Re: Strassen multiplication over the integers

[Bill Allombert <Bill.Allombert@math.u-bordeaux.fr>]

On Thu, May 07, 2015 at 12:09:45PM +0200, Peter Bruin wrote:
> Hello Bill,
> 
> [Sorry if there is any double post; I tried to send a reply before, but
> it seems to have disappeared.]
> 
> >> > - ideally the program src/test/tune.c should deal with the tuning.
> >> > This program deal with dependencies between tuning parameters.
> >> 
> >> OK, I hope to have some time soon to try to write an addition for
> >> src/test/tune.c for this.
> >
> > Do not waste too much time with tune.c. The most important is to find
> > out how the tuning is affected by the coefficient size.
> 
> (I have not yet done this, but hope to look at it soon.)
> 
> > By the way, what matrix size your project is using ?
> 
> In a currently reasonable case (computations with a curve of genus 22
> over a finite field), one has to do matrix multiplications up to size
> (361 x 69) * (69 x 67).  In a harder but possibly reachable case (a
> curve of genus 40), all sizes will be about twice as large.
> 
> >> > - Most of the code is fairly generic, so maybe there could be a
> >> > gen_matmul_sw function.
> 
> I have now implemented this; see the attached patch.  It is completely
> analogous to ZM_mul_sw.  There are some new tests to ensure that the new
> code is covered.

Thanks, I have applied your patch in 23b45ac3e8601.
(yes I took my time...)

Cheers,
Bill.