http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55838
--- Comment #7 from Steven Bosscher <steven at gcc dot gnu.org> 2013-01-01 19:52:48 UTC --- ... but actually it looks like it exists even before that: (gdb) p debug_rtx(desc->niter_expr) (udiv:QI (minus:QI (const_int -126 [0xffffffffffffff82]) (subreg:QI (reg:SI 113 [ ivtmp.14 ]) 0)) (const_int 129 [0x81])) That's a nonsense QImode udiv from loop-iv.