* [PATCH] Fix PR49078
@ 2011-05-24 15:28 Richard Guenther
0 siblings, 0 replies; only message in thread
From: Richard Guenther @ 2011-05-24 15:28 UTC (permalink / raw)
To: gcc-patches
This fixes PR49078 by restricting the assert in get_alias_set to
the case where it matters. We know that how types generated during
optimization have TYPE_CANONICAL computed differs from what the LTO
machinery does, the patch documents that and reverts a previous attempt
to fix parts of this inconsistency.
LTO profile-bootstrapped and tested on x86_64-unknown-linux-gnu,
SPEC 2k6 built, applied to trunk.
Richard.
2011-05-24 Richard Guenther <rguenther@suse.de>
PR bootstrap/49078
* gimple.c (gimple_register_canonical_type): Revert
previous change.
* alias.c (get_alias_set): Only assert that TYPE_CANONICAL
does not for a tree for the case where it matters. Cache
pointer-type alias-sets.
Index: gcc/gimple.c
===================================================================
--- gcc/gimple.c (revision 174106)
+++ gcc/gimple.c (working copy)
@@ -4779,7 +4802,12 @@ gimple_canonical_type_eq (const void *p1
/* Register type T in the global type table gimple_types.
If another type T', compatible with T, already existed in
gimple_types then return T', otherwise return T. This is used by
- LTO to merge identical types read from different TUs. */
+ LTO to merge identical types read from different TUs.
+
+ ??? This merging does not exactly match how the tree.c middle-end
+ functions will assign TYPE_CANONICAL when new types are created
+ during optimization (which at least happens for pointer and array
+ types). */
tree
gimple_register_canonical_type (tree t)
@@ -4800,25 +4828,6 @@ gimple_register_canonical_type (tree t)
if (TYPE_CANONICAL (t))
return TYPE_CANONICAL (t);
- /* For pointer and reference types do as the middle-end does - the
- canonical type is a pointer to the canonical pointed-to type. */
- if (TREE_CODE (t) == POINTER_TYPE)
- {
- TYPE_CANONICAL (t)
- = build_pointer_type_for_mode
- (gimple_register_canonical_type (TREE_TYPE (t)),
- TYPE_MODE (t), TYPE_REF_CAN_ALIAS_ALL (t));
- return TYPE_CANONICAL (t);
- }
- else if (TREE_CODE (t) == REFERENCE_TYPE)
- {
- TYPE_CANONICAL (t)
- = build_reference_type_for_mode
- (gimple_register_canonical_type (TREE_TYPE (t)),
- TYPE_MODE (t), TYPE_REF_CAN_ALIAS_ALL (t));
- return TYPE_CANONICAL (t);
- }
-
if (gimple_canonical_types == NULL)
gimple_canonical_types = htab_create_ggc (16381, gimple_canonical_type_hash,
gimple_canonical_type_eq, 0);
Index: gcc/alias.c
===================================================================
--- gcc/alias.c (revision 174106)
+++ gcc/alias.c (working copy)
@@ -709,10 +709,8 @@ get_alias_set (tree t)
t = TYPE_CANONICAL (t);
- /* Canonical types shouldn't form a tree nor should the canonical
- type require structural equality checks. */
- gcc_checking_assert (TYPE_CANONICAL (t) == t
- && !TYPE_STRUCTURAL_EQUALITY_P (t));
+ /* The canonical type should not require structural equality checks. */
+ gcc_checking_assert (!TYPE_STRUCTURAL_EQUALITY_P (t));
/* If this is a type with a known alias set, return it. */
if (TYPE_ALIAS_SET_KNOWN_P (t))
@@ -813,11 +811,19 @@ get_alias_set (tree t)
That's simple and avoids all the above problems. */
else if (POINTER_TYPE_P (t)
&& t != ptr_type_node)
- return get_alias_set (ptr_type_node);
+ set = get_alias_set (ptr_type_node);
/* Otherwise make a new alias set for this type. */
else
- set = new_alias_set ();
+ {
+ /* Each canonical type gets its own alias set, so canonical types
+ shouldn't form a tree. It doesn't really matter for types
+ we handle specially above, so only check it where it possibly
+ would result in a bogus alias set. */
+ gcc_checking_assert (TYPE_CANONICAL (t) == t);
+
+ set = new_alias_set ();
+ }
TYPE_ALIAS_SET (t) = set;
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2011-05-24 12:37 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2011-05-24 15:28 [PATCH] Fix PR49078 Richard Guenther
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).