3 This file is part of the HandBrake source code.
4 Homepage: <http://handbrake.fr/>.
5 It may be used under the terms of the GNU General Public License. */
14 #include "a52dec/a52.h"
15 #include "mp4v2/mp4v2.h"
17 #define min(a, b) a < b ? a : b
20 * This table defines how ISO MPEG stream type codes map to HandBrake
21 * codecs. It is indexed by the 8 bit stream type and contains the codec
22 * worker object id and a parameter for that worker proc (ignored except
23 * for the ffmpeg-based codecs in which case it is the ffmpeg codec id).
25 * Entries with a worker proc id of 0 or a kind of 'U' indicate that HB
26 * doesn't handle the stream type.
29 enum { N, U, A, V } kind; /* not handled / unknown / audio / video */
30 int codec; /* HB worker object id of codec */
31 int codec_param; /* param for codec (usually ffmpeg codec id) */
32 const char* name; /* description of type */
35 #define st(id, kind, codec, codec_param, name) \
36 [id] = { kind, codec, codec_param, name }
38 static const stream2codec_t st2codec[256] = {
39 st(0x01, V, WORK_DECMPEG2, 0, "MPEG1"),
40 st(0x02, V, WORK_DECMPEG2, 0, "MPEG2"),
41 st(0x03, A, HB_ACODEC_MPGA, CODEC_ID_MP2, "MPEG1"),
42 st(0x04, A, HB_ACODEC_MPGA, CODEC_ID_MP2, "MPEG2"),
43 st(0x05, N, 0, 0, "ISO 13818-1 private section"),
44 st(0x06, U, 0, 0, "ISO 13818-1 PES private data"),
45 st(0x07, N, 0, 0, "ISO 13522 MHEG"),
46 st(0x08, N, 0, 0, "ISO 13818-1 DSM-CC"),
47 st(0x09, N, 0, 0, "ISO 13818-1 auxiliary"),
48 st(0x0a, N, 0, 0, "ISO 13818-6 encap"),
49 st(0x0b, N, 0, 0, "ISO 13818-6 DSM-CC U-N msgs"),
50 st(0x0c, N, 0, 0, "ISO 13818-6 Stream descriptors"),
51 st(0x0d, N, 0, 0, "ISO 13818-6 Sections"),
52 st(0x0e, N, 0, 0, "ISO 13818-1 auxiliary"),
53 st(0x0f, A, HB_ACODEC_MPGA, CODEC_ID_AAC, "ISO 13818-7 AAC Audio"),
54 st(0x10, V, WORK_DECAVCODECV, CODEC_ID_MPEG4, "MPEG4"),
55 st(0x11, A, HB_ACODEC_MPGA, CODEC_ID_AAC_LATM, "MPEG4 LATM AAC"),
56 st(0x12, U, 0, 0, "MPEG4 generic"),
58 st(0x14, N, 0, 0, "ISO 13818-6 DSM-CC download"),
60 st(0x1b, V, WORK_DECAVCODECV, CODEC_ID_H264, "H.264"),
62 st(0x80, N, 0, 0, "DigiCipher II Video"),
63 st(0x81, A, HB_ACODEC_AC3, 0, "AC-3"),
64 st(0x82, A, HB_ACODEC_DCA, 0, "HDMV DTS"),
65 st(0x83, A, HB_ACODEC_LPCM, 0, "LPCM"),
66 st(0x84, A, 0, 0, "SDDS"),
67 st(0x85, U, 0, 0, "ATSC Program ID"),
68 st(0x86, A, HB_ACODEC_DCA, 0, "DTS-HD"),
69 st(0x87, A, 0, 0, "E-AC-3"),
71 st(0x8a, A, HB_ACODEC_DCA, 0, "DTS"),
73 st(0x91, A, HB_ACODEC_AC3, 0, "AC-3"),
74 st(0x92, N, 0, 0, "Subtitle"),
76 st(0x94, A, 0, 0, "SDDS"),
77 st(0xa0, V, 0, 0, "MSCODEC"),
79 st(0xea, V, WORK_DECAVCODECV, CODEC_ID_VC1, "VC1"),
84 hb_stream_type_unknown = 0,
91 #define kMaxNumberVideoPIDS 1
92 #define kMaxNumberAudioPIDS 31
93 #define kMaxNumberDecodeStreams (kMaxNumberVideoPIDS+kMaxNumberAudioPIDS)
94 #define kMaxNumberPMTStreams 32
99 int frames; /* video frames so far */
100 int errors; /* total errors so far */
101 int last_error_frame; /* frame # at last error message */
102 int last_error_count; /* # errors at last error message */
103 int packetsize; /* Transport Stream packet size */
105 uint8_t need_keyframe; // non-zero if want to start at a keyframe
106 uint8_t ts_found_pcr; // non-zero if we've found at least one input pcr
107 int ts_pcr_out; // sequence number of most recent output pcr
108 int ts_pcr_in; // sequence number of most recent input pcr
109 int64_t ts_pcr; // most recent input pcr
110 int64_t ts_pcrhist[4]; // circular buffer of output pcrs
112 uint8_t *ts_packet; /* buffer for one TS packet */
113 hb_buffer_t *ts_buf[kMaxNumberDecodeStreams];
114 int ts_pos[kMaxNumberDecodeStreams];
115 int8_t ts_skipbad[kMaxNumberDecodeStreams];
116 int8_t ts_streamcont[kMaxNumberDecodeStreams];
118 hb_buffer_t *fwrite_buf; /* PS buffer (set by hb_ts_stream_decode) */
120 int chapter; /* Chapter that we are currently in */
121 int64_t chapter_end; /* HB time that the current chapter ends */
124 * Stuff before this point is dynamic state updated as we read the
125 * stream. Stuff after this point is stream description state that
126 * we learn during the initial scan but cache so it can be
127 * reused during the conversion read.
129 uint8_t ts_number_video_pids;
130 uint8_t ts_number_audio_pids;
131 uint8_t ts_flags; // stream characteristics:
132 #define TS_HAS_PCR (1 << 0) // at least one PCR seen
133 #define TS_HAS_RAP (1 << 1) // Random Access Point bit seen
134 #define TS_HAS_RSEI (1 << 2) // "Restart point" SEI seen
135 uint8_t ts_IDRs; // # IDRs found during duration scan
137 int16_t ts_video_pids[kMaxNumberVideoPIDS];
138 int16_t ts_audio_pids[kMaxNumberAudioPIDS];
140 uint32_t ts_format_id[kMaxNumberDecodeStreams];
141 #define TS_FORMAT_ID_AC3 (('A' << 24) | ('C' << 16) | ('-' << 8) | '3')
142 uint8_t ts_stream_type[kMaxNumberDecodeStreams];
143 uint8_t ts_multiplexed[kMaxNumberDecodeStreams];
147 hb_stream_type_t hb_stream_type;
150 AVFormatContext *ffmpeg_ic;
151 AVPacket *ffmpeg_pkt;
152 double ffmpeg_tsconv[MAX_STREAMS];
153 uint8_t ffmpeg_video_id;
160 } a52_info[kMaxNumberAudioPIDS];
164 unsigned short program_number;
165 unsigned short program_map_PID;
166 } pat_info[kMaxNumberPMTStreams];
167 int ts_number_pat_entries;
172 unsigned char *tablebuf;
173 unsigned int tablepos;
174 unsigned char current_continuity_counter;
178 unsigned int PCR_PID;
179 int program_info_length;
180 unsigned char *progam_info_descriptor_data;
183 unsigned char stream_type;
184 unsigned short elementary_PID;
185 unsigned short ES_info_length;
186 unsigned char *es_info_descriptor_data;
187 } pmt_stream_info[kMaxNumberPMTStreams];
191 /***********************************************************************
193 **********************************************************************/
194 static void hb_stream_duration(hb_stream_t *stream, hb_title_t *inTitle);
195 static void hb_ts_stream_init(hb_stream_t *stream);
196 static void hb_ts_stream_find_pids(hb_stream_t *stream);
197 static int hb_ts_stream_decode(hb_stream_t *stream, hb_buffer_t *obuf);
198 static void hb_ts_stream_reset(hb_stream_t *stream);
199 static hb_audio_t *hb_ts_stream_set_audio_id_and_codec(hb_stream_t *stream,
201 static void hb_ps_stream_find_audio_ids(hb_stream_t *stream, hb_title_t *title);
202 static off_t align_to_next_packet(hb_stream_t *stream);
204 static int ffmpeg_open( hb_stream_t *stream, hb_title_t *title );
205 static void ffmpeg_close( hb_stream_t *d );
206 static hb_title_t *ffmpeg_title_scan( hb_stream_t *stream );
207 static int ffmpeg_read( hb_stream_t *stream, hb_buffer_t *buf );
208 static int ffmpeg_seek( hb_stream_t *stream, float frac );
211 * streams have a bunch of state that's learned during the scan. We don't
212 * want to throw away the state when scan does a close then relearn
213 * everything when reader does an open. So we save the stream state on
214 * the close following a scan and reuse it when 'reader' does an open.
216 static hb_list_t *stream_state_list;
218 static hb_stream_t *hb_stream_lookup( const char *path )
220 if ( stream_state_list == NULL )
226 while ( ( ss = hb_list_item( stream_state_list, i++ ) ) != NULL )
228 if ( strcmp( path, ss->path ) == 0 )
236 static void hb_stream_state_delete( hb_stream_t *ss )
238 hb_list_rem( stream_state_list, ss );
245 * these frontend hb_log because transport streams can have a lot of errors
246 * so we want to rate limit messages. this routine limits the number of
247 * messages to at most one per minute of video. other errors that occur
248 * during the minute are counted & the count is output with the next
249 * error msg we print.
251 static void ts_warn_helper( hb_stream_t *stream, char *log, va_list args )
253 // limit error printing to at most one per minute of video (at 30fps)
255 if ( stream->frames - stream->last_error_frame >= 30*60 )
259 vsnprintf( msg, sizeof(msg), log, args );
261 if ( stream->errors - stream->last_error_count < 10 )
263 hb_log( "stream: error near frame %d: %s", stream->frames, msg );
267 int Edelta = stream->errors - stream->last_error_count;
268 double Epcnt = (double)Edelta * 100. /
269 (stream->frames - stream->last_error_frame);
270 hb_log( "stream: %d new errors (%.0f%%) up to frame %d: %s",
271 Edelta, Epcnt, stream->frames, msg );
273 stream->last_error_frame = stream->frames;
274 stream->last_error_count = stream->errors;
278 static void ts_warn( hb_stream_t*, char*, ... ) HB_WPRINTF(2,3);
279 static void ts_err( hb_stream_t*, int, char*, ... ) HB_WPRINTF(3,4);
281 static void ts_warn( hb_stream_t *stream, char *log, ... )
284 va_start( args, log );
285 ts_warn_helper( stream, log, args );
289 static void ts_err( hb_stream_t *stream, int curstream, char *log, ... )
292 va_start( args, log );
293 ts_warn_helper( stream, log, args );
296 stream->ts_skipbad[curstream] = 1;
297 stream->ts_pos[curstream] = 0;
298 stream->ts_streamcont[curstream] = -1;
301 static int check_ps_sync(const uint8_t *buf)
303 // a legal MPEG program stream must start with a Pack header in the
305 return (buf[0] == 0x00) && (buf[1] == 0x00) &&
306 (buf[2] == 0x01) && (buf[3] == 0xba);
309 static int check_ps_sc(const uint8_t *buf)
311 // a legal MPEG program stream must start with a Pack followed by a
312 // some other start code. If we've already verified the pack, this skip
313 // it and checks for a start code prefix.
314 int pos = 14 + ( buf[13] & 0x7 ); // skip over the PACK
315 return (buf[pos+0] == 0x00) && (buf[pos+1] == 0x00) && (buf[pos+2] == 0x01);
318 static int check_ts_sync(const uint8_t *buf)
320 // must have initial sync byte, no scrambling & a legal adaptation ctrl
321 return (buf[0] == 0x47) && ((buf[3] >> 6) == 0) && ((buf[3] >> 4) > 0);
324 static int have_ts_sync(const uint8_t *buf, int psize)
326 return check_ts_sync(&buf[0*psize]) && check_ts_sync(&buf[1*psize]) &&
327 check_ts_sync(&buf[2*psize]) && check_ts_sync(&buf[3*psize]) &&
328 check_ts_sync(&buf[4*psize]) && check_ts_sync(&buf[5*psize]) &&
329 check_ts_sync(&buf[6*psize]) && check_ts_sync(&buf[7*psize]);
332 static int hb_stream_check_for_ts(const uint8_t *buf)
334 // transport streams should have a sync byte every 188 bytes.
335 // search the first 8KB of buf looking for at least 8 consecutive
336 // correctly located sync patterns.
339 for ( offset = 0; offset < 8*1024-8*188; ++offset )
341 if ( have_ts_sync( &buf[offset], 188) )
342 return 188 | (offset << 8);
343 if ( have_ts_sync( &buf[offset], 192) )
344 return 192 | (offset << 8);
345 if ( have_ts_sync( &buf[offset], 204) )
346 return 204 | (offset << 8);
347 if ( have_ts_sync( &buf[offset], 208) )
348 return 208 | (offset << 8);
353 static int hb_stream_check_for_ps(const uint8_t *buf)
355 // program streams should start with a PACK then some other mpeg start
356 // code (usually a SYS but that might be missing if we only have a clip).
357 return check_ps_sync(buf) && check_ps_sc(buf);
360 static int hb_stream_check_for_dvd_ps(const uint8_t *buf)
362 // DVD program streams should have a Pack header every 2048 bytes.
363 // check that we have 4 of these in a row.
364 return check_ps_sync(&buf[0*2048]) && check_ps_sync(&buf[1*2048]) &&
365 check_ps_sync(&buf[2*2048]) && check_ps_sync(&buf[3*2048]);
368 static int hb_stream_get_type(hb_stream_t *stream)
372 if ( fread(buf, 1, sizeof(buf), stream->file_handle) == sizeof(buf) )
375 if ( ( psize = hb_stream_check_for_ts(buf) ) != 0 )
377 int offset = psize >> 8;
379 hb_log("file is MPEG Transport Stream with %d byte packets"
380 " offset %d bytes", psize, offset);
381 stream->packetsize = psize;
382 stream->hb_stream_type = transport;
383 hb_ts_stream_init(stream);
384 if ( !stream->ts_number_video_pids || !stream->ts_number_audio_pids )
390 if ( hb_stream_check_for_dvd_ps(buf) != 0 )
392 hb_log("file is MPEG DVD Program Stream");
393 stream->hb_stream_type = dvd_program;
396 if ( hb_stream_check_for_ps(buf) != 0 )
398 hb_log("file is MPEG Program Stream");
399 stream->hb_stream_type = program;
406 static void hb_stream_delete_dynamic( hb_stream_t *d )
410 fclose( d->file_handle );
411 d->file_handle = NULL;
418 free( d->ts_packet );
421 for (i = 0; i < kMaxNumberDecodeStreams; i++)
425 hb_buffer_close(&(d->ts_buf[i]));
431 static void hb_stream_delete( hb_stream_t *d )
433 hb_stream_delete_dynamic( d );
438 static int audio_inactive( hb_stream_t *stream, int indx )
440 int aud_indx = indx - 1;
442 if ( stream->ts_audio_pids[aud_indx] < 0 )
444 // PID declared inactive by hb_stream_title_scan
447 if ( stream->ts_audio_pids[aud_indx] == stream->pmt_info.PCR_PID )
449 // PCR PID is always active
453 // see if we should make the stream inactive because scan.c didn't
454 // find a valid audio bitstream.
456 for ( i = 0; i < hb_list_count( stream->title->list_audio ); ++i )
458 hb_audio_t *audio = hb_list_item( stream->title->list_audio, i );
459 if ( audio->id == indx )
464 // not in the title's audio list - declare the PID inactive
465 stream->ts_audio_pids[aud_indx] = -stream->ts_audio_pids[aud_indx];
469 /***********************************************************************
471 ***********************************************************************
473 **********************************************************************/
474 hb_stream_t * hb_stream_open( char *path, hb_title_t *title )
476 FILE *f = fopen( path, "rb" );
479 hb_log( "hb_stream_open: open %s failed", path );
483 hb_stream_t *d = calloc( sizeof( hb_stream_t ), 1 );
487 hb_log( "hb_stream_open: can't allocate space for %s stream state", path );
492 * if we're opening the stream to read & convert, we need
493 * the state we saved when we scanned the stream. if we're
494 * opening the stream to scan it we want to rebuild the state
495 * (even if we have saved state, the stream may have changed).
497 hb_stream_t *ss = hb_stream_lookup( path );
498 if ( title && ss && ss->hb_stream_type != ffmpeg )
501 * copy the saved state since we might be encoding the same stream
504 memcpy( d, ss, sizeof(*d) );
507 d->path = strdup( path );
509 if ( d->hb_stream_type == transport )
511 d->ts_packet = malloc( d->packetsize );
514 for ( ; i < d->ts_number_video_pids + d->ts_number_audio_pids; i++)
516 if ( i && audio_inactive( d, i ) )
518 // this PID isn't wanted (we don't have a codec for it
519 // or scan didn't find audio parameters)
522 d->ts_buf[i] = hb_buffer_init(d->packetsize);
523 d->ts_buf[i]->size = 0;
525 hb_stream_seek( d, 0. );
531 * opening for scan - delete any saved state then (re)scan the stream.
532 * If it's something we can deal with (MPEG2 PS or TS) return a stream
533 * reference structure & null otherwise.
537 hb_stream_state_delete( ss );
541 d->path = strdup( path );
542 if (d->path != NULL )
544 if ( hb_stream_get_type( d ) != 0 )
548 fclose( d->file_handle );
549 d->file_handle = NULL;
550 if ( ffmpeg_open( d, title ) )
555 if ( d->file_handle )
557 fclose( d->file_handle );
563 hb_log( "hb_stream_open: open %s failed", path );
568 /***********************************************************************
570 ***********************************************************************
571 * Closes and frees everything
572 **********************************************************************/
573 void hb_stream_close( hb_stream_t ** _d )
575 hb_stream_t *stream = * _d;
577 if ( stream->hb_stream_type == ffmpeg )
579 ffmpeg_close( stream );
580 hb_stream_delete( stream );
585 if ( stream->frames )
587 hb_log( "stream: %d good frames, %d errors (%.0f%%)", stream->frames,
588 stream->errors, (double)stream->errors * 100. /
589 (double)stream->frames );
593 * if the stream was opened for a scan, cache the result, otherwise delete
596 if ( stream->title == NULL )
598 hb_stream_delete_dynamic( stream );
599 if ( stream_state_list == NULL )
601 stream_state_list = hb_list_init();
603 hb_list_add( stream_state_list, stream );
607 hb_stream_delete( stream );
612 /* when the file was first opened we made entries for all the audio elementary
613 * streams we found in it. Streams that were later found during the preview scan
614 * now have an audio codec, type, rate, etc., associated with them. At the end
615 * of the scan we delete all the audio entries that weren't found by the scan
616 * or don't have a format we support. This routine deletes audio entry 'indx'
617 * by setting its PID to an invalid value so no packet will match it. (We can't
618 * move any of the entries since the index of the entry is used as the id
619 * of the media stream for HB. */
620 static void hb_stream_delete_audio_entry(hb_stream_t *stream, int indx)
622 if ( stream->ts_audio_pids[indx] > 0 )
624 stream->ts_audio_pids[indx] = -stream->ts_audio_pids[indx];
628 static int index_of_pid(int pid, hb_stream_t *stream)
632 if ( pid == stream->ts_video_pids[0] )
635 for ( i = 0; i < stream->ts_number_audio_pids; ++i )
636 if ( pid == stream->ts_audio_pids[i] )
642 /***********************************************************************
643 * hb_ps_stream_title_scan
644 ***********************************************************************
646 **********************************************************************/
647 hb_title_t * hb_stream_title_scan(hb_stream_t *stream)
649 if ( stream->hb_stream_type == ffmpeg )
650 return ffmpeg_title_scan( stream );
653 hb_title_t *aTitle = hb_title_init( stream->path, 0 );
656 // Copy part of the stream path to the title name
657 char *sep = strrchr(stream->path, '/');
659 strcpy(aTitle->name, sep+1);
660 char *dot_term = strrchr(aTitle->name, '.');
664 // Height, width, rate and aspect ratio information is filled in when the previews are built
666 hb_stream_duration(stream, aTitle);
669 hb_chapter_t * chapter;
670 chapter = calloc( sizeof( hb_chapter_t ), 1 );
672 chapter->duration = aTitle->duration;
673 chapter->hours = aTitle->hours;
674 chapter->minutes = aTitle->minutes;
675 chapter->seconds = aTitle->seconds;
676 hb_list_add( aTitle->list_chapter, chapter );
678 // Figure out how many audio streams we really have:
679 // - For transport streams, for each PID listed in the PMT (whether
680 // or not it was an audio stream type) read the bitstream until we
681 // find an packet from that PID containing a PES header and see if
682 // the elementary stream is an audio type.
683 // - For program streams read the first 4MB and take every unique
684 // audio stream we find.
685 if (stream->hb_stream_type == transport)
689 for (i=0; i < stream->ts_number_audio_pids; i++)
691 hb_audio_t *audio = hb_ts_stream_set_audio_id_and_codec(stream, i);
692 if (audio->config.in.codec)
693 hb_list_add( aTitle->list_audio, audio );
697 hb_stream_delete_audio_entry(stream, i);
701 // make sure we're grabbing the PCR PID
702 if ( index_of_pid( stream->pmt_info.PCR_PID, stream ) < 0 )
704 stream->ts_audio_pids[stream->ts_number_audio_pids++] =
705 stream->pmt_info.PCR_PID;
708 // set the video id, codec & muxer
709 aTitle->video_id = 0;
710 aTitle->video_codec = st2codec[stream->ts_stream_type[0]].codec;
711 aTitle->video_codec_param = st2codec[stream->ts_stream_type[0]].codec_param;
712 aTitle->demuxer = HB_MPEG2_TS_DEMUXER;
714 if ( ( stream->ts_flags & TS_HAS_PCR ) == 0 )
716 hb_log( "transport stream missing PCRs - using video DTS instead" );
719 if ( stream->ts_IDRs < 1 )
721 hb_log( "transport stream doesn't seem to have video IDR frames" );
722 aTitle->flags |= HBTF_NO_IDR;
727 hb_ps_stream_find_audio_ids(stream, aTitle);
734 * read the next transport stream packet from 'stream'. Return NULL if
735 * we hit eof & a pointer to the sync byte otherwise.
737 static const uint8_t *next_packet( hb_stream_t *stream )
739 uint8_t *buf = stream->ts_packet + stream->packetsize - 188;
743 if ( fread(stream->ts_packet, 1, stream->packetsize, stream->file_handle) !=
752 // lost sync - back up to where we started then try to re-establish.
753 off_t pos = ftello(stream->file_handle) - stream->packetsize;
754 off_t pos2 = align_to_next_packet(stream);
757 hb_log( "next_packet: eof while re-establishing sync @ %"PRId64, pos );
760 ts_warn( stream, "next_packet: sync lost @ %"PRId64", regained after %"PRId64" bytes",
766 * skip to the start of the next PACK header in program stream src_stream.
768 static void skip_to_next_pack( hb_stream_t *src_stream )
770 // scan forward until we find the start of the next pack
771 uint32_t strt_code = -1;
774 flockfile( src_stream->file_handle );
775 while ( ( c = getc_unlocked( src_stream->file_handle ) ) != EOF )
777 strt_code = ( strt_code << 8 ) | c;
778 if ( strt_code == 0x000001ba )
779 // we found the start of the next pack
782 funlockfile( src_stream->file_handle );
784 // if we didn't terminate on an eof back up so the next read
785 // starts on the pack boundary.
788 fseeko( src_stream->file_handle, -4, SEEK_CUR );
792 static int isIframe( hb_stream_t *stream, const uint8_t *buf, int adapt_len )
794 // For mpeg2: look for a gop start or i-frame picture start
795 // for h.264: look for idr nal type or a slice header for an i-frame
796 // for vc1: look for a Sequence header
801 if ( stream->ts_stream_type[0] <= 2 )
803 // This section of the code handles MPEG-1 and MPEG-2 video streams
804 for (i = 13 + adapt_len; i < 188; i++)
806 strid = (strid << 8) | buf[i];
807 if ( ( strid >> 8 ) == 1 )
809 // we found a start code
813 case 0xB8: // group_start_code (GOP header)
814 case 0xB3: // sequence_header code
817 case 0x00: // picture_start_code
818 // picture_header, let's see if it's an I-frame
821 // check if picture_coding_type == 1
822 if ((buf[i+2] & (0x7 << 3)) == (1 << 3))
824 // found an I-frame picture
832 // didn't find an I-frame
835 if ( stream->ts_stream_type[0] == 0x1b )
837 // we have an h.264 stream
838 for (i = 13 + adapt_len; i < 188; i++)
840 strid = (strid << 8) | buf[i];
841 if ( ( strid >> 8 ) == 1 )
843 // we found a start code - remove the ref_idc from the nal type
844 uint8_t nal_type = strid & 0x1f;
845 if ( nal_type == 0x05 )
846 // h.264 IDR picture start
850 // didn't find an I-frame
853 if ( stream->ts_stream_type[0] == 0xea )
855 // we have an vc1 stream
856 for (i = 13 + adapt_len; i < 188; i++)
858 strid = (strid << 8) | buf[i];
859 if ( strid == 0x10f )
861 // the ffmpeg vc1 decoder requires a seq hdr code in the first
866 // didn't find an I-frame
870 // we don't understand the stream type so just say "yes" otherwise
871 // we'll discard all the video.
876 * scan the next MB of 'stream' to find the next start packet for
877 * the Packetized Elementary Stream associated with TS PID 'pid'.
879 static const uint8_t *hb_ts_stream_getPEStype(hb_stream_t *stream, uint32_t pid)
881 int npack = 300000; // max packets to read
885 const uint8_t *buf = next_packet( stream );
888 hb_log("hb_ts_stream_getPEStype: EOF while searching for PID 0x%x", pid);
892 // while we're reading the stream, check if it has valid PCRs
893 // and/or random access points.
894 uint32_t pack_pid = ( (buf[1] & 0x1f) << 8 ) | buf[2];
895 if ( pack_pid == stream->pmt_info.PCR_PID )
897 if ( ( buf[5] & 0x10 ) &&
898 ( ( ( buf[3] & 0x30 ) == 0x20 ) ||
899 ( ( buf[3] & 0x30 ) == 0x30 && buf[4] > 6 ) ) )
901 stream->ts_flags |= TS_HAS_PCR;
906 stream->ts_flags |= TS_HAS_RAP;
910 * The PES header is only in TS packets with 'start' set so we check
911 * that first then check for the right PID.
913 if ((buf[1] & 0x40) == 0 || pack_pid != pid )
915 // not a start packet or not the pid we want
919 /* skip over the TS hdr to return a pointer to the PES hdr */
921 switch (buf[3] & 0x30)
923 case 0x00: // illegal
924 case 0x20: // fill packet
927 case 0x30: // adaptation
930 hb_log("hb_ts_stream_getPEStype: invalid adaptation field length %d for PID 0x%x", buf[4], pid);
936 /* PES hdr has to begin with an mpeg start code */
937 if (buf[udata+0] == 0x00 && buf[udata+1] == 0x00 && buf[udata+2] == 0x01)
947 static uint64_t hb_ps_stream_getVideoPTS(hb_stream_t *stream)
949 hb_buffer_t *buf = hb_buffer_init(HB_DVD_READ_BUFFER_SIZE);
950 hb_list_t *list = hb_list_init();
951 // how many blocks we read while searching for a video PES header
955 while (--blksleft >= 0 && hb_stream_read(stream, buf) == 1)
959 // 'buf' contains an MPEG2 PACK - get a list of all it's elementary streams
960 hb_demux_ps( buf, list, 0 );
962 while ( ( es = hb_list_item( list, 0 ) ) )
964 hb_list_rem( list, es );
965 if ( es->id == 0xe0 )
967 // this PES contains video - if there's a PTS we're done
968 // hb_demux_ps left the PTS in buf_es->start.
969 if ( es->start != ~0 )
976 hb_buffer_close( &es );
979 hb_list_empty( &list );
980 hb_buffer_close(&buf);
984 /***********************************************************************
986 ***********************************************************************
988 * Finding stream duration is difficult. One issue is that the video file
989 * may have chunks from several different program fragments (main feature,
990 * commercials, station id, trailers, etc.) all with their own base pts
991 * value. We can't find the piece boundaries without reading the entire
992 * file but if we compute a rate based on time stamps from two different
993 * pieces the result will be meaningless. The second issue is that the
994 * data rate of compressed video normally varies by 5-10x over the length
995 * of the video. This says that we want to compute the rate over relatively
996 * long segments to get a representative average but long segments increase
997 * the likelihood that we'll cross a piece boundary.
999 * What we do is take time stamp samples at several places in the file
1000 * (currently 16) then compute the average rate (i.e., ticks of video per
1001 * byte of the file) for all pairs of samples (N^2 rates computed for N
1002 * samples). Some of those rates will be absurd because the samples came
1003 * from different segments. Some will be way low or high because the
1004 * samples came from a low or high motion part of the segment. But given
1005 * that we're comparing *all* pairs the majority of the computed rates
1006 * should be near the overall average. So we median filter the computed
1007 * rates to pick the most representative value.
1009 **********************************************************************/
1011 uint64_t pos; /* file position of this PTS sample */
1012 uint64_t pts; /* PTS from video stream */
1015 #define NDURSAMPLES 128
1017 // get one (position, timestamp) sampple from a transport or program
1019 static struct pts_pos hb_sample_pts(hb_stream_t *stream, uint64_t fpos)
1021 struct pts_pos pp = { 0, 0 };
1023 if ( stream->hb_stream_type == transport )
1026 fseeko( stream->file_handle, fpos, SEEK_SET );
1027 align_to_next_packet( stream );
1028 buf = hb_ts_stream_getPEStype( stream, stream->ts_video_pids[0] );
1031 hb_log("hb_sample_pts: couldn't find video packet near %"PRIu64, fpos);
1034 if ( ( buf[7] >> 7 ) != 1 )
1036 hb_log("hb_sample_pts: no PTS in video packet near %"PRIu64, fpos);
1039 pp.pts = ( ( (uint64_t)buf[9] >> 1 ) & 7 << 30 ) |
1040 ( (uint64_t)buf[10] << 22 ) |
1041 ( ( (uint64_t)buf[11] >> 1 ) << 15 ) |
1042 ( (uint64_t)buf[12] << 7 ) |
1043 ( (uint64_t)buf[13] >> 1 );
1045 if ( isIframe( stream, buf, -4 ) )
1047 if ( stream->ts_IDRs < 255 )
1055 // round address down to nearest dvd sector start
1056 fpos &=~ ( HB_DVD_READ_BUFFER_SIZE - 1 );
1057 fseeko( stream->file_handle, fpos, SEEK_SET );
1058 if ( stream->hb_stream_type == program )
1060 skip_to_next_pack( stream );
1062 pp.pts = hb_ps_stream_getVideoPTS( stream );
1064 pp.pos = ftello(stream->file_handle);
1068 static int dur_compare( const void *a, const void *b )
1070 const double *aval = a, *bval = b;
1071 return ( *aval < *bval ? -1 : ( *aval == *bval ? 0 : 1 ) );
1074 // given an array of (position, time) samples, compute a max-likelihood
1075 // estimate of the average rate by computing the rate between all pairs
1076 // of samples then taking the median of those rates.
1077 static double compute_stream_rate( struct pts_pos *pp, int n )
1080 double rates[NDURSAMPLES * NDURSAMPLES / 8];
1083 // the following nested loops compute the rates between all pairs.
1085 for ( i = 0; i < n-1; ++i )
1087 // Bias the median filter by not including pairs that are "far"
1088 // from one another. This is to handle cases where the file is
1089 // made of roughly equal size pieces where a symmetric choice of
1090 // pairs results in having the same number of intra-piece &
1091 // inter-piece rate estimates. This would mean that the median
1092 // could easily fall in the inter-piece part of the data which
1093 // would give a bogus estimate. The 'ns' index creates an
1094 // asymmetry that favors locality.
1095 int ns = i + ( n >> 3 );
1098 for ( j = i+1; j < ns; ++j )
1100 if ( (uint64_t)(pp[j].pts - pp[i].pts) > 90000LL*3600*6 )
1102 if ( pp[j].pts != pp[i].pts && pp[j].pos > pp[i].pos )
1104 *rp = ((double)( pp[j].pts - pp[i].pts )) /
1105 ((double)( pp[j].pos - pp[i].pos ));
1110 // now compute and return the median of all the (n*n/2) rates we computed
1112 int nrates = rp - rates;
1113 qsort( rates, nrates, sizeof (rates[0] ), dur_compare );
1114 return rates[nrates >> 1];
1117 static void hb_stream_duration(hb_stream_t *stream, hb_title_t *inTitle)
1119 struct pts_pos ptspos[NDURSAMPLES];
1120 struct pts_pos *pp = ptspos;
1123 fseeko(stream->file_handle, 0, SEEK_END);
1124 uint64_t fsize = ftello(stream->file_handle);
1125 uint64_t fincr = fsize / NDURSAMPLES;
1126 uint64_t fpos = fincr / 2;
1127 for ( i = NDURSAMPLES; --i >= 0; fpos += fincr )
1129 *pp++ = hb_sample_pts(stream, fpos);
1131 uint64_t dur = compute_stream_rate( ptspos, pp - ptspos ) * (double)fsize;
1132 inTitle->duration = dur;
1134 inTitle->hours = dur / 3600;
1135 inTitle->minutes = ( dur % 3600 ) / 60;
1136 inTitle->seconds = dur % 60;
1138 rewind(stream->file_handle);
1141 /***********************************************************************
1143 ***********************************************************************
1145 **********************************************************************/
1146 int hb_stream_read( hb_stream_t * src_stream, hb_buffer_t * b )
1148 if ( src_stream->hb_stream_type == ffmpeg )
1150 return ffmpeg_read( src_stream, b );
1152 if ( src_stream->hb_stream_type == dvd_program )
1154 size_t amt_read = fread(b->data, HB_DVD_READ_BUFFER_SIZE, 1,
1155 src_stream->file_handle);
1156 return (amt_read > 0);
1158 if ( src_stream->hb_stream_type == program )
1160 // a general program stream has arbitrary sized pack's. we're
1161 // currently positioned at the start of a pack so read up to but
1162 // not including the start of the next, expanding the buffer
1164 uint8_t *cp = b->data;
1165 uint8_t *ep = cp + b->alloc;
1166 uint32_t strt_code = -1;
1169 // consume the first byte of the initial pack so we don't match on
1170 // it in the loop below.
1171 if ( ( c = getc( src_stream->file_handle ) ) == EOF )
1176 flockfile( src_stream->file_handle );
1177 while ( ( c = getc_unlocked( src_stream->file_handle ) ) != EOF )
1179 strt_code = ( strt_code << 8 ) | c;
1180 if ( strt_code == 0x000001ba )
1181 // we found the start of the next pack
1185 // need to expand the buffer
1186 int curSize = cp - b->data;
1187 hb_buffer_realloc( b, curSize * 2 );
1188 cp = b->data + curSize;
1189 ep = b->data + b->alloc;
1193 funlockfile( src_stream->file_handle );
1195 // if we didn't terminate on an eof back up so the next read
1196 // starts on the pack boundary.
1197 b->size = cp - b->data;
1200 fseeko( src_stream->file_handle, -4, SEEK_CUR );
1205 return hb_ts_stream_decode( src_stream, b );
1208 int hb_stream_seek_chapter( hb_stream_t * stream, int chapter_num )
1211 if ( stream->hb_stream_type != ffmpeg )
1213 // currently meaningliess for transport and program streams
1216 if ( !stream || !stream->title ||
1217 chapter_num > hb_list_count( stream->title->list_chapter ) )
1222 int64_t sum_dur = 0;
1223 hb_chapter_t *chapter = NULL;
1225 for ( i = 0; i < chapter_num; ++i)
1227 chapter = hb_list_item( stream->title->list_chapter, i );
1228 sum_dur += chapter->duration;
1230 stream->chapter = chapter_num - 1;
1231 stream->chapter_end = sum_dur;
1233 int64_t pos = ( ( ( sum_dur - chapter->duration ) * AV_TIME_BASE ) / 90000 );
1235 hb_deep_log( 2, "Seeking to chapter %d: starts %"PRId64", ends %"PRId64", AV pos %"PRId64,
1236 chapter_num, sum_dur - chapter->duration, sum_dur, pos);
1238 if ( chapter_num > 1 && pos > 0 )
1240 av_seek_frame( stream->ffmpeg_ic, -1, pos, 0);
1245 /***********************************************************************
1247 ***********************************************************************
1248 * Return the number of the chapter that we are currently in. We store
1249 * the chapter number starting from 0, so + 1 for the real chpater num.
1250 **********************************************************************/
1251 int hb_stream_chapter( hb_stream_t * src_stream )
1253 return( src_stream->chapter + 1 );
1256 /***********************************************************************
1258 ***********************************************************************
1260 **********************************************************************/
1261 int hb_stream_seek( hb_stream_t * stream, float f )
1263 if ( stream->hb_stream_type == ffmpeg )
1265 return ffmpeg_seek( stream, f );
1267 off_t stream_size, cur_pos, new_pos;
1268 double pos_ratio = f;
1269 cur_pos = ftello( stream->file_handle );
1270 fseeko( stream->file_handle, 0, SEEK_END );
1271 stream_size = ftello( stream->file_handle );
1272 new_pos = (off_t) ((double) (stream_size) * pos_ratio);
1273 new_pos &=~ (HB_DVD_READ_BUFFER_SIZE - 1);
1275 int r = fseeko( stream->file_handle, new_pos, SEEK_SET );
1278 fseeko( stream->file_handle, cur_pos, SEEK_SET );
1282 if ( stream->hb_stream_type == transport )
1284 // We need to drop the current decoder output and move
1285 // forwards to the next transport stream packet.
1286 hb_ts_stream_reset(stream);
1289 if ( stream->ts_IDRs )
1291 // the stream has IDRs so look for one.
1292 stream->need_keyframe = 1;
1297 // we're at the beginning - say we have video sync so that we
1298 // won't drop initial SPS & PPS data on an AVC stream.
1299 stream->need_keyframe = 0;
1302 else if ( stream->hb_stream_type == program )
1304 skip_to_next_pack( stream );
1310 static const char* make_upper( const char* s )
1312 static char name[8];
1314 char *ep = cp + sizeof(name)-1;
1316 while ( *s && cp < ep )
1318 *cp++ = islower(*s)? toupper(*s) : *s;
1325 static void set_audio_description( hb_audio_t *audio, iso639_lang_t *lang )
1328 * This is a duplicate of code in dvd.c - it should get factored out
1329 * into a common routine. We probably should only be putting the lang
1330 * code or a lang pointer into the audio config & let the common description
1331 * formatting routine in scan.c do all the stuff below.
1333 const char *codec_name;
1336 if ( audio->config.in.codec == HB_ACODEC_FFMPEG &&
1337 ( cc = hb_ffmpeg_context( audio->config.in.codec_param ) ) &&
1338 avcodec_find_decoder( cc->codec_id ) )
1340 codec_name = make_upper( avcodec_find_decoder( cc->codec_id )->name );
1341 if ( !strcmp( codec_name, "LIBFAAD" ) )
1346 else if ( audio->config.in.codec == HB_ACODEC_MPGA &&
1347 avcodec_find_decoder( audio->config.in.codec_param ) )
1349 codec_name = avcodec_find_decoder( audio->config.in.codec_param )->name;
1353 codec_name = audio->config.in.codec == HB_ACODEC_AC3 ? "AC3" :
1354 audio->config.in.codec == HB_ACODEC_DCA ? "DTS" :
1355 audio->config.in.codec == HB_ACODEC_MPGA ? "MPEG" :
1356 audio->config.in.codec == HB_ACODEC_LPCM ? "LPCM" :
1357 audio->config.in.codec == HB_ACODEC_FFMPEG ? "FFMPEG" :
1360 snprintf( audio->config.lang.description,
1361 sizeof( audio->config.lang.description ), "%s (%s)",
1362 strlen(lang->native_name) ? lang->native_name : lang->eng_name,
1364 snprintf( audio->config.lang.simple, sizeof( audio->config.lang.simple ), "%s",
1365 strlen(lang->native_name) ? lang->native_name : lang->eng_name );
1366 snprintf( audio->config.lang.iso639_2, sizeof( audio->config.lang.iso639_2 ),
1367 "%s", lang->iso639_2);
1370 static hb_audio_t *hb_ts_stream_set_audio_id_and_codec(hb_stream_t *stream,
1373 off_t cur_pos = ftello(stream->file_handle);
1374 hb_audio_t *audio = calloc( sizeof( hb_audio_t ), 1 );
1377 fseeko(stream->file_handle, 0, SEEK_SET);
1378 align_to_next_packet(stream);
1379 buf = hb_ts_stream_getPEStype(stream, stream->ts_audio_pids[aud_pid_index]);
1381 /* check that we found a PES header */
1383 if (buf && buf[0] == 0x00 && buf[1] == 0x00 && buf[2] == 0x01)
1385 stype = stream->ts_stream_type[1 + aud_pid_index];
1387 // 0xbd ("private stream 1") is the normal container for non-ISO
1388 // media - AC3/DCA/PCM/etc.
1389 if ( buf[3] == 0xbd )
1391 if ( st2codec[stype].kind == U )
1393 // XXX assume unknown stream types are AC-3 (if they're not
1394 // audio we'll find that out during the scan but if they're
1395 // some other type of audio we'll end up ignoring them).
1397 stream->ts_stream_type[1 + aud_pid_index] = 0x81;
1400 else if ( buf[3] == 0xfd )
1402 // 0xfd indicates an extended stream id (ISO 13818-1(2007)).
1403 // the blu ray consortium apparently forgot to read the portion
1404 // of the MPEG spec that says one PID should map to one media
1405 // stream and multiplexed multiple types of audio into one PID
1406 // using the extended stream identifier of the PES header to
1407 // distinguish them. So we have to check if that's happening and
1408 // if so tell the runtime what esid we want.
1409 if ( st2codec[stype].kind == A && stype == 0x83 &&
1410 stream->ts_format_id[1 + aud_pid_index] == TS_FORMAT_ID_AC3 )
1412 // This is an interleaved TrueHD/AC-3 stream and the esid of
1414 stream->ts_multiplexed[1 + aud_pid_index] = 0x76;
1416 stream->ts_stream_type[1 + aud_pid_index] = 0x81;
1418 if ( st2codec[stype].kind == A && stype == 0x86 )
1420 // This is an interleaved DTS-HD/DTS stream and the esid of
1422 stream->ts_multiplexed[1 + aud_pid_index] = 0x71;
1424 stream->ts_stream_type[1 + aud_pid_index] = 0x82;
1427 else if ((buf[3] & 0xe0) == 0xc0)
1429 // 0xC0 - 0xCF are the normal containers for ISO-standard
1430 // media (mpeg2 audio and mpeg4 AAC).
1431 if ( st2codec[stype].kind == U )
1433 // XXX assume unknown stream types are MPEG audio
1435 stream->ts_stream_type[1 + aud_pid_index] = 0x03;
1443 // if we found an audio stream type & HB has a codec that can decode it
1444 // finish configuring the audio so we'll add it to the title's list.
1445 if ( st2codec[stype].kind == A && st2codec[stype].codec )
1447 audio->id = 1 + aud_pid_index;
1448 audio->config.in.codec = st2codec[stype].codec;
1449 audio->config.in.codec_param = st2codec[stype].codec_param;
1450 set_audio_description( audio,
1451 lang_for_code( stream->a52_info[aud_pid_index].lang_code ) );
1452 hb_log("transport stream pid 0x%x (type 0x%x) may be %s audio (id 0x%x)",
1453 stream->ts_audio_pids[aud_pid_index],
1454 stype, st2codec[stype].name, audio->id);
1460 hb_log("transport stream pid 0x%x (type 0x%x, substream 0x%x) "
1461 "isn't audio", stream->ts_audio_pids[aud_pid_index],
1462 stream->ts_stream_type[1 + aud_pid_index], buf[3]);
1466 hb_log("transport stream pid 0x%x (type 0x%x) isn't audio",
1467 stream->ts_audio_pids[aud_pid_index],
1468 stream->ts_stream_type[1 + aud_pid_index]);
1471 fseeko(stream->file_handle, cur_pos, SEEK_SET);
1475 static void add_audio_to_title(hb_title_t *title, int id)
1477 hb_audio_t *audio = calloc( sizeof( hb_audio_t ), 1 );
1483 audio->config.in.codec = HB_ACODEC_MPGA;
1484 hb_log("add_audio_to_title: added MPEG audio stream 0x%x", id);
1487 // type 2 is a DVD subtitle stream - just ignore it */
1491 audio->config.in.codec = HB_ACODEC_AC3;
1492 hb_log("add_audio_to_title: added AC3 audio stream 0x%x", id);
1495 audio->config.in.codec = HB_ACODEC_LPCM;
1496 hb_log("add_audio_to_title: added LPCM audio stream 0x%x", id);
1499 hb_log("add_audio_to_title: unknown audio stream type 0x%x", id);
1504 set_audio_description( audio, lang_for_code( 0 ) );
1505 hb_list_add( title->list_audio, audio );
1508 static void hb_ps_stream_find_audio_ids(hb_stream_t *stream, hb_title_t *title)
1510 off_t cur_pos = ftello(stream->file_handle);
1511 hb_buffer_t *buf = hb_buffer_init(HB_DVD_READ_BUFFER_SIZE);
1512 hb_list_t *list = hb_list_init();
1513 // how many blocks we read while searching for audio streams
1514 int blksleft = 4096;
1515 // there can be at most 16 unique streams in an MPEG PS (8 in a DVD)
1516 // so we use a bitmap to keep track of the ones we've already seen.
1517 // Bit 'i' of smap is set if we've already added the audio for
1518 // audio substream id 'i' to the title's audio list.
1521 // start looking 20% into the file since there's occasionally no
1522 // audio at the beginning (particularly for vobs).
1523 hb_stream_seek(stream, 0.2f);
1525 while (--blksleft >= 0 && hb_stream_read(stream, buf) == 1)
1529 // 'buf' contains an MPEG2 PACK - get a list of all it's elementary streams
1530 hb_demux_ps( buf, list, 0 );
1532 while ( ( es = hb_list_item( list, 0 ) ) )
1534 hb_list_rem( list, es );
1535 if ( (es->id & 0xff) == 0xbd || (es->id & 0xe0) == 0xc0 )
1537 // this PES contains some kind of audio - get the substream id
1538 // and check if we've seen it already.
1539 int ssid = (es->id > 0xff ? es->id >> 8 : es->id) & 0xf;
1540 if ( (smap & (1 << ssid)) == 0 )
1542 // we haven't seen this stream before - add it to the
1543 // title's list of audio streams.
1544 smap |= (1 << ssid);
1545 add_audio_to_title(title, es->id);
1548 hb_buffer_close( &es );
1551 hb_list_empty( &list );
1552 hb_buffer_close(&buf);
1553 fseeko(stream->file_handle, cur_pos, SEEK_SET);
1556 /***********************************************************************
1558 ***********************************************************************
1560 **********************************************************************/
1562 static void hb_ts_stream_init(hb_stream_t *stream)
1566 for (i=0; i < kMaxNumberDecodeStreams; i++)
1568 stream->ts_streamcont[i] = -1;
1570 stream->ts_video_pids[0] = -1;
1571 for ( i = 0; i < stream->ts_number_audio_pids; i++ )
1573 stream-> ts_audio_pids[i] = -1;
1576 stream->ts_packet = malloc( stream->packetsize );
1578 // Find the audio and video pids in the stream
1579 hb_ts_stream_find_pids(stream);
1581 for (i = 0; i < stream->ts_number_video_pids + stream->ts_number_audio_pids; i++)
1583 // demuxing buffer for TS to PS conversion
1584 stream->ts_buf[i] = hb_buffer_init(stream->packetsize);
1585 stream->ts_buf[i]->size = 0;
1589 #define MAX_HOLE 208*80
1591 static off_t align_to_next_packet(hb_stream_t *stream)
1593 uint8_t buf[MAX_HOLE];
1595 off_t start = ftello(stream->file_handle);
1597 if ( start >= stream->packetsize ) {
1598 start -= stream->packetsize;
1599 fseeko(stream->file_handle, start, SEEK_SET);
1602 if (fread(buf, sizeof(buf), 1, stream->file_handle) == 1)
1604 const uint8_t *bp = buf;
1607 for ( i = sizeof(buf); --i >= 0; ++bp )
1609 if ( have_ts_sync( bp, stream->packetsize ) )
1616 pos = ( bp - buf ) - stream->packetsize + 188;
1621 fseeko(stream->file_handle, start+pos, SEEK_SET);
1632 static const unsigned int bitmask[] = {
1633 0x0,0x1,0x3,0x7,0xf,0x1f,0x3f,0x7f,0xff,
1634 0x1ff,0x3ff,0x7ff,0xfff,0x1fff,0x3fff,0x7fff,0xffff,
1635 0x1ffff,0x3ffff,0x7ffff,0xfffff,0x1fffff,0x3fffff,0x7fffff,0xffffff,
1636 0x1ffffff,0x3ffffff,0x7ffffff,0xfffffff,0x1fffffff,0x3fffffff,0x7fffffff,0xffffffff};
1638 static inline void set_buf(bitbuf_t *bb, uint8_t* buf, int bufsize, int clear)
1642 bb->val = (bb->buf[0] << 24) | (bb->buf[1] << 16) |
1643 (bb->buf[2] << 8) | bb->buf[3];
1645 memset(bb->buf, 0, bufsize);
1648 static inline int buf_size(bitbuf_t *bb)
1650 return bb->pos >> 3;
1653 static inline unsigned int get_bits(bitbuf_t *bb, int bits)
1656 int left = 32 - (bb->pos & 31);
1660 val = (bb->val >> (left - bits)) & bitmask[bits];
1665 val = (bb->val & bitmask[left]) << (bits - left);
1669 int pos = bb->pos >> 3;
1670 bb->val = (bb->buf[pos] << 24) | (bb->buf[pos + 1] << 16) | (bb->buf[pos + 2] << 8) | bb->buf[pos + 3];
1674 val |= (bb->val >> (32 - bits)) & bitmask[bits];
1682 // extract what useful information we can from the elementary stream
1683 // descriptor list at 'dp' and add it to the stream at 'esindx'.
1684 // Descriptors with info we don't currently use are ignored.
1685 // The descriptor list & descriptor item formats are defined in
1686 // ISO 13818-1 (2000E) section 2.6 (pg. 62).
1687 static void decode_element_descriptors(hb_stream_t* stream, int esindx,
1688 const uint8_t *dp, uint8_t dlen)
1690 const uint8_t *ep = dp + dlen;
1696 case 5: // Registration descriptor
1697 stream->ts_format_id[esindx] = (dp[2] << 24) | (dp[3] << 16) |
1698 (dp[4] << 8) | dp[5];
1701 case 10: // ISO_639_language descriptor
1702 stream->a52_info[esindx].lang_code = lang_to_code(lang_for_code2((const char *)&dp[2]));
1705 case 0x6a: // DVB AC-3 descriptor
1706 stream->ts_stream_type[esindx+1] = 0x81;
1716 static const char *stream_type_name (uint8_t stream_type)
1718 return st2codec[stream_type].name? st2codec[stream_type].name : "Unknown";
1721 int decode_program_map(hb_stream_t* stream)
1724 set_buf(&bb, stream->pmt_info.tablebuf, stream->pmt_info.tablepos, 0);
1726 get_bits(&bb, 8); // table_id
1728 unsigned int section_length = get_bits(&bb, 12);
1729 stream->pmt_info.section_length = section_length;
1731 unsigned int program_number = get_bits(&bb, 16);
1732 stream->pmt_info.program_number = program_number;
1734 get_bits(&bb, 5); // version_number
1736 get_bits(&bb, 8); // section_number
1737 get_bits(&bb, 8); // last_section_number
1739 unsigned int PCR_PID = get_bits(&bb, 13);
1740 stream->pmt_info.PCR_PID = PCR_PID;
1742 unsigned int program_info_length = get_bits(&bb, 12);
1743 stream->pmt_info.program_info_length = program_info_length;
1746 unsigned char *descriptor_buf = (unsigned char *) malloc(program_info_length);
1747 for (i = 0; i < program_info_length; i++)
1749 descriptor_buf[i] = get_bits(&bb, 8);
1752 int cur_pos = 9 /* data after the section length field*/ + program_info_length;
1753 int done_reading_stream_types = 0;
1754 while (!done_reading_stream_types)
1756 unsigned char stream_type = get_bits(&bb, 8);
1758 unsigned int elementary_PID = get_bits(&bb, 13);
1760 unsigned int ES_info_length = get_bits(&bb, 12);
1763 unsigned char *ES_info_buf = (unsigned char *) malloc(ES_info_length);
1764 for (i=0; i < ES_info_length; i++)
1766 ES_info_buf[i] = get_bits(&bb, 8);
1770 if ( index_of_pid( elementary_PID, stream ) < 0 )
1772 // already have this pid - do nothing
1774 if (stream->ts_number_video_pids == 0 && st2codec[stream_type].kind == V )
1776 stream->ts_video_pids[0] = elementary_PID;
1777 stream->ts_stream_type[0] = stream_type;
1778 stream->ts_number_video_pids = 1;
1782 // Defined audio stream types are 0x81 for AC-3/A52 audio and 0x03
1783 // for mpeg audio. But content producers seem to use other
1784 // values (0x04 and 0x06 have both been observed) so at this point
1785 // we say everything that isn't a video pid is audio then at the end
1786 // of hb_stream_title_scan we'll figure out which are really audio
1787 // by looking at the PES headers.
1788 i = stream->ts_number_audio_pids;
1789 if (i < kMaxNumberAudioPIDS)
1791 stream->ts_audio_pids[i] = elementary_PID;
1792 stream->ts_stream_type[1 + i] = stream_type;
1793 if (ES_info_length > 0)
1795 decode_element_descriptors(stream, i, ES_info_buf,
1798 ++stream->ts_number_audio_pids;
1802 cur_pos += 5 /* stream header */ + ES_info_length;
1806 if (cur_pos >= section_length - 4 /* stop before the CRC */)
1807 done_reading_stream_types = 1;
1810 free(descriptor_buf);
1814 static int build_program_map(const uint8_t *buf, hb_stream_t *stream)
1816 // Get adaption header info
1818 int adaption = (buf[3] & 0x30) >> 4;
1821 else if (adaption == 0x2)
1823 else if (adaption == 0x3)
1824 adapt_len = buf[4] + 1;
1825 if (adapt_len > 184)
1828 // Get payload start indicator
1830 start = (buf[1] & 0x40) != 0;
1832 // Get pointer length - only valid in packets with a start flag
1833 int pointer_len = 0;
1837 pointer_len = buf[4 + adapt_len] + 1;
1838 stream->pmt_info.tablepos = 0;
1840 // Get Continuity Counter
1841 int continuity_counter = buf[3] & 0x0f;
1842 if (!start && (stream->pmt_info.current_continuity_counter + 1 != continuity_counter))
1844 hb_log("build_program_map - Continuity Counter %d out of sequence - expected %d", continuity_counter, stream->pmt_info.current_continuity_counter+1);
1847 stream->pmt_info.current_continuity_counter = continuity_counter;
1848 stream->pmt_info.reading |= start;
1850 // Add the payload for this packet to the current buffer
1851 int amount_to_copy = 184 - adapt_len - pointer_len;
1852 if (stream->pmt_info.reading && (amount_to_copy > 0))
1854 stream->pmt_info.tablebuf = realloc(stream->pmt_info.tablebuf, stream->pmt_info.tablepos + amount_to_copy);
1856 memcpy(stream->pmt_info.tablebuf + stream->pmt_info.tablepos, buf + 4 + adapt_len + pointer_len, amount_to_copy);
1857 stream->pmt_info.tablepos += amount_to_copy;
1859 if (stream->pmt_info.tablepos > 3)
1861 // We have enough to check the section length
1863 length = ((stream->pmt_info.tablebuf[1] << 8) +
1864 stream->pmt_info.tablebuf[2]) & 0xFFF;
1865 if (stream->pmt_info.tablepos > length + 1)
1867 // We just finished a bunch of packets - parse the program map details
1869 if (stream->pmt_info.tablebuf[0] == 0x02)
1870 decode_ok = decode_program_map(stream);
1871 free(stream->pmt_info.tablebuf);
1872 stream->pmt_info.tablebuf = NULL;
1873 stream->pmt_info.tablepos = 0;
1874 stream->pmt_info.reading = 0;
1884 static int decode_PAT(const uint8_t *buf, hb_stream_t *stream)
1886 unsigned char tablebuf[1024];
1887 unsigned int tablepos = 0;
1892 // Get adaption header info
1894 int adaption = (buf[3] & 0x30) >> 4;
1897 else if (adaption == 0x2)
1899 else if (adaption == 0x3)
1900 adapt_len = buf[4] + 1;
1901 if (adapt_len > 184)
1904 // Get pointer length
1905 int pointer_len = buf[4 + adapt_len] + 1;
1907 // Get payload start indicator
1909 start = (buf[1] & 0x40) != 0;
1914 // Add the payload for this packet to the current buffer
1915 if (reading && (184 - adapt_len) > 0)
1917 if (tablepos + 184 - adapt_len - pointer_len > 1024)
1919 hb_log("decode_PAT - Bad program section length (> 1024)");
1922 memcpy(tablebuf + tablepos, buf + 4 + adapt_len + pointer_len, 184 - adapt_len - pointer_len);
1923 tablepos += 184 - adapt_len - pointer_len;
1926 if (start && reading)
1928 memcpy(tablebuf + tablepos, buf + 4 + adapt_len + 1, pointer_len - 1);
1931 unsigned int pos = 0;
1932 //while (pos < tablepos)
1935 set_buf(&bb, tablebuf + pos, tablepos - pos, 0);
1937 unsigned char section_id = get_bits(&bb, 8);
1939 unsigned int section_len = get_bits(&bb, 12);
1940 get_bits(&bb, 16); // transport_id
1942 get_bits(&bb, 5); // version_num
1943 get_bits(&bb, 1); // current_next
1944 get_bits(&bb, 8); // section_num
1945 get_bits(&bb, 8); // last_section
1951 // Program Association Section
1952 section_len -= 5; // Already read transport stream ID, version num, section num, and last section num
1953 section_len -= 4; // Ignore the CRC
1955 stream->ts_number_pat_entries = 0;
1956 while ((curr_pos < section_len) && (stream->ts_number_pat_entries < kMaxNumberPMTStreams))
1958 unsigned int pkt_program_num = get_bits(&bb, 16);
1959 stream->pat_info[stream->ts_number_pat_entries].program_number = pkt_program_num;
1961 get_bits(&bb, 3); // Reserved
1962 if (pkt_program_num == 0)
1964 get_bits(&bb, 13); // pkt_network_id
1968 unsigned int pkt_program_map_PID = get_bits(&bb, 13);
1969 stream->pat_info[stream->ts_number_pat_entries].program_map_PID = pkt_program_map_PID;
1972 stream->ts_number_pat_entries++;
1986 pos += 3 + section_len;
1994 static void hb_ts_stream_find_pids(hb_stream_t *stream)
1996 // To be different from every other broadcaster in the world, New Zealand TV
1997 // changes PMTs (and thus video & audio PIDs) when 'programs' change. Since
1998 // we may have the tail of the previous program at the beginning of this
1999 // file, take our PMT from the middle of the file.
2000 fseeko(stream->file_handle, 0, SEEK_END);
2001 uint64_t fsize = ftello(stream->file_handle);
2002 fseeko(stream->file_handle, fsize >> 1, SEEK_SET);
2003 align_to_next_packet(stream);
2005 // Read the Transport Stream Packets (188 bytes each) looking at first for PID 0 (the PAT PID), then decode that
2006 // to find the program map PID and then decode that to get the list of audio and video PIDs
2010 const uint8_t *buf = next_packet( stream );
2014 hb_log("hb_ts_stream_find_pids - end of file");
2019 int pid = (((buf[1] & 0x1F) << 8) | buf[2]) & 0x1FFF;
2021 if ((pid == 0x0000) && (stream->ts_number_pat_entries == 0))
2023 decode_PAT(buf, stream);
2028 for (pat_index = 0; pat_index < stream->ts_number_pat_entries; pat_index++)
2030 // There are some streams where the PAT table has multiple entries as if their are
2031 // multiple programs in the same transport stream, and yet there's actually only one
2032 // program really in the stream. This seems to be true for transport streams that
2033 // originate in the HDHomeRun but have been output by EyeTV's export utility. What I think
2034 // is happening is that the HDHomeRun is sending the entire transport stream as broadcast,
2035 // but the EyeTV is only recording a single (selected) program number and not rewriting the
2036 // PAT info on export to match what's actually on the stream.
2037 // Until we have a way of handling multiple programs per transport stream elegantly we'll match
2038 // on the first pat entry for which we find a matching program map PID. The ideal solution would
2039 // be to build a title choice popup from the PAT program number details and then select from
2040 // their - but right now the API's not capable of that.
2041 if (stream->pat_info[pat_index].program_number != 0 &&
2042 pid == stream->pat_info[pat_index].program_map_PID)
2044 if (build_program_map(buf, stream) > 0)
2048 // Keep going until we have a complete set of PIDs
2049 if ((stream->ts_number_video_pids > 0) && (stream->ts_number_audio_pids > 0))
2053 hb_log("hb_ts_stream_find_pids - found the following PIDS");
2054 hb_log(" Video PIDS : ");
2056 for (i=0; i < stream->ts_number_video_pids; i++)
2058 hb_log( " 0x%x type %s (0x%x)",
2059 stream->ts_video_pids[i],
2060 stream_type_name(stream->ts_stream_type[i]),
2061 stream->ts_stream_type[i]);
2063 hb_log(" Audio PIDS : ");
2064 for (i = 0; i < stream->ts_number_audio_pids; i++)
2066 hb_log( " 0x%x type %s (0x%x)",
2067 stream->ts_audio_pids[i],
2068 stream_type_name(stream->ts_stream_type[i+1]),
2069 stream->ts_stream_type[i+1] );
2074 static void fwrite64( hb_stream_t *stream, void *buf, int len )
2078 int pos = stream->fwrite_buf->size;
2079 if ( pos + len > stream->fwrite_buf->alloc )
2081 int size = MAX(stream->fwrite_buf->alloc * 2, pos + len);
2082 hb_buffer_realloc(stream->fwrite_buf, size);
2084 memcpy( &(stream->fwrite_buf->data[pos]), buf, len );
2085 stream->fwrite_buf->size += len;
2089 // convert a PES PTS or DTS to an int64
2090 static int64_t pes_timestamp( const uint8_t *pes )
2092 int64_t ts = ( (uint64_t)(pes[0] & 0xe ) << 29 );
2093 ts |= ( pes[1] << 22 ) | ( ( pes[2] >> 1 ) << 15 ) |
2094 ( pes[3] << 7 ) | ( pes[4] >> 1 );
2098 static void generate_output_data(hb_stream_t *stream, int curstream)
2100 hb_buffer_t *buf = stream->fwrite_buf;
2101 uint8_t *tdat = stream->ts_buf[curstream]->data;
2103 buf->id = curstream;
2105 // check if this packet was referenced to an older pcr and if that
2106 // pcr was significantly different than the one we're using now.
2107 // (the reason for the uint cast on the pcr difference is that the
2108 // difference is significant if it advanced by more than 200ms or if
2109 // it went backwards by any amount. The negative numbers look like huge
2110 // unsigned ints so the cast allows both conditions to be checked at once.
2111 int bufpcr = stream->ts_buf[curstream]->cur;
2112 int curpcr = stream->ts_pcr_out;
2113 if ( bufpcr && bufpcr < curpcr &&
2114 (uint64_t)(stream->ts_pcrhist[curpcr & 3] - stream->ts_pcrhist[bufpcr & 3]) > 200*90LL )
2116 // we've sent up a new pcr but have a packet referenced to an
2117 // old pcr and the difference was enough to trigger a discontinuity
2118 // correction. smash the timestamps or we'll mess up the correction.
2120 buf->renderOffset = -1;
2124 if ( stream->ts_pcr_out != stream->ts_pcr_in )
2126 // we have a new pcr
2127 stream->ts_pcr_out = stream->ts_pcr_in;
2128 buf->stop = stream->ts_pcr;
2129 stream->ts_pcrhist[stream->ts_pcr_out & 3] = stream->ts_pcr;
2136 // put the PTS & possible DTS into 'start' & 'renderOffset' then strip
2137 // off the PES header.
2138 if ( tdat[7] & 0xc0 )
2140 buf->start = pes_timestamp( tdat + 9 );
2141 buf->renderOffset = ( tdat[7] & 0x40 )? pes_timestamp( tdat + 14 ) :
2147 buf->renderOffset = -1;
2150 int hlen = tdat[8] + 9;
2152 fwrite64( stream, tdat + hlen, stream->ts_pos[curstream] - hlen );
2154 stream->ts_pos[curstream] = 0;
2155 stream->ts_buf[curstream]->size = 0;
2158 static void hb_ts_stream_append_pkt(hb_stream_t *stream, int idx, const uint8_t *buf, int len)
2160 if (stream->ts_pos[idx] + len > stream->ts_buf[idx]->alloc)
2164 size = MAX(stream->ts_buf[idx]->alloc * 2, stream->ts_pos[idx] + len);
2165 hb_buffer_realloc(stream->ts_buf[idx], size);
2167 memcpy(stream->ts_buf[idx]->data + stream->ts_pos[idx], buf, len);
2168 stream->ts_pos[idx] += len;
2169 stream->ts_buf[idx]->size += len;
2172 /***********************************************************************
2173 * hb_ts_stream_decode
2174 ***********************************************************************
2176 **********************************************************************/
2177 static int hb_ts_stream_decode( hb_stream_t *stream, hb_buffer_t *obuf )
2180 * stash the output buffer pointer in our stream so we don't have to
2181 * pass it & its original value to everything we call.
2184 stream->fwrite_buf = obuf;
2186 // spin until we get a packet of data from some stream or hit eof
2191 const uint8_t *buf = next_packet(stream);
2194 // end of file - we didn't finish filling our ps write buffer
2195 // so just discard the remainder (the partial buffer is useless)
2196 hb_log("hb_ts_stream_decode - eof");
2200 /* This next section validates the packet */
2202 // Get pid and use it to find stream state.
2203 int pid = ((buf[1] & 0x1F) << 8) | buf[2];
2204 if ( ( curstream = index_of_pid( pid, stream ) ) < 0 )
2208 int errorbit = (buf[1] & 0x80) != 0;
2211 ts_err( stream, curstream, "packet error bit set");
2215 // Get adaption header info
2216 int adaption = (buf[3] & 0x30) >> 4;
2220 ts_err( stream, curstream, "adaptation code 0");
2223 else if (adaption == 0x2)
2225 else if (adaption == 0x3)
2227 adapt_len = buf[4] + 1;
2228 if (adapt_len > 184)
2230 ts_err( stream, curstream, "invalid adapt len %d", adapt_len);
2235 if ( adapt_len > 0 )
2237 if ( buf[5] & 0x40 )
2239 // found a random access point
2241 // if there's an adaptation header & PCR_flag is set
2242 // get the PCR (Program Clock Reference)
2243 if ( adapt_len > 7 && ( buf[5] & 0x10 ) != 0 )
2245 stream->ts_pcr = ( (uint64_t)buf[6] << (33 - 8) ) |
2246 ( (uint64_t)buf[7] << (33 - 16) ) |
2247 ( (uint64_t)buf[8] << (33 - 24) ) |
2248 ( (uint64_t)buf[9] << (33 - 32) ) |
2250 ++stream->ts_pcr_in;
2251 stream->ts_found_pcr = 1;
2255 // If we don't have a PCR yet but the stream has PCRs just loop
2256 // so we don't process anything until we have a clock reference.
2257 // Unfortunately the HD Home Run appears to null out the PCR so if
2258 // we didn't detect a PCR during scan keep going and we'll use
2259 // the video stream DTS for the PCR.
2261 if ( !stream->ts_found_pcr && ( stream->ts_flags & TS_HAS_PCR ) )
2267 // Continuity only increments for adaption values of 0x3 or 0x01
2268 // and is not checked for start packets.
2270 int start = (buf[1] & 0x40) != 0;
2272 if ( (adaption & 0x01) != 0 )
2274 int continuity = (buf[3] & 0xF);
2275 if ( continuity == stream->ts_streamcont[curstream] )
2277 // we got a duplicate packet (usually used to introduce
2278 // a PCR when one is needed). The only thing that can
2279 // change in the dup is the PCR which we grabbed above
2280 // so ignore the rest.
2283 if ( !start && (stream->ts_streamcont[curstream] != -1) &&
2284 !stream->ts_skipbad[curstream] &&
2285 (continuity != ( (stream->ts_streamcont[curstream] + 1) & 0xf ) ) )
2287 ts_err( stream, curstream, "continuity error: got %d expected %d",
2289 (stream->ts_streamcont[curstream] + 1) & 0xf );
2290 stream->ts_streamcont[curstream] = continuity;
2293 stream->ts_streamcont[curstream] = continuity;
2296 /* If we get here the packet is valid - process its data */
2300 // Found a random access point (now we can start a frame/audio packet..)
2302 if ( stream->need_keyframe )
2304 // we're looking for the first video frame because we're
2305 // doing random access during 'scan'
2306 if ( curstream != 0 || !isIframe( stream, buf, adapt_len ) )
2308 // not the video stream or didn't find an I frame
2309 // but we'll only wait 255 video frames for an I frame.
2310 if ( curstream != 0 || ++stream->need_keyframe )
2315 stream->need_keyframe = 0;
2318 // If we were skipping a bad packet, start fresh on this new PES packet..
2319 if (stream->ts_skipbad[curstream] == 1)
2321 stream->ts_skipbad[curstream] = 0;
2324 if ( curstream == 0 )
2328 // if we don't have a pcr yet use the dts from this frame
2329 if ( !stream->ts_found_pcr )
2331 // PES must begin with an mpeg start code & contain
2333 const uint8_t *pes = buf + adapt_len + 4;
2334 if ( pes[0] != 0x00 || pes[1] != 0x00 || pes[2] != 0x01 ||
2335 ( pes[7] >> 6 ) == 0 )
2339 // if we have a dts use it otherwise use the pts
2340 stream->ts_pcr = pes_timestamp( pes + ( pes[7] & 0x40? 14 : 9 ) );
2341 ++stream->ts_pcr_in;
2345 // if this is a multiplexed stream make sure this is the
2346 // substream we want.
2347 if ( stream->ts_multiplexed[curstream] )
2349 // PES must begin with an mpeg start code & contain
2351 const uint8_t *pes = buf + adapt_len + 4;
2352 if ( pes[0] != 0x00 || pes[1] != 0x00 || pes[2] != 0x01 ||
2355 stream->ts_skipbad[curstream] = 1;
2358 // the last byte of the header is the extension id. see if
2359 // it's the one we want.
2360 if ( pes[pes[8]+8] != stream->ts_multiplexed[curstream] )
2362 stream->ts_skipbad[curstream] = 1;
2367 // If we have some data already on this stream, turn it into
2368 // a program stream packet. Then add the payload for this
2369 // packet to the current pid's buffer.
2370 if ( stream->ts_pos[curstream] )
2372 // we have to ship the old packet before updating the pcr
2373 // since the packet we've been accumulating is referenced
2375 generate_output_data(stream, curstream);
2377 // remember the pcr that was in effect when we started
2379 stream->ts_buf[curstream]->cur = stream->ts_pcr_in;
2380 hb_ts_stream_append_pkt(stream, curstream, buf + 4 + adapt_len,
2384 // remember the pcr that was in effect when we started this packet.
2385 stream->ts_buf[curstream]->cur = stream->ts_pcr_in;
2388 // Add the payload for this packet to the current buffer
2389 if (!stream->ts_skipbad[curstream] && (184 - adapt_len) > 0)
2391 hb_ts_stream_append_pkt(stream, curstream, buf + 4 + adapt_len,
2393 // see if we've hit the end of this PES packet
2394 const uint8_t *pes = stream->ts_buf[curstream]->data;
2395 int len = ( pes[4] << 8 ) + pes[5] + 6;
2396 if ( len > 6 && stream->ts_pos[curstream] == len &&
2397 pes[0] == 0x00 && pes[1] == 0x00 && pes[2] == 0x01 )
2399 generate_output_data(stream, curstream);
2406 static void hb_ts_stream_reset(hb_stream_t *stream)
2410 for (i=0; i < kMaxNumberDecodeStreams; i++)
2412 stream->ts_pos[i] = 0;
2413 stream->ts_skipbad[i] = 1;
2414 stream->ts_streamcont[i] = -1;
2417 stream->need_keyframe = 0;
2419 stream->ts_found_pcr = 0;
2420 stream->ts_pcr_out = 0;
2421 stream->ts_pcr_in = 0;
2426 stream->last_error_frame = -10000;
2427 stream->last_error_count = 0;
2429 align_to_next_packet(stream);
2432 // ------------------------------------------------------------------
2433 // Support for reading media files via the ffmpeg libraries.
2435 static void ffmpeg_add_codec( hb_stream_t *stream, int stream_index )
2437 // add a codec to the context here so it will be there when we
2438 // read the first packet.
2439 AVCodecContext *context = stream->ffmpeg_ic->streams[stream_index]->codec;
2440 context->workaround_bugs = FF_BUG_AUTODETECT;
2441 context->error_recognition = 1;
2442 context->error_concealment = FF_EC_GUESS_MVS|FF_EC_DEBLOCK;
2443 AVCodec *codec = avcodec_find_decoder( context->codec_id );
2444 hb_avcodec_open( context, codec );
2447 // The ffmpeg stream reader / parser shares a lot of state with the
2448 // decoder via a codec context kept in the AVStream of the reader's
2449 // AVFormatContext. Since decoding is done in a different thread we
2450 // have to somehow pass this codec context to the decoder and we have
2451 // to do it before the first packet is read (so we can't put the info
2452 // in the buf we'll send downstream). Decoders don't have any way to
2453 // get to the stream directly (they're not passed the title or job
2454 // pointers during a scan) so this is a back door for the decoder to
2455 // get the codec context. We just stick the stream pointer in the next
2456 // slot an array of pointers maintained as a circular list then return
2457 // the index into the list combined with the ffmpeg stream index as the
2458 // codec_param that will be passed to the decoder init routine. We make
2459 // the list 'big' (enough for 1024 simultaneously open ffmpeg streams)
2460 // so that we don't have to do a complicated allocator or worry about
2461 // deleting entries on close.
2463 // Entries can only be added to this list during a scan and are never
2464 // deleted so the list access doesn't require locking.
2465 static hb_stream_t **ffmpeg_streams; // circular list of stream pointers
2466 static int ffmpeg_stream_cur; // where we put the last stream pointer
2467 #define ffmpeg_sl_bits (10) // log2 stream list size (in entries)
2468 #define ffmpeg_sl_size (1 << ffmpeg_sl_bits)
2470 // add a stream to the list & return the appropriate codec_param to access it
2471 static int ffmpeg_codec_param( hb_stream_t *stream, int stream_index )
2473 if ( !ffmpeg_streams )
2475 ffmpeg_streams = calloc( ffmpeg_sl_size, sizeof(stream) );
2478 // the title scan adds all the ffmpeg media streams at once so we
2479 // only add a new entry to our stream list if the stream is different
2481 int slot = ffmpeg_stream_cur;
2482 if ( ffmpeg_streams[slot] != stream )
2484 // new stream - put it in the next slot of the stream list
2485 slot = ++ffmpeg_stream_cur & (ffmpeg_sl_size - 1);
2486 ffmpeg_streams[slot] = stream;
2489 ffmpeg_add_codec( stream, stream_index );
2491 return ( stream_index << ffmpeg_sl_bits ) | slot;
2494 // we're about to open 'title' to convert it - remap the stream associated
2495 // with the video & audio codec params of the title to refer to 'stream'
2496 // (the original scan stream was closed and no longer exists).
2497 static void ffmpeg_remap_stream( hb_stream_t *stream, hb_title_t *title )
2499 // all the video & audio came from the same stream so remapping
2500 // the video's stream slot takes care of everything.
2501 int slot = title->video_codec_param & (ffmpeg_sl_size - 1);
2502 ffmpeg_streams[slot] = stream;
2504 // add codecs for all the streams used by the title
2505 ffmpeg_add_codec( stream, title->video_codec_param >> ffmpeg_sl_bits );
2509 for ( i = 0; ( audio = hb_list_item( title->list_audio, i ) ); ++i )
2511 if ( audio->config.in.codec == HB_ACODEC_FFMPEG )
2513 ffmpeg_add_codec( stream,
2514 audio->config.in.codec_param >> ffmpeg_sl_bits );
2519 void *hb_ffmpeg_context( int codec_param )
2521 int slot = codec_param & (ffmpeg_sl_size - 1);
2522 int stream_index = codec_param >> ffmpeg_sl_bits;
2523 return ffmpeg_streams[slot]->ffmpeg_ic->streams[stream_index]->codec;
2526 void *hb_ffmpeg_avstream( int codec_param )
2528 int slot = codec_param & (ffmpeg_sl_size - 1);
2529 int stream_index = codec_param >> ffmpeg_sl_bits;
2530 return ffmpeg_streams[slot]->ffmpeg_ic->streams[stream_index];
2533 static AVFormatContext *ffmpeg_deferred_close;
2535 static int ffmpeg_open( hb_stream_t *stream, hb_title_t *title )
2537 if ( ffmpeg_deferred_close )
2539 av_close_input_file( ffmpeg_deferred_close );
2540 ffmpeg_deferred_close = NULL;
2542 AVFormatContext *ic;
2544 av_log_set_level( AV_LOG_ERROR );
2545 if ( av_open_input_file( &ic, stream->path, NULL, 0, NULL ) < 0 )
2549 if ( av_find_stream_info( ic ) < 0 )
2552 stream->ffmpeg_ic = ic;
2553 stream->hb_stream_type = ffmpeg;
2554 stream->ffmpeg_pkt = malloc(sizeof(*stream->ffmpeg_pkt));
2555 av_init_packet( stream->ffmpeg_pkt );
2556 stream->chapter_end = INT64_MAX;
2560 // we're opening for read. scan passed out codec params that
2561 // indexed its stream so we need to remap them so they point
2563 ffmpeg_remap_stream( stream, title );
2564 av_log_set_level( AV_LOG_ERROR );
2568 // we're opening for scan. let ffmpeg put some info into the
2569 // log about what we've got.
2570 av_log_set_level( AV_LOG_INFO );
2571 dump_format( ic, 0, stream->path, 0 );
2572 av_log_set_level( AV_LOG_ERROR );
2574 // accept this file if it has at least one video stream we can decode
2576 for (i = 0; i < ic->nb_streams; ++i )
2578 if ( ic->streams[i]->codec->codec_type == CODEC_TYPE_VIDEO )
2583 if ( i >= ic->nb_streams )
2589 av_close_input_file( ic );
2593 static void ffmpeg_close( hb_stream_t *d )
2595 // XXX since we're sharing the CodecContext with the downstream
2596 // decoder proc we can't close the stream. We need to reference count
2597 // this so we can close it when both are done with their instance but
2598 // for now just defer the close until the next stream open or close.
2599 if ( ffmpeg_deferred_close )
2601 av_close_input_file( ffmpeg_deferred_close );
2603 ffmpeg_deferred_close = d->ffmpeg_ic;
2604 if ( d->ffmpeg_pkt != NULL )
2606 free( d->ffmpeg_pkt );
2607 d->ffmpeg_pkt = NULL;
2611 static void add_ffmpeg_audio( hb_title_t *title, hb_stream_t *stream, int id )
2613 AVStream *st = stream->ffmpeg_ic->streams[id];
2614 AVCodecContext *codec = st->codec;
2616 // scan will ignore any audio without a bitrate. Since we've already
2617 // typed the audio in order to determine its codec we set up the audio
2619 if ( codec->bit_rate || codec->sample_rate )
2621 static const int chan2layout[] = {
2622 HB_INPUT_CH_LAYOUT_MONO, // We should allow no audio really.
2623 HB_INPUT_CH_LAYOUT_MONO,
2624 HB_INPUT_CH_LAYOUT_STEREO,
2625 HB_INPUT_CH_LAYOUT_2F1R,
2626 HB_INPUT_CH_LAYOUT_2F2R,
2627 HB_INPUT_CH_LAYOUT_3F2R,
2628 HB_INPUT_CH_LAYOUT_4F2R,
2629 HB_INPUT_CH_LAYOUT_STEREO,
2630 HB_INPUT_CH_LAYOUT_STEREO,
2633 hb_audio_t *audio = calloc( 1, sizeof(*audio) );;
2636 if ( codec->codec_id == CODEC_ID_AC3 )
2638 audio->config.in.codec = HB_ACODEC_AC3;
2640 else if ( codec->codec_id == CODEC_ID_DTS )
2642 audio->config.in.codec = HB_ACODEC_DCA;
2646 audio->config.in.codec = HB_ACODEC_FFMPEG;
2647 audio->config.in.codec_param = ffmpeg_codec_param( stream, id );
2649 audio->config.in.bitrate = codec->bit_rate? codec->bit_rate : 1;
2650 audio->config.in.samplerate = codec->sample_rate;
2651 audio->config.in.channel_layout = chan2layout[codec->channels & 7];
2654 set_audio_description( audio, lang_for_code2( st->language ) );
2656 hb_list_add( title->list_audio, audio );
2660 static hb_title_t *ffmpeg_title_scan( hb_stream_t *stream )
2662 AVFormatContext *ic = stream->ffmpeg_ic;
2664 // 'Barebones Title'
2665 hb_title_t *title = hb_title_init( stream->path, 0 );
2668 // Copy part of the stream path to the title name
2669 char *sep = strrchr(stream->path, '/');
2671 strcpy(title->name, sep+1);
2672 char *dot_term = strrchr(title->name, '.');
2676 uint64_t dur = ic->duration * 90000 / AV_TIME_BASE;
2677 title->duration = dur;
2679 title->hours = dur / 3600;
2680 title->minutes = ( dur % 3600 ) / 60;
2681 title->seconds = dur % 60;
2683 // set the title to decode the first video stream in the file
2684 title->demuxer = HB_NULL_DEMUXER;
2685 title->video_codec = 0;
2687 for (i = 0; i < ic->nb_streams; ++i )
2689 if ( ic->streams[i]->codec->codec_type == CODEC_TYPE_VIDEO &&
2690 avcodec_find_decoder( ic->streams[i]->codec->codec_id ) &&
2691 title->video_codec == 0 )
2693 title->video_id = i;
2694 stream->ffmpeg_video_id = i;
2696 // We have to use the 'internal' avcodec decoder because
2697 // it needs to share the codec context from this video
2698 // stream. The parser internal to av_read_frame
2699 // passes a bunch of state info to the decoder via the context.
2700 title->video_codec = WORK_DECAVCODECVI;
2701 title->video_codec_param = ffmpeg_codec_param( stream, i );
2703 else if ( ic->streams[i]->codec->codec_type == CODEC_TYPE_AUDIO &&
2704 avcodec_find_decoder( ic->streams[i]->codec->codec_id ) )
2706 add_ffmpeg_audio( title, stream, i );
2710 title->container_name = strdup( ic->iformat->name );
2711 title->data_rate = ic->bit_rate;
2713 hb_deep_log( 2, "Found ffmpeg %d chapters, container=%s", ic->nb_chapters, ic->iformat->name );
2715 if( ic->nb_chapters != 0 )
2718 uint64_t duration_sum = 0;
2719 for( i = 0; i < ic->nb_chapters; i++ )
2720 if( ( m = ic->chapters[i] ) != NULL )
2722 hb_chapter_t * chapter;
2723 chapter = calloc( sizeof( hb_chapter_t ), 1 );
2724 chapter->index = i+1;
2725 chapter->duration = ( m->end / ( (double) m->time_base.num * m->time_base.den ) ) * 90000 - duration_sum;
2726 duration_sum += chapter->duration;
2727 chapter->hours = chapter->duration / 90000 / 3600;
2728 chapter->minutes = ( ( chapter->duration / 90000 ) % 3600 ) / 60;
2729 chapter->seconds = ( chapter->duration / 90000 ) % 60;
2730 strcpy( chapter->title, m->title );
2731 hb_deep_log( 2, "Added chapter %i, name='%s', dur=%"PRIu64", (%02i:%02i:%02i)",
2732 chapter->index, chapter->title,
2733 chapter->duration, chapter->hours,
2734 chapter->minutes, chapter->seconds );
2735 hb_list_add( title->list_chapter, chapter );
2740 * Fill the metadata.
2742 decmetadata( title );
2744 if( hb_list_count( title->list_chapter ) == 0 )
2746 // Need at least one chapter
2747 hb_chapter_t * chapter;
2748 chapter = calloc( sizeof( hb_chapter_t ), 1 );
2750 chapter->duration = title->duration;
2751 chapter->hours = title->hours;
2752 chapter->minutes = title->minutes;
2753 chapter->seconds = title->seconds;
2754 hb_list_add( title->list_chapter, chapter );
2760 static int64_t av_to_hb_pts( int64_t pts, double conv_factor )
2762 if ( pts == AV_NOPTS_VALUE )
2764 return (int64_t)( (double)pts * conv_factor );
2767 static int ffmpeg_is_keyframe( hb_stream_t *stream )
2771 switch ( stream->ffmpeg_ic->streams[stream->ffmpeg_video_id]->codec->codec_id )
2774 // XXX the VC1 codec doesn't mark key frames so to get previews
2775 // we do it ourselves here. The decoder gets messed up if it
2776 // doesn't get a SEQ header first so we consider that to be a key frame.
2777 pkt = stream->ffmpeg_pkt->data;
2778 if ( pkt[0] && pkt[1] && pkt[2] == 1 && pkt[3] == 0x0f )
2784 // XXX the ffmpeg WMV3 codec doesn't mark key frames.
2785 // Only M$ could make I-frame detection this complicated: there
2786 // are two to four bits of unused junk ahead of the frame type
2787 // so we have to look at the sequence header to find out how much
2788 // to skip. Then there are three different ways of coding the type
2789 // depending on whether it's main or advanced profile then whether
2790 // there are bframes or not so we have to look at the sequence
2791 // header to get that.
2792 pkt = stream->ffmpeg_pkt->data;
2793 uint8_t *seqhdr = stream->ffmpeg_ic->streams[stream->ffmpeg_video_id]->codec->extradata;
2795 if ( ( seqhdr[3] & 0x02 ) == 0 )
2798 if ( ( seqhdr[3] & 0x80 ) == 0 )
2799 // no RANGEREDUCTION
2801 if ( seqhdr[3] & 0x70 )
2802 // stream has b-frames
2803 return ( ( pkt[0] >> pshift ) & 0x3 ) == 0x01;
2805 return ( ( pkt[0] >> pshift ) & 0x2 ) == 0;
2810 return ( stream->ffmpeg_pkt->flags & PKT_FLAG_KEY );
2813 static int ffmpeg_read( hb_stream_t *stream, hb_buffer_t *buf )
2817 if ( ( err = av_read_frame( stream->ffmpeg_ic, stream->ffmpeg_pkt )) < 0 )
2819 // XXX the following conditional is to handle avi files that
2820 // use M$ 'packed b-frames' and occasionally have negative
2821 // sizes for the null frames these require.
2822 if ( err != AVERROR_NOMEM || stream->ffmpeg_pkt->size >= 0 )
2826 if ( stream->ffmpeg_pkt->size <= 0 )
2828 // M$ "invalid and inefficient" packed b-frames require 'null frames'
2829 // following them to preserve the timing (since the packing puts two
2830 // or more frames in what looks like one avi frame). The contents and
2831 // size of these null frames are ignored by the ff_h263_decode_frame
2832 // as long as they're < 20 bytes. We need a positive size so we use
2833 // one byte if we're given a zero or negative size. We don't know
2834 // if the pkt data points anywhere reasonable so we just stick a
2835 // byte of zero in our outbound buf.
2841 if ( stream->ffmpeg_pkt->size > buf->alloc )
2843 // sometimes we get absurd sizes from ffmpeg
2844 if ( stream->ffmpeg_pkt->size >= (1 << 25) )
2846 hb_log( "ffmpeg_read: pkt too big: %d bytes", stream->ffmpeg_pkt->size );
2847 av_free_packet( stream->ffmpeg_pkt );
2848 return ffmpeg_read( stream, buf );
2850 // need to expand buffer
2851 hb_buffer_realloc( buf, stream->ffmpeg_pkt->size );
2853 memcpy( buf->data, stream->ffmpeg_pkt->data, stream->ffmpeg_pkt->size );
2854 buf->size = stream->ffmpeg_pkt->size;
2856 buf->id = stream->ffmpeg_pkt->stream_index;
2857 if ( buf->id == stream->ffmpeg_video_id )
2859 if ( stream->need_keyframe )
2861 // we've just done a seek (generally for scan or live preview) and
2862 // want to start at a keyframe. Some ffmpeg codecs seek to a key
2863 // frame but most don't. So we spin until we either get a keyframe
2864 // or we've looked through 50 video frames without finding one.
2865 if ( ! ffmpeg_is_keyframe( stream ) && ++stream->need_keyframe < 50 )
2867 av_free_packet( stream->ffmpeg_pkt );
2870 stream->need_keyframe = 0;
2875 // if we haven't done it already, compute a conversion factor to go
2876 // from the ffmpeg timebase for the stream to HB's 90KHz timebase.
2877 double tsconv = stream->ffmpeg_tsconv[stream->ffmpeg_pkt->stream_index];
2880 AVStream *s = stream->ffmpeg_ic->streams[stream->ffmpeg_pkt->stream_index];
2881 tsconv = 90000. * (double)s->time_base.num / (double)s->time_base.den;
2882 stream->ffmpeg_tsconv[stream->ffmpeg_pkt->stream_index] = tsconv;
2885 buf->start = av_to_hb_pts( stream->ffmpeg_pkt->pts, tsconv );
2886 buf->renderOffset = av_to_hb_pts( stream->ffmpeg_pkt->dts, tsconv );
2887 if ( buf->renderOffset >= 0 && buf->start == -1 )
2889 buf->start = buf->renderOffset;
2893 * Check to see whether this video buffer is on a chapter
2894 * boundary, if so mark it as such in the buffer then advance
2895 * chapter_end to the end of the next chapter.
2896 * If there are no chapters, chapter_end is always initialized to INT64_MAX
2897 * (roughly 3 million years at our 90KHz clock rate) so the test
2898 * below handles both the chapters & no chapters case.
2900 if ( buf->id == stream->ffmpeg_video_id && buf->start >= stream->chapter_end )
2902 hb_chapter_t *chapter = hb_list_item( stream->title->list_chapter,
2903 ++stream->chapter );
2906 stream->chapter_end += chapter->duration;
2907 buf->new_chap = stream->chapter + 1;
2908 hb_deep_log( 2, "ffmpeg_read starting chapter %i at %"PRId64,
2909 buf->new_chap, buf->start);
2911 // Must have run out of chapters, stop looking.
2912 stream->chapter_end = INT64_MAX;
2917 av_free_packet( stream->ffmpeg_pkt );
2921 static int ffmpeg_seek( hb_stream_t *stream, float frac )
2923 AVFormatContext *ic = stream->ffmpeg_ic;
2926 int64_t pos = (double)ic->duration * (double)frac;
2927 if ( ic->start_time != AV_NOPTS_VALUE && ic->start_time > 0 )
2929 pos += ic->start_time;
2931 av_seek_frame( ic, -1, pos, 0 );
2932 stream->need_keyframe = 1;
2936 av_seek_frame( ic, -1, 0LL, AVSEEK_FLAG_BACKWARD );