Trying to find source of gfortran problems in Ubuntu

[James J. Ramsey]

----- Original Message ----
> From: Tobias Schlüter <Tobias.Schlueter@physik.uni-muenchen.de>
> To: James J. Ramsey <jjramsey@pobox.com>
> Cc: fortran@gcc.gnu.org
> Sent: Thursday, March 6, 2008 8:57:22 AM
> Subject: Re: Trying to find source of gfortran problems in Ubuntu
> 
> James J. Ramsey wrote:
> > ----- Original Message ----
> >> From: Jerry DeLisle  To: James J. Ramsey
> >>  Cc: fortran@gcc.gnu.org Sent: Wednesday, March
> >> 5, 2008 9:27:48 PM
> > 
> >> I can try try on Linux x86-64 or did you try already?
> > 
> > I did try, but considering that this problem seems to be dependent
> > not only on whether the Linux system is x86-64 but also on some
> > unknown particulars of how gfortran is set up, trying it out on
> > another Linux system would be helpful anyway. Even knowing if you
> > *can't* reproduce the problem might be useful.
> 
> And if you can produce a minimal testcase that fails on one of your 
> configurations, but not the other, we could add it to the testsuite to 
> make sure that we don't regress.

Unfortunately, I still don't know exactly what is wrong. I *think* I've narrowed it down to a problem generated in one particular file of the ABINIT code, but I don't know how to drill down further so that I can make a minimal testcase. For example, I don't quite know why adding WRITE statements can make the problem go away, or how the NaNs are produced without causing a floating-point exception. (Yes, I did compile with the '--ffpe-trap=invalid' option.) This is a really weird bug.



      ____________________________________________________________________________________
Never miss a thing.  Make Yahoo your home page. 
http://www.yahoo.com/r/hs