How can I make sure that OpenMP threads are running in parallel?

Question

I am working on some OpenMP program and I have the following code:

   #pragma omp parallel sections shared(a, b, r1, r2)
   {
        #pragma omp section
        {
           a = 1;
           r1 = b;
        }

        #pragma omp section
        {
           b = 1;
           r2 = a;
        }
   }

The main thing I want to observe here is that r1 = 1or r2 = 1 (this is allowed by the specification) but I have run this code snippet for more than 1000000 times and I cannot observe it. The only result I got is r1 = 0, r2 = 1 (I was expecting at least some of them will be r1 = 1, r2 = 0. I am wondering if the implementation is serialising my program. I used omp_set_num_threads(2); to force two threads, but how I can be sure that the two threads are actually run on two physical cores (I am using a dual-core Macbook pro, and I am using gcc-4.9 installed by homebrew).

Answer 1

Your example is perfectly fine. The problem is the minimal workload for the two threads. Section 1 is scheduled to thread #0 and Section 2 to thread #1. However, thread #0 finished his work before the thread #1 has started. It just look like sequential execution of your sections.

I modified your example:

#include <iostream>
#include <unistd.h>
int main()
{
    int a, b, r1=0, r2=0;
    #pragma omp parallel sections shared(a, b, r1, r2)
    {
        #pragma omp section
        {
            a = 1;
            sleep(10);
            std::cout << "section #1" << std::endl;
            r1 = b;
        }
        #pragma omp section
        {
            b = 1;
            sleep(5);
            std::cout << "section #2 "<< std::endl;
            r2 = a;
        }
   }
   std::cout << r1 << r2 << std::endl;
}

The output shows

section #2

section #1

11

Just as you might have expected.