Not working OpenMP in C [closed] - c

Question

Closed. This question needs debugging details. It is not currently accepting answers.
Edit the question to include desired behavior, a specific problem or error, and the shortest code necessary to reproduce the problem. This will help others answer the question.
Closed 7 years ago.
Improve this question
I use a function with a multithread process in a while. The code runs fine sometimes but other times just stops. I have identified that the problem is in the multithread process. Im newbie with multithread and OpenMP... if someone have a tip to solve that... I'll be very grateful. xP
void paralelSim(PetriNet *p, Matrix *mconflit, int steps){
Matrix ring;
choicesRing(p, mconflit, &ring);
clock_t t;
t = clock();
while((ring.row!=0) && ((steps>0) || (steps == CONTINUE))){
omp_set_dynamic(0);
omp_set_num_threads(ring.col);
#pragma omp parallel shared(p)
{
firethread(p, ring.m[0][omp_get_thread_num()]);
}
if(steps != CONTINUE){
steps--;
}
choicesRing(p, mconflit, &ring);
}
t= clock() - t;
printf("%f seconds.\n",t,((float)t)/CLOCKS_PER_SEC);
printf("m:\n");
showMatrix(&p->m);
printf("steps: %d\n", p->steps);
}

Answer 1

Too many information are missing on your code snippet to be sure of anything, but I can at least give you some hints of what could go wrong...
You set the number of threads to an arbitrary value that, I assume, can be quite large. But that's not really in OpenMP's philosophy. Indeed, in OpenMP, you don't expect the number of threads to go far beyond the number of cores or hardware threads available on the machine. I'm sure the run time library can handle much more, but I'm also sure the performance penalty can be sever and I even suspect there is a limit to what it can manage.
You repeatedly forbid nested parallelism. I guess doing it once is enough, unless firethread() set it back on.
You time your runs with clock() which is a bad idea, since it times the CPU time of the current threads and all its children and sums it, instead of reporting the wall time. So you'll never see if you have any speed-up, and you'll even experience reports of slow-downs. Use omp_get_wtime() instead.
Your time printing statement is wrong with 2 values packed for only one in the format list.
Here is a tentative re-write of your code, which I can't compile or test, but which I feel more in-line with what one would expect of an OpenMP code. Maybe it will improve / solve your issue.
void paralelSim(PetriNet *p, Matrix *mconflit, int steps){
Matrix ring;
omp_set_dynamic(0);
choicesRing(p, mconflit, &ring);
double t = omp_get_wtime();
while((ring.row!=0) && ((steps>0) || (steps == CONTINUE))){
#pragma omp parallel for
for(int i=0; i<ring.col; ++i)
firethread(p, ring.m[0][i]);
if(steps != CONTINUE){
steps--;
}
choicesRing(p, mconflit, &ring);
}
t = omp_get_wtime() - t;
printf("%f seconds.\n",t);
printf("m:\n");
showMatrix(&p->m);
printf("steps: %d\n", p->steps);
}
Again, I didn't even compile this so there might (likely) be some typos. Moreover, should it work but not give expected performance, you could consider move the omp parallel part outside of the while loop, and use some omp single where needed.
Finally, since I didn't know how many cores you plan to run this on, I didn't explicitly set the number of threads. This will for the moment rely on either your environment's default, or your setting of the OMP_NUM_THREADS environment variable.