C++多线程编程#pragma omp parallel

通常创建线程通过pthread_create来进行线程创建

创建线程

下面的程序，我们可以用它来创建一个 POSIX 线程：

#include <pthread.h>
pthread_create (thread, attr, start_routine, arg) 

在这里，pthread_create 创建一个新的线程，并让它可执行。下面是关于参数的说明：

创建线程成功时，函数返回 0，若返回值不为 0 则说明创建线程失败。

终止线程

使用下面的程序，我们可以用它来终止一个 POSIX 线程：

#include <pthread.h>
pthread_exit (status) 

在这里，pthread_exit 用于显式地退出一个线程。通常情况下，pthread_exit() 函数是在线程完成工作后无需继续存在时被调用

这样创建线程较为复杂而繁琐，下面介绍通过#pragma omp parallel简单而高效的创建线程

#pragma omp parallel创建线程

#pragma omp parallel通过定义代码块创建多线程，如下面的方式指定哪部分代码创建多线程

#include <omp.h>
int main(){
    print(“The output:\n”);
    #pragma omp parallel     /* define multi-thread section */
    {
        printf(“Hello World\n”);
    }
    /* Resume Serial section*/
    printf(“Done\n”);
}

 下面是一个创建多线程的实例：

 #include<stdio.h>
 #include<stdlib.h>
 
 void main(int argc, int *argv[]){
   int width = 1280;
   int height = 1280;
   float *imageBuffer = new float[3 * width* height];
   #pragma omp parallel for num_threads(3)
   {
      int tid = omp_get_thread_num();
      for(int i=0;i< width * height;i++){
            imageBuffer[i] = 0;
            imageBuffer[width * height + i] = 255;
            imageBuffer[width * height * 2 + i] = 0;
      }
   }
}

这种创建多线程的方式简单高效，但是有一点必须注意，#pragma omp parallel关键字创建多线程必须在编译时加上-fopenmp选

项，否则起不到并行的效果，

g++ a.cc -fopenmp

首先，如何使一段代码并行处理呢？omp中使用parallel制导指令标识代码中的并行段，形式为：

           #pragma omp parallel

           {

             每个线程都会执行大括号里的代码

            }

 如果想将for循环用多个线程去执行，可以用for制导语句

for制导语句是将for循环分配给各个线程执行，这里要求数据不存在依赖。

 使用形式为：

（1）#pragma omp parallel for

         for()

（2）#pragma omp parallel

        {//注意：大括号必须要另起一行

         #pragma omp for

          for()

        }

指定代码分块，每个分块开一个线程去执行，例如

 #pragma omp parallel sections // starts a new team
 {
   { Work1(); }
   #pragma omp section
   { Work2();
     Work3(); }
   #pragma omp section
   { Work4(); }
 }
or
 
 
 #pragma omp parallel // starts a new team
 {
   //Work0(); // this function would be run by all threads.
   
   #pragma omp sections // divides the team into sections
   { 
     // everything herein is run only once.
     { Work1(); }
     #pragma omp section
     { Work2();
       Work3(); }
     #pragma omp section
     { Work4(); }
   }
   
   //Work5(); // this function would be run by all threads.
 }

以shared,private的修饰为例：

#include <stdlib.h>   //malloc and free
#include <stdio.h>    //printf
#include <omp.h>      //OpenMP
 
// Very small values for this simple illustrative example
#define ARRAY_SIZE 8     //Size of arrays whose elements will be added together.
#define NUM_THREADS 4    //Number of threads to use for vector addition.
 
/*
 *  Classic vector addition using openMP default data decomposition.
 *
 *  Compile using gcc like this:
 *  	gcc -o va-omp-simple VA-OMP-simple.c -fopenmp
 *
 *  Execute:
 *  	./va-omp-simple
 */
int main (int argc, char *argv[]) 
{
	// elements of arrays a and b will be added
	// and placed in array c
	int * a;
	int * b; 
	int * c;
        
        int n = ARRAY_SIZE;                 // number of array elements
	int n_per_thread;                   // elements per thread
	int total_threads = NUM_THREADS;    // number of threads to use  
	int i;       // loop index
        
        // allocate spce for the arrays
        a = (int *) malloc(sizeof(int)*n);
	b = (int *) malloc(sizeof(int)*n);
	c = (int *) malloc(sizeof(int)*n);
 
        // initialize arrays a and b with consecutive integer values
	// as a simple example
        for(i=0; i<n; i++) {
            a[i] = i;
        }
        for(i=0; i<n; i++) {
            b[i] = i;
        }   
        
	// Additional work to set the number of threads.
	// We hard-code to 4 for illustration purposes only.
	omp_set_num_threads(total_threads);
	
	// determine how many elements each process will work on
	n_per_thread = n/total_threads;
	
        // Compute the vector addition
	// Here is where the 4 threads are specifically 'forked' to
	// execute in parallel. This is directed by the pragma and
	// thread forking is compiled into the resulting exacutable.
	// Here we use a 'static schedule' so each thread works on  
	// a 2-element chunk of the original 8-element arrays.
	#pragma omp parallel for shared(a, b, c) private(i) schedule(static, n_per_thread)
        for(i=0; i<n; i++) {
		c[i] = a[i]+b[i];
		// Which thread am I? Show who works on what for this samll example
		printf("Thread %d works on element%d\n", omp_get_thread_num(), i);
        }
	
	// Check for correctness (only plausible for small vector size)
	// A test we would eventually leave out
	printf("i\ta[i]\t+\tb[i]\t=\tc[i]\n");
        for(i=0; i<n; i++) {
		printf("%d\t%d\t\t%d\t\t%d\n", i, a[i], b[i], c[i]);
        }
	
        // clean up memory
        free(a);  free(b); free(c);
	
	return 0;
}

对于递归函数也可以使用task并行：

参考：http://akira.ruc.dk/~keld/teaching/IPDC_f10/Slides/pdf/4_Performance.pdf

openMP参考：https://www.cnblogs.com/mfryf/p/12744547.html

                        https://scc.ustc.edu.cn/zlsc/cxyy/200910/W020121113517997951933.pdf

参考：https://blog.csdn.net/zhongkejingwang/article/details/40350027

可参考：https://stackoverflow.com/questions/24417145/pragma-omp-parallel-num-threads-is-not-working

参考：https://people.cs.pitt.edu/~melhem/courses/xx45p/OpenMp.pdf