https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124141
--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #2)
> <bb 4> [local count: 143434726]:
> # .MEM_66 = VDEF <.MEM_65>
> _21 = sqlite3AtoF.part.0 (_20, &rValue, _19, enc_18);
> if (_21 <= 0)
> goto <bb 5>; [26.36%]
> else
> goto <bb 6>; [73.64%]
>
> <bb 5> [local count: 111700011]:
> # .MEM_60 = PHI <.MEM_65(3), .MEM_66(4)>
> # .MEM_75 = VDEF <.MEM_60>
> rValue ={v} {CLOBBER(eos)};
> goto <bb 18>; [100.00%]
>
>
> bb 6 never reaches a block which has a clobber for rValue.
that is there is no path from bb 5 to bb 6 at all. the only common block is bb
4.
> also bb 5 is not dominiating bb 6 at all, it looks like dangling pointer
> code is getting confused because of that.
