https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120629
--- Comment #20 from Jakub Jelinek <jakub at gcc dot gnu.org> --- In all 3 cases the addition of the BB_RTL bbs looks like if we'd split one of the edges (in the first case the ENTRY->succs[0] edge, in the latter case the EDGE_FALSE_VALUE edge I think. So handling it in the ranger as an empty bb should reflect original IL.
