Intel 17 ICE, unique to v17

[Ronald_G_2]

Steve, Kevin,

I've isolated an ICE occurring in 17 and 17 Update 1.  Doesn't seem to be in 16 Update 3, but haven't check the later v16 compilers fully.  Looks like IFORT Front-end. 

cat intel17bug.f90

     module ice_transport_remap

      contains

      subroutine horizontal_remap ( mm)
         real (kind=8), intent(inout), dimension(:,:,0:,:) :: mm           
         ! local variables
         integer (kind=4) ::    iblk, nghost
         call make_masks ( nghost, mm(:,:,:,iblk) )
      end subroutine horizontal_remap

      subroutine make_masks ( nghost, mm)
      integer (kind=4), intent(in) :: nghost
      real (kind=8), dimension (:,:,:), contiguous, intent(in) :: mm  
      end subroutine make_masks

      end module ice_transport_remap

/users/green> ifort -c -O0 intel17bug.f90
intel17bug.f90(9): catastrophic error: **Internal compiler error: internal abort** 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 intel17bug.f90 (code 1)

Can you send me the CQ number, either here or in private message?  Gotta document this for LANL.

Thanks Old Friends!

Ron

[Steven_L_Intel1]

Hi, Ron. I think this is the same as DPD200415576 so I am going to attach this example there. The one I had for that issue is larger and pickier about options than this one, so this is very helpful.

[Ronald_G_2]

We have a work around: removing CONTIGUOUS from 'mm' in make_masks.  We did that in the actual code.  Or changing the lower bound from 0:  in horizontal_remap, however that breaks the actual code.

thanks Steve.
 

[TimP]

I had to set -O1 to see this failure in 17.0.1.  Maybe CONTIGUOUS is ignored at -O0.

[Ronald_G_2]

Tim, are you talking about a bug you have or the attached example here?  This example gets and ICE at -O0.  It depends on the CONTIGUOUS attribute on mm.  If you remove contiguous this does compile w/o error.