[FFmpeg-devel] [PATCH 03/10] lavc/vp8dsp: R-V V put_bilin_v

[Rémi Denis-Courmont]

Le lauantaina 4. toukokuuta 2024, 17.48.32 EEST uk7b at foxmail.com a écrit :
> From: sunyuechi <sunyuechi at iscas.ac.cn>
> 
> C908:
> vp8_put_bilin4_v_c: 383.5
> vp8_put_bilin4_v_rvv_i32: 139.7
> vp8_put_bilin8_v_c: 1455.7
> vp8_put_bilin8_v_rvv_i32: 299.7
> vp8_put_bilin16_v_c: 2863.7
> vp8_put_bilin16_v_rvv_i32: 347.7
> ---
>  libavcodec/riscv/vp8dsp_init.c |  7 +++++++
>  libavcodec/riscv/vp8dsp_rvv.S  | 36 ++++++++++++++++++++++++++++++++++
>  2 files changed, 43 insertions(+)
> 
> diff --git a/libavcodec/riscv/vp8dsp_init.c b/libavcodec/riscv/vp8dsp_init.c
> index 32cb4893a4..10e1498d01 100644
> --- a/libavcodec/riscv/vp8dsp_init.c
> +++ b/libavcodec/riscv/vp8dsp_init.c
> @@ -58,6 +58,13 @@ av_cold void ff_vp78dsp_init_riscv(VP8DSPContext *c)
>          c->put_vp8_bilinear_pixels_tab[1][0][2] = ff_put_vp8_bilin8_h_rvv;
>          c->put_vp8_bilinear_pixels_tab[2][0][1] = ff_put_vp8_bilin4_h_rvv;
>          c->put_vp8_bilinear_pixels_tab[2][0][2] = ff_put_vp8_bilin4_h_rvv;
> +
> +        c->put_vp8_bilinear_pixels_tab[0][1][0] = ff_put_vp8_bilin16_v_rvv;
> +        c->put_vp8_bilinear_pixels_tab[0][2][0] =
> ff_put_vp8_bilin16_v_rvv; +        c->put_vp8_bilinear_pixels_tab[1][1][0]
> = ff_put_vp8_bilin8_v_rvv; +        c->put_vp8_bilinear_pixels_tab[1][2][0]
> = ff_put_vp8_bilin8_v_rvv; +        c->put_vp8_bilinear_pixels_tab[2][1][0]
> = ff_put_vp8_bilin4_v_rvv; +        c->put_vp8_bilinear_pixels_tab[2][2][0]
> = ff_put_vp8_bilin4_v_rvv; }
>  #endif
>  }
> diff --git a/libavcodec/riscv/vp8dsp_rvv.S b/libavcodec/riscv/vp8dsp_rvv.S
> index c8d265e516..a58c197ba1 100644
> --- a/libavcodec/riscv/vp8dsp_rvv.S
> +++ b/libavcodec/riscv/vp8dsp_rvv.S
> @@ -152,3 +152,39 @@ endfunc
>  func ff_put_vp8_bilin4_h_rvv, zve32x
>          put_vp8_bilin_h 4
>  endfunc
> +
> +.macro put_vp8_bilin_v
> +        li              t1, 8
> +        li              t4, 4
> +        sub             t1, t1, a6
> +1:
> +        add             t2, a2, a3
> +        addi            a4, a4, -1
> +        vle8.v          v0, (a2)
> +        vle8.v          v2, (t2)
> +        vwmulu.vx       v28, v0, t1
> +        vwmaccu.vx      v28, a6, v2
> +        vwaddu.wx       v24, v28, t4
> +        vnsra.wi        v0, v24, 3
> +        vse8.v          v0, (a0)
> +        add             a2, a2, a3
> +        add             a0, a0, a1
> +        bnez            a4, 1b
> +
> +        ret
> +.endm

This looks like it can be factored with a branch rather than a macro.

> +
> +func ff_put_vp8_bilin16_v_rvv, zve32x
> +        vsetivli        zero, 16, e8, m1, ta, ma
> +        put_vp8_bilin_v
> +endfunc
> +
> +func ff_put_vp8_bilin8_v_rvv, zve32x
> +        vsetivli        zero, 8, e8, mf2, ta, ma
> +        put_vp8_bilin_v
> +endfunc
> +
> +func ff_put_vp8_bilin4_v_rvv, zve32x
> +        vsetivli        zero, 4, e8, mf4, ta, ma
> +        put_vp8_bilin_v
> +endfunc


-- 
レミ・デニ-クールモン
http://www.remlab.net/