This is the mail archive of the
mailing list for the GCC project.
Re: [patch] Disable tree-ssa DSE for now
- From: Daniel Berlin <dberlin at dberlin dot org>
- To: Diego Novillo <dnovillo at redhat dot com>
- Cc: Steven Bosscher <stevenb at suse dot de>, gcc-patches at gcc dot gnu dot org, law at redhat dot com
- Date: Tue, 18 Jan 2005 22:09:55 -0500 (EST)
- Subject: Re: [patch] Disable tree-ssa DSE for now
- References: <firstname.lastname@example.org> <41EDCA24.email@example.com>
On Tue, 18 Jan 2005, Diego Novillo wrote:
Steven Bosscher wrote:
Thoughts?Well, given your analysis, disabling it is not a bad idea. OTOH, DSE doesn't
take a long time to run, does it? I don't think I've ever noticed it.
Also, just because a pass is fast doesn't mean we should keep it enabled
Down this path, slowness (because it's unnecessary work) lies.
We didn't get slower 30 or 40% a time, it was .7% or .3%
Jeff, maybe there's a couple of things we can do for 4.0? And then do a
complete rewrite for 4.1.
The algorithm DSE is using could probably be replaced with a modified
version of GVN-PRE (we need a kill set, but we need this for load PRE
anyway), on the reverse flowgraph.
I'll look into it for 4.1.