Question

I have a problem i would like parallelize two for loops with openmp.

how to optimize this loop with openmp :

void test(float** m,tab* t,int n){

    int i,j;
    float gain;
    for (i = 1; i < n; i++)
    {
        for (j = i + 1;] j < n; j++)
        {
            if (i != j)
            {
                gain=m[t[(i+n-1)%n]][t[j]] + m[t[i]][t[(j+1)%n]] - m[t[(i+n -1)%n]][t[i]] - m[t[j]][t[(j+1)%n]] 
                if (gain< 0)
                {
                    swapTab(t,i,j);
                }
            }
        }
    }
}

thx.

Answer 1

As Oli Charlesworth points out the order of traversal of the elements in your matrices matters, so simply slapping a parallel for directive around the outermost loop won’t work.

One option you do have would be to trade some space for time. Make a copy of your array T (call it T'). In iteration 1 you would replace your line:

swapTab(t,i,j)

with something like

 T'[j][i] = T[i][j]

(I’m not much of a C++ programmer so ignore defects in the syntax.)

In the second iteration you would want first to copy T' back to T and then carry on. DON’T do that — set up pointers (Tnew, Told, perhaps) to T and T' and switch the pointers around so that Told always points to the array to be read, and Tnew always points to the array to be written.