This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Value Range Propagation Pass Status
- To: John Wehle <john at feith dot com>
- Subject: Re: Value Range Propagation Pass Status
- From: Jeffrey A Law <law at cygnus dot com>
- Date: Mon, 07 Feb 2000 09:57:52 -0700
- cc: gcc at gcc dot gnu dot org
- Reply-To: law at cygnus dot com
In message <200002050005.TAA25324@jwlab.FEITH.COM>you write:
> > I doubt there's much benefit in running it twice. I suspect we can get 9
> 9%
> > of the benefit by running it once, probably before loop.
>
> Running a second time after loop occurred to me in order to handle
> unnecessary condition jumps resulting from loop unrolling which I've
> seen happen. For example using -O2 -funroll-all-loops on a x86 platform
> to compile:
It may be the case that we want to run a second pass after loop if we're
unrolled any loops.
I've been thinking about ways to restructure the main loop in toplev.c so
that each optimization sets/resets flags so that we can intelligently rerun
passes only when they're likely to really help (or selectively invalidate
information like the cfg, life info, etc). But I haven't actually tried to
do this yet :-)
jeff