使用并行任务减少 OpenMP

Question

I tried to use parallel task to perform reduction, but always get 0. Here is my code:

int sum = 0;
#pragma omp parallel reduction(+:sum)
#pragma omp single
for(int i=0; i<10; i++)
{
    #pragma omp task
    {
        printf("Thread: %d\n", omp_get_thread_num());
        int y = 5;
        sum += y;
    }
}
printf("%d\n", sum);

But when I use parallel for, the result is right, which is 50. Can anyone tell me how to modify the parallel task code? This is my parallel for code, which works well:

int sum = 0;
#pragma omp parallel for reduction(+:sum)
for(int i=0; i<10; i++)
{
    printf("Thread: %d\n", omp_get_thread_num());
    int y = 5;
    sum += y;
}
printf("%d\n", sum);

Answer 1

According to the OpenMP standard 4.5 , you cannot reduce a variable that is used in the task constructor.

A list item that appears in a reduction clause of the innermost enclosing worksharing or parallel construct may not be accessed in an explicit task.

Nevertheless, it looks like that feature will be covered by OpenMP 5.0. Looking at your code the better approach is actually to use the parallel for with the reduction clause, anyway.

Answer 2

dreamcrash's answer is spot on. OpenMP 5.0 has added task reduction support, but OpenMP 5.0 is not widely supported yet, afaik only the Intel compiler supports a preview of OpenMP 5.0 that includes task reduction.

If you really must use reduction with tasks with OpenMP previous to 5.0, you can basically declare the reduction variable as threadprivate and do a manual reduction as outlined in this talk .

It would look something like this:

int sum = 0;
int thread_sum = 0;
#pragma omp threadprivate(sum)
#pragma omp parallel
{
    #pragma omp single
    for(int i=0; i<10; i++)
    {
        #pragma omp task
        {
            printf("Thread: %d\n", omp_get_thread_num());
            int y = 5;
            thread_sum += y;
        }
    }
    #pragma omp atomic
    sum += thread_sum
}