[Bug tree-optimization/40550] Segmentation fault caused by alignment error in sse code

[rguenther at suse dot de]

------- Comment #12 from rguenther at suse dot de  2009-06-28 11:07 -------
Subject: Re:  Segmentation fault caused by
 alignment error in sse code

On Sun, 28 Jun 2009, ubizjak at gmail dot com wrote:

> ------- Comment #11 from ubizjak at gmail dot com  2009-06-28 11:04 -------
> Patch in testing:
> 
> Index: tree-vect-generic.c
> ===================================================================
> --- tree-vect-generic.c (revision 148947)
> +++ tree-vect-generic.c (working copy)
> @@ -481,8 +481,10 @@
>        tree vector_compute_type
>          = type_for_widest_vector_mode (TYPE_MODE (TREE_TYPE (type)), op,
>                                        TYPE_SATURATING (TREE_TYPE (type)));
> -      if (vector_compute_type != NULL_TREE)
> -        compute_type = vector_compute_type;
> +      if (vector_compute_type != NULL_TREE
> +         && (int_size_in_bytes (vector_compute_type)
> +             <= int_size_in_bytes (compute_type)))
> +       compute_type = vector_compute_type;
>      }
> 
>    /* If we are breaking a BLKmode vector into smaller pieces,

Shouldn't we be able to use mmx regs here?  Thus not rely on
type_for_widest_vector_mode, but instead see if the original vector
mode is supported by the backend?

Richard.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40550