[Bug tree-optimization/66856] [6 Regression] ICE in compute_live_loop_exits, at tree-ssa-loop-manip.c:234

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66856

--- Comment #10 from Richard Biener <rguenth at gcc dot gnu.org> ---
Ok.  So this is caused by operand swapping and a shared stmt (I knew we'd
eventually hit such case ... :/)

t.c:15:11: note: node
t.c:15:11: note:        stmt 0 _30 = _29 & 1;
t.c:15:11: note:        stmt 1 a_15 = g_14(D) & _21;

This has bogus ordering - the g_14(D) should be 2nd place to match up with the
1

t.c:15:11: note: node
t.c:15:11: note:        stmt 0 _29 = _13 >> 5;
t.c:15:11: note:        stmt 1 _21 = _20 >> 3;
t.c:15:11: note: node
t.c:15:11: note:        stmt 0 _13 = *f_39;
t.c:15:11: note:        stmt 1 _20 = MEM[(int *)f_39 + 4B];
t.c:15:11: note: node
t.c:15:11: note:        stmt 0 _27 = a_26 >> 5;
t.c:15:11: note:        stmt 1 _12 = a_15 >> 5;
t.c:15:11: note: node
t.c:15:11: note:        stmt 0 a_26 = g_14(D) & _25;
t.c:15:11: note:        stmt 1 a_15 = g_14(D) & _21;

Because we swapped the stmt when seeing it used here (dumping missing for the
case in vect_get_and_check_slp_defs).