#include <mpi.h>
#include <stdio.h>
#include <malloc.h>
#include <string.h>

#define LOOP 300        /* Execute the message passing 300 times */

main(argc,argv)
int argc;
char **argv; 
{ int rank;
  int size;
  int i,j;
  int packetlength;
  char *message;
  char *rmessage;
  double start[LOOP], middle[LOOP], end[LOOP];
  double *mitimes;
  double avs, avr, av;


  if (argc<2) {
    printf("Syntax: gather <msg size in chars>\n");
    exit(-1);
  }
  
  MPI_Init(&argc,&argv);

  MPI_Comm_rank(MPI_COMM_WORLD,&rank);
  MPI_Comm_size(MPI_COMM_WORLD,&size);

  packetlength=atoi(argv[1]); /* param. 1 = mes. len */
  message=(char *)malloc(sizeof(char)*packetlength);
  rmessage=(char *)malloc(sizeof(char)*packetlength*size);

  memset(message, 'A'+rank, packetlength);
  memset(rmessage, '.', packetlength*size);

  /* rmessage[(packetlength*size)-1]=0;
  printf("%u [%s]\n", rank, rmessage); */

  mitimes=(double *)malloc(sizeof(double)*size);
  memset(mitimes, 0, sizeof(double)*size);

  /* if (!rank) printf ("Size : %d\n",size); */

  for (i=0; i < LOOP; i++) {
    if (!rank) {
      MPI_Barrier(MPI_COMM_WORLD);
      start[i]=MPI_Wtime();
      MPI_Gather(message, packetlength, MPI_CHAR,
                rmessage, packetlength, MPI_CHAR, 0, MPI_COMM_WORLD);
      end[i]=MPI_Wtime();
    }
    else {
      MPI_Barrier(MPI_COMM_WORLD);
      start[i]=MPI_Wtime();
      MPI_Gather(message, packetlength, MPI_CHAR,
                rmessage, packetlength, MPI_CHAR, 0, MPI_COMM_WORLD);
      end[i]=MPI_Wtime();
    }
  }

/*
   rmessage[packetlength*size-1]=0;
   printf("%u %s\n", rank, rmessage); */
 /* Calculate and print the overheads */ 

 if (!rank) {
  avs=avr=0.0;
  for (i=0; i < LOOP; i++) {
    avs=avs+end[i]-start[i];
  }

  avs=avs/LOOP;

  printf("%f ", avs);

  MPI_Gather(&avr, 1, MPI_DOUBLE, mitimes, 1, MPI_DOUBLE, 0, MPI_COMM_WORLD);

  avr=avs=av=mitimes[1];
  for(i=2; i<size; i++) {
    av=av+mitimes[i];
    if (mitimes[i]<avr) avr=mitimes[i];
    if (mitimes[i]>avs) avs=mitimes[i];
  }
  printf ("%f %f %f\n", avr, av/(size-1), avs);

 }
 else {
  avs=avr=0;
  for (i=0; i < LOOP; i++) {
    avr=avr+end[i]-start[i];
  }

  avr=avr/LOOP;

  MPI_Gather(&avr, 1, MPI_DOUBLE, mitimes, 1, MPI_DOUBLE, 0, MPI_COMM_WORLD);

 }


MPI_Finalize();

}