This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [boehms-gc] Performance results


Le Tue, Jul 25, 2006 at 12:30:04PM -0500, David Nicol écrivait/wrote:
> On 7/24/06, Laurynas Biveinis <laurynas.biveinis@gmail.com> wrote:
> 
> >[How is it that setting pointers] to NULL can
> >actually increase peak GC memory usage?
> 
> I'll guess that during collection phases, the list of
> collectible structures becomes longer.
> 
> GCC is just too huge to try and implement reference counting
> just to see what it would do, right?  All managed structures would have
> to include a base class with the count in it and all new references would
> have to be through a macro...

Refcouting is also a nightmare for circular references. I suspect that
GCC gimple trees has some circularities (I might be wrong, not having
dived yet into the code).

A theoritical alternative might be to use a precise copying collector
like Qish.  http://starynkevitch.net/Basile/qishintro.html In
practice, this would be impractical (even if the GTY marker could
help) because it requires changing a lot of code.

I've got no explanation regarding zero-ing pointers penalties. Maybe
(just a guess) some cache effect?

Regards

-- 
Basile STARYNKEVITCH         http://starynkevitch.net/Basile/ 
email: basile<at>starynkevitch<dot>net 
aliases: basile<at>tunes<dot>org = bstarynk<at>nerim<dot>net
8, rue de la Faïencerie, 92340 Bourg La Reine, France


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]