At 04:16 02-12-01 +0000, Jun Xiong wrote:


>Hello,
>     I'm a novice in Fortran 90. In my opinion, once an allocatable or
>pointer array is allocated, it performs as efficient, i.e. fast, as a
>static or automatic array. But the fact surprises me. I wrote a simple test
>program as below:
>
>PROGRAM test
>    IMPLICIT NONE
>    INTEGER, PARAMETER :: N=5000
>    REAL, DIMENSION(N, N) :: x1, x2
>    REAL, DIMENSION(:, :), ALLOCATABLE :: y1, y2
>    INTEGER :: i
>    REAL, DIMENSION(0:10) :: t
>
>    allocate(y1(N, N), y2(N, N))
>
>    t(0)=secnds(0.0)
>
>    call random_number(x1)
>    t(1)=secnds(0.0)
>    call random_number(y1)
>    t(2)=secnds(0.0)
>
>    x2=x1**2
>    t(3)=secnds(0.0)
>    y2=y1**2
>    t(4)=secnds(0.0)
>
>    do i=1, 3, 2
>       write(*, *) t(i)-t(i-1), t(i+1)-t(i), (t(i+1)-t(i))/(t(i)-t(i-1))
>    enddo
>
>    deallocate(y1, y2)
>END PROGRAM test
>
>     I compiled it with Compaq Visual Fortran 6.6 under Win2000
>professional, and executed it for several times. The results are almost the
>same as:
>3.304688   3.765625   1.139480
>0.000000   0.9804688  Infinity
>
>My questions are:
>(1) Why do the allocatable arrays y1 and y2 need more time to
>accomplish the same operations than the static arrays x1 and x2 in this
>program?
>(2) Do allocatable or pointer arrays ALWAYS perform at least 10%
>percent slower than static or automatic arrays?
>(3) If the answer to the question (2) is positive, any suggestions
>concerning the efficiency when using allocatable or pointer arrays?
>
>Sincerely,
>Jun Xiong


I cannot provide you with a general answer to these questions.
The subject arises every now and then on this list.

But I did the test again with the slightly modified program
to prevent cpu time beging zero.
The answer to qustion 2 is NO.

!   [Jun Xiong]     2001-12-02     timing.f95
  PROGRAM test
    IMPLICIT NONE
    INTEGER, PARAMETER :: N=1000
    REAL, DIMENSION(N, N) :: x1, x2
    REAL, DIMENSION(:, :), ALLOCATABLE :: y1, y2
    INTEGER :: i
    REAL, DIMENSION(0:10) :: t

    allocate(y1(N, N), y2(N, N))

    call cpu_time(t(0))
    call cpu_time(t(0))
    do i = 1, 10
      call random_number(x1)
      x1(n,n-1) = i
    end do
    call cpu_time(t(1))
    do i = 1, 10
      call random_number(y1)
     y1(n,n-1) = i
    end do
    call cpu_time(t(2))

    do i = 1, 10
      x2=x1**2
      x2(n,n-1) = i
    end do
    call cpu_time(t(3))
    do i = 1, 10
      y2=y1**2
     y2(n, n-1) = i
    end do
    call cpu_time(t(4))

    do i = 1, 3, 2
       if(t(i) /= t(i-1) ) &
       write(*, *) t(i)-t(i-1), t(i+1)-t(i), (t(i+1)-t(i))/(t(i)-t(i-1))
    enddo

    deallocate(y1, y2)
  END PROGRAM test

! Results from one system:
!E:\Fortran\Test>atiming
!    1.15527  0.834961  0.722739
!   0.821289  0.968750  1.17955

! Another compiler:
!E:\Fortran\Test>stiming
!      2.36000         1.71000        0.724576
!     0.710000        0.710000         1.00000
!E:\Fortran\Test>stiming
!      2.36000         1.82000        0.771186
!     0.660000        0.760000         1.15152



Best regards,

--

Jan van Oosterwijk
Computing Centre
Delft University of Technology
Phone: +31 15 278 5017
Fax:   +31 15 278 3787
mailto:[log in to unmask]
http://huizen.dto.tudelft.nl/vanOosterwijk/