https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81442
--- Comment #7 from Tom de Vries <vries at gcc dot gnu.org> --- (In reply to Tom de Vries from comment #5) > ... the patches handle the issue in different ways: > - handle the case that there's an edge with uninitialized probability > - initialize probability on some edges > > At first glance I'd say that one doesn't necessarily exclude the other. Having reflected a bit further, while the point solution from comment 1 is something that always works, it also hides the fact that we have inconsistent edge info. It's probably better to check more aggressively for the inconsistency: I've submitted a tentative patch at https://gcc.gnu.org/ml/gcc-patches/2017-08/msg00194.html