[FFmpeg-devel] [PATCH v1] aarch64/avc: Fix ff_pred16x16_plane_neon_10

[Martin Storsjö]

On Fri, 13 Dec 2024, Bin Peng wrote:

> Fix test failure on aarch64:
> ./tests/checkasm/checkasm --test=h264pred   367840
>
> Signed-off-by: Peng Bin <pengbin at visionular.com>
> ---
> libavcodec/aarch64/h264pred_neon.S | 15 +++++++--------
> 1 file changed, 7 insertions(+), 8 deletions(-)
>
> diff --git a/libavcodec/aarch64/h264pred_neon.S 
> b/libavcodec/aarch64/h264pred_neon.S
> index ea37689f34..bc185683e4 100644
> --- a/libavcodec/aarch64/h264pred_neon.S
> +++ b/libavcodec/aarch64/h264pred_neon.S
> @@ -502,28 +502,27 @@ function ff_pred16x16_plane_neon_10, export=1
>         add             v7.4h,  v7.4h,  v0.4h
>         shl             v2.4h,  v7.4h,  #4
>         ssubl           v2.4s,  v2.4h,  v3.4h
> -        shl             v3.4h,  v4.4h,  #4
>         ext             v0.16b, v0.16b, v0.16b, #14
> -        ssubl           v6.4s,  v5.4h,  v3.4h
> +        sshll           v6.4s,  v5.4h,  #0     /** c */

sshll #0 is equal to sxtl, which would be more idiomatic IMO.

The patch wasn't possible to apply, I'm not sure what happened to the 
whitespaces etc.

I applied the changes manually (while also changing the added comments to 
use // rather than /** */, to match the existing style) and pushed the 
commit, thanks!

// Martin