https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108463
--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> --- The first two differences are between insns 1204 vs. 1116 and 1204 vs. 1095: (insn 1095 1094 1097 2 (set (mem/c:V4SI (plus:DI (reg/f:DI 7 sp) (const_int -104 [0xffffffffffffff98])) [1 S16 A128]) (reg:V4SI 22 xmm2 [227])) "pr106746.c":27:5 1809 {movv4si_internal} (expr_list:REG_DEAD (reg:V4SI 22 xmm2 [227]) (nil))) (insn 1116 1115 1118 2 (set (mem/c:V4SI (plus:DI (reg/f:DI 7 sp) (const_int -88 [0xffffffffffffffa8])) [1 S16 A128]) (reg:V4SI 22 xmm2 [247])) "pr106746.c":27:5 1809 {movv4si_internal} (expr_list:REG_DEAD (reg:V4SI 22 xmm2 [247]) (nil))) (insn 1204 1118 1189 2 (set (reg:SI 0 ax [250]) (mem:SI (plus:DI (reg/f:DI 7 sp) (const_int 120 [0x78])) [1 S4 A128])) "pr106746.c":27:5 83 {*movsi_internal} (expr_list:REG_EQUIV (mem:SI (plus:DI (reg/f:DI 19 frame) (const_int -208 [0xffffffffffffff30])) [1 S4 A128]) (nil))) 4 bytes at sp + 120 can't alias with 16 bytes at sp - 104 or 16 bytes at sp - 88 when sp isn't modified in between and all 3 are in the same bb. So 1 returned from true_dependence looks incorrect (the -g case).