> On 3/19/19 2:02 AM, Martin Liška wrote:
> > Hi.
> >
> > When calling set_even_probabilities, the function assumes that an edge
> > can't live in both sets ({un,}likely_edges). When such situation happens,
> > clear just the sets.
> >
> > Patch can bootstrap on x86_64-linux-gnu and survives regression tests.
> >
> > Ready to be installed?
> > Thanks,
> > Martin
> >
> > gcc/ChangeLog:
> >
> > 2019-03-18 Martin Liska <[email protected]>
> >
> > PR middle-end/89737
> > * predict.c (combine_predictions_for_bb): Empty likely_edges and
> > unlikely_edges if there's an edge that belongs to both these sets.
> >
> > gcc/testsuite/ChangeLog:
> >
> > 2019-03-18 Martin Liska <[email protected]>
> >
> > PR middle-end/89737
> > * gcc.dg/pr89737.c: New test.
> But is having the edge in both sets a valid state? You know this code
> better than I, so I'll go with your recommendation here.
It is the situation you add both hot and cold attribute which I guess
can happen.
I think it is OK to ignore both hint then, so patch is OK.
Honza
>
> jeff
