ffmpeg/libavformat/movenchint.c

472 lines
16 KiB
C
Raw Normal View History

/*
* MOV, 3GP, MP4 muxer RTP hinting
* Copyright (c) 2010 Martin Storsjo
*
* This file is part of Libav.
*
* Libav is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* Libav is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with Libav; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "movenc.h"
#include "libavutil/intreadwrite.h"
#include "internal.h"
#include "rtpenc_chain.h"
#include "avio_internal.h"
#include "rtp.h"
int ff_mov_init_hinting(AVFormatContext *s, int index, int src_index)
{
MOVMuxContext *mov = s->priv_data;
MOVTrack *track = &mov->tracks[index];
MOVTrack *src_track = &mov->tracks[src_index];
AVStream *src_st = s->streams[src_index];
int ret = AVERROR(ENOMEM);
track->tag = MKTAG('r','t','p',' ');
track->src_track = src_index;
lavf: replace AVStream.codec with AVStream.codecpar Currently, AVStream contains an embedded AVCodecContext instance, which is used by demuxers to export stream parameters to the caller and by muxers to receive stream parameters from the caller. It is also used internally as the codec context that is passed to parsers. In addition, it is also widely used by the callers as the decoding (when demuxer) or encoding (when muxing) context, though this has been officially discouraged since Libav 11. There are multiple important problems with this approach: - the fields in AVCodecContext are in general one of * stream parameters * codec options * codec state However, it's not clear which ones are which. It is consequently unclear which fields are a demuxer allowed to set or a muxer allowed to read. This leads to erratic behaviour depending on whether decoding or encoding is being performed or not (and whether it uses the AVStream embedded codec context). - various synchronization issues arising from the fact that the same context is used by several different APIs (muxers/demuxers, parsers, bitstream filters and encoders/decoders) simultaneously, with there being no clear rules for who can modify what and the different processes being typically delayed with respect to each other. - avformat_find_stream_info() making it necessary to support opening and closing a single codec context multiple times, thus complicating the semantics of freeing various allocated objects in the codec context. Those problems are resolved by replacing the AVStream embedded codec context with a newly added AVCodecParameters instance, which stores only the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 18:42:52 +00:00
track->par = avcodec_parameters_alloc();
if (!track->par)
goto fail;
lavf: replace AVStream.codec with AVStream.codecpar Currently, AVStream contains an embedded AVCodecContext instance, which is used by demuxers to export stream parameters to the caller and by muxers to receive stream parameters from the caller. It is also used internally as the codec context that is passed to parsers. In addition, it is also widely used by the callers as the decoding (when demuxer) or encoding (when muxing) context, though this has been officially discouraged since Libav 11. There are multiple important problems with this approach: - the fields in AVCodecContext are in general one of * stream parameters * codec options * codec state However, it's not clear which ones are which. It is consequently unclear which fields are a demuxer allowed to set or a muxer allowed to read. This leads to erratic behaviour depending on whether decoding or encoding is being performed or not (and whether it uses the AVStream embedded codec context). - various synchronization issues arising from the fact that the same context is used by several different APIs (muxers/demuxers, parsers, bitstream filters and encoders/decoders) simultaneously, with there being no clear rules for who can modify what and the different processes being typically delayed with respect to each other. - avformat_find_stream_info() making it necessary to support opening and closing a single codec context multiple times, thus complicating the semantics of freeing various allocated objects in the codec context. Those problems are resolved by replacing the AVStream embedded codec context with a newly added AVCodecParameters instance, which stores only the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 18:42:52 +00:00
track->par->codec_type = AVMEDIA_TYPE_DATA;
track->par->codec_tag = track->tag;
ret = ff_rtp_chain_mux_open(&track->rtp_ctx, s, src_st, NULL,
RTP_MAX_PACKET_SIZE, src_index);
if (ret < 0)
goto fail;
/* Copy the RTP AVStream timebase back to the hint AVStream */
track->timescale = track->rtp_ctx->streams[0]->time_base.den;
/* Mark the hinted track that packets written to it should be
* sent to this track for hinting. */
src_track->hint_track = index;
return 0;
fail:
av_log(s, AV_LOG_WARNING,
"Unable to initialize hinting of stream %d\n", src_index);
lavf: replace AVStream.codec with AVStream.codecpar Currently, AVStream contains an embedded AVCodecContext instance, which is used by demuxers to export stream parameters to the caller and by muxers to receive stream parameters from the caller. It is also used internally as the codec context that is passed to parsers. In addition, it is also widely used by the callers as the decoding (when demuxer) or encoding (when muxing) context, though this has been officially discouraged since Libav 11. There are multiple important problems with this approach: - the fields in AVCodecContext are in general one of * stream parameters * codec options * codec state However, it's not clear which ones are which. It is consequently unclear which fields are a demuxer allowed to set or a muxer allowed to read. This leads to erratic behaviour depending on whether decoding or encoding is being performed or not (and whether it uses the AVStream embedded codec context). - various synchronization issues arising from the fact that the same context is used by several different APIs (muxers/demuxers, parsers, bitstream filters and encoders/decoders) simultaneously, with there being no clear rules for who can modify what and the different processes being typically delayed with respect to each other. - avformat_find_stream_info() making it necessary to support opening and closing a single codec context multiple times, thus complicating the semantics of freeing various allocated objects in the codec context. Those problems are resolved by replacing the AVStream embedded codec context with a newly added AVCodecParameters instance, which stores only the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 18:42:52 +00:00
avcodec_parameters_free(&track->par);
/* Set a default timescale, to avoid crashes in av_dump_format */
track->timescale = 90000;
return ret;
}
/**
* Remove the first sample from the sample queue.
*/
static void sample_queue_pop(HintSampleQueue *queue)
{
if (queue->len <= 0)
return;
if (queue->samples[0].own_data)
av_free(queue->samples[0].data);
queue->len--;
memmove(queue->samples, queue->samples + 1, sizeof(HintSample)*queue->len);
}
/**
* Empty the sample queue, releasing all memory.
*/
static void sample_queue_free(HintSampleQueue *queue)
{
int i;
for (i = 0; i < queue->len; i++)
if (queue->samples[i].own_data)
av_free(queue->samples[i].data);
av_freep(&queue->samples);
queue->len = 0;
queue->size = 0;
}
/**
* Add a reference to the sample data to the sample queue. The data is
* not copied. sample_queue_retain should be called before pkt->data
* is reused/freed.
*/
static void sample_queue_push(HintSampleQueue *queue, uint8_t *data, int size,
int sample)
{
/* No need to keep track of smaller samples, since describing them
* with immediates is more efficient. */
if (size <= 14)
return;
if (!queue->samples || queue->len >= queue->size) {
queue->size += 10;
if (av_reallocp(&queue->samples, sizeof(*queue->samples) * queue->size) < 0) {
queue->len = queue->size = 0;
return;
}
}
queue->samples[queue->len].data = data;
queue->samples[queue->len].size = size;
queue->samples[queue->len].sample_number = sample;
queue->samples[queue->len].offset = 0;
queue->samples[queue->len].own_data = 0;
queue->len++;
}
/**
* Make local copies of all referenced sample data in the queue.
*/
static void sample_queue_retain(HintSampleQueue *queue)
{
int i;
for (i = 0; i < queue->len; ) {
HintSample *sample = &queue->samples[i];
if (!sample->own_data) {
uint8_t *ptr = av_malloc(sample->size);
if (!ptr) {
/* Unable to allocate memory for this one, remove it */
memmove(queue->samples + i, queue->samples + i + 1,
sizeof(HintSample)*(queue->len - i - 1));
queue->len--;
continue;
}
memcpy(ptr, sample->data, sample->size);
sample->data = ptr;
sample->own_data = 1;
}
i++;
}
}
/**
* Find matches of needle[n_pos ->] within haystack. If a sufficiently
* large match is found, matching bytes before n_pos are included
* in the match, too (within the limits of the arrays).
*
* @param haystack buffer that may contain parts of needle
* @param h_len length of the haystack buffer
* @param needle buffer containing source data that have been used to
* construct haystack
* @param n_pos start position in needle used for looking for matches
* @param n_len length of the needle buffer
* @param match_h_offset_ptr offset of the first matching byte within haystack
* @param match_n_offset_ptr offset of the first matching byte within needle
* @param match_len_ptr length of the matched segment
* @return 0 if a match was found, < 0 if no match was found
*/
static int match_segments(const uint8_t *haystack, int h_len,
const uint8_t *needle, int n_pos, int n_len,
int *match_h_offset_ptr, int *match_n_offset_ptr,
int *match_len_ptr)
{
int h_pos;
for (h_pos = 0; h_pos < h_len; h_pos++) {
int match_len = 0;
int match_h_pos, match_n_pos;
/* Check how many bytes match at needle[n_pos] and haystack[h_pos] */
while (h_pos + match_len < h_len && n_pos + match_len < n_len &&
needle[n_pos + match_len] == haystack[h_pos + match_len])
match_len++;
if (match_len <= 8)
continue;
/* If a sufficiently large match was found, try to expand
* the matched segment backwards. */
match_h_pos = h_pos;
match_n_pos = n_pos;
while (match_n_pos > 0 && match_h_pos > 0 &&
needle[match_n_pos - 1] == haystack[match_h_pos - 1]) {
match_n_pos--;
match_h_pos--;
match_len++;
}
if (match_len <= 14)
continue;
*match_h_offset_ptr = match_h_pos;
*match_n_offset_ptr = match_n_pos;
*match_len_ptr = match_len;
return 0;
}
return -1;
}
/**
* Look for segments in samples in the sample queue matching the data
* in ptr. Samples not matching are removed from the queue. If a match
* is found, the next time it will look for matches starting from the
* end of the previous matched segment.
*
* @param data data to find matches for in the sample queue
* @param len length of the data buffer
* @param queue samples used for looking for matching segments
* @param pos the offset in data of the matched segment
* @param match_sample the number of the sample that contained the match
* @param match_offset the offset of the matched segment within the sample
* @param match_len the length of the matched segment
* @return 0 if a match was found, < 0 if no match was found
*/
static int find_sample_match(const uint8_t *data, int len,
HintSampleQueue *queue, int *pos,
int *match_sample, int *match_offset,
int *match_len)
{
while (queue->len > 0) {
HintSample *sample = &queue->samples[0];
/* If looking for matches in a new sample, skip the first 5 bytes,
* since they often may be modified/removed in the output packet. */
if (sample->offset == 0 && sample->size > 5)
sample->offset = 5;
if (match_segments(data, len, sample->data, sample->offset,
sample->size, pos, match_offset, match_len) == 0) {
*match_sample = sample->sample_number;
/* Next time, look for matches at this offset, with a little
* margin to this match. */
sample->offset = *match_offset + *match_len + 5;
if (sample->offset + 10 >= sample->size)
sample_queue_pop(queue); /* Not enough useful data left */
return 0;
}
if (sample->offset < 10 && sample->size > 20) {
/* No match found from the start of the sample,
* try from the middle of the sample instead. */
sample->offset = sample->size/2;
} else {
/* No match for this sample, remove it */
sample_queue_pop(queue);
}
}
return -1;
}
static void output_immediate(const uint8_t *data, int size,
AVIOContext *out, int *entries)
{
while (size > 0) {
int len = size;
if (len > 14)
len = 14;
avio_w8(out, 1); /* immediate constructor */
avio_w8(out, len); /* amount of valid data */
avio_write(out, data, len);
data += len;
size -= len;
for (; len < 14; len++)
avio_w8(out, 0);
(*entries)++;
}
}
static void output_match(AVIOContext *out, int match_sample,
int match_offset, int match_len, int *entries)
{
avio_w8(out, 2); /* sample constructor */
avio_w8(out, 0); /* track reference */
avio_wb16(out, match_len);
avio_wb32(out, match_sample);
avio_wb32(out, match_offset);
avio_wb16(out, 1); /* bytes per block */
avio_wb16(out, 1); /* samples per block */
(*entries)++;
}
static void describe_payload(const uint8_t *data, int size,
AVIOContext *out, int *entries,
HintSampleQueue *queue)
{
/* Describe the payload using different constructors */
while (size > 0) {
int match_sample, match_offset, match_len, pos;
if (find_sample_match(data, size, queue, &pos, &match_sample,
&match_offset, &match_len) < 0)
break;
output_immediate(data, pos, out, entries);
data += pos;
size -= pos;
output_match(out, match_sample, match_offset, match_len, entries);
data += match_len;
size -= match_len;
}
output_immediate(data, size, out, entries);
}
/**
* Write an RTP hint (that may contain one or more RTP packets)
* for the packets in data. data contains one or more packets with a
* BE32 size header.
*
* @param out buffer where the hints are written
* @param data buffer containing RTP packets
* @param size the size of the data buffer
* @param trk the MOVTrack for the hint track
* @param dts pointer where the timestamp for the written RTP hint is stored
* @return the number of RTP packets in the written hint
*/
static int write_hint_packets(AVIOContext *out, const uint8_t *data,
int size, MOVTrack *trk, int64_t *dts)
{
int64_t curpos;
int64_t count_pos, entries_pos;
int count = 0, entries;
count_pos = avio_tell(out);
/* RTPsample header */
avio_wb16(out, 0); /* packet count */
avio_wb16(out, 0); /* reserved */
while (size > 4) {
uint32_t packet_len = AV_RB32(data);
uint16_t seq;
uint32_t ts;
int32_t ts_diff;
data += 4;
size -= 4;
if (packet_len > size || packet_len <= 12)
break;
if (RTP_PT_IS_RTCP(data[1])) {
/* RTCP packet, just skip */
data += packet_len;
size -= packet_len;
continue;
}
if (packet_len > trk->max_packet_size)
trk->max_packet_size = packet_len;
seq = AV_RB16(&data[2]);
ts = AV_RB32(&data[4]);
if (trk->prev_rtp_ts == 0)
trk->prev_rtp_ts = ts;
/* Unwrap the 32-bit RTP timestamp that wraps around often
* into a not (as often) wrapping 64-bit timestamp. */
ts_diff = ts - trk->prev_rtp_ts;
if (ts_diff > 0) {
trk->cur_rtp_ts_unwrapped += ts_diff;
trk->prev_rtp_ts = ts;
ts_diff = 0;
}
if (*dts == AV_NOPTS_VALUE)
*dts = trk->cur_rtp_ts_unwrapped;
count++;
/* RTPpacket header */
avio_wb32(out, 0); /* relative_time */
avio_write(out, data, 2); /* RTP header */
avio_wb16(out, seq); /* RTPsequenceseed */
avio_wb16(out, ts_diff ? 4 : 0); /* reserved + flags (extra_flag) */
entries_pos = avio_tell(out);
avio_wb16(out, 0); /* entry count */
if (ts_diff) { /* if extra_flag is set */
avio_wb32(out, 16); /* extra_information_length */
avio_wb32(out, 12); /* rtpoffsetTLV box */
avio_write(out, "rtpo", 4);
avio_wb32(out, ts_diff);
}
data += 12;
size -= 12;
packet_len -= 12;
entries = 0;
/* Write one or more constructors describing the payload data */
describe_payload(data, packet_len, out, &entries, &trk->sample_queue);
data += packet_len;
size -= packet_len;
curpos = avio_tell(out);
avio_seek(out, entries_pos, SEEK_SET);
avio_wb16(out, entries);
avio_seek(out, curpos, SEEK_SET);
}
curpos = avio_tell(out);
avio_seek(out, count_pos, SEEK_SET);
avio_wb16(out, count);
avio_seek(out, curpos, SEEK_SET);
return count;
}
int ff_mov_add_hinted_packet(AVFormatContext *s, AVPacket *pkt,
int track_index, int sample,
uint8_t *sample_data, int sample_size)
{
MOVMuxContext *mov = s->priv_data;
MOVTrack *trk = &mov->tracks[track_index];
AVFormatContext *rtp_ctx = trk->rtp_ctx;
uint8_t *buf = NULL;
int size;
AVIOContext *hintbuf = NULL;
AVPacket hint_pkt;
int ret = 0, count;
if (!rtp_ctx)
return AVERROR(ENOENT);
if (!rtp_ctx->pb)
return AVERROR(ENOMEM);
if (sample_data)
sample_queue_push(&trk->sample_queue, sample_data, sample_size, sample);
else
sample_queue_push(&trk->sample_queue, pkt->data, pkt->size, sample);
/* Feed the packet to the RTP muxer */
ff_write_chained(rtp_ctx, 0, pkt, s);
/* Fetch the output from the RTP muxer, open a new output buffer
* for next time. */
size = avio_close_dyn_buf(rtp_ctx->pb, &buf);
if ((ret = ffio_open_dyn_packet_buf(&rtp_ctx->pb,
RTP_MAX_PACKET_SIZE)) < 0)
goto done;
if (size <= 0)
goto done;
/* Open a buffer for writing the hint */
if ((ret = avio_open_dyn_buf(&hintbuf)) < 0)
goto done;
av_init_packet(&hint_pkt);
count = write_hint_packets(hintbuf, buf, size, trk, &hint_pkt.dts);
av_freep(&buf);
/* Write the hint data into the hint track */
hint_pkt.size = size = avio_close_dyn_buf(hintbuf, &buf);
hint_pkt.data = buf;
hint_pkt.pts = hint_pkt.dts;
hint_pkt.stream_index = track_index;
if (pkt->flags & AV_PKT_FLAG_KEY)
hint_pkt.flags |= AV_PKT_FLAG_KEY;
if (count > 0)
ff_mov_write_packet(s, &hint_pkt);
done:
av_free(buf);
sample_queue_retain(&trk->sample_queue);
return ret;
}
void ff_mov_close_hinting(MOVTrack *track)
{
AVFormatContext *rtp_ctx = track->rtp_ctx;
lavf: replace AVStream.codec with AVStream.codecpar Currently, AVStream contains an embedded AVCodecContext instance, which is used by demuxers to export stream parameters to the caller and by muxers to receive stream parameters from the caller. It is also used internally as the codec context that is passed to parsers. In addition, it is also widely used by the callers as the decoding (when demuxer) or encoding (when muxing) context, though this has been officially discouraged since Libav 11. There are multiple important problems with this approach: - the fields in AVCodecContext are in general one of * stream parameters * codec options * codec state However, it's not clear which ones are which. It is consequently unclear which fields are a demuxer allowed to set or a muxer allowed to read. This leads to erratic behaviour depending on whether decoding or encoding is being performed or not (and whether it uses the AVStream embedded codec context). - various synchronization issues arising from the fact that the same context is used by several different APIs (muxers/demuxers, parsers, bitstream filters and encoders/decoders) simultaneously, with there being no clear rules for who can modify what and the different processes being typically delayed with respect to each other. - avformat_find_stream_info() making it necessary to support opening and closing a single codec context multiple times, thus complicating the semantics of freeing various allocated objects in the codec context. Those problems are resolved by replacing the AVStream embedded codec context with a newly added AVCodecParameters instance, which stores only the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 18:42:52 +00:00
avcodec_parameters_free(&track->par);
sample_queue_free(&track->sample_queue);
if (!rtp_ctx)
return;
if (rtp_ctx->pb) {
av_write_trailer(rtp_ctx);
ffio_free_dyn_buf(&rtp_ctx->pb);
}
avformat_free_context(rtp_ctx);
}