[FFmpeg-devel] [PATCH] swscale/aarch64/output.S: refactor ff_yuv2plane1_8_neon

[Martin Storsjö]

On Fri, 31 Jan 2025, Krzysztof Pyrkosz via ffmpeg-devel wrote:

> The benchmarks (before vs after) were gathered using
> ./tests/checkasm/checkasm --test=sw_scale --bench --runs=6 | grep yuv2yuv1

Thanks for the numbers! To keep it more readable and concise, we'd only 
need to see the numbers for the _neon versions here - and we probably 
don't need to see all the variants/combinations of it, just one or a 
couple would be fine.

Anyway, the numbers look great, thanks!

> yuv2yuv1_16_512_accurate_c:                           1439.9 ( 1.00x)
> yuv2yuv1_16_512_accurate_neon:                          77.6 (18.55x)
> yuv2yuv1_16_512_approximate_c:                        1422.1 ( 1.00x)
> yuv2yuv1_16_512_approximate_neon:                       78.1 (18.20x)
> yuv2yuv1_19_512_accurate_c:                           1447.1 ( 1.00x)
> yuv2yuv1_19_512_accurate_neon:                          78.1 (18.52x)
> yuv2yuv1_19_512_approximate_c:                        1474.4 ( 1.00x)
> yuv2yuv1_19_512_approximate_neon:                       78.1 (18.87x)
>
> Krzysztof
>
> ---
> libswscale/aarch64/output.S | 18 ++++++------------
> 1 file changed, 6 insertions(+), 12 deletions(-)

Your signature above does get included as part of the commit message, 
while it's probably not needed/intended :-) I'm editing it out before 
merging though.

(I also end up needing to apply your actual email address and name, as the 
mailing list adds the "via ffmpeg-devel" to the name and rewrites the from 
address. But that's obviously outside of your control.)

>
> diff --git a/libswscale/aarch64/output.S b/libswscale/aarch64/output.S
> index 934d62dfd0..190c438870 100644
> --- a/libswscale/aarch64/output.S
> +++ b/libswscale/aarch64/output.S
> @@ -214,21 +214,15 @@ function ff_yuv2plane1_8_neon, export=1
>         and             w4, w4, #7
>         cbz             w4, 1f                          // check if offsetting present
>         ext             v0.8b, v0.8b, v0.8b, #3         // honor offsetting which can be 0 or 3 only
> -1:      uxtl            v0.8h, v0.8b                    // extend dither to 32-bit
> -        uxtl            v1.4s, v0.4h
> -        uxtl2           v2.4s, v0.8h
> +1:
> +        uxtl            v0.8h, v0.8b                    // extend dither to 32-bit
> 2:
>         ld1             {v3.8h}, [x0], #16              // read 8x16-bit @ src[j  ][i + {0..7}]: A,B,C,D,E,F,G,H
> -        sxtl            v4.4s, v3.4h
> -        sxtl2           v5.4s, v3.8h
> -        add             v4.4s, v4.4s, v1.4s
> -        add             v5.4s, v5.4s, v2.4s
> -        sqshrun         v4.4h, v4.4s, #6
> -        sqshrun2        v4.8h, v5.4s, #6
> -
> -        uqshrn          v3.8b, v4.8h, #1                // clip8(val>>7)
>         subs            w2, w2, #8                      // dstW -= 8
> -        st1             {v3.8b}, [x1], #8               // write to destination
> +        shadd           v1.8h, v0.8h, v3.8h             // v1 = (v0 + v3) >> 1
> +        sqshrun         v2.8b, v1.8h, #6                // clip_uint8(v1 >> 6)
> +
> +        st1             {v2.8b}, [x1], #8               // write to destination
>         b.gt            2b                              // loop until width consumed
>         ret
> endfunc
> --

Thanks, these changes look good - will push.

// Martin