[Bug tree-optimization/59779] [4.9 Regression] FAIL: gcc.dg/autopar/outer-1.c scan-tree-dump-times parloops "parallelizing outer loop"

["jakub at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59779

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
The following fixes this for me:
--- gcc/tree-dfa.c.jj    2014-01-03 11:40:57.000000000 +0100
+++ gcc/tree-dfa.c    2014-03-13 11:22:20.727848114 +0100
@@ -615,6 +615,15 @@ get_ref_base_and_extent (tree exp, HOST_
     maxsize = -1;

  done:
+  if (bitsize < 0)
+    {
+      *poffset = 0;
+      *psize = -1;
+      *pmax_size = -1;
+
+      return exp;
+    }
+
   if (!bit_offset.fits_shwi ())
     {
       *poffset = 0;
@@ -645,6 +654,9 @@ get_ref_base_and_extent (tree exp, HOST_
     maxsize = tree_to_uhwi (TYPE_SIZE (TREE_TYPE (exp))) - hbit_offset;
     }

+  if (maxsize < 0)
+    maxsize = -1;
+
   /* ???  Due to negative offsets in ARRAY_REF we can end up with
      negative bit_offset here.  We might want to store a zero offset
      in this case.  */

The problem is that the array is 0xbebc2000 bits long, and most of the places
in get_ref_base_and_extent work with maxsize (and bitsize) as uhwi, so we can
easily end up with negative, but not -1, maxsize (or bitsize), but callers
obviously expect sizes to be positive or -1 (special magic value for unknown).