[Bug tree-optimization/126925] ICE at -O3 on x86_64-linux-gnu: Segmentation fault during GIMPLE pass: ivopts

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126925

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |pinskia at gcc dot gnu.org
   Last reconfirmed|                            |2026-08-18
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org
           Keywords|                            |ice-on-valid-code,
                   |                            |missed-optimization
             Status|UNCONFIRMED                 |ASSIGNED
     Ever confirmed|0                           |1

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.  The issue is that we analyzed cand->var_after (f_44) as being
invariant:

# f_42 = PHI <f_44(21), 6(20)>
h_24 = (short int) d_15(D);
_25 = (long unsigned int) d_15(D);
_27 = (long unsigned int) h_24;
_36 = _25 + 18446744073709551612;
_21 = _36 - _27;
_22 = _27 - _36;
_7 = f_42 + _22;
f_44 = _7 + _21;

here f_44 simplifies to f_42 meaning the f_42 PHI simplifies to '6'.

But that's not expected, the IV candidate is analyzed as

 {6, +, ((long unsigned int) d_15(D) - (long unsigned int) (short int) d_15(D))
+ ((long unsigned int) (short int) d_15(D) - (long unsigned int) d_15(D)) }

which failed to simplify.

I'll fix the ICE.