[FFmpeg-devel] [PATCH 02/10] lavc/vp8dsp: R-V V put_bilin_h

[Rémi Denis-Courmont]

Le sunnuntaina 5. toukokuuta 2024, 19.45.28 EEST uk7b at foxmail.com a écrit :
> From: sunyuechi <sunyuechi at iscas.ac.cn>
> 
> C908:
> vp8_put_bilin4_h_c: 373.5
> vp8_put_bilin4_h_rvv_i32: 158.7
> vp8_put_bilin8_h_c: 1437.7
> vp8_put_bilin8_h_rvv_i32: 318.7
> vp8_put_bilin16_h_c: 2845.7
> vp8_put_bilin16_h_rvv_i32: 374.7
> ---
>  libavcodec/riscv/vp8dsp_init.c | 14 +++++++++++
>  libavcodec/riscv/vp8dsp_rvv.S  | 45 ++++++++++++++++++++++++++++++++++
>  2 files changed, 59 insertions(+)
> 
> diff --git a/libavcodec/riscv/vp8dsp_init.c b/libavcodec/riscv/vp8dsp_init.c
> index fa3feeacf7..778d5ceb29 100644
> --- a/libavcodec/riscv/vp8dsp_init.c
> +++ b/libavcodec/riscv/vp8dsp_init.c
> @@ -34,6 +34,10 @@ VP8_EPEL(16, rvi);
>  VP8_EPEL(8,  rvi);
>  VP8_EPEL(4,  rvi);
> 
> +VP8_BILIN(16, rvv);
> +VP8_BILIN(8,  rvv);
> +VP8_BILIN(4,  rvv);
> +
>  av_cold void ff_vp78dsp_init_riscv(VP8DSPContext *c)
>  {
>  #if HAVE_RV
> @@ -48,6 +52,16 @@ av_cold void ff_vp78dsp_init_riscv(VP8DSPContext *c)
>          c->put_vp8_epel_pixels_tab[2][0][0] = ff_put_vp8_pixels4_rvi;
>          c->put_vp8_bilinear_pixels_tab[2][0][0] = ff_put_vp8_pixels4_rvi;
>      }
> +#if HAVE_RVV
> +    if (flags & AV_CPU_FLAG_RVV_I32 && ff_get_rv_vlenb() >= 16) {
> +        c->put_vp8_bilinear_pixels_tab[0][0][1] = ff_put_vp8_bilin16_h_rvv;
> +        c->put_vp8_bilinear_pixels_tab[0][0][2] =
> ff_put_vp8_bilin16_h_rvv; +        c->put_vp8_bilinear_pixels_tab[1][0][1]
> = ff_put_vp8_bilin8_h_rvv; +        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; +    }
> +#endif
>  #endif
>  }
> 
> diff --git a/libavcodec/riscv/vp8dsp_rvv.S b/libavcodec/riscv/vp8dsp_rvv.S
> index 8a0773f964..760d9d3871 100644
> --- a/libavcodec/riscv/vp8dsp_rvv.S
> +++ b/libavcodec/riscv/vp8dsp_rvv.S
> @@ -20,6 +20,18 @@
> 
>  #include "libavutil/riscv/asm.S"
> 
> +.macro vsetvlstatic8 len
> +.if \len <= 4
> +        vsetivli        zero, \len, e8, mf4, ta, ma
> +.elseif \len <= 8
> +        vsetivli        zero, \len, e8, mf2, ta, ma
> +.elseif \len <= 16
> +        vsetivli        zero, \len, e8, m1, ta, ma
> +.elseif \len <= 31
> +        vsetivli        zero, \len, e8, m2, ta, ma
> +.endif
> +.endm
> +
>  .macro vp8_idct_dc_add
>          vlse32.v      v0, (a0), a2
>          lh            a5, 0(a1)
> @@ -71,3 +83,36 @@ func ff_vp8_idct_dc_add4uv_rvv, zve32x
> 
>          ret
>  endfunc
> +
> +.macro bilin_h_load dst len
> +        vsetvlstatic8   \len + 1
> +        vle8.v          \dst, (a2)
> +        vslide1down.vx  v2, \dst, t5
> +        vsetvlstatic8   \len

Doesn't this effectively discard the last element, t5?
Can't we skip the slide and just load the vector at a2+1? Also then, we can 
keep VL=len and halve the multipler.

> +        vwmulu.vx       v28, \dst, t1
> +        vwmaccu.vx      v28, a5, v2
> +        vwaddu.wx       v24, v28, t4
> +        vnsra.wi        \dst, v24, 3
> +.endm
> +
> +.macro put_vp8_bilin_h len
> +func ff_put_vp8_bilin\len\()_h_rvv, zve32x
> +        li              t1, 8
> +        li              t4, 4
> +        li              t5, 1
> +        sub             t1, t1, a5
> +1:
> +        addi            a4, a4, -1
> +        bilin_h_load    v0, \len
> +        vse8.v          v0, (a0)
> +        add             a2, a2, a3
> +        add             a0, a0, a1
> +        bnez            a4, 1b
> +
> +        ret
> +endfunc
> +.endm
> +
> +.irp len 16,8,4
> +put_vp8_bilin_h \len
> +.endr


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