[Bug tree-optimization/106025] [13 Regression] Incorrect optimization at -O2 leads to infinite test execution time since r13-469-g9a53101caadae1b5

[rguenth at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106025

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
It looks like the pattern just exposes an existing issue. 
-fdbg-cnt=match:0-104 breaks while -fdbg-cnt=match:0-103 is OK but the MAX
generation for this looks correct.

Disabling VRP fixes it but avoids the MAX generation as we end up with

 if (_22 < _33)
    goto <bb 8>; [50.00%]
  else
    goto <bb 7>; [50.00%]

  <bb 7> [local count: 536870913]:
  _36 = (short int) _22;
  # RANGE [-32768, 32767]
  _38 = (int) _36;

  <bb 8> [local count: 1073741824]:
  # RANGE [-32768, 32767]
  # prephitmp_39 = PHI <_1(6), _38(7)>

_22 is indeed small.  We recognize ushortA < ushortB ? (int)ushortB :
(int)ushortA as (int)MAX (ushortA, ushortB)

Not sure what goes wrong here.