[TOC]

Introduce

Author: steve

Email: [email protected]

Date: 2022-6-10

This is a concise tutorial to illustrate the process of compiling C program. The platform is Deepin 20.6,

community, a Debian based Linux distribution. Complier utilize g++.

$ g++ --version
>>> g++ (Uos 8.3.0.3-3+rebuild) 8.3.0
Copyright (C) 2018 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

1. Single compile process

// @file: test.c
#include <iostream>
using namespace std;

#define PI 3.14
#ifdef USING_FLOAT64
typedef double scalar;
#else
typedef float scalar;
#endif

int main() {
  scalar n;
  cout << "Hello world!" << endl;
  return 0;
}

Directly compile and generate executable file by command:

$ g++ test.c -o test

option -o means the output executable file name. If you don't specify it, the executable file name is a.out by default.

Then run the executable file,

$ ./test
>>> Hello world!

We hope to dive into the details when run command g++ test.c -o main to figure out the process when compile program.

Actually, g++ test.c contains 4 stages:

• Preprocess
• Compile
• Assemble
• Link

1.1 Preprocess

This stage generate test.i from test.c, which consists of 4 steps.

1. expanding header file
2. macro replacement
3. uncomment
4. condition compile

$ g++ test.c -E -o test.i

you could open both files to distinguish difference.

$ wc -l test.c
>>> 16 test.c
$ wc -l test.i
>>> 28279 test.i

Most part of test.i is include iostream file, whose last few lines are shown as follows:

using namespace std;





typedef float scalar;


int main() {
  scalar n;
  cout << "Hello world!" << endl;
  return 0;
}

1.2 Compile

Grammar check happened in this stage. At the same time, generate assembly code test.s

$ gcc test.c -S -o test.s

1.3 Assemble

Convert assembly code to machine code.

$ g++ test.s -c -o test.o

1.4 Link

This stage is to generate binary executable file from machine code. If you use the library, such as OpenCV, PCL, etc, the concerned library will be linked to test.o files.

$ g++ test.o -o test

*static and dynamic library

At the stage of link, two different kinds of libraries are supplied.

Both of them exist in the form of files, no math which platform you choose. Essentially both of them is a binary format of executable code that can be loaded into memory for execution. Both dynamic libraries and static link libraries provide variables, functions and classes to their callers.

The static library will be embedded into the executable file so that normally executable file size is large than dynamic library. In other word, the static library becomes the part of the executable file. On the contrary, dynamic library only embed some description information (redirection info normally) into executable file. Only the program is loaded into memory, the program will link the concerned libraries.

As for a single file , the program linked by static library can run anywhere. The dynamic library and executable file should keep all together in case that program run successfully.

However, if multiple file load the same library, it is more worthy to make use of dynamic library. For example, program A and B use the same function alignment() in XXX library. We only need to load once in memory rather than twice if dynamic library suppied.

---

Win32 platform static library file extension is .lib, dynamic file extension is .dll

Linux platform static library file extension is .a, dynamic file extension is .so

This image show the naming rule about library.

2. Generate library file

Before start to learn how to generate library file, we just make a summary about the compile options:

• -shared Generate dynamic library file

• -static Generate static library file

• -fPIC Position Independent Code

• -g gdb option for DEBUG

• -c Only generate .o file from .c

• -I include folder

• -L link library

2.1 Static library

If we have two files, add.c , add.h

// @file add.h
int add(int x, int y);
int sub(int x, int y);

// @file add.c
#pragma once
#include "add.h"

int add(int x, int y) {
  return x + y;
}

int sub(int x, int y) {
  return x - y;
}

$ g++ -c add.c
$ ls
>>> add.c add.h add.o

Generate static library by command ``

$ ar -rc libadd.a add.o
$ ls
>>> add.c add.h add.o libadd.a

the libadd.a can placed in any location only if the complier find it when linking. Later we'll discuss the locations the complier search for. For convenience, we touch a new file main.c located in the same folder with add.h

$ g++ main.c -o main libadd.a
$ ./main
>>> add(5,4) is 9
sub(5,4) is 1

Let's see libadd.a ,add.h and main.c are not in the same place, see source in folder add_sub_a_folder

$ g++ main.c -o main -I ./include ./lib/libadd.a

or

$ g++ main.c -o main -I ./include -L ./lib -ladd

remove -I , -L see what happened.

2.2 Dynamic library

folder add_sub_so

generate dynamic library by two commands

$ g++ -fPIC -c add.c -o libadd.o
$ g++ -shared -o libadd.so libadd.o

or single one command:

$ g++ -fPIC -shared -o libadd.so add.c

link

$ g++ main.c -o main libadd.so

If we move the libadd.so in another place, what will happened?

$ mv libadd.so ../
$ ./main
>>> /main: error while loading shared libraries: libadd.so: cannot open shared object file: No such file or directory

2.3 How to install dynamic library?

1. copy *.so to /lib or /usr/lib

$ sudo cp -d ../libadd.so /lib

or

$ sudo cp -d ../libadd.so /usr/lib

Actually /lib is a soft link of /usr/lib

$ ll /lib
>>> ll /lib lrwxrwxrwx 1 root root 7 Mar 31 15:11 /lib -> usr/lib

Then

$ ./main
>>> add(5,4) is 9
sub(5,4) is 1

Note: /usr/local/lib not work

2. Macro LD_LIBRARY_PATH

export LD_LIBRARY_PATH=/home/*****/*****/

Permanently enable

vim /etc/profile
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
source /etc/profile

or ~/.bashrc

3. /etc/ld.so.conf

sudo vim /etc/ld.so.conf

add /usr/local/lib

copy *.so to /usr/local/lib

sudo ldconfig

CMake version

notice main.c:3:10: fatal error: iostream: No such file or directory

change main.c to main.cpp could solve this problem.

I just simply #include<stdio.h> to solve it.

Tips

cmake -DCMAKE_INSTALL_PREFIX=~/.local
export LD_LIBRARY_PATH=~/.local:/$LD_LIBRARY_PATH

make -j16 # 执行编译操作，我的CPU有16个线程，所以我写16以加快编译速度
sudo make install # 安装 PCL
cd lib/ # 该文件夹在build文件夹下
sudo cp -d libpcl_* /usr/lib/x86_64-linux-gnu/ # 将这些so文件复制到之前清理过的这个目录，否则编译自己的程序时找不到so文件，-d参数是保持文件之间的软链接关系

花了50大洋买回来的教训

cudart

Linux 下编译安装软件，找不到共享库 xx.so 的解决办法

TODO

/usr/lib /lib/x86_64-linux-gnu/ /usr/lib/gcc/x86_64-linux-gnu/8/ 区别

i686-linux-gnu/ x86_64-linux-gnu/ 的区别