[PATCH] [ARM] PR68532: Fix VUZP and VZIP recognition on big endian

[Charles Baylis]

On 1 February 2016 at 17:14, Kyrill Tkachov <kyrylo.tkachov@foss.arm.com> wrote:

> Indeed I see the new passes on armeb-none-eabi.
> However, the new FAILs that I see are ICEs, not just vectorisation failures,
> so they need to be looked at.
>
> The ICEs that I see are:
> FAIL: gcc.dg/torture/vshuf-v4hi.c   -O2  (internal compiler error)
> FAIL: gcc.dg/torture/vshuf-v8qi.c   -O2  (internal compiler error)

Thanks. I hadn't seen these because I wasn't running the "expensive" tests.

> Seems that the code in expr.c asserts that expand_vec_perm returned a
> non-NULL result.

It seems that my implementation of arm_evpc_neon_vuzp doesn't handle
the one vector case correctly. I'm testing a fix.

> I'll look at the patches in more detail, but in the meantime I notice that
> there are some
> GNU style issues that should be resolved, like starting comments with a
> capital letter,
> two spaces after full stop, two spaces between full stop and close comment,
> as well as some
> lines over 80 characters. The check_GNU_style.sh script in the contrib/
> directory can help
> catch some (if not all) of these.

OK, I'll fix those.

> Also, can you please send any follow-up versions of the two patches as
> separate emails,
> so that we can more easily keep track of what's comment goes to which patch.

Will do.