Re: [Caml-list] Closing the performance gap to C

[Mark Shinwell <mshinwell@janestreet.com>]

I agree with Gabriel, but, we are planning substantial work within
Flambda during the coming year to implement unboxing transformations
therein.  I do think we should spend a little time, before diving into
this particular issue, convincing ourselves that these two tracks of
work are going to be complementary rather than conflicting.

Mark

On 21 December 2016 at 16:47, Gabriel Scherer <gabriel.scherer@gmail.com> wrote:
> I think there is an information asymmetry here. Alain has done a lot of work
> on float unboxing in old and recent OCaml releases, and if he says that some
> approach is a low-hanging fruit it is reasonable to trust him. I suspect,
> Yotam, that you have different issues in mind that correspond to the fact
> that inlining and specialization at the flambda level could be complementary
> with the work that Alain is describing.
>
> (I think it is difficult to accurately discuss these optimization questions
> at a high/conceptual level only.)
>
> On Wed, Dec 21, 2016 at 11:39 AM, Yotam Barnoy <yotambarnoy@gmail.com>
> wrote:
>>
>> On Wed, Dec 21, 2016 at 11:31 AM, Gerd Stolpmann <info@gerd-stolpmann.de>
>> wrote:
>> > Dumb question because you are effectively suggesting an alternate
>> > calling convention in addition to the existing one: wouldn't it make
>> > more sense to switch to a completely different convention? Like: we
>> > pass fp values always in fp registers so far available?
>> >
>> >
>> > Gerd
>>
>> As far as I understand, the current calling convention has to do with
>> generic functions. Having to be polymorphic over all types, including
>> primitive ones, such as floating point and ints of different widths,
>> is really difficult. This is why OOP languages don't do it -- only
>> classes are polymorphic.
>>
>> And Alain, even if we create these alternate calling conventions, the
>> tricky part is still detecting when it's ok to call using direct
>> calling conventions. That's the hard part, and it's best done by
>> Flambda with its inlining.
>>
>> -Yotam
>>
>> >
>> > Am Mittwoch, den 21.12.2016, 17:06 +0100 schrieb Alain Frisch:
>> >> On 21/12/2016 15:45, Yotam Barnoy wrote:
>> >> >
>> >> > I think it's not worth the effort. You need to examine all the code
>> >> > dealing with a parameter (ie. its flow) to see if any generic
>> >> > function
>> >> > is called on that parameter.
>> >> This would be treated a bit like the stubs for optional arguments
>> >> with a
>> >> default value.  Any function taking float arguments or returning a
>> >> float
>> >> would be compiled to a specialized version with an unboxed calling
>> >> convention plus a stub which implements the generic calling
>> >> convention
>> >> and delegate the call to the specialized version (or copy its body if
>> >> it
>> >> is small enough).  On call site, when the called function is known,
>> >> one
>> >> calls the specialized version instead.  This is a systematic, local
>> >> compilation scheme, similar to other optimizations made in
>> >> closure/cmmgen; it does not require a more global analysis nor a
>> >> radically different representation of the code.
>> >>
>> >> About the "it's not worth the effort": the effort has largely been
>> >> done,
>> >> since the ticket came with a working patch (some more effort would
>> >> be
>> >> needed to bring it up to date, though).  In my opinion, this seems
>> >> like
>> >> a rather low-hanging fruit with very immediate and significant
>> >> gains.
>> >> I'd rather have this soon than wait for flambda to become stable and
>> >> usable on large projects.
>> >>
>> >>
>> >>
>> >> -- Alain
>> >>
>> > --
>> > ------------------------------------------------------------
>> > Gerd Stolpmann, Darmstadt, Germany    gerd@gerd-stolpmann.de
>> > My OCaml site:          http://www.camlcity.org
>> > Contact details:        http://www.camlcity.org/contact.html
>> > Company homepage:       http://www.gerd-stolpmann.de
>> > ------------------------------------------------------------
>> >
>> >
>>
>> --
>> Caml-list mailing list.  Subscription management and archives:
>> https://sympa.inria.fr/sympa/arc/caml-list
>> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
>> Bug reports: http://caml.inria.fr/bin/caml-bugs
>
>