Catostrophic error: **Internal compiler error: internal abort** ifort v. 12.1.4 20120410

[losalamos]

We have an important code that has been unable to compile with the Intel Fortran compiler for Linux since version 10. We just tried the latest release and it, too, fails to compile. The reproducer code is given below but here is the error message:

$ ifort --version

ifort (IFORT) 12.1.4 20120410

Copyright (C) 1985-2012 Intel Corporation. All rights reserved.

$ ifort -c -fp-model strict -I . reproducer.F

0_12459

: 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 reproducer.F (code 1)

Here is the code that causes this error:

$ cat reproducer.F

FUNCTION EOS_REAL (X, shift, bnd, dim)

implicit none

include "param.h"

integer(kb) :: shift, dim

real(kb_r) :: bnd

real(kb_r) :: eos_real(0:mx,0:my,0:mz)

real(kb_r) :: X(0:mx,0:my,0:mz)

integer(kb) :: i,j,k, my_id

eos_real = eoshift(X, SHIFT=shift, BOUNDARY=bnd, DIM=1)

end function eos_real

FUNCTION EOS_LOG (X, shift, bnd, dim)

include "param.h"

integer(kb) :: shift, dim

logical(kb) :: bnd

logical(kb) :: eos_log(0:mx,0:my,0:mz)

logical(kb) :: X(0:mx,0:my,0:mz)

integer(kb) :: i,j,k

eos_log = eoshift(X, SHIFT=shift, BOUNDARY=bnd, DIM=1)

end function eos_log

$cat param.h

include "kind.h"

integer(kb) nx, ny, nz

integer(kb) mx, my, mz

integer(kb) nc, nv

integer(kb) nmat, mmat

integer(kb) nstr, mstr

integer(kb) nmts, mmts

integer(kb) nfrc, mfrc

integer(kb) ntm, mtm

integer(kb) nemm, memm

integer(kb) nprog, mprog

integer(kb) ndelq, mdelq

integer(kb) nmsff, mmsff

integer(kb) ncrs, mcrs

integer(kb) ndsd, mdsd

integer(kb) ndsdjwl, mdsdjwl

common /param/ nx, ny, nz,

& mx, my, mz,

& nc, nv,

& nmat, mmat,

& nstr, mstr,

& nmts, mmts,

& nfrc, mfrc,

& ntm, mtm,

& nemm, memm,

& nprog, mprog,

& ndelq, mdelq,

& nmsff, mmsff,

& ndsd, mdsd,

& ndsdjwl, mdsdjwl,

& ncrs, mcrs

integer(kb) nx_m1, ny_m1, nz_m1

integer(kb) mx_m1, my_m1, mz_m1

common /param_m1/ nx_m1, ny_m1, nz_m1,

& mx_m1, my_m1, mz_m1

$cat kind.h

integer, parameter :: kb_r = 4 ! single precision

integer, parameter :: kb = 4 ! single precision

integer, parameter :: kb4 = 4 ! single precision

integer, parameter :: kb8 = 8 ! double precision

This could be related to this problem (http://software.intel.com/en-us/forums/showthread.php?t=102565) for which in January was stated it was fixed in a future release. Any idea when that would appear?

This is a critical bug for us. Thanks.

-david

[Steven_L_Intel1]

I can reproduce the problem, but only if -fp-model strict is used. It does appear to be fixed in our next major release, scheduled for later this year. Can you avoid using -fp-model strict when compiling this source?

[losalamos]

Unfortunately, no, we cannot remove the -fp-model strict requirement. The sensative nature of the code and its results are something I cann't bring up, but suffice it to say that it is of extreme significance to the Department of Energy's mission. :-/

[Steven_L_Intel1]

Perhaps you have access to our beta of the next version? If not, let me know.