https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117964
--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> --- In principle I'd place the duplication at a point in the pass pipeline where we do _not_ need the outgoing edges at all (we're already adding a BARRIER!). Which likely means no accurate dataflow. Possibly a new PROP_* we can assert on when asked to do DF. But of course separating the actual indirect jump instruction (which we want to duplicate) and the "fake" edge producing instruction might be a better solution.
