4、OpenMP的临界区critical、atomic、omp_lock_t

基本思想：OpenMP和多线程的互斥锁很像

#pragma omp parallel for
    for(int i=0;i<num;i++)
    {
#pragma omp critical
//#pragma omp critical(sum)
//#pragma omp critical(i)
        sum=sum+i;
    }
    std::cout<<"sum="<<sum<<std::endl;


}

测试代码 :critical 可以对代码块进行临界区设置，而atomic只能对代码语句进行加持

#include <iostream>
#include <omp.h>
#include<chrono>
using namespace std;
using namespace chrono;


void sequentialProgram(int num)
{
    int sum=0;
    int doubleI=0;
    for(int i=0;i<num;i++)
    {
        doubleI=i+i;
        sum=doubleI+sum;
    }
    std::cout<<"sum="<<sum<<std::endl;
}

void  parallelProgram(int num) {


    int sum=0;
    int doubleI=0;
#pragma omp parallel for
    for(int i=0;i<num;i++)
    {
        { doubleI=i+i;
            sum=doubleI+sum;}

    }
    std::cout<<"sum="<<sum<<std::endl;
     sum=0;
    doubleI=0;

#pragma omp parallel for
    for(int i=0;i<num;i++)
    {
#pragma omp critical
//#pragma omp critical(sum)
//#pragma omp critical(doubleI)
        { doubleI=i+i; //故意增加一个doubleI变量
            sum=doubleI+sum;}

    }
    std::cout<<"sum="<<sum<<std::endl;


}

int main() {


    int num=omp_get_num_procs()*10;//12

    auto start_time=std::chrono::steady_clock::now();
    sequentialProgram(num);
    auto end_time=std::chrono::steady_clock::now();
    std::cout<<"sequentialProgram elapse time: "<<std::chrono::duration<double>(end_time-start_time).count()<<" seconds"<<std::endl;

    start_time=std::chrono::steady_clock::now();
    parallelProgram(num);
    end_time=std::chrono::steady_clock::now();
    std::cout<<"parallelProgram elapse time: "<<std::chrono::duration<double>(end_time-start_time).count()<<" seconds"<<std::endl;
    return 0;
}

测试结果：发现不添加临界区，计算的值并不正确，因此在累加和累乘运算过程中，需要临界区的加持

F:\OpenMP\cmake-build-debug\OpenMP.exe
sum=14280
sequentialProgram elapse time: 0.0004372 seconds
sum=13790
sum=14280
parallelProgram elapse time: 0.0023326 seconds

Process finished with exit code 0

如果将错误的代码注释掉，会发现临界区的执行时间还是长于串行求和的程序的，因此并不划算~ 具体方法参考reduction

F:\OpenMP\cmake-build-debug\OpenMP.exe
sum=14280
sequentialProgram elapse time: 0.0005795 seconds
sum=14280
parallelProgram elapse time: 0.0031314 seconds

Process finished with exit code 0

（2）atomic 对代码语句进行临界设置

 atomic 只能针对单条语句线程互斥访问，它的使用和 #pragra omp parallel for 一样 只能直接接语句 不能在套一个大括号

测试代码

#include <iostream>
#include <omp.h>
#include<chrono>
using namespace std;
using namespace chrono;


void sequentialProgram(int num)
{
    int sum=0;
    int doubleI=0;
    for(int i=0;i<num;i++)
    {
        doubleI=i+i;
        sum=doubleI+sum;
    }
    std::cout<<"sum="<<sum<<std::endl;
}

void  parallelProgram(int num) {


    int sum=0;
    int doubleI=0;


#pragma omp parallel for
    for(int i=0;i<num;i++)
    {
#pragma omp atomic
        sum=i+i+sum;
        
    }
    std::cout<<"sum="<<sum<<std::endl;


}

int main() {


    int num=omp_get_num_procs()*10;//12

    auto start_time=std::chrono::steady_clock::now();
    sequentialProgram(num);
    auto end_time=std::chrono::steady_clock::now();
    std::cout<<"sequentialProgram elapse time: "<<std::chrono::duration<double>(end_time-start_time).count()<<" seconds"<<std::endl;

    start_time=std::chrono::steady_clock::now();
    parallelProgram(num);
    end_time=std::chrono::steady_clock::now();
    std::cout<<"parallelProgram elapse time: "<<std::chrono::duration<double>(end_time-start_time).count()<<" seconds"<<std::endl;
    return 0;
}

测试结果

F:\OpenMP\cmake-build-debug\OpenMP.exe
sum=14280
sequentialProgram elapse time: 0.0005045 seconds
sum=14280
parallelProgram elapse time: 0.0015284 seconds

Process finished with exit code 0

其中可以使用原子操作的运算符有 +、-、*、/、|、^、&、>>、<<

(3)添加锁

static omp_lock_t m_lock;
omp_init_lock(&m_lock);
#pragma omp parallel for
    for(int i=0;i<num;i++)
    {
        omp_set_lock(&m_lock);
        {
....
}
   omp_unset_lock(&m_lock);
    }
   
omp_destroy_lock(&m_lock);

测试代码

#include <iostream>
#include <omp.h>
#include<chrono>
using namespace std;
using namespace chrono;

static omp_lock_t m_lock;
void sequentialProgram(int num)
{
    int sum=0;
    int doubleI=0;
    for(int i=0;i<num;i++)
    {
        doubleI=i+i;
        sum=doubleI+sum;
    }
    std::cout<<"sum="<<sum<<std::endl;
}

void  parallelProgram(int num) {

omp_init_lock(&m_lock);
    int sum=0;
    int doubleI=0;


#pragma omp parallel for
    for(int i=0;i<num;i++)
    {
        omp_set_lock(&m_lock);
        doubleI=i+i;
        sum=doubleI+sum;
        omp_unset_lock(&m_lock);

    }
    std::cout<<"sum="<<sum<<std::endl;
omp_destroy_lock(&m_lock);

}

int main() {


    int num=omp_get_num_procs()*10;//12

    auto start_time=std::chrono::steady_clock::now();
    sequentialProgram(num);
    auto end_time=std::chrono::steady_clock::now();
    std::cout<<"sequentialProgram elapse time: "<<std::chrono::duration<double>(end_time-start_time).count()<<" seconds"<<std::endl;

    start_time=std::chrono::steady_clock::now();
    parallelProgram(num);
    end_time=std::chrono::steady_clock::now();
    std::cout<<"parallelProgram elapse time: "<<std::chrono::duration<double>(end_time-start_time).count()<<" seconds"<<std::endl;
    return 0;
}

测试结果 如果计算量很大的话，这样写会消耗的时间高于串行程序

F:\OpenMP\cmake-build-debug\OpenMP.exe
sum=14280
sequentialProgram elapse time: 0.0341926 seconds
sum=14280
parallelProgram elapse time: 0.0020649 seconds

Process finished with exit code 0