On Thu, Mar 5, 2015 at 2:30 AM, Francesco Biscani <[email protected]> wrote:
> Hello Ondrej,
>
> sorry, I just noticed today the message (I am subscribed to the SymPy
> mailing list and the gmail tagging filter made this message skip my inbox).

No problem. I didn't know that you were subscribed.

>
> On 2 March 2015 at 18:29, Ondřej Čertík <[email protected]> wrote:
>>
>> Hi Jason,
>>
>> On Mon, Mar 2, 2015 at 9:37 AM, Jason Moore <[email protected]> wrote:
>> > Though this may be interesting to folks here:
>> >
>> > https://peerj.com/preprints/504/
>>
>> Thanks a lot for sharing it. I've added a link to the paper to our
>> "benchmark" issue for CSymPy:
>>
>> https://github.com/sympy/csympy/issues/364
>>
>> Very nice paper. This is a subsequent paper to:
>>
>> Monagan, M., & Pearce, R. (2011). Sparse polynomial division using a
>> heap. Journal of Symbolic Computation, 46(7), 807–822.
>> doi:10.1016/j.jsc.2010.08.014
>>
>> Anyway, to compare with Table 1., I used Piranha
>> (https://github.com/bluescarni/piranha). The p1 takes 0.061s (1 core)
>> and 0.024s (4 cores) on my computer, compared to Maple's 0.041s (1
>> core) and 0.013s (4 cores) from the Table 1. Francesco, the authors of
>> Piranha (CCed) and I are actually playing with various integer
>> implementations, I think there is a way to speed Piranha as well on
>> this benchmark.
>>
>> For p2, I got 0.065s (1 core) and 0.021s (4 cores), compared Maple's
>> to 0.042s(1 core) and 0.017s (4 cores)
>>
>> For p4, Piranha gets better.
>
>
> Thanks for the pointer to the paper, I was not aware of it. Indeed Piranha
> was optimised targetting larger polynomials (~10**6 terms), so I am not too
> surprised that it gives better performance with larger examples.
>
> The article does not seem to specify exactly how the integer coefficients
> are represented. It just mentions that only integer coefficients are
> supported (e.g., no rationals, floating-point, etc.). The representation of
> the coefficients is going to have a rather big impact on overall performance
> for these benchmarks (which are rather dense).
>
> I am also quite positive we can improve Piranha's timings on smaller test
> cases - possibly via improvements for the integer class and the tweaking of
> various internal bits of the multiplication routine.

Agreed.

Btw, I got a clarification from the author, and they benchmark the
following structure:

( (1+x+y+z)^20 + 1 ) * ( (1+x+y+z)^20 + 2 )

Which is a bit slower than the Fateman benchmark, which is just
(1+x+y+z)^20 * ( (1+x+y+z)^20 + 1 ).

Ondrej

-- 
You received this message because you are subscribed to the Google Groups 
"sympy" 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 http://groups.google.com/group/sympy.
To view this discussion on the web visit 
https://groups.google.com/d/msgid/sympy/CADDwiVAVEtqAbKoxEk-XAYyoH8EX8hXvgRRWyMaMb%3DQEX2fZgA%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.

Reply via email to