[tree-ssa] verify_stmts and verify_ssa

[Jan Hubicka]

> On Fri, Nov 21, 2003 at 02:41:53AM +0100, Jan Hubicka wrote:
> > +       if (TREE_CODE (x) == ARRAY_REF
> > + 	  || TREE_CODE (x) == COMPONENT_REF
> > + 	  || TREE_CODE (x) == REALPART_EXPR
> > + 	  || TREE_CODE (x) == IMAGPART_EXPR)
> > + 	x = TREE_OPERAND (x, 0);
> 
> This must be a while loop.  At least for the ARRAY_REF and
> COMPONENT_REF parts.
> 
> > +       if (TREE_CODE (x) == STRING_CST
> > + 	  || TREE_CODE (x) == LABEL_DECL
> > + 	  || TREE_CODE (x) == FUNCTION_DECL)
> > + 	return NULL;
> > +       if (!TREE_ADDRESSABLE (x))
> > +         return x;
> 
> You'd better have a VAR_DECL or PARM_DECL here.
> 
> > +   if ((TREE_CODE (t) == ARRAY_REF
> > +        || TREE_CODE (t) == COMPONENT_REF)
> > +       && DECL_P (TREE_OPERAND (t, 0))
> > +       && is_gimple_min_invariant (TREE_OPERAND (t, 1)))
> 
> Again a while loop.  Here you have to be more clever though,
> since you've got to check is_gimple_min_invariant for each
> ARRAY_REF index.  Also, is_gimple_min_invariant is wrong to
How these indexes are represented? There is only one operand for
ARRAY_REF for indexes.  OK, I will check.
> apply to the FIELD_DECL of a COMPONENT_REF; I'm surprised
> nothing went wrong.
THis is because we apparently never create these at the moment.  I am
preparing updated patch.

Honza
> 
> 
> 
> r~