* Drop ref2_is_decl from aliasing_component_refs_p
@ 2019-06-20 14:16 Jan Hubicka
2019-06-21 12:15 ` Richard Biener
0 siblings, 1 reply; 2+ messages in thread
From: Jan Hubicka @ 2019-06-20 14:16 UTC (permalink / raw)
To: gcc-patches
Hi,
as discussed earlier this patch drops ref2_is_decl from
aliasing_component_refs. This parameter makes it to assume that all
access path may not continue by reference to DECL which is not true
because in gimple memory model we allow to store value of completely
unrealted type to it. I have also constructed an (invalid C IMO)
testcase (which fails on GCC for ages, but works for ICC)
Bootstrapped/regtested x86_64-linux, will commit it shortly.
* tree-ssa-alias.c (aliasing_component_refs_p): Remove ref2_is_decl
parameter; it has no use in gimple memory model.
(indirect_ref_may_alias_decl_p): Update.
* gcc.c-torture/execute/alias-access-path-1.c: New testcase.
Index: tree-ssa-alias.c
===================================================================
--- tree-ssa-alias.c (revision 272507)
+++ tree-ssa-alias.c (working copy)
@@ -850,8 +850,7 @@ type_has_components_p (tree type)
/* Determine if the two component references REF1 and REF2 which are
based on access types TYPE1 and TYPE2 and of which at least one is based
- on an indirect reference may alias. REF2 is the only one that can
- be a decl in which case REF2_IS_DECL is true.
+ on an indirect reference may alias.
REF1_ALIAS_SET, BASE1_ALIAS_SET, REF2_ALIAS_SET and BASE2_ALIAS_SET
are the respective alias sets. */
@@ -863,8 +862,7 @@ aliasing_component_refs_p (tree ref1,
tree ref2,
alias_set_type ref2_alias_set,
alias_set_type base2_alias_set,
- poly_int64 offset2, poly_int64 max_size2,
- bool ref2_is_decl)
+ poly_int64 offset2, poly_int64 max_size2)
{
/* If one reference is a component references through pointers try to find a
common base and apply offset based disambiguation. This handles
@@ -982,7 +980,7 @@ aliasing_component_refs_p (tree ref1,
if (TREE_CODE (TREE_TYPE (base1)) == ARRAY_TYPE
&& (!TYPE_SIZE (TREE_TYPE (base1))
|| TREE_CODE (TYPE_SIZE (TREE_TYPE (base1))) != INTEGER_CST
- || (ref == base2 && !ref2_is_decl)))
+ || ref == base2))
{
++alias_stats.aliasing_component_refs_p_may_alias;
return true;
@@ -1041,7 +1039,7 @@ aliasing_component_refs_p (tree ref1,
if (TREE_CODE (TREE_TYPE (base2)) == ARRAY_TYPE
&& (!TYPE_SIZE (TREE_TYPE (base2))
|| TREE_CODE (TYPE_SIZE (TREE_TYPE (base2))) != INTEGER_CST
- || (ref == base1 && !ref2_is_decl)))
+ || ref == base1))
{
++alias_stats.aliasing_component_refs_p_may_alias;
return true;
@@ -1089,8 +1087,7 @@ aliasing_component_refs_p (tree ref1,
return true;
}
/* If this is ptr vs. decl then we know there is no ptr ... decl path. */
- if (!ref2_is_decl
- && compare_type_sizes (TREE_TYPE (ref1), type2) >= 0
+ if (compare_type_sizes (TREE_TYPE (ref1), type2) >= 0
&& (!end_struct_ref2
|| compare_type_sizes (TREE_TYPE (ref1),
TREE_TYPE (end_struct_ref2)) >= 0)
@@ -1554,13 +1551,7 @@ indirect_ref_may_alias_decl_p (tree ref1
offset1, max_size1,
ref2,
ref2_alias_set, base2_alias_set,
- offset2, max_size2,
- /* Only if the other reference is actual
- decl we can safely check only toplevel
- part of access path 1. */
- same_type_for_tbaa (TREE_TYPE (dbase2),
- TREE_TYPE (base2))
- == 1);
+ offset2, max_size2);
return true;
}
@@ -1675,7 +1666,7 @@ indirect_refs_may_alias_p (tree ref1 ATT
offset1, max_size1,
ref2,
ref2_alias_set, base2_alias_set,
- offset2, max_size2, false);
+ offset2, max_size2);
return true;
}
Index: testsuite/gcc.c-torture/execute/alias-access-path-1.c
===================================================================
--- testsuite/gcc.c-torture/execute/alias-access-path-1.c (nonexistent)
+++ testsuite/gcc.c-torture/execute/alias-access-path-1.c (working copy)
@@ -0,0 +1,19 @@
+/* Test that variable
+ int val;
+ may hold value of tyope "struct c" which has same size.
+ This is valid in GIMPLE memory model. */
+
+struct a {int val;} a={1},a2;
+struct b {struct a a;};
+int val;
+struct c {struct b b;} *cptr=(void *)&val;
+
+int
+main(void)
+{
+ cptr->b.a=a;
+ val = 2;
+ a2=cptr->b.a;
+ if (a2.val == a.val)
+ __builtin_abort ();
+}
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: Drop ref2_is_decl from aliasing_component_refs_p
2019-06-20 14:16 Drop ref2_is_decl from aliasing_component_refs_p Jan Hubicka
@ 2019-06-21 12:15 ` Richard Biener
0 siblings, 0 replies; 2+ messages in thread
From: Richard Biener @ 2019-06-21 12:15 UTC (permalink / raw)
To: Jan Hubicka; +Cc: GCC Patches
On Thu, Jun 20, 2019 at 4:16 PM Jan Hubicka <hubicka@ucw.cz> wrote:
>
> Hi,
> as discussed earlier this patch drops ref2_is_decl from
> aliasing_component_refs. This parameter makes it to assume that all
> access path may not continue by reference to DECL which is not true
> because in gimple memory model we allow to store value of completely
> unrealted type to it. I have also constructed an (invalid C IMO)
> testcase (which fails on GCC for ages, but works for ICC)
>
> Bootstrapped/regtested x86_64-linux, will commit it shortly.
OK.
Richard.
> * tree-ssa-alias.c (aliasing_component_refs_p): Remove ref2_is_decl
> parameter; it has no use in gimple memory model.
> (indirect_ref_may_alias_decl_p): Update.
> * gcc.c-torture/execute/alias-access-path-1.c: New testcase.
> Index: tree-ssa-alias.c
> ===================================================================
> --- tree-ssa-alias.c (revision 272507)
> +++ tree-ssa-alias.c (working copy)
> @@ -850,8 +850,7 @@ type_has_components_p (tree type)
>
> /* Determine if the two component references REF1 and REF2 which are
> based on access types TYPE1 and TYPE2 and of which at least one is based
> - on an indirect reference may alias. REF2 is the only one that can
> - be a decl in which case REF2_IS_DECL is true.
> + on an indirect reference may alias.
> REF1_ALIAS_SET, BASE1_ALIAS_SET, REF2_ALIAS_SET and BASE2_ALIAS_SET
> are the respective alias sets. */
>
> @@ -863,8 +862,7 @@ aliasing_component_refs_p (tree ref1,
> tree ref2,
> alias_set_type ref2_alias_set,
> alias_set_type base2_alias_set,
> - poly_int64 offset2, poly_int64 max_size2,
> - bool ref2_is_decl)
> + poly_int64 offset2, poly_int64 max_size2)
> {
> /* If one reference is a component references through pointers try to find a
> common base and apply offset based disambiguation. This handles
> @@ -982,7 +980,7 @@ aliasing_component_refs_p (tree ref1,
> if (TREE_CODE (TREE_TYPE (base1)) == ARRAY_TYPE
> && (!TYPE_SIZE (TREE_TYPE (base1))
> || TREE_CODE (TYPE_SIZE (TREE_TYPE (base1))) != INTEGER_CST
> - || (ref == base2 && !ref2_is_decl)))
> + || ref == base2))
> {
> ++alias_stats.aliasing_component_refs_p_may_alias;
> return true;
> @@ -1041,7 +1039,7 @@ aliasing_component_refs_p (tree ref1,
> if (TREE_CODE (TREE_TYPE (base2)) == ARRAY_TYPE
> && (!TYPE_SIZE (TREE_TYPE (base2))
> || TREE_CODE (TYPE_SIZE (TREE_TYPE (base2))) != INTEGER_CST
> - || (ref == base1 && !ref2_is_decl)))
> + || ref == base1))
> {
> ++alias_stats.aliasing_component_refs_p_may_alias;
> return true;
> @@ -1089,8 +1087,7 @@ aliasing_component_refs_p (tree ref1,
> return true;
> }
> /* If this is ptr vs. decl then we know there is no ptr ... decl path. */
> - if (!ref2_is_decl
> - && compare_type_sizes (TREE_TYPE (ref1), type2) >= 0
> + if (compare_type_sizes (TREE_TYPE (ref1), type2) >= 0
> && (!end_struct_ref2
> || compare_type_sizes (TREE_TYPE (ref1),
> TREE_TYPE (end_struct_ref2)) >= 0)
> @@ -1554,13 +1551,7 @@ indirect_ref_may_alias_decl_p (tree ref1
> offset1, max_size1,
> ref2,
> ref2_alias_set, base2_alias_set,
> - offset2, max_size2,
> - /* Only if the other reference is actual
> - decl we can safely check only toplevel
> - part of access path 1. */
> - same_type_for_tbaa (TREE_TYPE (dbase2),
> - TREE_TYPE (base2))
> - == 1);
> + offset2, max_size2);
>
> return true;
> }
> @@ -1675,7 +1666,7 @@ indirect_refs_may_alias_p (tree ref1 ATT
> offset1, max_size1,
> ref2,
> ref2_alias_set, base2_alias_set,
> - offset2, max_size2, false);
> + offset2, max_size2);
>
> return true;
> }
> Index: testsuite/gcc.c-torture/execute/alias-access-path-1.c
> ===================================================================
> --- testsuite/gcc.c-torture/execute/alias-access-path-1.c (nonexistent)
> +++ testsuite/gcc.c-torture/execute/alias-access-path-1.c (working copy)
> @@ -0,0 +1,19 @@
> +/* Test that variable
> + int val;
> + may hold value of tyope "struct c" which has same size.
> + This is valid in GIMPLE memory model. */
> +
> +struct a {int val;} a={1},a2;
> +struct b {struct a a;};
> +int val;
> +struct c {struct b b;} *cptr=(void *)&val;
> +
> +int
> +main(void)
> +{
> + cptr->b.a=a;
> + val = 2;
> + a2=cptr->b.a;
> + if (a2.val == a.val)
> + __builtin_abort ();
> +}
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2019-06-21 12:15 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2019-06-20 14:16 Drop ref2_is_decl from aliasing_component_refs_p Jan Hubicka
2019-06-21 12:15 ` Richard Biener
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).