Re: [PATCH, 4.5?] Fix get_ref_base_and_extent() handling of unions and variable sized arrays

[Jan Hubicka <hubicka at ucw dot cz>]

> Hi,
> 
> On Fri, Dec 12, 2008 at 07:17:02PM +0100, Martin Jambor wrote:
> > Hi,
> > 
> > this  patch  fixes bug  in  get_ref_base_and_extent()  which makes  it
> > return wrong  max_size when dealing  with variable size  arrays within
> > unions.
> > 
> > Please see discussion in the gcc mailing list, particularly the
> > following message for details:
> > 
> > http://gcc.gnu.org/ml/gcc/2008-12/msg00192.html
> > 
> > Even though this is technically a  bug in trunk, I'm happy to postpone
> > this until 4.5.   It may be part of the  pretty-ipa branch sooner than
> > stage1.
> > 
> > Bootstrapped and tested on x86_64.
> > 
> 
> I have now also bootstrapped  and tested it against pretty-ipa branch.
> It is not ipa related per se but my current development on that branch
> requires the fix.
> 
> Honza, can  I commit  it to the  branch?  (Alternatively, I  can still
> keep it separate  in my quilt patch  set until Richi has a  look at it
> but I do not want to intermingle it with the rest of the work).

In case I didn't OKed it already, it is OK :)

Honza
> 
> Thanks,
> 
> Martin
> 
> > 
> > 2008-12-12  Martin Jambor  <mjambor@suse.cz>
> > 
> > 	* tree-dfa.c (get_ref_base_and_extent): Return -1 maxsize if
> > 	seen_variable_array_ref while also traversing a union.
> > 
> > Index: isra/gcc/tree-dfa.c
> > ===================================================================
> > --- isra.orig/gcc/tree-dfa.c
> > +++ isra/gcc/tree-dfa.c
> > @@ -797,6 +797,7 @@ get_ref_base_and_extent (tree exp, HOST_
> >    tree size_tree = NULL_TREE;
> >    HOST_WIDE_INT bit_offset = 0;
> >    bool seen_variable_array_ref = false;
> > +  bool seen_union = false;
> >  
> >    gcc_assert (!SSA_VAR_P (exp));
> >  
> > @@ -840,6 +841,9 @@ get_ref_base_and_extent (tree exp, HOST_
> >  	    tree field = TREE_OPERAND (exp, 1);
> >  	    tree this_offset = component_ref_field_offset (exp);
> >  
> > +	    if (TREE_CODE (TREE_TYPE (TREE_OPERAND (exp, 0))) == UNION_TYPE)
> > +	      seen_union = true;
> > +
> >  	    if (this_offset && TREE_CODE (this_offset) == INTEGER_CST)
> >  	      {
> >  		HOST_WIDE_INT hthis_offset = tree_low_cst (this_offset, 0);
> > @@ -930,12 +934,16 @@ get_ref_base_and_extent (tree exp, HOST_
> >       where we do not know maxsize for variable index accesses to
> >       the array.  The simplest way to conservatively deal with this
> >       is to punt in the case that offset + maxsize reaches the
> > -     base type boundary.  */
> > +     base type boundary.
> > +
> > +     Unfortunately this is difficult to determine reliably when unions are
> > +     involved and so we are conservative in such cases.  */
> >    if (seen_variable_array_ref
> > -      && maxsize != -1
> > -      && host_integerp (TYPE_SIZE (TREE_TYPE (exp)), 1)
> > -      && bit_offset + maxsize
> > -	   == (signed)TREE_INT_CST_LOW (TYPE_SIZE (TREE_TYPE (exp))))
> > +      && (seen_union
> > +	  || (maxsize != -1
> > +	      && host_integerp (TYPE_SIZE (TREE_TYPE (exp)), 1)
> > +	      && bit_offset + maxsize
> > +	      == (signed) TREE_INT_CST_LOW (TYPE_SIZE (TREE_TYPE (exp))))))
> >      maxsize = -1;
> >  
> >    /* ???  Due to negative offsets in ARRAY_REF we can end up with