On 02/26/16 06:37, Wilco Dijkstra wrote:
> Evandro Menezes <e.menezes@samsung.com> wrote:
>> I have a question though: is it necessary to add the "fp" and "simd"
>> attributes to both movsf_aarch64 and movdf_aarch64 as well?
> You need at least the "simd" attribute, but providing "fp" as well is clearer
> (in principle the TARGET_FLOAT check in the pattern condition is
> redundant as a result, but the movhf and movtf patterns already do both).
>
> Also you want to use the smallest possible SIMD size as these are
> scalar operations and some microarchitectures execute 64-bit operations
> more efficiently than 128-bit ones, so:
>
>      mov\\t%0.h[0], %w1
> +   movi\\t%0.4h, #0
>      umov\\t%w0, %1.h[0]
>
>      fmov\\t%s0, %w1
> +   movi\\t%0.2s, #0
>      fmov\\t%w0, %s1
>
> With those changes it should be ready for commit once you get the OK from James/Marcus.

         Replace insn to zero up SIMD registers

         gcc/
             * config/aarch64/aarch64.md
             (*movhf_aarch64): Add "movi %0, #0" to zero up register.
             (*movsf_aarch64): Likewise and add "simd" and "fp" attributes.
             (*movdf_aarch64): Likewise.

Please, verify the new "simd" and "fp" attributes for SF and DF.

Thank you,

-- 
Evandro Menezes