sqrt algorithm

Marco Bodrato bodrato at mail.dm.unipi.it
Wed Jul 29 22:00:50 UTC 2015


On Wed, July 29, 2015 9:24 pm, Niels Möller wrote:
> It could likely be made to work if the input is properly normalized,
> does the current code do that?

Right, current code does normalise in such a way that at most the highest
bit of the input is zero

> And for small sizes, one ought to sqrlo rather than mulmod_bnm1 to
> exploit the cancellation. So the tuning would be between sqrlo and
> sqrmod_bnm1. But it seems we only have mullo, not sqrlo?

Right, we currently don't have sqrlo...



More information about the gmp-devel mailing list