Re: Changes in Round4

[Xavier-Francois Roblot <roblot@math.univ-lyon1.fr>]

On Mon, 2007-08-20 at 10:34 -0400, Igor Schein wrote:
> Hi Xavier,
> 
> nfdisc(x^64+2^16) is almost 5 times slower now.

Hi Igor,

Thanks for the report. I have committed a new version that should make
this example now complete in time similar to the old version. Please
keep on testing for other possible regressions.

Xavier

> On 8/20/07, Xavier-Francois Roblot <roblot@math.univ-lyon1.fr> wrote:
> > Dear all,
> >
> > I have committed a number of changes to the CVS version of PARI/GP
> > Round4 algorithm (essentially to fix some precision issues). This is the
> > algorithm used to compute maximal orders of number fields (functions:
> > nfbasis, nfdisc, nfinit...) and factorize polynomials over the p-adic
> > numbers (function: factorpadic). I'll be glad to hear about any
> > bugs/regressions resulting from these changes. Note that since I fixed
> > parts of the old version where it was cutting corners, the new version
> > could show a small performance loss in some examples. Significant
> > improvements are not impossible (though not very likely) as I also fixed
> > some slight inefficiencies. I'll be happy to hear about those too. In
> > general, I'm very much interested in examples of polynomials for which
> > the algorithm performs poorly (compared to the degree and the size of
> > the discriminant).
> >
> > Have fun,
> >
> > Xavier
> >
> >
--