Thank you for your quick

janusz_andrzejewski · ‎02-16-2016

Hi,

this little program cause compiler error when it is compiled with -fopenmp switch.

MODULE aM
IMPLICIT NONE

CONTAINS
SUBROUTINE a1(tab1, tab2)
REAL, INTENT(IN), DIMENSION(:) :: tab1
REAL, INTENT(OUT), DIMENSION(:) :: tab2

REAL, DIMENSION(10) :: t1

!$OMP PARALLEL
!$OMP SINGLE
    !error1 !$OMP TASK DEPEND(IN: tab1) !<= unmarked lines causes internal compiler error
    !$OMP TASK DEPEND(OUT: tab2) !<= unmarked lines causes internal compiler error
    !OK $OMP TASK DEPEND(OUT: t1) !<= this is OK
        tab2(:)=2.8*tab1(:)
        t1(:)=tab1(:)
    !$OMP END TASK
!$OMP END SINGLE
!$OMP END PARALLEL

END SUBROUTINE a1

END MODULE aM

Program aa
USE aM
IMPLICIT NONE
REAL, DIMENSION(10) :: tab1, tab2

tab1=1.0
CALL a1(tab1, tab2)

END

I tested this program on ifort 15.0.5 and 16.0.1 on Debian

Best regards

Janusz Andrzejewski

Steven_L_Intel1 · ‎02-16-2016

Thanks for the problem report. I can reproduce this in 16.0.1, but not in 16.0.2 which should be out any day now. When you get 16.0.2, please try your program again and let me know if the problem is resolved.

janusz_andrzejewski · ‎02-17-2016

Thank you for your quick replay.

The following program compiles now, but crashes.

MODULE aM
IMPLICIT NONE

CONTAINS
SUBROUTINE a1(n)
INTEGER, INTENT(INOUT) :: n

!$OMP PARALLEL DEFAULT(SHARED)
!$OMP SINGLE
!$OMP TASK FIRSTPRIVATE(n)
!$OMP END TASK
!$OMP END SINGLE
!$OMP END PARALLEL

END SUBROUTINE a1

END MODULE aM

Program aa
USE aM
IMPLICIT NONE
INTEGER :: n

n=10
CALL a1(n)

END

There is a problem within the FIRSTPRIVATE. If you replace FIRSTPRIVATE(n) with FIRSTPRIVATE(n1),

where n1 is a local INTEGER variable everything is OK. The bug is insensitive on attribute INTENET in subroutine.

janusz_andrzejewski · ‎02-18-2016

I've just update intel compiler to the version 16.0.2.

After update, the first program compiles now. But the second one (with FIRSTPRIVATE) still causes runtime segmentation fault.

Bug with TASK directive in OpenMP