Hi, Martin,

I modified it according to your comments. Please review again.

And here are the checkasm benchmark results of the related functions:


put_hevc_epel_uni_w_v4_8_c: 116.1
put_hevc_epel_uni_w_v4_8_neon: 48.6
put_hevc_epel_uni_w_v6_8_c: 248.9
put_hevc_epel_uni_w_v6_8_neon: 80.6
put_hevc_epel_uni_w_v8_8_c: 383.9
put_hevc_epel_uni_w_v8_8_neon: 91.9
put_hevc_epel_uni_w_v12_8_c: 806.1
put_hevc_epel_uni_w_v12_8_neon: 202.9
put_hevc_epel_uni_w_v16_8_c: 1411.1
put_hevc_epel_uni_w_v16_8_neon: 289.9
put_hevc_epel_uni_w_v24_8_c: 3168.9
put_hevc_epel_uni_w_v24_8_neon: 619.4
put_hevc_epel_uni_w_v32_8_c: 5632.9
put_hevc_epel_uni_w_v32_8_neon: 1161.1
put_hevc_epel_uni_w_v48_8_c: 12406.1
put_hevc_epel_uni_w_v48_8_neon: 2476.4
put_hevc_epel_uni_w_v64_8_c: 22001.4
put_hevc_epel_uni_w_v64_8_neon: 4343.9


在 2023/6/12 16:09, Martin Storsjö 写道:
> On Sun, 4 Jun 2023, Logan.Lyu@myais.com.cn wrote:
>
>> From: Logan Lyu <Logan.Lyu@myais.com.cn>
>>
>> Signed-off-by: Logan Lyu <Logan.Lyu@myais.com.cn>
>> ---
>> libavcodec/aarch64/hevcdsp_epel_neon.S    | 504 ++++++++++++++++++++++
>> libavcodec/aarch64/hevcdsp_init_aarch64.c |   6 +
>> 2 files changed, 510 insertions(+)
>>
>> diff --git a/libavcodec/aarch64/hevcdsp_epel_neon.S 
>> b/libavcodec/aarch64/hevcdsp_epel_neon.S
>> index fe494dd843..4841f49dab 100644
>> --- a/libavcodec/aarch64/hevcdsp_epel_neon.S
>> +++ b/libavcodec/aarch64/hevcdsp_epel_neon.S
>
>
>> +function ff_hevc_put_hevc_epel_uni_w_v48_8_neon, export=1
>> +        stp             q8, q9, [sp, #-32]
>> +        stp             q10, q11, [sp, #-64]
>
> This backs up values on the stack without decrementing the stack 
> pointer, i.e. storing it in the red zone. Whether this is supported 
> depends on the platform ABI. Linux and macOS have a 128 byte red zone 
> on aarch64, while Windows only has 16 bytes. So for portability, don't 
> rely on a red zone at all.
>
> I.e., here please decrement the stack pointer like in a previous patch:
>
>     stp q8,  q9,  [sp, #-64]!
>     stp q10, q11, [sp, #32]
>
> And inversely when restoring it.
>
>> +        EPEL_UNI_W_V_HEADER
>> +
>> +        ld1             {v16.16b, v17.16b, v18.16b}, [x2], x3
>> +        ld1             {v19.16b, v20.16b, v21.16b}, [x2], x3
>> +        ld1             {v22.16b, v23.16b, v24.16b}, [x2], x3
>> +1:
>> +        ld1             {v25.16b, v26.16b, v27.16b}, [x2], x3
>> +
>> +        EPEL_UNI_W_V16_CALC v4, v6, v16, v19, v22, v25, v8, v9, v10, 
>> v11
>> +        EPEL_UNI_W_V16_CALC v5, v7, v17, v20, v23, v26, v8, v9, v10, 
>> v11
>> +        EPEL_UNI_W_V16_CALC v6, v7, v18, v21, v24, v27, v8, v9, v10, 
>> v11
>> +        st1             {v4.16b, v5.16b, v6.16b}, [x0], x1
>> +        subs            w4, w4, #1
>> +        b.eq            2f
>> +        ld1             {v16.16b, v17.16b, v18.16b}, [x2], x3
>> +        EPEL_UNI_W_V16_CALC v4, v6, v19, v22, v25, v16, v8, v9, v10, 
>> v11
>> +        EPEL_UNI_W_V16_CALC v5, v7, v20, v23, v26, v17, v8, v9, v10, 
>> v11
>> +        EPEL_UNI_W_V16_CALC v6, v7, v21, v24, v27, v18, v8, v9, v10, 
>> v11
>> +        st1             {v4.16b, v5.16b, v6.16b}, [x0], x1
>> +        subs            w4, w4, #1
>> +        b.eq            2f
>> +        ld1             {v19.16b, v20.16b, v21.16b}, [x2], x3
>> +        EPEL_UNI_W_V16_CALC v4, v6,  v22, v25, v16, v19, v8, v9, 
>> v10, v11
>> +        EPEL_UNI_W_V16_CALC v5, v7,  v23, v26, v17, v20, v8, v9, 
>> v10, v11
>> +        EPEL_UNI_W_V16_CALC v6, v7,  v24, v27, v18, v21, v8, v9, 
>> v10, v11
>> +        st1             {v4.16b, v5.16b, v6.16b}, [x0], x1
>> +        subs            w4, w4, #1
>> +        b.eq            2f
>> +        ld1             {v22.16b, v23.16b, v24.16b}, [x2], x3
>> +        EPEL_UNI_W_V16_CALC v4, v6,  v25, v16, v19, v22, v8, v9, 
>> v10, v11
>> +        EPEL_UNI_W_V16_CALC v5, v7,  v26, v17, v20, v23, v8, v9, 
>> v10, v11
>> +        EPEL_UNI_W_V16_CALC v6, v7,  v27, v18, v21, v24, v8, v9, 
>> v10, v11
>> +        st1             {v4.16b, v5.16b, v6.16b}, [x0], x1
>> +        subs            w4, w4, #1
>> +        b.hi            1b
>> +2:
>> +        ldp             q8, q9, [sp, #-32]
>> +        ldp             q10, q11, [sp, #-64]
>> +        ret
>> +endfunc
>> +
>> +function ff_hevc_put_hevc_epel_uni_w_v64_8_neon, export=1
>> +        stp             q8, q9, [sp, #-32]
>> +        stp             q10, q11, [sp, #-64]
>> +        stp             q12, q13, [sp, #-96]
>> +        stp             q14, q15, [sp, #-128]
>
> Same thing here.
>
> // Martin
>