C******************************************************************************
C FILE: mem01.f
C DESCRIPTION:
C   Demonstrates effect of stride
C******************************************************************************

       program mem01

       IMPLICIT NONE
       integer ARRAY_SIZE
       parameter(ARRAY_SIZE = 1000)
       integer I, J
       real A(ARRAY_SIZE,ARRAY_SIZE), AA(ARRAY_SIZE,ARRAY_SIZE)
       real B(ARRAY_SIZE,ARRAY_SIZE), BB(ARRAY_SIZE,ARRAY_SIZE)
       real C(ARRAY_SIZE,ARRAY_SIZE), CC(ARRAY_SIZE,ARRAY_SIZE)
       integer(8) T1, T2, IRTC


C*****************************************************************************
C*                             Initialize Data                               *
C*****************************************************************************
       DO I = 1, 1000
         DO J = 1, 1000
           A(J,I) = RAND()
           B(J,I) = RAND()
           C(J,I) = RAND()
           AA(J,I) = RAND()
           BB(J,I) = RAND()
           CC(J,I) = RAND()
         END DO
       END DO
   
C*****************************************************************************
C*                             Stride 1000 Loop                             *
C*****************************************************************************
       T1 = IRTC()
       DO I = 1, 1000
         DO J = 1, 1000
           C(I,J) = C(I,J) + A(I,J) * B(I,J)
         END DO
       END DO
       T2 = IRTC()
       write(*,*)'Stride 1000 loop took', (T2-T1)/1000000, 'msec'

C*****************************************************************************
C*                             Stride 1 Loop                                *
C*****************************************************************************
       T1 = IRTC()
       DO J = 1, 1000
         DO I = 1, 1000
           C(I,J) = C(I,J) + A(I,J) * B(I,J)
         END DO
       END DO
       T2 = IRTC()
       write(*,*)'Stride 1 loop 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.
      I = RAND() * 10
      J = RAND() * 10
      print *, ' '
      print *, 'Ignore this next line of output'
      print *, A(I,J), B(I,J),C(I,J),AA(I,J),BB(I,J),CC(I,J)

      END