[Bug target/125106] [17 Regression] 5% slowdown of milc on Zen2 with PGO since r17-140-gf8d911e6ae3fc1

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125106

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
There was a small followup fix in r17-207-gc392d64098cc67, let's see if that
resolves it (also, jump up-down).

The series is expected to make some BB vectorizations profitable that were not
previously.