[Bug middle-end/122004] x86 optimization: (x > INT_MAX - c) can utilize overflow flag to save code size

[law at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122004

--- Comment #2 from Jeffrey A. Law <law at gcc dot gnu.org> ---
ITSM that if something like a .SUB_OVERFLOW's result isn't used, then we could
optimize away the .SUB_OVERFLOW and just emit the comparison necessary to check
for overflow.

  _3 = .SUB_OVERFLOW (x_2(D), -127);
  _1 = IMAGPART_EXPR <_3>;
  _4 = (_Bool) _1;
  return _4;

So just _1 = -128 < x_2 and drop the assignment to _3 entirely. Or somehing
along those lines.   It's kind of like throwing away part of divmod.  So there
may also be something we can do over in RTL, though it's likely harder there.