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->codecpar->codec_type = AVMEDIA_TYPE_VIDEO;
163 st->codecpar->codec_id = AV_CODEC_ID_H264;
164 st->codecpar->width = 640;
165 st->codecpar->height = 480;
166 st->time_base.num = 1;
167 st->time_base.den = 30;
168 st->codecpar->extradata_size = sizeof(h264_extradata);
169 st->codecpar->extradata = av_mallocz(st->codecpar->extradata_size + AV_INPUT_BUFFER_PADDING_SIZE);
170 if (!st->codecpar->extradata)
172 memcpy(st->codecpar->extradata, h264_extradata, sizeof(h264_extradata));
175 st = avformat_new_stream(ctx, NULL);
178 st->codecpar->codec_type = AVMEDIA_TYPE_AUDIO;
179 st->codecpar->codec_id = AV_CODEC_ID_AAC;
180 st->codecpar->sample_rate = 44100;
181 st->codecpar->channels = 2;
182 st->time_base.num = 1;
183 st->time_base.den = 44100;
184 st->codecpar->extradata_size = sizeof(aac_extradata);
185 st->codecpar->extradata = av_mallocz(st->codecpar->extradata_size + AV_INPUT_BUFFER_PADDING_SIZE);
186 if (!st->codecpar->extradata)
188 memcpy(st->codecpar->extradata, aac_extradata, sizeof(aac_extradata));
191 if (avformat_write_header(ctx, &opts) < 0)
197 duration = video_st->time_base.den / fps;
198 audio_duration = 1024LL * audio_st->time_base.den / audio_st->codecpar->sample_rate;
200 audio_preroll = 2048LL * audio_st->time_base.den / audio_st->codecpar->sample_rate;
203 video_dts = bframes ? -duration : 0;
204 audio_dts = -audio_preroll;
207 static void init(int bf, int audio_preroll)
209 init_fps(bf, audio_preroll, 30);
212 static void mux_frames(int n)
214 int end_frames = frames + n;
217 uint8_t pktdata[8] = { 0 };
218 av_init_packet(&pkt);
220 if (av_compare_ts(audio_dts, audio_st->time_base, video_dts, video_st->time_base) < 0) {
221 pkt.dts = pkt.pts = audio_dts;
222 pkt.stream_index = 1;
223 pkt.duration = audio_duration;
224 audio_dts += audio_duration;
226 if (frames == end_frames)
229 pkt.stream_index = 0;
230 pkt.duration = duration;
231 if ((frames % gop_size) == 0) {
232 pkt.flags |= AV_PKT_FLAG_KEY;
233 last_picture = AV_PICTURE_TYPE_I;
234 pkt.pts = pkt.dts + duration;
237 if (last_picture == AV_PICTURE_TYPE_P) {
238 last_picture = AV_PICTURE_TYPE_B;
240 video_dts = next_p_pts;
242 last_picture = AV_PICTURE_TYPE_P;
243 if (((frames + 1) % gop_size) == 0) {
244 pkt.pts = pkt.dts + duration;
247 next_p_pts = pkt.pts = pkt.dts + 2 * duration;
248 video_dts += duration;
259 AV_WB32(pktdata + 4, pkt.pts);
264 if (skip_write_audio && pkt.stream_index == 1)
266 av_write_frame(ctx, &pkt);
270 static void mux_gops(int n)
272 mux_frames(gop_size * n);
275 static void skip_gops(int n)
282 static void signal_init_ts(void)
285 av_init_packet(&pkt);
289 pkt.stream_index = 0;
292 av_write_frame(ctx, &pkt);
294 pkt.stream_index = 1;
295 pkt.dts = pkt.pts = audio_dts;
296 av_write_frame(ctx, &pkt);
299 static void finish(void)
301 av_write_trailer(ctx);
303 avformat_free_context(ctx);
307 static void help(void)
309 printf("movenc-test [-w]\n"
310 "-w write output into files\n");
313 int main(int argc, char **argv)
316 uint8_t header[HASH_SIZE];
317 uint8_t content[HASH_SIZE];
322 c = getopt(argc, argv, "wh");
338 md5 = av_md5_alloc();
342 // Write a fragmented file with an initial moov that actually contains some
343 // samples. One moov+mdat with 1 second of data and one moof+mdat with 1
345 init_out("non-empty-moov");
346 av_dict_set(&opts, "movflags", "frag_keyframe", 0);
352 // Write a similar file, but with b-frames and audio preroll, handled
354 init_out("non-empty-moov-elst");
355 av_dict_set(&opts, "movflags", "frag_keyframe", 0);
356 av_dict_set(&opts, "use_editlist", "1", 0);
362 // Use b-frames but no audio-preroll, but without an edit list.
363 // Due to avoid_negative_ts == AVFMT_AVOID_NEG_TS_MAKE_ZERO, the dts
364 // of the first audio packet is > 0, but it is set to zero since edit
365 // lists aren't used, increasing the duration of the first packet instead.
366 init_out("non-empty-moov-no-elst");
367 av_dict_set(&opts, "movflags", "frag_keyframe", 0);
368 av_dict_set(&opts, "use_editlist", "0", 0);
375 // Write an ISMV, with b-frames and audio preroll.
377 av_dict_set(&opts, "movflags", "frag_keyframe", 0);
384 // An initial moov that doesn't contain any samples, followed by two
386 init_out("empty-moov");
387 av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
388 av_dict_set(&opts, "use_editlist", "0", 0);
393 memcpy(content, hash, HASH_SIZE);
395 // Similar to the previous one, but with input that doesn't start at
396 // pts/dts 0. avoid_negative_ts behaves in the same way as
397 // in non-empty-moov-no-elst above.
398 init_out("empty-moov-no-elst");
399 av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
405 // Same as the previous one, but disable avoid_negative_ts (which
406 // would require using an edit list, but with empty_moov, one can't
407 // write a sensible edit list, when the start timestamps aren't known).
408 // This should trigger a warning - we check that the warning is produced.
409 init_count_warnings();
410 init_out("empty-moov-no-elst-no-adjust");
411 av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
412 av_dict_set(&opts, "avoid_negative_ts", "0", 0);
418 reset_count_warnings();
419 check(num_warnings > 0, "No warnings printed for unhandled start offset");
421 // Verify that delay_moov produces the same as empty_moov for
423 init_out("delay-moov");
424 av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov", 0);
425 av_dict_set(&opts, "use_editlist", "0", 0);
430 check(!memcmp(hash, content, HASH_SIZE), "delay_moov differs from empty_moov");
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);
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);
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
453 check(out_size > 0, "delay_moov didn't output anything");
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);
461 av_write_frame(ctx, NULL); // Force writing the moov
462 check(out_size > 0, "No moov written");
463 av_write_frame(ctx, NULL);
465 av_write_frame(ctx, NULL);
469 skip_write_audio = 0;
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 av_dict_set(&opts, "use_editlist", "0", 0);
480 memcpy(header, hash, HASH_SIZE);
481 init_out("empty-moov-content");
483 // Written 2 seconds of content, with an automatic flush after 1 second.
484 check(out_size > 0, "No automatic flush?");
485 empty_moov_pos = prev_pos = out_size;
486 // Manually flush the second fragment
487 av_write_frame(ctx, NULL);
488 check(out_size > prev_pos, "No second fragment flushed?");
490 // Check that an extra flush doesn't output any more data
491 av_write_frame(ctx, NULL);
492 check(out_size == prev_pos, "More data written?");
494 memcpy(content, hash, HASH_SIZE);
495 // Ignore the trailer written here
498 init_out("delay-moov-header");
499 av_dict_set(&opts, "movflags", "frag_custom+delay_moov", 0);
500 av_dict_set(&opts, "use_editlist", "0", 0);
502 check(out_size == 0, "Output written during init with delay_moov");
503 mux_gops(1); // Write 1 second of content
504 av_write_frame(ctx, NULL); // Force writing the moov
506 check(!memcmp(hash, header, HASH_SIZE), "delay_moov header differs from empty_moov");
507 init_out("delay-moov-content");
508 av_write_frame(ctx, NULL); // Flush the first fragment
509 check(out_size == empty_moov_pos, "Manually flushed content differs from automatically flushed, %d vs %d", out_size, empty_moov_pos);
510 mux_gops(1); // Write the rest of the content
511 av_write_frame(ctx, NULL); // Flush the second fragment
513 check(!memcmp(hash, content, HASH_SIZE), "delay_moov content differs from empty_moov");
517 // Verify that we can produce an identical second fragment without
518 // writing the first one. First write the reference fragments that
519 // we want to reproduce.
520 av_dict_set(&opts, "movflags", "frag_custom+empty_moov+dash", 0);
523 av_write_frame(ctx, NULL); // Output the first fragment
524 init_out("empty-moov-second-frag");
526 av_write_frame(ctx, NULL); // Output the second fragment
528 memcpy(content, hash, HASH_SIZE);
531 // Produce the same second fragment without actually writing the first
533 av_dict_set(&opts, "movflags", "frag_custom+empty_moov+dash+frag_discont", 0);
534 av_dict_set(&opts, "fragment_index", "2", 0);
535 av_dict_set(&opts, "avoid_negative_ts", "0", 0);
536 av_dict_set(&opts, "use_editlist", "0", 0);
539 init_out("empty-moov-second-frag-discont");
541 av_write_frame(ctx, NULL); // Output the second fragment
543 check(!memcmp(hash, content, HASH_SIZE), "discontinuously written fragment differs");
546 // Produce the same thing by using delay_moov, which requires a slightly
547 // different call sequence.
548 av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash+frag_discont", 0);
549 av_dict_set(&opts, "fragment_index", "2", 0);
553 av_write_frame(ctx, NULL); // Output the moov
554 init_out("delay-moov-second-frag-discont");
555 av_write_frame(ctx, NULL); // Output the second fragment
557 check(!memcmp(hash, content, HASH_SIZE), "discontinuously written fragment differs");
561 // Test discontinously written fragments with b-frames (where the
562 // assumption of starting at pts=0 works) but not with audio preroll
563 // (which can't be guessed).
564 av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash", 0);
567 init_out("delay-moov-elst-init");
568 av_write_frame(ctx, NULL); // Output the moov
570 memcpy(header, hash, HASH_SIZE);
571 av_write_frame(ctx, NULL); // Output the first fragment
572 init_out("delay-moov-elst-second-frag");
574 av_write_frame(ctx, NULL); // Output the second fragment
576 memcpy(content, hash, HASH_SIZE);
579 av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash+frag_discont", 0);
580 av_dict_set(&opts, "fragment_index", "2", 0);
583 mux_gops(1); // Write the second fragment
584 init_out("delay-moov-elst-init-discont");
585 av_write_frame(ctx, NULL); // Output the moov
587 check(!memcmp(hash, header, HASH_SIZE), "discontinuously written header differs");
588 init_out("delay-moov-elst-second-frag-discont");
589 av_write_frame(ctx, NULL); // Output the second fragment
591 check(!memcmp(hash, content, HASH_SIZE), "discontinuously written fragment differs");
595 // Test discontinously written fragments with b-frames and audio preroll,
596 // properly signaled.
597 av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash", 0);
600 init_out("delay-moov-elst-signal-init");
601 av_write_frame(ctx, NULL); // Output the moov
603 memcpy(header, hash, HASH_SIZE);
604 av_write_frame(ctx, NULL); // Output the first fragment
605 init_out("delay-moov-elst-signal-second-frag");
607 av_write_frame(ctx, NULL); // Output the second fragment
609 memcpy(content, hash, HASH_SIZE);
612 av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash+frag_discont", 0);
613 av_dict_set(&opts, "fragment_index", "2", 0);
617 mux_gops(1); // Write the second fragment
618 init_out("delay-moov-elst-signal-init-discont");
619 av_write_frame(ctx, NULL); // Output the moov
621 check(!memcmp(hash, header, HASH_SIZE), "discontinuously written header differs");
622 init_out("delay-moov-elst-signal-second-frag-discont");
623 av_write_frame(ctx, NULL); // Output the second fragment
625 check(!memcmp(hash, content, HASH_SIZE), "discontinuously written fragment differs");
629 // Test VFR content, with sidx atoms (which declare the pts duration
630 // of a fragment, forcing overriding the start pts of the next one).
631 // Here, the fragment duration in pts is significantly different from
632 // the duration in dts. The video stream starts at dts=-10,pts=0, and
633 // the second fragment starts at dts=155,pts=156. The trun duration sum
634 // of the first fragment is 165, which also is written as
635 // baseMediaDecodeTime in the tfdt in the second fragment. The sidx for
636 // the first fragment says earliest_presentation_time = 0 and
637 // subsegment_duration = 156, which also matches the sidx in the second
638 // fragment. For the audio stream, the pts and dts durations also don't
639 // match - the input stream starts at pts=-2048, but that part is excluded
642 av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov+dash", 0);
644 mux_frames(gop_size/2);
646 mux_frames(gop_size/2);
651 // Test VFR content, with cleared duration fields. In these cases,
652 // the muxer must guess the duration of the last packet of each
653 // fragment. As long as the framerate doesn't vary (too much) at the
654 // fragment edge, it works just fine. Additionally, when automatically
655 // cutting fragments, the muxer already know the timestamps of the next
656 // packet for one stream (in most cases the video stream), avoiding
657 // having to use guesses for that one.
658 init_count_warnings();
660 init_out("vfr-noduration");
661 av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov+dash", 0);
663 mux_frames(gop_size/2);
665 mux_frames(gop_size/2);
670 reset_count_warnings();
671 check(num_warnings > 0, "No warnings printed for filled in durations");
675 return check_faults > 0 ? 1 : 0;