summaryrefslogtreecommitdiff
path: root/libavformat/movenc-test.c
diff options
context:
space:
mode:
authorDiego Biurrun <diego@biurrun.de>2016-04-20 11:40:40 +0200
committerDiego Biurrun <diego@biurrun.de>2016-05-13 14:55:56 +0200
commita6a750c7ef240b72ce01e9653343a0ddf247d196 (patch)
tree5c472bb92b5264f19b9270b0caf8a8565b071cef /libavformat/movenc-test.c
parent257f00ec1ab06a2a161f535036c6512f3fc8e801 (diff)
tests: Move all test programs to a subdirectory
Diffstat (limited to 'libavformat/movenc-test.c')
-rw-r--r--libavformat/movenc-test.c672
1 files changed, 0 insertions, 672 deletions
diff --git a/libavformat/movenc-test.c b/libavformat/movenc-test.c
deleted file mode 100644
index e7b67e1e7a..0000000000
--- a/libavformat/movenc-test.c
+++ /dev/null
@@ -1,672 +0,0 @@
-/*
- * Copyright (c) 2015 Martin Storsjo
- *
- * This file is part of Libav.
- *
- * Libav is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation; either
- * version 2.1 of the License, or (at your option) any later version.
- *
- * Libav is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
- * Lesser General Public License for more details.
- *
- * You should have received a copy of the GNU Lesser General Public
- * License along with Libav; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
- */
-
-#include "config.h"
-
-#include "libavutil/intreadwrite.h"
-#include "libavutil/mathematics.h"
-#include "libavutil/md5.h"
-
-#include "avformat.h"
-
-#if HAVE_UNISTD_H
-#include <unistd.h>
-#endif
-
-#if !HAVE_GETOPT
-#include "compat/getopt.c"
-#endif
-
-#define HASH_SIZE 16
-
-static const uint8_t h264_extradata[] = {
- 0x01, 0x4d, 0x40, 0x1e, 0xff, 0xe1, 0x00, 0x02, 0x67, 0x4d, 0x01, 0x00, 0x02, 0x68, 0xef
-};
-static const uint8_t aac_extradata[] = {
- 0x12, 0x10
-};
-
-
-const char *format = "mp4";
-AVFormatContext *ctx;
-uint8_t iobuf[32768];
-AVDictionary *opts;
-
-int write_file;
-const char *cur_name;
-FILE* out;
-int out_size;
-struct AVMD5* md5;
-uint8_t hash[HASH_SIZE];
-
-AVStream *video_st, *audio_st;
-int64_t audio_dts, video_dts;
-
-int bframes;
-int64_t duration;
-int64_t audio_duration;
-int frames;
-int gop_size;
-int64_t next_p_pts;
-enum AVPictureType last_picture;
-int skip_write;
-int skip_write_audio;
-int clear_duration;
-
-int num_warnings;
-
-int check_faults;
-
-
-static void count_warnings(void *avcl, int level, const char *fmt, va_list vl)
-{
- if (level == AV_LOG_WARNING)
- num_warnings++;
-}
-
-static void init_count_warnings(void)
-{
- av_log_set_callback(count_warnings);
- num_warnings = 0;
-}
-
-static void reset_count_warnings(void)
-{
- av_log_set_callback(av_log_default_callback);
-}
-
-static int io_write(void *opaque, uint8_t *buf, int size)
-{
- out_size += size;
- av_md5_update(md5, buf, size);
- if (out)
- fwrite(buf, 1, size, out);
- return size;
-}
-
-static void init_out(const char *name)
-{
- char buf[100];
- cur_name = name;
- snprintf(buf, sizeof(buf), "%s.%s", cur_name, format);
-
- av_md5_init(md5);
- if (write_file) {
- out = fopen(buf, "wb");
- if (!out)
- perror(buf);
- }
- out_size = 0;
-}
-
-static void close_out(void)
-{
- int i;
- av_md5_final(md5, hash);
- for (i = 0; i < HASH_SIZE; i++)
- printf("%02x", hash[i]);
- printf(" %d %s\n", out_size, cur_name);
- if (out)
- fclose(out);
- out = NULL;
-}
-
-static void check_func(int value, int line, const char *msg, ...)
-{
- if (!value) {
- va_list ap;
- va_start(ap, msg);
- printf("%d: ", line);
- vprintf(msg, ap);
- printf("\n");
- check_faults++;
- va_end(ap);
- }
-}
-#define check(value, ...) check_func(value, __LINE__, __VA_ARGS__)
-
-static void init_fps(int bf, int audio_preroll, int fps)
-{
- AVStream *st;
- ctx = avformat_alloc_context();
- if (!ctx)
- exit(1);
- ctx->oformat = av_guess_format(format, NULL, NULL);
- if (!ctx->oformat)
- exit(1);
- ctx->pb = avio_alloc_context(iobuf, sizeof(iobuf), AVIO_FLAG_WRITE, NULL, NULL, io_write, NULL);
- if (!ctx->pb)
- exit(1);
- ctx->flags |= AVFMT_FLAG_BITEXACT;
-
- st = avformat_new_stream(ctx, NULL);
- if (!st)
- exit(1);
- st->codecpar->codec_type = AVMEDIA_TYPE_VIDEO;
- st->codecpar->codec_id = AV_CODEC_ID_H264;
- st->codecpar->width = 640;
- st->codecpar->height = 480;
- st->time_base.num = 1;
- st->time_base.den = 30;
- st->codecpar->extradata_size = sizeof(h264_extradata);
- st->codecpar->extradata = av_mallocz(st->codecpar->extradata_size + AV_INPUT_BUFFER_PADDING_SIZE);
- if (!st->codecpar->extradata)
- exit(1);
- memcpy(st->codecpar->extradata, h264_extradata, sizeof(h264_extradata));
- video_st = st;
-
- st = avformat_new_stream(ctx, NULL);
- if (!st)
- exit(1);
- st->codecpar->codec_type = AVMEDIA_TYPE_AUDIO;
- st->codecpar->codec_id = AV_CODEC_ID_AAC;
- st->codecpar->sample_rate = 44100;
- st->codecpar->channels = 2;
- st->time_base.num = 1;
- st->time_base.den = 44100;
- st->codecpar->extradata_size = sizeof(aac_extradata);
- st->codecpar->extradata = av_mallocz(st->codecpar->extradata_size + AV_INPUT_BUFFER_PADDING_SIZE);
- if (!st->codecpar->extradata)
- exit(1);
- memcpy(st->codecpar->extradata, aac_extradata, sizeof(aac_extradata));
- audio_st = st;
-
- if (avformat_write_header(ctx, &opts) < 0)
- exit(1);
- av_dict_free(&opts);
-
- frames = 0;
- gop_size = 30;
- duration = video_st->time_base.den / fps;
- audio_duration = 1024LL * audio_st->time_base.den / audio_st->codecpar->sample_rate;
- if (audio_preroll)
- audio_preroll = 2048LL * audio_st->time_base.den / audio_st->codecpar->sample_rate;
-
- bframes = bf;
- video_dts = bframes ? -duration : 0;
- audio_dts = -audio_preroll;
-}
-
-static void init(int bf, int audio_preroll)
-{
- init_fps(bf, audio_preroll, 30);
-}
-
-static void mux_frames(int n)
-{
- int end_frames = frames + n;
- while (1) {
- AVPacket pkt;
- uint8_t pktdata[4];
- av_init_packet(&pkt);
-
- if (av_compare_ts(audio_dts, audio_st->time_base, video_dts, video_st->time_base) < 0) {
- pkt.dts = pkt.pts = audio_dts;
- pkt.stream_index = 1;
- pkt.duration = audio_duration;
- audio_dts += audio_duration;
- } else {
- if (frames == end_frames)
- break;
- pkt.dts = video_dts;
- pkt.stream_index = 0;
- pkt.duration = duration;
- if ((frames % gop_size) == 0) {
- pkt.flags |= AV_PKT_FLAG_KEY;
- last_picture = AV_PICTURE_TYPE_I;
- pkt.pts = pkt.dts + duration;
- video_dts = pkt.pts;
- } else {
- if (last_picture == AV_PICTURE_TYPE_P) {
- last_picture = AV_PICTURE_TYPE_B;
- pkt.pts = pkt.dts;
- video_dts = next_p_pts;
- } else {
- last_picture = AV_PICTURE_TYPE_P;
- if (((frames + 1) % gop_size) == 0) {
- pkt.pts = pkt.dts + duration;
- video_dts = pkt.pts;
- } else {
- next_p_pts = pkt.pts = pkt.dts + 2 * duration;
- video_dts += duration;
- }
- }
- }
- if (!bframes)
- pkt.pts = pkt.dts;
- frames++;
- }
-
- if (clear_duration)
- pkt.duration = 0;
- AV_WB32(pktdata, pkt.pts);
- pkt.data = pktdata;
- pkt.size = 4;
- if (skip_write)
- continue;
- if (skip_write_audio && pkt.stream_index == 1)
- continue;
- av_write_frame(ctx, &pkt);
- }
-}
-
-static void mux_gops(int n)
-{
- mux_frames(gop_size * n);
-}
-
-static void skip_gops(int n)
-{
- skip_write = 1;
- mux_gops(n);
- skip_write = 0;
-}
-
-static void signal_init_ts(void)
-{
- AVPacket pkt;
- av_init_packet(&pkt);
- pkt.size = 0;
- pkt.data = NULL;
-
- pkt.stream_index = 0;
- pkt.dts = video_dts;
- pkt.pts = 0;
- av_write_frame(ctx, &pkt);
-
- pkt.stream_index = 1;
- pkt.dts = pkt.pts = audio_dts;
- av_write_frame(ctx, &pkt);
-}
-
-static void finish(void)
-{
- av_write_trailer(ctx);
- av_free(ctx->pb);
- avformat_free_context(ctx);
- ctx = NULL;
-}
-
-static void help(void)
-{
- printf("movenc-test [-w]\n"
- "-w write output into files\n");
-}
-
-int main(int argc, char **argv)
-{
- int c;
- uint8_t header[HASH_SIZE];
- uint8_t content[HASH_SIZE];
- int empty_moov_pos;
- int prev_pos;
-
- for (;;) {
- c = getopt(argc, argv, "wh");
- if (c == -1)
- break;
- switch (c) {
- case 'w':
- write_file = 1;
- break;
- default:
- case 'h':
- help();
- return 0;
- }
- }
-
- av_register_all();
-
- md5 = av_md5_alloc();
- if (!md5)
- return 1;
-
- // Write a fragmented file with an initial moov that actually contains some
- // samples. One moov+mdat with 1 second of data and one moof+mdat with 1
- // second of data.
- init_out("non-empty-moov");
- av_dict_set(&opts, "movflags", "frag_keyframe", 0);
- init(0, 0);
- mux_gops(2);
- finish();
- close_out();
-
- // Write a similar file, but with B-frames and audio preroll, handled
- // via an edit list.
- init_out("non-empty-moov-elst");
- av_dict_set(&opts, "movflags", "frag_keyframe", 0);
- av_dict_set(&opts, "use_editlist", "1", 0);
- init(1, 1);
- mux_gops(2);
- finish();
- close_out();
-
- // Use B-frames but no audio-preroll, but without an edit list.
- // Due to avoid_negative_ts == AVFMT_AVOID_NEG_TS_MAKE_ZERO, the dts
- // of the first audio packet is > 0, but it is set to zero since edit
- // lists aren't used, increasing the duration of the first packet instead.
- init_out("non-empty-moov-no-elst");
- av_dict_set(&opts, "movflags", "frag_keyframe", 0);
- av_dict_set(&opts, "use_editlist", "0", 0);
- init(1, 0);
- mux_gops(2);
- finish();
- close_out();
-
- format = "ismv";
- // Write an ISMV, with B-frames and audio preroll.
- init_out("ismv");
- av_dict_set(&opts, "movflags", "frag_keyframe", 0);
- init(1, 1);
- mux_gops(2);
- finish();
- close_out();
- format = "mp4";
-
- // An initial moov that doesn't contain any samples, followed by two
- // moof+mdat pairs.
- init_out("empty-moov");
- av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
- init(0, 0);
- mux_gops(2);
- finish();
- close_out();
- memcpy(content, hash, HASH_SIZE);
-
- // Similar to the previous one, but with input that doesn't start at
- // pts/dts 0. avoid_negative_ts behaves in the same way as
- // in non-empty-moov-no-elst above.
- init_out("empty-moov-no-elst");
- av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
- init(1, 0);
- mux_gops(2);
- finish();
- close_out();
-
- // Same as the previous one, but disable avoid_negative_ts (which
- // would require using an edit list, but with empty_moov, one can't
- // write a sensible edit list, when the start timestamps aren't known).
- // This should trigger a warning - we check that the warning is produced.
- init_count_warnings();
- init_out("empty-moov-no-elst-no-adjust");
- av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
- av_dict_set(&opts, "avoid_negative_ts", "0", 0);
- init(1, 0);
- mux_gops(2);
- finish();
- close_out();
-
- reset_count_warnings();
- check(num_warnings > 0, "No warnings printed for unhandled start offset");
-
- // Verify that delay_moov produces the same as empty_moov for
- // simple input
- init_out("delay-moov");
- av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov", 0);
- init(0, 0);
- mux_gops(2);
- finish();
- close_out();
- check(!memcmp(hash, content, HASH_SIZE), "delay_moov differs from empty_moov");
-
- // Test writing content that requires an edit list using delay_moov
- init_out("delay-moov-elst");
- av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov", 0);
- init(1, 1);
- mux_gops(2);
- finish();
- close_out();
-
- // Test writing a file with one track lacking packets, with delay_moov.
- skip_write_audio = 1;
- init_out("delay-moov-empty-track");
- av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov", 0);
- init(0, 0);
- mux_gops(2);
- // The automatic flushing shouldn't output anything, since we're still
- // waiting for data for some tracks
- check(out_size == 0, "delay_moov flushed prematurely");
- // When closed (or manually flushed), all the written data should still
- // be output.
- finish();
- close_out();
- check(out_size > 0, "delay_moov didn't output anything");
-
- // Check that manually flushing still outputs things as expected. This
- // produces two fragments, while the one above produces only one.
- init_out("delay-moov-empty-track-flush");
- av_dict_set(&opts, "movflags", "frag_custom+delay_moov", 0);
- init(0, 0);
- mux_gops(1);
- av_write_frame(ctx, NULL); // Force writing the moov
- check(out_size > 0, "No moov written");
- av_write_frame(ctx, NULL);
- mux_gops(1);
- av_write_frame(ctx, NULL);
- finish();
- close_out();
-
- skip_write_audio = 0;
-
-
-
- // Verify that the header written by delay_moov when manually flushed
- // is identical to the one by empty_moov.
- init_out("empty-moov-header");
- av_dict_set(&opts, "movflags", "frag_keyframe+empty_moov", 0);
- init(0, 0);
- close_out();
- memcpy(header, hash, HASH_SIZE);
- init_out("empty-moov-content");
- mux_gops(2);
- // Written 2 seconds of content, with an automatic flush after 1 second.
- check(out_size > 0, "No automatic flush?");
- empty_moov_pos = prev_pos = out_size;
- // Manually flush the second fragment
- av_write_frame(ctx, NULL);
- check(out_size > prev_pos, "No second fragment flushed?");
- prev_pos = out_size;
- // Check that an extra flush doesn't output any more data
- av_write_frame(ctx, NULL);
- check(out_size == prev_pos, "More data written?");
- close_out();
- memcpy(content, hash, HASH_SIZE);
- // Ignore the trailer written here
- finish();
-
- init_out("delay-moov-header");
- av_dict_set(&opts, "movflags", "frag_custom+delay_moov", 0);
- init(0, 0);
- check(out_size == 0, "Output written during init with delay_moov");
- mux_gops(1); // Write 1 second of content
- av_write_frame(ctx, NULL); // Force writing the moov
- close_out();
- check(!memcmp(hash, header, HASH_SIZE), "delay_moov header differs from empty_moov");
- init_out("delay-moov-content");
- av_write_frame(ctx, NULL); // Flush the first fragment
- check(out_size == empty_moov_pos, "Manually flushed content differs from automatically flushed, %d vs %d", out_size, empty_moov_pos);
- mux_gops(1); // Write the rest of the content
- av_write_frame(ctx, NULL); // Flush the second fragment
- close_out();
- check(!memcmp(hash, content, HASH_SIZE), "delay_moov content differs from empty_moov");
- finish();
-
-
- // Verify that we can produce an identical second fragment without
- // writing the first one. First write the reference fragments that
- // we want to reproduce.
- av_dict_set(&opts, "movflags", "frag_custom+empty_moov+dash", 0);
- init(0, 0);
- mux_gops(1);
- av_write_frame(ctx, NULL); // Output the first fragment
- init_out("empty-moov-second-frag");
- mux_gops(1);
- av_write_frame(ctx, NULL); // Output the second fragment
- close_out();
- memcpy(content, hash, HASH_SIZE);
- finish();
-
- // Produce the same second fragment without actually writing the first
- // one before.
- av_dict_set(&opts, "movflags", "frag_custom+empty_moov+dash+frag_discont", 0);
- av_dict_set(&opts, "fragment_index", "2", 0);
- av_dict_set(&opts, "avoid_negative_ts", "0", 0);
- av_dict_set(&opts, "use_editlist", "0", 0);
- init(0, 0);
- skip_gops(1);
- init_out("empty-moov-second-frag-discont");
- mux_gops(1);
- av_write_frame(ctx, NULL); // Output the second fragment
- close_out();
- check(!memcmp(hash, content, HASH_SIZE), "discontinuously written fragment differs");
- finish();
-
- // Produce the same thing by using delay_moov, which requires a slightly
- // different call sequence.
- av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash+frag_discont", 0);
- av_dict_set(&opts, "fragment_index", "2", 0);
- init(0, 0);
- skip_gops(1);
- mux_gops(1);
- av_write_frame(ctx, NULL); // Output the moov
- init_out("delay-moov-second-frag-discont");
- av_write_frame(ctx, NULL); // Output the second fragment
- close_out();
- check(!memcmp(hash, content, HASH_SIZE), "discontinuously written fragment differs");
- finish();
-
-
- // Test discontinuously written fragments with B-frames (where the
- // assumption of starting at pts=0 works) but not with audio preroll
- // (which can't be guessed).
- av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash", 0);
- init(1, 0);
- mux_gops(1);
- init_out("delay-moov-elst-init");
- av_write_frame(ctx, NULL); // Output the moov
- close_out();
- memcpy(header, hash, HASH_SIZE);
- av_write_frame(ctx, NULL); // Output the first fragment
- init_out("delay-moov-elst-second-frag");
- mux_gops(1);
- av_write_frame(ctx, NULL); // Output the second fragment
- close_out();
- memcpy(content, hash, HASH_SIZE);
- finish();
-
- av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash+frag_discont", 0);
- av_dict_set(&opts, "fragment_index", "2", 0);
- init(1, 0);
- skip_gops(1);
- mux_gops(1); // Write the second fragment
- init_out("delay-moov-elst-init-discont");
- av_write_frame(ctx, NULL); // Output the moov
- close_out();
- check(!memcmp(hash, header, HASH_SIZE), "discontinuously written header differs");
- init_out("delay-moov-elst-second-frag-discont");
- av_write_frame(ctx, NULL); // Output the second fragment
- close_out();
- check(!memcmp(hash, content, HASH_SIZE), "discontinuously written fragment differs");
- finish();
-
-
- // Test discontinuously written fragments with B-frames and audio preroll,
- // properly signaled.
- av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash", 0);
- init(1, 1);
- mux_gops(1);
- init_out("delay-moov-elst-signal-init");
- av_write_frame(ctx, NULL); // Output the moov
- close_out();
- memcpy(header, hash, HASH_SIZE);
- av_write_frame(ctx, NULL); // Output the first fragment
- init_out("delay-moov-elst-signal-second-frag");
- mux_gops(1);
- av_write_frame(ctx, NULL); // Output the second fragment
- close_out();
- memcpy(content, hash, HASH_SIZE);
- finish();
-
- av_dict_set(&opts, "movflags", "frag_custom+delay_moov+dash+frag_discont", 0);
- av_dict_set(&opts, "fragment_index", "2", 0);
- init(1, 1);
- signal_init_ts();
- skip_gops(1);
- mux_gops(1); // Write the second fragment
- init_out("delay-moov-elst-signal-init-discont");
- av_write_frame(ctx, NULL); // Output the moov
- close_out();
- check(!memcmp(hash, header, HASH_SIZE), "discontinuously written header differs");
- init_out("delay-moov-elst-signal-second-frag-discont");
- av_write_frame(ctx, NULL); // Output the second fragment
- close_out();
- check(!memcmp(hash, content, HASH_SIZE), "discontinuously written fragment differs");
- finish();
-
-
- // Test VFR content, with sidx atoms (which declare the pts duration
- // of a fragment, forcing overriding the start pts of the next one).
- // Here, the fragment duration in pts is significantly different from
- // the duration in dts. The video stream starts at dts=-10,pts=0, and
- // the second fragment starts at dts=155,pts=156. The trun duration sum
- // of the first fragment is 165, which also is written as
- // baseMediaDecodeTime in the tfdt in the second fragment. The sidx for
- // the first fragment says earliest_presentation_time = 0 and
- // subsegment_duration = 156, which also matches the sidx in the second
- // fragment. For the audio stream, the pts and dts durations also don't
- // match - the input stream starts at pts=-2048, but that part is excluded
- // by the edit list.
- init_out("vfr");
- av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov+dash", 0);
- init_fps(1, 1, 3);
- mux_frames(gop_size/2);
- duration /= 10;
- mux_frames(gop_size/2);
- mux_gops(1);
- finish();
- close_out();
-
- // Test VFR content, with cleared duration fields. In these cases,
- // the muxer must guess the duration of the last packet of each
- // fragment. As long as the framerate doesn't vary (too much) at the
- // fragment edge, it works just fine. Additionally, when automatically
- // cutting fragments, the muxer already know the timestamps of the next
- // packet for one stream (in most cases the video stream), avoiding
- // having to use guesses for that one.
- init_count_warnings();
- clear_duration = 1;
- init_out("vfr-noduration");
- av_dict_set(&opts, "movflags", "frag_keyframe+delay_moov+dash", 0);
- init_fps(1, 1, 3);
- mux_frames(gop_size/2);
- duration /= 10;
- mux_frames(gop_size/2);
- mux_gops(1);
- finish();
- close_out();
- clear_duration = 0;
- reset_count_warnings();
- check(num_warnings > 0, "No warnings printed for filled in durations");
-
- av_free(md5);
-
- return check_faults > 0 ? 1 : 0;
-}