[FFmpeg-devel] [PATCH] lavfi: port decimate libmpcodecs filter

Nicolas George nicolas.george at normalesup.org
Sun Aug 19 20:22:20 CEST 2012


Le tridi 3 fructidor, an CCXX, Stefano Sabatini a écrit :
> >From 7379459869bb671282e7f50d4bf7ba44c05ef0a5 Mon Sep 17 00:00:00 2001
> From: Stefano Sabatini <stefasab at gmail.com>
> Date: Sat, 17 Mar 2012 15:49:51 +0100
> Subject: [PATCH] lavfi: port decimate libmpcodecs filter
> 
> ---
>  configure                 |    1 +
>  doc/filters.texi          |   34 +++++++
>  libavfilter/Makefile      |    1 +
>  libavfilter/allfilters.c  |    1 +
>  libavfilter/diff.c        |   35 +++++++
>  libavfilter/diff.h        |   27 +++++
>  libavfilter/vf_decimate.c |  243 +++++++++++++++++++++++++++++++++++++++++++++
>  libavfilter/x86/Makefile  |    2 +
>  libavfilter/x86/diff.c    |   63 ++++++++++++
>  9 files changed, 407 insertions(+), 0 deletions(-)
>  create mode 100644 libavfilter/diff.c
>  create mode 100644 libavfilter/diff.h
>  create mode 100644 libavfilter/vf_decimate.c
>  create mode 100644 libavfilter/x86/diff.c
> 
> diff --git a/configure b/configure
> index 679fae0..e67278c 100755
> --- a/configure
> +++ b/configure
> @@ -1830,6 +1830,7 @@ blackframe_filter_deps="gpl"
>  boxblur_filter_deps="gpl"
>  colormatrix_filter_deps="gpl"
>  cropdetect_filter_deps="gpl"
> +decimate_filter_deps="gpl"
>  delogo_filter_deps="gpl"
>  deshake_filter_deps="avcodec"
>  drawtext_filter_deps="libfreetype"
> diff --git a/doc/filters.texi b/doc/filters.texi
> index 8847990..1ac88c2 100644
> --- a/doc/filters.texi
> +++ b/doc/filters.texi
> @@ -1539,6 +1539,40 @@ indicates never reset and return the largest area encountered during
>  playback.
>  @end table
>  
> + at section decimate
> +
> +This filter drops frames that do not differ greatly from the previous
> +frame in order to reduce framerate.  The main use of this filter is
> +for very-low-bitrate encoding (e.g. streaming over dialup modem), but
> +it could in theory be used for fixing movies that were
> +inverse-telecined incorrectly.
> +
> +It accepts the following parameters:
> + at var{max}:@var{hi}:@var{lo}:@var{frac}.
> +
> + at table @option
> +
> + at item max
> +Set the maximum number of consecutive frames which can be dropped (if
> +positive), or the minimum interval between dropped frames (if
> +negative). Default value is 0.
> +
> + at item hi, lo, frac
> +Set the dropping threshold values.
> +
> +Values for @var{hi} and @var{lo} are for 8x8 pixel blocks and
> +represent actual pixel value differences, so a threshold of 64
> +corresponds to 1 unit of difference for each pixel, or the same spread
> +out differently over the block.
> +
> +A frame is a candidate for dropping if no 8x8 blocks differ by more
> +than a threshold of @var{hi}, and if no more than @var{frac} blocks (1
> +meaning the whole image) differ by more than a threshold of @var{lo}.
> +
> +Default value for @var{hi} is 64*12, default value for @var{lo} is
> +64*5, and default value for @var{frac} is 0.33.
> + at end table
> +
>  @section delogo
>  
>  Suppress a TV station logo by a simple interpolation of the surrounding
> diff --git a/libavfilter/Makefile b/libavfilter/Makefile
> index af4fde6..4cae706 100644
> --- a/libavfilter/Makefile
> +++ b/libavfilter/Makefile
> @@ -87,6 +87,7 @@ OBJS-$(CONFIG_COLORMATRIX_FILTER)            += vf_colormatrix.o
>  OBJS-$(CONFIG_COPY_FILTER)                   += vf_copy.o
>  OBJS-$(CONFIG_CROP_FILTER)                   += vf_crop.o
>  OBJS-$(CONFIG_CROPDETECT_FILTER)             += vf_cropdetect.o
> +OBJS-$(CONFIG_DECIMATE_FILTER)               += vf_decimate.o diff.o
>  OBJS-$(CONFIG_DELOGO_FILTER)                 += vf_delogo.o
>  OBJS-$(CONFIG_DESHAKE_FILTER)                += vf_deshake.o
>  OBJS-$(CONFIG_DRAWBOX_FILTER)                += vf_drawbox.o
> diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c
> index 6defed4..be49e23 100644
> --- a/libavfilter/allfilters.c
> +++ b/libavfilter/allfilters.c
> @@ -78,6 +78,7 @@ void avfilter_register_all(void)
>      REGISTER_FILTER (COPY,        copy,        vf);
>      REGISTER_FILTER (CROP,        crop,        vf);
>      REGISTER_FILTER (CROPDETECT,  cropdetect,  vf);
> +    REGISTER_FILTER (DECIMATE,    decimate,    vf);
>      REGISTER_FILTER (DELOGO,      delogo,      vf);
>      REGISTER_FILTER (DESHAKE,     deshake,     vf);
>      REGISTER_FILTER (DRAWBOX,     drawbox,     vf);
> diff --git a/libavfilter/diff.c b/libavfilter/diff.c
> new file mode 100644
> index 0000000..4e748ec
> --- /dev/null
> +++ b/libavfilter/diff.c
> @@ -0,0 +1,35 @@
> +/*

> + * Copyright (c) 2003 Rich Felker
> + *
> + * 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.

The original code is under GPL, but you clearly reimplemented it. Probably
alter the copyright notice.

> + *
> + * 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 "diff.h"
> +
> +int ff_diff_8x8_c(const uint8_t *cur, int cur_linesize,
> +                  const uint8_t *ref, int ref_linesize)
> +{
> +    int x, y, d = 0;
> +

> +    for (y = 8; y; y--) {
> +        for (x = 8; x; x--)
> +            d += abs(cur[x] - ref[x]);

Looks strange: the data will be accessed at indices 1..8, instead of 0..7.
And in reverse order, which may be not good for the cache. Maybe switch to a
more natural loop?

> +        cur += cur_linesize;
> +        ref += ref_linesize;
> +    }
> +    return d;
> +}
> diff --git a/libavfilter/diff.h b/libavfilter/diff.h
> new file mode 100644
> index 0000000..c27128d
> --- /dev/null
> +++ b/libavfilter/diff.h
> @@ -0,0 +1,27 @@
> +/*
> + * 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
> + */
> +
> +#ifndef AVFILTER_DIFF_H
> +#define AVFILTER_DIFF_H
> +
> +#include "avfilter.h"
> +
> +int ff_diff_8x8_c  (const uint8_t *cur, int cur_linesize, const uint8_t *ref, int ref_linesize);
> +int ff_diff_8x8_mmx(const uint8_t *cur, int cur_linesize, const uint8_t *ref, int ref_linesize);
> +
> +#endif /* AVFILTER_DIFF_H */
> diff --git a/libavfilter/vf_decimate.c b/libavfilter/vf_decimate.c
> new file mode 100644
> index 0000000..d94eed6
> --- /dev/null
> +++ b/libavfilter/vf_decimate.c
> @@ -0,0 +1,243 @@
> +/*
> + * Copyright (c) 2003 Rich Felker
> + * Copyright (c) 2012 Stefano Sabatini
> + *
> + * This file is part of FFmpeg.
> + *
> + * FFmpeg is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License as published by
> + * the Free Software Foundation; either version 2 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 General Public License for more details.
> + *
> + * You should have received a copy of the GNU 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/pixdesc.h"
> +#include "libavutil/timestamp.h"
> +#include "avfilter.h"
> +#include "diff.h"
> +#include "internal.h"
> +#include "formats.h"
> +#include "video.h"
> +
> +typedef struct {
> +    int lo, hi;                    ///< lower and higher threshold number of differences
> +                                   /// values for 8x8 blocks

> +    float frac;                    ///< threshold of changed pixels over the total fraction

Maybe store it as an integer? That would help making the filter bit-exact.

> +
> +    int max_drop_count;            ///< if positive: maximum number of sequential frames to drop
> +                                   ///  if negative: minimum number of frames between two drops
> +    int drop_count;                ///< if positive: number of frames sequentially dropped
> +                                   /// if negative: number of sequential frame which were not dropped
> +
> +    int hsub, vsub;                ///< chroma subsampling values
> +    AVFilterBufferRef *ref;        ///< reference picture
> +    /* diff function used to compute difference between lines */
> +    int (* diff)(const uint8_t *cur, int cur_linesize,
> +                 const uint8_t *ref, int ref_linesize);
> +} DecimateContext;
> +
> +/**
> + * Return 1 in case the two frames are different, 0 otherwise.
> + */
> +static int diff_planes(AVFilterContext *ctx,
> +                       uint8_t *cur, int cur_linesize,
> +                       uint8_t *ref, int ref_linesize,
> +                       int w, int h)
> +{
> +    DecimateContext *decimate = ctx->priv;
> +
> +    int x, y;
> +    int d, c = 0;
> +    int t = (w/16)*(h/16)*decimate->frac;
> +
> +    /* compute difference for blocks of 8x8 bytes */

> +    for (y = 0; y < h-7; y += 4) {
> +        for (x = 8; x < w-7; x += 4) {
> +            d = decimate->diff(cur+x+y*cur_linesize, cur_linesize,
> +                               ref+x+y*ref_linesize, ref_linesize);

I wonder if the benefit of the MMX optimization of the diff function is not
completely wasted by the fact that all computations are done four times.

> +            if (d > decimate->hi)
> +                return 1;
> +            if (d > decimate->lo) {
> +                c++;
> +                if (c > t)
> +                    return 1;
> +            }
> +        }
> +    }
> +    return 0;
> +}
> +
> +/**
> + * Tell if the frame should be decimated, that is if it is no much
> + * different with respect to the reference frame ref.
> + */
> +static int decimate_frame(AVFilterContext *ctx,
> +                          AVFilterBufferRef *cur, AVFilterBufferRef *ref)
> +{
> +    DecimateContext *decimate = ctx->priv;
> +    int plane;
> +
> +    if (decimate->max_drop_count == 0 ||
> +        decimate->drop_count >= decimate->max_drop_count)
> +        return 0;
> +
> +    for (plane = 0; ref->data[plane] && ref->linesize[plane]; plane++) {
> +        int vsub = plane == 1 || plane == 2 ? decimate->vsub : 0;
> +        int hsub = plane == 1 || plane == 2 ? decimate->hsub : 0;
> +        if (diff_planes(ctx,
> +                        cur->data[plane], cur->linesize[plane],
> +                        ref->data[plane], ref->linesize[plane],
> +                        ref->video->w>>hsub, ref->video->h>>vsub))
> +            return 0;
> +    }
> +
> +    return 1;
> +}
> +
> +static av_cold int init(AVFilterContext *ctx, const char *args)
> +{
> +    DecimateContext *decimate = ctx->priv;
> +    int cpu_flags = av_get_cpu_flags();
> +
> +    /* set default values */
> +    decimate->drop_count = decimate->max_drop_count = 0;
> +    decimate->lo = 64*5;
> +    decimate->hi = 64*12;
> +    decimate->frac = 0.33;
> +
> +    if (args)
> +        sscanf(args, "%d:%d:%d:%f",
> +               &decimate->max_drop_count,
> +               &decimate->hi, &decimate->lo, &decimate->frac);
> +
> +    av_log(ctx, AV_LOG_VERBOSE, "max_drop_count:%d hi:%d lo:%d frac:%f\n",
> +           decimate->max_drop_count, decimate->hi, decimate->lo, decimate->frac);
> +
> +    if (HAVE_MMX && cpu_flags&AV_CPU_FLAG_MMX2 && HAVE_EBX_AVAILABLE)
> +        decimate->diff = ff_diff_8x8_mmx;
> +    else
> +        decimate->diff = ff_diff_8x8_c;
> +
> +    return 0;
> +}
> +
> +static av_cold void uninit(AVFilterContext *ctx)
> +{
> +    DecimateContext *decimate = ctx->priv;
> +    avfilter_unref_bufferp(&decimate->ref);
> +}
> +
> +static int query_formats(AVFilterContext *ctx)
> +{
> +    static const enum PixelFormat pix_fmts[] = {
> +        PIX_FMT_YUV444P,      PIX_FMT_YUV422P,
> +        PIX_FMT_YUV420P,      PIX_FMT_YUV411P,
> +        PIX_FMT_YUV410P,      PIX_FMT_YUV440P,
> +        PIX_FMT_YUVJ444P,     PIX_FMT_YUVJ422P,
> +        PIX_FMT_YUVJ420P,     PIX_FMT_YUVJ440P,
> +        PIX_FMT_YUVA420P,
> +        PIX_FMT_NONE
> +    };
> +
> +    ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
> +
> +    return 0;
> +}
> +
> +static int config_input(AVFilterLink *inlink)
> +{
> +    AVFilterContext *ctx = inlink->dst;
> +    DecimateContext *decimate = ctx->priv;
> +    const AVPixFmtDescriptor *pix_desc = &av_pix_fmt_descriptors[inlink->format];
> +    decimate->hsub = pix_desc->log2_chroma_w;
> +    decimate->vsub = pix_desc->log2_chroma_h;
> +    return 0;
> +}
> +
> +static int start_frame(AVFilterLink *inlink, AVFilterBufferRef *picref) { return 0; }
> +
> +static int draw_slice(AVFilterLink *inlink, int y, int h, int slice_dir) { return 0; }
> +
> +static int end_frame(AVFilterLink *inlink)
> +{
> +    DecimateContext *decimate = inlink->dst->priv;
> +    AVFilterBufferRef *cur = inlink->cur_buf;
> +    AVFilterLink *outlink = inlink->dst->outputs[0];
> +    int ret;
> +
> +    if (decimate->ref && decimate_frame(inlink->dst, cur, decimate->ref)) {
> +        decimate->drop_count = FFMAX(1, decimate->drop_count+1);
> +    } else {
> +        avfilter_unref_buffer(decimate->ref);
> +        decimate->ref = cur;
> +        inlink->cur_buf = NULL;
> +        decimate->drop_count = FFMIN(-1, decimate->drop_count-1);
> +
> +        if ((ret = ff_start_frame(outlink,
> +                                  avfilter_ref_buffer(cur, ~AV_PERM_WRITE)) < 0) ||
> +            (ret = ff_draw_slice(outlink, 0, inlink->h, 1)) < 0 ||
> +            (ret = ff_end_frame(outlink)) < 0)
> +            return ret;
> +    }
> +
> +    av_log(inlink->dst, AV_LOG_DEBUG,
> +           "%s pts:%s pts_time:%s drop_count:%d\n",
> +           decimate->drop_count > 0 ? "drop" : "keep",
> +           av_ts2str(cur->pts), av_ts2timestr(cur->pts, &inlink->time_base),
> +           decimate->drop_count);
> +
> +    return 0;
> +}
> +
> +static int request_frame(AVFilterLink *outlink)
> +{
> +    DecimateContext *decimate = outlink->src->priv;
> +    AVFilterLink *inlink = outlink->src->inputs[0];
> +    int ret;
> +
> +    do {
> +        ret = ff_request_frame(inlink);
> +    } while (decimate->drop_count < 0 && ret >= 0);
> +
> +    return ret;
> +}
> +
> +AVFilter avfilter_vf_decimate = {
> +    .name        = "decimate",
> +    .description = NULL_IF_CONFIG_SMALL("Remove near-duplicate frames."),
> +    .init        = init,
> +    .uninit      = uninit,
> +
> +    .priv_size = sizeof(DecimateContext),
> +    .query_formats = query_formats,
> +
> +    .inputs = (const AVFilterPad[]) {
> +        {
> +            .name             = "default",
> +            .type             = AVMEDIA_TYPE_VIDEO,
> +            .get_video_buffer = ff_null_get_video_buffer,
> +            .config_props     = config_input,
> +            .start_frame      = start_frame,
> +            .draw_slice       = draw_slice,
> +            .end_frame        = end_frame,

> +            .min_perms        = AV_PERM_READ,

You need PRESERVE.

> +        },
> +        { .name = NULL }
> +    },
> +    .outputs = (const AVFilterPad[]) {
> +        {
> +            .name          = "default",
> +            .type          = AVMEDIA_TYPE_VIDEO,
> +            .request_frame = request_frame,
> +        },
> +        { .name = NULL }
> +    },
> +};
> diff --git a/libavfilter/x86/Makefile b/libavfilter/x86/Makefile
> index e98693d..1bbfe68 100644
> --- a/libavfilter/x86/Makefile
> +++ b/libavfilter/x86/Makefile
> @@ -1,2 +1,4 @@
>  MMX-OBJS-$(CONFIG_YADIF_FILTER)              += x86/yadif.o
>  MMX-OBJS-$(CONFIG_GRADFUN_FILTER)            += x86/gradfun.o

> +
> +MMX-OBJS-$(HAVE_MMX)                         += x86/diff.o

This looks strange.

> \ No newline at end of file
> diff --git a/libavfilter/x86/diff.c b/libavfilter/x86/diff.c
> new file mode 100644
> index 0000000..b4ad088
> --- /dev/null
> +++ b/libavfilter/x86/diff.c
> @@ -0,0 +1,63 @@
> +/*
> + * Copyright (c) 2003 Rich Felker
> + *
> + * This file is part of FFmpeg.
> + *
> + * FFmpeg is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License as published by
> + * the Free Software Foundation; either version 2 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 General Public License for more details.
> + *
> + * You should have received a copy of the GNU 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/cpu.h"
> +#include "libavfilter/diff.h"
> +
> +int ff_diff_8x8_mmx(const uint8_t *cur, int cur_linesize,
> +                    const uint8_t *ref, int ref_linesize)
> +{
> +    volatile short out[4];
> +    __asm__ (
> +        "movl $8, %%ecx \n\t"
> +        "pxor %%mm4, %%mm4 \n\t"
> +        "pxor %%mm7, %%mm7 \n\t"
> +
> +        ".p2align 4 \n\t"
> +        "1: \n\t"
> +
> +        "movq (%%"REG_S"), %%mm0 \n\t"
> +        "movq (%%"REG_S"), %%mm2 \n\t"
> +        "add %%"REG_a", %%"REG_S" \n\t"
> +        "movq (%%"REG_D"), %%mm1 \n\t"
> +        "add %%"REG_b", %%"REG_D" \n\t"
> +        "psubusb %%mm1, %%mm2 \n\t"
> +        "psubusb %%mm0, %%mm1 \n\t"
> +        "movq %%mm2, %%mm0 \n\t"
> +        "movq %%mm1, %%mm3 \n\t"
> +        "punpcklbw %%mm7, %%mm0 \n\t"
> +        "punpcklbw %%mm7, %%mm1 \n\t"
> +        "punpckhbw %%mm7, %%mm2 \n\t"
> +        "punpckhbw %%mm7, %%mm3 \n\t"
> +        "paddw %%mm0, %%mm4 \n\t"
> +        "paddw %%mm1, %%mm4 \n\t"
> +        "paddw %%mm2, %%mm4 \n\t"
> +        "paddw %%mm3, %%mm4 \n\t"
> +
> +        "decl %%ecx \n\t"
> +        "jnz 1b \n\t"
> +        "movq %%mm4, (%%"REG_d") \n\t"
> +        "emms \n\t"
> +        :
> +        : "S" (cur), "D" (ref), "a" ((long)cur_linesize), "b" ((long)ref_linesize), "d" (out)
> +        : "%ecx", "memory"
> +        );
> +    return out[0]+out[1]+out[2]+out[3];
> +}

Regards,

-- 
  Nicolas George
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20120819/3d6ab172/attachment.asc>


More information about the ffmpeg-devel mailing list