Inconsistent math behavior on ifort 11.1

[Luiz_Felipe_S_]

HI All!

          I'm compiling the code below in gfortan, ifort 11.1 and ifort 14.0.3.

PROGRAM TEST
  REAL :: x, x2(5), z
  COMPLEX :: y, y2(5) 
  z = -1.
  z=sqrt(z)
  x=5./z 
  x2=5./z
  y = (1.,1.)*x
  y2= (1.,1.)*x2
  WRITE(*,*) x  
  WRITE(*,*) x2
  WRITE(*,*) y
  WRITE(*,*) y2
  WRITE(*,*) x /= x  
  WRITE(*,*) ANY(x2 /= x2)  
  WRITE(*,*) y /= y  
  WRITE(*,*) ANY(y2 /= y2)  
END PROGRAM TEST

     The problem is when ifort 11.1 compares an array of NaNs with another array of NaNs, it returns false instead of true. Is it a compiler bug? Is there a way to set ifort 11.1 in a way it outputs the same results as in ifort 14.3?

$ gfortran -V

gcc version 4.4.7 20120313 (Red Hat 4.4.7-4) (GCC)

$ ./teste_gfortran
             NaN
             NaN             NaN             NaN             NaN             NaN
 (            NaN,            NaN)
 (            NaN,            NaN) (            NaN,            NaN) (            NaN,            NaN) (            NaN,            NaN) (            NaN,            NaN)
 T
 T
 T
 T

$ ifort -V
Intel(R) Fortran Intel(R) 64 Compiler Professional for applications running on Intel(R) 64, Version 11.1    Build 20090827 Package ID: l_cprof_p_11.1.056
Copyright (C) 1985-2009 Intel Corporation.  All rights reserved.

$ ./teste_ifort_11_1
 NaN           
 NaN            NaN            NaN            NaN            NaN           
 (NaN,NaN)
 (NaN,NaN) (NaN,NaN) (NaN,NaN) (NaN,NaN) (NaN,NaN)
 T
 T
 F
 F

$ ifort -V
Intel(R) Fortran Intel(R) 64 Compiler XE for applications running on Intel(R) 64, Version 14.0.3.174 Build 20140422
Copyright (C) 1985-2014 Intel Corporation.  All rights reserved.

t$ ./teste_ifort_14_03
            NaN
            NaN            NaN            NaN            NaN            NaN
 (NaN,NaN)
 (NaN,NaN) (NaN,NaN) (NaN,NaN) (NaN,NaN) (NaN,NaN)
 T
 T
 T
 T

[Steven_L_Intel1]

Your program is actually asking if any of the array elements is not equal to the others, and for NaN, a NaN isn't equal to anything else, even another NaN. However, we don't guarantee such comparisons unless you compile with "-fp strict", so try that.

[Luiz_Felipe_S_]

Hi Steve!

      I got the same thing, no change...

$ ifort -fp-model strict teste.f90 -O -o teste_ifort3
$ ./teste_ifort3 
 NaN           
 NaN            NaN            NaN            NaN            NaN           
 (NaN,NaN)
 (NaN,NaN) (NaN,NaN) (NaN,NaN) (NaN,NaN) (NaN,NaN)
 T
 T
 F
 F

[Steven_L_Intel1]

Well, I don't know then. That's an old, unsupported version. You should be using IEEE_ISNAN from intrinsic module IEEE_ARITHMETIC instead of comparing.