------- Comment #5 from ubizjak at gmail dot com 2010-03-26 11:22 -------
(In reply to comment #4)
> > There is a mismatch in a guard expression. Following patch fixes ICE for me
> > (I'm not sure if it makes any sense, though):
>
> No, it isn't correct, because:
>
> /* Return 1 if T is an INTEGER_CST that can be manipulated efficiently on
> the host. If POS is zero, the value can be represented in a single
> HOST_WIDE_INT. If POS is nonzero, the value must be non-negative and can
> be represented in a single unsigned HOST_WIDE_INT. */
>
> int
> host_integerp (const_tree t, int pos)
>
>
> If you pass 1 to host_integerp/tree_low_cst, you must use unsigned HWI.
If this is indeed required for tree_low_cst (which is defined as signed HWI!),
then the patch that reverses the semantics of my previous patch works as well.
Index: stor-layout.c
===================================================================
--- stor-layout.c (revision 157742)
+++ stor-layout.c (working copy)
@@ -1349,9 +1349,9 @@ place_field (record_layout_info rli, tre
&& host_integerp (TYPE_SIZE (TREE_TYPE (field)), 0)
&& host_integerp (DECL_SIZE (field), 0))
{
- HOST_WIDE_INT bitsize = tree_low_cst (DECL_SIZE (field), 1);
+ HOST_WIDE_INT bitsize = tree_low_cst (DECL_SIZE (field), 0);
HOST_WIDE_INT typesize
- = tree_low_cst (TYPE_SIZE (TREE_TYPE (field)), 1);
+ = tree_low_cst (TYPE_SIZE (TREE_TYPE (field)), 0);
if (typesize < bitsize)
rli->remaining_in_alignment = 0;
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43528
