Multithreaded recursive fibonacci with OpenMP Tasks - c

Question

I know fibonacci is fundamentally sequential. But I just want to test OpenMP Tasks for the recursive implementation of fibonacci series. The following code in C works fine but my Problem is, instead of getting faster results with more threads, it gets worse. Why? You can try it on your self. I want best scalability.
Compile this code with "gcc -O3 -fopenmp -o fib fib.c" and run it.
#include <stdio.h>
#include <stdlib.h>
#include <omp.h>
double serialFib(int n, double* a) {
if (n < 2) {
a[n] = n;
return n;
}
double x = serialFib(n - 1, a);
double y = serialFib(n - 2, a);
a[n] = x + y;
return x + y;
}
double fib(int n, double* a) {
if (n < 2) {
a[n] = n;
return n;
}
if (n <= 30) { // avoid task creation overhead
return serialFib(n, a);
}
double x, y;
#pragma omp task shared(x, a) firstprivate(n)
{
x = fib(n - 1, a);
}
#pragma omp task shared(y, a) firstprivate(n)
{
y = fib(n - 2, a);
}
#pragma omp taskwait
a[n] = x + y;
return x + y;
}
int main(int argc, char *argv[]) {
double t0, t1;
// To test scalability of recursive approach
// we take N = 40. Otherwise it will take too long.
int N = 40, i, nthreads;
printf("Starting benchmark...\n");
nthreads = atoi(argv[1]);
omp_set_num_threads(nthreads);
double* a = (double *) calloc(N, sizeof(double));
t0 = omp_get_wtime();
#pragma omp parallel
{
#pragma omp single
{
fib(N, a);
}
}
t1 = omp_get_wtime();
for (i = 0; i < N; ++i) {
printf("a[%d] = %.2f\n", i, a[i]);
}
printf("Execution time: %f\n", t1 - t0);
free(a);
return 0;
}