This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: tree-ssa branch: Call for preliminary testing
- From: Andrew MacLeod <amacleod at redhat dot com>
- To: Daniel Berlin <dberlin at dberlin dot org>
- Cc: Steven Bosscher <s dot bosscher at student dot tudelft dot nl>,Diego Novillo <dnovillo at redhat dot com>,Richard Guenther <rguenth at tat dot physik dot uni-tuebingen dot de>,gcc mailing list <gcc at gcc dot gnu dot org>
- Date: 12 May 2003 10:12:51 -0400
- Subject: Re: tree-ssa branch: Call for preliminary testing
- References: <E4499C0C-8481-11D7-87A1-000A95A34564@dberlin.org>
On Mon, 2003-05-12 at 09:59, Daniel Berlin wrote:
>
> Apparently you've never had the joy of working with some form of
> interference graph.
> :)
>
> Andrew, if this relationship is symmetric (ie a interferes with b
> implies b interferes with a), you can halve the size and represent it
> as a triangular bitmatrix like we do in new-regalloc.
> grep for igraph and igraph_index.
>
>
It reuses the conflict.c interference graph. Its more likely some of the
live range bitvectors. I think live-on-exit is a #BB x #vars matrix. I
have a patch laying around somewhere that uses sbitmaps when the size is
small, and bitmaps when its large, but when I originally tested it, it
was for time, and it didn't buy anything. Perhaps it will be useful
afterall. We'll see :-)
Im also working on reducing the number of variables. There will be a
note about it later today. There are a lot of variables which are used
for aliasing and stuff that never need to be rewritten and can be
trimmed out of the graph and live range vectors.. the whole cycle.
Andrew