Question

How can I speed up a recursive code using OpenMP? Basically, I have to speed up the S function in the program here. The code is as below.

void S(Oct* oct, unsigned int l) {
    S(oct, l+1);
    A(oct, l);
    S(oct, l+1);
    AR(oct,l);
}

Answer 1

Looking at your code, I would say the code is not parallelizable. Since S() is a recursive function, adding openmp pragma inside the function would lead to overhead of creating threads. It wouldn’t really improve the performance. Moreover, you shouldn’t parallelize those other functions, A() and AR(), as well. It will also create the same overhead performance problem.

I would suggest either to parallelize the source code that invokes S() in the first place, or to break down the code to eliminate recursion (then you might have a possibility to parallelize using openmp).