This is the mail archive of the
guile@sourceware.cygnus.com
mailing list for the Guile project.
Re: markbits out of cells. 1st attempt.
Jim Blandy <jimb@savonarola.red-bean.com> writes:
> If the hype is true, and memory references are really so much more
> expensive than computation in registers, the best way to locate the
> mark bits is with something like this:
> [ schnipp ]
> I don't think theoretical arguments will resolve much; we should just
> try things and see what's fastest.
OK, so I did (sorry Han-Wen, I didn't use your code, as I judged it
would be less effort to finish mine -- the NIH phenomenon in action).
results: Greg Harvey's scheme with the markbits in their own dedicated
pages really rocks. like, totally. at least on the simple GC
benchmark from guile-modules and without using any swap, and I only
checked on UltraSparc and i686.
so I'm waiting for the end of the code freeze, and meanwhile Han-Wen
is going to be a guinea pig, whether he likes it or not ;).
--
SIGTHTBABW: a signal sent from Unix to its programmers at random
intervals to make them remember that There Has To Be A Better Way.
-- Erik Naggum