c++

How can I thread the nested for-loop below safely in order to run the programme in parallel on a core with 8 threads and still output data in the correct order. I have tried using the #pragma omp for command but that gives me an error message: work-sharing region may not be closely nested inside of work-sharing, critical or explicit task region .

Note: This code is for an introduction to parallel programming, so it is poorly written for the sake of being optimised

#pragma omp parallel private(t, i, j) shared(nx, ny, nt)
{
    // main loop

    for (int t = 0; t < nt; t++)
    {
        cout << "n" << t;
        cout.flush();

        // first block 

        for (int i = 0; i < nx; i++)
        {
            for(int j=0; j < ny ;j++)
            {
                if (i> 0 && i < nx - 1 && j >0 && j < ny - 1) 
                {
                    vr[i][j] = (vi[i+1][j]+vi[i-1][j]+vi[i][j-1]+vi[i][j+1]) / 4.;
                } 
                else if (i == 0 && i < nx - 1 && j > 0 && j < ny - 1) 
                {
                    vr[i][j] = (vi[i+1][j]+10.+vi[i][j-1]+vi[i][j+1]) / 4.;
                } 
                else if (i > 0 && i == nx - 1 && j > 0 && j < ny - 1) 
                {
                    vr[i][j] = (5.+vi[i-1][j]+vi[i][j-1]+vi[i][j+1]) / 4.;
                } 
                else if (i > 0 && i < nx - 1 && j == 0 && j < ny - 1) 
                {
                    vr[i][j] = (vi[i+1][j]+vi[i-1][j]+15.45+vi[i][j+1]) / 4.;
                } 
                else if (i > 0 && i < nx - 1 && j > 0 && j == ny - 1) 
                {
                    vr[i][j] = (vi[i+1][j]+vi[i-1][j]+vi[i][j-1]-6.7) / 4.;
                }
            }
        }

        // second block

        for (int i = 0; i < nx; i++) 
        {
            for (int j = 0; j < ny; j++)
            {
                if (fabs(fabs(vr[i][j]) - fabs(vi[i][j])) < 1e-2) 
                {
                    fout << "n" << t << " " << i << " " << j << " "
                         << fabs(vi[i][j]) << " " << fabs(vr[i][j]);
                }
            }

            #pragma omp for schedule(static,100)

            // third block

            for (int i = 0; i < nx; i++) 
            {
                for (int j = 0; j < ny; j++) 
                {
                    vi[i][j] = vi[i][j] / 2. + vr[i][j] / 2.;
                }
            }
        }
    }

---

You cannot nest OMP regions in this way. From the OMP documentation (Intel):

Two OpenMP constructs are improperly (dynamically) nested. The OpenMP specification imposes several restrictions on how OpenMP constructs can be dynamically nested, that is, which OpenMP constructs can be legally encountered during execution of another region. OpenMP parallel regions can be nested within one another, but some restrictions apply. Generally speaking, two parallel regions can only be nested if there is an intermediate single threaded region, as created by a SINGLE, CRITICAL, or MASTER directive.

To be precise, the following restrictions apply. In the following, the term "worksharing region" is shorthand for any one of the following constructs: loop (FOR/DO), SECTIONS, SINGLE, or WORKSHARE. The term "closely nested region" means a region that is dynamically nested inside another region with no parallel region nested between them.

A worksharing region may not be closely nested inside a worksharing, explicit TASK, CRITICAL, ORDERED, or MASTER region.

A BARRIER region may not be closely nested inside a worksharing, explicit TASK, CRITICAL, ORDERED, or MASTER region.

A MASTER region may not be closely nested inside a worksharing or explicit TASK region.

An ORDERED region may not be closely nested inside a CRITICAL or explicit TASK region.

An ORDERED region must be closely nested inside a loop region (or parallel loop region) with an ORDERED clause.

A CRITICAL region may not be nested (closely or otherwise) inside a CRITICAL region with the same name (although violations of this restriction are reported as a different error type than this one).

Similar questions have been asked and answered before on SO.

OpenMP, for loop inside section

OpenMP for loop with master region: "master region may not be closely nested inside of work-sharing or explicit task region"

链接地址: http://www.djcxy.com/p/79238.html

上一篇: openmp等待内部for

下一篇: C ++