/* File:      global_sum_div0.c
 * Purpose:   Driver for MPI global sum function
 * Input:     None
 * Output:    Random values generated by processes and sum of these
 *            values.
 *
 * Algorithm: 
 *    1. Each process computes a random int x from 0 to MAX_CONTRIB-1
 *    2. Processes call Global_sum function which sums the local
 *       values on process 0
 *    3. Process 0 prints global sum
 *
 * Compile:  mpicc -g -Wall -o global_sum0 global_sum0.c
 * Run:      mpiexec -n <number of processes> global_sum0
 *
 * Notes:
 *    1. The number of processes should be a power of 2
 *    2. This version uses modular arithmetic rather than bitmasks
 *    3. The value returned by global_sum on processes other
 *       than 0 is meaningless.
 *    4. DEBUG flag can be used to print information on pairing
 *       of processes.
 */
#include <stdio.h>
#include <stdlib.h>
#include "mpi.h"

#define MAX_CONTRIB 20

int Global_sum(int myContrib, int my_rank, int p, MPI_Comm comm);

int main(void) {
   int p, my_rank;
   MPI_Comm comm;
   int x;
   int total;

   MPI_Init(NULL, NULL);
   comm = MPI_COMM_WORLD;
   MPI_Comm_size(comm, &p);
   MPI_Comm_rank(comm, &my_rank);

   srandom(my_rank);
   x = random() % MAX_CONTRIB;
   printf("Proc %d > x = %d, p = %d\n", my_rank, x, p);

   total = Global_sum(x, my_rank, p, comm);

   if (my_rank == 0)
      printf("The total is %d\n", total);

   MPI_Finalize();
   return 0;
}  /* main */


/*---------------------------------------------------------------
 * Function:  Global_sum
 * Purpose:   Compute global sum of values distributed across 
 *            processes
 * Input args:
 *    my_contrib: the calling process' contribution to the global sum
 *    my_rank:    the calling process' rank in the communicator
 *    p:          the number of processes in the communicator
 *    comm:       the communicator used for sends and receives
 *
 * Return val:  the sum of the my_contrib values contributed by
 *    each process.
 *
 * Algorithm:  Use tree structured communication, pairing processes
 *    to communicate.
 *
 * Notes:
 *    1. The value returned by global_sum on processes other
 *       than 0 is meaningless.
 *    2. The pairing of the processes is done using modular
 *       arithmetic.
 */
int Global_sum(int my_contrib, int my_rank, int p, MPI_Comm comm) {
   int sum = my_contrib;
   int temp;
   int divisor = 2;
   int proc_diff = 1;
   int partner;
   int done = 0;
   int i_send;

#  ifdef DEBUG
   int my_pass = -1;
   partner = -1;
   printf("Proc %d > partner = %d, pass = %d, p = %d, done = %d\n", 
      my_rank, partner, my_pass, p, done);
   fflush(stdout);
#  endif

   while (!done && divisor <= p) {
      i_send = my_rank % divisor;
      if (i_send) {
         partner = my_rank - proc_diff;
#        ifdef DEBUG
         my_pass++;
         printf("Proc %d > partner = %d, pass = %d, i_send = %d, sum = %d\n", 
            my_rank, partner, my_pass, i_send, sum);
         fflush(stdout);
#        endif
         MPI_Send(&sum, 1, MPI_INT, partner, 0, comm); 
         done = 1;
      } else {  /* I'm receiving */
         partner = my_rank + proc_diff;
#        ifdef DEBUG
         my_pass++;
         printf("Proc %d > partner = %d, pass = %d, i_send = %d \n", 
            my_rank, partner, my_pass, i_send);
         fflush(stdout);
#        endif
         MPI_Recv(&temp, 1, MPI_INT, partner, 0, comm, 
               MPI_STATUS_IGNORE);
         sum += temp;
#        ifdef DEBUG
         printf("Proc %d > partner = %d, pass = %d, received = %d, sum = %d\n", 
            my_rank, partner, my_pass, temp, sum);
         fflush(stdout);
#        endif
         divisor *= 2;
         proc_diff *= 2;
      }
   }

   /* Valid only on 0 */
   return sum;
}  /* Global_sum */