On Thu, 15 Dec 2022 at 12:19, Dima Pasechnik <[email protected]> wrote:
>
> On Thu, Dec 15, 2022 at 12:05 PM John Cremona <[email protected]> wrote:
> >
> > On Thu, 15 Dec 2022 at 12:00, Dima Pasechnik <[email protected]> wrote:
> > >
> > > On Thu, Dec 15, 2022 at 9:30 AM Vincent Delecroix
> > > <[email protected]> wrote:
> > > >
> > > > On Thu, 15 Dec 2022 at 01:42, Dima Pasechnik <[email protected]> wrote:
> > > > >
> > > > >
> > > > >
> > > > > On Wed, 14 Dec 2022, 21:06 Vincent Delecroix, 
> > > > > <[email protected]> wrote:
> > > > >>
> > > > >> Note that in #34850 I execute the very same command, namely
> > > > >> pari("quadclassunit(1 - 2^100)"), once in a fresh python session and
> > > > >> once preceeded by "import sage.all". So whatever linking and whatever
> > > > >> threading option has been used to compile pari these are the same for
> > > > >> both executions.
> > > > >
> > > > >
> > > > >
> > > > > oops, sorry, I was too quick. I guess here it is a change in memory 
> > > > > layout that causes the slowdown.
> > > >
> > > > I don't quite believe it. The PARI/GP stack is allocated before the
> > > > `import sage.all` and is not affected by this call. What kind of
> > > > change in memory could cause the slowdown here?
> > >
> > > It's caused by changes in FPU status, or something like this (the
> > > latter happened at least once earlier on in Sage, was caused by a
> > > clang bug, in fact). This time, it is  caused by sympy.
> > >
> > > If you replace
> > >
> > >     import sage.all
> > >
> > > with
> > >
> > >     import sympy
> > >
> > > you'll get the same slowdown...
> > >
> > > You guys should invite me to Luminy and buy me a beer there ;-)
> >
> > A propos, which Sage developers will be in Luminy for the COUNT
> > meeting?  I will.
> >
> I'm still waiting to hear from them. Or perhaps they don't notify
> people who haven't made it on the list, I have no idea.
> I should form a 1-person union and declare myself on strike w.r.t.,
> quoting their announcement,  "informal collaboration and for coding
> projects related to PARI/GP, SageMath,..." :-)

I only had confirmation a day or two ago, they are working through the
applicants.  It may help that one of the organisers was my postdoc for
3 years...

>
> Dima
>
> > John
> >
> > >
> > >
> > >
> > >
> > > >
> > > > > You also mention that it might be ipython - can this be tested by 
> > > > > building Sage without it?
> > > >
> > > > I meant the base kernel from ipython that would be responsible for the
> > > > slowdown in pari_jupyter. That is just a supposition.
> > >
> > > This one I can rule out - I ran your tests in iPython shell, with
> > > basically the same results, so no, iPython has
> > > an alibi. I also tried cProfile, to see if there are  extra things
> > > executed for some reason
> > > if sage.all is imported - but no, it's a genuine Pari slowdown.
> > >
> > >
> > >
> > >
> > > >
> > > > >> In particular, I do not see any relations between
> > > > >> #31572 (which tests two pari versions) and #34850 (which tests a
> > > > >> single pari version). Could you either read carefully the description
> > > > >> of #34850 or explain the link between #31572 and #34850 if any?
> > > > >>
> > > > >> On Wed, 14 Dec 2022 at 21:58, Dima Pasechnik <[email protected]> 
> > > > >> wrote:
> > > > >> >
> > > > >> > Does #31572 actually mean that pari threads are more harmful than 
> > > > >> > useful?
> > > > >> > If so, perhaps, we can create a special multithreaded, 
> > > > >> > differerently
> > > > >> > named, libpari for giac, and
> > > > >> > build our "main" libpari single-threaded?
> > > > >> >
> > > > >> > Dima
> > > > >> >
> > > > >> > On Wed, Dec 14, 2022 at 8:38 PM David Roe <[email protected]> 
> > > > >> > wrote:
> > > > >> > >
> > > > >> > > It could be related to #31572, which notes a dramatic speed 
> > > > >> > > regression based on pari being compiled with pthread.
> > > > >> > > David
> > > > >> > >
> > > > >> > > On Wed, Dec 14, 2022 at 3:30 PM Dima Pasechnik 
> > > > >> > > <[email protected]> wrote:
> > > > >> > >>
> > > > >> > >> I vaguely recall something about the way gp is built, without 
> > > > >> > >> linking libpari dynamically, and something about gp getting 
> > > > >> > >> slower if built with threads.
> > > > >> > >>
> > > > >> > >> So, probably, more info is needed on how exactly your 
> > > > >> > >> comparisons are done, in regard of using, or not, threads, and 
> > > > >> > >> dynamic linking.
> > > > >> > >>
> > > > >> > >>
> > > > >> > >>
> > > > >> > >>
> > > > >> > >>
> > > > >> > >> On Wed, 14 Dec 2022, 20:23 Vincent Delecroix, 
> > > > >> > >> <[email protected]> wrote:
> > > > >> > >>>
> > > > >> > >>> Dear all,
> > > > >> > >>>
> > > > >> > >>> Some strange phenomenon makes execution of PARI/GP up to twice 
> > > > >> > >>> slower
> > > > >> > >>> when ran inside a Python process compared to its execution in 
> > > > >> > >>> GP.
> > > > >> > >>> Though this is not systematic, in a fresh python environment 
> > > > >> > >>> timing
> > > > >> > >>> are comparable with GP.
> > > > >> > >>>
> > > > >> > >>> The slowdown has been first observed in the context of 
> > > > >> > >>> pari_jupyter
> > > > >> > >>> [1] and then with cypari2 when sage.all is loaded [2]. There 
> > > > >> > >>> is a
> > > > >> > >>> short code snippet in the description of [2] that you can try 
> > > > >> > >>> on your
> > > > >> > >>> own sage version (must be ran with "sage -python" and not 
> > > > >> > >>> "sage").
> > > > >> > >>>
> > > > >> > >>> Any idea on what could be the source of the slowdown or some 
> > > > >> > >>> profiling
> > > > >> > >>> advices would be very welcome.
> > > > >> > >>>
> > > > >> > >>> [1] https://github.com/sagemath/pari-jupyter/issues/27
> > > > >> > >>> [2] https://trac.sagemath.org/ticket/34850
> > > > >> > >>>
> > > > >> > >>> Best
> > > > >> > >>> Vincent
> > > > >> > >>>
> > > > >> > >>> --
> > > > >> > >>> You received this message because you are subscribed to the 
> > > > >> > >>> Google Groups "sage-devel" group.
> > > > >> > >>> To unsubscribe from this group and stop receiving emails from 
> > > > >> > >>> it, send an email to [email protected].
> > > > >> > >>> To view this discussion on the web visit 
> > > > >> > >>> https://groups.google.com/d/msgid/sage-devel/CAGEwAAm_%2BVfg%3D6po8-7B4eb7kW2UT%3DHjvXaR7yfUOxL930HhQQ%40mail.gmail.com.
> > > > >> > >>
> > > > >> > >> --
> > > > >> > >> You received this message because you are subscribed to the 
> > > > >> > >> Google Groups "sage-devel" group.
> > > > >> > >> To unsubscribe from this group and stop receiving emails from 
> > > > >> > >> it, send an email to [email protected].
> > > > >> > >> To view this discussion on the web visit 
> > > > >> > >> https://groups.google.com/d/msgid/sage-devel/CAAWYfq2yJEvXbRa8znksXtDm%2B-hkXN_EVpwN09X4tb8EY-0-kA%40mail.gmail.com.
> > > > >> > >
> > > > >> > > --
> > > > >> > > You received this message because you are subscribed to the 
> > > > >> > > Google Groups "sage-devel" group.
> > > > >> > > To unsubscribe from this group and stop receiving emails from 
> > > > >> > > it, send an email to [email protected].
> > > > >> > > To view this discussion on the web visit 
> > > > >> > > https://groups.google.com/d/msgid/sage-devel/CAChs6_mqWOpnHAe4-cYTxg63y5ri4vFuJxZRasMUzsEMYzTNAQ%40mail.gmail.com.
> > > > >> >
> > > > >> > --
> > > > >> > You received this message because you are subscribed to the Google 
> > > > >> > Groups "sage-devel" group.
> > > > >> > To unsubscribe from this group and stop receiving emails from it, 
> > > > >> > send an email to [email protected].
> > > > >> > To view this discussion on the web visit 
> > > > >> > https://groups.google.com/d/msgid/sage-devel/CAAWYfq0dH8%3DTp1z_owfAbCw-ED_W%3DHjuh0NQp%2B9MUatapfxKXw%40mail.gmail.com.
> > > > >>
> > > > >> --
> > > > >> You received this message because you are subscribed to the Google 
> > > > >> Groups "sage-devel" group.
> > > > >> To unsubscribe from this group and stop receiving emails from it, 
> > > > >> send an email to [email protected].
> > > > >> To view this discussion on the web visit 
> > > > >> https://groups.google.com/d/msgid/sage-devel/CAGEwAAmj4FkYzJC07v80S%3DqU%3DVOH8vtysU%3Dg5tOs10-qV8Sqxg%40mail.gmail.com.
> > > > >
> > > > > --
> > > > > You received this message because you are subscribed to the Google 
> > > > > Groups "sage-devel" group.
> > > > > To unsubscribe from this group and stop receiving emails from it, 
> > > > > send an email to [email protected].
> > > > > To view this discussion on the web visit 
> > > > > https://groups.google.com/d/msgid/sage-devel/CAAWYfq1QWbtX07O_ZD4Jotjf1XrT9u_BdRuMiS_d7mPXjz2Z7w%40mail.gmail.com.
> > > >
> > > > --
> > > > You received this message because you are subscribed to the Google 
> > > > Groups "sage-devel" group.
> > > > To unsubscribe from this group and stop receiving emails from it, send 
> > > > an email to [email protected].
> > > > To view this discussion on the web visit 
> > > > https://groups.google.com/d/msgid/sage-devel/CAGEwAA%3DmHuOF%3DbERuFY2s0gN4RkWdAxxBKObPZDfie-tfqWQyQ%40mail.gmail.com.
> > >
> > > --
> > > You received this message because you are subscribed to the Google Groups 
> > > "sage-devel" group.
> > > To unsubscribe from this group and stop receiving emails from it, send an 
> > > email to [email protected].
> > > To view this discussion on the web visit 
> > > https://groups.google.com/d/msgid/sage-devel/CAAWYfq0_WYBe9DEZ85kCG_x7xVnG3fcF60bo%2BLgnYjR%2BjpV1Kw%40mail.gmail.com.
> >
> > --
> > You received this message because you are subscribed to the Google Groups 
> > "sage-devel" group.
> > To unsubscribe from this group and stop receiving emails from it, send an 
> > email to [email protected].
> > To view this discussion on the web visit 
> > https://groups.google.com/d/msgid/sage-devel/CAD0p0K6qLK51iayjBArBzgxh3%2Bt40SrbVeRKpDaWAJBu5Kj2TA%40mail.gmail.com.
>
> --
> You received this message because you are subscribed to the Google Groups 
> "sage-devel" group.
> To unsubscribe from this group and stop receiving emails from it, send an 
> email to [email protected].
> To view this discussion on the web visit 
> https://groups.google.com/d/msgid/sage-devel/CAAWYfq3PvEQZaRjO2kjYKv5Cm%2Bnq0b9%2BDmMR8WiPSbsFdUfOMQ%40mail.gmail.com.

-- 
You received this message because you are subscribed to the Google Groups 
"sage-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion on the web visit 
https://groups.google.com/d/msgid/sage-devel/CAD0p0K5uHLf9fq5s9x27tdEyjRpBTQeq5RaCJ%2BwL0jZmLyrPvw%40mail.gmail.com.

Reply via email to