B11 - orphaned directive

void work() {
    #pragma omp for
    for (int i = 0; i < 100; i++) {
        // Work
    }
}

int main() {
    #pragma omp parallel
    {
        work();
    }
}

• here, #pragma omp for inside work() is called an orphaned directive, because it isn't lexically enclosed by a parallel directive in the source code

• when the function executes, it is still inside the active parallel region

• static or lexical extent is the code physically enclosed by the parallel directive

• dynamic extent includes the code executed through function calls made from the parallel region

// dynamic extent
void work() {
    #pragma omp for
    for (int i = 0; i < 100; i++) {
        // Work
    }
}

// static extent
#pragma omp parallel
{
    work();
}

#include <stdio.h>
#include <omp.h>

void work() {
    #pragma omp for
    for (int i = 0; i < 8; i++) {
        printf("Thread %d executes i=%d\n",
               omp_get_thread_num(), i);
    }
}

int main() {
    #pragma omp parallel num_threads(4)
    {
        work();
    }

    return 0;
}

• all four threads call work(), but OpenMP distributes the eight loop iterations among them

• the function is called four times, but the loop is not executed four times in full

• if the function is called from outside any parallel region, the loop executes using the initial team of one thread

• only if called from within a parallel construct, threads are created