https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109303
--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> --- BTW, I think I got that offset >= (HOST_WIDE_INT) UINT_MAX * BITS_PER_UNIT wrong last time, if offset is equal to that, it is still representable (as UINT_MAX). So perhaps either offset > (HOST_WIDE_INT) UINT_MAX * BITS_PER_UNIT or >= (UINT_MAX + HOST_WIDE_INT_1) * BITS_PER_UNIT, depending on whether we expect the divisions to be exact or truncating. On the other side, if it is just an optimization, who cares...
