OpenMP and Nested for loops with continue statements

Question

I have the following the code:

int myArray[] = {0, 0, 0, 0, 0, 0};
double EV = 0;
for(short a1 = 1; a1 <= 6; ++a1)
{
    ++myArray[a1-1];
    if(....)
    {
        --myArray[a1-1];
        continue;
    }
    EV = myEVFunc();
    if(EV...)
    {

        for(short a2 = 1; a2 <=6 ; ++a2)
        {
            ++myArray[a2-1];
            if(....)
            {
                --myArray[a2-1];
                continue;
            }
            EV = myEVFunc();
            if(EV...)
            {
                for(short a3 = 1; a3 <= 6; ++a3)
                {
                    ++myArray[a3-1];
                    if(....)
                    {
                        --myArray[a3-1];
                        continue;
                    }
                    EV = myEVFunc();
                }
            }
        }
    }
}

I am trying to use OpenMP to parallelize the loops. the code compiles fine when i place #pragma omp parallel for in front of the outermost for loop. However it gives incorrect results. I suspect two issues the continue statements inside the loops and the fact that there are shared variables in the nested loops. Is it possible to use OpenMP with this code snippet, if so can anyone please give me the correct syntax. Thanks in advance.

Mysticial · Accepted Answer · 2012-02-01 05:12:33Z

2

I'll point out a couple obvious things:

1.) double EV = 0; is declared outside the outer loop. Therefore it will be shared by all the threads. So you'll have a race condition at EV = myEVFunc(); and at each access to EV.

The solution to this is to declare it inside the loop. That will make it private to each thread.

#pragma omp parallel for
    for(short a1 = 1; a1 <= 6; ++a1)
    {
        ++myArray[a1-1];
        if(....)
        {
            --myArray[a1-1];
            continue;
        }
        double EV = myEVFunc();
        if(EV...)
        ...

2.) Another (sorta) issue is that your outer-loop only has 6 iterations. So you won't get more than 6 threads. Furthermore, you could get load-balancing issues with say 4 cores...

edited Feb 1, 2012 at 5:12

answered Feb 1, 2012 at 5:03

Mysticial

473k46 gold badges343 silver badges337 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

Manos Papadakis Over a year ago

You could use "#pragma omp atomic" for the ++ and -- statements.

Robert Hembree · Accepted Answer · 2012-02-01 05:36:57Z

1

You can also use thread private if that is more of what you need

It will give a private copy of the variable EV to each thread. Initially it will be uninitialized in these threads unless you use copyin as well.

With out copyin the directive would look like:

#pragma omp parallel for private(EV)

This avoids the race condition and allows EV to act independntly in each thread

edited Feb 1, 2012 at 5:36

answered Feb 1, 2012 at 5:23

Robert Hembree

831 silver badge7 bronze badges

Comments

Bort · Accepted Answer · 2012-02-01 12:28:16Z

0

you also have a race condition with the array access e.g. ++myArray[a1-1]; and ++myArray[a2-1]; from different threads. additionally you need to declare all loop counter private, a1 (is default private because of pragma omp for, but not), a2, a3..

Maybe you can find another way of writing your loops because all these conditional statements within the loop is usually not very efficient.

answered Feb 1, 2012 at 12:28

Bort

2,49115 silver badges24 bronze badges

Collectives™ on Stack Overflow

OpenMP and Nested for loops with continue statements

3 Answers 3

1 Comment

Comments

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

3 Answers 3

1 Comment

Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Related