[FFmpeg-devel] [PATCH 1/2] aarch64/hevcdsp_idct_neon: Optimize idct dc

[Zhao Zhili]

> On Mar 2, 2025, at 05:39, Martin Storsjö <martin at martin.st> wrote:
> 
> On Thu, 20 Feb 2025, Zhao Zhili wrote:
> 
>> From: Zhao Zhili <zhilizhao at tencent.com>
>> 
>> clang does better than the assembly code before the patch, especially
>> for small size:
>> 
>> hevc_idct_4x4_dc_8_c:                                   11.2 ( 1.00x)
>> hevc_idct_4x4_dc_8_neon:                                15.5 ( 0.73x)
>> hevc_idct_4x4_dc_10_c:                                  12.0 ( 1.00x)
>> hevc_idct_4x4_dc_10_neon:                               15.2 ( 0.79x)
>> hevc_idct_8x8_dc_8_c:                                   13.2 ( 1.00x)
>> hevc_idct_8x8_dc_8_neon:                                18.2 ( 0.73x)
>> hevc_idct_8x8_dc_10_c:                                  13.5 ( 1.00x)
>> hevc_idct_8x8_dc_10_neon:                               17.2 ( 0.78x)
>> hevc_idct_16x16_dc_8_c:                                 41.8 ( 1.00x)
>> hevc_idct_16x16_dc_8_neon:                              37.8 ( 1.11x)
>> hevc_idct_16x16_dc_10_c:                                41.8 ( 1.00x)
>> hevc_idct_16x16_dc_10_neon:                             37.8 ( 1.11x)
>> hevc_idct_32x32_dc_8_c:                                130.2 ( 1.00x)
>> hevc_idct_32x32_dc_8_neon:                             132.2 ( 0.98x)
>> hevc_idct_32x32_dc_10_c:                               130.2 ( 1.00x)
>> hevc_idct_32x32_dc_10_neon:                            132.2 ( 0.98x)
>> 
>> This patch basically clone what the compiler does, so the performance
>> is the same.
>> ---
>> libavcodec/aarch64/hevcdsp_idct_neon.S | 59 ++++++++++++++------------
>> 1 file changed, 33 insertions(+), 26 deletions(-)
>> 
>> diff --git a/libavcodec/aarch64/hevcdsp_idct_neon.S b/libavcodec/aarch64/hevcdsp_idct_neon.S
>> index 3cac6e6db9..4543ab6b07 100644
>> --- a/libavcodec/aarch64/hevcdsp_idct_neon.S
>> +++ b/libavcodec/aarch64/hevcdsp_idct_neon.S
>> @@ -888,38 +888,45 @@ function ff_hevc_transform_luma_4x4_neon_8, export=1
>>        ret
>> endfunc
>> 
>> +.macro idct_8x8_dc_store offset
>> +.irp    i, 0x0, 0x20, 0x40, 0x60
>> +        stp             q0, q0, [x0, #(\offset + \i)]
>> +.endr
>> +.endm
>> +
>> +.macro idct_16x16_dc_store
>> +.irp    index, 0x0, 0x80, 0x100, 0x180
>> +        idct_8x8_dc_store offset=\index
>> +.endr
>> +.endm
>> +
>> // void ff_hevc_idct_NxN_dc_DEPTH_neon(int16_t *coeffs)
>> .macro idct_dc size, bitdepth
>> function ff_hevc_idct_\size\()x\size\()_dc_\bitdepth\()_neon, export=1
>> -        ld1r            {v4.8h}, [x0]
>> -        srshr           v4.8h,  v4.8h,  #1
>> -        srshr           v0.8h,  v4.8h,  #(14 - \bitdepth)
>> -        srshr           v1.8h,  v4.8h,  #(14 - \bitdepth)
>> -.if \size > 4
>> -        srshr           v2.8h,  v4.8h,  #(14 - \bitdepth)
>> -        srshr           v3.8h,  v4.8h,  #(14 - \bitdepth)
>> -.if \size > 16 /* dc 32x32 */
>> -        mov             x2,  #4
>> +        ldrsh           w1, [x0]
>> +        add             w1, w1, #1
>> +        asr             w1, w1, #1
>> +        add             w1, w1, #(1 << (13 - \bitdepth))
>> +        asr             w1, w1, #(14 - \bitdepth)
>> +        dup             v0.8h, w1
>> +
>> +.if \size < 8
>> +        stp             q0, q0, [x0]
>> +.else
>> +.if \size < 16
> 
> You can use .elseif instead of .else and a nested .if.

Modified locally and pushed, thanks for the review.

> 
> Other than that, this patch looks good to me, thanks!
> 
> // Martin
> 
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org <mailto:ffmpeg-devel at ffmpeg.org>
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> 
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org <mailto:ffmpeg-devel-request at ffmpeg.org> with subject "unsubscribe".