h264_metadata: Add option to delete filler data
[libav.git] / libavfilter / af_asyncts.c
1 /*
2 * This file is part of Libav.
3 *
4 * Libav 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 * Libav 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 Libav; if not, write to the Free Software
16 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18
19 #include <stdint.h>
20
21 #include "libavresample/avresample.h"
22 #include "libavutil/attributes.h"
23 #include "libavutil/audio_fifo.h"
24 #include "libavutil/common.h"
25 #include "libavutil/mathematics.h"
26 #include "libavutil/opt.h"
27 #include "libavutil/samplefmt.h"
28
29 #include "audio.h"
30 #include "avfilter.h"
31 #include "internal.h"
32
33 typedef struct ASyncContext {
34 const AVClass *class;
35
36 AVAudioResampleContext *avr;
37 int64_t pts; ///< timestamp in samples of the first sample in fifo
38 int min_delta; ///< pad/trim min threshold in samples
39 int first_frame; ///< 1 until filter_frame() has processed at least 1 frame with a pts != AV_NOPTS_VALUE
40 int64_t first_pts; ///< user-specified first expected pts, in samples
41 int comp; ///< current resample compensation
42
43 /* options */
44 int resample;
45 float min_delta_sec;
46 int max_comp;
47
48 /* set by filter_frame() to signal an output frame to request_frame() */
49 int got_output;
50 } ASyncContext;
51
52 #define OFFSET(x) offsetof(ASyncContext, x)
53 #define A AV_OPT_FLAG_AUDIO_PARAM
54 static const AVOption options[] = {
55 { "compensate", "Stretch/squeeze the data to make it match the timestamps", OFFSET(resample), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, A },
56 { "min_delta", "Minimum difference between timestamps and audio data "
57 "(in seconds) to trigger padding/trimmin the data.", OFFSET(min_delta_sec), AV_OPT_TYPE_FLOAT, { .dbl = 0.1 }, 0, INT_MAX, A },
58 { "max_comp", "Maximum compensation in samples per second.", OFFSET(max_comp), AV_OPT_TYPE_INT, { .i64 = 500 }, 0, INT_MAX, A },
59 { "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 },
60 { NULL },
61 };
62
63 static const AVClass async_class = {
64 .class_name = "asyncts filter",
65 .item_name = av_default_item_name,
66 .option = options,
67 .version = LIBAVUTIL_VERSION_INT,
68 };
69
70 static av_cold int init(AVFilterContext *ctx)
71 {
72 ASyncContext *s = ctx->priv;
73
74 s->pts = AV_NOPTS_VALUE;
75 s->first_frame = 1;
76
77 return 0;
78 }
79
80 static av_cold void uninit(AVFilterContext *ctx)
81 {
82 ASyncContext *s = ctx->priv;
83
84 if (s->avr) {
85 avresample_close(s->avr);
86 avresample_free(&s->avr);
87 }
88 }
89
90 static int config_props(AVFilterLink *link)
91 {
92 ASyncContext *s = link->src->priv;
93 int ret;
94
95 s->min_delta = s->min_delta_sec * link->sample_rate;
96 link->time_base = (AVRational){1, link->sample_rate};
97
98 s->avr = avresample_alloc_context();
99 if (!s->avr)
100 return AVERROR(ENOMEM);
101
102 av_opt_set_int(s->avr, "in_channel_layout", link->channel_layout, 0);
103 av_opt_set_int(s->avr, "out_channel_layout", link->channel_layout, 0);
104 av_opt_set_int(s->avr, "in_sample_fmt", link->format, 0);
105 av_opt_set_int(s->avr, "out_sample_fmt", link->format, 0);
106 av_opt_set_int(s->avr, "in_sample_rate", link->sample_rate, 0);
107 av_opt_set_int(s->avr, "out_sample_rate", link->sample_rate, 0);
108
109 if (s->resample)
110 av_opt_set_int(s->avr, "force_resampling", 1, 0);
111
112 if ((ret = avresample_open(s->avr)) < 0)
113 return ret;
114
115 return 0;
116 }
117
118 /* get amount of data currently buffered, in samples */
119 static int64_t get_delay(ASyncContext *s)
120 {
121 return avresample_available(s->avr) + avresample_get_delay(s->avr);
122 }
123
124 static void handle_trimming(AVFilterContext *ctx)
125 {
126 ASyncContext *s = ctx->priv;
127
128 if (s->pts < s->first_pts) {
129 int delta = FFMIN(s->first_pts - s->pts, avresample_available(s->avr));
130 av_log(ctx, AV_LOG_VERBOSE, "Trimming %d samples from start\n",
131 delta);
132 avresample_read(s->avr, NULL, delta);
133 s->pts += delta;
134 } else if (s->first_frame)
135 s->pts = s->first_pts;
136 }
137
138 static int request_frame(AVFilterLink *link)
139 {
140 AVFilterContext *ctx = link->src;
141 ASyncContext *s = ctx->priv;
142 int ret = 0;
143 int nb_samples;
144
145 s->got_output = 0;
146 while (ret >= 0 && !s->got_output)
147 ret = ff_request_frame(ctx->inputs[0]);
148
149 /* flush the fifo */
150 if (ret == AVERROR_EOF) {
151 if (s->first_pts != AV_NOPTS_VALUE)
152 handle_trimming(ctx);
153
154 if (nb_samples = get_delay(s)) {
155 AVFrame *buf = ff_get_audio_buffer(link, nb_samples);
156 if (!buf)
157 return AVERROR(ENOMEM);
158 ret = avresample_convert(s->avr, buf->extended_data,
159 buf->linesize[0], nb_samples, NULL, 0, 0);
160 if (ret <= 0) {
161 av_frame_free(&buf);
162 return (ret < 0) ? ret : AVERROR_EOF;
163 }
164
165 buf->pts = s->pts;
166 return ff_filter_frame(link, buf);
167 }
168 }
169
170 return ret;
171 }
172
173 static int write_to_fifo(ASyncContext *s, AVFrame *buf)
174 {
175 int ret = avresample_convert(s->avr, NULL, 0, 0, buf->extended_data,
176 buf->linesize[0], buf->nb_samples);
177 av_frame_free(&buf);
178 return ret;
179 }
180
181 static int filter_frame(AVFilterLink *inlink, AVFrame *buf)
182 {
183 AVFilterContext *ctx = inlink->dst;
184 ASyncContext *s = ctx->priv;
185 AVFilterLink *outlink = ctx->outputs[0];
186 int nb_channels = av_get_channel_layout_nb_channels(buf->channel_layout);
187 int64_t pts = (buf->pts == AV_NOPTS_VALUE) ? buf->pts :
188 av_rescale_q(buf->pts, inlink->time_base, outlink->time_base);
189 int out_size, ret;
190 int64_t delta;
191 int64_t new_pts;
192
193 /* buffer data until we get the next timestamp */
194 if (s->pts == AV_NOPTS_VALUE || pts == AV_NOPTS_VALUE) {
195 if (pts != AV_NOPTS_VALUE) {
196 s->pts = pts - get_delay(s);
197 }
198 return write_to_fifo(s, buf);
199 }
200
201 if (s->first_pts != AV_NOPTS_VALUE) {
202 handle_trimming(ctx);
203 if (!avresample_available(s->avr))
204 return write_to_fifo(s, buf);
205 }
206
207 /* when we have two timestamps, compute how many samples would we have
208 * to add/remove to get proper sync between data and timestamps */
209 delta = pts - s->pts - get_delay(s);
210 out_size = avresample_available(s->avr);
211
212 if (llabs(delta) > s->min_delta ||
213 (s->first_frame && delta && s->first_pts != AV_NOPTS_VALUE)) {
214 av_log(ctx, AV_LOG_VERBOSE, "Discontinuity - %"PRId64" samples.\n", delta);
215 out_size = av_clipl_int32((int64_t)out_size + delta);
216 } else {
217 if (s->resample) {
218 // adjust the compensation if delta is non-zero
219 int delay = get_delay(s);
220 int comp = s->comp + av_clip(delta * inlink->sample_rate / delay,
221 -s->max_comp, s->max_comp);
222 if (comp != s->comp) {
223 av_log(ctx, AV_LOG_VERBOSE, "Compensating %d samples per second.\n", comp);
224 if (avresample_set_compensation(s->avr, comp, inlink->sample_rate) == 0) {
225 s->comp = comp;
226 }
227 }
228 }
229 // adjust PTS to avoid monotonicity errors with input PTS jitter
230 pts -= delta;
231 delta = 0;
232 }
233
234 if (out_size > 0) {
235 AVFrame *buf_out = ff_get_audio_buffer(outlink, out_size);
236 if (!buf_out) {
237 ret = AVERROR(ENOMEM);
238 goto fail;
239 }
240
241 if (s->first_frame && delta > 0) {
242 int planar = av_sample_fmt_is_planar(buf_out->format);
243 int planes = planar ? nb_channels : 1;
244 int block_size = av_get_bytes_per_sample(buf_out->format) *
245 (planar ? 1 : nb_channels);
246
247 int ch;
248
249 av_samples_set_silence(buf_out->extended_data, 0, delta,
250 nb_channels, buf->format);
251
252 for (ch = 0; ch < planes; ch++)
253 buf_out->extended_data[ch] += delta * block_size;
254
255 avresample_read(s->avr, buf_out->extended_data, out_size);
256
257 for (ch = 0; ch < planes; ch++)
258 buf_out->extended_data[ch] -= delta * block_size;
259 } else {
260 avresample_read(s->avr, buf_out->extended_data, out_size);
261
262 if (delta > 0) {
263 av_samples_set_silence(buf_out->extended_data, out_size - delta,
264 delta, nb_channels, buf->format);
265 }
266 }
267 buf_out->pts = s->pts;
268 ret = ff_filter_frame(outlink, buf_out);
269 if (ret < 0)
270 goto fail;
271 s->got_output = 1;
272 } else if (avresample_available(s->avr)) {
273 av_log(ctx, AV_LOG_WARNING, "Non-monotonous timestamps, dropping "
274 "whole buffer.\n");
275 }
276
277 /* drain any remaining buffered data */
278 avresample_read(s->avr, NULL, avresample_available(s->avr));
279
280 new_pts = pts - avresample_get_delay(s->avr);
281 /* check for s->pts monotonicity */
282 if (new_pts > s->pts) {
283 s->pts = new_pts;
284 ret = avresample_convert(s->avr, NULL, 0, 0, buf->extended_data,
285 buf->linesize[0], buf->nb_samples);
286 } else {
287 av_log(ctx, AV_LOG_WARNING, "Non-monotonous timestamps, dropping "
288 "whole buffer.\n");
289 ret = 0;
290 }
291
292 s->first_frame = 0;
293 fail:
294 av_frame_free(&buf);
295
296 return ret;
297 }
298
299 static const AVFilterPad avfilter_af_asyncts_inputs[] = {
300 {
301 .name = "default",
302 .type = AVMEDIA_TYPE_AUDIO,
303 .filter_frame = filter_frame,
304 },
305 { NULL }
306 };
307
308 static const AVFilterPad avfilter_af_asyncts_outputs[] = {
309 {
310 .name = "default",
311 .type = AVMEDIA_TYPE_AUDIO,
312 .config_props = config_props,
313 .request_frame = request_frame
314 },
315 { NULL }
316 };
317
318 AVFilter ff_af_asyncts = {
319 .name = "asyncts",
320 .description = NULL_IF_CONFIG_SMALL("Sync audio data to timestamps"),
321
322 .init = init,
323 .uninit = uninit,
324
325 .priv_size = sizeof(ASyncContext),
326 .priv_class = &async_class,
327
328 .inputs = avfilter_af_asyncts_inputs,
329 .outputs = avfilter_af_asyncts_outputs,
330 };