https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85180
--- Comment #13 from Richard Biener <rguenth at gcc dot gnu.org> --- Created attachment 43851 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=43851&action=edit other alternative Like this. 11.00user 0.01system 0:11.01elapsed 99%CPU (0avgtext+0avgdata 43204maxresident)k 0inputs+144outputs (0major+7848minor)pagefaults 0swaps needed to avoid calling cselib_sp_based_value_p because that didn't like the "messed up" val_rtx. Now the question is whether that's too dangerous given we don't control FIND_BASE_TERM as defined by targets... (I could guard that with !VALUE and re-do that below in the VALUE case after the caching, but eventually that function may recurse itself and wreck the whole thing anyways... it also looks like it is invented to do some RTX massaging before finding the base value, not find it itself?)
