[Bug tree-optimization/40052] New: missed optimizations on logical types: (x | 1) --> 1

[Bug tree-optimization/40052] New: missed optimizations on logical types: (x | 1) --> 1

[matz at gcc dot gnu dot org]

The testcase from PR40021:
 http://gcc.gnu.org/bugzilla/attachment.cgi?id=17800&action=view
shows a current problem in optimizing away logical operations on logical
(boolean) types, like "x | 1" --> "1".  This folding is done in some contexts,
and not in others.  In particular the testcase
 (with -m32 -O2 -ftree-vectorize -msse2 -mfpmath=sse -ffast-math )
leaves this until expand (it's generated quite early already, so there
are multiple possibilities to fold it):

  D.1650_109 = D.1648_107 | 1;
  if (D.1650_109 != 0)
    goto <bb 6>;
  else
    goto <bb 5>;

Richi already fiddled a bit with this, so CCing him.


-- 
           Summary: missed optimizations on logical types: (x | 1) --> 1
           Product: gcc
           Version: 4.5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: matz at gcc dot gnu dot org
  GCC host triplet: x86_64-unknown-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40052

[Bug tree-optimization/40052] missed optimizations on (extended) logical types: (x | 1) --> 1

[rguenth at gcc dot gnu dot org]

------- Comment #1 from rguenth at gcc dot gnu dot org  2009-05-07 08:58 -------
C testcase:

int foo(_Bool b)
{
  return b | 1;
}

int bar(_Bool b)
{
  return b & -2;
}

should both be folded to return a constant but instead are never optimized,
not even at RTL level.

Due to C semantics we get

  return (int) b | 1;

and

  return (int) b & -2;

here.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |enhancement
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2009-05-07 08:58:29
               date|                            |
            Summary|missed optimizations on     |missed optimizations on
                   |logical types: (x | 1) --> 1|(extended) logical types: (x
                   |                            || 1) --> 1


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40052

[Bug tree-optimization/40052] missed optimizations on (extended) logical types: (x | 1) --> 1

[rguenth at gcc dot gnu dot org]

------- Comment #2 from rguenth at gcc dot gnu dot org  2009-05-07 09:36 -------
Of course this only works if logicals are zero-extended.

We manage to fold

int foo(unsigned char b)
{
  return b | 0xff;
}

but not

int bar(unsigned char b)
{
  return b & (~0 << 8);
}

appearantly because the C FE for the former generates (int) (b | 255)
because of some premature optimization in build_binary_op.

We still do not fold

int foo(unsigned char b)
{
  return b | 0xfff;
}


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40052