#include <math.h>
#include "mpi.h"
#include <stdio.h>
#define SIZE 1000000
int main(argc,argv)
int argc; char *argv[];
{
int i,nprocs, rank, remainder;
float *A, *B;
int *lsize,less;
float ldot,dot;
                   /*INITIALIZE THE MPI ENVIRONMENT*/
MPI_Init(&argc,&argv);
MPI_Comm_rank(MPI_COMM_WORLD,&rank);
MPI_Comm_size(MPI_COMM_WORLD,&nprocs);

/*   DEVIDE THE SIZE FOR PROCESSORS AND STORE IN AN ARRAY LSIZE OF SIZE
        NPROCS-1 */

      lsize = (int *)malloc(nprocs*sizeof(int));

      remainder = SIZE%nprocs;
      for (i=0;i<nprocs;i++){
                 if(i < remainder){
                       lsize[i] = 1+(SIZE-remainder)/nprocs;
                       }
                 else
                      lsize[i]=(SIZE-remainder)/nprocs;
                      }

          /*  ALLOCATE THE ARRAYS WITH REDUCED SIZE */      

      A = (float *)malloc(lsize[rank]*sizeof(float *));
      B = (float *)malloc(lsize[rank]*sizeof(float *));

      /* COMPUTE THE STARTING POINT FOR LOCAL WORK */

      less=0;
      if(rank > 0) for (i=0;i < rank; i++) less=less+lsize[i];

      /* INITIALIZE THE LOCAL ARRAY AFTER ADJUSTMENT */

      for(i=0;i<lsize[rank];i++){
                                A[i]=sin((float)(i+less+1));
                                B[i]=cos((float)(i+less+1));
              }

       /*  COMPUTE LOCAT DOT PRODUCT */

      ldot=0.0;
      for(i=0;i<lsize[rank];i++)ldot=ldot+A[i]*B[i];

       /* COMPUTE THE DOT PRODUCT AS IN SERIAL CODE */

      MPI_Allreduce(&ldot,&dot,1,MPI_FLOAT,MPI_SUM,MPI_COMM_WORLD);
      printf("dot= %f\n",dot);

       /* TERMINATE THE MPI ENVIRONMENT */

      MPI_Finalize();
}