Re: underflows, overflows, ...

[pancho at galaxy dot lca dot uevora dot pt]

Hi Dave,

thansk for your email.

I ran a fortran code, to test the ability to run with floating
points, underflows and overflows within a compiler, compiled with the
egcs-1.1-prerelease.

When the code comes to the first floating point it cores dump. can't run
the code anymore.

Whereas with digital fortran I only need to enable the flag -fpe2 and
things run nicely.


I would really try to have this kind of codes running in alpha machines
running linux and having egcs as a compiler.

cheers

Miguel



On 2 Sep 1998, Dave Love wrote:

> >>>>> "pancho" == pancho  <pancho@galaxy.lca.uevora.pt> writes:
> 
>  pancho> What are the flags to be used during compilation tha allow
>  pancho> for overflows, underflows and floating points in the
>  pancho> calculations using egcs? in such a way that the codes dont
>  pancho> stop, make any core dump, and dont stall.
> 
> I think only the Alpha has code generation options which affect this.
> (Maybe other targets do, but it isn't generally the case).
> 
> The g77 documentation gives an example of changing the floating point
> trap behaviour on GNU x86 systems.  I have similar code that works on
> various other systems to which I've had access, but I don't know what
> system you're using; most I've seen actually default to the
> IEEE-recommended (?) behaviour of _not_ trapping floating point
> exceptions.
> 
> [I'd like to try again after the egcs 1.1 dust settles to get a
> tidied-up version of this included as a standard (link-time) facility
> with either g77 or (preferably) egcs itself, but it was rejected for
> g77 previously.]
>