My change on the probability of builtin_expect does have an impact on the partial inline (more outlined functions will get inline back to the original function). I think this triggers an existing issue. Dehao will explain this in his coming email.
-Rong On Fri, Oct 4, 2013 at 6:05 AM, Ramana Radhakrishnan <ramra...@arm.com> wrote: > On 10/02/13 23:49, Rong Xu wrote: >> >> Here is the new patch. Honaz: Could you take a look? >> >> Thanks, >> >> -Rong >> >> On Wed, Oct 2, 2013 at 2:31 PM, Jan Hubicka <hubi...@ucw.cz> wrote: >>>> >>>> Thanks for the suggestion. This is much cleaner than to use binary >>>> parameter. >>>> >>>> Just want to make sure I understand it correctly about the orginal >>>> hitrate: >>>> you want to retire the hitrate in PRED_BUILTIN_EXPECT and always use >>>> the one specified in the biniltin-expect-probability parameter. >>> >>> >>> Yes. >>>> >>>> >>>> Should I use 90% as the default? It's hard to fit current value 0.9996 >>>> in percent form. >>> >>> >>> Yes, 90% seems fine. The original value was set quite arbitrarily and no >>> real >>> performance study was made as far as I know except yours. I think users >>> that >>> are sure they use expect to gueard completely cold edges may just use >>> 100% >>> instead of 0.9996, so I would not worry much about the precision. >>> >>> Honza >>>> >>>> >>>> -Rong >>>>> >>>>> >>>>> OK with that change. >>>>> >>>>> Honza > > > > This broke arm-linux-gnueabihf building libstdc++-v3. I haven't dug further > yet but still reducing the testcase. > > See > > http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58619 > > for details. > > Can you please deal with this appropriately ? > > regards > Ramana > >
