Re: floating point test suite ?

["C. Bergström" <codestr0m%osunix.org@localhost>]

Stathis Kamperis wrote:
> 2010/10/9 Manuel Bouyer <bouyer%antioche.eu.org@localhost>:
>   
> > Hello,
> >     
> Hi Manuel
>
>   
> > I strongly suspect that the MIPS floating point emulation has some
> > issues (the numbers printed by top makes no sense, and sometimes
> > it hangs because the computed time to sleep makes no sense either).
> >
> > Do we have some floating point test suite around (or something available
> > from pkgsrc) ?
> > [...]
> >     
>
> There are quite few tests online for floating-point testing, such as
> http://www.tybor.com/ or
> http://www.math.utah.edu/~beebe/software/ieee/. I bet you can find
> many more.
>
> Another idea is to install some CAS from pkgsrc and run its own test
> suite, if it has one. E.g., math/maxima.
>
> Last, we have a math test suite for libm. Here you can find
> information on how to deploy it:
> http://netbsd-soc.sourceforge.net/projects/mathlib/#_how_to_get_you_started
>
> You could give it a shot and compare your results with those for other
> architectures:
> http://leaf.dragonflybsd.org/~beket/logs/NetBSD/
>
> If there are many more failures in MIPS than in other archs, that
> could be indicative of a problem in MIPS.
>   
Hi

If you find the problem with MIPS to be related to gcc there's an 
alternative[1] compiler that's recently been ported over to NBSD and may 
offer better results in correctness and performance.

(Disclaimer - Emailing from my open source email, but I'm paid to work 
on this project.)


Best,

Christopher


[1] http://github.com/path64/compiler