This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [tree-profiling] Pre-inline optimizations
> Jan Hubicka wrote:
> >>On Thu, 2005-03-03 at 08:51 +0100, Jan Hubicka wrote:
> >>
> >>>Hi,
> >>>this patch makes it possible to execute optimizations early (ie before
> >>>IPA is
> >>>performed).
>
> I'd be curious what the existing prior art is in this area. Have you
> talked to Kenny Z. or Danny B. about this at all? I'm not oppposed, and
Yes, we talked about this. In general we all agree that optimizations
are needed, but of course in longer term we don't want to go in and out
SSA.
> I can certainly see how this would help; I'm just curious. I also
> wonder whether we should fix the in-and-out-of-SSA problem before
> putting this in; that seems like something that we're starting to run up
> against in multiple contexts, and going back and forth is pretty
> expensive in terms of both time and memory.
Yes, however staying in SSA form is expensive too - we do have pretty
bloated representation. Zdenek did some calculations and it looks like
the peak memory usage would be increased roughly twice compared to
existing compiler. So both approaches have their advantages and
disadvantages on the moment and this one is easier to implement. My
plan is to get this one stable, try to save something from memory usage
and then start pushing the SSA further into inliner. We discussed with
Kenny that we need to do some datastructure changes - like the variable
UIDs being local for each functions seems to be stopper for IPA, while
making them global would ruin locality in the arrays indexed by them and
such.
Honza
>
> --
> Mark Mitchell
> CodeSourcery, LLC
> mark@codesourcery.com
> (916) 791-8304