[Bug tree-optimization/125749] New: `MAX == 0` -> (A|B) == 0 should happen for signed types if a and b are nonnegative

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125749

            Bug ID: 125749
           Summary: `MAX<a,b> == 0` -> (A|B) == 0 should happen for signed
                    types if a and b are nonnegative
           Product: gcc
           Version: 16.0
            Status: UNCONFIRMED
          Keywords: easyhack, missed-optimization
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: pinskia at gcc dot gnu.org
  Target Milestone: ---

Take:
```
int
f (unsigned short a, unsigned short b) {
  int aa = a;
  int bb = b;
  int t = aa > bb ? aa : bb;
  return t == 0;
}

unsigned
g (unsigned short a, unsigned short b) {
  unsigned aa = a;
  unsigned bb = b;
  unsigned t = aa > bb ? aa : bb;
  return t == 0;
}
```

These 2 functions should have the same code produce but currently f is not
optimized because the pattern:
```
(for cmp (eq ne)
 (simplify
  (cmp (max @0 @1) integer_zerop)
  (if (TYPE_UNSIGNED (TREE_TYPE (@0)))
   (with { tree utype = unsigned_type_for (TREE_TYPE (@0)); }
    (cmp (bit_ior (convert:utype @0) (convert:utype @1))
         { build_zero_cst (utype); } )))))
```

Only checks TYPE_UNSIGNED  rather than if @0 and @1 are both nonnegative.