[FFmpeg-devel] [PATCH] lavc/vvc_mc: reduce sequential dependency in R-V V sad

[flow gg]

> Don't clobber v8 here.
> Use vsub.vv here to avoid the sequential dependency.

Updated.

<uk7b-at-foxmail.com at ffmpeg.org> 于2024年12月21日周六 20:22写道：

> From: sunyuechi <sunyuechi at iscas.ac.cn>
>
> ---
>  libavcodec/riscv/vvc/vvc_sad_rvv.S | 10 +++++-----
>  1 file changed, 5 insertions(+), 5 deletions(-)
>
> diff --git a/libavcodec/riscv/vvc/vvc_sad_rvv.S
> b/libavcodec/riscv/vvc/vvc_sad_rvv.S
> index 341167be1f..f325deee17 100644
> --- a/libavcodec/riscv/vvc/vvc_sad_rvv.S
> +++ b/libavcodec/riscv/vvc/vvc_sad_rvv.S
> @@ -36,20 +36,20 @@ func ff_vvc_sad_rvv_\vlen, zve32x, zbb, zba
>  SADVSET\vlen\w:
>          vsetvlstatic32    \w, \vlen
>          vmv.v.i           v0, 0
> -        vmv.s.x           v24, zero
>          vsetvlstatic16    \w, \vlen
>  SAD\vlen\w:
>          addi              a5, a5, -2
>          vle16.v           v8, (a0)
>          vle16.v           v16, (a1)
> -        vsub.vv           v8, v8, v16
> -        vneg.v            v16, v8
> +        vsub.vv           v24, v8, v16
> +        vsub.vv           v16, v16, v8
>          addi              a0, a0, 2 * 128 * 2
> -        vmax.vv           v8, v8, v16
> -        vwaddu.wv         v0, v0, v8
> +        vmax.vv           v8, v24, v16
>          addi              a1, a1, 2 * 128 * 2
> +        vwaddu.wv         v0, v0, v8
>          bnez              a5, SAD\vlen\w
>          vsetvlstatic32    \w, \vlen
> +        vmv.s.x           v24, zero
>          vredsum.vs        v24, v0, v24
>          vmv.x.s           a0, v24
>          ret
> --
> 2.47.1
>
> _______________________________________________
> ffmpeg-devel mailing list
> 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 with subject "unsubscribe".
>