Question

我的代码如下，由于某些原因，我的程序似乎永远不会退出parallel区域。

我的代码进入

cout<< omp_get_thread_num()<<" done"<< endl;

行，但它只是由root（0）进程执行，它永远不会到达

cout<<"done2?"<< endl;

一部分。

它已进入无限循环或其他东西。对我来说，每个线程都没有cout这样做是没有意义的，因为我在它之前添加了omp barrier所以每个线程都必须在它之前到达它root甚至可以说已经完成了。感谢您的帮助。

#pragma omp parallel num_threads(desiredNumThreads)
{
    double startTime;
    #pragma omp master
    {
        startTime = omp_get_wtime();
    }           
    while (!Converged)
    {
        #pragma omp master
        {
            iteration++;
        }
        #pragma omp for
        for (int i = 0; i < matrixsize; i++)
        {
            oldX[i] = newX[i];
            //newX[i] = Bmatrix[i];
        }
        #pragma omp for
        for (int i = 0; i < matrixsize; i++)
        {
            newX[i] = Bmatrix[i]; //new part
            for (int j = 0; j < matrixsize; j++)
            {
                if (j != i)
                {
                    newX[i] = newX[i] - matrix[i][j] * oldX[j];
                }
            }
            newX[i] = newX[i] / matrix[i][i]; //new part
        }
        #pragma omp master
        {
            Converged = true;
        }
        #pragma omp barrier 
        #pragma omp for
        for (int i = 0; i<matrixsize; i++)
        {
            if (abs(oldX[i] - newX[i]) > Tolerance)
            {
                Converged = false;
            }
        }   
    }
    #pragma omp master
    {
        double endTime = omp_get_wtime();
        cout << "This took " << iteration << " iterations. " << endl;
        cout << "This took " << endTime - startTime << " seconds" << endl;
        cout << "File to Print answer to? ";
        cin >> filename;
        ofstream fout(filename);
        for (int i = 0; i < matrixsize; i++)
        {
            fout << fixed << newX[i] << endl;
        }                   
    }
    #pragma omp barrier 
    cout<<omp_get_thread_num()<<" done"<<endl;  
}
cout<<"done2?"<<endl;

Answer 1

根据MSDN，最大线程数为10.将desiredNumThreads更改为10或更少，它应该可以正常工作。

Answer 2

根据评论，代码在Debug配置中工作，但在Release配置中不起作用。这向我表明编译器正在进行优化以阻止代码运行。这些优化可能来自OpenMP内存模型，我建议阅读有关OpenMP内存模型的内容。原始代码中存在若干缺陷（语句中的竞争条件（＆＃34; Converged = true＆＃34;）。添加适当的同步将强制编译器正确更新变量Converged，并可能使问题消失。 / p>

干杯， -Michael

Answer 3

我通过移动最后一节来修复它

＆＃13;

#pragma omp master
    {
        double endTime = omp_get_wtime();
        cout << "This took " << iteration << " iterations. " << endl;
        cout << "This took " << endTime - startTime << " seconds" << endl;
        cout << "File to Print answer to? ";
        cin >> filename;
        ofstream fout(filename);
        for (int i = 0; i < matrixsize; i++)
        {
            fout << fixed << newX[i] << endl;
        }                   
    }

＆＃13;

离开并行部分并按顺序执行（因为它已经是顺序的）。为什么这解决了我无能为力的问题。

#pragma omp parallel永远不会结束

3 个答案: