Set number of threads using omp_set_num_threads() to 2, but omp_get_num_threads() returns 1

Question

I have the following C/C++ code using OpenMP:

    int nProcessors=omp_get_max_threads();
    if(argv[4]!=NULL){
        printf("argv[4]: %s\n",argv[4]);
        nProcessors=atoi(argv[4]);
        printf("nProcessors: %d\n",nProcessors);
    }
    omp_set_num_threads(nProcessors);
    printf("omp_get_num_threads(): %d\n",omp_get_num_threads());
    exit(0);

As you can see, I'm trying to set the number of processors to use based on an argument passed on the command line.

However, I'm getting the following output:

argv[4]: 2   //OK
nProcessors: 2   //OK
omp_get_num_threads(): 1   //WTF?!

Why isn't omp_get_num_threads() returning 2?!!!

As has been pointed out, I'm calling omp_get_num_threads() in a serial region, hence the function returns 1.

However, I have the following parallel code:

#pragma omp parallel for private(i,j,tid,_hash) firstprivate(firstTime) reduction(+:nChunksDetected)
    for(i=0;i<fileLen-CHUNKSIZE;i++){
        tid=omp_get_thread_num();
        printf("%d\n",tid);
        int nThreads=omp_get_num_threads();
        printf("%d\n",nThreads);
...

which outputs:

0   //tid
1   //nThreads - this should be 2!
0
1
0
1
0
1
...

Are you calling this from a serial portion of your program? If yes, the number of threads is actually 1. — stephan, Jan 23 '12 at 09:58
He stephan - many thanks for that insight. I've edited the OP with another related question. — Eamorr, Jan 23 '12 at 10:06
If you use Intel IPP libraries with OpenMP enabled, I've seen some cases in which they collide... — Diego Sevilla, Jan 23 '12 at 10:18
That's interesting. I'm using Intel thread building blocks... — Eamorr, Jan 23 '12 at 10:20

score 33 · Accepted Answer · edited May 19 '17 at 06:09

33

The omp_get_num_threads() call returns 1 in the serial section of the code. See Link

So you need to have parallel code to get the correct value, here how your code should look like:

#include <iostream>
#include <omp.h>

int main (int argc, const char * argv[])
{
    int nProcessors = omp_get_max_threads();

    std::cout<<nProcessors<<std::endl;

    omp_set_num_threads(nProcessors);

    std::cout<<omp_get_num_threads()<<std::endl;

#pragma omp parallel for 
    for(int i = 0; i < 5; i++){
        int tid = omp_get_thread_num();
        std::cout<<tid<<"\t tid"<<std::endl;
        int nThreads = omp_get_num_threads();
        std::cout<<nThreads<<"\t nThreads"<<std::endl;
    }

    exit(0);
}

This code produces:

2

1
0    tid
2    nThreads
0    tid
2    nThreads
0    tid
2    nThreads
1    tid
2    nThreads
1    tid
2    nThreads

It seems that you have either open mp not enabled or your loop is not in the form that can be parallized by openmp

edited May 19 '17 at 06:09

meJustAndrew

6,011
8
50
76

answered Jan 23 '12 at 09:58

tune2fs

7,605
5
41
57

OK, but when I go into a parallel region, it's still returning 1... Please see my edit. Many thanks... – Eamorr Jan 23 '12 at 10:03
Where are you starting your parallel region, i do not see it in the edit. – tune2fs Jan 23 '12 at 10:05
Thanks, do you change the variable i or fileLen inside your parallel region? As it might be that your for loop cannot be parallised by openmp, due side effects. – tune2fs Jan 23 '12 at 10:09
No - don't change those variables – Eamorr Jan 23 '12 at 10:11
2

I have edited my code above and it works for me, please show some minimal running example where your code produces this error. – tune2fs Jan 23 '12 at 10:21
2

Hey, thanks for that. It turns out that I needed to pass -fopenmp to LIBS and CXXFLAGS in my Makefile... ;( – Eamorr Jan 23 '12 at 10:36

score 13 · Answer 2 · answered Jan 23 '12 at 12:44

13

you are using the wrong function. use omp_get_max_threads to check for the maximum number of allowed threads.

answered Jan 23 '12 at 12:44

Bort

2,423
14
22

score 3 · Answer 3 · answered Mar 13 '17 at 09:29

It has been already pointed out that omp_get_num_threads() returns 1 in sequential sections of the code. Accordingly, even if setting, by omp_set_num_threads(), an overall number of threads larger than 1, any call to omp_get_num_threads() will return 1, unless we are in a parallel section. The example below tries to clarify this point

#include <stdio.h>

#include <omp.h>

int main() {

    const int maxNumThreads = omp_get_max_threads();

    printf("Maximum number of threads for this machine: %i\n", maxNumThreads);

    printf("Not yet started a parallel Section: the number of threads is %i\n", omp_get_num_threads());

    printf("Setting the maximum number of threads...\n");
    omp_set_num_threads(maxNumThreads);

    printf("Once again, not yet started a parallel Section: the number of threads is still %i\n", omp_get_num_threads());

    printf("Starting a parallel Section...\n");

#pragma omp parallel for 
    for (int i = 0; i < maxNumThreads; i++) {
        int tid = omp_get_thread_num();
        printf("This is thread %i announcing that the number of launched threads is %i\n", tid, omp_get_num_threads());
    }

}

Set number of threads using omp_set_num_threads() to 2, but omp_get_num_threads() returns 1

3 Answers3

Linked