http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45820
--- Comment #1 from John David Anglin <danglin at gcc dot gnu.org> 2010-10-02 17:20:26 UTC --- Actually, the insn doesn't satisfy its constraints because %r31 should be %r1. Have a patch. This is an old regression caused by a change to pa_secondary_reload.
