[PATCH, middle-end] Ping: Switch initializations conversion
Martin Jambor
mjambor@suse.cz
Sun Jul 13 22:07:00 GMT 2008
Hi,
On Sun, Jul 13, 2008 at 10:38:41PM +0200, Richard Guenther wrote:
> > I believe the cleanest solution is indeed to generate the arithmetics
> > in a safe super-type. Based on comments in tree.def I have come up
> > with the following four line patch which prevents the assert from
> > being hit. I was considering replacing the if statement with a while
> > cycle but it does not seem to make any difference.
> >
> > I will bootstrap and test it soon. Is it OK if it passes?
>
> I think it's better to make sure to use the base-type already here:
>
> tree utype = unsigned_type_for (TREE_TYPE (info.index_expr));
>
> Ok with using the base-type of TREE_TYPE (info.index_expr) to generate
> utype.
>
You are right the order of TREE_TYPE and unsigned_type_for should be
the opposite. Therefore I will test this patch instead. Is it OK?
Thanks.
Martin
2008-07-13 Martin Jambor <mjambor@suse.cz>
* tree-switch-conversion.c (gen_inbound_check): Make sure the type
in which we generate arithmetics is not a subrange.
Index: gcc/tree-switch-conversion.c
===================================================================
--- gcc/tree-switch-conversion.c (revision 137717)
+++ gcc/tree-switch-conversion.c (working copy)
@@ -622,7 +622,7 @@
tree label_decl3 = create_artificial_label ();
tree label1, label2, label3;
- tree utype = unsigned_type_for (TREE_TYPE (info.index_expr));
+ tree utype;
tree tmp_u;
tree cast, cast_assign;
tree ulb, minus, minus_assign;
@@ -638,6 +638,12 @@
gcc_assert (info.default_values);
bb0 = bb_for_stmt (swtch);
+ /* Make sure we do not generate arithmetics in a subrange. */
+ if (TREE_TYPE (TREE_TYPE (info.index_expr)))
+ utype = unsigned_type_for (TREE_TYPE (TREE_TYPE (info.index_expr)));
+ else
+ utype = unsigned_type_for (TREE_TYPE (info.index_expr));
+
/* (end of) block 0 */
bsi = bsi_for_stmt (info.arr_ref_first);
tmp_u = make_rename_temp (utype, "csui");
More information about the Gcc-patches
mailing list