https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125571
--- Comment #20 from Richard Biener <rguenth at gcc dot gnu.org> --- (In reply to Eric Botcazou from comment #19) > > The only option is likely to go for a different PHI insertion algorithm that > > does not require iterated dominance frontiers (but by construction is then > > less optimal). > > I suppose that it is too late to go back in time and do not go into SSA form > for cases like this? Yes, I think so. There's also -O1 which IMO should behave reasonably even for unreasonable machine-generated code.
