/* File:      global_sum2.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_sum global_sum.c
 * Run:      mpiexec -n <number of processes> global_sum
 *
 * Notes:
 *    1. This version has no restriction on the number of processes.
 *    2. The value returned by global_sum on processes other
 *       than 0 is meaningless.
 *    3. 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\n", my_rank, x);

   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 bitwise
 *       exclusive or.  Here's a table showing the rule for
 *       for bitwise exclusive or
 *           X Y X^Y
 *           0 0  0
 *           0 1  1
 *           1 0  1
 *           1 1  0
 *       Here's a table showing the process pairing with 8
 *       processes (r = my_rank, other column heads are bitmask)
 *           r     001 010 100 
 *           -     --- --- ---
 *           0 000 001 010 110  
 *           1 001 000  x   x
 *           2 010 011 000  x
 *           3 011 010  x   x
 *           4 100 101 110 000
 *           5 101 100  x   x
 *           6 110 111 100  x
 *           7 111 110  x   x              
 */
int Global_sum(int my_contrib, int my_rank, int p, MPI_Comm comm) {
    int sum = my_contrib;
    int temp;
    int partner;
    int done = 0;
    unsigned bitmask = (unsigned) 1;

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

    while (!done && bitmask < p) {
        partner = my_rank ^ bitmask;
#       ifdef DEBUG
        my_pass++;
        printf("Proc %d > partner = %d, bitmask = %d, pass = %d\n", 
           my_rank, partner, bitmask, my_pass);
        fflush(stdout);
#       endif
        if (my_rank < partner) {
            if (partner < p) {
                MPI_Recv(&temp, 1, MPI_INT, partner, 0, comm, 
                      MPI_STATUS_IGNORE);
                sum += temp;
            }
            bitmask <<= 1;
        } else {
            MPI_Send(&sum, 1, MPI_INT, partner, 0, comm); 
            done = 1;
        }
    }

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