https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125788
--- Comment #8 from Guoce Feng <fengguoce at hygon dot cn> --- I agree that a PHI-driven approach should fit GCC better than the strict lockstep predecessor-tail scan I described, especially because it avoids being too sensitive to local statement ordering. Also agreed about the example in comment #6; it is too simple since PRE, phiopt and ifcvt already handle it. The sqrt/div examples in this PR are better motivating cases. Thanks for the explanation about how PRE hoisting, phiopt commonization and ifcvt interact here.
