oldk1331 wrote:
> Scratch this patch.
>
> UPOLYC has POLYCAT has GPOLCAT has FAMR has pomopo!,
> which is basicly the same with fmecg:
> pomopo!(p1, r, e, p2) == fmecg(p1, e, -r, p2)
> differences are args order, minus sign, possibly change of p1.
pomopo! modifies p1, so it can be faster, but harder to use.
Also pomopo! makes sense for semirings (no '-').
> fmecg should go into FAMR, or better, merge with pomopo!.
>
> The implementation of pomopo! in PR is a bit messy. I do a simple
> benchmark, for 1 million elements dense polynomial, fmecg is
> 0.06s, pomopo! is 0.04s. Run a few times and GC takes 0.5s.
0.06s to 0.04s if repeatable is significant. And GC times
are correlated with amount of allocation, so GC time using
pomopo! should be smaller. Actually, IME when limiting
allocation _all_ times go down (probably due to better
cache use).
pomopo! in FAMR probably is faster for large polynomials. And
it could be easily speed up by calling addms! when appropriate.
OTOH we care very much about performance for small polynomials.
That is, how fast can we do 1 million polynomials of degree say
5.
> Thus, I propose, merge fmecg and pomopo! into fmecg.
> What do you think?
pomopo! in PR probably gives no advantages. OTOH the
one in FAMR is cheap (just exposes internal routine)
and offers possibility for faster code, so I would keep
it. fmecg has to be sepatate but probably can go into FAMR.
--
Waldek Hebisch
--
You received this message because you are subscribed to the Google Groups
"FriCAS - computer algebra system" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To post to this group, send email to [email protected].
Visit this group at https://groups.google.com/group/fricas-devel.
For more options, visit https://groups.google.com/d/optout.
