Re: [FFmpeg-devel] [PATCH] libswresample/riscv:add RVV optimized for conv_flt_to_s16

["Rémi Denis-Courmont" <remi@remlab.net>]

Le torstaina 20. maaliskuuta 2025, 11.30.01 UTC+2 daichengrong@iscas.ac.cn a 
écrit :
> From: daichengrong <daichengrong@iscas.ac.cn>
> 
> This patch introduces RVV optimized for conv_flt_to_s16.
> On Banana PI F3, it gets an average improvement of 5% for 20000 SAMPLES.
> ---
>  libswresample/audioconvert.c             |  2 +
>  libswresample/riscv/Makefile             |  3 ++
>  libswresample/riscv/audio_convert_init.c | 50 ++++++++++++++++++++++++
>  libswresample/riscv/audio_convert_rvv.S  | 46 ++++++++++++++++++++++
>  libswresample/swresample_internal.h      |  4 ++
>  5 files changed, 105 insertions(+)
>  create mode 100644 libswresample/riscv/Makefile
>  create mode 100644 libswresample/riscv/audio_convert_init.c
>  create mode 100644 libswresample/riscv/audio_convert_rvv.S
> 
> diff --git a/libswresample/audioconvert.c b/libswresample/audioconvert.c
> index 04108fb966..49b56b6b5e 100644
> --- a/libswresample/audioconvert.c
> +++ b/libswresample/audioconvert.c
> @@ -182,6 +182,8 @@ AudioConvert *swri_audio_convert_alloc(enum
> AVSampleFormat out_fmt, swri_audio_convert_init_arm(ctx, out_fmt, in_fmt,
> channels);
>  #elif ARCH_AARCH64
>      swri_audio_convert_init_aarch64(ctx, out_fmt, in_fmt, channels);
> +#elif ARCH_RISCV
> +    swri_audio_convert_init_riscv(ctx, out_fmt, in_fmt, channels);
>  #endif
> 
>      return ctx;
> diff --git a/libswresample/riscv/Makefile b/libswresample/riscv/Makefile
> new file mode 100644
> index 0000000000..01943cec64
> --- /dev/null
> +++ b/libswresample/riscv/Makefile
> @@ -0,0 +1,3 @@
> +OBJS                             += riscv/audio_convert_init.o
> +
> +RVV-OBJS                        += riscv/audio_convert_rvv.o
> diff --git a/libswresample/riscv/audio_convert_init.c
> b/libswresample/riscv/audio_convert_init.c new file mode 100644
> index 0000000000..7bea7e6eb4
> --- /dev/null
> +++ b/libswresample/riscv/audio_convert_init.c
> @@ -0,0 +1,50 @@
> +/*
> + * This file is part of libswresample.
> + *
> + * libswresample 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.
> + *
> + * libswresample 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 libswresample; if not, write to the Free Software
> + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301
> USA + */
> +
> +#include <stdint.h>
> +
> +#include "config.h"
> +#include "libavutil/attributes.h"
> +#include "libavutil/cpu.h"
> +#include "libavutil/riscv/cpu.h"
> +#include "libavutil/samplefmt.h"
> +#include "libswresample/swresample_internal.h"
> +#include "libswresample/audioconvert.h"
> +
> +void swri_oldapi_conv_flt_to_s16_rvv(int16_t *dst, const float *src, int
> len); +
> +static void conv_flt_to_s16_rvv(uint8_t **dst, const uint8_t **src, int
> len){ +    swri_oldapi_conv_flt_to_s16_rvv((int16_t*)*dst, (const
> float*)*src, len); +}
> +
> +av_cold void swri_audio_convert_init_riscv(struct AudioConvert *ac,
> +                                       enum AVSampleFormat out_fmt,
> +                                       enum AVSampleFormat in_fmt,
> +                                       int channels)
> +{
> +    int flags = av_get_cpu_flags();

This will cause warnings if !HAVE_RVV. Better to guard the whole function 
body.

> +
> +    ac->simd_f= NULL;

I don't suppose that this is necessary, otherwise how would it work in the 
current code base?

> +
> +#if HAVE_RVV
> +    if (flags & AV_CPU_FLAG_RVV_F32) {
> +        if(out_fmt == AV_SAMPLE_FMT_S16 && in_fmt == AV_SAMPLE_FMT_FLT ||
> out_fmt == AV_SAMPLE_FMT_S16P && in_fmt == AV_SAMPLE_FMT_FLTP) +           
> ac->simd_f = conv_flt_to_s16_rvv;
> +    }
> +#endif
> +}
> diff --git a/libswresample/riscv/audio_convert_rvv.S
> b/libswresample/riscv/audio_convert_rvv.S new file mode 100644
> index 0000000000..d9d58d6d5e
> --- /dev/null
> +++ b/libswresample/riscv/audio_convert_rvv.S
> @@ -0,0 +1,46 @@
> +/*
> + * Copyright (c) 2025 daichengrong <daichengrong@iscas.ac.cn>
> + *
> + * 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 "config.h"
> +#include "libavutil/riscv/asm.S"
> +
> +func swri_oldapi_conv_flt_to_s16_rvv, zve32f
> +        mv      t1, a0
> +        mv      t2, a1

These moves and the one at the end are unnecessary. Just use the argument 
registers directly.

> +        #mv t3, a2
> +1:      vsetvli a4,a2,e32,m8,ta,ma

Not: we use at least one space between operands.

> +        vle32.v v8,(t2)
> +        sub     a2, a2, a4
> +        li      t0, (1<<15)

You are computing the same value at each loop iteration.

> +        sext.w  t0,t0

Isn't this a no-op?

> +        fcvt.s.w        fa2, t0

Again, same input, same output at every iteration.

> +        vfmul.vf        v16, v8, fa2
> +        vfcvt.x.f.v     v8, v16

You need to initialise the round mode before you can do that.

> +        vsetvli zero,zero,e16,m4,ta,ma
> +        vnclip.wi       v16, v8, 0

Isn't VFNCVT more straightforward here?

> +        vse16.v v16,(t1)
> +        sll     a4,a4,0x1
> +        add     t1, t1, a4
> +        sll     a4, a4, 0x1
> +        add     t2, t2, a4

You can advantageously use Zba instructions here.

> +        bnez    a2, 1b
> +        mv      a0, t1
> +        ret
> +endfunc
> \ No newline at end of file
> diff --git a/libswresample/swresample_internal.h
> b/libswresample/swresample_internal.h index 7e46b16fb2..257f69f6dd 100644
> --- a/libswresample/swresample_internal.h
> +++ b/libswresample/swresample_internal.h
> @@ -216,5 +216,9 @@ void swri_audio_convert_init_x86(struct AudioConvert
> *ac, enum AVSampleFormat out_fmt,
>                                   enum AVSampleFormat in_fmt,
>                                   int channels);
> +void swri_audio_convert_init_riscv(struct AudioConvert *ac,
> +                                 enum AVSampleFormat out_fmt,
> +                                 enum AVSampleFormat in_fmt,
> +                                 int channels);
> 
>  #endif


-- 
Rémi Denis-Courmont
Tapiolan uusi kaupunki, Uudenmaan entinen Suomen tasavalta



_______________________________________________
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".