2 * Copyright (c) 2015 Martin Storsjo
4 * This file is part of FFmpeg.
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.
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. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 #include "libavutil/intreadwrite.h"
24 #include "libavutil/mathematics.h"
25 #include "libavutil/md5.h"
27 #include "libavformat/avformat.h"
34 #include "compat/getopt.c"
39 static const uint8_t h264_extradata[] = {
40 0x01, 0x4d, 0x40, 0x1e, 0xff, 0xe1, 0x00, 0x02, 0x67, 0x4d, 0x01, 0x00, 0x02, 0x68, 0xef
42 static const uint8_t aac_extradata[] = {
47 static const char *format = "mp4";
57 uint8_t hash[HASH_SIZE];
59 AVStream *video_st, *audio_st;
60 int64_t audio_dts, video_dts;
64 int64_t audio_duration;
68 enum AVPictureType last_picture;
74 int fake_pkt_duration;
81 static void count_warnings(void *avcl, int level, const char *fmt, va_list vl)
83 if (level == AV_LOG_WARNING)
87 static void init_count_warnings(void)
89 av_log_set_callback(count_warnings);
93 static void reset_count_warnings(void)
95 av_log_set_callback(av_log_default_callback);
98 static int io_write(void *opaque, uint8_t *buf, int size)
101 av_md5_update(md5, buf, size);
103 fwrite(buf, 1, size, out);
107 static int io_write_data_type(void *opaque, uint8_t *buf, int size,
108 enum AVIODataMarkerType type, int64_t time)
110 char timebuf[30], content[5] = { 0 };
113 case AVIO_DATA_MARKER_HEADER: str = "header"; break;
114 case AVIO_DATA_MARKER_SYNC_POINT: str = "sync"; break;
115 case AVIO_DATA_MARKER_BOUNDARY_POINT: str = "boundary"; break;
116 case AVIO_DATA_MARKER_UNKNOWN: str = "unknown"; break;
117 case AVIO_DATA_MARKER_TRAILER: str = "trailer"; break;
119 if (time == AV_NOPTS_VALUE)
120 snprintf(timebuf, sizeof(timebuf), "nopts");
122 snprintf(timebuf, sizeof(timebuf), "%"PRId64, time);
123 // There can be multiple header/trailer callbacks, only log the box type
124 // for header at out_size == 0
125 if (type != AVIO_DATA_MARKER_UNKNOWN &&
126 type != AVIO_DATA_MARKER_TRAILER &&
127 (type != AVIO_DATA_MARKER_HEADER || out_size == 0) &&
129 memcpy(content, &buf[4], 4);
131 snprintf(content, sizeof(content), "-");
132 printf("write_data len %d, time %s, type %s atom %s\n", size, timebuf, str, content);
133 return io_write(opaque, buf, size);
136 static void init_out(const char *name)
140 snprintf(buf, sizeof(buf), "%s.%s", cur_name, format);
144 out = fopen(buf, "wb");
151 static void close_out(void)
154 av_md5_final(md5, hash);
155 for (i = 0; i < HASH_SIZE; i++)
156 printf("%02x", hash[i]);
157 printf(" %d %s\n", out_size, cur_name);
163 static void check_func(int value, int line, const char *msg, ...)
168 printf("%d: ", line);
175 #define check(value, ...) check_func(value, __LINE__, __VA_ARGS__)
177 static void init_fps(int bf, int audio_preroll, int fps)
180 int iobuf_size = force_iobuf_size ? force_iobuf_size : sizeof(iobuf);
181 ctx = avformat_alloc_context();
184 ctx->oformat = av_guess_format(format, NULL, NULL);
187 ctx->pb = avio_alloc_context(iobuf, iobuf_size, AVIO_FLAG_WRITE, NULL, NULL, io_write, NULL);
190 ctx->pb->write_data_type = io_write_data_type;
191 ctx->flags |= AVFMT_FLAG_BITEXACT;
193 st = avformat_new_stream(ctx, NULL);
196 st->codecpar->codec_type = AVMEDIA_TYPE_VIDEO;
197 st->codecpar->codec_id = AV_CODEC_ID_H264;
198 st->codecpar->width = 640;
199 st->codecpar->height = 480;
200 st->time_base.num = 1;
201 st->time_base.den = 30;
202 st->codecpar->extradata_size = sizeof(h264_extradata);
203 st->codecpar->extradata = av_mallocz(st->codecpar->extradata_size + AV_INPUT_BUFFER_PADDING_SIZE);
204 if (!st->codecpar->extradata)
206 memcpy(st->codecpar->extradata, h264_extradata, sizeof(h264_extradata));
209 st = avformat_new_stream(ctx, NULL);
212 st->codecpar->codec_type = AVMEDIA_TYPE_AUDIO;
213 st->codecpar->codec_id = AV_CODEC_ID_AAC;
214 st->codecpar->sample_rate = 44100;
215 st->codecpar->channels = 2;
216 st->time_base.num = 1;
217 st->time_base.den = 44100;
218 st->codecpar->extradata_size = sizeof(aac_extradata);
219 st->codecpar->extradata = av_mallocz(st->codecpar->extradata_size + AV_INPUT_BUFFER_PADDING_SIZE);
220 if (!st->codecpar->extradata)
222 memcpy(st->codecpar->extradata, aac_extradata, sizeof(aac_extradata));
225 if (avformat_write_header(ctx, &opts) < 0)
231 duration = video_st->time_base.den / fps;
232 audio_duration = 1024LL * audio_st->time_base.den / audio_st->codecpar->sample_rate;
234 audio_preroll = 2048LL * audio_st->time_base.den / audio_st->codecpar->sample_rate;
237 video_dts = bframes ? -duration : 0;
238 audio_dts = -audio_preroll;
241 static void init(int bf, int audio_preroll)
243 init_fps(bf, audio_preroll, 30);
246 static void mux_frames(int n, int c)
248 int end_frames = frames + n;
251 uint8_t pktdata[8] = { 0 };
252 av_init_packet(&pkt);
254 if (av_compare_ts(audio_dts, audio_st->time_base, video_dts, video_st->time_base) < 0) {
255 pkt.dts = pkt.pts = audio_dts;
256 pkt.stream_index = 1;
257 pkt.duration = audio_duration;
258 audio_dts += audio_duration;
260 if (frames == end_frames)
263 pkt.stream_index = 0;
264 pkt.duration = duration;
265 if ((frames % gop_size) == 0) {
266 pkt.flags |= AV_PKT_FLAG_KEY;
267 last_picture = AV_PICTURE_TYPE_I;
268 pkt.pts = pkt.dts + duration;
271 if (last_picture == AV_PICTURE_TYPE_P) {
272 last_picture = AV_PICTURE_TYPE_B;
274 video_dts = next_p_pts;
276 last_picture = AV_PICTURE_TYPE_P;
277 if (((frames + 1) % gop_size) == 0) {
278 pkt.pts = pkt.dts + duration;
281 next_p_pts = pkt.pts = pkt.dts + 2 * duration;
282 video_dts += duration;
288 if (fake_pkt_duration)
289 pkt.duration = fake_pkt_duration;
295 AV_WB32(pktdata + 4, pkt.pts);
300 if (skip_write_audio && pkt.stream_index == 1)
304 pkt.pts += (1LL<<32);
305 pkt.dts += (1LL<<32);
309 av_interleaved_write_frame(ctx, &pkt);
311 av_write_frame(ctx, &pkt);
315 static void mux_gops(int n)
317 mux_frames(gop_size * n, 0);
320 static void skip_gops(int n)
327 static void signal_init_ts(void)
330 av_init_packet(&pkt);
334 pkt.stream_index = 0;
337 av_write_frame(ctx, &pkt);
339 pkt.stream_index = 1;
340 pkt.dts = pkt.pts = audio_dts;
341 av_write_frame(ctx, &pkt);
344 static void finish(void)
346 av_write_trailer(ctx);
348 avformat_free_context(ctx);
352 static void help(void)
354 printf("movenc-test [-w]\n"
355 "-w write output into files\n");
358 int main(int argc, char **argv)
361 uint8_t header[HASH_SIZE];
362 uint8_t content[HASH_SIZE];
367 c = getopt(argc, argv, "wh");
383 md5 = av_md5_alloc();
387 // Write a fragmented file with an initial moov that actually contains some
388 // samples. One moov+mdat with 1 second of data and one moof+mdat with 1
390 init_out("non-empty-moov");
391 av_dict_set(&opts, "movflags", "frag_keyframe", 0);
397 // Write a similar file, but with B-frames and audio preroll, handled
399 init_out("non-empty-moov-elst");
400 av_dict_set(&opts, "movflags", "frag_keyframe", 0);
401 av_dict_set(&opts, "use_editlist", "1", 0);
407 // Use B-frames but no audio-preroll, but without an edit list.
408 // Due to avoid_negative_ts == AVFMT_AVOID_NEG_TS_MAKE_ZERO, the dts
409 // of the first audio packet is > 0, but it is set to zero since edit
410 // lists aren't used, increasing the duration of the first packet instead.
411 init_out("non-empty-moov-no-elst");
412 av_dict_set(&opts, "movflags", "frag_keyframe", 0);
413 av_dict_set(&opts, "use_editlist", "0", 0);
420 // Write an ISMV, with B-frames and audio preroll.
422 av_dict_set(&opts, "movflags", "frag_keyframe", 0);
429 // An initial moov that doesn't contain any samples, followed by two
431 init_out("empty-moov");
432 av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
433 av_dict_set(&opts, "use_editlist", "0", 0);
438 memcpy(content, hash, HASH_SIZE);
440 // Similar to the previous one, but with input that doesn't start at
441 // pts/dts 0. avoid_negative_ts behaves in the same way as
442 // in non-empty-moov-no-elst above.
443 init_out("empty-moov-no-elst");
444 av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
450 // Same as the previous one, but disable avoid_negative_ts (which
451 // would require using an edit list, but with empty_moov, one can't
452 // write a sensible edit list, when the start timestamps aren't known).
453 // This should trigger a warning - we check that the warning is produced.
454 init_count_warnings();
455 init_out("empty-moov-no-elst-no-adjust");
456 av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
457 av_dict_set(&opts, "avoid_negative_ts", "0", 0);
463 reset_count_warnings();
464 check(num_warnings > 0, "No warnings printed for unhandled start offset");
466 // Verify that delay_moov produces the same as empty_moov for
468 init_out("delay-moov");
469 av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov", 0);
470 av_dict_set(&opts, "use_editlist", "0", 0);
475 check(!memcmp(hash, content, HASH_SIZE), "delay_moov differs from empty_moov");
477 // Test writing content that requires an edit list using delay_moov
478 init_out("delay-moov-elst");
479 av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov", 0);
485 // Test writing a file with one track lacking packets, with delay_moov.
486 skip_write_audio = 1;
487 init_out("delay-moov-empty-track");
488 av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov", 0);
491 // The automatic flushing shouldn't output anything, since we're still
492 // waiting for data for some tracks
493 check(out_size == 0, "delay_moov flushed prematurely");
494 // When closed (or manually flushed), all the written data should still
498 check(out_size > 0, "delay_moov didn't output anything");
500 // Check that manually flushing still outputs things as expected. This
501 // produces two fragments, while the one above produces only one.
502 init_out("delay-moov-empty-track-flush");
503 av_dict_set(&opts, "movflags", "frag_custom+delay_moov", 0);
506 av_write_frame(ctx, NULL); // Force writing the moov
507 check(out_size > 0, "No moov written");
508 av_write_frame(ctx, NULL);
510 av_write_frame(ctx, NULL);
514 skip_write_audio = 0;
518 // Verify that the header written by delay_moov when manually flushed
519 // is identical to the one by empty_moov.
520 init_out("empty-moov-header");
521 av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
522 av_dict_set(&opts, "use_editlist", "0", 0);
525 memcpy(header, hash, HASH_SIZE);
526 init_out("empty-moov-content");
528 // Written 2 seconds of content, with an automatic flush after 1 second.
529 check(out_size > 0, "No automatic flush?");
530 empty_moov_pos = prev_pos = out_size;
531 // Manually flush the second fragment
532 av_write_frame(ctx, NULL);
533 check(out_size > prev_pos, "No second fragment flushed?");
535 // Check that an extra flush doesn't output any more data
536 av_write_frame(ctx, NULL);
537 check(out_size == prev_pos, "More data written?");
539 memcpy(content, hash, HASH_SIZE);
540 // Ignore the trailer written here
543 init_out("delay-moov-header");
544 av_dict_set(&opts, "movflags", "frag_custom+delay_moov", 0);
545 av_dict_set(&opts, "use_editlist", "0", 0);
547 check(out_size == 0, "Output written during init with delay_moov");
548 mux_gops(1); // Write 1 second of content
549 av_write_frame(ctx, NULL); // Force writing the moov
551 check(!memcmp(hash, header, HASH_SIZE), "delay_moov header differs from empty_moov");
552 init_out("delay-moov-content");
553 av_write_frame(ctx, NULL); // Flush the first fragment
554 check(out_size == empty_moov_pos, "Manually flushed content differs from automatically flushed, %d vs %d", out_size, empty_moov_pos);
555 mux_gops(1); // Write the rest of the content
556 av_write_frame(ctx, NULL); // Flush the second fragment
558 check(!memcmp(hash, content, HASH_SIZE), "delay_moov content differs from empty_moov");
562 // Verify that we can produce an identical second fragment without
563 // writing the first one. First write the reference fragments that
564 // we want to reproduce.
565 av_dict_set(&opts, "movflags", "frag_custom+empty_moov+dash", 0);
568 av_write_frame(ctx, NULL); // Output the first fragment
569 init_out("empty-moov-second-frag");
571 av_write_frame(ctx, NULL); // Output the second fragment
573 memcpy(content, hash, HASH_SIZE);
576 // Produce the same second fragment without actually writing the first
578 av_dict_set(&opts, "movflags", "frag_custom+empty_moov+dash+frag_discont", 0);
579 av_dict_set(&opts, "fragment_index", "2", 0);
580 av_dict_set(&opts, "avoid_negative_ts", "0", 0);
581 av_dict_set(&opts, "use_editlist", "0", 0);
584 init_out("empty-moov-second-frag-discont");
586 av_write_frame(ctx, NULL); // Output the second fragment
588 check(!memcmp(hash, content, HASH_SIZE), "discontinuously written fragment differs");
591 // Produce the same thing by using delay_moov, which requires a slightly
592 // different call sequence.
593 av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash+frag_discont", 0);
594 av_dict_set(&opts, "fragment_index", "2", 0);
598 av_write_frame(ctx, NULL); // Output the moov
599 init_out("delay-moov-second-frag-discont");
600 av_write_frame(ctx, NULL); // Output the second fragment
602 check(!memcmp(hash, content, HASH_SIZE), "discontinuously written fragment differs");
606 // Test discontinuously written fragments with B-frames (where the
607 // assumption of starting at pts=0 works) but not with audio preroll
608 // (which can't be guessed).
609 av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash", 0);
612 init_out("delay-moov-elst-init");
613 av_write_frame(ctx, NULL); // Output the moov
615 memcpy(header, hash, HASH_SIZE);
616 av_write_frame(ctx, NULL); // Output the first fragment
617 init_out("delay-moov-elst-second-frag");
619 av_write_frame(ctx, NULL); // Output the second fragment
621 memcpy(content, hash, HASH_SIZE);
624 av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash+frag_discont", 0);
625 av_dict_set(&opts, "fragment_index", "2", 0);
628 mux_gops(1); // Write the second fragment
629 init_out("delay-moov-elst-init-discont");
630 av_write_frame(ctx, NULL); // Output the moov
632 check(!memcmp(hash, header, HASH_SIZE), "discontinuously written header differs");
633 init_out("delay-moov-elst-second-frag-discont");
634 av_write_frame(ctx, NULL); // Output the second fragment
636 check(!memcmp(hash, content, HASH_SIZE), "discontinuously written fragment differs");
640 // Test discontinuously written fragments with B-frames and audio preroll,
641 // properly signaled.
642 av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash", 0);
645 init_out("delay-moov-elst-signal-init");
646 av_write_frame(ctx, NULL); // Output the moov
648 memcpy(header, hash, HASH_SIZE);
649 av_write_frame(ctx, NULL); // Output the first fragment
650 init_out("delay-moov-elst-signal-second-frag");
652 av_write_frame(ctx, NULL); // Output the second fragment
654 memcpy(content, hash, HASH_SIZE);
657 av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash+frag_discont", 0);
658 av_dict_set(&opts, "fragment_index", "2", 0);
662 mux_gops(1); // Write the second fragment
663 init_out("delay-moov-elst-signal-init-discont");
664 av_write_frame(ctx, NULL); // Output the moov
666 check(!memcmp(hash, header, HASH_SIZE), "discontinuously written header differs");
667 init_out("delay-moov-elst-signal-second-frag-discont");
668 av_write_frame(ctx, NULL); // Output the second fragment
670 check(!memcmp(hash, content, HASH_SIZE), "discontinuously written fragment differs");
674 // Test muxing discontinuous fragments with very large (> (1<<31)) timestamps.
675 av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash+frag_discont", 0);
676 av_dict_set(&opts, "fragment_index", "2", 0);
680 mux_frames(gop_size, 1); // Write the second fragment
681 init_out("delay-moov-elst-signal-init-discont-largets");
682 av_write_frame(ctx, NULL); // Output the moov
684 init_out("delay-moov-elst-signal-second-frag-discont-largets");
685 av_write_frame(ctx, NULL); // Output the second fragment
689 // Test VFR content, with sidx atoms (which declare the pts duration
690 // of a fragment, forcing overriding the start pts of the next one).
691 // Here, the fragment duration in pts is significantly different from
692 // the duration in dts. The video stream starts at dts=-10,pts=0, and
693 // the second fragment starts at dts=155,pts=156. The trun duration sum
694 // of the first fragment is 165, which also is written as
695 // baseMediaDecodeTime in the tfdt in the second fragment. The sidx for
696 // the first fragment says earliest_presentation_time = 0 and
697 // subsegment_duration = 156, which also matches the sidx in the second
698 // fragment. For the audio stream, the pts and dts durations also don't
699 // match - the input stream starts at pts=-2048, but that part is excluded
702 av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov+dash", 0);
704 mux_frames(gop_size/2, 0);
706 mux_frames(gop_size/2, 0);
711 // Test VFR content, with cleared duration fields. In these cases,
712 // the muxer must guess the duration of the last packet of each
713 // fragment. As long as the framerate doesn't vary (too much) at the
714 // fragment edge, it works just fine. Additionally, when automatically
715 // cutting fragments, the muxer already know the timestamps of the next
716 // packet for one stream (in most cases the video stream), avoiding
717 // having to use guesses for that one.
718 init_count_warnings();
720 init_out("vfr-noduration");
721 av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov+dash", 0);
723 mux_frames(gop_size/2, 0);
725 mux_frames(gop_size/2, 0);
730 reset_count_warnings();
731 check(num_warnings > 0, "No warnings printed for filled in durations");
733 // Test with an IO buffer size that is too small to hold a full fragment;
734 // this will cause write_data_type to be called with the type unknown.
735 force_iobuf_size = 1500;
736 init_out("large_frag");
737 av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov", 0);
742 force_iobuf_size = 0;
744 // Test VFR content with bframes with interleaving.
745 // Here, using av_interleaved_write_frame allows the muxer to get the
746 // fragment end durations right. We always set the packet duration to
747 // the expected, but we simulate dropped frames at one point.
749 init_out("vfr-noduration-interleave");
750 av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov", 0);
751 av_dict_set(&opts, "frag_duration", "650000", 0);
753 mux_frames(gop_size/2, 0);
754 // Pretend that the packet duration is the normal, even if
755 // we actually skip a bunch of frames. (I.e., simulate that
756 // we don't know of the framedrop in advance.)
757 fake_pkt_duration = duration;
760 fake_pkt_duration = 0;
762 mux_frames(gop_size/2 - 1, 0);
772 return check_faults > 0 ? 1 : 0;