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"
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;
78 static void count_warnings(void *avcl, int level, const char *fmt, va_list vl)
80 if (level == AV_LOG_WARNING)
84 static void init_count_warnings(void)
86 av_log_set_callback(count_warnings);
90 static void reset_count_warnings(void)
92 av_log_set_callback(av_log_default_callback);
95 static int io_write(void *opaque, uint8_t *buf, int size)
98 av_md5_update(md5, buf, size);
100 fwrite(buf, 1, size, out);
104 static void init_out(const char *name)
108 snprintf(buf, sizeof(buf), "%s.%s", cur_name, format);
112 out = fopen(buf, "wb");
119 static void close_out(void)
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);
131 static void check_func(int value, int line, const char *msg, ...)
136 printf("%d: ", line);
143 #define check(value, ...) check_func(value, __LINE__, __VA_ARGS__)
145 static void init_fps(int bf, int audio_preroll, int fps)
148 ctx = avformat_alloc_context();
151 ctx->oformat = av_guess_format(format, NULL, NULL);
154 ctx->pb = avio_alloc_context(iobuf, sizeof(iobuf), AVIO_FLAG_WRITE, NULL, NULL, io_write, NULL);
157 ctx->flags |= AVFMT_FLAG_BITEXACT;
159 st = avformat_new_stream(ctx, NULL);
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)
172 memcpy(st->codec->extradata, h264_extradata, sizeof(h264_extradata));
173 st->codec->flags |= AV_CODEC_FLAG_GLOBAL_HEADER;
176 st = avformat_new_stream(ctx, NULL);
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)
189 memcpy(st->codec->extradata, aac_extradata, sizeof(aac_extradata));
190 st->codec->flags |= AV_CODEC_FLAG_GLOBAL_HEADER;
193 if (avformat_write_header(ctx, &opts) < 0)
199 duration = video_st->time_base.den / fps;
200 audio_duration = 1024LL * audio_st->time_base.den / audio_st->codec->sample_rate;
202 audio_preroll = 2048LL * audio_st->time_base.den / audio_st->codec->sample_rate;
205 video_dts = bframes ? -duration : 0;
206 audio_dts = -audio_preroll;
209 static void init(int bf, int audio_preroll)
211 init_fps(bf, audio_preroll, 30);
214 static void mux_frames(int n)
216 int end_frames = frames + n;
219 uint8_t pktdata[8] = { 0 };
220 av_init_packet(&pkt);
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;
228 if (frames == end_frames)
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;
239 if (last_picture == AV_PICTURE_TYPE_P) {
240 last_picture = AV_PICTURE_TYPE_B;
242 video_dts = next_p_pts;
244 last_picture = AV_PICTURE_TYPE_P;
245 if (((frames + 1) % gop_size) == 0) {
246 pkt.pts = pkt.dts + duration;
249 next_p_pts = pkt.pts = pkt.dts + 2 * duration;
250 video_dts += duration;
261 AV_WB32(pktdata + 4, pkt.pts);
266 if (skip_write_audio && pkt.stream_index == 1)
268 av_write_frame(ctx, &pkt);
272 static void mux_gops(int n)
274 mux_frames(gop_size * n);
277 static void skip_gops(int n)
284 static void signal_init_ts(void)
287 av_init_packet(&pkt);
291 pkt.stream_index = 0;
294 av_write_frame(ctx, &pkt);
296 pkt.stream_index = 1;
297 pkt.dts = pkt.pts = audio_dts;
298 av_write_frame(ctx, &pkt);
301 static void finish(void)
303 av_write_trailer(ctx);
305 avformat_free_context(ctx);
309 static void help(void)
311 printf("movenc-test [-w]\n"
312 "-w write output into files\n");
315 int main(int argc, char **argv)
318 uint8_t header[HASH_SIZE];
319 uint8_t content[HASH_SIZE];
324 c = getopt(argc, argv, "wh");
340 md5 = av_md5_alloc();
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
347 init_out("non-empty-moov");
348 av_dict_set(&opts, "movflags", "frag_keyframe", 0);
354 // Write a similar file, but with b-frames and audio preroll, handled
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);
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);
377 // Write an ISMV, with b-frames and audio preroll.
379 av_dict_set(&opts, "movflags", "frag_keyframe", 0);
386 // An initial moov that doesn't contain any samples, followed by two
388 init_out("empty-moov");
389 av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
390 av_dict_set(&opts, "use_editlist", "0", 0);
395 memcpy(content, hash, HASH_SIZE);
397 // Similar to the previous one, but with input that doesn't start at
398 // pts/dts 0. avoid_negative_ts behaves in the same way as
399 // in non-empty-moov-no-elst above.
400 init_out("empty-moov-no-elst");
401 av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
407 // Same as the previous one, but disable avoid_negative_ts (which
408 // would require using an edit list, but with empty_moov, one can't
409 // write a sensible edit list, when the start timestamps aren't known).
410 // This should trigger a warning - we check that the warning is produced.
411 init_count_warnings();
412 init_out("empty-moov-no-elst-no-adjust");
413 av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
414 av_dict_set(&opts, "avoid_negative_ts", "0", 0);
420 reset_count_warnings();
421 check(num_warnings > 0, "No warnings printed for unhandled start offset");
423 // Verify that delay_moov produces the same as empty_moov for
425 init_out("delay-moov");
426 av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov", 0);
427 av_dict_set(&opts, "use_editlist", "0", 0);
432 check(!memcmp(hash, content, HASH_SIZE), "delay_moov differs from empty_moov");
434 // Test writing content that requires an edit list using delay_moov
435 init_out("delay-moov-elst");
436 av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov", 0);
442 // Test writing a file with one track lacking packets, with delay_moov.
443 skip_write_audio = 1;
444 init_out("delay-moov-empty-track");
445 av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov", 0);
448 // The automatic flushing shouldn't output anything, since we're still
449 // waiting for data for some tracks
450 check(out_size == 0, "delay_moov flushed prematurely");
451 // When closed (or manually flushed), all the written data should still
455 check(out_size > 0, "delay_moov didn't output anything");
457 // Check that manually flushing still outputs things as expected. This
458 // produces two fragments, while the one above produces only one.
459 init_out("delay-moov-empty-track-flush");
460 av_dict_set(&opts, "movflags", "frag_custom+delay_moov", 0);
463 av_write_frame(ctx, NULL); // Force writing the moov
464 check(out_size > 0, "No moov written");
465 av_write_frame(ctx, NULL);
467 av_write_frame(ctx, NULL);
471 skip_write_audio = 0;
475 // Verify that the header written by delay_moov when manually flushed
476 // is identical to the one by empty_moov.
477 init_out("empty-moov-header");
478 av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
479 av_dict_set(&opts, "use_editlist", "0", 0);
482 memcpy(header, hash, HASH_SIZE);
483 init_out("empty-moov-content");
485 // Written 2 seconds of content, with an automatic flush after 1 second.
486 check(out_size > 0, "No automatic flush?");
487 empty_moov_pos = prev_pos = out_size;
488 // Manually flush the second fragment
489 av_write_frame(ctx, NULL);
490 check(out_size > prev_pos, "No second fragment flushed?");
492 // Check that an extra flush doesn't output any more data
493 av_write_frame(ctx, NULL);
494 check(out_size == prev_pos, "More data written?");
496 memcpy(content, hash, HASH_SIZE);
497 // Ignore the trailer written here
500 init_out("delay-moov-header");
501 av_dict_set(&opts, "movflags", "frag_custom+delay_moov", 0);
502 av_dict_set(&opts, "use_editlist", "0", 0);
504 check(out_size == 0, "Output written during init with delay_moov");
505 mux_gops(1); // Write 1 second of content
506 av_write_frame(ctx, NULL); // Force writing the moov
508 check(!memcmp(hash, header, HASH_SIZE), "delay_moov header differs from empty_moov");
509 init_out("delay-moov-content");
510 av_write_frame(ctx, NULL); // Flush the first fragment
511 check(out_size == empty_moov_pos, "Manually flushed content differs from automatically flushed, %d vs %d", out_size, empty_moov_pos);
512 mux_gops(1); // Write the rest of the content
513 av_write_frame(ctx, NULL); // Flush the second fragment
515 check(!memcmp(hash, content, HASH_SIZE), "delay_moov content differs from empty_moov");
519 // Verify that we can produce an identical second fragment without
520 // writing the first one. First write the reference fragments that
521 // we want to reproduce.
522 av_dict_set(&opts, "movflags", "frag_custom+empty_moov+dash", 0);
525 av_write_frame(ctx, NULL); // Output the first fragment
526 init_out("empty-moov-second-frag");
528 av_write_frame(ctx, NULL); // Output the second fragment
530 memcpy(content, hash, HASH_SIZE);
533 // Produce the same second fragment without actually writing the first
535 av_dict_set(&opts, "movflags", "frag_custom+empty_moov+dash+frag_discont", 0);
536 av_dict_set(&opts, "fragment_index", "2", 0);
537 av_dict_set(&opts, "avoid_negative_ts", "0", 0);
538 av_dict_set(&opts, "use_editlist", "0", 0);
541 init_out("empty-moov-second-frag-discont");
543 av_write_frame(ctx, NULL); // Output the second fragment
545 check(!memcmp(hash, content, HASH_SIZE), "discontinuously written fragment differs");
548 // Produce the same thing by using delay_moov, which requires a slightly
549 // different call sequence.
550 av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash+frag_discont", 0);
551 av_dict_set(&opts, "fragment_index", "2", 0);
555 av_write_frame(ctx, NULL); // Output the moov
556 init_out("delay-moov-second-frag-discont");
557 av_write_frame(ctx, NULL); // Output the second fragment
559 check(!memcmp(hash, content, HASH_SIZE), "discontinuously written fragment differs");
563 // Test discontinously written fragments with b-frames (where the
564 // assumption of starting at pts=0 works) but not with audio preroll
565 // (which can't be guessed).
566 av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash", 0);
569 init_out("delay-moov-elst-init");
570 av_write_frame(ctx, NULL); // Output the moov
572 memcpy(header, hash, HASH_SIZE);
573 av_write_frame(ctx, NULL); // Output the first fragment
574 init_out("delay-moov-elst-second-frag");
576 av_write_frame(ctx, NULL); // Output the second fragment
578 memcpy(content, hash, HASH_SIZE);
581 av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash+frag_discont", 0);
582 av_dict_set(&opts, "fragment_index", "2", 0);
585 mux_gops(1); // Write the second fragment
586 init_out("delay-moov-elst-init-discont");
587 av_write_frame(ctx, NULL); // Output the moov
589 check(!memcmp(hash, header, HASH_SIZE), "discontinuously written header differs");
590 init_out("delay-moov-elst-second-frag-discont");
591 av_write_frame(ctx, NULL); // Output the second fragment
593 check(!memcmp(hash, content, HASH_SIZE), "discontinuously written fragment differs");
597 // Test discontinously written fragments with b-frames and audio preroll,
598 // properly signaled.
599 av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash", 0);
602 init_out("delay-moov-elst-signal-init");
603 av_write_frame(ctx, NULL); // Output the moov
605 memcpy(header, hash, HASH_SIZE);
606 av_write_frame(ctx, NULL); // Output the first fragment
607 init_out("delay-moov-elst-signal-second-frag");
609 av_write_frame(ctx, NULL); // Output the second fragment
611 memcpy(content, hash, HASH_SIZE);
614 av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash+frag_discont", 0);
615 av_dict_set(&opts, "fragment_index", "2", 0);
619 mux_gops(1); // Write the second fragment
620 init_out("delay-moov-elst-signal-init-discont");
621 av_write_frame(ctx, NULL); // Output the moov
623 check(!memcmp(hash, header, HASH_SIZE), "discontinuously written header differs");
624 init_out("delay-moov-elst-signal-second-frag-discont");
625 av_write_frame(ctx, NULL); // Output the second fragment
627 check(!memcmp(hash, content, HASH_SIZE), "discontinuously written fragment differs");
631 // Test VFR content, with sidx atoms (which declare the pts duration
632 // of a fragment, forcing overriding the start pts of the next one).
633 // Here, the fragment duration in pts is significantly different from
634 // the duration in dts. The video stream starts at dts=-10,pts=0, and
635 // the second fragment starts at dts=155,pts=156. The trun duration sum
636 // of the first fragment is 165, which also is written as
637 // baseMediaDecodeTime in the tfdt in the second fragment. The sidx for
638 // the first fragment says earliest_presentation_time = 0 and
639 // subsegment_duration = 156, which also matches the sidx in the second
640 // fragment. For the audio stream, the pts and dts durations also don't
641 // match - the input stream starts at pts=-2048, but that part is excluded
644 av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov+dash", 0);
646 mux_frames(gop_size/2);
648 mux_frames(gop_size/2);
653 // Test VFR content, with cleared duration fields. In these cases,
654 // the muxer must guess the duration of the last packet of each
655 // fragment. As long as the framerate doesn't vary (too much) at the
656 // fragment edge, it works just fine. Additionally, when automatically
657 // cutting fragments, the muxer already know the timestamps of the next
658 // packet for one stream (in most cases the video stream), avoiding
659 // having to use guesses for that one.
660 init_count_warnings();
662 init_out("vfr-noduration");
663 av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov+dash", 0);
665 mux_frames(gop_size/2);
667 mux_frames(gop_size/2);
672 reset_count_warnings();
673 check(num_warnings > 0, "No warnings printed for filled in durations");
677 return check_faults > 0 ? 1 : 0;