Can't get a performance boots when using openmp/gomp

[Ruud Schramp <spam2@xxxxxxxxxxx>]

Hi All,

I am trying to familiarize myself with openmp and gomp.

Just for testing purposes I made some small apps that I compile on a FC6 system which has gcc4.1 with gomp pacthes.


I noticed it is almost imposible to get a performance boost. I do notice that things are
in different threads when e.g. I parrellize a small loop of sleep(1)'s.

However when trying to optimize any of the examples from the several presentations
performance gain is negative. Even when making constructions with a big inner loop and
a small outerloop.

E.g.

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

int main ()
{
    char buf[10000];
    int i, q;
    int sum = 0;

#ifdef OMP
    omp_set_num_threads (2);
#endif

    for (q = 0; q < 2 ; q++)
    {
#pragma omp parallel for schedule(static,20)
        for (i = 0; i < 100; i++)
        {
            int j;
            int k = i;
            int sum = 0;
            for (j = 0; j < 10000000; j++)
                sum += j - k;
            buf[k] = sum;
        };
    };

    for (i = 0; i < 100; i++)
        sum += buf[i];

    printf ("%d", sum);

    return 0;
}




gcc -fopenmp -O2 -DOMP test2.c -o test2mp ; gcc -O2 test2.c -o test2; time ./test2mp; time ./test2
0
real    0m1.469s
user    0m2.700s
sys     0m0.000s
0
real    0m1.485s
user    0m1.484s
sys     0m0.000s


I read some threads on cache conflict on L1 and L2 caches, so I made the program not to access the global variables too often, even to take the iterator in a local variable as well. All made no significant difference.


Anyone can shed a light on what I am doing wrong?

Best regards,

Ruud