleParams : public SampleParams{ std::string uffFileName; //!< Filename of uff file of a network};//!//! /brief Struct to maintain command-line arguments.//!//parseArgs解析完命令行的參數後,會把得到的資訊存入此結構體內struct Args{ bool runInInt8{ false}; bool runInFp16{ false}; bool help{ false}; int useDLACore{ -1}; int batch{ 1}; std::vector
dataDirs; //? bool useILoop{ false};};//!//! \brief Populates the Args struct with the provided command-line parameters.//!//! \throw invalid_argument if any of the arguments are not valid//!//! \return boolean If return value is true, execution can continue, otherwise program should exit//!//解析argv,將所得到的資訊存入args這個struct Args型別的變數內//如果出錯會回傳false,反之則回傳trueinline bool parseArgs(Args& args, int argc, char* argv[]){ //無窮迴圈。等到參數解析完了,arg會自動變成-1,接著跳出迴圈 while (1) { int arg; static struct option long_options[] = { { "help", no_argument, 0, 'h'}, { "datadir", required_argument, 0, 'd'}, { "int8", no_argument, 0, 'i'}, { "fp16", no_argument, 0, 'f'}, { "useILoop", no_argument, 0, 'l'}, { "useDLACore", required_argument, 0, 'u'}, { "batch", required_argument, 0, 'b'}, { nullptr, 0, nullptr, 0}}; int option_index = 0; // getopt_long用於解析傳入的參數,定義於Linux系統自帶的getopt.h, // 亦或是samples/common/windows/getopt.c // 在getopt_long被呼叫的過程中,option_index會隨之變化,代表該次解析的是long_options裡的第幾個參數 //getopt_long的回傳值是struct option裡最後一個元素,即'h','d','i','f','l','u','b'或0 //注意到d後面跟著一個冒號,這代表-d後面還跟著一個參數 //為何這裡optstring只有'h','d','i','u',少了'f','l','b'及0? arg = getopt_long(argc, argv, "hd:iu", long_options, &option_index); // 如果所有可選參數都解析完了,則getopt_long回傳-1 if (arg == -1) { break; } switch (arg) { //parseArgs的參數args傳入時是空的,在這個block裡依不同情況來為args設值 case 'h': args.help = true; return true; case 'd': //optarg定義於samples/common/windows/getopt.c,由getopt_long間接地被賦值,表示可選參數 //-d後跟著的參數會被解析到optarg內 if (optarg) { args.dataDirs.push_back(optarg); } else { std::cerr << "ERROR: --datadir requires option argument" << std::endl; return false; } break; case 'i': args.runInInt8 = true; break; //'f'不在optstring內,能被解析出來? case 'f': args.runInFp16 = true; break; //? //'l'不在optstring內,能被解析出來? case 'l': args.useILoop = true; break; case 'u': //在optstring中,u後面沒加冒號,為何這裡可以有參數? if (optarg) { args.useDLACore = std::stoi(optarg); } break; case 'b': //'b'不在optstring內,能被解析出來? if (optarg) { args.batch = std::stoi(optarg); } break; //如果傳入的參數不是上面的任一個,則程序執行失敗 default: return false; } } return true;}} // namespace samplesCommon#endif // TENSORRT_ARGS_PARSER_H

namespace

代碼中定義了名為samplesCommon的namespace。關於namespace,詳見。

struct的繼承

CaffeSampleParams結構體是繼承自SampleParams結構體,

關於struct的繼承,詳見。

caffe特有參數

  • prototxtFileName:預期的參數是以.prototxt結尾的檔名,代表網路結構定義

  • weightsFileName:預期的參數是以.caffemodel結尾的檔名,代表網路權重

  • meanFileName:預期的參數是以.binaryproto結尾的檔名,代表圖像的平均值

UFF格式

摘自:

In order to run a TensorFlow network with TensorRT, you must first convert it to the UFF format.

UFF格式似乎是為了將TensorFlow與TensorRT的模型互相轉換而定義的一種中間表示方式?

不需要typedef struct xxx yyy?

記得在C語言中,定義完一個struct之後,我們通常還會加上typedef,用以簡化之後與該struct相關的代碼。但是在argsParser.h中,定義完Args這個結構體後,我們卻可以直接在parseArgs函數裡使用它,這是為什麼呢?詳見。

struct初始化

parseArgs函數中定義了一個struct option型別的變數long_options,對它初始化的方式就像在初始化array一樣。關於struct的初始化方式,詳見。

struct option,getopt_long

parseArgs函數中定義了一個struct option型別的變數long_options,還用到了getopt_long這個函數,詳見。

inline

parseArgs函數被宣告為inline function,關於inline function,詳見。

DLA

NVDLA是NVIDIA開源的深度學習加速架構,包含了及部份。

參考連結

上一篇:C++ smart pointer,unique_ptr,shared_ptr
下一篇:C++ typedef struct v.s. struct

发表评论

最新留言

路过,博主的博客真漂亮。。
[***.116.15.85]2025年04月13日 10时42分46秒