program pardotproduct
         include "mpif.h"
         INTEGER SIZE
         PARAMETER(SIZE=1000000)
         REAL,ALLOCATABLE:: A(:),B(:)
         INTEGER NPROCS, RANK, REMAINDER, I, J,IERR
         INTEGER stat(MPI_STATUS_SIZE) 
         INTEGER,ALLOCATABLE:: LSIZE(:)
         REAL LDOT,DOT
          
!        INITIALIZE THE MPI ENVIRONMENT

         CALL MPI_INIT(IERR)
         CALL MPI_COMM_RANK(MPI_COMM_WORLD, RANK,IERR)
         CALL MPI_COMM_SIZE(MPI_COMM_WORLD,NPROCS,IERR)

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

         REMAINDER= MOD(SIZE,NPROCS)
         ALLOCATE(LSIZE(0:NPROCS-1))
         DO I=0,NPROCS-1
         IF(I.LT.REMAINDER)THEN
         LSIZE(I)=1+(SIZE-REMAINDER)/NPROCS
         ELSE
         LSIZE(I)=(SIZE-REMAINDE)/NPROCS
         ENDIF
         ENDDO

!        ALLOCATE THE ARRAYS WITH REDUCED SIZE

         ALLOCATE (A(LSIZE(RANK)),B(LSIZE(RANK)))
!
!        COMPUTE THE STARTING POINT FOR LOCAL WORK
!
         LESS=0
         IF(RANK.GT.0)THEN
         DO I=0,RANK-1
         LESS=LESS+LSIZE(I)
         ENDDO
         ENDIF
!
!        INITIALIZE THE LOCAL ARRAY AFTER ADJUSTMENT 
!
         DO I=1,LSIZE(RANK)
         A(I)=SIN((I+LESS)*1.)
         B(I)=COS((I+LESS)*1.)
         ENDDO
!
!        COMPUTE LOCAT DOT PRODUCT
!
         LDOT=0.0
         DO I=1,LSIZE(RANK)
         LDOT=LDOT+A(I)*B(I)
         ENDDO
!
!        COMPUTE THE DOT PRODUCT AS IN SERIAL CODE
!
         CALL MPI_ALLREDUCE(LDOT,DOT,1,MPI_REAL,MPI_SUM,
     &   MPI_COMM_WORLD,ierr)
         print*,'DOT PRODUCT OF A and B IS',DOT

!        TERMINATE THE MPI ENVIRONMENT

         CALL MPI_FINALIZE(IERR)
         END