Collapse clause

The collapse clause is used to convert a prefect nested loop into a single loop then parallelize it. Check the following example:

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

int main(void)
{    
    #pragma omp parallel for
    for (int i = 0; i < 4; i++)
    {
        for (int j = 0; j < 5; j++)
        {
            printf("Thread number is %d\n", omp_get_thread_num());
        }
    }

    return 0;
}  

Compile and run it on my 24-core machine:

# gcc -fopenmp parallel.c
# ./a.out
Thread number is 0
Thread number is 0
Thread number is 0
Thread number is 0
Thread number is 0
Thread number is 3
Thread number is 3
Thread number is 3
Thread number is 3
Thread number is 3
Thread number is 1
Thread number is 1
Thread number is 1
Thread number is 1
Thread number is 1
Thread number is 2
Thread number is 2
Thread number is 2
Thread number is 2
Thread number is 2

Every iteration of outer loop will be dispatched to 1 every thread to run:

#pragma omp parallel for
for (int i = 0; i < 4; i++)
{
    ......
}

Every thread will execute the inner for-loop sequentially:

for (int j = 0; j < 5; j++)
{
    ......
}

So there are only 4 threads in active state actually.

Modify the directive from

#pragma omp parallel for

to

#pragma omp parallel for collapse(2)

Then build and run it again:

# gcc -fopenmp parallel.c
# ./a.out
Thread number is 0
Thread number is 2
Thread number is 18
Thread number is 16
Thread number is 6
Thread number is 8
Thread number is 7
Thread number is 10
Thread number is 14
Thread number is 12
Thread number is 13
Thread number is 17
Thread number is 15
Thread number is 9
Thread number is 11
Thread number is 19
Thread number is 4
Thread number is 3
Thread number is 5
Thread number is 1

This time we can see 20 threads are utilized. The integer after collapse (i.e., 2 in this example) identifies how many loops to be parallelized, and counted from outer side to inner side. Please be aware that collapse(1) and no collapse take the same effect for loop parallelism.