[Bug tree-optimization/109638] New: unsigned > 1 ? 0 : n is not optimized to n == 1

[Bug tree-optimization/109638] New: unsigned > 1 ? 0 : n is not optimized to n == 1

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109638

            Bug ID: 109638
           Summary: unsigned > 1 ? 0 : n is not optimized to n == 1
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: enhancement
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: pinskia at gcc dot gnu.org
  Target Milestone: ---

Take:
```
unsigned test_cst (unsigned n)
{
  if (n > 1)
    n = 0;
  return n;
}
unsigned test_cst1 (unsigned n)
{
  return (n == 1);
}
```
For gimple at least the second one is smaller. I suspect for x86 the second
version is also faster because it uses sete rather than cmov.
But these 2 functions should produce the same code.

I found this on accident while looking at gcc.dg/tree-ssa/builtin-snprintf-2.c 
testcase.

This can only be done for 1 too.

[Bug tree-optimization/109638] unsigned > 1 ? 0 : n is not optimized to n == 1

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109638

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
This should do it, I think:

(simplify
 (cond (lt @1 integer_onep@2) integer_zerop @1)
 (if (TYPE_UNSIGNED (type))
  (convert (eq @1 @2))))

[Bug tree-optimization/109638] unsigned > 1 ? 0 : n is not optimized to n == 1

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109638

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Hmm, but more generally

 if (n != x)
   n = 0;
 return n != 0;

could be handled as well.  Or

 if (n != x)
   n = 0;
 return n;

->

 return (n == x) * x;


-1 might be another special case that CPUs can efficiently handle.

[Bug tree-optimization/109638] unsigned > 1 ? 0 : n is not optimized to n == 1

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109638

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |FIXED
   Target Milestone|---                         |14.0
             Status|UNCONFIRMED                 |RESOLVED

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
This got fixed by r14-3036-gb57bd27cb68fdb .