This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [tree-ssa] Avoid nondeterminism in tree-ssanames
- From: Jan Hubicka <jh at suse dot cz>
- To: law at redhat dot com
- Cc: Andrew MacLeod <amacleod at redhat dot com>,Gerald Pfeifer <gp at suse dot de>, Jan Hubicka <jh at suse dot cz>,Diego Novillo <dnovillo at redhat dot com>,Geoff Keating <geoffk at geoffk dot org>,gcc-patches <gcc-patches at gcc dot gnu dot org>
- Date: Wed, 3 Dec 2003 23:34:41 +0100
- Subject: Re: [tree-ssa] Avoid nondeterminism in tree-ssanames
- References: <1070476847.9315.2389.camel@p4> <200312031848.hB3ImUfW010515@speedy.slc.redhat.com>
> In message <1070476847.9315.2389.camel@p4>, Andrew MacLeod writes:
> >On Wed, 2003-12-03 at 13:39, Gerald Pfeifer wrote:
> >> On Mon, 1 Dec 2003, Jan Hubicka wrote:
> >> > It makes 11% performance difference on Gerald's testcase on my setup,
> >> > just FYI.
> >>
> >> If I read this correctly...
> >>
> >> On Mon, 1 Dec 2003, Diego Novillo wrote:
> >> > Unless it's a huge win, I would *really* want to go back to the previous
> >> > scheme of having unique SSA numbers across the whole compilation unit.
> >>
> >> ...you might be able to convince Diego. :-)
> >>
> >> 11% on PR8361 is quite impressive an improvement!
> >
> >I'd want to know where that 11% came from, it sounds like one (or more)
> >of the optimizations are doing something pretty inefficient and ought to
> >be looked at first. Just reducing the number of SSA_NAME ought not have
> >that significant an effect if everyone is being reasonably intelligent
> >about how they do things.
> >
> >Its definately worth investigating.
> Well, I'm not sure I believe the 11% number to start with. If it is
> believable, then like you, I'd think we've got an optimizer that is
> doing something dumb.
I don't trust it really either. My system used to go into swap while
compiling the testcase (don't do anymore after PHI and
cgraph_preserve_function_body changes) and I re-tested with not very
noticeable differences now.
Lets wait for zones before going crazy here :)
Honza
>
> Jeff
>