[Bug tree-optimization/121006] New: if (float_var == CST || float_var== -CST) is known to be true, __builtin_copysignf(CST, float_var) can be optimized to just float_var

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121006

            Bug ID: 121006
           Summary: if (float_var == CST || float_var== -CST) is known to
                    be true,  __builtin_copysignf(CST, float_var) can be
                    optimized to just float_var
           Product: gcc
           Version: 16.0
            Status: UNCONFIRMED
          Keywords: 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:
```

float f1(float a)
{
  if (a == 1.0 || a == -1.0)
      return __builtin_copysignf(1.0, a);
  return a;
}

float f0(float a)
{
  if (a == 0.0 || a == -0.0)
      return __builtin_copysignf(0.0, a);
  return a;
}


float fpi(float a)
{
  float x = 3.141592;
  if (a == x || a == -x)
      return __builtin_copysignf(x, a);
  return a;
}

```


These all should be optimized to just `return a`. LLVM is able to do each one
except for f0 (but I think that is because it is pattern matching here).

With frange we have for fpi:
```
a_4(D)  [frange] float [-3.1415920257568359375e+0 (-0x0.c90fd8p+2),
3.1415920257568359375e+0 (0x0.c90fd8p+2)]
    <bb 4> :
    _6 = __builtin_copysignf (3.1415920257568359375e+0, a_4(D));
    // predicted unlikely by early return (on trees) predictor.
```