This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: basic-block and profile-based optimizing (was Re: New attribute "infrequent"?)
- To: Daniel Berlin <dan at cgsoftware dot com>
- Subject: Re: basic-block and profile-based optimizing (was Re: New attribute "infrequent"?)
- From: Jan Hubicka <jh at suse dot cz>
- Date: Fri, 31 Aug 2001 15:33:39 +0200
- Cc: mike stump <mrs at windriver dot com>, aj at suse dot de, crosby at qwes dot math dot cmu dot edu, apfk at fuchs dot offl dot uni-jena dot de, gcc at gcc dot gnu dot org, jh at suse dot cz
- References: <200108311319.GAA29220@kankakee.wrs.com> <87u1yos6tb.fsf@cgsoftware.com>
> mike stump <mrs@windriver.com> writes:
>
> >> To: Scott A Crosby <crosby@qwes.math.cmu.edu>
> >> From: Andreas Jaeger <aj@suse.de>
> >> Date: Fri, 31 Aug 2001 14:11:01 +0200
> >
> >> FYI Jan is working on profile-based optimizations at the moment, we've
> >> just did a small announcement at
> >> http://gcc.gnu.org/news/profiledriven.html
> >
> > When to inline in C++ might best be answered by profile based
> > techniques. Love to see any results, if someone wires it up and tries
> > it out.
>
> Not that easy.
> Jan's stuff works on RTL.
> The C++ inliner works on trees.
> For basic blocks on trees, you'd need to work on the ast-optimizer
> branch.
Yes - to get inliner working on profile feedback we first need
1) finish the transformation of backed for CFG (loop optimizer is main showstopper now)
2) integrate the ast-optimizer
3) Map the CFG on ast to CFG on RTL and instrument code eitghter on AST level or
in some way postnote it to RTL, but make it readable on AST.
Honza
>
> --
> "I invented the cordless extension cord.
> "-Steven Wright