https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107561
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> --- This probably boils down to the object size machinery ignoring [0, 0] (aka not picking a convex hull of the range). The range implementation of all these passes should be using irange<> (but with wide_ints, not trees) instead of doing their own thing.
