This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH][RFC] Do not error on inconsistent profiles (PR38977)


On Tue, 27 Jan 2009, Ian Lance Taylor wrote:

> Richard Guenther <rguenther@suse.de> writes:
> 
> > It happens that some openSUSE packages do no longer build with GCC 4.4
> > as they error on inconsistent profiles.  In 4.3 this kind of
> > inconsistency was simply ignored, now we have the possibility to
> > "correct" them by smoothing - but this is not enabled by default, instead
> > we now error by default.
> >
> > I suggest we revert to ignoring this kind of problem for 4.4 and
> > turn on profile smoothing by default for 4.5 (it is quite late for 4.4
> > now, and obviously this code-path has been not extensively tested).
> >
> > Comments?
> >
> > Thanks,
> > Richard.
> >
> > 2009-01-26  Richard Guenther  <rguenther@suse.de>
> >
> > 	PR tree-optimization/38977
> > 	* profile.c (compute_branch_probabilities): Do not error on
> > 	inconsistent profiles.
> 
> 
> What if we change the error to a warning?  As I understand it, the
> corrupted profile may lead to optimizing based on meaningless data,
> and it might be just as good to simply ignore the profiling data
> entirely.
> 
> Or if the warning leads to -Werror issues, I suppose I'm OK with just
> ignoring the corruption.

Actually the patch didn't fix my problems, we just get other
errors later.  So I won't pursue this particular patch further.

It hints at that we have some broken profile stuff around though.

Richard.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]