Segmentation Fault with OpenMP

[trobitaille]

Hi,

If I compile the following code:

[fortran]module routine

  implicit none

contains

  subroutine sub()
    implicit none
    integer,parameter :: nmax=1000000
    complex(8),dimension(nmax) :: d
    d = 0.
  end subroutine sub

end module routine

program test

  use routine

  !$OMP PARALLEL DO
  do iw=1,100
     call sub()
  end do
  !$OMP END PARALLEL DO

end program test
[/fortran]

with the following compile command:

ifort -heap-arrays -O0 -openmp test.f90 -static-intel

Then run it, I get:

$ ./a.out

Segmentation fault

Notes:

• If I change complex(8) to real(8) I get:

forrtl: severe (180): SIGBUS, bus error occurred

Image PC Routine Line Source

a.out 00000001000011AA Unknown Unknown Unknown

a.out 0000000100001423 Unknown Unknown Unknown

a.out 00000001000BED13 Unknown Unknown Unknown

• If I set nmax <=262061 then the code runs properly (with complex(8))
• If I set262061 < nmax <=262308 then I get

Illegal instruction

• If I set nmax >262308 then I get a segfault as before

Details

$ ifort -V

Intel Fortran Intel 64 Compiler XE for applications running on Intel 64, Version 12.0.2.142 Build 20110112

Copyright (C) 1985-2011 Intel Corporation. All rights reserved.

System: MacOS 10.6.6, XCode 3.2.5

Thanks in advance for any help!

[jimdempseyatthecove]

As an experiment, declare sub as a recursive subroutine

recursive subroutine sub()
...


Jim Dempsey

[TimP]

It's probably better to attempt to design separate experiments for each abuse of the compiler.
Are you asking why the compiler doesn't complain, or which issues prevent dead code elimination?

[trobitaille]

Adding 'recursive' does not seem to change anything - I still get a Segmentation fault. Any ideas?

[Steven_L_Intel1]

You are getting a stack overflow, even with -heap-arrays. Try this substitute code which works for me:

complex(8),dimension(:),allocatable :: d
allocate (d(nmax))