Best regards,
Thomas
> -Original Message-
> From: Richard Biener [mailto:rguent...@suse.de]
> Sent: Friday, January 23, 2015 6:19 PM
> To: Thomas Preud'homme
> Cc: gcc-patches@gcc.gnu.org
> Subject: RE: [PATCH] Fix PR64718: bad 16-bit bswap replacement
>
&
On Fri, 23 Jan 2015, Thomas Preud'homme wrote:
> > From: Richard Biener [mailto:rguent...@suse.de]
> > Sent: Friday, January 23, 2015 6:01 PM
> > > + if (bswap && n->range == 16)
> > > +bswap_type = TYPE_UNSIGNED (TREE_TYPE (src)) ?
> > short_unsigned_type_node
> > > +
> From: Richard Biener [mailto:rguent...@suse.de]
> Sent: Friday, January 23, 2015 6:01 PM
> > + if (bswap && n->range == 16)
> > +bswap_type = TYPE_UNSIGNED (TREE_TYPE (src)) ?
> short_unsigned_type_node
> > +:
> short_integer_type_node;
>
> I don'
On Fri, 23 Jan 2015, Thomas Preud'homme wrote:
> Hi Richard,
>
> When detecting a 16-bit hand-crafted byte swap, the bswap pass replaces the
> sequence of statements by a rotation left of 8 bits of the same type as the
> source. For source whose type size is bigger than 16 bits this is wrong. F
Hi Richard,
When detecting a 16-bit hand-crafted byte swap, the bswap pass replaces the
sequence of statements by a rotation left of 8 bits of the same type as the
source. For source whose type size is bigger than 16 bits this is wrong. For
instance,
int swap(int x)
{
return (unsigned short) (
