[Bug sanitizer/108481] [13 Regression] UBsan missed a signed integer overflow

["pinskia at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108481

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #2)
>       i = i - 6822162149299574294;
> 
> Is not being invoked on the executable code.
> 
> If we look at look at the original code:
> 
> 
>     if ((i * (unsigned long)7 <= 1) << j)
>       ;
>     else {
>       i = i - 6822162149299574294;
>       if (j) {
>         if (*g)
>           break;
>         continue;
>       }
>       return 8;
>     }
> ...
>   return h;
> 
> The only path where the undefined behavior even matters is inside the path
> that had continue in it. Which is where the subtraction is pushed to now.
> 
> I don't know if we should declear this as a valid thing to do or not.

This is what the IR looks in the end:
  _27 = _17 << _20;
  if (_27 != 0)
    goto <bb 12>; [50.00%]
  else
    goto <bb 9>; [50.00%]

  <bb 9> [local count: 1560331071]:
  if (_20 != 0)
    goto <bb 10>; [94.50%]
  else
    goto <bb 13>; [5.50%]

  <bb 10> [local count: 1474512859]:
  _30 = a;
  if (_30 != 0)
    goto <bb 13>; [67.00%]
  else
    goto <bb 11>; [33.00%]

  <bb 11> [local count: 486589241]:
  i_28 = .UBSAN_CHECK_SUB (i_13, 6822162149299574294);

  <bb 12> [local count: 2046920311]:
  # i_34 = PHI <i_13(8), i_28(11)>
  (Loop back)
...

  <bb 13> [local count: 1073741830]:
  return 0;

As you can see the only path where the ubsan_check_sub is executed is right
before the continue which is seems like a very valid thing to do.