Problems with array subsets

jeffludwig · ‎12-20-2005

Hello all. I'm wondering if this is a compiler bug or a nuance of fortran 90 I'm not familiar with. If I have a two dimensional array in F90, a, and I pass a subset of the array to a subroutine:

call calc_norm(a(1:3,2),a(4:6,2))

the intel compiler only seems to get the reference of the first element of a, no matter what the index is. That is in the function:

subroutine calc_norm(r1,r2)
real(kind(0d0)), dimension(3) :: r1, r2

r1 and r2 are equal in the subroutine. Even if i specify a(3:5,2) as the first arugment calc_norm still only gets a(1:3,2). Interestingly enough if I print the subsets they are correct. Am I missing something? At the very least the results are inconsistent. Using:

Intel Fortran Compiler for 32-bit applications, Version 9.0 Build 20050430Z Package ID: 
Copyright (C) 1985-2005 Intel Corporation. 
All rights reserved. FOR NON-COMMERCIAL USE ONLY
Centos 4.2, 2.6.9-22.0.1.EL #1

Attached is a sample program to illustrate what I mean:


module testsplice
    implicit none
contains
    subroutine check(chunk1, chunk2)
        implicit none
        real(kind(0d0)), dimension(5) :: chunk1, chunk2

        print *, '1', chunk1
        print *, '2', chunk2
    end subroutine check
end module testsplice

program splice
    use testsplice
    implicit none
    real(kind(0d0)), dimension(10,10) :: myarray
    integer :: i, j, counter

    counter=1
    do i=1,10
        do j=1,10
            myarray(j,i)=float(counter)
            counter=counter+1
        enddo
    enddo

    print *, myarray
    call check(myarray(1:5,2), myarray(6:10,2))

end program splice

Steven_L_Intel1 · ‎12-20-2005

You don't show what results you're getting, but I do vaguely recall a bug in this area that was fixed a long time ago. You're using a compiler from last April. I tried a current compiler and it looked ok.

However, your statement that "if I print the subsets they are correct" is confusing - how are you determining that the arrays are not correct?

jeffludwig · ‎12-20-2005

Thanks for the quick reply.

For example, if I add a line:


print *, myarray(6:10,2)

To the program above, it reports (correctly):


16.0000000000000        17.0000000000000        18.0000000000000     
   19.0000000000000        20.0000000000000

However in the subroutine, I obtain:


 1   11.0000000000000        12.0000000000000        13.0000000000000     
   14.0000000000000        15.0000000000000     
 2   11.0000000000000        12.0000000000000        13.0000000000000     
   14.0000000000000        15.0000000000000

As for using an old compiler, I'm using the latest non-commerical version you are offering (downloaded two or three weeks ago). It seems silly to be distributing this compiler (even for free) if there are major bugs like this? At the very least is makes it hard to evaluate it for purchase. Perhaps this is a platform specific issue or a language technicality I'm missing?

Jeff Ludwig

University of Delaware

Steven_L_Intel1 · ‎12-20-2005

Ah, yes. The compiler on the FTP server gets updated infrequently. It was, however, just updated last week. But if you register the serial number, you can download a current compiler from Intel Premier Support if your license is still valid.

I agree with you that asking people to evaluate an old compiler is perhaps not the best strategy.

The current compiler gets this right.

Message Edited by Steve_Lionel on 12-20-2005 03:39 PM