diff mbox

[rs6000] Skip another test case for little endian

Message ID 1386027144.25660.26.camel@gnopaine
State New
Headers show

Commit Message

Bill Schmidt Dec. 2, 2013, 11:32 p.m. UTC
The test case gcc.dg/vect/costmodel/ppc/costmodel-slp-34.c fails if a
loop isn't vectorized.  When compiled for little endian, the cost of
vectorizing the loop is deemed too high to vectorize due to unaligned
vector accesses within the loop.  Therefore we should skip this test for
LE.

Verified on powerpc64le-unknown-linux-gnu.  Is this ok for trunk?

Thanks,
Bill


2013-12-02  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

	* gcc.dg/vect/costmodel/ppc/costmodel-slp-34.c: Skip for little
	endian.

Comments

David Edelsohn Dec. 3, 2013, 12:54 a.m. UTC | #1
On Mon, Dec 2, 2013 at 6:32 PM, Bill Schmidt
<wschmidt@linux.vnet.ibm.com> wrote:
> The test case gcc.dg/vect/costmodel/ppc/costmodel-slp-34.c fails if a
> loop isn't vectorized.  When compiled for little endian, the cost of
> vectorizing the loop is deemed too high to vectorize due to unaligned
> vector accesses within the loop.  Therefore we should skip this test for
> LE.
>
> Verified on powerpc64le-unknown-linux-gnu.  Is this ok for trunk?
>
> Thanks,
> Bill
>
>
> 2013-12-02  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
>
>         * gcc.dg/vect/costmodel/ppc/costmodel-slp-34.c: Skip for little
>         endian.

LGTM.

Thanks, David
Mike Stump Dec. 3, 2013, 12:54 a.m. UTC | #2
On Dec 2, 2013, at 3:32 PM, Bill Schmidt <wschmidt@linux.vnet.ibm.com> wrote:
> The test case gcc.dg/vect/costmodel/ppc/costmodel-slp-34.c fails if a
> loop isn't vectorized.  When compiled for little endian, the cost of
> vectorizing the loop is deemed too high

> Index: gcc/testsuite/gcc.dg/vect/costmodel/ppc/costmodel-slp-34.c
> ===================================================================
> --- gcc/testsuite/gcc.dg/vect/costmodel/ppc/costmodel-slp-34.c	(revision 205585)
> +++ gcc/testsuite/gcc.dg/vect/costmodel/ppc/costmodel-slp-34.c	(working copy)
> @@ -1,4 +1,5 @@
> /* { dg-require-effective-target vect_int } */
> +/* { dg-skip-if "" { powerpc*le-*-* } { "*" } { "" } } */

We like noting comments somewhere why we want to skip, the idea being if another target has high costs, they can just know that this is a comment failure mode.  Maybe something like:

+/* { dg-skip-if "cost to high" { powerpc*le-*-* } { "*" } { "" } } */

?
Bill Schmidt Dec. 3, 2013, 1:03 a.m. UTC | #3
Good idea, Mike, I'll make that change.

Thanks,
Bill

On Mon, 2013-12-02 at 16:54 -0800, Mike Stump wrote:
> On Dec 2, 2013, at 3:32 PM, Bill Schmidt <wschmidt@linux.vnet.ibm.com> wrote:
> > The test case gcc.dg/vect/costmodel/ppc/costmodel-slp-34.c fails if a
> > loop isn't vectorized.  When compiled for little endian, the cost of
> > vectorizing the loop is deemed too high
> 
> > Index: gcc/testsuite/gcc.dg/vect/costmodel/ppc/costmodel-slp-34.c
> > ===================================================================
> > --- gcc/testsuite/gcc.dg/vect/costmodel/ppc/costmodel-slp-34.c	(revision 205585)
> > +++ gcc/testsuite/gcc.dg/vect/costmodel/ppc/costmodel-slp-34.c	(working copy)
> > @@ -1,4 +1,5 @@
> > /* { dg-require-effective-target vect_int } */
> > +/* { dg-skip-if "" { powerpc*le-*-* } { "*" } { "" } } */
> 
> We like noting comments somewhere why we want to skip, the idea being if another target has high costs, they can just know that this is a comment failure mode.  Maybe something like:
> 
> +/* { dg-skip-if "cost to high" { powerpc*le-*-* } { "*" } { "" } } */
> 
> ?
diff mbox

Patch

Index: gcc/testsuite/gcc.dg/vect/costmodel/ppc/costmodel-slp-34.c
===================================================================
--- gcc/testsuite/gcc.dg/vect/costmodel/ppc/costmodel-slp-34.c	(revision 205585)
+++ gcc/testsuite/gcc.dg/vect/costmodel/ppc/costmodel-slp-34.c	(working copy)
@@ -1,4 +1,5 @@ 
 /* { dg-require-effective-target vect_int } */
+/* { dg-skip-if "" { powerpc*le-*-* } { "*" } { "" } } */
 
 #include <stdarg.h>
 #include "../../tree-vect.h"