This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Testsuite patch for gcc.dg/vect/no-vfa-vect-43.c on IA64



gcc-patches-owner@gcc.gnu.org wrote on 22/04/2009 12:22:47:

> On Tue, Apr 21, 2009 at 6:56 PM, Steve Ellcey <sje@cup.hp.com> wrote:
> > The alias improvement branch, r145494, is causing a failure of
> > gcc.dg/vect/no-vfa-vect-43.c on IA64 HP-UX and Linux. ?Looking at the
> > test changes done with the alias improvement checkin, the number of
> > expected vectorized loops was increased from 1 to 2 for this test but
> > the number of loops that are versioned for alignment requirements was
> > left at 1 and should now be 2. ?This patch changes the number of
expected
> > versionings to 2 and allows the test to pass on IA64 HP-UX and Linux.
> >
> > It might look like we should only have to version the loop in main2 and
> > not the one in main1 because the input arrays in main1 have aligned
> > attributes on them. ?But the output array in main1 does not have an
> > alignment attribute and GCC doesn't seem to allow alignment attributes
> > on function arguments so there should be 2 versionings.
> >
> > Tested on IA64 HP-UX and Linux, OK for checkin?
>
> Hmm.  But that will now FAIL on for example x86_64 where the alignment
> is probably ensured by peeling(?).

Right, but I don't think it will fail on x86_64 because the versioning
check is only for target vect_no_align.

Thanks,
Ira


> I suggest to instead do
>
> /*  { dg-final { scan-tree-dump-times "Alignment of access forced
> using versioning" 1 "vect" { target vect_no_align } { xfail ia64-*-* }
>  } } */
> /*  { dg-final { scan-tree-dump-times "Alignment of access forced
> using versioning" 2 "vect" { target ia64-*-* } } } */
>
> Thanks,
> Richard.
>
> > Steve Ellcey
> > sje@cup.hp.com
> >
> >
> > 2009-04-21 ?Steve Ellcey ?<sje@cup.hp.com>
> >
> > ? ? ? ?PR testsuite/39623
> > ? ? ? ?* gcc.dg/vect/no-vfa-vect-43.c Fix vect_no_align check.
> >
> >
> >
> > Index: gcc.dg/vect/no-vfa-vect-43.c
> > ===================================================================
> > --- gcc.dg/vect/no-vfa-vect-43.c ? ? ? ?(revision 146528)
> > +++ gcc.dg/vect/no-vfa-vect-43.c ? ? ? ?(working copy)
> > @@ -76,5 +76,5 @@ int main (void)
> > ?}
> >
> > ?/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 2 "vect" } }
*/
> > -/* ?{ dg-final { scan-tree-dump-times "Alignment of access forced
> using versioning" 1 "vect" { target vect_no_align } } } */
> > +/* ?{ dg-final { scan-tree-dump-times "Alignment of access forced
> using versioning" 2 "vect" { target vect_no_align } } } */
> > ?/* { dg-final { cleanup-tree-dump "vect" } } */
> >


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]