From mboxrd@z Thu Jan  1 00:00:00 1970
From: Richard Henderson <rth@cygnus.com>
To: John Carr <jfc@mit.edu>
Cc: Richard Henderson <rth@cygnus.com>, egcs@cygnus.com
Subject: Re: an experiment with garbage collection
Date: Sat, 09 May 1998 16:35:00 -0000
Message-id: <19980509154259.19062@dot.cygnus.com>
References: <19980509012500.06283@dot.cygnus.com> <199805092232.SAA26244@jfc.>
X-SW-Source: 1998-05/msg00335.html

On Sat, May 09, 1998 at 06:32:24PM -0400, John Carr wrote:
> Can you guess how much work it would take to be able to run a
> not-conservative GC in the middle of compiling a function?

That is what I am doing, actually.  The only thing extra you would
need from the information I am collecting now is the address of the
pointers.  This could be trivially gotten with

	#define ggc_mark_rtx(x)	  ggc_mark_rtxp (&(x))

or similar.

> Even if you don't like obstack semantics, I think it is still useful to
> have a special allocator for rtl to avoid malloc overhead (both memory and
> time).  At least special-case the smallest, most common size: 8 bytes (16
> on 64 bit systems).

This is something I planned to look into after it works.

Oh, I've been able to do some more testing.  GC is currently eating
about 15% of the run time, which is a bit more than I expected.  No
data on size reduction yet.


r~