I have this code, parallel odd-even merge sort. I am now trying to optimize it more, and I think a good idea is to parallelize the for loop that is commented, as that comparations are independent. But, when I parallelize that loop, only worse performance is achieved. I printed the thread ID inside the loop, and only one thread is looping, even if I write num_threads(2)

#pragma omp parallel
{
    #pragma omp single
    oddEvenMergeSort(newArr, newSize, 0, limit);
}


void oddEvenMergeSort(int *arr, int n, int level, int limit) {
if (n > 1) {
    int m = n / 2;

    if (level < limit)
    {
        level += 1;
        #pragma omp task shared(arr)
        oddEvenMergeSort(arr, m, level, limit);

        #pragma omp task shared(arr)
        oddEvenMergeSort(arr + m, n - m, level, limit);

        #pragma omp taskwait
        oddEvenMerge(arr, n);
    }
    else {
        oddEvenMergeSort(arr, m, level, limit);

        oddEvenMergeSort(arr + m, n - m, level, limit);

        oddEvenMerge(arr, n);
    }
}

}

void oddEvenMerge(int *arr, int n) {
if (n == 1) return;
if (n == 2) {
    compare(arr, 0, 1);
    return;
}
else {
    int *odd = (int *)malloc(sizeof(int) * (n / 2));
    int *even = (int *)malloc(sizeof(int) * (n / 2));
    oddEvenSplit(arr, odd, even, n);

    oddEvenMerge(odd, n / 2);

    oddEvenMerge(even, n / 2);

    oddEvenJoin(arr, odd, even, n);


    // #pragma omp parallel for schedule(static, 1)
    for (int i = 1; i < n / 2; i++) {
        printf("Thread %dn", omp_get_thread_num());
        compare(arr, 2 * i - 1, 2 * i);
    }

    free(odd);
    free(even);
}

}