http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54742
--- Comment #9 from Steve Ellcey <sje at gcc dot gnu.org> 2013-05-02 00:11:52 UTC --- See http://gcc.gnu.org/ml/gcc/2013-05/msg00009.html for a dynamically loadable pass to do this optimization. It is not a finished product but it seems to work for coremark.