1/*
2 * Copyright (c) 2012 Nicolas George
3 *
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
14 * See the GNU Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public License
17 * along with FFmpeg; if not, write to the Free Software Foundation, Inc.,
18 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21/**
22 * @file
23 * concat audio-video filter
24 */
25
26#include "libavutil/avassert.h"
27#include "libavutil/avstring.h"
28#include "libavutil/channel_layout.h"
29#include "libavutil/opt.h"
30#include "avfilter.h"
31#define FF_BUFQUEUE_SIZE 256
32#include "bufferqueue.h"
33#include "internal.h"
34#include "video.h"
35#include "audio.h"
36
37#define TYPE_ALL 2
38
39typedef struct {
40    const AVClass *class;
41    unsigned nb_streams[TYPE_ALL]; /**< number of out streams of each type */
42    unsigned nb_segments;
43    unsigned cur_idx; /**< index of the first input of current segment */
44    int64_t delta_ts; /**< timestamp to add to produce output timestamps */
45    unsigned nb_in_active; /**< number of active inputs in current segment */
46    unsigned unsafe;
47    struct concat_in {
48        int64_t pts;
49        int64_t nb_frames;
50        unsigned eof;
51        struct FFBufQueue queue;
52    } *in;
53} ConcatContext;
54
55#define OFFSET(x) offsetof(ConcatContext, x)
56#define A AV_OPT_FLAG_AUDIO_PARAM
57#define F AV_OPT_FLAG_FILTERING_PARAM
58#define V AV_OPT_FLAG_VIDEO_PARAM
59
60static const AVOption concat_options[] = {
61    { "n", "specify the number of segments", OFFSET(nb_segments),
62      AV_OPT_TYPE_INT, { .i64 = 2 }, 2, INT_MAX, V|A|F},
63    { "v", "specify the number of video streams",
64      OFFSET(nb_streams[AVMEDIA_TYPE_VIDEO]),
65      AV_OPT_TYPE_INT, { .i64 = 1 }, 0, INT_MAX, V|F },
66    { "a", "specify the number of audio streams",
67      OFFSET(nb_streams[AVMEDIA_TYPE_AUDIO]),
68      AV_OPT_TYPE_INT, { .i64 = 0 }, 0, INT_MAX, A|F},
69    { "unsafe", "enable unsafe mode",
70      OFFSET(unsafe),
71      AV_OPT_TYPE_INT, { .i64 = 0 }, 0, INT_MAX, V|A|F},
72    { NULL }
73};
74
75AVFILTER_DEFINE_CLASS(concat);
76
77static int query_formats(AVFilterContext *ctx)
78{
79    ConcatContext *cat = ctx->priv;
80    unsigned type, nb_str, idx0 = 0, idx, str, seg;
81    AVFilterFormats *formats, *rates = NULL;
82    AVFilterChannelLayouts *layouts = NULL;
83
84    for (type = 0; type < TYPE_ALL; type++) {
85        nb_str = cat->nb_streams[type];
86        for (str = 0; str < nb_str; str++) {
87            idx = idx0;
88
89            /* Set the output formats */
90            formats = ff_all_formats(type);
91            if (!formats)
92                return AVERROR(ENOMEM);
93            ff_formats_ref(formats, &ctx->outputs[idx]->in_formats);
94            if (type == AVMEDIA_TYPE_AUDIO) {
95                rates = ff_all_samplerates();
96                if (!rates)
97                    return AVERROR(ENOMEM);
98                ff_formats_ref(rates, &ctx->outputs[idx]->in_samplerates);
99                layouts = ff_all_channel_layouts();
100                if (!layouts)
101                    return AVERROR(ENOMEM);
102                ff_channel_layouts_ref(layouts, &ctx->outputs[idx]->in_channel_layouts);
103            }
104
105            /* Set the same formats for each corresponding input */
106            for (seg = 0; seg < cat->nb_segments; seg++) {
107                ff_formats_ref(formats, &ctx->inputs[idx]->out_formats);
108                if (type == AVMEDIA_TYPE_AUDIO) {
109                    ff_formats_ref(rates, &ctx->inputs[idx]->out_samplerates);
110                    ff_channel_layouts_ref(layouts, &ctx->inputs[idx]->out_channel_layouts);
111                }
112                idx += ctx->nb_outputs;
113            }
114
115            idx0++;
116        }
117    }
118    return 0;
119}
120
121static int config_output(AVFilterLink *outlink)
122{
123    AVFilterContext *ctx = outlink->src;
124    ConcatContext *cat   = ctx->priv;
125    unsigned out_no = FF_OUTLINK_IDX(outlink);
126    unsigned in_no  = out_no, seg;
127    AVFilterLink *inlink = ctx->inputs[in_no];
128
129    /* enhancement: find a common one */
130    outlink->time_base           = AV_TIME_BASE_Q;
131    outlink->w                   = inlink->w;
132    outlink->h                   = inlink->h;
133    outlink->sample_aspect_ratio = inlink->sample_aspect_ratio;
134    outlink->format              = inlink->format;
135    for (seg = 1; seg < cat->nb_segments; seg++) {
136        inlink = ctx->inputs[in_no += ctx->nb_outputs];
137        if (!outlink->sample_aspect_ratio.num)
138            outlink->sample_aspect_ratio = inlink->sample_aspect_ratio;
139        /* possible enhancement: unsafe mode, do not check */
140        if (outlink->w                       != inlink->w                       ||
141            outlink->h                       != inlink->h                       ||
142            outlink->sample_aspect_ratio.num != inlink->sample_aspect_ratio.num &&
143                                                inlink->sample_aspect_ratio.num ||
144            outlink->sample_aspect_ratio.den != inlink->sample_aspect_ratio.den) {
145            av_log(ctx, AV_LOG_ERROR, "Input link %s parameters "
146                   "(size %dx%d, SAR %d:%d) do not match the corresponding "
147                   "output link %s parameters (%dx%d, SAR %d:%d)\n",
148                   ctx->input_pads[in_no].name, inlink->w, inlink->h,
149                   inlink->sample_aspect_ratio.num,
150                   inlink->sample_aspect_ratio.den,
151                   ctx->input_pads[out_no].name, outlink->w, outlink->h,
152                   outlink->sample_aspect_ratio.num,
153                   outlink->sample_aspect_ratio.den);
154            if (!cat->unsafe)
155                return AVERROR(EINVAL);
156        }
157    }
158
159    return 0;
160}
161
162static int push_frame(AVFilterContext *ctx, unsigned in_no, AVFrame *buf)
163{
164    ConcatContext *cat = ctx->priv;
165    unsigned out_no = in_no % ctx->nb_outputs;
166    AVFilterLink * inlink = ctx-> inputs[ in_no];
167    AVFilterLink *outlink = ctx->outputs[out_no];
168    struct concat_in *in = &cat->in[in_no];
169
170    buf->pts = av_rescale_q(buf->pts, inlink->time_base, outlink->time_base);
171    in->pts = buf->pts;
172    in->nb_frames++;
173    /* add duration to input PTS */
174    if (inlink->sample_rate)
175        /* use number of audio samples */
176        in->pts += av_rescale_q(buf->nb_samples,
177                                av_make_q(1, inlink->sample_rate),
178                                outlink->time_base);
179    else if (in->nb_frames >= 2)
180        /* use mean duration */
181        in->pts = av_rescale(in->pts, in->nb_frames, in->nb_frames - 1);
182
183    buf->pts += cat->delta_ts;
184    return ff_filter_frame(outlink, buf);
185}
186
187static int process_frame(AVFilterLink *inlink, AVFrame *buf)
188{
189    AVFilterContext *ctx  = inlink->dst;
190    ConcatContext *cat    = ctx->priv;
191    unsigned in_no = FF_INLINK_IDX(inlink);
192
193    if (in_no < cat->cur_idx) {
194        av_log(ctx, AV_LOG_ERROR, "Frame after EOF on input %s\n",
195               ctx->input_pads[in_no].name);
196        av_frame_free(&buf);
197    } else if (in_no >= cat->cur_idx + ctx->nb_outputs) {
198        ff_bufqueue_add(ctx, &cat->in[in_no].queue, buf);
199    } else {
200        return push_frame(ctx, in_no, buf);
201    }
202    return 0;
203}
204
205static AVFrame *get_video_buffer(AVFilterLink *inlink, int w, int h)
206{
207    AVFilterContext *ctx = inlink->dst;
208    unsigned in_no = FF_INLINK_IDX(inlink);
209    AVFilterLink *outlink = ctx->outputs[in_no % ctx->nb_outputs];
210
211    return ff_get_video_buffer(outlink, w, h);
212}
213
214static AVFrame *get_audio_buffer(AVFilterLink *inlink, int nb_samples)
215{
216    AVFilterContext *ctx = inlink->dst;
217    unsigned in_no = FF_INLINK_IDX(inlink);
218    AVFilterLink *outlink = ctx->outputs[in_no % ctx->nb_outputs];
219
220    return ff_get_audio_buffer(outlink, nb_samples);
221}
222
223static int filter_frame(AVFilterLink *inlink, AVFrame *buf)
224{
225    return process_frame(inlink, buf);
226}
227
228static void close_input(AVFilterContext *ctx, unsigned in_no)
229{
230    ConcatContext *cat = ctx->priv;
231
232    cat->in[in_no].eof = 1;
233    cat->nb_in_active--;
234    av_log(ctx, AV_LOG_VERBOSE, "EOF on %s, %d streams left in segment.\n",
235           ctx->input_pads[in_no].name, cat->nb_in_active);
236}
237
238static void find_next_delta_ts(AVFilterContext *ctx, int64_t *seg_delta)
239{
240    ConcatContext *cat = ctx->priv;
241    unsigned i = cat->cur_idx;
242    unsigned imax = i + ctx->nb_outputs;
243    int64_t pts;
244
245    pts = cat->in[i++].pts;
246    for (; i < imax; i++)
247        pts = FFMAX(pts, cat->in[i].pts);
248    cat->delta_ts += pts;
249    *seg_delta = pts;
250}
251
252static int send_silence(AVFilterContext *ctx, unsigned in_no, unsigned out_no,
253                        int64_t seg_delta)
254{
255    ConcatContext *cat = ctx->priv;
256    AVFilterLink *outlink = ctx->outputs[out_no];
257    int64_t base_pts = cat->in[in_no].pts + cat->delta_ts - seg_delta;
258    int64_t nb_samples, sent = 0;
259    int frame_nb_samples, ret;
260    AVRational rate_tb = { 1, ctx->inputs[in_no]->sample_rate };
261    AVFrame *buf;
262    int nb_channels = av_get_channel_layout_nb_channels(outlink->channel_layout);
263
264    if (!rate_tb.den)
265        return AVERROR_BUG;
266    nb_samples = av_rescale_q(seg_delta - cat->in[in_no].pts,
267                              outlink->time_base, rate_tb);
268    frame_nb_samples = FFMAX(9600, rate_tb.den / 5); /* arbitrary */
269    while (nb_samples) {
270        frame_nb_samples = FFMIN(frame_nb_samples, nb_samples);
271        buf = ff_get_audio_buffer(outlink, frame_nb_samples);
272        if (!buf)
273            return AVERROR(ENOMEM);
274        av_samples_set_silence(buf->extended_data, 0, frame_nb_samples,
275                               nb_channels, outlink->format);
276        buf->pts = base_pts + av_rescale_q(sent, rate_tb, outlink->time_base);
277        ret = ff_filter_frame(outlink, buf);
278        if (ret < 0)
279            return ret;
280        sent       += frame_nb_samples;
281        nb_samples -= frame_nb_samples;
282    }
283    return 0;
284}
285
286static int flush_segment(AVFilterContext *ctx)
287{
288    int ret;
289    ConcatContext *cat = ctx->priv;
290    unsigned str, str_max;
291    int64_t seg_delta;
292
293    find_next_delta_ts(ctx, &seg_delta);
294    cat->cur_idx += ctx->nb_outputs;
295    cat->nb_in_active = ctx->nb_outputs;
296    av_log(ctx, AV_LOG_VERBOSE, "Segment finished at pts=%"PRId64"\n",
297           cat->delta_ts);
298
299    if (cat->cur_idx < ctx->nb_inputs) {
300        /* pad audio streams with silence */
301        str = cat->nb_streams[AVMEDIA_TYPE_VIDEO];
302        str_max = str + cat->nb_streams[AVMEDIA_TYPE_AUDIO];
303        for (; str < str_max; str++) {
304            ret = send_silence(ctx, cat->cur_idx - ctx->nb_outputs + str, str,
305                               seg_delta);
306            if (ret < 0)
307                return ret;
308        }
309        /* flush queued buffers */
310        /* possible enhancement: flush in PTS order */
311        str_max = cat->cur_idx + ctx->nb_outputs;
312        for (str = cat->cur_idx; str < str_max; str++) {
313            while (cat->in[str].queue.available) {
314                ret = push_frame(ctx, str, ff_bufqueue_get(&cat->in[str].queue));
315                if (ret < 0)
316                    return ret;
317            }
318        }
319    }
320    return 0;
321}
322
323static int request_frame(AVFilterLink *outlink)
324{
325    AVFilterContext *ctx = outlink->src;
326    ConcatContext *cat   = ctx->priv;
327    unsigned out_no = FF_OUTLINK_IDX(outlink);
328    unsigned in_no  = out_no + cat->cur_idx;
329    unsigned str, str_max;
330    int ret;
331
332    while (1) {
333        if (in_no >= ctx->nb_inputs)
334            return AVERROR_EOF;
335        if (!cat->in[in_no].eof) {
336            ret = ff_request_frame(ctx->inputs[in_no]);
337            if (ret != AVERROR_EOF)
338                return ret;
339            close_input(ctx, in_no);
340        }
341        /* cycle on all inputs to finish the segment */
342        /* possible enhancement: request in PTS order */
343        str_max = cat->cur_idx + ctx->nb_outputs - 1;
344        for (str = cat->cur_idx; cat->nb_in_active;
345             str = str == str_max ? cat->cur_idx : str + 1) {
346            if (cat->in[str].eof)
347                continue;
348            ret = ff_request_frame(ctx->inputs[str]);
349            if (ret == AVERROR_EOF)
350                close_input(ctx, str);
351            else if (ret < 0)
352                return ret;
353        }
354        ret = flush_segment(ctx);
355        if (ret < 0)
356            return ret;
357        in_no += ctx->nb_outputs;
358    }
359}
360
361static av_cold int init(AVFilterContext *ctx)
362{
363    ConcatContext *cat = ctx->priv;
364    unsigned seg, type, str;
365
366    /* create input pads */
367    for (seg = 0; seg < cat->nb_segments; seg++) {
368        for (type = 0; type < TYPE_ALL; type++) {
369            for (str = 0; str < cat->nb_streams[type]; str++) {
370                AVFilterPad pad = {
371                    .type             = type,
372                    .get_video_buffer = get_video_buffer,
373                    .get_audio_buffer = get_audio_buffer,
374                    .filter_frame     = filter_frame,
375                };
376                pad.name = av_asprintf("in%d:%c%d", seg, "va"[type], str);
377                ff_insert_inpad(ctx, ctx->nb_inputs, &pad);
378            }
379        }
380    }
381    /* create output pads */
382    for (type = 0; type < TYPE_ALL; type++) {
383        for (str = 0; str < cat->nb_streams[type]; str++) {
384            AVFilterPad pad = {
385                .type          = type,
386                .config_props  = config_output,
387                .request_frame = request_frame,
388            };
389            pad.name = av_asprintf("out:%c%d", "va"[type], str);
390            ff_insert_outpad(ctx, ctx->nb_outputs, &pad);
391        }
392    }
393
394    cat->in = av_calloc(ctx->nb_inputs, sizeof(*cat->in));
395    if (!cat->in)
396        return AVERROR(ENOMEM);
397    cat->nb_in_active = ctx->nb_outputs;
398    return 0;
399}
400
401static av_cold void uninit(AVFilterContext *ctx)
402{
403    ConcatContext *cat = ctx->priv;
404    unsigned i;
405
406    for (i = 0; i < ctx->nb_inputs; i++) {
407        av_freep(&ctx->input_pads[i].name);
408        ff_bufqueue_discard_all(&cat->in[i].queue);
409    }
410    for (i = 0; i < ctx->nb_outputs; i++)
411        av_freep(&ctx->output_pads[i].name);
412    av_free(cat->in);
413}
414
415AVFilter ff_avf_concat = {
416    .name          = "concat",
417    .description   = NULL_IF_CONFIG_SMALL("Concatenate audio and video streams."),
418    .init          = init,
419    .uninit        = uninit,
420    .query_formats = query_formats,
421    .priv_size     = sizeof(ConcatContext),
422    .inputs        = NULL,
423    .outputs       = NULL,
424    .priv_class    = &concat_class,
425    .flags         = AVFILTER_FLAG_DYNAMIC_INPUTS | AVFILTER_FLAG_DYNAMIC_OUTPUTS,
426};
427