Internal compiler error at -O2,-O3 with 14.0.2.144

[Harald]

Hello,

the code below fails to compile with ifort 14.0.2.144 at -O2 and -O3 and with the specified checking options, but not with -O0 or -O1 and not with older versions.

module ifort14_ice
  implicit none
contains
  subroutine sub ()
    integer :: nc5
    integer :: len_rep
    integer :: min_rep
    min_rep = 10
    min_rep = 30
    min_rep = 60
    nc5 = min(min_rep, len_rep)
  end subroutine sub
end module ifort14_ice

% ifort -check all,noarg_temp_created,nooutput_conversion,nostack -c ifort14_ice.f90 -O2 -V
Intel(R) Fortran Compiler XE for applications running on IA-32, Version 14.0.2.144 Build 20140120
Copyright (C) 1985-2014 Intel Corporation.  All rights reserved.
FOR NON-COMMERCIAL USE ONLY

 Intel(R) Fortran 14.0-1624
ifort14_ice.f90: catastrophic error: **Internal compiler error: segmentation violation signal raised** Please report this error along with the circumstances in which it occurred in a Software Problem Report.  Note: File and line given may not be explicit cause of this error.
compilation aborted for ifort14_ice.f90 (code 1)

[FortranFan]

You should also report it to Intel Premier Support to have an incident created for the internal compiler error (ICE).

But also, can you recheck your code, especially lines 8 through 10?  What's their purpose?  And how is len_rep assigned for the statement in line 11 to be evaluated?

[Steven_L_Intel1]

You don't have to use Intel Premier Support - here is fine and we'll investigate.

[Steven_L_Intel1]

It's specifically -check uninit that is the trigger for this error. I have escalated it as issue DPD200253731 - thanks for the small example! You can exclude uninit from the checks if you want the compile to continue.

[FortranFan]

Steve Lionel (Intel) wrote:

It's specifically -check uninit that is the trigger for this error. I have escalated it as issue DPD200253731 - thanks for the small example! You can exclude uninit from the checks if you want the compile to continue.

Would initializing and defining len_rep before its use in line 11 be another way to allow the compiler to continue?  This way, -check uninit which is too good an option to exclude, can be retained?

[Steven_L_Intel1]

Not in this case. But removing the redundant assignments to min_rep does eliminate the error.

[Harald]

FortranFan wrote:

But also, can you recheck your code, especially lines 8 through 10?  What's their purpose?  And how is len_rep assigned for the statement in line 11 to be evaluated?

Steve correctly guessed that the reproducer was reduced from some larger code.  If you're concerned that len_rep might be uninitialized, you can move its declaration to the module level.  The ICE does not disappear.

Steve's suggestion to use -check nouninit works.  I also have another workaround.

Thanks,
Harald
 

[Harald]

Here's another snippet that leads to a similar crash.  Don't know whether it is related, but here we go:

module my_matrix
  implicit none
  integer, parameter :: wp = kind (1.d0)
contains
  subroutine check (a, y)
    real(wp) ,intent(in)            :: a(:,:)
    real(wp) ,intent(out) ,optional :: y(:,:)
    real(wp)          :: zw(size(a,1),size(a,1))
    integer           :: n, nb, lwork
    integer, external :: ilaenv

    n     = size(a,1)
    nb    = ilaenv( 1, 'DSYTRD', 'U', n, -1, -1, -1 )
    lwork = (nb+2)*n
    if (present(y)) then
      call dgemm ('n','t', n, n, n, 1._wp, zw, n, a, n, 0._wp, &
                  y, size (y,dim=1))
    endif
  end subroutine check
end module my_matrix

% ifort -V -O -check uninit -c my_matrix.f90
Intel(R) Fortran Compiler XE for applications running on IA-32, Version 14.0.2.144 Build 20140120
Copyright (C) 1985-2014 Intel Corporation.  All rights reserved.
FOR NON-COMMERCIAL USE ONLY

 Intel(R) Fortran 14.0-1624
catastrophic error: **Internal compiler error: segmentation violation signal raised** Please report this error along with the circumstances in which it occurred in a Software Problem Report.  Note: File and line given may not be explicit cause of this error.
compilation aborted for my_matrix.f90 (code 1)

[Harald]

It's nice to see that the above ICEs are fixed in

Intel(R) Fortran Compiler XE for applications running on IA-32, Version 15.0.0.090 Build 20140723

Thanks,

Harald

[Steven_L_Intel1]

Glad to hear it.