Strange behavior with simple loop and array

trobitaille · ‎07-21-2010

I am encountering strange behavior, probably due to optimization, in the following code:


program test
  
  implicit none
  
  integer :: n_x,ia
  real,allocatable :: x(:)
  
  n_x = 19
  allocate(x(n_x))

  do ia=1,9
    x(10+ia) = real(ia-1) / real(n_x)
  end do
  
  x(1) = 0.
  x(2) = 1.e-10
  x(3) = 1.e-9
  x(4) = 1.e-8
  x(5) = 1.e-7
  x(6) = 1.e-6
  x(7) = 1.e-5
  x(8) = 1.e-4
  x(9) = 1.e-3
  x(10) = 1.e-2
  x(11) = 1.e-1
  
  print *,x

end program test

When compiled with ifort test.f90, the output is


  1.4699509E-39  1.4012985E-45  0.0000000E+00  0.0000000E+00  0.0000000E+00
  0.0000000E+00  7.0064923E-45  0.0000000E+00  1.4012985E-45  0.0000000E+00
  0.0000000E+00  5.2631579E-02 NaN            NaN            NaN           
 NaN             0.3157895      0.3684210      0.4210526

When compiled with ifort -O0 test.f90, the output is


  0.0000000E+00  1.0000000E-10  9.9999997E-10  9.9999999E-09  1.0000000E-07
  1.0000000E-06  9.9999997E-06  9.9999997E-05  1.0000000E-03  9.9999998E-03
  0.0000000E+00  5.2631579E-02  0.1052632      0.1578947      0.2105263    
  0.2631579      0.3157895      0.3684210      0.4210526

(i.e. the correct output)

When compiled with ifort -check all test.f90 the output is also correct. Finally, if I comment out the following line:


x(11) = 1.e-1

then the code does produce the correct output when compiled with ifort test.f90:


  0.0000000E+00  1.0000000E-10  9.9999997E-10  9.9999999E-09  1.0000000E-07
  1.0000000E-06  9.9999997E-06  9.9999997E-05  1.0000000E-03  9.9999998E-03
  0.0000000E+00  5.2631579E-02  0.1052632      0.1578947      0.2105263    
  0.2631579      0.3157895      0.3684210      0.4210526

Is this a compiler bug?I am using


ifort (IFORT) 11.1 20091130
Copyright (C) 1985-2009 Intel Corporation.  All rights reserved.

Thanks for any help,

Thomas

mecej4 · ‎07-21-2010

The reported behavior does not occur with 11.1.065 on Windows-32.

Are you, by some chance, running the compiler on OSX with one of the problematic versions of XCode? Please see Ron Green's sticky note at the top of the postings in this forum.

Kevin_D_Intel · ‎07-22-2010

mecej4 nailed it. You're using Xcode 3.2.2 and this Mac OS Fortran compiler:

$ ifort -V
Intel Fortran Intel 64 Compiler Professional for applications running on Intel 64, Version 11.1 Build 20091130 Package ID: m_cprof_p_11.1.080
Copyright (C) 1985-2009 Intel Corporation. All rights reserved.

--version is almost inadequate, so I advise using -V

As suggested in the sticky post mecej4 referred to here, your available work arounds (also discussed in the Knowledge Base article here) are to downgrade to Xcode 3.2.1 or compile with -use-asm.

Good work mecej4!

trobitaille · ‎07-22-2010

Thanks for your quick replies! Using -use-asm does indeed fix this.

Thomas