> Jan Hubicka <hubi...@ucw.cz> writes: > > > > Now for longer term, we want to make function CFGs independent of gimple > > body > > and we want to decide on instrumentation at linktime, so we won't require > > user > > to rebuild with -fprofile-use, just relink. I plan to work on this, but > > not for 4.9. > > It may be already obsolete with AutoFDO? > > Longer term I hope more and more of the explicit instrumentation > functionality can be implemented using the hardware PMU. > > For example (auto fdo doesn't do this today) it would be quite > possible to do x86 value profiling without explicit instrumentation > too by looking at the PEBS records, and likely similar functionality > elsewhere.
Well, AutoFDO is great feature, but I do not think it is going to replace FDO really soon. FDO has advantages in letting you to measure things exactly and reproducibly w/o hardware dependencies. But yep, with debug info based AutoFDO we may be able to unify comdat function profiles w/o the issues we currently see with FO. I am curious how both features will develop... Honza > > -Andi > > -- > a...@linux.intel.com -- Speaking for myself only
