This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH, PowerPC] altivec_expand_vec_perm_const reverses pack pattern arguments in little endian mode
- From: Bill Schmidt <wschmidt at linux dot vnet dot ibm dot com>
- To: David Edelsohn <dje dot gcc at gmail dot com>
- Cc: GCC Patches <gcc-patches at gcc dot gnu dot org>, David Edelsohn <dje at gcc dot gnu dot org>
- Date: Mon, 22 Jul 2013 19:56:15 -0500
- Subject: Re: [PATCH, PowerPC] altivec_expand_vec_perm_const reverses pack pattern arguments in little endian mode
- References: <1374534309 dot 3633 dot 96 dot camel at gnopaine> <CAGWvnym0zOTbn_7CnqeRVUQf+LsmdyLnSD1oUT6afTHqbw1jhw at mail dot gmail dot com>
On Mon, 2013-07-22 at 19:49 -0400, David Edelsohn wrote:
> On Mon, Jul 22, 2013 at 7:05 PM, Bill Schmidt
> <wschmidt@linux.vnet.ibm.com> wrote:
> > This patch is another fix for vector handling in little endian mode.
> > The first two operands for a pack pattern are two vector registers that
> > form a contiguous array of inputs. In LE mode the order of the operands
> > must be reversed so that the array remains contiguous in the reverse
> > order.
> >
> > This fixes a failure in the testsuite when run little-endian
> > (gcc.dg/vect/no-scevccp-outer-18.c). Bootstrapped and tested big-endian
> > on powerpc64-unknown-linux-gnu with no new regressions. Ok for trunk?
> >
> > Patch by Anton Blanchard.
> >
> > Thanks,
> > Bill
> >
> >
> > 2013-07-22 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
> > Anton Blanchard <anton@au1.ibm.com>
> >
> > * config/rs6000/rs6000.c (altivec_expand_vec_perm_const): Reverse
> > two operands for little-endian.
>
> Wouldn't it be better to handle this where the code is performing a
> swap a few lines above?
Hm, I don't think so. The reason for that swap has nothing to do with
endianness, so I think it would just confuse matters. Also there's a
three-way swap going on with x, op0, and op1 there, and we just want to
swap op0 and op1. I think the patch as it stands is probably easier to
grok.
Thanks,
Bill
>
> Thanks, David
>