http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59208
--- Comment #7 from Jan Hubicka <hubicka at gcc dot gnu.org> --- Created attachment 31322 --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=31322&action=edit Proposed fix I am testing Hi, the problem here is that update_stmt is called with cfun being set to NULL. It uses ssa_operands_active (cfun) call that return false. This patch makes us to set cfun correctly when updating the function body. It may be interesting to investigate why some testcases fails on PPC and not on x86 - they may be cases where we missed devirtualization during early optimization but for whatever reason we now handle it during IPA Honza