[Bug middle-end/52578] New: Fails to fold another size difference

["rguenth at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52578

             Bug #: 52578
           Summary: Fails to fold another size difference
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: rguenth@gcc.gnu.org


gnat.dg/opt9.adb will fail with the bitfield handling patch with -m32 because
we do not fold

long bar (long i)
{
  return (long)((unsigned long)i + 2) - (long)i;
}
long foo (int i)
{
  return (long)((unsigned long)i + 2) - (long)i;
}

to constants.  the associate: path splits the above to
var0 == (long unsigned int) i, lit0 == 2, var1 == -i

which fails to associate because of


          /* With undefined overflow we can only associate constants with one
             variable, and constants whose association doesn't overflow.  */
          if ((POINTER_TYPE_P (type) && POINTER_TYPE_OVERFLOW_UNDEFINED)
              || (INTEGRAL_TYPE_P (type) && !TYPE_OVERFLOW_WRAPS (type)))
            {
              if (var0 && var1)
                {
                  tree tmp0 = var0;
                  tree tmp1 = var1;

                  if (TREE_CODE (tmp0) == NEGATE_EXPR)
                    tmp0 = TREE_OPERAND (tmp0, 0);
                  if (TREE_CODE (tmp1) == NEGATE_EXPR)
                    tmp1 = TREE_OPERAND (tmp1, 0);
                  /* The only case we can still associate with two variables
                     is if they are the same, modulo negation.  */
                  if (!operand_equal_p (tmp0, tmp1, 0))
                    ok = false;

but we can as well strip widening and sign-changing conversions I think.