[patch, 4.1.2, 4.2.1, 4.3] Fix a bug in pointer dependency test

[Ira Rosen]

As pointed out by Daniel, there is a bug in pointers' dependency test in
data-refs analysis. Currently, only tags are compared, while two different
tags  can have common symbols in their may-alias sets. Therefore, in the
case of different tags,  may-alias sets must be examined too.

This patch fixes this problem using pointer-sets, as suggested by Diego.
(We can't compare vops, since we don't have the original statements at this
stage).

Bootstrapped and tested on ppc-linux.

O.K. for mainline (and 4.1.2 and 4.2.1) ?

Thanks,
Ira


ChangeLog entry:

      * tree-ssa-alias.c (may_aliases_intersect): New function.
      * tree-data-ref.c (ptr_ptr_may_alias_p): Call may_aliases_intersect
for different tags.
      * tree-flow.h (may_aliases_intersect): Add function declaration.

The patch:

Index: tree-ssa-alias.c
===================================================================
--- tree-ssa-alias.c    (revision 120765)
+++ tree-ssa-alias.c    (working copy)
@@ -2609,6 +2609,35 @@ is_aliased_with (tree tag, tree sym)
   return false;
 }

+/* Given two tags return TRUE if their may-alias sets intersect.  */
+
+bool
+may_aliases_intersect (tree tag1, tree tag2)
+{
+  struct pointer_set_t *set1 = pointer_set_create ();
+  unsigned i;
+  VEC(tree,gc) *may_aliases1 = may_aliases (tag1);
+  VEC(tree,gc) *may_aliases2 = may_aliases (tag2);
+  tree sym;
+
+  /* Insert all the symbols from the first may-alias set into the
+     pointer-set.  */
+  for (i = 0; VEC_iterate (tree, may_aliases1, i, sym); i++)
+    pointer_set_insert (set1, sym);
+
+  /* Go through the second may-alias set and check if it contains symbols
that
+     are common with the first set.  */
+  for (i = 0; VEC_iterate (tree, may_aliases2, i, sym); i++)
+    if (pointer_set_contains (set1, sym))
+      {
+       pointer_set_destroy (set1);
+       return true;
+      }
+
+  pointer_set_destroy (set1);
+  return false;
+}
+
 /* The following is based on code in add_stmt_operand to ensure that the
    same defs/uses/vdefs/vuses will be found after replacing a reference
    to var (or ARRAY_REF to var) with an INDIRECT_REF to ptr whose value
Index: tree-data-ref.c
===================================================================
--- tree-data-ref.c     (revision 120765)
+++ tree-data-ref.c     (working copy)
@@ -192,7 +192,12 @@ ptr_ptr_may_alias_p (tree ptr_a, tree pt
       if (!tag_b)
        return false;
     }
-  *aliased = (tag_a == tag_b);
+
+  if (tag_a == tag_b)
+    *aliased = true;
+  else
+    *aliased = may_aliases_intersect (tag_a, tag_b);
+
   return true;
 }

Index: tree-flow.h
===================================================================
--- tree-flow.h (revision 120765)
+++ tree-flow.h (working copy)
@@ -729,6 +729,7 @@ extern void dump_points_to_info_for (FIL
 extern void debug_points_to_info_for (tree);
 extern bool may_be_aliased (tree);
 extern bool is_aliased_with (tree, tree);
+extern bool may_aliases_intersect (tree, tree);
 extern struct ptr_info_def *get_ptr_info (tree);
 extern void new_type_alias (tree, tree, tree);
 extern void count_uses_and_derefs (tree, tree, unsigned *, unsigned *,
bool *);


New test case:

(See attached file: vect-106.c.txt)

::ADDPATCH SSA:
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: vect-106.c.txt
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20070115/30dbda61/attachment.txt>