http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46590
--- Comment #35 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-09-04 13:17:53 UTC --- With -O1 FRE uses loads of memory and compile-time. Not the value-numbering itself but compute_avail ()! Bah. Of course AVAIL sets get bigger and bigger going down the dominator tree. For FRE a single domwalk after SCCVN computing "avail" and doing elimiation would be enough. Especially as all the overhead in AVAIL is just SSA DEFs and their value.