This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: GCC 3.3, GCC 3.4
- From: Daniel Berlin <dberlin at dberlin dot org>
- To: Richard Henderson <rth at redhat dot com>
- Cc: Tim Josling <tej at melbpc dot org dot au>, "" <gcc at gcc dot gnu dot org>, Neil Booth <neil at daikokuya dot co dot uk>, Mike Stump <mrs at apple dot com>, Benjamin Kosnik <bkoz at redhat dot com>
- Date: Mon, 3 Feb 2003 23:14:14 -0500 (EST)
- Subject: Re: GCC 3.3, GCC 3.4
- References: <1043976898.27601.ezmlm@gcc.gnu.org> <3E3E226F.1DA63EDF@melbpc.org.au><20030204020323.GB14743@redhat.com>
On Mon, 3 Feb 2003, Richard Henderson wrote:
> On Mon, Feb 03, 2003 at 07:04:00PM +1100, Tim Josling wrote:
> > 1. Allocating storage of similar sizes together, sacrificing locality.
> > 2. Wasting space by rounding up allocations to the next power of 2, wasting
> > space and increasing working set sizes.
> > 3. Having bits of GC code all over the place impacting code locality and
> > working set sizes.
>
> Number 2 is already done. We special case the few non-power-of-2
> cases that actually get used.
>
> As for the rest, I had a patch (before the pch merge) that changed
> the allocator such that there were only two size buckets: larger
> and smaller than 1 page. I did get somewhat better peak memory
> usage, but the compiler actually slowed down some more (!), which
> is why I havn't pushed the code in.
Err, the generational copying collector patch i posted actually solves 1 +
2 + part of 3, but only for things besides rtl and trees (which are
pointed to by non-roots, and thus, would require stack scanning, etc).
--Dan