FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
af_asyncts.c
Go to the documentation of this file.
1 /*
2  * This file is part of FFmpeg.
3  *
4  * FFmpeg is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU Lesser General Public
6  * License as published by the Free Software Foundation; either
7  * version 2.1 of the License, or (at your option) any later version.
8  *
9  * FFmpeg is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12  * Lesser General Public License for more details.
13  *
14  * You should have received a copy of the GNU Lesser General Public
15  * License along with FFmpeg; if not, write to the Free Software
16  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17  */
18 
20 #include "libavutil/attributes.h"
21 #include "libavutil/audio_fifo.h"
22 #include "libavutil/common.h"
23 #include "libavutil/mathematics.h"
24 #include "libavutil/opt.h"
25 #include "libavutil/samplefmt.h"
26 
27 #include "audio.h"
28 #include "avfilter.h"
29 #include "internal.h"
30 
31 typedef struct ASyncContext {
32  const AVClass *class;
33 
35  int64_t pts; ///< timestamp in samples of the first sample in fifo
36  int min_delta; ///< pad/trim min threshold in samples
37  int first_frame; ///< 1 until filter_frame() has processed at least 1 frame with a pts != AV_NOPTS_VALUE
38  int64_t first_pts; ///< user-specified first expected pts, in samples
39  int comp; ///< current resample compensation
40 
41  /* options */
42  int resample;
44  int max_comp;
45 
46  /* set by filter_frame() to signal an output frame to request_frame() */
48 } ASyncContext;
49 
50 #define OFFSET(x) offsetof(ASyncContext, x)
51 #define A AV_OPT_FLAG_AUDIO_PARAM
52 #define F AV_OPT_FLAG_FILTERING_PARAM
53 static const AVOption asyncts_options[] = {
54  { "compensate", "Stretch/squeeze the data to make it match the timestamps", OFFSET(resample), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, A|F },
55  { "min_delta", "Minimum difference between timestamps and audio data "
56  "(in seconds) to trigger padding/trimmin the data.", OFFSET(min_delta_sec), AV_OPT_TYPE_FLOAT, { .dbl = 0.1 }, 0, INT_MAX, A|F },
57  { "max_comp", "Maximum compensation in samples per second.", OFFSET(max_comp), AV_OPT_TYPE_INT, { .i64 = 500 }, 0, INT_MAX, A|F },
58  { "first_pts", "Assume the first pts should be this value.", OFFSET(first_pts), AV_OPT_TYPE_INT64, { .i64 = AV_NOPTS_VALUE }, INT64_MIN, INT64_MAX, A|F },
59  { NULL }
60 };
61 
62 AVFILTER_DEFINE_CLASS(asyncts);
63 
64 static av_cold int init(AVFilterContext *ctx)
65 {
66  ASyncContext *s = ctx->priv;
67 
68  s->pts = AV_NOPTS_VALUE;
69  s->first_frame = 1;
70 
71  return 0;
72 }
73 
74 static av_cold void uninit(AVFilterContext *ctx)
75 {
76  ASyncContext *s = ctx->priv;
77 
78  if (s->avr) {
80  avresample_free(&s->avr);
81  }
82 }
83 
84 static int config_props(AVFilterLink *link)
85 {
86  ASyncContext *s = link->src->priv;
87  int ret;
88 
89  s->min_delta = s->min_delta_sec * link->sample_rate;
90  link->time_base = (AVRational){1, link->sample_rate};
91 
93  if (!s->avr)
94  return AVERROR(ENOMEM);
95 
96  av_opt_set_int(s->avr, "in_channel_layout", link->channel_layout, 0);
97  av_opt_set_int(s->avr, "out_channel_layout", link->channel_layout, 0);
98  av_opt_set_int(s->avr, "in_sample_fmt", link->format, 0);
99  av_opt_set_int(s->avr, "out_sample_fmt", link->format, 0);
100  av_opt_set_int(s->avr, "in_sample_rate", link->sample_rate, 0);
101  av_opt_set_int(s->avr, "out_sample_rate", link->sample_rate, 0);
102 
103  if (s->resample)
104  av_opt_set_int(s->avr, "force_resampling", 1, 0);
105 
106  if ((ret = avresample_open(s->avr)) < 0)
107  return ret;
108 
109  return 0;
110 }
111 
112 /* get amount of data currently buffered, in samples */
113 static int64_t get_delay(ASyncContext *s)
114 {
116 }
117 
119 {
120  ASyncContext *s = ctx->priv;
121 
122  if (s->pts < s->first_pts) {
123  int delta = FFMIN(s->first_pts - s->pts, avresample_available(s->avr));
124  av_log(ctx, AV_LOG_VERBOSE, "Trimming %d samples from start\n",
125  delta);
126  avresample_read(s->avr, NULL, delta);
127  s->pts += delta;
128  } else if (s->first_frame)
129  s->pts = s->first_pts;
130 }
131 
132 static int request_frame(AVFilterLink *link)
133 {
134  AVFilterContext *ctx = link->src;
135  ASyncContext *s = ctx->priv;
136  int ret = 0;
137  int nb_samples;
138 
139  s->got_output = 0;
140  while (ret >= 0 && !s->got_output)
141  ret = ff_request_frame(ctx->inputs[0]);
142 
143  /* flush the fifo */
144  if (ret == AVERROR_EOF) {
145  if (s->first_pts != AV_NOPTS_VALUE)
146  handle_trimming(ctx);
147 
148  if (nb_samples = get_delay(s)) {
149  AVFrame *buf = ff_get_audio_buffer(link, nb_samples);
150  if (!buf)
151  return AVERROR(ENOMEM);
152  ret = avresample_convert(s->avr, buf->extended_data,
153  buf->linesize[0], nb_samples, NULL, 0, 0);
154  if (ret <= 0) {
155  av_frame_free(&buf);
156  return (ret < 0) ? ret : AVERROR_EOF;
157  }
158 
159  buf->pts = s->pts;
160  return ff_filter_frame(link, buf);
161  }
162  }
163 
164  return ret;
165 }
166 
168 {
169  int ret = avresample_convert(s->avr, NULL, 0, 0, buf->extended_data,
170  buf->linesize[0], buf->nb_samples);
171  av_frame_free(&buf);
172  return ret;
173 }
174 
175 static int filter_frame(AVFilterLink *inlink, AVFrame *buf)
176 {
177  AVFilterContext *ctx = inlink->dst;
178  ASyncContext *s = ctx->priv;
179  AVFilterLink *outlink = ctx->outputs[0];
181  int64_t pts = (buf->pts == AV_NOPTS_VALUE) ? buf->pts :
182  av_rescale_q(buf->pts, inlink->time_base, outlink->time_base);
183  int out_size, ret;
184  int64_t delta;
185  int64_t new_pts;
186 
187  /* buffer data until we get the next timestamp */
188  if (s->pts == AV_NOPTS_VALUE || pts == AV_NOPTS_VALUE) {
189  if (pts != AV_NOPTS_VALUE) {
190  s->pts = pts - get_delay(s);
191  }
192  return write_to_fifo(s, buf);
193  }
194 
195  if (s->first_pts != AV_NOPTS_VALUE) {
196  handle_trimming(ctx);
197  if (!avresample_available(s->avr))
198  return write_to_fifo(s, buf);
199  }
200 
201  /* when we have two timestamps, compute how many samples would we have
202  * to add/remove to get proper sync between data and timestamps */
203  delta = pts - s->pts - get_delay(s);
204  out_size = avresample_available(s->avr);
205 
206  if (labs(delta) > s->min_delta ||
207  (s->first_frame && delta && s->first_pts != AV_NOPTS_VALUE)) {
208  av_log(ctx, AV_LOG_VERBOSE, "Discontinuity - %"PRId64" samples.\n", delta);
209  out_size = av_clipl_int32((int64_t)out_size + delta);
210  } else {
211  if (s->resample) {
212  // adjust the compensation if delta is non-zero
213  int delay = get_delay(s);
214  int comp = s->comp + av_clip(delta * inlink->sample_rate / delay,
215  -s->max_comp, s->max_comp);
216  if (comp != s->comp) {
217  av_log(ctx, AV_LOG_VERBOSE, "Compensating %d samples per second.\n", comp);
218  if (avresample_set_compensation(s->avr, comp, inlink->sample_rate) == 0) {
219  s->comp = comp;
220  }
221  }
222  }
223  // adjust PTS to avoid monotonicity errors with input PTS jitter
224  pts -= delta;
225  delta = 0;
226  }
227 
228  if (out_size > 0) {
229  AVFrame *buf_out = ff_get_audio_buffer(outlink, out_size);
230  if (!buf_out) {
231  ret = AVERROR(ENOMEM);
232  goto fail;
233  }
234 
235  if (s->first_frame && delta > 0) {
236  int planar = av_sample_fmt_is_planar(buf_out->format);
237  int planes = planar ? nb_channels : 1;
238  int block_size = av_get_bytes_per_sample(buf_out->format) *
239  (planar ? 1 : nb_channels);
240 
241  int ch;
242 
243  av_samples_set_silence(buf_out->extended_data, 0, delta,
244  nb_channels, buf->format);
245 
246  for (ch = 0; ch < planes; ch++)
247  buf_out->extended_data[ch] += delta * block_size;
248 
249  avresample_read(s->avr, buf_out->extended_data, out_size);
250 
251  for (ch = 0; ch < planes; ch++)
252  buf_out->extended_data[ch] -= delta * block_size;
253  } else {
254  avresample_read(s->avr, buf_out->extended_data, out_size);
255 
256  if (delta > 0) {
257  av_samples_set_silence(buf_out->extended_data, out_size - delta,
258  delta, nb_channels, buf->format);
259  }
260  }
261  buf_out->pts = s->pts;
262  ret = ff_filter_frame(outlink, buf_out);
263  if (ret < 0)
264  goto fail;
265  s->got_output = 1;
266  } else if (avresample_available(s->avr)) {
267  av_log(ctx, AV_LOG_WARNING, "Non-monotonous timestamps, dropping "
268  "whole buffer.\n");
269  }
270 
271  /* drain any remaining buffered data */
273 
274  new_pts = pts - avresample_get_delay(s->avr);
275  /* check for s->pts monotonicity */
276  if (new_pts > s->pts) {
277  s->pts = new_pts;
278  ret = avresample_convert(s->avr, NULL, 0, 0, buf->extended_data,
279  buf->linesize[0], buf->nb_samples);
280  } else {
281  av_log(ctx, AV_LOG_WARNING, "Non-monotonous timestamps, dropping "
282  "whole buffer.\n");
283  ret = 0;
284  }
285 
286  s->first_frame = 0;
287 fail:
288  av_frame_free(&buf);
289 
290  return ret;
291 }
292 
294  {
295  .name = "default",
296  .type = AVMEDIA_TYPE_AUDIO,
297  .filter_frame = filter_frame
298  },
299  { NULL }
300 };
301 
303  {
304  .name = "default",
305  .type = AVMEDIA_TYPE_AUDIO,
306  .config_props = config_props,
307  .request_frame = request_frame
308  },
309  { NULL }
310 };
311 
313  .name = "asyncts",
314  .description = NULL_IF_CONFIG_SMALL("Sync audio data to timestamps"),
315  .init = init,
316  .uninit = uninit,
317  .priv_size = sizeof(ASyncContext),
318  .priv_class = &asyncts_class,
319  .inputs = avfilter_af_asyncts_inputs,
320  .outputs = avfilter_af_asyncts_outputs,
321 };