https://gcc.gnu.org/bugzilla/show_bug.cgi?id=46590
--- Comment #53 from Richard Biener <rguenth at gcc dot gnu.org> --- After both are fixed there's still non-linearity from loop-invariant, this time in the DF RD problem. Doubling the testcase goes from 9.1s to 25.5s, doubling again to 82.2s with RD going from df reaching defs : 0.67 ( 7%) 0.01 ( 3%) 0.72 ( 8%) 0 kB ( 0%) to df reaching defs : 3.79 ( 15%) 0.05 ( 7%) 3.79 ( 14%) 0 kB ( 0%) and df reaching defs : 11.47 ( 14%) 46.48 ( 88%) 58.41 ( 43%) 0 kB ( 0%)