Re: [PrimeNumbers] has SB been tested?
- Décio Luiz Gazzoni Filho wrote:
> On Tuesday 01 April 2003 12:53, Michael Bell wrote:Except that George's code uses different assembler routines for each FFT
> > Hi,
> > I think we are getting confused here between roundoff errors (which are
> > implementation independent assuming the exact same FFT algorithm, limb
> > size, etc. being run on a processor with the same precision) and coding
> > errors. The latter are nothing much to do with size.
> Exactly. Since SoB already found 5 or 6 primes, I assume their code has no
> errors of the latter kind. The only errors that might creep up now are
> of the
> roundoff kind, which I argued against in my other posts.
length, so that doesn't really hold.
>Indeed, I should think this is pretty unlikely.
> I also agree with David, but IMO no one should hope that SoB is not finding
> any new primes due to the multiplication code pushing the limits of
> errors (and even less due to outright buggy code).