2 * General DV muxer/demuxer
3 * Copyright (c) 2003 Roman Shaposhnik
5 * Many thanks to Dan Dennedy <dan@dennedy.org> for providing wealth
6 * of DV technical info.
9 * Copyright (c) 2002 Fabrice Bellard
11 * 50 Mbps (DVCPRO50) and 100 Mbps (DVCPRO HD) support
12 * Copyright (c) 2006 Daniel Maas <dmaas@maasdigital.com>
13 * Funded by BBC Research & Development
15 * This file is part of Libav.
17 * Libav is free software; you can redistribute it and/or
18 * modify it under the terms of the GNU Lesser General Public
19 * License as published by the Free Software Foundation; either
20 * version 2.1 of the License, or (at your option) any later version.
22 * Libav is distributed in the hope that it will be useful,
23 * but WITHOUT ANY WARRANTY; without even the implied warranty of
24 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
25 * Lesser General Public License for more details.
27 * You should have received a copy of the GNU Lesser General Public
28 * License along with Libav; if not, write to the Free Software
29 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
34 #include "libavcodec/dvdata.h"
35 #include "libavutil/intreadwrite.h"
36 #include "libavutil/mathematics.h"
39 struct DVDemuxContext {
40 const DVprofile* sys; /* Current DV profile. E.g.: 525/60, 625/50 */
41 AVFormatContext* fctx;
44 AVPacket audio_pkt[4];
45 uint8_t audio_buf[4][8192];
51 static inline uint16_t dv_audio_12to16(uint16_t sample)
53 uint16_t shift, result;
55 sample = (sample < 0x800) ? sample : sample | 0xf000;
56 shift = (sample & 0xf00) >> 8;
58 if (shift < 0x2 || shift > 0xd) {
60 } else if (shift < 0x8) {
62 result = (sample - (256 * shift)) << shift;
65 result = ((sample + ((256 * shift) + 1)) << shift) - 1;
72 * This is the dumbest implementation of all -- it simply looks at
73 * a fixed offset and if pack isn't there -- fails. We might want
74 * to have a fallback mechanism for complete search of missing packs.
76 static const uint8_t* dv_extract_pack(uint8_t* frame, enum dv_pack_type t)
82 offs = (80*6 + 80*16*3 + 3);
84 case dv_audio_control:
85 offs = (80*6 + 80*16*4 + 3);
87 case dv_video_control:
88 offs = (80*5 + 48 + 5);
94 return frame[offs] == t ? &frame[offs] : NULL;
98 * There's a couple of assumptions being made here:
99 * 1. By default we silence erroneous (0x8000/16bit 0x800/12bit) audio samples.
100 * We can pass them upwards when libavcodec will be ready to deal with them.
101 * 2. We don't do software emphasis.
102 * 3. Audio is always returned as 16bit linear samples: 12bit nonlinear samples
103 * are converted into 16bit linear ones.
105 static int dv_extract_audio(uint8_t* frame, uint8_t* ppcm[4],
106 const DVprofile *sys)
108 int size, chan, i, j, d, of, smpls, freq, quant, half_ch;
110 const uint8_t* as_pack;
113 as_pack = dv_extract_pack(frame, dv_audio_source);
114 if (!as_pack) /* No audio ? */
117 smpls = as_pack[1] & 0x3f; /* samples in this frame - min. samples */
118 freq = (as_pack[4] >> 3) & 0x07; /* 0 - 48kHz, 1 - 44,1kHz, 2 - 32kHz */
119 quant = as_pack[4] & 0x07; /* 0 - 16bit linear, 1 - 12bit nonlinear */
122 return -1; /* unsupported quantization */
124 size = (sys->audio_min_samples[freq] + smpls) * 4; /* 2ch, 2bytes */
125 half_ch = sys->difseg_size / 2;
127 /* We work with 720p frames split in half, thus even frames have
128 * channels 0,1 and odd 2,3. */
129 ipcm = (sys->height == 720 && !(frame[1] & 0x0C)) ? 2 : 0;
131 /* for each DIF channel */
132 for (chan = 0; chan < sys->n_difchan; chan++) {
133 /* next stereo channel (50Mbps and 100Mbps only) */
138 /* for each DIF segment */
139 for (i = 0; i < sys->difseg_size; i++) {
140 frame += 6 * 80; /* skip DIF segment header */
141 if (quant == 1 && i == half_ch) {
142 /* next stereo channel (12bit mode only) */
148 /* for each AV sequence */
149 for (j = 0; j < 9; j++) {
150 for (d = 8; d < 80; d += 2) {
151 if (quant == 0) { /* 16bit quantization */
152 of = sys->audio_shuffle[i][j] + (d - 8) / 2 * sys->audio_stride;
156 pcm[of*2] = frame[d+1]; // FIXME: maybe we have to admit
157 pcm[of*2+1] = frame[d]; // that DV is a big-endian PCM
158 if (pcm[of*2+1] == 0x80 && pcm[of*2] == 0x00)
160 } else { /* 12bit quantization */
161 lc = ((uint16_t)frame[d] << 4) |
162 ((uint16_t)frame[d+2] >> 4);
163 rc = ((uint16_t)frame[d+1] << 4) |
164 ((uint16_t)frame[d+2] & 0x0f);
165 lc = (lc == 0x800 ? 0 : dv_audio_12to16(lc));
166 rc = (rc == 0x800 ? 0 : dv_audio_12to16(rc));
168 of = sys->audio_shuffle[i%half_ch][j] + (d - 8) / 3 * sys->audio_stride;
172 pcm[of*2] = lc & 0xff; // FIXME: maybe we have to admit
173 pcm[of*2+1] = lc >> 8; // that DV is a big-endian PCM
174 of = sys->audio_shuffle[i%half_ch+half_ch][j] +
175 (d - 8) / 3 * sys->audio_stride;
176 pcm[of*2] = rc & 0xff; // FIXME: maybe we have to admit
177 pcm[of*2+1] = rc >> 8; // that DV is a big-endian PCM
182 frame += 16 * 80; /* 15 Video DIFs + 1 Audio DIF */
190 static int dv_extract_audio_info(DVDemuxContext* c, uint8_t* frame)
192 const uint8_t* as_pack;
193 int freq, stype, smpls, quant, i, ach;
195 as_pack = dv_extract_pack(frame, dv_audio_source);
196 if (!as_pack || !c->sys) { /* No audio ? */
201 smpls = as_pack[1] & 0x3f; /* samples in this frame - min. samples */
202 freq = (as_pack[4] >> 3) & 0x07; /* 0 - 48kHz, 1 - 44,1kHz, 2 - 32kHz */
203 stype = (as_pack[3] & 0x1f); /* 0 - 2CH, 2 - 4CH, 3 - 8CH */
204 quant = as_pack[4] & 0x07; /* 0 - 16bit linear, 1 - 12bit nonlinear */
207 av_log(c->fctx, AV_LOG_ERROR, "stype %d is invalid\n", stype);
212 /* note: ach counts PAIRS of channels (i.e. stereo channels) */
213 ach = ((int[4]){ 1, 0, 2, 4})[stype];
214 if (ach == 1 && quant && freq == 2)
217 /* Dynamic handling of the audio streams in DV */
218 for (i = 0; i < ach; i++) {
220 c->ast[i] = avformat_new_stream(c->fctx, NULL);
223 avpriv_set_pts_info(c->ast[i], 64, 1, 30000);
224 c->ast[i]->codec->codec_type = AVMEDIA_TYPE_AUDIO;
225 c->ast[i]->codec->codec_id = CODEC_ID_PCM_S16LE;
227 av_init_packet(&c->audio_pkt[i]);
228 c->audio_pkt[i].size = 0;
229 c->audio_pkt[i].data = c->audio_buf[i];
230 c->audio_pkt[i].stream_index = c->ast[i]->index;
231 c->audio_pkt[i].flags |= AV_PKT_FLAG_KEY;
233 c->ast[i]->codec->sample_rate = dv_audio_frequency[freq];
234 c->ast[i]->codec->channels = 2;
235 c->ast[i]->codec->bit_rate = 2 * dv_audio_frequency[freq] * 16;
236 c->ast[i]->start_time = 0;
240 return (c->sys->audio_min_samples[freq] + smpls) * 4; /* 2ch, 2bytes */;
243 static int dv_extract_video_info(DVDemuxContext *c, uint8_t* frame)
245 const uint8_t* vsc_pack;
246 AVCodecContext* avctx;
251 avctx = c->vst->codec;
253 avpriv_set_pts_info(c->vst, 64, c->sys->time_base.num,
254 c->sys->time_base.den);
255 avctx->time_base= c->sys->time_base;
257 avctx->width = c->sys->width;
258 avctx->height = c->sys->height;
260 avctx->pix_fmt = c->sys->pix_fmt;
262 /* finding out SAR is a little bit messy */
263 vsc_pack = dv_extract_pack(frame, dv_video_control);
264 apt = frame[4] & 0x07;
265 is16_9 = (vsc_pack && ((vsc_pack[2] & 0x07) == 0x02 ||
266 (!apt && (vsc_pack[2] & 0x07) == 0x07)));
267 c->vst->sample_aspect_ratio = c->sys->sar[is16_9];
268 avctx->bit_rate = av_rescale_q(c->sys->frame_size, (AVRational){8,1},
270 size = c->sys->frame_size;
276 * The following 3 functions constitute our interface to the world
279 DVDemuxContext* avpriv_dv_init_demux(AVFormatContext *s)
283 c = av_mallocz(sizeof(DVDemuxContext));
287 c->vst = avformat_new_stream(s, NULL);
295 memset(c->ast, 0, sizeof(c->ast));
300 c->vst->codec->codec_type = AVMEDIA_TYPE_VIDEO;
301 c->vst->codec->codec_id = CODEC_ID_DVVIDEO;
302 c->vst->codec->bit_rate = 25000000;
303 c->vst->start_time = 0;
308 int avpriv_dv_get_packet(DVDemuxContext *c, AVPacket *pkt)
313 for (i = 0; i < c->ach; i++) {
314 if (c->ast[i] && c->audio_pkt[i].size) {
315 *pkt = c->audio_pkt[i];
316 c->audio_pkt[i].size = 0;
325 int avpriv_dv_produce_packet(DVDemuxContext *c, AVPacket *pkt,
326 uint8_t* buf, int buf_size)
329 uint8_t *ppcm[4] = {0};
331 if (buf_size < DV_PROFILE_BYTES ||
332 !(c->sys = avpriv_dv_frame_profile(c->sys, buf, buf_size)) ||
333 buf_size < c->sys->frame_size) {
334 return -1; /* Broken frame, or not enough data */
337 /* Queueing audio packet */
338 /* FIXME: in case of no audio/bad audio we have to do something */
339 size = dv_extract_audio_info(c, buf);
340 for (i = 0; i < c->ach; i++) {
341 c->audio_pkt[i].size = size;
342 c->audio_pkt[i].pts = c->abytes * 30000*8 / c->ast[i]->codec->bit_rate;
343 ppcm[i] = c->audio_buf[i];
346 dv_extract_audio(buf, ppcm, c->sys);
348 /* We work with 720p frames split in half, thus even frames have
349 * channels 0,1 and odd 2,3. */
350 if (c->sys->height == 720) {
352 c->audio_pkt[2].size = c->audio_pkt[3].size = 0;
354 c->audio_pkt[0].size = c->audio_pkt[1].size = 0;
361 /* Now it's time to return video packet */
362 size = dv_extract_video_info(c, buf);
366 pkt->flags |= AV_PKT_FLAG_KEY;
367 pkt->stream_index = c->vst->id;
368 pkt->pts = c->frames;
375 static int64_t dv_frame_offset(AVFormatContext *s, DVDemuxContext *c,
376 int64_t timestamp, int flags)
378 // FIXME: sys may be wrong if last dv_read_packet() failed (buffer is junk)
379 const DVprofile* sys = avpriv_dv_codec_profile(c->vst->codec);
381 int64_t size = avio_size(s->pb) - s->data_offset;
382 int64_t max_offset = ((size-1) / sys->frame_size) * sys->frame_size;
384 offset = sys->frame_size * timestamp;
386 if (size >= 0 && offset > max_offset) offset = max_offset;
387 else if (offset < 0) offset = 0;
389 return offset + s->data_offset;
392 void dv_offset_reset(DVDemuxContext *c, int64_t frame_offset)
394 c->frames= frame_offset;
396 c->abytes= av_rescale_q(c->frames, c->sys->time_base,
397 (AVRational){8, c->ast[0]->codec->bit_rate});
398 c->audio_pkt[0].size = c->audio_pkt[1].size = 0;
399 c->audio_pkt[2].size = c->audio_pkt[3].size = 0;
402 /************************************************************
403 * Implementation of the easiest DV storage of all -- raw DV.
404 ************************************************************/
406 typedef struct RawDVContext {
407 DVDemuxContext* dv_demux;
408 uint8_t buf[DV_MAX_FRAME_SIZE];
411 static int dv_read_header(AVFormatContext *s)
413 unsigned state, marker_pos = 0;
414 RawDVContext *c = s->priv_data;
416 c->dv_demux = avpriv_dv_init_demux(s);
420 state = avio_rb32(s->pb);
421 while ((state & 0xffffff7f) != 0x1f07003f) {
422 if (s->pb->eof_reached) {
423 av_log(s, AV_LOG_ERROR, "Cannot find DV header.\n");
426 if (state == 0x003f0700 || state == 0xff3f0700)
427 marker_pos = avio_tell(s->pb);
428 if (state == 0xff3f0701 && avio_tell(s->pb) - marker_pos == 80) {
429 avio_seek(s->pb, -163, SEEK_CUR);
430 state = avio_rb32(s->pb);
433 state = (state << 8) | avio_r8(s->pb);
435 AV_WB32(c->buf, state);
437 if (avio_read(s->pb, c->buf + 4, DV_PROFILE_BYTES - 4) <= 0 ||
438 avio_seek(s->pb, -DV_PROFILE_BYTES, SEEK_CUR) < 0)
441 c->dv_demux->sys = avpriv_dv_frame_profile(c->dv_demux->sys, c->buf, DV_PROFILE_BYTES);
442 if (!c->dv_demux->sys) {
443 av_log(s, AV_LOG_ERROR, "Can't determine profile of DV input stream.\n");
447 s->bit_rate = av_rescale_q(c->dv_demux->sys->frame_size, (AVRational){8,1},
448 c->dv_demux->sys->time_base);
454 static int dv_read_packet(AVFormatContext *s, AVPacket *pkt)
457 RawDVContext *c = s->priv_data;
459 size = avpriv_dv_get_packet(c->dv_demux, pkt);
462 if (!c->dv_demux->sys)
464 size = c->dv_demux->sys->frame_size;
465 if (avio_read(s->pb, c->buf, size) <= 0)
468 size = avpriv_dv_produce_packet(c->dv_demux, pkt, c->buf, size);
474 static int dv_read_seek(AVFormatContext *s, int stream_index,
475 int64_t timestamp, int flags)
477 RawDVContext *r = s->priv_data;
478 DVDemuxContext *c = r->dv_demux;
479 int64_t offset = dv_frame_offset(s, c, timestamp, flags);
481 dv_offset_reset(c, offset / c->sys->frame_size);
483 offset = avio_seek(s->pb, offset, SEEK_SET);
484 return (offset < 0) ? offset : 0;
487 static int dv_read_close(AVFormatContext *s)
489 RawDVContext *c = s->priv_data;
490 av_free(c->dv_demux);
494 static int dv_probe(AVProbeData *p)
496 unsigned state, marker_pos = 0;
499 int secondary_matches = 0;
504 state = AV_RB32(p->buf);
505 for (i = 4; i < p->buf_size; i++) {
506 if ((state & 0xffffff7f) == 0x1f07003f)
508 // any section header, also with seq/chan num != 0,
509 // should appear around every 12000 bytes, at least 10 per frame
510 if ((state & 0xff07ff7f) == 0x1f07003f)
512 if (state == 0x003f0700 || state == 0xff3f0700)
514 if (state == 0xff3f0701 && i - marker_pos == 80)
516 state = (state << 8) | p->buf[i];
519 if (matches && p->buf_size / matches < 1024*1024) {
520 if (matches > 4 || (secondary_matches >= 10 && p->buf_size / secondary_matches < 24000))
521 return AVPROBE_SCORE_MAX*3/4; // not max to avoid dv in mov to match
522 return AVPROBE_SCORE_MAX/4;
527 #if CONFIG_DV_DEMUXER
528 AVInputFormat ff_dv_demuxer = {
530 .long_name = NULL_IF_CONFIG_SMALL("DV video format"),
531 .priv_data_size = sizeof(RawDVContext),
532 .read_probe = dv_probe,
533 .read_header = dv_read_header,
534 .read_packet = dv_read_packet,
535 .read_close = dv_read_close,
536 .read_seek = dv_read_seek,
537 .extensions = "dv,dif",