http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54629
--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-09-20 07:41:01 UTC --- Created attachment 28225 --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28225 gcc48-pr54629.patch IMHO there are two issues - one is that the loop nest is freed too early, before it is still used, and another one that on success the ddr isn't freed at all, so we leak memory.