diff mbox series

[FFmpeg-devel,v3,1/9] lavc/vp9dsp: R-V ipred vert

Message ID tencent_912C61313C2FAA5C16254339C567733E5808@qq.com
State New
Headers show
Series [FFmpeg-devel,v3,1/9] lavc/vp9dsp: R-V ipred vert | expand

Checks

Context Check Description
andriy/make_x86 success Make finished
andriy/make_fate_x86 success Make fate finished

Commit Message

uk7b@foxmail.com May 13, 2024, 4:59 p.m. UTC
From: sunyuechi <sunyuechi@iscas.ac.cn>

C908:
vp9_vert_8x8_8bpp_c: 22.0
vp9_vert_8x8_8bpp_rvi: 15.7
vp9_vert_16x16_8bpp_c: 71.2
vp9_vert_16x16_8bpp_rvi: 39.0
vp9_vert_32x32_8bpp_c: 300.2
vp9_vert_32x32_8bpp_rvi: 135.2
---
 libavcodec/riscv/Makefile        |  1 +
 libavcodec/riscv/vp9_intra_rvi.S | 71 ++++++++++++++++++++++++++++++++
 libavcodec/riscv/vp9dsp.h        |  6 +++
 libavcodec/riscv/vp9dsp_init.c   | 15 +++++--
 4 files changed, 90 insertions(+), 3 deletions(-)
 create mode 100644 libavcodec/riscv/vp9_intra_rvi.S

Comments

flow gg May 13, 2024, 5:01 p.m. UTC | #1
just rebase

<uk7b@foxmail.com> 于2024年5月14日周二 01:00写道:

> From: sunyuechi <sunyuechi@iscas.ac.cn>
>
> C908:
> vp9_vert_8x8_8bpp_c: 22.0
> vp9_vert_8x8_8bpp_rvi: 15.7
> vp9_vert_16x16_8bpp_c: 71.2
> vp9_vert_16x16_8bpp_rvi: 39.0
> vp9_vert_32x32_8bpp_c: 300.2
> vp9_vert_32x32_8bpp_rvi: 135.2
> ---
>  libavcodec/riscv/Makefile        |  1 +
>  libavcodec/riscv/vp9_intra_rvi.S | 71 ++++++++++++++++++++++++++++++++
>  libavcodec/riscv/vp9dsp.h        |  6 +++
>  libavcodec/riscv/vp9dsp_init.c   | 15 +++++--
>  4 files changed, 90 insertions(+), 3 deletions(-)
>  create mode 100644 libavcodec/riscv/vp9_intra_rvi.S
>
> diff --git a/libavcodec/riscv/Makefile b/libavcodec/riscv/Makefile
> index 89273b1cad..ccd060c666 100644
> --- a/libavcodec/riscv/Makefile
> +++ b/libavcodec/riscv/Makefile
> @@ -62,6 +62,7 @@ OBJS-$(CONFIG_VP8DSP) += riscv/vp8dsp_init.o
>  RV-OBJS-$(CONFIG_VP8DSP) += riscv/vp8dsp_rvi.o
>  RVV-OBJS-$(CONFIG_VP8DSP) += riscv/vp8dsp_rvv.o
>  OBJS-$(CONFIG_VP9_DECODER) += riscv/vp9dsp_init.o
> +RV-OBJS-$(CONFIG_VP9_DECODER) += riscv/vp9_intra_rvi.o
>  RVV-OBJS-$(CONFIG_VP9_DECODER) += riscv/vp9_intra_rvv.o
>  OBJS-$(CONFIG_VORBIS_DECODER) += riscv/vorbisdsp_init.o
>  RVV-OBJS-$(CONFIG_VORBIS_DECODER) += riscv/vorbisdsp_rvv.o
> diff --git a/libavcodec/riscv/vp9_intra_rvi.S
> b/libavcodec/riscv/vp9_intra_rvi.S
> new file mode 100644
> index 0000000000..16b6bdb25a
> --- /dev/null
> +++ b/libavcodec/riscv/vp9_intra_rvi.S
> @@ -0,0 +1,71 @@
> +/*
> + * Copyright (c) 2024 Institue of Software Chinese Academy of Sciences
> (ISCAS).
> + *
> + * This file is part of FFmpeg.
> + *
> + * FFmpeg is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU Lesser General Public
> + * License as published by the Free Software Foundation; either
> + * version 2.1 of the License, or (at your option) any later version.
> + *
> + * FFmpeg is distributed in the hope that it will be useful,
> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> + * Lesser General Public License for more details.
> + *
> + * You should have received a copy of the GNU Lesser General Public
> + * License along with FFmpeg; if not, write to the Free Software
> + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
> 02110-1301 USA
> + */
> +
> +#include "libavutil/riscv/asm.S"
> +
> +#if __riscv_xlen >= 64
> +func ff_v_32x32_rvi
> +        ld           t0, (a3)
> +        ld           t1, 8(a3)
> +        ld           t2, 16(a3)
> +        ld           t3, 24(a3)
> +        .rept 16
> +        add          a7, a0, a1
> +        sd           t0, (a0)
> +        sd           t1, 8(a0)
> +        sd           t2, 16(a0)
> +        sd           t3, 24(a0)
> +        sh1add       a0, a1, a0
> +        sd           t0, (a7)
> +        sd           t1, 8(a7)
> +        sd           t2, 16(a7)
> +        sd           t3, 24(a7)
> +        .endr
> +
> +        ret
> +endfunc
> +
> +func ff_v_16x16_rvi
> +        ld           t0, (a3)
> +        ld           t1, 8(a3)
> +        .rept 8
> +        add          a7, a0, a1
> +        sd           t0, (a0)
> +        sd           t1, 8(a0)
> +        sh1add       a0, a1, a0
> +        sd           t0, (a7)
> +        sd           t1, 8(a7)
> +        .endr
> +
> +        ret
> +endfunc
> +
> +func ff_v_8x8_rvi
> +        ld           t0, (a3)
> +        .rept 4
> +        add          a7, a0, a1
> +        sd           t0, (a0)
> +        sh1add       a0, a1, a0
> +        sd           t0, (a7)
> +        .endr
> +
> +        ret
> +endfunc
> +#endif
> diff --git a/libavcodec/riscv/vp9dsp.h b/libavcodec/riscv/vp9dsp.h
> index 25047ed507..f8bc6563a5 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_rvi(uint8_t *dst, ptrdiff_t stride, const uint8_t *l,
> +                    const uint8_t *a);
> +void ff_v_16x16_rvi(uint8_t *dst, ptrdiff_t stride, const uint8_t *l,
> +                    const uint8_t *a);
> +void ff_v_8x8_rvi(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 dd418bd5bf..0f64afc6d2 100644
> --- a/libavcodec/riscv/vp9dsp_init.c
> +++ b/libavcodec/riscv/vp9dsp_init.c
> @@ -24,11 +24,19 @@
>  #include "libavcodec/vp9dsp.h"
>  #include "vp9dsp.h"
>
> -static av_cold void vp9dsp_intrapred_init_rvv(VP9DSPContext *dsp, int bpp)
> +static av_cold void vp9dsp_intrapred_init_riscv(VP9DSPContext *dsp, int
> bpp)
>  {
> -#if HAVE_RVV
> +#if HAVE_RV
>      int flags = av_get_cpu_flags();
>
> +    if (bpp == 8 && (flags & AV_CPU_FLAG_RV_MISALIGNED) && (flags &
> AV_CPU_FLAG_RVB_ADDR)) {
> +# if __riscv_xlen >= 64
> +        dsp->intra_pred[TX_32X32][VERT_PRED] = ff_v_32x32_rvi;
> +        dsp->intra_pred[TX_16X16][VERT_PRED] = ff_v_16x16_rvi;
> +        dsp->intra_pred[TX_8X8][VERT_PRED] = ff_v_8x8_rvi;
> +# endif
> +    }
> +#if HAVE_RVV
>      if (bpp == 8 && flags & AV_CPU_FLAG_RVV_I64 && ff_rv_vlen_least(128))
> {
>          dsp->intra_pred[TX_8X8][DC_PRED] = ff_dc_8x8_rvv;
>          dsp->intra_pred[TX_8X8][LEFT_DC_PRED] = ff_dc_left_8x8_rvv;
> @@ -53,9 +61,10 @@ static av_cold void
> vp9dsp_intrapred_init_rvv(VP9DSPContext *dsp, int bpp)
>          dsp->intra_pred[TX_16X16][TOP_DC_PRED] = ff_dc_top_16x16_rvv;
>      }
>  #endif
> +#endif
>  }
>
>  av_cold void ff_vp9dsp_init_riscv(VP9DSPContext *dsp, int bpp, int
> bitexact)
>  {
> -    vp9dsp_intrapred_init_rvv(dsp, bpp);
> +    vp9dsp_intrapred_init_riscv(dsp, bpp);
>  }
> --
> 2.45.0
>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel@ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
>
Rémi Denis-Courmont May 13, 2024, 7:47 p.m. UTC | #2
Le maanantaina 13. toukokuuta 2024, 19.59.18 EEST uk7b@foxmail.com a écrit :
> From: sunyuechi <sunyuechi@iscas.ac.cn>
> 
> C908:
> vp9_vert_8x8_8bpp_c: 22.0
> vp9_vert_8x8_8bpp_rvi: 15.7
> vp9_vert_16x16_8bpp_c: 71.2
> vp9_vert_16x16_8bpp_rvi: 39.0
> vp9_vert_32x32_8bpp_c: 300.2
> vp9_vert_32x32_8bpp_rvi: 135.2
> ---
>  libavcodec/riscv/Makefile        |  1 +
>  libavcodec/riscv/vp9_intra_rvi.S | 71 ++++++++++++++++++++++++++++++++
>  libavcodec/riscv/vp9dsp.h        |  6 +++
>  libavcodec/riscv/vp9dsp_init.c   | 15 +++++--
>  4 files changed, 90 insertions(+), 3 deletions(-)
>  create mode 100644 libavcodec/riscv/vp9_intra_rvi.S
> 
> diff --git a/libavcodec/riscv/Makefile b/libavcodec/riscv/Makefile
> index 89273b1cad..ccd060c666 100644
> --- a/libavcodec/riscv/Makefile
> +++ b/libavcodec/riscv/Makefile
> @@ -62,6 +62,7 @@ OBJS-$(CONFIG_VP8DSP) += riscv/vp8dsp_init.o
>  RV-OBJS-$(CONFIG_VP8DSP) += riscv/vp8dsp_rvi.o
>  RVV-OBJS-$(CONFIG_VP8DSP) += riscv/vp8dsp_rvv.o
>  OBJS-$(CONFIG_VP9_DECODER) += riscv/vp9dsp_init.o
> +RV-OBJS-$(CONFIG_VP9_DECODER) += riscv/vp9_intra_rvi.o
>  RVV-OBJS-$(CONFIG_VP9_DECODER) += riscv/vp9_intra_rvv.o
>  OBJS-$(CONFIG_VORBIS_DECODER) += riscv/vorbisdsp_init.o
>  RVV-OBJS-$(CONFIG_VORBIS_DECODER) += riscv/vorbisdsp_rvv.o
> diff --git a/libavcodec/riscv/vp9_intra_rvi.S
> b/libavcodec/riscv/vp9_intra_rvi.S new file mode 100644
> index 0000000000..16b6bdb25a
> --- /dev/null
> +++ b/libavcodec/riscv/vp9_intra_rvi.S
> @@ -0,0 +1,71 @@
> +/*
> + * Copyright (c) 2024 Institue of Software Chinese Academy of Sciences
> (ISCAS). + *
> + * This file is part of FFmpeg.
> + *
> + * FFmpeg is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU Lesser General Public
> + * License as published by the Free Software Foundation; either
> + * version 2.1 of the License, or (at your option) any later version.
> + *
> + * FFmpeg is distributed in the hope that it will be useful,
> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> + * Lesser General Public License for more details.
> + *
> + * You should have received a copy of the GNU Lesser General Public
> + * License along with FFmpeg; if not, write to the Free Software
> + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301
> USA + */
> +
> +#include "libavutil/riscv/asm.S"
> +
> +#if __riscv_xlen >= 64
> +func ff_v_32x32_rvi
> +        ld           t0, (a3)
> +        ld           t1, 8(a3)
> +        ld           t2, 16(a3)
> +        ld           t3, 24(a3)
> +        .rept 16
> +        add          a7, a0, a1
> +        sd           t0, (a0)
> +        sd           t1, 8(a0)
> +        sd           t2, 16(a0)
> +        sd           t3, 24(a0)
> +        sh1add       a0, a1, a0
> +        sd           t0, (a7)
> +        sd           t1, 8(a7)
> +        sd           t2, 16(a7)
> +        sd           t3, 24(a7)
> +        .endr
> +
> +        ret
> +endfunc
> +
> +func ff_v_16x16_rvi
> +        ld           t0, (a3)
> +        ld           t1, 8(a3)
> +        .rept 8
> +        add          a7, a0, a1
> +        sd           t0, (a0)
> +        sd           t1, 8(a0)
> +        sh1add       a0, a1, a0
> +        sd           t0, (a7)
> +        sd           t1, 8(a7)
> +        .endr
> +
> +        ret
> +endfunc
> +
> +func ff_v_8x8_rvi
> +        ld           t0, (a3)
> +        .rept 4
> +        add          a7, a0, a1
> +        sd           t0, (a0)
> +        sh1add       a0, a1, a0
> +        sd           t0, (a7)
> +        .endr
> +
> +        ret
> +endfunc
> +#endif
> diff --git a/libavcodec/riscv/vp9dsp.h b/libavcodec/riscv/vp9dsp.h
> index 25047ed507..f8bc6563a5 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_rvi(uint8_t *dst, ptrdiff_t stride, const uint8_t *l,
> +                    const uint8_t *a);
> +void ff_v_16x16_rvi(uint8_t *dst, ptrdiff_t stride, const uint8_t *l,
> +                    const uint8_t *a);
> +void ff_v_8x8_rvi(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 dd418bd5bf..0f64afc6d2 100644
> --- a/libavcodec/riscv/vp9dsp_init.c
> +++ b/libavcodec/riscv/vp9dsp_init.c
> @@ -24,11 +24,19 @@
>  #include "libavcodec/vp9dsp.h"
>  #include "vp9dsp.h"
> 
> -static av_cold void vp9dsp_intrapred_init_rvv(VP9DSPContext *dsp, int bpp)
> +static av_cold void vp9dsp_intrapred_init_riscv(VP9DSPContext *dsp, int
> bpp) {
> -#if HAVE_RVV
> +#if HAVE_RV
>      int flags = av_get_cpu_flags();
> 
> +    if (bpp == 8 && (flags & AV_CPU_FLAG_RV_MISALIGNED) && (flags &
> AV_CPU_FLAG_RVB_ADDR)) { +# if __riscv_xlen >= 64
> +        dsp->intra_pred[TX_32X32][VERT_PRED] = ff_v_32x32_rvi;
> +        dsp->intra_pred[TX_16X16][VERT_PRED] = ff_v_16x16_rvi;
> +        dsp->intra_pred[TX_8X8][VERT_PRED] = ff_v_8x8_rvi;
> +# endif

AFAICT, those are aligned accesses? At least in checkasm, they are?

> +    }
> +#if HAVE_RVV
>      if (bpp == 8 && flags & AV_CPU_FLAG_RVV_I64 && ff_rv_vlen_least(128)) {
> dsp->intra_pred[TX_8X8][DC_PRED] = ff_dc_8x8_rvv;
>          dsp->intra_pred[TX_8X8][LEFT_DC_PRED] = ff_dc_left_8x8_rvv;
> @@ -53,9 +61,10 @@ static av_cold void
> vp9dsp_intrapred_init_rvv(VP9DSPContext *dsp, int bpp)
> dsp->intra_pred[TX_16X16][TOP_DC_PRED] = ff_dc_top_16x16_rvv; }
>  #endif
> +#endif
>  }
> 
>  av_cold void ff_vp9dsp_init_riscv(VP9DSPContext *dsp, int bpp, int
> bitexact) {
> -    vp9dsp_intrapred_init_rvv(dsp, bpp);
> +    vp9dsp_intrapred_init_riscv(dsp, bpp);
>  }
Rémi Denis-Courmont May 13, 2024, 7:53 p.m. UTC | #3
Le maanantaina 13. toukokuuta 2024, 19.59.18 EEST uk7b@foxmail.com a écrit :
> From: sunyuechi <sunyuechi@iscas.ac.cn>
> 
> C908:
> vp9_vert_8x8_8bpp_c: 22.0
> vp9_vert_8x8_8bpp_rvi: 15.7
> vp9_vert_16x16_8bpp_c: 71.2
> vp9_vert_16x16_8bpp_rvi: 39.0
> vp9_vert_32x32_8bpp_c: 300.2
> vp9_vert_32x32_8bpp_rvi: 135.2

Not sure how you get that. It should be reported under rvb_a (Zba) or 
misaligned, AFAIU.
flow gg May 14, 2024, 4:45 a.m. UTC | #4
I am locally using:
    if (bpp == 8 && (flags & AV_CPU_FLAG_RVI) && (flags &
AV_CPU_FLAG_RVB_ADDR)) {
this performs better on k230/banana_f3 than C.
For email, refer to [FFmpeg-devel] [PATCH 2/2] lavc/vp8dsp: restrict RVI
optimisations and change it to
    if (bpp == 8 && (flags & AV_CPU_FLAG_RV_MISALIGNED) && (flags &
AV_CPU_FLAG_RVB_ADDR)) {
So no output, but I think the same modification should be made here?

Rémi Denis-Courmont <remi@remlab.net> 于2024年5月14日周二 03:54写道:

> Le maanantaina 13. toukokuuta 2024, 19.59.18 EEST uk7b@foxmail.com a
> écrit :
> > From: sunyuechi <sunyuechi@iscas.ac.cn>
> >
> > C908:
> > vp9_vert_8x8_8bpp_c: 22.0
> > vp9_vert_8x8_8bpp_rvi: 15.7
> > vp9_vert_16x16_8bpp_c: 71.2
> > vp9_vert_16x16_8bpp_rvi: 39.0
> > vp9_vert_32x32_8bpp_c: 300.2
> > vp9_vert_32x32_8bpp_rvi: 135.2
>
> Not sure how you get that. It should be reported under rvb_a (Zba) or
> misaligned, AFAIU.
>
> --
> レミ・デニ-クールモン
> 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-request@ffmpeg.org with subject "unsubscribe".
>
Rémi Denis-Courmont May 14, 2024, 5 p.m. UTC | #5
Le tiistaina 14. toukokuuta 2024, 7.45.29 EEST flow gg a écrit :
> I am locally using:
>     if (bpp == 8 && (flags & AV_CPU_FLAG_RVI) && (flags &
> AV_CPU_FLAG_RVB_ADDR)) {

There is no point testing the I flag if you test any other flag. The I flag is 
always set (since we don't, and probably never will, support RV32E) and only 
exists for the benefit of checkasm.

> this performs better on k230/banana_f3 than C.

It also performs better than C on SiFive U74, even though that design has 
veeeeeery slow unaligned access (emulated in SBI). Of course, it could just be 
that checkasm only tests aligned accesses and unaligned accesses are legal, 
hence my earlier question.
flow gg May 14, 2024, 5:22 p.m. UTC | #6
Okay, learned it

Rémi Denis-Courmont <remi@remlab.net> 于2024年5月15日周三 01:00写道:

> Le tiistaina 14. toukokuuta 2024, 7.45.29 EEST flow gg a écrit :
> > I am locally using:
> >     if (bpp == 8 && (flags & AV_CPU_FLAG_RVI) && (flags &
> > AV_CPU_FLAG_RVB_ADDR)) {
>
> There is no point testing the I flag if you test any other flag. The I
> flag is
> always set (since we don't, and probably never will, support RV32E) and
> only
> exists for the benefit of checkasm.
>
> > this performs better on k230/banana_f3 than C.
>
> It also performs better than C on SiFive U74, even though that design has
> veeeeeery slow unaligned access (emulated in SBI). Of course, it could
> just be
> that checkasm only tests aligned accesses and unaligned accesses are
> legal,
> hence my earlier question.
>
> --
> レミ・デニ-クールモン
> 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-request@ffmpeg.org with subject "unsubscribe".
>
diff mbox series

Patch

diff --git a/libavcodec/riscv/Makefile b/libavcodec/riscv/Makefile
index 89273b1cad..ccd060c666 100644
--- a/libavcodec/riscv/Makefile
+++ b/libavcodec/riscv/Makefile
@@ -62,6 +62,7 @@  OBJS-$(CONFIG_VP8DSP) += riscv/vp8dsp_init.o
 RV-OBJS-$(CONFIG_VP8DSP) += riscv/vp8dsp_rvi.o
 RVV-OBJS-$(CONFIG_VP8DSP) += riscv/vp8dsp_rvv.o
 OBJS-$(CONFIG_VP9_DECODER) += riscv/vp9dsp_init.o
+RV-OBJS-$(CONFIG_VP9_DECODER) += riscv/vp9_intra_rvi.o
 RVV-OBJS-$(CONFIG_VP9_DECODER) += riscv/vp9_intra_rvv.o
 OBJS-$(CONFIG_VORBIS_DECODER) += riscv/vorbisdsp_init.o
 RVV-OBJS-$(CONFIG_VORBIS_DECODER) += riscv/vorbisdsp_rvv.o
diff --git a/libavcodec/riscv/vp9_intra_rvi.S b/libavcodec/riscv/vp9_intra_rvi.S
new file mode 100644
index 0000000000..16b6bdb25a
--- /dev/null
+++ b/libavcodec/riscv/vp9_intra_rvi.S
@@ -0,0 +1,71 @@ 
+/*
+ * Copyright (c) 2024 Institue of Software Chinese Academy of Sciences (ISCAS).
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/riscv/asm.S"
+
+#if __riscv_xlen >= 64
+func ff_v_32x32_rvi
+        ld           t0, (a3)
+        ld           t1, 8(a3)
+        ld           t2, 16(a3)
+        ld           t3, 24(a3)
+        .rept 16
+        add          a7, a0, a1
+        sd           t0, (a0)
+        sd           t1, 8(a0)
+        sd           t2, 16(a0)
+        sd           t3, 24(a0)
+        sh1add       a0, a1, a0
+        sd           t0, (a7)
+        sd           t1, 8(a7)
+        sd           t2, 16(a7)
+        sd           t3, 24(a7)
+        .endr
+
+        ret
+endfunc
+
+func ff_v_16x16_rvi
+        ld           t0, (a3)
+        ld           t1, 8(a3)
+        .rept 8
+        add          a7, a0, a1
+        sd           t0, (a0)
+        sd           t1, 8(a0)
+        sh1add       a0, a1, a0
+        sd           t0, (a7)
+        sd           t1, 8(a7)
+        .endr
+
+        ret
+endfunc
+
+func ff_v_8x8_rvi
+        ld           t0, (a3)
+        .rept 4
+        add          a7, a0, a1
+        sd           t0, (a0)
+        sh1add       a0, a1, a0
+        sd           t0, (a7)
+        .endr
+
+        ret
+endfunc
+#endif
diff --git a/libavcodec/riscv/vp9dsp.h b/libavcodec/riscv/vp9dsp.h
index 25047ed507..f8bc6563a5 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_rvi(uint8_t *dst, ptrdiff_t stride, const uint8_t *l,
+                    const uint8_t *a);
+void ff_v_16x16_rvi(uint8_t *dst, ptrdiff_t stride, const uint8_t *l,
+                    const uint8_t *a);
+void ff_v_8x8_rvi(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 dd418bd5bf..0f64afc6d2 100644
--- a/libavcodec/riscv/vp9dsp_init.c
+++ b/libavcodec/riscv/vp9dsp_init.c
@@ -24,11 +24,19 @@ 
 #include "libavcodec/vp9dsp.h"
 #include "vp9dsp.h"
 
-static av_cold void vp9dsp_intrapred_init_rvv(VP9DSPContext *dsp, int bpp)
+static av_cold void vp9dsp_intrapred_init_riscv(VP9DSPContext *dsp, int bpp)
 {
-#if HAVE_RVV
+#if HAVE_RV
     int flags = av_get_cpu_flags();
 
+    if (bpp == 8 && (flags & AV_CPU_FLAG_RV_MISALIGNED) && (flags & AV_CPU_FLAG_RVB_ADDR)) {
+# if __riscv_xlen >= 64
+        dsp->intra_pred[TX_32X32][VERT_PRED] = ff_v_32x32_rvi;
+        dsp->intra_pred[TX_16X16][VERT_PRED] = ff_v_16x16_rvi;
+        dsp->intra_pred[TX_8X8][VERT_PRED] = ff_v_8x8_rvi;
+# endif
+    }
+#if HAVE_RVV
     if (bpp == 8 && flags & AV_CPU_FLAG_RVV_I64 && ff_rv_vlen_least(128)) {
         dsp->intra_pred[TX_8X8][DC_PRED] = ff_dc_8x8_rvv;
         dsp->intra_pred[TX_8X8][LEFT_DC_PRED] = ff_dc_left_8x8_rvv;
@@ -53,9 +61,10 @@  static av_cold void vp9dsp_intrapred_init_rvv(VP9DSPContext *dsp, int bpp)
         dsp->intra_pred[TX_16X16][TOP_DC_PRED] = ff_dc_top_16x16_rvv;
     }
 #endif
+#endif
 }
 
 av_cold void ff_vp9dsp_init_riscv(VP9DSPContext *dsp, int bpp, int bitexact)
 {
-    vp9dsp_intrapred_init_rvv(dsp, bpp);
+    vp9dsp_intrapred_init_riscv(dsp, bpp);
 }