This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Clean up early inliner
On Wed, 14 Apr 2010, Jan Hubicka wrote:
> > Hi,
> > for some reason the patch seems to regress compile time for flatten tramp3d
> > tester as well as runtime. I think runtime is caused by fact that overall size
> > for small inlining should be based on the unit size after flattening (so
> > flattening won't blow up all unit growth limits.).
> >
> > This has chance to fix the compile time issues too, if we get lucky.
> > I will test patch after testers pick up my yesterday inliner fix.
> ... and I guess this is also good reason for making early inliner to
> handle flattening.
> Even with profile feedback flattening can be used to get better profile
> generation this way (as well as better profile estimates without)
Well, it doesn't fit nicely into incremental inlining when you do that
in postorder. We should just adjust the unit size adjustments during
flattening.
Richard.