Re: [PATCH 6/17][ARM] Add data processing intrinsics for float16_t.

[Matthew Wahab <matthew dot wahab at foss dot arm dot com>]

On 26/09/16 14:03, Ramana Radhakrishnan wrote:
> On Mon, Sep 26, 2016 at 1:48 PM, Christophe Lyon
> <christophe.lyon@linaro.org> wrote:
> > On 26 September 2016 at 11:43, Matthew Wahab <matthew.wahab@foss.arm.com> wrote:
> > > Hello,
> > >
> > > On 25/09/16 14:00, Christophe Lyon wrote:
> > > > > > This patch adds the new intrinsics:
> > > > > >    vbsl_f16, vbslq_f16, vdup_n_f16, vdupq_n_f16, vdup_lane_f16,
> > > > > >    vdupq_lane_f16, vext_f16, vextq_f16, vmov_n_f16, vmovq_n_f16,
> > > > > >    vrev64_f16, vrev64q_f16, vtrn_f16, vtrnq_f16, vuzp_f16, vuzpq_f16,
> > > > > >    vzip_f16, vzipq_f16.
> > > > > >
> > > > > > This patch also updates the advsimd-intrinsics testsuite to test the f16
> > > > > > variants for ARM targets. These intrinsics are only implemented in the
> > > > > > ARM target so the tests are disabled for AArch64 using an extra
> > > > > > condition on a new convenience macro FP16_SUPPORTED. This patch also
> > > > > > disables, for the ARM target, the testsuite defined macro vdup_n_f16 as
> > > > > > it is no longer needed.
> > > >
> > > >
> > > > Since you committed this patch, I've noticed that libgcc fails to build
> > > > when GCC is configured:
> > > > --target arm-none-eabi and default cpu
> > > > /tmp/9649048_29.tmpdir/ccuBwQJJ.s: Assembler messages:
> > > > /tmp/9649048_29.tmpdir/ccuBwQJJ.s:64: Error: selected processor does
> > > > not support ARM mode `movwlt r0,32768'
> > > > /tmp/9649048_29.tmpdir/ccuBwQJJ.s:65: Error: selected processor does
> > > > not support ARM mode `movwge r0,32767'
> > > > make[4]: *** [_ssaddHQ.o] Error 1
> > > > make[4]: Leaving directory
> > > >
> > > > `/tmp/9649048_29.tmpdir/aci-gcc-fsf/builds/gcc-fsf-gccsrc/obj-arm-none-eabi/gcc1/arm-none-eabi/fpu/libgcc'
> > >
> > >
> > > I can't reproduce the failure, could you send the configure arguments for
> > > the build.
> >
> > If I'm not mistaken, that is:
> >   --target=arm-none-eabi  --disable-nls --disable-libgomp
> > --disable-libmudflap --disable-libcilkrts --enable-checking
> > --enable-languages=c,c++ --with-newlib
> >
> > Maybe you've disabled multilibs?
>
>
> I'm pretty sure I built this as part of reviewing all these patches
> with --with-mutlib-list=aprofile and didnt' see any failures. Not sure
> what's going on here.

I think the problem is that the new patterns use MOVW, which is a Thumb-2 
instruction, but don't check for Thumb-2 support in the target. I'm testing a 
patch to fix this.

Matthew