Re: [PATCH 9/17][ARM] Add NEON FP16 arithmetic instructions.

[Ramana Radhakrishnan <ramana.gcc@googlemail.com>]

On Mon, Jul 4, 2016 at 3:09 PM, Matthew Wahab
<matthew.wahab@foss.arm.com> wrote:
> On 18/05/16 01:58, Joseph Myers wrote:
>> On Tue, 17 May 2016, Matthew Wahab wrote:
>>
>>> As with the VFP FP16 arithmetic instructions, operations on __fp16
>>> values are done by conversion to single-precision. Any new optimization
>>> supported by the instruction descriptions can only apply to code
>>> generated using intrinsics added in this patch series.
>>
>> As with the scalar instructions, I think it is legitimate in most cases to
>> optimize arithmetic via single precision to work direct on __fp16 values
>> (and this would be natural for vectorization of __fp16 arithmetic).
>>
>>> A number of the instructions are modelled as two variants, one using
>>> UNSPEC and the other using RTL operations, with the model used decided
>>> by the funsafe-math-optimizations flag. This follows the
>>> single-precision instructions and is due to the half-precision
>>> operations having the same conditions and restrictions on their use in
>>> optmizations (when they are enabled).
>>
>> (Of course, these restrictions still apply.)
>
> The F16 support generally follows the F32 implementation and, for F32,
> direct arithmetic vector operations are only available when
> unsafe-math-optimizations is enabled. I want to check the behaviour of
> the F16 operations when unsafe-math is enabled so I'll defer to a follow
> up patch the change to use standard names for the vector operations.
>
> There are still some changes from the previous patch:
>
> - Two fma/fmsub patterns *fma<VH:mode>4 and <*fmsub<VH:mode>4 are
>   dropped since they just duplicated *fma<VH:mode>4_intrinsic and
>   <*fmsub<VH:mode>4_intrinsic.
>
> - Patterns neon_vadd<mode>_unspec and neon_vsub<mode>_unspec are
>   dropped, they were redundant.
>
> - <absneg_str><mode>2_fp16 is renamed to <absneg_str><mode>2. This
>   implements the abs and neg operations which are always safe to use.
>
> - neon_vsqrte<mode> is renamed to neon_vrsqrte<mode>. This is a
>   misspelled intrinsic that wasn't caught in testing because the
>   relevant test case is missing. The intrinsic is fixed here and in
>   other patches and an advsimd-intrinsics test added later in the
>   (updated) series.
>
> - neon_vcvt<sup>_n<mode: The bounds on the scalar were wrong, the
>   correct range for f16 is 0-17.
>
> - Test armv8_2-fp16-arith-1.c is updated to expect f16 arithmetic
>   instructions rather then f32 and to use the neon command line options.
>
> Tested the series for arm-none-linux-gnueabihf with native bootstrap and
> make check and for arm-none-eabi and armeb-none-eabi with make check on
> an ARMv8.2-A emulator.
>
> Ok for trunk?

OK.

Ramana
> Matthew
>
> 2016-07-04  Matthew Wahab  <matthew.wahab@arm.com>
>
>         * config/arm/iterators.md (VCVTHI): New.
>         (NEON_VCMP): Add UNSPEC_VCLT and UNSPEC_VCLE.  Fix a long line.
>         (NEON_VAGLTE): New.
>         (VFM_LANE_AS): New.
>         (VH_CVTTO): New.
>         (V_reg): Add HF, V4HF and V8HF.  Fix white-space.
>         (V_HALF): Add V4HF.  Fix white-space.
>         (V_if_elem): Add HF, V4HF and V8HF.  Fix white-space.
>         (V_s_elem): Likewise.
>         (V_sz_elem): Fix white-space.
>         (V_elem_ch): Likewise.
>         (VH_elem_ch): New.
>         (scalar_mul_constraint): Add V8HF and V4HF.
>         (Is_float_mode): Fix white-space.
>         (Is_d_reg): Fix white-space.
>         (q): Add HF.  Fix white-space.
>         (float_sup): New.
>         (float_SUP): New.
>         (cmp_op_unsp): Add UNSPEC_VCALE and UNSPEC_VCALT.
>         (neon_vfm_lane_as): New.
>         * config/arm/neon.md (add<mode>3_fp16): New.
>         (sub<mode>3_fp16): New.
>         (mul<mode>3add<mode>_neon): New.
>         (fma<VH:mode>4_intrinsic): New.
>         (fmsub<VCVTF:mode>4_intrinsic): Fix white-space.
>         (fmsub<VH:mode>4_intrinsic): New.
>         (<absneg_str><mode>2): New.
>         (neon_v<absneg_str><mode>): New.
>         (neon_v<fp16_rnd_str><mode>): New.
>         (neon_vrsqrte<mode>): New.
>         (neon_vpaddv4hf): New.
>         (neon_vadd<mode>): New.
>         (neon_vsub<mode>): New.
>         (neon_vmulf<mode>): New.
>         (neon_vfma<VH:mode>): New.
>         (neon_vfms<VH:mode>): New.
>         (neon_vc<cmp_op><mode>): New.
>         (neon_vc<cmp_op><mode>_fp16insn): New
>         (neon_vc<cmp_op_unsp><mode>_fp16insn_unspec): New.
>         (neon_vca<cmp_op><mode>): New.
>         (neon_vca<cmp_op><mode>_fp16insn): New.
>         (neon_vca<cmp_op_unsp><mode>_fp16insn_unspec): New.
>         (neon_vc<cmp_op>z<mode>): New.
>         (neon_vabd<mode>): New.
>         (neon_v<maxmin>f<mode>): New.
>         (neon_vp<maxmin>fv4hf: New.
>         (neon_<fmaxmin_op><mode>): New.
>         (neon_vrecps<mode>): New.
>         (neon_vrsqrts<mode>): New.
>         (neon_vrecpe<mode>): New (VH variant).
>         (neon_vdup_lane<mode>_internal): New.
>         (neon_vdup_lane<mode>): New.
>         (neon_vcvt<sup><mode>): New (VCVTHI variant).
>         (neon_vcvt<sup><mode>): New (VH variant).
>         (neon_vcvt<sup>_n<mode>): New (VH variant).
>         (neon_vcvt<sup>_n<mode>): New (VCVTHI variant).
>         (neon_vcvt<vcvth_op><sup><mode>): New.
>         (neon_vmul_lane<mode>): New.
>         (neon_vmul_n<mode>): New.
>         * config/arm/unspecs.md (UNSPEC_VCALE): New
>         (UNSPEC_VCALT): New.
>         (UNSPEC_VFMA_LANE): New.
>         (UNSPECS_VFMS_LANE): New.
>
> testsuite/
> 2016-07-04  Matthew Wahab  <matthew.wahab@arm.com>
>
>         * gcc.target/arm/armv8_2-fp16-arith-1.c: Use arm_v8_2a_fp16_neon
>         options.  Add tests for float16x4_t and float16x8_t.
>