[Bug tree-optimization/77536] Vectorizer not maintaining relationship of relative block frequencies in absence of real profile data

[amker at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77536

--- Comment #5 from amker at gcc dot gnu.org ---
(In reply to amker from comment #4)
> Looks like generic loop unrolling code used by predcom spends quite a lot
> maintaining profiling counter, I will check if that's correct and we shall
> do the same in vectorizer.  At least simply scaling down vect_loop is wrong.
> Thanks.

Testing patch for this issue.