This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [tree-ssa] Remove semi-pruned support
- From: Chris Lattner <sabre at nondot dot org>
- To: law at redhat dot com
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Tue, 20 Jan 2004 10:38:39 -0600 (CST)
- Subject: Re: [tree-ssa] Remove semi-pruned support
On Tue, 20 Jan 2004 law@redhat.com wrote:
> > Have you found anything that would be made more powerful or easier with it?
> IIRC Daniel had a code snippet which might have benefited from using
> semi-pruned and I've got vague memories of seeing a mention in a paper
> that the extra PHIs were sometimes useful. I've never personally run into
> any code or transformation which was easier/better in semi-pruned form.
Ok, just checking. I remember in the original papers there was some vauge
hand-wavy suggestions on how it could be useful, but it seemed like more
general techniques subsumed them anyway.
> Beyond those, the objection from Diego and Dan to using fully pruned all the
> time was that it could hurt compile-time performance. My experiments from
> a long time showed that fully pruned all the time was a slight compile-time
> loser -- but since then Andrew made some changes to improve life analysis and
> we're completely avoiding the life computation for a large subset of the
> variables. The combination of those two improvements makes the compile time
> issue basically go away.
That matches my exerience as well. It's much cheaper to decide not to
place PHI's than to place them and subsequently delete them: especially
for high-degree basic blocks.
Thanks!
-Chris
--
http://llvm.cs.uiuc.edu/
http://www.nondot.org/~sabre/Projects/