Question

The following code produces wrong and inconsistent output with gcc (4.1.2 20080704) but correct and expected output with icc (Version 11.1) . But when I moved the thread_data_array[] definition from main() to global (immediately after the struct thread_data definition) it works fine with both compilers. I do not see why this change should make any difference. I would like to generate threads recursively so I need to call it from a function but not define as global. Could someone explain what is wrong the code please?

#include <pthread.h>
#include <stdio.h>
#include <stdlib.h>

#define NUM_THREADS 4

struct thread_data {
    int  thread_id;
    int  sum;
};

/* struct thread_data thread_data_array[NUM_THREADS]; */

void *p_task(void *threadarg)
{
    struct thread_data *my_data;
    int taskid;
    int sum;

    my_data = (struct thread_data *) threadarg;
    taskid = my_data->thread_id;
    sum = my_data->sum;

    printf("Thread #%d with sum %d\n", taskid, sum);

    for ( sum = 0; sum < 000000000; sum++ ) {
        for ( taskid = 0; taskid < 000000000; taskid++ ) {
            sum+=taskid;
        }
    }

    return my_data;
}

int main ()
{
    struct thread_data thread_data_array[NUM_THREADS]; /*this does not work*/

    pthread_t threads[NUM_THREADS];
    int rc;
    long t;

    for ( t = 0; t < NUM_THREADS; t++ ) {
        thread_data_array[t].thread_id = t;
        thread_data_array[t].sum = (int) t*2;
        rc = pthread_create( &threads[t], NULL, p_task, (void *) &thread_data_array[t] );
    }

    pthread_exit(NULL);

    return 0;

}

Answer 1

Well, my first thought would be that the main thread is actually exiting, and taking its stack frame with it, before the other threads are finished. From what I remember about pthreads (and this is over a decade and a half ago, using them with DCE), there’s absolutely nothing special about the main thread other than the stack size.

If the array is declared global, the exit from main will have no effect on it but, if it’s in the stack frame for main, I’d be very careful using it after it exits.

I’m basing this partially on guesswork since you haven’t actually explained what the behavior is that you’re seeing.

I would suggest inserting the following code before the pthread_exit in main:

for (t = 0; t < NUM_THREADS; t++ )
    pthread_join (threads[t], NULL);