[FFmpeg-devel] [PATCH] lavfi: USPP filter
Michael Niedermayer
michaelni at gmx.at
Tue Dec 9 19:47:53 CET 2014
On Tue, Dec 09, 2014 at 10:13:36PM +0530, arwa arif wrote:
> On Tue, Dec 9, 2014 at 8:27 PM, Michael Niedermayer <michaelni at gmx.at>
> wrote:
>
> > On Tue, Dec 09, 2014 at 04:03:56PM +0530, arwa arif wrote:
> > [...]
> >
> > > +static void filter(USPPContext *p, uint8_t *dst[3], uint8_t *src[3],
> > > + int dst_stride[3], int src_stride[3], int width,
> > > + int height, uint8_t *qp_store, int qp_stride)
> > > +{
> > > + int x, y, i, j;
> > > + const int count = 1<<p->log2_count;
> > > + for(i = 0 ; i < 3 ; i++) {
> > > + int is_chroma = !!i;
> > > + int w = width >>is_chroma;
> > > + int h = height>>is_chroma;
> > > + int stride = p->temp_stride[i];
> > > + int block = BLOCK>>is_chroma;
> > > +
> > > + if (!src[i] || !dst[i])
> > > + continue;
> > > + for(y = 0 ; y < h ; y++) {
> > > + int index = block + block*stride + y*stride;
> > > + memcpy( p->src[i] + index , src[i] + y*src_stride[i] , w );
> > > + for(x = 0 ; x < block ; x++) {
> > > + p->src[i][index - x - 1]= p->src[i][index + x
> > ];
> > > + p->src[i][index + w + x ]= p->src[i][index + w - x -
> > 1];
> > > + }
> > > + }
> > > + for(y = 0 ; y < block ; y++){
> > > + memcpy(p->src[i] + ( block-1-y)*stride , p->src[i] + (
> > y+block )*stride , stride);
> > > + memcpy(p->src[i] + (h+block +y)*stride , p->src[i] +
> > (h-y+block-1)*stride , stride);
> > > + }
> > > +
> > > + p->frame->linesize[i]= stride;
> > > + memset(p->temp[i] , 0 , (h + 2*block)*stride*sizeof(int16_t));
> > > + }
> > > +
> > > + if(p->qp)
> > > + p->frame->quality= p->qp * FF_QP2LAMBDA;
> > > + else
> > > + p->frame->quality= norm_qscale(qp_store[0] , p->qscale_type) *
> > FF_QP2LAMBDA;
> > > +// init per MB qscale stuff FIXME
> > > + p->frame->height = height;
> > > + p->frame->width = width;
> > > +
> > > + for(i = 0 ; i < count ; i++) {
> > > + const int x1 = offset[i+count-1][0];
> > > + const int y1 = offset[i+count-1][1];
> > > + int offset;
> > > + p->frame->data[0] = p->src[0] + x1 + y1 *
> > p->frame->linesize[0];
> > > + p->frame->data[1] = p->src[1] + x1/2 + y1/2 *
> > p->frame->linesize[1];
> > > + p->frame->data[2] = p->src[2] + x1/2 + y1/2 *
> > p->frame->linesize[2];
> > > +
> > > + AVPacket pkt = {.data = p->outbuf, .size = p->outbuf_size};
> >
> > the packet should be cleared with av_init_packet() before setting
> > data/size
> >
> >
> > > + avcodec_encode_video2(p->avctx_enc[i] , &pkt , p->frame ,
> > &p->outbuf_size );
> >
> > the last argument is not the size but a flag that indicates if a
> > packet was encoded. If you use outbuf_size there, it will be
> > overwritten and cause problems
> >
> >
> > > +
> > > + p->frame_dec = p->avctx_enc[i]->coded_frame;
> > > + p->frame->format = p->avctx_enc[i]->pix_fmt;
> >
> > p->frame is an input to avcodec_encode_video2()
> > avcodec_encode_video2() takes the AVFrame and produces a AVPacket
> > thus format must be set before calling avcodec_encode_video2()
> >
> >
> > [...]
> > > +static int config_input(AVFilterLink *inlink)
> > > +{
> > > +
> > > + AVFilterContext *ctx = inlink->dst;
> > > + USPPContext *uspp = ctx->priv;
> > > + const int height = inlink->h;
> > > + const int width = inlink->w;
> > > +
> > > + AVCodec *enc= avcodec_find_encoder(AV_CODEC_ID_SNOW);
> > > + int i;
> > > +
> > > + if (!uspp->use_bframe_qp) {
> > > + /* we are assuming here the qp blocks will not be smaller that
> > 16x16 */
> > > + uspp->non_b_qp_alloc_size = FF_CEIL_RSHIFT(width, 4) *
> > FF_CEIL_RSHIFT(height, 4);
> > > + uspp->non_b_qp_table = av_calloc(uspp->non_b_qp_alloc_size,
> > sizeof(*uspp->non_b_qp_table));
> > > +
> > > + if (!uspp->non_b_qp_table)
> > > + return AVERROR(ENOMEM);
> > > + }
> > > +
> > > + for(i = 0 ; i < 3; i++) {
> > > + int is_chroma = !!i;
> > > + int w = ((width + 4*BLOCK-1) & (~(2*BLOCK-1)))>>is_chroma;
> > > + int h = ((height + 4*BLOCK-1) & (~(2*BLOCK-1)))>>is_chroma;
> > > +
> > > + uspp->temp_stride[i] = w;
> > > + uspp->temp[i] =
> > av_malloc(uspp->temp_stride[i]*h*sizeof(int16_t));
> > > + uspp->src [i] =
> > av_malloc(uspp->temp_stride[i]*h*sizeof(uint8_t));
> > > + }
> > > +
> > > + for(i = 0 ; i < (1<<uspp->log2_count) ; i++) {
> > > + AVCodecContext *avctx_enc;
> > > + AVDictionary *opts = NULL;
> > > + int ret;
> > > +
> > > + avctx_enc = uspp->avctx_enc[i]= avcodec_alloc_context3(NULL);
> > > + avctx_enc->width = width + BLOCK;
> > > + avctx_enc->height = height + BLOCK;
> > > + avctx_enc->time_base = (AVRational){1,25}; // meaningless
> > > + avctx_enc->gop_size = 300;
> > > + avctx_enc->max_b_frames = 0;
> > > + avctx_enc->pix_fmt = AV_PIX_FMT_YUV420P;
> > > + avctx_enc->flags = CODEC_FLAG_QSCALE | CODEC_FLAG_LOW_DELAY;
> > > + avctx_enc->strict_std_compliance = FF_COMPLIANCE_EXPERIMENTAL;
> > > + avctx_enc->global_quality = 123;
> > > + av_dict_set(&opts, "no_bitstream", "1", 0);
> > > + ret = avcodec_open2(avctx_enc, enc, &opts);
> > > + if (ret < 0)
> > > + return ret;
> > > + av_dict_free(&opts);
> > > + av_assert0(avctx_enc->codec);
> > > + }
> >
> > > + uspp->frame = av_frame_alloc();
> > > + uspp->frame_dec = av_frame_alloc();
> >
> > these two need to be freed with av_frame_free() in uninit
> >
> >
> > > +
> > > + uspp->outbuf_size = (width + BLOCK)*(height + BLOCK)*10;
> > > + uspp->outbuf = av_malloc(uspp->outbuf_size);
> >
> > outbuf has to be freed in uninit()
> >
> >
> > [...]
> > > +static av_cold void uninit(AVFilterContext *ctx)
> > > +{
> > > + USPPContext *uspp = ctx->priv;
> > > + if(!uspp) return;
> > > +
> > > + av_freep(&uspp->temp);
> > > + av_freep(&uspp->src);
> > > +
> >
> > > + if (uspp->avctx_enc) {
> > > + avcodec_close(*uspp->avctx_enc);
> >
> > this only deallocates the first encoder like using uspp->avctx_enc[0]
> > would
> > they all must be deallocated
> >
> >
> > you can use something like
> > valgrind --leak-check=full ./ffmpeg_g -f lavfi -i testsrc -vf uspp=1
> > -vframes 1 -f null -
> >
> > to find out what is being allocated but not deallocated
> >
> > [...]
> >
> > --
> > Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
> >
> > You can kill me, but you cannot change the truth.
> >
> > _______________________________________________
> > ffmpeg-devel mailing list
> > ffmpeg-devel at ffmpeg.org
> > http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> >
> >
>
> Updated the patch.
> doc/filters.texi | 24 ++
> libavfilter/Makefile | 1
> libavfilter/allfilters.c | 1
> libavfilter/vf_uspp.c | 472 +++++++++++++++++++++++++++++++++++++++++++++++
> libavfilter/vf_uspp.h | 52 +++++
> 5 files changed, 549 insertions(+), 1 deletion(-)
> b60172409c840581ffc068df874eb5f81b484df9 0001-lavfi-USPP-Filter.patch
> From 1c5bc86e3dd457de87927dbdeb805a17cf9e586b Mon Sep 17 00:00:00 2001
> From: Arwa Arif <arwaarif1994 at gmail.com>
> Date: Sun, 7 Dec 2014 18:56:46 +0530
> Subject: [PATCH] lavfi: USPP Filter
>
> ---
> doc/filters.texi | 24 ++-
> libavfilter/Makefile | 1 +
> libavfilter/allfilters.c | 1 +
> libavfilter/vf_uspp.c | 472 ++++++++++++++++++++++++++++++++++++++++++++++
> libavfilter/vf_uspp.h | 52 +++++
> 5 files changed, 549 insertions(+), 1 deletion(-)
> create mode 100644 libavfilter/vf_uspp.c
> create mode 100644 libavfilter/vf_uspp.h
>
> diff --git a/doc/filters.texi b/doc/filters.texi
> index 8c16c7a..322899d 100644
> --- a/doc/filters.texi
> +++ b/doc/filters.texi
> @@ -6120,7 +6120,6 @@ The list of the currently supported filters follows:
> @item ilpack
> @item pp7
> @item softpulldown
> - at item uspp
> @end table
>
> The parameter syntax and behavior for the listed filters are the same
> @@ -8804,6 +8803,29 @@ unsharp=7:7:-2:7:7:-2
> @end example
> @end itemize
>
> + at section uspp
> +
> +Apply ultra slow/simple postprocessing filter that compresses and decompresses
> +the image at several (or - in the case of @option{quality} level @code{8} - all)
> +shifts and average the results.The way this differs from the behavior of spp is
> +that uspp actually encodes & decodes each case with libavcodec Snow, whereas spp
> +uses a simplified intra only 8x8 DCT similar to MJPEG.
> +
> +The filter accepts the following options:
> +
> + at table @option
> + at item quality
> +Set quality. This option defines the number of levels for averaging. It accepts
> +an integer in the range 0-8. If set to @code{0}, the filter will have no
> +effect. A value of @code{8} means the higher quality. For each increment of
> +that value the speed drops by a factor of approximately 2. Default value is
> + at code{3}.
> +
> + at item qp
> +Force a constant quantization parameter. If not set, the filter will use the QP
> +from the video stream (if available).
> + at end table
> +
> @anchor{vidstabdetect}
> @section vidstabdetect
>
> diff --git a/libavfilter/Makefile b/libavfilter/Makefile
> index 2c56e38..62940fe 100644
> --- a/libavfilter/Makefile
> +++ b/libavfilter/Makefile
> @@ -193,6 +193,7 @@ OBJS-$(CONFIG_TINTERLACE_FILTER) += vf_tinterlace.o
> OBJS-$(CONFIG_TRANSPOSE_FILTER) += vf_transpose.o
> OBJS-$(CONFIG_TRIM_FILTER) += trim.o
> OBJS-$(CONFIG_UNSHARP_FILTER) += vf_unsharp.o
> +OBJS-$(CONFIG_USPP_FILTER) += vf_uspp.o
> OBJS-$(CONFIG_VFLIP_FILTER) += vf_vflip.o
> OBJS-$(CONFIG_VIDSTABDETECT_FILTER) += vidstabutils.o vf_vidstabdetect.o
> OBJS-$(CONFIG_VIDSTABTRANSFORM_FILTER) += vidstabutils.o vf_vidstabtransform.o
> diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c
> index 2352d44..adb86be 100644
> --- a/libavfilter/allfilters.c
> +++ b/libavfilter/allfilters.c
> @@ -208,6 +208,7 @@ void avfilter_register_all(void)
> REGISTER_FILTER(TRANSPOSE, transpose, vf);
> REGISTER_FILTER(TRIM, trim, vf);
> REGISTER_FILTER(UNSHARP, unsharp, vf);
> + REGISTER_FILTER(USPP, uspp, vf);
> REGISTER_FILTER(VFLIP, vflip, vf);
> REGISTER_FILTER(VIDSTABDETECT, vidstabdetect, vf);
> REGISTER_FILTER(VIDSTABTRANSFORM, vidstabtransform, vf);
> diff --git a/libavfilter/vf_uspp.c b/libavfilter/vf_uspp.c
> new file mode 100644
> index 0000000..b9f92d9
> --- /dev/null
> +++ b/libavfilter/vf_uspp.c
> @@ -0,0 +1,472 @@
> +/*
> + * Copyright (c) 2003 Michael Niedermayer <michaelni at gmx.at>
> + * Copyright (c) 2014 Arwa Arif <arwaarif1994 at gmail.com>
> + *
> + * 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.
> + */
> +
> +/**
> + * @file
> + * Ultra Slow/Simple Post-processing filter.
> + *
> + * Originally written by Michael Niedermayer for the MPlayer project, and
> + * ported by Arwa Arif for FFmpeg.
> + */
> +
> +#include "libavutil/avassert.h"
> +#include "libavutil/imgutils.h"
> +#include "libavutil/opt.h"
> +#include "libavutil/pixdesc.h"
> +#include "internal.h"
> +#include "vf_uspp.h"
> +
> +#define OFFSET(x) offsetof(USPPContext, x)
> +#define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
> +static const AVOption uspp_options[] = {
> + { "quality", "set quality", OFFSET(log2_count), AV_OPT_TYPE_INT, {.i64 = 3}, 0, MAX_LEVEL, FLAGS },
> + { "qp", "force a constant quantizer parameter", OFFSET(qp), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 63, FLAGS },
> + { "use_bframe_qp", "use B-frames' QP", OFFSET(use_bframe_qp), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, FLAGS },
> + { NULL }
> +};
> +
> +static const AVClass uspp_class = {
> + .class_name = "uspp",
> + .item_name = av_default_item_name,
> + .option = uspp_options,
> + .version = LIBAVUTIL_VERSION_INT,
> + .category = AV_CLASS_CATEGORY_FILTER,
> +};
> +
> +//===========================================================================//
> +DECLARE_ALIGNED(8, static const uint8_t, dither)[8][8] = {
> + { 0*4, 48*4, 12*4, 60*4, 3*4, 51*4, 15*4, 63*4, },
> + { 32*4, 16*4, 44*4, 28*4, 35*4, 19*4, 47*4, 31*4, },
> + { 8*4, 56*4, 4*4, 52*4, 11*4, 59*4, 7*4, 55*4, },
> + { 40*4, 24*4, 36*4, 20*4, 43*4, 27*4, 39*4, 23*4, },
> + { 2*4, 50*4, 14*4, 62*4, 1*4, 49*4, 13*4, 61*4, },
> + { 34*4, 18*4, 46*4, 30*4, 33*4, 17*4, 45*4, 29*4, },
> + { 10*4, 58*4, 6*4, 54*4, 9*4, 57*4, 5*4, 53*4, },
> + { 42*4, 26*4, 38*4, 22*4, 41*4, 25*4, 37*4, 21*4, },
> +};
> +
> +static const uint8_t offset[511][2]= {
> + { 0, 0},
> + { 0, 0}, { 8, 8}, // quality 1
> + { 0, 0}, { 4, 4}, {12, 8}, { 8,12}, // quality 2
> + { 0, 0}, {10, 2}, { 4, 4}, {14, 6}, { 8, 8}, { 2,10}, {12,12}, { 6,14}, // quality 3
> +
> + { 0, 0}, {10, 2}, { 4, 4}, {14, 6}, { 8, 8}, { 2,10}, {12,12}, { 6,14},
> + { 5, 1}, {15, 3}, { 9, 5}, { 3, 7}, {13, 9}, { 7,11}, { 1,13}, {11,15}, // quality 4
> +
> + { 0, 0}, { 8, 0}, { 0, 8}, { 8, 8}, { 5, 1}, {13, 1}, { 5, 9}, {13, 9},
> + { 2, 2}, {10, 2}, { 2,10}, {10,10}, { 7, 3}, {15, 3}, { 7,11}, {15,11},
> + { 4, 4}, {12, 4}, { 4,12}, {12,12}, { 1, 5}, { 9, 5}, { 1,13}, { 9,13},
> + { 6, 6}, {14, 6}, { 6,14}, {14,14}, { 3, 7}, {11, 7}, { 3,15}, {11,15}, // quality 5
> +
> + { 0, 0}, { 8, 0}, { 0, 8}, { 8, 8}, { 4, 0}, {12, 0}, { 4, 8}, {12, 8},
> + { 1, 1}, { 9, 1}, { 1, 9}, { 9, 9}, { 5, 1}, {13, 1}, { 5, 9}, {13, 9},
> + { 3, 2}, {11, 2}, { 3,10}, {11,10}, { 7, 2}, {15, 2}, { 7,10}, {15,10},
> + { 2, 3}, {10, 3}, { 2,11}, {10,11}, { 6, 3}, {14, 3}, { 6,11}, {14,11},
> + { 0, 4}, { 8, 4}, { 0,12}, { 8,12}, { 4, 4}, {12, 4}, { 4,12}, {12,12},
> + { 1, 5}, { 9, 5}, { 1,13}, { 9,13}, { 5, 5}, {13, 5}, { 5,13}, {13,13},
> + { 3, 6}, {11, 6}, { 3,14}, {11,14}, { 7, 6}, {15, 6}, { 7,14}, {15,14},
> + { 2, 7}, {10, 7}, { 2,15}, {10,15}, { 6, 7}, {14, 7}, { 6,15}, {14,15}, // quality 6
> +
> + { 0, 0}, { 8, 0}, { 0, 8}, { 8, 8}, { 0, 2}, { 8, 2}, { 0,10}, { 8,10},
> + { 0, 4}, { 8, 4}, { 0,12}, { 8,12}, { 0, 6}, { 8, 6}, { 0,14}, { 8,14},
> + { 1, 1}, { 9, 1}, { 1, 9}, { 9, 9}, { 1, 3}, { 9, 3}, { 1,11}, { 9,11},
> + { 1, 5}, { 9, 5}, { 1,13}, { 9,13}, { 1, 7}, { 9, 7}, { 1,15}, { 9,15},
> + { 2, 0}, {10, 0}, { 2, 8}, {10, 8}, { 2, 2}, {10, 2}, { 2,10}, {10,10},
> + { 2, 4}, {10, 4}, { 2,12}, {10,12}, { 2, 6}, {10, 6}, { 2,14}, {10,14},
> + { 3, 1}, {11, 1}, { 3, 9}, {11, 9}, { 3, 3}, {11, 3}, { 3,11}, {11,11},
> + { 3, 5}, {11, 5}, { 3,13}, {11,13}, { 3, 7}, {11, 7}, { 3,15}, {11,15},
> + { 4, 0}, {12, 0}, { 4, 8}, {12, 8}, { 4, 2}, {12, 2}, { 4,10}, {12,10},
> + { 4, 4}, {12, 4}, { 4,12}, {12,12}, { 4, 6}, {12, 6}, { 4,14}, {12,14},
> + { 5, 1}, {13, 1}, { 5, 9}, {13, 9}, { 5, 3}, {13, 3}, { 5,11}, {13,11},
> + { 5, 5}, {13, 5}, { 5,13}, {13,13}, { 5, 7}, {13, 7}, { 5,15}, {13,15},
> + { 6, 0}, {14, 0}, { 6, 8}, {14, 8}, { 6, 2}, {14, 2}, { 6,10}, {14,10},
> + { 6, 4}, {14, 4}, { 6,12}, {14,12}, { 6, 6}, {14, 6}, { 6,14}, {14,14},
> + { 7, 1}, {15, 1}, { 7, 9}, {15, 9}, { 7, 3}, {15, 3}, { 7,11}, {15,11},
> + { 7, 5}, {15, 5}, { 7,13}, {15,13}, { 7, 7}, {15, 7}, { 7,15}, {15,15}, // quality 7
> +
> + { 0, 0}, { 8, 0}, { 0, 8}, { 8, 8}, { 4, 4}, {12, 4}, { 4,12}, {12,12},
> + { 0, 4}, { 8, 4}, { 0,12}, { 8,12}, { 4, 0}, {12, 0}, { 4, 8}, {12, 8},
> + { 2, 2}, {10, 2}, { 2,10}, {10,10}, { 6, 6}, {14, 6}, { 6,14}, {14,14},
> + { 2, 6}, {10, 6}, { 2,14}, {10,14}, { 6, 2}, {14, 2}, { 6,10}, {14,10},
> + { 0, 2}, { 8, 2}, { 0,10}, { 8,10}, { 4, 6}, {12, 6}, { 4,14}, {12,14},
> + { 0, 6}, { 8, 6}, { 0,14}, { 8,14}, { 4, 2}, {12, 2}, { 4,10}, {12,10},
> + { 2, 0}, {10, 0}, { 2, 8}, {10, 8}, { 6, 4}, {14, 4}, { 6,12}, {14,12},
> + { 2, 4}, {10, 4}, { 2,12}, {10,12}, { 6, 0}, {14, 0}, { 6, 8}, {14, 8},
> + { 1, 1}, { 9, 1}, { 1, 9}, { 9, 9}, { 5, 5}, {13, 5}, { 5,13}, {13,13},
> + { 1, 5}, { 9, 5}, { 1,13}, { 9,13}, { 5, 1}, {13, 1}, { 5, 9}, {13, 9},
> + { 3, 3}, {11, 3}, { 3,11}, {11,11}, { 7, 7}, {15, 7}, { 7,15}, {15,15},
> + { 3, 7}, {11, 7}, { 3,15}, {11,15}, { 7, 3}, {15, 3}, { 7,11}, {15,11},
> + { 1, 3}, { 9, 3}, { 1,11}, { 9,11}, { 5, 7}, {13, 7}, { 5,15}, {13,15},
> + { 1, 7}, { 9, 7}, { 1,15}, { 9,15}, { 5, 3}, {13, 3}, { 5,11}, {13,11}, // quality 8
> + { 3, 1}, {11, 1}, { 3, 9}, {11, 9}, { 7, 5}, {15, 5}, { 7,13}, {15,13},
> + { 3, 5}, {11, 5}, { 3,13}, {11,13}, { 7, 1}, {15, 1}, { 7, 9}, {15, 9},
> + { 0, 1}, { 8, 1}, { 0, 9}, { 8, 9}, { 4, 5}, {12, 5}, { 4,13}, {12,13},
> + { 0, 5}, { 8, 5}, { 0,13}, { 8,13}, { 4, 1}, {12, 1}, { 4, 9}, {12, 9},
> + { 2, 3}, {10, 3}, { 2,11}, {10,11}, { 6, 7}, {14, 7}, { 6,15}, {14,15},
> + { 2, 7}, {10, 7}, { 2,15}, {10,15}, { 6, 3}, {14, 3}, { 6,11}, {14,11},
> + { 0, 3}, { 8, 3}, { 0,11}, { 8,11}, { 4, 7}, {12, 7}, { 4,15}, {12,15},
> + { 0, 7}, { 8, 7}, { 0,15}, { 8,15}, { 4, 3}, {12, 3}, { 4,11}, {12,11},
> + { 2, 1}, {10, 1}, { 2, 9}, {10, 9}, { 6, 5}, {14, 5}, { 6,13}, {14,13},
> + { 2, 5}, {10, 5}, { 2,13}, {10,13}, { 6, 1}, {14, 1}, { 6, 9}, {14, 9},
> + { 1, 0}, { 9, 0}, { 1, 8}, { 9, 8}, { 5, 4}, {13, 4}, { 5,12}, {13,12},
> + { 1, 4}, { 9, 4}, { 1,12}, { 9,12}, { 5, 0}, {13, 0}, { 5, 8}, {13, 8},
> + { 3, 2}, {11, 2}, { 3,10}, {11,10}, { 7, 6}, {15, 6}, { 7,14}, {15,14},
> + { 3, 6}, {11, 6}, { 3,14}, {11,14}, { 7, 2}, {15, 2}, { 7,10}, {15,10},
> + { 1, 2}, { 9, 2}, { 1,10}, { 9,10}, { 5, 6}, {13, 6}, { 5,14}, {13,14},
> + { 1, 6}, { 9, 6}, { 1,14}, { 9,14}, { 5, 2}, {13, 2}, { 5,10}, {13,10},
> + { 3, 0}, {11, 0}, { 3, 8}, {11, 8}, { 7, 4}, {15, 4}, { 7,12}, {15,12},
> + { 3, 4}, {11, 4}, { 3,12}, {11,12}, { 7, 0}, {15, 0}, { 7, 8}, {15, 8},
> +};
> +
> +static void store_slice_c(uint8_t *dst, int16_t *src,
> + int dst_stride, int src_stride,
> + int width, int height, int log2_scale )
> +{
> + int y, x;
> +
> +#define STORE(pos) do { \
> + temp = ((src[x + y*src_stride + pos]<<log2_scale) + d[pos])>>8; \
> + if(temp & 0x100) temp = ~(temp>>31); \
> + dst[x + y*dst_stride + pos] = temp; \
> +} while (0);
> +
> + for(y = 0 ; y < height ; y++) {
> + const uint8_t *d = dither[y&7];
> + for(x = 0 ; x < width ; x += 8) {
> + int temp;
> + STORE(0);
> + STORE(1);
> + STORE(2);
> + STORE(3);
> + STORE(4);
> + STORE(5);
> + STORE(6);
> + STORE(7);
> + }
> + }
> +}
> +
> +static inline int norm_qscale(int qscale, int type)
> +{
> + switch (type) {
> + case FF_QSCALE_TYPE_MPEG1: return qscale;
> + case FF_QSCALE_TYPE_MPEG2: return qscale >> 1;
> + case FF_QSCALE_TYPE_H264: return qscale >> 2;
> + case FF_QSCALE_TYPE_VP56: return (63 - qscale + 2) >> 2;
> + }
> + return qscale;
> +}
> +
> +static void filter(USPPContext *p, uint8_t *dst[3], uint8_t *src[3],
> + int dst_stride[3], int src_stride[3], int width,
> + int height, uint8_t *qp_store, int qp_stride)
> +{
> + int x, y, i, j;
> + const int count = 1<<p->log2_count;
> + for(i = 0 ; i < 3 ; i++) {
> + int is_chroma = !!i;
> + int w = width >>is_chroma;
> + int h = height>>is_chroma;
> + int stride = p->temp_stride[i];
> + int block = BLOCK>>is_chroma;
> +
> + if (!src[i] || !dst[i])
> + continue;
> + for(y = 0 ; y < h ; y++) {
> + int index = block + block*stride + y*stride;
> + memcpy( p->src[i] + index , src[i] + y*src_stride[i] , w );
> + for(x = 0 ; x < block ; x++) {
> + p->src[i][index - x - 1]= p->src[i][index + x ];
> + p->src[i][index + w + x ]= p->src[i][index + w - x - 1];
> + }
> + }
> + for(y = 0 ; y < block ; y++){
> + memcpy(p->src[i] + ( block-1-y)*stride , p->src[i] + ( y+block )*stride , stride);
> + memcpy(p->src[i] + (h+block +y)*stride , p->src[i] + (h-y+block-1)*stride , stride);
> + }
> +
> + p->frame->linesize[i]= stride;
> + memset(p->temp[i] , 0 , (h + 2*block)*stride*sizeof(int16_t));
> + }
> +
> + if(p->qp)
> + p->frame->quality= p->qp * FF_QP2LAMBDA;
> + else
> + p->frame->quality= norm_qscale(qp_store[0] , p->qscale_type) * FF_QP2LAMBDA;
> +// init per MB qscale stuff FIXME
> + p->frame->height = height;
> + p->frame->width = width;
> +
> + for(i = 0 ; i < count ; i++) {
> + const int x1 = offset[i+count-1][0];
> + const int y1 = offset[i+count-1][1];
> + int offset;
> + AVPacket pkt;// = {.data = p->outbuf , .size = p->outbuf_size };
> + int got_pkt_ptr;
> +
> + av_init_packet(&pkt);
> + pkt.data = p->outbuf;
> + pkt.size = p->outbuf_size;
> +
> + p->frame->data[0] = p->src[0] + x1 + y1 * p->frame->linesize[0];
> + p->frame->data[1] = p->src[1] + x1/2 + y1/2 * p->frame->linesize[1];
> + p->frame->data[2] = p->src[2] + x1/2 + y1/2 * p->frame->linesize[2];
> + p->frame->format = p->avctx_enc[i]->pix_fmt;
> +
> + avcodec_encode_video2(p->avctx_enc[i] , &pkt , p->frame , &got_pkt_ptr);
> + p->frame_dec = p->avctx_enc[i]->coded_frame;
actually, as you overwrite frame_dec here, you dont need to alloc or
free it, i missed that
> +
> + offset = (BLOCK-x1) + (BLOCK-y1)*p->frame_dec->linesize[0];
> +
> + for(y = 0 ; y < height ; y++)
> + for(x = 0 ; x < width ; x++)
> + p->temp[0][ x + y*p->temp_stride[0] ] += p->frame_dec->data[0][ x + y*p->frame_dec->linesize[0] + offset ];
> +
> + offset = (BLOCK/2-x1/2) + (BLOCK/2-y1/2)*p->frame_dec->linesize[1];
> +
> + for(y = 0 ; y < height/2 ; y++) {
> + for(x = 0; x < width/2 ; x++) {
> + p->temp[1][ x + y*p->temp_stride[1] ] += p->frame_dec->data[1][ x + y*p->frame_dec->linesize[1] + offset ];
> + p->temp[2][ x + y*p->temp_stride[2] ] += p->frame_dec->data[2][ x + y*p->frame_dec->linesize[2] + offset ];
> + }
> + }
> + }
> +
> + for(j = 0 ; j < 3 ; j++) {
> + int is_chroma = !!j;
> + if (!dst[j])
> + continue;
> + store_slice_c(dst[j] , p->temp[j] , dst_stride[j] , p->temp_stride[j] , width>>is_chroma , height>>is_chroma , 8-p->log2_count);
> + }
> +}
> +
> +static int query_formats(AVFilterContext *ctx)
> +{
> + static const enum PixelFormat pix_fmts[] = {
> + AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV422P,
> + AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV411P,
> + AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV440P,
> + AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ422P,
> + AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ440P,
> + AV_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;
> + USPPContext *uspp = ctx->priv;
> + const int height = inlink->h;
> + const int width = inlink->w;
> +
> + AVCodec *enc= avcodec_find_encoder(AV_CODEC_ID_SNOW);
> + int i;
> +
> + if (!uspp->use_bframe_qp) {
> + /* we are assuming here the qp blocks will not be smaller that 16x16 */
> + uspp->non_b_qp_alloc_size = FF_CEIL_RSHIFT(width, 4) * FF_CEIL_RSHIFT(height, 4);
> + uspp->non_b_qp_table = av_calloc(uspp->non_b_qp_alloc_size, sizeof(*uspp->non_b_qp_table));
> +
> + if (!uspp->non_b_qp_table)
> + return AVERROR(ENOMEM);
> + }
> +
> + for(i = 0 ; i < 3; i++) {
> + int is_chroma = !!i;
> + int w = ((width + 4*BLOCK-1) & (~(2*BLOCK-1)))>>is_chroma;
> + int h = ((height + 4*BLOCK-1) & (~(2*BLOCK-1)))>>is_chroma;
> +
> + uspp->temp_stride[i] = w;
> + uspp->temp[i] = av_malloc(uspp->temp_stride[i]*h*sizeof(int16_t));
> + uspp->src [i] = av_malloc(uspp->temp_stride[i]*h*sizeof(uint8_t));
> + }
> +
> + for(i = 0 ; i < (1<<uspp->log2_count) ; i++) {
> + AVCodecContext *avctx_enc;
> + AVDictionary *opts = NULL;
> + int ret;
> +
> + avctx_enc = uspp->avctx_enc[i]= avcodec_alloc_context3(NULL);
> + avctx_enc->width = width + BLOCK;
> + avctx_enc->height = height + BLOCK;
> + avctx_enc->time_base = (AVRational){1,25}; // meaningless
> + avctx_enc->gop_size = 300;
> + avctx_enc->max_b_frames = 0;
> + avctx_enc->pix_fmt = AV_PIX_FMT_YUV420P;
> + avctx_enc->flags = CODEC_FLAG_QSCALE | CODEC_FLAG_LOW_DELAY;
> + avctx_enc->strict_std_compliance = FF_COMPLIANCE_EXPERIMENTAL;
> + avctx_enc->global_quality = 123;
> + av_dict_set(&opts, "no_bitstream", "1", 0);
> + ret = avcodec_open2(avctx_enc, enc, &opts);
> + if (ret < 0)
> + return ret;
> + av_dict_free(&opts);
> + av_assert0(avctx_enc->codec);
> + }
> + uspp->frame = av_frame_alloc();
> + uspp->frame_dec = av_frame_alloc();
> +
> + uspp->outbuf_size = (width + BLOCK)*(height + BLOCK)*10;
> + uspp->outbuf = av_malloc(uspp->outbuf_size);
> + return 0;
> +}
> +
> +static int filter_frame(AVFilterLink *inlink, AVFrame *in)
> +{
> + AVFilterContext *ctx = inlink->dst;
> + USPPContext *uspp = ctx->priv;
> + AVFilterLink *outlink = ctx->outputs[0];
> + AVFrame *out = in;
> +
> + int qp_stride = 0;
> + uint8_t *qp_table = NULL;
> +
> + /* if we are not in a constant user quantizer mode and we don't want to use
> + * the quantizers from the B-frames (B-frames often have a higher QP), we
> + * need to save the qp table from the last non B-frame; this is what the
> + * following code block does */
> + if (!uspp->qp) {
> + qp_table = av_frame_get_qp_table(in, &qp_stride, &uspp->qscale_type);
> +
> + if (qp_table && !uspp->use_bframe_qp && in->pict_type != AV_PICTURE_TYPE_B) {
> + int w, h;
> +
> + /* if the qp stride is not set, it means the QP are only defined on
> + * a line basis */
> + if (!qp_stride) {
> + w = FF_CEIL_RSHIFT(inlink->w, 4);
> + h = 1;
> + } else {
> + w = FF_CEIL_RSHIFT(qp_stride, 4);
> + h = FF_CEIL_RSHIFT(inlink->h, 4);
> + }
> + av_assert0(w * h <= uspp->non_b_qp_alloc_size);
> + memcpy(uspp->non_b_qp_table , qp_table , w * h);
> + }
> + }
> +
> + if (uspp->log2_count && !ctx->is_disabled) {
> + if (!uspp->use_bframe_qp && uspp->non_b_qp_table)
> + qp_table = uspp->non_b_qp_table;
> +
> + if (qp_table || uspp->qp) {
> +
> + /* get a new frame if in-place is not possible or if the dimensions
> + * are not multiple of 8 */
> + if (!av_frame_is_writable(in) || (inlink->w & 7) || (inlink->h & 7)) {
> + const int aligned_w = FFALIGN(inlink->w, 8);
> + const int aligned_h = FFALIGN(inlink->h, 8);
> +
> + out = ff_get_video_buffer(outlink, aligned_w, aligned_h);
> + if (!out) {
> + av_frame_free(&in);
> + return AVERROR(ENOMEM);
> + }
> + av_frame_copy_props(out, in);
> + out->width = in->width;
> + out->height = in->height;
> + }
> +
> + filter(uspp, out->data, in->data, out->linesize, in->linesize, inlink->w, inlink->h, qp_table, qp_stride);
> + emms_c();
> + }
> + }
> +
> + if (in != out) {
> + if (in->data[3])
> + av_image_copy_plane(out->data[3], out->linesize[3],
> + in ->data[3], in ->linesize[3],
> + inlink->w, inlink->h);
> + av_frame_free(&in);
> + }
> + return ff_filter_frame(outlink, out);
> +}
> +
> +static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
> + char *res, int res_len, int flags)
> +{
> + USPPContext *uspp = ctx->priv;
> +
> + if (!strcmp(cmd, "level")) {
> + if (!strcmp(args, "max"))
> + uspp->log2_count = MAX_LEVEL;
> + else
> + uspp->log2_count = av_clip(strtol(args, NULL, 10), 0, MAX_LEVEL);
> + return 0;
> + }
> + return AVERROR(ENOSYS);
> +}
> +
> +static av_cold void uninit(AVFilterContext *ctx)
> +{
> + USPPContext *uspp = ctx->priv;
> + if(!uspp) return;
> +
> + av_freep(&uspp->temp);
> + av_freep(&uspp->src);
you allocate 3 of these, in config_input(), thus you also need to
free 3
> + av_freep(&uspp->outbuf);
> + av_frame_free(&uspp->frame);
> + av_frame_free(&uspp->frame_dec);
> +
> + if (uspp->avctx_enc) {
> + int i = 0;
> + while(uspp->avctx_enc[i]) {
> + avcodec_close(uspp->avctx_enc[i]);
> + av_freep(&uspp->avctx_enc[i]);
> + }
that loop differs from the allocation loop, it should look the same
that is:
for(i = 0 ; i < (1<<uspp->log2_count) ; i++)
> + }
> + av_freep(&uspp->non_b_qp_table);
> +}
> +
> +static const AVFilterPad uspp_inputs[] = {
> + {
> + .name = "default",
> + .type = AVMEDIA_TYPE_VIDEO,
> + .config_props = config_input,
> + .filter_frame = filter_frame,
> + },
> + { NULL }
> +};
> +
> +static const AVFilterPad uspp_outputs[] = {
> + {
> + .name = "default",
> + .type = AVMEDIA_TYPE_VIDEO,
> + },
> + { NULL }
> +};
> +
> +AVFilter ff_vf_uspp = {
> + .name = "uspp",
> + .description = NULL_IF_CONFIG_SMALL("Apply Ultra Simple / Slow Post-processing filter."),
> + .priv_size = sizeof(USPPContext),
> + .uninit = uninit,
> + .query_formats = query_formats,
> + .inputs = uspp_inputs,
> + .outputs = uspp_outputs,
> + .process_command = process_command,
> + .priv_class = &uspp_class,
> + .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL,
> +};
> diff --git a/libavfilter/vf_uspp.h b/libavfilter/vf_uspp.h
> new file mode 100644
> index 0000000..093167f
> --- /dev/null
> +++ b/libavfilter/vf_uspp.h
> @@ -0,0 +1,52 @@
> +/*
> + * Copyright (c) 2003 Michael Niedermayer <michaelni at gmx.at>
> + * Copyright (c) 2014 Arwa Arif <arwaarif1994 at gmail.com>
> + *
> + * 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.
> + */
> +
> +#ifndef AVFILTER_USPP_H
> +#define AVFILTER_USPP_H
> +
> +#include "libavcodec/avcodec.h"
> +#include "avfilter.h"
> +
> +#define MAX_LEVEL 8 /* quality levels */
> +#define BLOCK 16
> +
> +typedef struct {
> + const AVClass *av_class;
> + int log2_count;
> + int qp;
> + int qscale_type;
> + int temp_stride[3];
> + uint8_t *src[3];
> + int16_t *temp[3];
> + int outbuf_size;
> + uint8_t *outbuf;
> + AVCodecContext *avctx_enc[BLOCK*BLOCK];
> + AVFrame *frame;
> + AVFrame *frame_dec;
> + uint8_t *non_b_qp_table;
> + int non_b_qp_alloc_size;
> + int use_bframe_qp;
> +
> +} USPPContext;
> +
> +void ff_uspp_init_x86(USPPContext *s);
> +
> +#endif /* AVFILTER_USPP_H */
> --
> 1.7.9.5
>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
--
Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
The greatest way to live with honor in this world is to be what we pretend
to be. -- Socrates
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 181 bytes
Desc: Digital signature
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20141209/cea3660d/attachment.asc>
More information about the ffmpeg-devel
mailing list