[Bug tree-optimization/63445] [5 Regression] request: make -Wstrict-overflow avoid a class of false positives

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63445

--- Comment #6 from Eric Botcazou <ebotcazou at gcc dot gnu.org> ---
> Found new range for j_9: [i_15 + 1, +INF]
> 
> Visiting statement:
> _6 = j_9 - i_15;
> Found new range for _6: [1, +INF(OVF)]
> 
> i_15 could be negative and thus j_9 - i_15 could well overflow the input
> range at the +INF side.  (i_15 is [-INF, j_5(D) + -1])

Yes, the range is as expected here but...

> I believe we are just better in propagating the ranges here and 4.9
> computed _6 as VARYING (and thus not applying the optimization).

...we actually don't apply any optimization to the code:

Value ranges after VRP:

n_1: [1, +INF]  EQUIVALENCES: { n_7 c_8 n_16 } (3 elements)
_2: [1, +INF]  EQUIVALENCES: { n_7 n_16 } (2 elements)
pretmp_3: ~[0, 0]
i_4(D): VARYING
j_5(D): VARYING
_6: [1, +INF(OVF)]
n_7: [0, +INF]

[...]

  <bb 2>:
  if (i_4(D) < j_5(D))
    goto <bb 3>;
  else
    goto <bb 7>;

  <bb 3>:
  _6 = j_5(D) - i_4(D);
  n_7 = (unsigned int) _6;
  if (_6 != 0)
    goto <bb 4>;
  else
    goto <bb 7>;

The range of n_7 is suboptimal and the test on _6 can be eliminated.