[PATCH] Don't require TREE_OVERFLOW in int_fits_type_p

[Roger Sayle <roger at eyesopen dot com>]

The following patch tweaks int_fits_type_p to avoid testing the
result of force_fit_type for TREE_OVERFLOW, but instead to compare
the result's value with the original constant value.  This is the
same logic as currently used in force_fits_type to determine whether
to set TREE_OVERFLOW, so there's currently no change in functionality.
However, this moves the middle-end away from using/relying on the
TREE_OVERFLOW and TREE_CONSTANT_OVERFLOW flags, which are the source
of several PRs (including 4.0 regressions).

Decoupling this functionality, should prevent changes to when/if we
set TREE_OVERFLOW in force_fit_type from affecting int_fits_type_p.


The following patch has been tested on i686-pc-linux-gnu with a
full "make bootstrap", all default languages, and regression tested
with a top-level "make -k check" with no new failures.

Ok for mainline?



2004-12-29  Roger Sayle  <roger@eyesopen.com>

	* tree.c (int_fits_type_p): Compare the result of force_fit_type
	with the original constant rather than require TREE_OVERFLOW.


Index: tree.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/tree.c,v
retrieving revision 1.457
diff -c -3 -p -r1.457 tree.c
*** tree.c	17 Dec 2004 08:17:01 -0000	1.457
--- tree.c	29 Dec 2004 21:13:18 -0000
*************** int_fits_type_p (tree c, tree type)
*** 4914,4923 ****
    /* Or to force_fit_type, if nothing else.  */
    else
      {
!       c = copy_node (c);
!       TREE_TYPE (c) = type;
!       c = force_fit_type (c, -1, false, false);
!       return !TREE_OVERFLOW (c);
      }
  }

--- 4914,4924 ----
    /* Or to force_fit_type, if nothing else.  */
    else
      {
!       tree n = copy_node (c);
!       TREE_TYPE (n) = type;
!       n = force_fit_type (n, -1, false, false);
!       return TREE_INT_CST_HIGH (n) == TREE_INT_CST_HIGH (c)
!              && TREE_INT_CST_LOW (n) == TREE_INT_CST_LOW (c);
      }
  }


Roger
--