dmharvey at cims.nyu.edu
Wed Jun 10 14:08:54 CEST 2009
On Jun 10, 2009, at 3:00 AM, bodrato at mail.dm.unipi.it wrote:
> There are a lot of tricks like this, the first one in Karatsuba
> (due to?)
> saving half an addition. There is one in Toom-3, saving half a sum
> and it is implemented in the current interpolate_5pts, another in
> Toom-3.5, saving two half additions, implemented in interpolate_6pts.
At some point I skimmed over your paper with Zanoni on the optimality
of Toom-Cook matrices; I recall your automated search did not take
into account these kind of tricks that merge the interpolation and
recomposition stages. Is this correct? Have you tried this since
writing the paper? It seems that it shouldn't be difficult in theory
to extend the search method, although I have no idea if it's
practical (the search spaces are very large...?).
More information about the gmp-devel