On Jul 31, 2012, at 7:42 AM, Niels Möller wrote:
We currently have modular exponentation, powlo and regular powering
with no reduction of any kind. I'm suggesting a pow_modbnm1. For
euclidean square root, and for mpfr, it might also be useful with a
pow_high, keeping only the n most
On 04/04/2014, at 4:28 AM, Niels Möller ni...@lysator.liu.se wrote:
(1) the bit bounds for the coefficients get worse. For example if u =
5 then f(x) = x^5 + x^4 + x + 1, so when you compute say a(x) b(x) mod
f(x) (in Z[x]), every coefficient in the result is a sum of up to
*five* terms from
On 5 Jan 2015, at 10:08 am, Niels Möller ni...@lysator.liu.se wrote:
Of course there are also some drawbacks. It makes life more complicated
for applications, and the implementation of functions like mpn_mul_itch,
which interact with pretty complex algorithm choice machinery, is going
to be
On Tue, 2022-02-22 at 23:23 +0100, Marco Bodrato wrote:
> Ciao David,
>
> Il Mar, 22 Febbraio 2022 10:55 pm, David Harvey ha scritto:
> > On Tue, 2022-02-22 at 22:39 +0100, Marco Bodrato wrote:
> > > > E.g, in this case we could try a top-level B^66 - 1 product, split i
On Wed, 2022-02-23 at 09:53 +1100, David Harvey wrote:
> On Tue, 2022-02-22 at 23:23 +0100, Marco Bodrato wrote:
> > Ciao David,
> >
> > Il Mar, 22 Febbraio 2022 10:55 pm, David Harvey ha scritto:
> > > On Tue, 2022-02-22 at 22:39 +0100, Marco Bodrato wrote:
> &
On Tue, 2022-02-22 at 22:39 +0100, Marco Bodrato wrote:
>
> > E.g, in this case we could try a top-level B^66 - 1 product, split in
> > B^33+1 and B^33-1; then the former suits your new algorithm well, but
> > the former can't be recursively split (at least not on a B boundary). If
>
> I fully
