6520aea94b64d0e8d5d0a2470eb6c7592354cb35
[libav.git] / libavformat / movenc-test.c
1 /*
2 * Copyright (c) 2015 Martin Storsjo
3 *
4 * This file is part of Libav.
5 *
6 * Libav 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 * Libav 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. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with Libav; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21 #include "config.h"
22
23 #include "libavutil/intreadwrite.h"
24 #include "libavutil/mathematics.h"
25 #include "libavutil/md5.h"
26
27 #include "avformat.h"
28
29 #if HAVE_UNISTD_H
30 #include <unistd.h>
31 #endif
32
33 #if !HAVE_GETOPT
34 #include "compat/getopt.c"
35 #endif
36
37 #define HASH_SIZE 16
38
39 static const uint8_t h264_extradata[] = {
40 0x01, 0x4d, 0x40, 0x1e, 0xff, 0xe1, 0x00, 0x02, 0x67, 0x4d, 0x01, 0x00, 0x02, 0x68, 0xef
41 };
42 static const uint8_t aac_extradata[] = {
43 0x12, 0x10
44 };
45
46
47 const char *format = "mp4";
48 AVFormatContext *ctx;
49 uint8_t iobuf[32768];
50 AVDictionary *opts;
51
52 int write_file;
53 const char *cur_name;
54 FILE* out;
55 int out_size;
56 struct AVMD5* md5;
57 uint8_t hash[HASH_SIZE];
58
59 AVStream *video_st, *audio_st;
60 int64_t audio_dts, video_dts;
61
62 int bframes;
63 int64_t duration;
64 int64_t audio_duration;
65 int frames;
66 int gop_size;
67 int64_t next_p_pts;
68 enum AVPictureType last_picture;
69 int skip_write;
70 int skip_write_audio;
71 int clear_duration;
72
73 int num_warnings;
74
75 int check_faults;
76
77
78 static void count_warnings(void *avcl, int level, const char *fmt, va_list vl)
79 {
80 if (level == AV_LOG_WARNING)
81 num_warnings++;
82 }
83
84 static void init_count_warnings(void)
85 {
86 av_log_set_callback(count_warnings);
87 num_warnings = 0;
88 }
89
90 static void reset_count_warnings(void)
91 {
92 av_log_set_callback(av_log_default_callback);
93 }
94
95 static int io_write(void *opaque, uint8_t *buf, int size)
96 {
97 out_size += size;
98 av_md5_update(md5, buf, size);
99 if (out)
100 fwrite(buf, 1, size, out);
101 return size;
102 }
103
104 static void init_out(const char *name)
105 {
106 char buf[100];
107 cur_name = name;
108 snprintf(buf, sizeof(buf), "%s.%s", cur_name, format);
109
110 av_md5_init(md5);
111 if (write_file) {
112 out = fopen(buf, "wb");
113 if (!out)
114 perror(buf);
115 }
116 out_size = 0;
117 }
118
119 static void close_out(void)
120 {
121 int i;
122 av_md5_final(md5, hash);
123 for (i = 0; i < HASH_SIZE; i++)
124 printf("%02x", hash[i]);
125 printf(" %d %s\n", out_size, cur_name);
126 if (out)
127 fclose(out);
128 out = NULL;
129 }
130
131 static void check_func(int value, int line, const char *msg, ...)
132 {
133 if (!value) {
134 va_list ap;
135 va_start(ap, msg);
136 printf("%d: ", line);
137 vprintf(msg, ap);
138 printf("\n");
139 check_faults++;
140 va_end(ap);
141 }
142 }
143 #define check(value, ...) check_func(value, __LINE__, __VA_ARGS__)
144
145 static void init_fps(int bf, int audio_preroll, int fps)
146 {
147 AVStream *st;
148 ctx = avformat_alloc_context();
149 if (!ctx)
150 exit(1);
151 ctx->oformat = av_guess_format(format, NULL, NULL);
152 if (!ctx->oformat)
153 exit(1);
154 ctx->pb = avio_alloc_context(iobuf, sizeof(iobuf), AVIO_FLAG_WRITE, NULL, NULL, io_write, NULL);
155 if (!ctx->pb)
156 exit(1);
157 ctx->flags |= AVFMT_FLAG_BITEXACT;
158
159 st = avformat_new_stream(ctx, NULL);
160 if (!st)
161 exit(1);
162 st->codec->codec_type = AVMEDIA_TYPE_VIDEO;
163 st->codec->codec_id = AV_CODEC_ID_H264;
164 st->codec->width = 640;
165 st->codec->height = 480;
166 st->time_base.num = 1;
167 st->time_base.den = 30;
168 st->codec->extradata_size = sizeof(h264_extradata);
169 st->codec->extradata = av_mallocz(st->codec->extradata_size + AV_INPUT_BUFFER_PADDING_SIZE);
170 if (!st->codec->extradata)
171 exit(1);
172 memcpy(st->codec->extradata, h264_extradata, sizeof(h264_extradata));
173 st->codec->flags |= AV_CODEC_FLAG_GLOBAL_HEADER;
174 video_st = st;
175
176 st = avformat_new_stream(ctx, NULL);
177 if (!st)
178 exit(1);
179 st->codec->codec_type = AVMEDIA_TYPE_AUDIO;
180 st->codec->codec_id = AV_CODEC_ID_AAC;
181 st->codec->sample_rate = 44100;
182 st->codec->channels = 2;
183 st->time_base.num = 1;
184 st->time_base.den = 44100;
185 st->codec->extradata_size = sizeof(aac_extradata);
186 st->codec->extradata = av_mallocz(st->codec->extradata_size + AV_INPUT_BUFFER_PADDING_SIZE);
187 if (!st->codec->extradata)
188 exit(1);
189 memcpy(st->codec->extradata, aac_extradata, sizeof(aac_extradata));
190 st->codec->flags |= AV_CODEC_FLAG_GLOBAL_HEADER;
191 audio_st = st;
192
193 if (avformat_write_header(ctx, &opts) < 0)
194 exit(1);
195 av_dict_free(&opts);
196
197 frames = 0;
198 gop_size = 30;
199 duration = video_st->time_base.den / fps;
200 audio_duration = 1024LL * audio_st->time_base.den / audio_st->codec->sample_rate;
201 if (audio_preroll)
202 audio_preroll = 2048LL * audio_st->time_base.den / audio_st->codec->sample_rate;
203
204 bframes = bf;
205 video_dts = bframes ? -duration : 0;
206 audio_dts = -audio_preroll;
207 }
208
209 static void init(int bf, int audio_preroll)
210 {
211 init_fps(bf, audio_preroll, 30);
212 }
213
214 static void mux_frames(int n)
215 {
216 int end_frames = frames + n;
217 while (1) {
218 AVPacket pkt;
219 uint8_t pktdata[4];
220 av_init_packet(&pkt);
221
222 if (av_compare_ts(audio_dts, audio_st->time_base, video_dts, video_st->time_base) < 0) {
223 pkt.dts = pkt.pts = audio_dts;
224 pkt.stream_index = 1;
225 pkt.duration = audio_duration;
226 audio_dts += audio_duration;
227 } else {
228 if (frames == end_frames)
229 break;
230 pkt.dts = video_dts;
231 pkt.stream_index = 0;
232 pkt.duration = duration;
233 if ((frames % gop_size) == 0) {
234 pkt.flags |= AV_PKT_FLAG_KEY;
235 last_picture = AV_PICTURE_TYPE_I;
236 pkt.pts = pkt.dts + duration;
237 video_dts = pkt.pts;
238 } else {
239 if (last_picture == AV_PICTURE_TYPE_P) {
240 last_picture = AV_PICTURE_TYPE_B;
241 pkt.pts = pkt.dts;
242 video_dts = next_p_pts;
243 } else {
244 last_picture = AV_PICTURE_TYPE_P;
245 if (((frames + 1) % gop_size) == 0) {
246 pkt.pts = pkt.dts + duration;
247 video_dts = pkt.pts;
248 } else {
249 next_p_pts = pkt.pts = pkt.dts + 2 * duration;
250 video_dts += duration;
251 }
252 }
253 }
254 if (!bframes)
255 pkt.pts = pkt.dts;
256 frames++;
257 }
258
259 if (clear_duration)
260 pkt.duration = 0;
261 AV_WB32(pktdata, pkt.pts);
262 pkt.data = pktdata;
263 pkt.size = 4;
264 if (skip_write)
265 continue;
266 if (skip_write_audio && pkt.stream_index == 1)
267 continue;
268 av_write_frame(ctx, &pkt);
269 }
270 }
271
272 static void mux_gops(int n)
273 {
274 mux_frames(gop_size * n);
275 }
276
277 static void skip_gops(int n)
278 {
279 skip_write = 1;
280 mux_gops(n);
281 skip_write = 0;
282 }
283
284 static void signal_init_ts(void)
285 {
286 AVPacket pkt;
287 av_init_packet(&pkt);
288 pkt.size = 0;
289 pkt.data = NULL;
290
291 pkt.stream_index = 0;
292 pkt.dts = video_dts;
293 pkt.pts = 0;
294 av_write_frame(ctx, &pkt);
295
296 pkt.stream_index = 1;
297 pkt.dts = pkt.pts = audio_dts;
298 av_write_frame(ctx, &pkt);
299 }
300
301 static void finish(void)
302 {
303 av_write_trailer(ctx);
304 av_free(ctx->pb);
305 avformat_free_context(ctx);
306 ctx = NULL;
307 }
308
309 static void help(void)
310 {
311 printf("movenc-test [-w]\n"
312 "-w write output into files\n");
313 }
314
315 int main(int argc, char **argv)
316 {
317 int c;
318 uint8_t header[HASH_SIZE];
319 uint8_t content[HASH_SIZE];
320 int empty_moov_pos;
321 int prev_pos;
322
323 for (;;) {
324 c = getopt(argc, argv, "wh");
325 if (c == -1)
326 break;
327 switch (c) {
328 case 'w':
329 write_file = 1;
330 break;
331 default:
332 case 'h':
333 help();
334 return 0;
335 }
336 }
337
338 av_register_all();
339
340 md5 = av_md5_alloc();
341 if (!md5)
342 return 1;
343
344 // Write a fragmented file with an initial moov that actually contains some
345 // samples. One moov+mdat with 1 second of data and one moof+mdat with 1
346 // second of data.
347 init_out("non-empty-moov");
348 av_dict_set(&opts, "movflags", "frag_keyframe", 0);
349 init(0, 0);
350 mux_gops(2);
351 finish();
352 close_out();
353
354 // Write a similar file, but with b-frames and audio preroll, handled
355 // via an edit list.
356 init_out("non-empty-moov-elst");
357 av_dict_set(&opts, "movflags", "frag_keyframe", 0);
358 av_dict_set(&opts, "use_editlist", "1", 0);
359 init(1, 1);
360 mux_gops(2);
361 finish();
362 close_out();
363
364 // Use b-frames but no audio-preroll, but without an edit list.
365 // Due to avoid_negative_ts == AVFMT_AVOID_NEG_TS_MAKE_ZERO, the dts
366 // of the first audio packet is > 0, but it is set to zero since edit
367 // lists aren't used, increasing the duration of the first packet instead.
368 init_out("non-empty-moov-no-elst");
369 av_dict_set(&opts, "movflags", "frag_keyframe", 0);
370 av_dict_set(&opts, "use_editlist", "0", 0);
371 init(1, 0);
372 mux_gops(2);
373 finish();
374 close_out();
375
376 format = "ismv";
377 // Write an ISMV, with b-frames and audio preroll.
378 init_out("ismv");
379 av_dict_set(&opts, "movflags", "frag_keyframe", 0);
380 init(1, 1);
381 mux_gops(2);
382 finish();
383 close_out();
384 format = "mp4";
385
386 // An initial moov that doesn't contain any samples, followed by two
387 // moof+mdat pairs.
388 init_out("empty-moov");
389 av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
390 init(0, 0);
391 mux_gops(2);
392 finish();
393 close_out();
394 memcpy(content, hash, HASH_SIZE);
395
396 // Similar to the previous one, but with input that doesn't start at
397 // pts/dts 0. avoid_negative_ts behaves in the same way as
398 // in non-empty-moov-no-elst above.
399 init_out("empty-moov-no-elst");
400 av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
401 init(1, 0);
402 mux_gops(2);
403 finish();
404 close_out();
405
406 // Same as the previous one, but disable avoid_negative_ts (which
407 // would require using an edit list, but with empty_moov, one can't
408 // write a sensible edit list, when the start timestamps aren't known).
409 // This should trigger a warning - we check that the warning is produced.
410 init_count_warnings();
411 init_out("empty-moov-no-elst-no-adjust");
412 av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
413 av_dict_set(&opts, "avoid_negative_ts", "0", 0);
414 init(1, 0);
415 mux_gops(2);
416 finish();
417 close_out();
418
419 reset_count_warnings();
420 check(num_warnings > 0, "No warnings printed for unhandled start offset");
421
422 // Verify that delay_moov produces the same as empty_moov for
423 // simple input
424 init_out("delay-moov");
425 av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov", 0);
426 init(0, 0);
427 mux_gops(2);
428 finish();
429 close_out();
430 check(!memcmp(hash, content, HASH_SIZE), "delay_moov differs from empty_moov");
431
432 // Test writing content that requires an edit list using delay_moov
433 init_out("delay-moov-elst");
434 av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov", 0);
435 init(1, 1);
436 mux_gops(2);
437 finish();
438 close_out();
439
440 // Test writing a file with one track lacking packets, with delay_moov.
441 skip_write_audio = 1;
442 init_out("delay-moov-empty-track");
443 av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov", 0);
444 init(0, 0);
445 mux_gops(2);
446 // The automatic flushing shouldn't output anything, since we're still
447 // waiting for data for some tracks
448 check(out_size == 0, "delay_moov flushed prematurely");
449 // When closed (or manually flushed), all the written data should still
450 // be output.
451 finish();
452 close_out();
453 check(out_size > 0, "delay_moov didn't output anything");
454
455 // Check that manually flushing still outputs things as expected. This
456 // produces two fragments, while the one above produces only one.
457 init_out("delay-moov-empty-track-flush");
458 av_dict_set(&opts, "movflags", "frag_custom+delay_moov", 0);
459 init(0, 0);
460 mux_gops(1);
461 av_write_frame(ctx, NULL); // Force writing the moov
462 check(out_size > 0, "No moov written");
463 av_write_frame(ctx, NULL);
464 mux_gops(1);
465 av_write_frame(ctx, NULL);
466 finish();
467 close_out();
468
469 skip_write_audio = 0;
470
471
472
473 // Verify that the header written by delay_moov when manually flushed
474 // is identical to the one by empty_moov.
475 init_out("empty-moov-header");
476 av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
477 init(0, 0);
478 close_out();
479 memcpy(header, hash, HASH_SIZE);
480 init_out("empty-moov-content");
481 mux_gops(2);
482 // Written 2 seconds of content, with an automatic flush after 1 second.
483 check(out_size > 0, "No automatic flush?");
484 empty_moov_pos = prev_pos = out_size;
485 // Manually flush the second fragment
486 av_write_frame(ctx, NULL);
487 check(out_size > prev_pos, "No second fragment flushed?");
488 prev_pos = out_size;
489 // Check that an extra flush doesn't output any more data
490 av_write_frame(ctx, NULL);
491 check(out_size == prev_pos, "More data written?");
492 close_out();
493 memcpy(content, hash, HASH_SIZE);
494 // Ignore the trailer written here
495 finish();
496
497 init_out("delay-moov-header");
498 av_dict_set(&opts, "movflags", "frag_custom+delay_moov", 0);
499 init(0, 0);
500 check(out_size == 0, "Output written during init with delay_moov");
501 mux_gops(1); // Write 1 second of content
502 av_write_frame(ctx, NULL); // Force writing the moov
503 close_out();
504 check(!memcmp(hash, header, HASH_SIZE), "delay_moov header differs from empty_moov");
505 init_out("delay-moov-content");
506 av_write_frame(ctx, NULL); // Flush the first fragment
507 check(out_size == empty_moov_pos, "Manually flushed content differs from automatically flushed, %d vs %d", out_size, empty_moov_pos);
508 mux_gops(1); // Write the rest of the content
509 av_write_frame(ctx, NULL); // Flush the second fragment
510 close_out();
511 check(!memcmp(hash, content, HASH_SIZE), "delay_moov content differs from empty_moov");
512 finish();
513
514
515 // Verify that we can produce an identical second fragment without
516 // writing the first one. First write the reference fragments that
517 // we want to reproduce.
518 av_dict_set(&opts, "movflags", "frag_custom+empty_moov+dash", 0);
519 init(0, 0);
520 mux_gops(1);
521 av_write_frame(ctx, NULL); // Output the first fragment
522 init_out("empty-moov-second-frag");
523 mux_gops(1);
524 av_write_frame(ctx, NULL); // Output the second fragment
525 close_out();
526 memcpy(content, hash, HASH_SIZE);
527 finish();
528
529 // Produce the same second fragment without actually writing the first
530 // one before.
531 av_dict_set(&opts, "movflags", "frag_custom+empty_moov+dash+frag_discont", 0);
532 av_dict_set(&opts, "fragment_index", "2", 0);
533 av_dict_set(&opts, "avoid_negative_ts", "0", 0);
534 av_dict_set(&opts, "use_editlist", "0", 0);
535 init(0, 0);
536 skip_gops(1);
537 init_out("empty-moov-second-frag-discont");
538 mux_gops(1);
539 av_write_frame(ctx, NULL); // Output the second fragment
540 close_out();
541 check(!memcmp(hash, content, HASH_SIZE), "discontinuously written fragment differs");
542 finish();
543
544 // Produce the same thing by using delay_moov, which requires a slightly
545 // different call sequence.
546 av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash+frag_discont", 0);
547 av_dict_set(&opts, "fragment_index", "2", 0);
548 init(0, 0);
549 skip_gops(1);
550 mux_gops(1);
551 av_write_frame(ctx, NULL); // Output the moov
552 init_out("delay-moov-second-frag-discont");
553 av_write_frame(ctx, NULL); // Output the second fragment
554 close_out();
555 check(!memcmp(hash, content, HASH_SIZE), "discontinuously written fragment differs");
556 finish();
557
558
559 // Test discontinously written fragments with b-frames (where the
560 // assumption of starting at pts=0 works) but not with audio preroll
561 // (which can't be guessed).
562 av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash", 0);
563 init(1, 0);
564 mux_gops(1);
565 init_out("delay-moov-elst-init");
566 av_write_frame(ctx, NULL); // Output the moov
567 close_out();
568 memcpy(header, hash, HASH_SIZE);
569 av_write_frame(ctx, NULL); // Output the first fragment
570 init_out("delay-moov-elst-second-frag");
571 mux_gops(1);
572 av_write_frame(ctx, NULL); // Output the second fragment
573 close_out();
574 memcpy(content, hash, HASH_SIZE);
575 finish();
576
577 av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash+frag_discont", 0);
578 av_dict_set(&opts, "fragment_index", "2", 0);
579 init(1, 0);
580 skip_gops(1);
581 mux_gops(1); // Write the second fragment
582 init_out("delay-moov-elst-init-discont");
583 av_write_frame(ctx, NULL); // Output the moov
584 close_out();
585 check(!memcmp(hash, header, HASH_SIZE), "discontinuously written header differs");
586 init_out("delay-moov-elst-second-frag-discont");
587 av_write_frame(ctx, NULL); // Output the second fragment
588 close_out();
589 check(!memcmp(hash, content, HASH_SIZE), "discontinuously written fragment differs");
590 finish();
591
592
593 // Test discontinously written fragments with b-frames and audio preroll,
594 // properly signaled.
595 av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash", 0);
596 init(1, 1);
597 mux_gops(1);
598 init_out("delay-moov-elst-signal-init");
599 av_write_frame(ctx, NULL); // Output the moov
600 close_out();
601 memcpy(header, hash, HASH_SIZE);
602 av_write_frame(ctx, NULL); // Output the first fragment
603 init_out("delay-moov-elst-signal-second-frag");
604 mux_gops(1);
605 av_write_frame(ctx, NULL); // Output the second fragment
606 close_out();
607 memcpy(content, hash, HASH_SIZE);
608 finish();
609
610 av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash+frag_discont", 0);
611 av_dict_set(&opts, "fragment_index", "2", 0);
612 init(1, 1);
613 signal_init_ts();
614 skip_gops(1);
615 mux_gops(1); // Write the second fragment
616 init_out("delay-moov-elst-signal-init-discont");
617 av_write_frame(ctx, NULL); // Output the moov
618 close_out();
619 check(!memcmp(hash, header, HASH_SIZE), "discontinuously written header differs");
620 init_out("delay-moov-elst-signal-second-frag-discont");
621 av_write_frame(ctx, NULL); // Output the second fragment
622 close_out();
623 check(!memcmp(hash, content, HASH_SIZE), "discontinuously written fragment differs");
624 finish();
625
626
627 // Test VFR content, with sidx atoms (which declare the pts duration
628 // of a fragment, forcing overriding the start pts of the next one).
629 // Here, the fragment duration in pts is significantly different from
630 // the duration in dts. The video stream starts at dts=-10,pts=0, and
631 // the second fragment starts at dts=155,pts=156. The trun duration sum
632 // of the first fragment is 165, which also is written as
633 // baseMediaDecodeTime in the tfdt in the second fragment. The sidx for
634 // the first fragment says earliest_presentation_time = 0 and
635 // subsegment_duration = 156, which also matches the sidx in the second
636 // fragment. For the audio stream, the pts and dts durations also don't
637 // match - the input stream starts at pts=-2048, but that part is excluded
638 // by the edit list.
639 init_out("vfr");
640 av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov+dash", 0);
641 init_fps(1, 1, 3);
642 mux_frames(gop_size/2);
643 duration /= 10;
644 mux_frames(gop_size/2);
645 mux_gops(1);
646 finish();
647 close_out();
648
649 // Test VFR content, with cleared duration fields. In these cases,
650 // the muxer must guess the duration of the last packet of each
651 // fragment. As long as the framerate doesn't vary (too much) at the
652 // fragment edge, it works just fine. Additionally, when automatically
653 // cutting fragments, the muxer already know the timestamps of the next
654 // packet for one stream (in most cases the video stream), avoiding
655 // having to use guesses for that one.
656 init_count_warnings();
657 clear_duration = 1;
658 init_out("vfr-noduration");
659 av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov+dash", 0);
660 init_fps(1, 1, 3);
661 mux_frames(gop_size/2);
662 duration /= 10;
663 mux_frames(gop_size/2);
664 mux_gops(1);
665 finish();
666 close_out();
667 clear_duration = 0;
668 reset_count_warnings();
669 check(num_warnings > 0, "No warnings printed for filled in durations");
670
671 av_free(md5);
672
673 return check_faults > 0 ? 1 : 0;
674 }