Compiler Bug: contiguous attribute

[Rodrigo_R_]

Hi.

While working with the contiguous attribute, I got into this:

integer, allocatable, target :: a(:) 
integer, pointer, contiguous :: b(:) 
allocate(a(5)) 
a = 0 
b => a ! No problem here 
b => a(1:3) ! No problem here 
b => a(:3) ! Error 
associate(x => a(:3)) 
    b => x ! No problem here either 
end associate 

Ifort's error says I'm associating a contiguous pointer to a non contiguous target in the third case. I think in all the cases the target is simply contiguous. Therefore, its a bug to throw an error only in the third case. Am I in the correct?

(I am using Intel Compiler 18 update 2 for windows, environment initialized for x64.)

[Rodrigo_R_]

Any update on this?

[Steve_Lionel]

I suggest that you report this at https://supporttickets.intel.com/?lang=en-US Issues reported here may get overlooked by Intel support.

[Rodrigo_R_]

I just reported as you suggested. Thank you for the attention.

[Christoph_F_]

Dear all,

has this issue already been addressed by the developers? (Sorry, I do not know how to check the status of the ticket.)

I am having a very similar problem in two cases with the false error message
"A pointer with the contiguous attributes is being made to a non-contiguous target"
after compilation with "-check all" with two compiler versions (18.0.2 20180210 and 19.0.0.117 20180804).

In the first case the target is an automatic array on the stack:

complex(real64), target :: hmtpw(2,(maxlcut+1)**2,maxval(neq))
complex(real64), pointer :: hmtpw_p(:,:)
hmtpw_p => hmtpw(:,:lm2)

The parameters "maxlcut" and "neq(:)" are taken from a module. The variable lm2 is calculated at runtime.

In the other case the target is another pointer included from a module with the CONTIGUOUS attribute:

complex(real64), pointer, contiguous :: cpw(:,:,:,:)  ! in the module (it is allocated as shared MPI3 memory)

complex(real64), pointer, contiguous :: vec(:) ! in the respective subroutine
vec => cpw(:n1,n2,n3,n4)
 

I think in both cases the target is contiguous, and Ifort should not complain.

Thank you.

Best regards
Christoph

[FortranFan]

Christoph F. wrote:

.. has this issue already been addressed by the developers? (Sorry, I do not know how to check the status of the ticket.) ..

I think in both cases the target is contiguous, and Ifort should not complain. ..

@Christoph,

Is it possible for you to try out the recently released Update 3 of Intel Fortran 19.0 compiler version and check for yourself?  Per this thread, it appears Intel Fortran team has included fixes on some related matters at least:

https://software.intel.com/en-us/forums/intel-fortran-compiler-for-linux-and-mac-os-x/topic/746360#new

For any unresolved issues, your best option will be to submit support requests at Intel Online Support Center (OSC) using your own reproducer cases: https://supporttickets.intel.com/?lang=en-US

[Christoph_F_]

Hello FortranFan,

thank you very much for your quick reply.

Unfortunately, the latest version we have is 19.0.0.117 20180804. As soon as I can get my hands on Update 3, I will check it.

My question would be: Am I correct in assuming that in both described cases (target in stack memory and target as contiguous pointer array in a module) the target is, in fact, contiguous, and the compiler is wrong in giving this error message? Just to make sure that the code is correct.

With best regards
Christoph

[jimdempseyatthecove]

!  contiguous.f90 
program contiguous
    implicit none
    integer, allocatable, target :: a(:) 
    integer, pointer, contiguous :: b(:) 
    allocate(a(5)) 
    a = 0 
    b => a ! No problem here 
    b => a(1:3) ! No problem here 
    b => a(:3) ! Error 
    associate(x => a(:3)) 
        b => x ! No problem here either 
    end associate 
    print *, b
end program contiguous

Win32 Debug
1>------ Build started: Project: contiguous, Configuration: Debug Win32 ------
1>Compiling with Intel(R) Visual Fortran Compiler 19.0.1.144 [IA-32]...
1>contiguous.f90
1>Linking...
1>Embedding manifest...
1>
1>Build log written to  "file://c:\test\contiguous\contiguous\Debug\BuildLog.htm"
1>contiguous - 0 error(s), 0 warning(s)
========== Build: 1 succeeded, 0 failed, 0 up-to-date, 0 skipped ==========

x64 Debug
1>------ Build started: Project: contiguous, Configuration: Debug x64 ------
1>Compiling with Intel(R) Visual Fortran Compiler 19.0.1.144 [Intel(R) 64]...
1>contiguous.f90
1>Linking...
1>Embedding manifest...
1>
1>Build log written to  "file://c:\test\contiguous\contiguous\x64\Debug\BuildLog.htm"
1>contiguous - 0 error(s), 0 warning(s)
========== Build: 1 succeeded, 0 failed, 0 up-to-date, 0 skipped ==========

x64 Release
1>------ Build started: Project: contiguous, Configuration: Release x64 ------
1>Compiling with Intel(R) Visual Fortran Compiler 19.0.1.144 [Intel(R) 64]...
1>contiguous.f90
1>Linking...
1>Embedding manifest...
1>
1>Build log written to  "file://c:\test\contiguous\contiguous\x64\Release\BuildLog.htm"
1>contiguous - 0 error(s), 0 warning(s)
========== Build: 1 succeeded, 0 failed, 0 up-to-date, 0 skipped ==========

x64 Release and Run

           0           0           0
Press any key to continue . . .

Jim Dempsey

[Christoph_F_]

      subroutine sub(m,n)
      implicit none
      integer, intent(in)          :: m,n
      real,    target              :: array(2,m**2,n)
      real,    pointer, contiguous :: point(:,:)
      nullify(point)
      point => array (:,:,1)
      end

ifort -check all  -c -o sub.o sub.f

sub.f(7): error #5558: A pointer with the CONTIGUOUS attributes is being made to a non-contiguous target
      point => array (:,:,1)
------^
compilation aborted for sub.f (code 1)

ifort --version

ifort (IFORT) 19.0.3.199 20190206
Copyright (C) 1985-2019 Intel Corporation.  All rights reserved.

[FortranFan]

Christoph F. wrote:
 

.. sub.f(7): error #5558: A pointer with the CONTIGUOUS attributes is being made to a non-contiguous target
      point => array (:,:,1)
------^
compilation aborted for sub.f (code 1)

ifort --version

ifort (IFORT) 19.0.3.199 20190206 ..

@Christoph F.,

At first glance, the code looks conforming; the target is "simply contiguous" if I'm not mistaken.  But I suggest you try to get Intel Fortran team to look at this by submitting a support request with this reproducer: https://supporttickets.intel.com/?lang=en-US

[Christoph_F_]

Hi FortranFan,

thank you, I just submitted a support ticket.

Slightly changing the code (only two array dimensions or using m instead of m**2 as upper boundary) make the error go away. So, it is most certainly a compiler error.

Best
Christoph