Writing parallel programs using the commonly used pattern

  • Recall: a commonly used parallel programming pattern is: 

       main( )
   {
       Prepare data1;
       Spawn N "worker" threads to process data1;
       Wait for all threads to complete processing

       Prepare data2;
       Spawn N "worker" threads to process data2;
       Wait for all threads to complete processing

       ....
   }

  • Graphical depiction of each step:

Writing parallel programs using the commonly used pattern

  • This parallel programming pattern is created as follows: 

    void *worker(int *id)  // Not syntactically correct, but easier to code...
{
   // Use *id  as int parameter (The official way is: void * and cast)
}

int main(int argc, char *argv[])
{
    pthread_t tid[N];
    int arg[N];

    /* ------------------------------------
       Create the worker thread group
       ------------------------------------ */
    for ( int i = 0; i < N; i++ )
    {
        arg[i] = i;

        pthread_create(&tid[i], NULL, worker, &arg[i]);
    }

    /* ---------------------------------------
       Wait for the worker threads to finish
       --------------------------------------- */
    for ( int i = 0; i < N; i++ )
        pthread_join(tid[i], NULL);
}

Using global variables to pass input to worker threads in multi-threaded programs

  • All worker threads in a multi-threaded program must coorperate to solve the problem:

    • The job is divided among all the worker threads

    • Each worker thread works on some part/portion of the input data

  • Commonly used organization to provide input data to worker threads:

    • The input data of the job is stored in global variables

      (Because global variables are accessible by all threads)