This is the mail archive of the fortran@gcc.gnu.org mailing list for the GNU Fortran 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: XPASS: gfortran.dg/vect/vect-5.f90 on PPC64


>    Does anyone know if the fact that the gfortran.dg
> test case vect/vect-5.f90 is now showing a XPASS on
> indicates a regression? All three dg-final options that
> are marked to fail at lp64 now pass.
>
> Running target unix/-m64
> XPASS: gfortran.dg/vect/vect-5.f90  -O  scan-tree-dump-times
> vectorized 2 loops 1
> XPASS: gfortran.dg/vect/vect-5.f90  -O  scan-tree-dump-times
> Alignment of access forced using peeling 1
> XPASS: gfortran.dg/vect/vect-5.f90  -O  scan-tree-dump-times
> Vectorizing an unaligned access 2
>
> I see the same behavior for x86_64 but not for ia64.
> The sparc64-sun-solaris2.10 only shows...
>
> XPASS: gfortran.dg/vect/vect-5.f90  -O  scan-tree-dump-times
> vectorized 2 loops 1
>
> Again, I am mainly interested to know if passing these tests on PPC64
> is a good thing or indicative of a regression.

It's a good thing :-)
Looks like lp64 should be removed from the xfail.
That's why ppc64 and x86_64 XPASS on all three checks.
sparc64 XPASSes only the first check (and not the other two alignment
related checks) because it is a "vect_no_align" target.
For the same reason ia64 doesn't have the two alignment related XPASSes.
It remains to understand why ia64 still fails to vectorize the 2 loops -
potentially this is not ilp64 related.
Do you have the dump file generated by -fdump-tree-vect-details on ia64 for
this testcase?

dorit

>            Jack


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