Re: [wwwdocs] Mention cgraph stuff in news/changes?

[Jan Hubicka <jh at suse dot cz>]

> On Sun, 13 Jul 2003, Steven Bosscher wrote:
> > Gerald has said he would like to see big changes mentioned on the web.
> > Assuming the cgraph work is worth mentioning, how about this?
> 
> The changes.html part looks fine, thanks!
> 
> Concerning the announcement on the main page, Jan is on holidays now,
> and Andreas, Jan, and myself had planned to create a page under news/
> with more details on this aspect of Jan's work after his return, so I'd
> hold of that one (for now).
Actually I was planning to write something once all the stuff is on the
place together with some numbers.  I hoped it to be right now, but the
intermodule patch is bringing more funn (it mess up the callgraph
completely as the invariant that function/variable can be uniquely
identified by DECL_ASSMEBLER_NAME is broken so compiler crash and
misscompile).

It will be bit tricky to fix this but I will try to have it by tomorrow.

Also the scores of preriodic testers has regressed.  I duno what caused
this as there are regressions in SPECFP f77 tests that should not be
influenced by my work at all.  So I guess it should be at least partly
caused by some other commit same day.  I don't have time to figure out
more this week unforutnately :(

The few C regressions are probably because inlining limits are very low.
Bumping them by 50% I got about 1% speedup on SPECs, but the picture is
slighly different now as Richard suggested to walk the trees with
duplicates when estimating body size.  I hope that we will be able to
make unit-at-a-time default at -O2 and then we can increase the limits
without getting into degenerative cases on POOMA

Honza
> 
> Thanks for bringing this up!
> 
> Gerald
> 
> + This will be enabled by default at <code>-O3</code> in the GCC 3.4
> + release.
> 
> This I'd shorten to "This will be enabled at <code>-O3</code> in GCC 3.4".