This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [IPA] Inline before other IPA optimizations
> On Sun, 2005-08-14 at 19:36 +0200, Richard Guenther wrote:
> > On 8/14/05, Jan Hubicka <jh@suse.cz> wrote:
> > > Hi,
> > > this patch makes the long promised (and now finally easy ;) change to
> > > inline everything during the IPA inline pass instead of deferring it to
> > > tree_rest_of_compilation. The advantage of this scheme is mainly in the
> > > fact that other IPA passes don't need to worry about the possibly
> > > numberous inline clones around cgraph and we also order functions in a
> > > way so saving is done seldomly instead of usually so we produce fewer
> > > garbage.
> > >
> > > The disadvantage is increase of peak memory usage just after inlining.
> > > It is about 50% for Gerald's testcase (as the inline-unit-groth suggest
> > > after all), but don't seems to be as bad as I anticipated.
> >
> > Uh, oh. How much does it improve if you run some scalar cleanup
> > passes right after (individual) inlining?
>
> Memory usage will be high until we do something about the size of our
> IR. Even scalar cleanups aren't going to save us.
> That's just the way it's gonna be temporarily.
Sadly...
On the brighter side, it is just usage right after lowering that went
50% up. We consume a lot of memory later so overall numbers would be
more pleasant (I didn't measured them however).
Honza
>
> >
> > Richard.
>