parallelizing C++ code with OpenMP doesn't improove performance

Question

I've write a C++ code that compute some number. I'm trying to use the OpenMP library to parallelize it. It has 3 nested for loops, and the parallelized one is the outer. I work on Linux with G++ compiler. The code works, but the problem is that the parallelization hasn't improoved performance. Launching it with with 2 threads in my dual core laptop (intel i5 processor with multithreading switched off) takes a bit more time that executing it with just one.

Here it is the code (launched with: g++ source.cpp -fopenmp):

#include <time.h>
clock_t tStart = clock();
#include <stdlib.h>
#include <iostream>
#include <iomanip>
#include <omp.h>
using namespace std;

int nThreads= 2; // switch to change the number of threads.

int main () {
    int i, j, k, nz= 10;
    double T= 1000, deltaT= 0.005, g= 9.80665, h=10, dz= h/nz, t, z, V, A;
    int nT= T/deltaT;

    #pragma omp parallel for private (j, k, t, z, V, A) num_threads(nThreads)
    for (i=0; i<=nT; i++) {
        t= i*deltaT;
        for (j=0; j<=nz; ++j) {
            z= dz*j;
            for (k=0; k<=1000; k++) {
                V= t*z*g*k;
                A= z*g*k;
            }
        }
    }

    cout << "Time taken: " << setprecision(5) <<  (double)(clock() - tStart)/CLOCKS_PER_SEC << endl;
    return 0;
}

possible duplicate of OpenMP time and clock() calculates two different results — High Performance Mark
– High Performance Mark, Commented May 21, 2015 at 9:49
Pretty hard to say anything meaningful about this code, you are doing it wrong if you don't get 0. There's is nothing to parallelize, the compiler will remove the for-loops and simply compute V = nT * deltaT * dz * nz * g * 1000 since that is all that is needed. — Hans Passant
– Hans Passant, Commented May 21, 2015 at 10:11
how do you measure the execution time? If you measure the cpu time, what you got is to be expected. openmp adds some overhead and the calculations you are doing take in total roughly the same time as on one cpu. You should measure the real time. — 463035818_is_not_an_ai
– 463035818_is_not_an_ai, Commented May 21, 2015 at 10:31

teo2070 · Accepted Answer · 2015-05-21 12:40:40Z

1

Ok, perfect. I was measuring cpu time, and the result was coherent. time() function gives me the number i need to know. Problem solved. Thanks everyone.

answered May 21, 2015 at 12:40

teo2070

234 bronze badges

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

Collectives™ on Stack Overflow

parallelizing C++ code with OpenMP doesn't improove performance

1 Answer 1

Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related