https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124137
--- Comment #16 from Richard Biener <rguenth at gcc dot gnu.org> --- I verified the 2nd larger testcase also completes in time (54s) now, with machine dep reorg : 0.12 ( 0%) 774k ( 0%) There's still a correctness (IMO) patch pending. Possibly, as FLAGS_REG should not be live across edges and the relevant hardregs should only be live from function start (as the local live doesn't handle new live starts), all the iteration fluff can be removed and we should be able to assert we never need it ...
