C******************************************************************************
C FILE: arraypad.f
C DESCRIPTION:
C   Demonstrates array padding
C******************************************************************************

      PROGRAM PAD
      IMPLICIT NONE
      INTEGER N
      PARAMETER (N=256)
      REAL*8 A(N,N), AA(N+1,N)
      INTEGER I,J,K
      integer(8) T1, T2, IRTC 

C*****************************************************************************
C*                             Initialize Data                               *
C*****************************************************************************

      DO J=1,N
         DO I=1,N
            A(I,J)=1.23
            AA(I,J)=A(I,J)
         END DO
      END DO

C*****************************************************************************
C*                             Untuned                                      *
C*****************************************************************************

      T1 = IRTC()
      DO I=1,N
         DO J=1,N
            DO K=1,N
               A(I,J)=A(I,J)+A(J,K)-A(K,I)
            END DO
         END DO
      END DO
      T2 = IRTC()
      write(*,*)'Untuned version took', (T2-T1)/1000000, 'msec'


C*****************************************************************************
C*                             Tuned                                        *
C*****************************************************************************

      T1 = IRTC()
      DO I=1,N
         DO J=1,N
            DO K=1,N
               AA(I,J)=AA(I,J)+AA(J,K)-AA(K,I)
            END DO
         END DO
      END DO
      T2 = IRTC()
      write(*,*)'Tuned version took', (T2-T1)/1000000, 'msec'

C     Need to actually use the results of the calculations or else the
C     optimizing compiler may just skip doing them...so we'll just
C     print them.  This will force the optimizer to actually do the work.
      print *, ' '
      print *, 'Ignore this next line of output'
      PRINT *, A(1,1),A(128,128),A(256,256)
      PRINT *, AA(1,1),AA(128,128),AA(256,256)

      END