Le lauantaina 4. toukokuuta 2024, 18.03.04 EEST u...@foxmail.com a écrit :
> From: sunyuechi <sunyue...@iscas.ac.cn>
>
> C908:
> vp9_vert_8x8_8bpp_c: 22.0
> vp9_vert_8x8_8bpp_rvv_i64: 18.5
> vp9_vert_16x16_8bpp_c: 71.2
> vp9_vert_16x16_8bpp_rvv_i32: 50.7
> vp9_vert_32x32_8bpp_c: 300.2
> vp9_vert_32x32_8bpp_rvv_i32: 136.7
> ---
> libavcodec/riscv/vp9_intra_rvv.S | 35 ++++++++++++++++++++++++++++++++
> libavcodec/riscv/vp9dsp.h | 6 ++++++
> libavcodec/riscv/vp9dsp_init.c | 3 +++
> 3 files changed, 44 insertions(+)
>
> diff --git a/libavcodec/riscv/vp9_intra_rvv.S
> b/libavcodec/riscv/vp9_intra_rvv.S index db9774c263..b5f0f9d3c3 100644
> --- a/libavcodec/riscv/vp9_intra_rvv.S
> +++ b/libavcodec/riscv/vp9_intra_rvv.S
> @@ -113,3 +113,38 @@ func_dc dc_left 8 left 3 0 zve64x
> func_dc dc_top 32 top 5 1 zve32x
> func_dc dc_top 16 top 4 1 zve32x
> func_dc dc_top 8 top 3 0 zve64x
> +
> +func ff_v_32x32_rvv, zve32x
> + vsetivli zero, 8, e8, mf2, ta, ma
> + vle32.v v8, (a3)
> +
> + .rept 31
> + vse32.v v8, (a0)
> + add a0, a0, a1
> + .endr
> + vse32.v v8, (a0)
> +
> + ret
> +endfunc
This does not look like it needs vectors. Ditto the next one.
> +
> +func ff_v_16x16_rvv, zve32x
> + vsetivli zero, 4, e8, mf4, ta, ma
> + vle32.v v8, (a3)
> +
> + .rept 15
> + vse32.v v8, (a0)
> + add a0, a0, a1
> + .endr
> + vse32.v v8, (a0)
> +
> + ret
> +endfunc
> +
> +func ff_v_8x8_rvv, zve64x
> + ld t0, (a3)
> + vsetivli zero, 8, e64, m4, ta, ma
> + vmv.v.x v8, t0
> + vsse64.v v8, (a0), a1
> +
> + ret
> +endfunc
This is missing a guard against RV32.
> diff --git a/libavcodec/riscv/vp9dsp.h b/libavcodec/riscv/vp9dsp.h
> index 25047ed507..113397ce86 100644
> --- a/libavcodec/riscv/vp9dsp.h
> +++ b/libavcodec/riscv/vp9dsp.h
> @@ -60,6 +60,12 @@ void ff_dc_129_16x16_rvv(uint8_t *dst, ptrdiff_t stride,
> const uint8_t *l, const uint8_t *a);
> void ff_dc_129_8x8_rvv(uint8_t *dst, ptrdiff_t stride, const uint8_t *l,
> const uint8_t *a);
> +void ff_v_32x32_rvv(uint8_t *dst, ptrdiff_t stride, const uint8_t *l,
> + const uint8_t *a);
> +void ff_v_16x16_rvv(uint8_t *dst, ptrdiff_t stride, const uint8_t *l,
> + const uint8_t *a);
> +void ff_v_8x8_rvv(uint8_t *dst, ptrdiff_t stride, const uint8_t *l,
> + const uint8_t *a);
>
> #define VP9_8TAP_RISCV_RVV_FUNC(SIZE, type, type_idx)
> \ void ff_put_8tap_##type##_##SIZE##h_rvv(uint8_t *dst, ptrdiff_t
> dststride, \ diff --git a/libavcodec/riscv/vp9dsp_init.c
> b/libavcodec/riscv/vp9dsp_init.c index 69ab39004c..9c550d40b5 100644
> --- a/libavcodec/riscv/vp9dsp_init.c
> +++ b/libavcodec/riscv/vp9dsp_init.c
> @@ -36,6 +36,7 @@ static av_cold void
> vp9dsp_intrapred_init_rvv(VP9DSPContext *dsp, int bpp)
> dsp->intra_pred[TX_8X8][DC_128_PRED] = ff_dc_128_8x8_rvv;
> dsp->intra_pred[TX_8X8][DC_129_PRED] = ff_dc_129_8x8_rvv;
> dsp->intra_pred[TX_8X8][TOP_DC_PRED] = ff_dc_top_8x8_rvv; +
> dsp->intra_pred[TX_8X8][VERT_PRED] = ff_v_8x8_rvv;
> }
>
> if (bpp == 8 && flags & AV_CPU_FLAG_RVV_I32 && ff_get_rv_vlenb() >=
> 16) { @@ -51,6 +52,8 @@ static av_cold void
> vp9dsp_intrapred_init_rvv(VP9DSPContext *dsp, int bpp)
> dsp->intra_pred[TX_16X16][DC_129_PRED] = ff_dc_129_16x16_rvv;
> dsp->intra_pred[TX_32X32][TOP_DC_PRED] = ff_dc_top_32x32_rvv;
> dsp->intra_pred[TX_16X16][TOP_DC_PRED] = ff_dc_top_16x16_rvv; +
> dsp->intra_pred[TX_32X32][VERT_PRED] = ff_v_32x32_rvv; +
> dsp->intra_pred[TX_16X16][VERT_PRED] = ff_v_16x16_rvv; }
> #endif
> }
--
Rémi Denis-Courmont
http://www.remlab.net/
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
