[gfortran] setting variable to NaN... - Fortran

On 25 sep, 08:39, fred <fredantis...@free.fr> wrote:
> Hi,
>
> I want to set a variable to NaN.
>
> This does the trick, with the intel fortran compiler:
>
> real :: x
>
> x = sqrt(-1.0)
>
> print *, x
>
> But it does not compile with gfortran:
>
> gfortran -Wall foo.f90
>
> In file foo.f90:7
>
> x = sqrt(-1.0)
> 1
> Error: Argument of SQRT at (1) has a negative value
>
> I guess there is some option to add to the command line, but which one ?
>
> I have browsed google groups, but unsuccessful.
>
> All answers are related to test if a variable is a NaN,
> not to set a variable to NaN.
>
> TIA.
>
> Cheers,
>
> --http://scipy.org/FredericPetit

Hm, one way around it might be:

real :: arg = -1.0
real :: x

x = sqrt(arg)

That should fool the compiler or even use:

x = setnan(arg)

and

real function setnan(arg)
setnan = sqrt(arg)
end function

You may want to have a look at http://ftp.aset.psu.edu/pub/ger/fortran/hdk/nan.f90
as well for more information on how NaNs are treated
by different compilers.

Regards,

Arjen

Arjen Markus <arjen.markus@wldelft.nl> a écrit :

> real :: arg = -1.0
> real :: x
>
> x = sqrt(arg)
So obviuous... :-)

Thanks a lot !

Cheers,

--
http://scipy.org/FredericPetit

On 25 sep, 09:12, fred <fredantis...@free.fr> wrote:
> Arjen Markus <arjen.mar...@wldelft.nl> a écrit :
>
> > real :: arg = -1.0
> > real :: x
>
> > x = sqrt(arg)
>
> So obviuous... :-)
>
> Thanks a lot !
>
> Cheers,
>
> --http://scipy.org/FredericPetit

Mind you, a very very smart compiler might figure out
what is going on

Regards,

Arjen

In article <873ax3xn8g.fsf@free.fr>,
fred <fredantispam@free.fr> writes:
>
> Hi,
>
> I want to set a variable to NaN.
>
> This does the trick, with the intel fortran compiler:
>
> real :: x
>
> x = sqrt(-1.0)
>
> print *, x
>
> But it does not compile with gfortran:
>
> gfortran -Wall foo.f90
>
> In file foo.f90:7
>
> x = sqrt(-1.0)
> 1
> Error: Argument of SQRT at (1) has a negative value
>
> I guess there is some option to add to the command line, but which one ?
>
> I have browsed google groups, but unsuccessful.
>
> All answers are related to test if a variable is a NaN,
> not to set a variable to NaN.

program k
real nan
nan = 0.
nan = nan / nan
print *, nan
end program k

mobile:kargl[203] gfc -o z k.f90
mobile:kargl[204] ./z
NaN
mobile:kargl[205]


--
Steve
http://troutmask.apl.washington.edu/~kargl/

In article <1190709303.878795.191660@19g2000hsx.googlegroups.com>,
Arjen Markus <arjen.markus@wldelft.nl> wrote:
>On 25 sep, 09:12, fred <fredantis...@free.fr> wrote:
>> Arjen Markus <arjen.mar...@wldelft.nl> a écrit :
>>
>> > real :: arg = -1.0
>> > real :: x
>>
>> > x = sqrt(arg)
>
>Mind you, a very very smart compiler might figure out
>what is going on

So NAG, Compaq and Sun f95 are all very very smart. They all reported
a floating-point error at run time and core-dumped. I used only the
default compiler settings in each case.

-- John Harper, School of Mathematics, Statistics and Computer Science,
Victoria University, PO Box 600, Wellington 6140, New Zealand
e-mail john.harper@vuw.ac.nz phone (+64)(4)463 5341 fax (+64)(4)463 5045

On 25 sep, 23:26, har...@mcs.vuw.ac.nz (John Harper) wrote:
> In article <1190709303.878795.191...@19g2000hsx.googlegroups.com>,
> Arjen Markus <arjen.mar...@wldelft.nl> wrote:
>
> >On 25 sep, 09:12, fred <fredantis...@free.fr> wrote:
> >> Arjen Markus <arjen.mar...@wldelft.nl> a écrit :
>
> >> > real :: arg = -1.0
> >> > real :: x
>
> >> > x = sqrt(arg)
>
> >Mind you, a very very smart compiler might figure out
> >what is going on
>
> So NAG, Compaq and Sun f95 are all very very smart. They all reported
> a floating-point error at run time and core-dumped. I used only the
> default compiler settings in each case.
>
> -- John Harper, School of Mathematics, Statistics and Computer Science,
> Victoria University, PO Box 600, Wellington 6140, New Zealand
> e-mail john.har...@vuw.ac.nz phone (+64)(4)463 5341 fax (+64)(4)463 5045

Hm, not the type of smartness I was thinking about:
seeing that the argument to sqrt() would be negative and
therefore refuse to compile it .

However, the behaviour does show that it is not that easy
to get a (quiet?) NaN.

Regards,

Arjen

On Sep 25, 11:35 pm, Arjen Markus <arjen.mar...@wldelft.nl> wrote:
> On 25 sep, 23:26, har...@mcs.vuw.ac.nz (John Harper) wrote:
>
>
>
> > In article <1190709303.878795.191...@19g2000hsx.googlegroups.com>,
> > Arjen Markus <arjen.mar...@wldelft.nl> wrote:
>
> > >On 25 sep, 09:12, fred <fredantis...@free.fr> wrote:
> > >> Arjen Markus <arjen.mar...@wldelft.nl> a écrit :
>
> > >> > real :: arg = -1.0
> > >> > real :: x
>
> > >> > x = sqrt(arg)
>
> > >Mind you, a very very smart compiler might figure out
> > >what is going on
>
> > So NAG, Compaq and Sun f95 are all very very smart. They all reported
> > a floating-point error at run time and core-dumped. I used only the
> > default compiler settings in each case.
>
> > -- John Harper, School of Mathematics, Statistics and Computer Science,
> > Victoria University, PO Box 600, Wellington 6140, New Zealand
> > e-mail john.har...@vuw.ac.nz phone (+64)(4)463 5341 fax (+64)(4)463 5045
>
> Hm, not the type of smartness I was thinking about:
> seeing that the argument to sqrt() would be negative and
> therefore refuse to compile it .
>
> However, the behaviour does show that it is not that easy
> to get a (quiet?) NaN.

If the IEEE modules are available, it should be easy to get
a quiet NaN.

Bob Corbett

robert.corbett@sun.com a écrit :

> If the IEEE modules are available, it should be easy to get
> a quiet NaN.
Tried it with gfortran.
Unsuccessful.
Or maybe I did something wrong...

Cheers,

--
http://scipy.org/FredericPetit

>> If the IEEE modules are available, it should be easy to get
>> a quiet NaN.
>
> Tried it with gfortran.
> Unsuccessful.
> Or maybe I did something wrong...

No, nothing wrong, gfortran doesn't support IEEE modules yet. It should
be in the next version (4.4).

--
FX