ifort 10.1 compiler bug: allocated allocatable array is not allocated with openmp

[domel07]

ifort 10.1 compiler bug: allocated allocatable array is not allocated with openmp. The code does not violate the 2003 specs and executes fine in serial. Compiled with "-std03 -openmp -fpp -DDEBUG -g -debug all -C -CB -CU -module /tmp -error-limit 3" and executed generates an error:

forrtl: severe (408): fort: (8): Attempt to fetch from allocatable variable USEDD when it is not allocated

I am quite sure this is a bug, however, I am using a free non-commercial licence so can not access the premier support page.

Below is the code to reproduce the problem:
 

MODULE NS3T10
USE OMP_LIB
IMPLICIT NONE

CONTAINS

SUBROUTINE SETUP_A
 INTEGER, DIMENSION(:), POINTER :: USED
 INTEGER, DIMENSION(:), ALLOCATABLE :: USEDD
 INTEGER :: NTH=1, MYID=1

!$OMP PARALLEL DEFAULT(SHARED) PRIVATE(NTH, MYID)
 NTH = OMP_GET_NUM_THREADS()
 MYID = OMP_GET_THREAD_NUM()
!$OMP SINGLE
 PRINT *, NTH, MYID
 CALL RESIZE(USEDD,2)
 USEDD = 0
 PRINT *, USEDD
!$OMP END SINGLE
!$OMP END PARALLEL
END SUBROUTINE SETUP_A

SUBROUTINE RESIZE(V,S)
 INTEGER, DIMENSION(:), ALLOCATABLE, INTENT(INOUT) :: V
 INTEGER, INTENT(IN) :: S
 IF(ALLOCATED(V)) DEALLOCATE(V)
 ALLOCATE(V(S))
 PRINT *, 'RESIZE : ', S
END SUBROUTINE RESIZE

END MODULE NS3T10

PROGRAM NS3T
USE OMP_LIB
USE NS3T10
IMPLICIT NONE

CALL SETUP_A
END PROGRAM NS3T

[jimdempseyatthecove]

Domel,

I think you want to change

CONTAINS

SUBROUTINE SETUP_A
 INTEGER, DIMENSION(:), POINTER :: USED
 INTEGER, DIMENSION(:), ALLOCATABLE :: USEDD
  ...

to

INTEGER, DIMENSION(:), POINTER :: USED
INTEGER, DIMENSION(:), ALLOCATABLE :: USEDD
CONTAINS

SUBROUTINE SETUP_A
 ...

As-was, only setup would see USED and USEDD.

If that still give you problems then you may need to add an interface block

INTEGER, DIMENSION(:), POINTER :: USED
INTEGER, DIMENSION(:), ALLOCATABLE :: USEDD
CONTAINS

SUBROUTINE SETUP_A
  INTERFACE
  SUBROUTINE RESIZE(V,S)
   INTEGER, DIMENSION(:), ALLOCATABLE, INTENT(INOUT) :: V
   INTEGER, INTENT(IN) :: S
  END SUBROUTINE RESIZE
  END INTERFACE

  ...

Jim Dempsey

[domel07]

the problem occurs only when compiling with

-std03 -openmp -openmp-report1 -fpp -DDEBUG -g -debug all -C -CA -CB -CU -module /tmp -error-limit 3

and not

-openmp -openmp-report1 -fpp -DDEBUG -O3 -mtune=core2 -module /tmp -error-limit 3

Compiler bug most obviously.

[Steven_L_Intel1]

Any licensed user can submit issues through Premier Support. If you did not sign up for a support account initially, you can do so by reregistering your serial number and checking the box for a support account.

[Steven_L_Intel1]

I can reproduce this single-threaded. The key is that you have to ask for "-check pointer" (which is implicit in -C) as well as -openmp. We've had other issues in the past where this feature improperly decided that a variable was not allocated when it was. I'll pass this on to the developers.

The easy workaround is to not use -check pointer (or specify only other -check options.)

[Steven_L_Intel1]

Jim,

I don't understand your reply - why do you think restructuring the code is necessary? USEDD is used only in SETUP_A. It is passed as an argument to RESIZE but that doesn't require USEDD to be visible as a name in RESIZE.

[domel07]

Hi, I find this is actually -g that is causing the problem for me. I am able to compile this unlucky code with -C but not with -g -C. Exact lines to be sure:

-std03 -openmp -openmp-report1 -fpp -DDEBUG -mtune=core2 -C -module /tmp -error-limit 3

will work. '-g' inserted before '-C' will break the thing. I would be very glad to have it fixed as I prefer ALLOCATABLE arrays to POINTERs.

-Dominik

[Steven_L_Intel1]

-g is a side-effect in that it implies -Od. We will fix this, but it will take a while for you to see the fix. Let me suggest that you not use -C and instead specify the specific -check options you want, without "pointer".

[domel07]

Steve,
I can not find -Od option in the compiler's help. Once again, to make sure we converge, I have no problem with any of -C, -CA, -CB, -CU. I am having problems only with -g switch on the compile line, which implies -O0. Adding explicit -O1 (still keeping -g) will crash too, but not -O2 or -O3. So e.g.:

FFLAGS = -std03 -openmp -openmp-report1 -fpp -DDEBUG -debug all -g -mtune=core2 -O2 -C -CA -CB -CU -module /tmp -error-limit 3

will work

FFLAGS = -std03 -openmp -openmp-report1 -fpp -DDEBUG -debug all -g -mtune=core2 -O1 -C -CA -CB -CU -module /tmp -error-limit 3

will not, (neither will -O0).

-- Dominik

[domel07]

... and leaving out all -Cx switches completely will still crash, if -g is present. I forgot to add this important info in the previous post.

Dominik

[Steven_L_Intel1]

I can't reproduce that - you don't get that particular error unless -check pointer is enabled. Show me a log of the compile and run that shows the error.

[domel07]

Here the code again, now modified a bit to be all in one file.

MODULE NS3T10
USE OMP_LIB
IMPLICIT NONE

INTEGER, PARAMETER :: NK = 34

CONTAINS

SUBROUTINE SETUP_A
! INTEGER, DIMENSION(:), POINTER :: USED
INTEGER, DIMENSION(:), ALLOCATABLE :: USED
INTEGER :: NTH=1, MYID=1

! CALL RESIZE(USED,1)
! CALL PRINT_ADDRESS(USED)

!$OMP PARALLEL DEFAULT(SHARED) PRIVATE(NTH, MYID)
NTH = OMP_GET_NUM_THREADS()
MYID = OMP_GET_THREAD_NUM()
PRINT *, NTH, MYID
!$OMP SINGLE
! CALL PRINT_ADDRESS(USED)
! PRINT *, ALLOCATED(USED)
CALL RESIZE(USED,NTH)
! CALL PRINT_ADDRESS(USED)
USED = 10
PRINT *, USED
!$OMP END SINGLE
!$OMP END PARALLEL

! CALL PRINT_ADDRESS(USED)

END SUBROUTINE SETUP_A

SUBROUTINE RESIZE(V,S)
INTEGER, DIMENSION(:), ALLOCATABLE, INTENT(INOUT) :: V
INTEGER, INTENT(IN) :: S
IF(ALLOCATED(V)) DEALLOCATE(V)
ALLOCATE(V(S))
PRINT *, 'RESIZE : ', S
! CALL PRINT_ADDRESS(V)
END SUBROUTINE RESIZE
!
! SUBROUTINE RESIZE(V,S)
! INTEGER, DIMENSION(:), POINTER :: V
! INTEGER, INTENT(IN) :: S
! NULLIFY(V)
! ALLOCATE(V(S))
! PRINT *, 'RESIZE : ', S
! END SUBROUTINE RESIZE

END MODULE NS3T10

PROGRAM NS3T
USE OMP_LIB
USE NS3T10
IMPLICIT NONE

CALL SETUP_A
END PROGRAM NS3T


The compilation:

ifort -std03 -openmp -openmp-report1 -fpp -DDEBUG -debug all -g -mtune=core2 -module /tmp -error-limit 3 -I /scratch/domel/pack/hdf5-1.6.7/fortran/src -o /tmp/NS3T10_MAIN_TEST1.o -c NS3T10_MAIN_TEST1.f90
fortcom: Warning: NS3T10_MAIN_TEST1.f90, line 17: Fortran 2003 does not allow this statement or directive.
!$OMP PARALLEL DEFAULT(SHARED) PRIVATE(NTH, MYID)
------^
fortcom: Warning: NS3T10_MAIN_TEST1.f90, line 21: Fortran 2003 does not allow this statement or directive.
!$OMP SINGLE
------^
fortcom: Warning: NS3T10_MAIN_TEST1.f90, line 28: Fortran 2003 does not allow this statement or directive.
!$OMP END SINGLE
----------^
fortcom: Warning: NS3T10_MAIN_TEST1.f90, line 29: Fortran 2003 does not allow this statement or directive.
!$OMP END PARALLEL
------------^
NS3T10_MAIN_TEST1.f90(17): (col. 7) remark: OpenMP DEFINED REGION WAS PARALLELIZED.
icc -g -o /tmp/address.o -c address.c
ifort -std03 -openmp -openmp-report1 -fpp -DDEBUG -debug all -g -mtune=core2 -module /tmp -error-limit 3 /tmp/NS3T10_MAIN_TEST1.o /tmp/address.o -L /home/domel/pack/intel/mkl/10.0.2.018/lib/32 -lmkl_solver -lmkl_intel -lmkl_intel_thread -liomp5 -lmkl_core -lpthread -lz -o /tmp/NS3T10_TEST1



The output:


domel@x60:~/src/solve> /tmp/NS3T10_TEST1
2 0
RESIZE : 2
forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image PC Routine Line&nb sp; Source
. FFFFE410 Unknown Unknown Unknown
libiomp5.so 40398C7A Unknown Unknown Unknown
libpthread.so.0 4042A192 Unknown Unknown Unknown
libc.so.6 4059A02E Unknown Unknown Unknown
forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image PC Routine Line Source
. FFFFE410 Unknown Unknown Unknown
libiomp5.so 40398C7A Unknown Unknown Unknown
libpthread.so.0 4042A192 Unknown Unknown Unknown
libc.so.6 4059A02E Unknown Unknown Unknown


-- Dominik

[Steven_L_Intel1]

Interesting - note that that's a different error. I'll look into that.

[domel07]

If you could tell me how to see some more than 'Unknown' during the crash that I get despite -g I might be of some more help (more general question/remark).
Thanks,
Dominik

[Steven_L_Intel1]

-traceback