------- Additional Comments From dorit at il dot ibm dot com 2005-03-14 20:01
-------
The problem is that we take the size_type of a void type, and pass null to
fold_convert in vect_analyze_pointer_ref_access.
So one thing to do is to make sure that we're dealing with a complete type:
*************** vect_analyze_pointer_ref_access (tree me
*** 1321,1326 ****
--- 1403,1416 ----
*ptr_step = fold_convert (ssizetype, step);
innertype = TREE_TYPE (reftype);
+ if (!COMPLETE_TYPE_P (innertype))
+ {
+ if (vect_print_dump_info (REPORT_UNVECTORIZED_LOOPS,
+ LOOP_LOC (loop_vinfo)))
+ fprintf (vect_dump, "not vectorized: pointer to incomplete type.");
+ return NULL;
+ }
+
/* Check that STEP is a multiple of type size. */
if (!integer_zerop (size_binop (TRUNC_MOD_EXPR, *ptr_step,
fold_convert (ssizetype, TYPE_SIZE_UNIT (innertype)))))
This solves the ICE, but I noticed that we're actually looking at the wrong
type - we're interested in the type of the data-reference, not the type of
its 'init'. The following is what we want (also solves the ICE):
*************** vect_analyze_pointer_ref_access (tree me
*** 1310,1317 ****
return NULL;
}
! reftype = TREE_TYPE (init);
! if (!POINTER_TYPE_P (reftype))
{
if (vect_print_dump_info (REPORT_UNVECTORIZED_LOOPS,
LOOP_LOC (loop_vinfo)))
--- 1393,1399 ----
return NULL;
}
! if (!POINTER_TYPE_P (TREE_TYPE (init)))
{
if (vect_print_dump_info (REPORT_UNVECTORIZED_LOOPS,
LOOP_LOC (loop_vinfo)))
I'll bootstrap/test and submit.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20474
