[Bug tree-optimization/111542] New: [11/12/13/14 Regression]

[Bug tree-optimization/111542] New: [11/12/13/14 Regression]

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111542

            Bug ID: 111542
           Summary: [11/12/13/14 Regression]
           Product: gcc
           Version: 14.0
            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:
```
_Bool f(int x,int y)
{
  _Bool a = x == 0;
  _Bool b = y == 0;
  _Bool t = a & b;
  return t & !a;
}

```

This used to optimize to 0 in GCC 6 but starting with GCC 7, we get:
```
  _5 = x_2(D) | y_3(D);
  _6 = _5 == 0;
  _1 = x_2(D) != 0;
  _7 = _1 & _6;
```

[Bug tree-optimization/111542] [11/12/13/14 Regression]

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111542

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|unassigned at gcc dot gnu.org      |pinskia at gcc dot gnu.org
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2023-09-22
             Status|UNCONFIRMED                 |ASSIGNED

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Mine.

[Bug tree-optimization/111542] [11/12/13/14 Regression] (a==0)&(b==0) into `(a|b) == 0` optimization sometimes gets in the way of other optimizations

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111542

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |14.0
            Summary|[11/12/13/14 Regression]    |[11/12/13/14 Regression]
                   |                            |(a==0)&(b==0) into `(a|b)
                   |                            |== 0` optimization
                   |                            |sometimes gets in the way
                   |                            |of other optimizations
      Known to fail|                            |7.1.0
      Known to work|                            |6.1.0

[Bug tree-optimization/111542] [11/12/13/14 Regression] (a==0)&(b==0) into `(a|b) == 0` optimization sometimes gets in the way of other optimizations

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111542

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Note IOR with != 0 has the same issue.
Full testcase:
```
_Bool f_and(int x,int y)
{
  _Bool a = x == 0;
  _Bool b = y == 0;
  _Bool t = a & b;
  return t & !a;
}


_Bool f_or(int x,int y)
{
  _Bool a = x != 0;
  _Bool b = y != 0;
  _Bool t = a | b;
  return t | !a;
}
```

Match pattern:
```
(for bitop (bit_ior bit_and)
     cmp1  (eq      ne     )
     cmp2  (ne      eq     )
 (simplify
  (bitop:c
   (cmp1 @1 integer_zerop)
   (cmp2 (bit_ior:c @0 @1) integer_zerop))
  { constant_boolean_node (bitop == BIT_IOR_EXPR, type); }))
```

[Bug tree-optimization/111542] [11/12/13/14 Regression] (a==0)&(b==0) into `(a|b) == 0` optimization sometimes gets in the way of other optimizations

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111542

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #2) 
> Match pattern:
> ```
> (for bitop (bit_ior bit_and)
>      cmp1  (eq      ne     )
>      cmp2  (ne      eq     )
>  (simplify
>   (bitop:c
>    (cmp1 @1 integer_zerop)
>    (cmp2 (bit_ior:c @0 @1) integer_zerop))
>   { constant_boolean_node (bitop == BIT_IOR_EXPR, type); }))
> ```

But this breaks generic-match building of the function:
tree
generic_simplify_92 (location_t ARG_UNUSED (loc), const tree ARG_UNUSED (type),
 tree ARG_UNUSED (_p0), tree ARG_UNUSED (_p1), tree *captures,
 const enum tree_code ARG_UNUSED (bitop),
 const enum tree_code ARG_UNUSED (cmp1),
 const enum tree_code ARG_UNUSED (cmp2))
{
  const bool debug_dump = dump_file && (dump_flags & TDF_FOLDING);
  if (TREE_SIDE_EFFECTS (_p0)) goto next_after_fail192;
  if (TREE_SIDE_EFFECTS (_p1)) goto next_after_fail192;
  if (UNLIKELY (!dbg_cnt (match))) goto next_after_fail192;
  {
    tree _r;
    _r =  constant_boolean_node (bitop == BIT_IOR_EXPR, type);
    if (UNLIKELY (debug_dump)) generic_dump_logs ("match.pd", 136, __FILE__,
__LINE__, true);
    return _r;
  }
next_after_fail192:;
  return NULL_TREE;
}


captures is not used .

I guess I could check TREE_SIDE_EFFECTS on @0/@1 to workaround that issue.

Like so:
```
(for bitop (bit_ior bit_and)
     cmp1  (eq      ne     )
     cmp2  (ne      eq     )
 (simplify
  (bitop:c
   (cmp1 @1 integer_zerop)
   (cmp2 (bit_ior:c @0 @1) integer_zerop))
  (if (GIMPLE || (!TREE_SIDE_EFFECTS (@0)
                   && !TREE_SIDE_EFFECTS (@1)))
   { constant_boolean_node (bitop == BIT_IOR_EXPR, type); })))
```

I had assumed genmatch would generate that code that would use @0/@1 if they
had side effects.  Maybe because I match @0 twice, it assumed we would be using
the captures (I have not looked though).

[Bug tree-optimization/111542] [11/12/13/14 Regression] (a==0)&(b==0) into `(a|b) == 0` optimization sometimes gets in the way of other optimizations

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111542

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2

[Bug tree-optimization/111542] [11/12/13/14/15 Regression] (a==0)&(b==0) into `(a|b) == 0` optimization sometimes gets in the way of other optimizations

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111542

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|14.0                        |14.2

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 14.1 is being released, retargeting bugs to GCC 14.2.