You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

2861 lines
103 KiB

/*
* Matroska muxer
* Copyright (c) 2007 David Conrad
*
* This file is part of FFmpeg.
*
* FFmpeg 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.
*
* FFmpeg 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 FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include <stdint.h>
#include "av1.h"
#include "avc.h"
#include "hevc.h"
#include "avformat.h"
#include "avio_internal.h"
#include "avlanguage.h"
#include "flacenc.h"
#include "internal.h"
#include "isom.h"
#include "matroska.h"
#include "riff.h"
#include "vorbiscomment.h"
#include "wv.h"
#include "libavutil/avstring.h"
#include "libavutil/channel_layout.h"
#include "libavutil/crc.h"
#include "libavutil/dict.h"
#include "libavutil/intfloat.h"
#include "libavutil/intreadwrite.h"
#include "libavutil/lfg.h"
#include "libavutil/mastering_display_metadata.h"
#include "libavutil/mathematics.h"
#include "libavutil/opt.h"
avformat/matroskaenc: Write duration early during mkv_write_header (Rev #3) Rev #2: Fixes doubled header writing, checked FATE running without errors Rev #3: Fixed coding style This commit addresses the following scenario: we are using ffmpeg to transcode or remux mkv (or something else) to mkv. The result is being streamed on-the-fly to an HTML5 client (streaming starts while ffmpeg is still running). The problem here is that the client is unable to detect the duration because the duration is only written to the mkv at the end of the transcoding/remoxing process. In matroskaenc.c, the duration is only written during mkv_write_trailer but not during mkv_write_header. The approach: FFMPEG is currently putting quite some effort to estimate the durations of source streams, but in many cases the source stream durations are still left at 0 and these durations are nowhere mapped to or used for output streams. As much as I would have liked to deduct or estimate output durations based on input stream durations - I realized that this is a hard task (as Nicolas already mentioned in a previous conversation). It would involve changes to the duration calculation/estimation/deduction for input streams and propagating these durations to output streams or the output context in a correct way. So I looked for a simple and small solution with better chances to get accepted. In webmdashenc.c I found that a duration is written during write_header and this duration is taken from the streams' metadata, so I decided for a similar approach. And here's what it does: At first it is checking the duration of the AVFormatContext. In typical cases this value is not set, but: It is set in cases where the user has specified a recording_time or an end_time via the -t or -to parameters. Then it is looking for a DURATION metadata field in the metadata of the output context (AVFormatContext::metadata). This would only exist in case the user has explicitly specified a metadata DURATION value from the command line. Then it is iterating all streams looking for a "DURATION" metadata (this works unless the option "-map_metadata -1" has been specified) and determines the maximum value. The precendence is as follows: 1. Use duration of AVFormatContext - 2. Use explicitly specified metadata duration value - 3. Use maximum (mapped) metadata duration over all streams. To test this: 1. With explicit recording time: ffmpeg -i file:"src.mkv" -loglevel debug -t 01:38:36.000 -y "dest.mkv" 2. Take duration from metadata specified via command line parameters: ffmpeg -i file:"src.mkv" -loglevel debug -map_metadata -1 -metadata Duration="01:14:33.00" -y "dest.mkv" 3. Take duration from mapped input metadata: ffmpeg -i file:"src.mkv" -loglevel debug -y "dest.mkv" Regression risk: Very low IMO because it only affects the header while ffmpeg is still running. When ffmpeg completes the process, the duration is rewritten to the header with the usual value (same like without this commit). Signed-off-by: SoftWorkz <softworkz@hotmail.com> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
9 years ago
#include "libavutil/parseutils.h"
#include "libavutil/random_seed.h"
#include "libavutil/rational.h"
#include "libavutil/samplefmt.h"
#include "libavutil/stereo3d.h"
#include "libavcodec/xiph.h"
#include "libavcodec/mpeg4audio.h"
avformat/matroskaenc: Avoid allocations for SeekHead Up until e7ddafd5, the Matroska muxer wrote two SeekHeads: One at the beginning referencing the main level 1 elements (i.e. not the Clusters) and one at the end, referencing the Clusters. This second SeekHead was useless and has therefore been removed. Yet the SeekHead-related functions and structures are still geared towards this usecase: They are built around an allocated array of variable size that gets reallocated every time an element is added to it although the maximum number of Seek entries is a small compile-time constant, so that one should rather include the array in the SeekHead structure itself; and said structure should be contained in the MatroskaMuxContext instead of being allocated separately. The earlier code reserved space for a SeekHead with 10 entries, although we currently write at most 6. Reducing said number implied that every Matroska/Webm file will be 84 bytes smaller and required to adapt several FATE tests; furthermore, the reserved amount overestimated the amount needed for for the SeekHead's length field and how many bytes need to be reserved to write a EBML Void element, bringing the total reduction to 89 bytes. This also fixes a potential segfault: If !mkv->is_live and if the AVIOContext is initially unseekable when writing the header, the SeekHead is already written when writing the header and this used to free the SeekHead-related structures that have been allocated. But if the AVIOContext happens to be seekable when writing the trailer, it will be attempted to write the SeekHead again which will lead to segfaults because the corresponding structures have already been freed. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
/* Level 1 elements we create a SeekHead entry for:
* Info, Tracks, Chapters, Attachments, Tags and Cues */
#define MAX_SEEKHEAD_ENTRIES 6
enum {
DEFAULT_MODE_INFER,
DEFAULT_MODE_INFER_NO_SUBS,
DEFAULT_MODE_PASSTHROUGH,
};
typedef struct ebml_master {
int64_t pos; ///< absolute offset in the containing AVIOContext where the master's elements start
int sizebytes; ///< how many bytes were reserved for the size
} ebml_master;
typedef struct mkv_seekhead_entry {
uint32_t elementid;
uint64_t segmentpos;
} mkv_seekhead_entry;
typedef struct mkv_seekhead {
int64_t filepos;
avformat/matroskaenc: Avoid allocations for SeekHead Up until e7ddafd5, the Matroska muxer wrote two SeekHeads: One at the beginning referencing the main level 1 elements (i.e. not the Clusters) and one at the end, referencing the Clusters. This second SeekHead was useless and has therefore been removed. Yet the SeekHead-related functions and structures are still geared towards this usecase: They are built around an allocated array of variable size that gets reallocated every time an element is added to it although the maximum number of Seek entries is a small compile-time constant, so that one should rather include the array in the SeekHead structure itself; and said structure should be contained in the MatroskaMuxContext instead of being allocated separately. The earlier code reserved space for a SeekHead with 10 entries, although we currently write at most 6. Reducing said number implied that every Matroska/Webm file will be 84 bytes smaller and required to adapt several FATE tests; furthermore, the reserved amount overestimated the amount needed for for the SeekHead's length field and how many bytes need to be reserved to write a EBML Void element, bringing the total reduction to 89 bytes. This also fixes a potential segfault: If !mkv->is_live and if the AVIOContext is initially unseekable when writing the header, the SeekHead is already written when writing the header and this used to free the SeekHead-related structures that have been allocated. But if the AVIOContext happens to be seekable when writing the trailer, it will be attempted to write the SeekHead again which will lead to segfaults because the corresponding structures have already been freed. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
mkv_seekhead_entry entries[MAX_SEEKHEAD_ENTRIES];
int num_entries;
avformat/matroskaenc: Avoid allocations for SeekHead Up until e7ddafd5, the Matroska muxer wrote two SeekHeads: One at the beginning referencing the main level 1 elements (i.e. not the Clusters) and one at the end, referencing the Clusters. This second SeekHead was useless and has therefore been removed. Yet the SeekHead-related functions and structures are still geared towards this usecase: They are built around an allocated array of variable size that gets reallocated every time an element is added to it although the maximum number of Seek entries is a small compile-time constant, so that one should rather include the array in the SeekHead structure itself; and said structure should be contained in the MatroskaMuxContext instead of being allocated separately. The earlier code reserved space for a SeekHead with 10 entries, although we currently write at most 6. Reducing said number implied that every Matroska/Webm file will be 84 bytes smaller and required to adapt several FATE tests; furthermore, the reserved amount overestimated the amount needed for for the SeekHead's length field and how many bytes need to be reserved to write a EBML Void element, bringing the total reduction to 89 bytes. This also fixes a potential segfault: If !mkv->is_live and if the AVIOContext is initially unseekable when writing the header, the SeekHead is already written when writing the header and this used to free the SeekHead-related structures that have been allocated. But if the AVIOContext happens to be seekable when writing the trailer, it will be attempted to write the SeekHead again which will lead to segfaults because the corresponding structures have already been freed. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
int reserved_size;
} mkv_seekhead;
typedef struct mkv_cuepoint {
uint64_t pts;
int stream_idx;
int64_t cluster_pos; ///< offset of the cluster containing the block relative to the segment
int64_t relative_pos; ///< relative offset from the position of the cluster containing the block
int64_t duration; ///< duration of the block according to time base
} mkv_cuepoint;
typedef struct mkv_cues {
mkv_cuepoint *entries;
int num_entries;
} mkv_cues;
typedef struct mkv_track {
int write_dts;
int has_cue;
uint64_t uid;
unsigned track_num;
int track_num_size;
int sample_rate;
int64_t sample_rate_offset;
int64_t last_timestamp;
int64_t duration;
int64_t duration_offset;
int64_t codecpriv_offset;
int64_t ts_offset;
} mkv_track;
#define MODE_MATROSKAv2 0x01
#define MODE_WEBM 0x02
typedef struct MatroskaMuxContext {
const AVClass *class;
int mode;
AVIOContext *tags_bc;
int64_t tags_pos;
AVIOContext *info_bc;
int64_t info_pos;
AVIOContext *tracks_bc;
int64_t tracks_pos;
ebml_master segment;
int64_t segment_offset;
AVIOContext *cluster_bc;
int64_t cluster_pos; ///< file offset of the current cluster
int64_t cluster_pts;
int64_t duration_offset;
int64_t duration;
avformat/matroskaenc: Avoid allocations for SeekHead Up until e7ddafd5, the Matroska muxer wrote two SeekHeads: One at the beginning referencing the main level 1 elements (i.e. not the Clusters) and one at the end, referencing the Clusters. This second SeekHead was useless and has therefore been removed. Yet the SeekHead-related functions and structures are still geared towards this usecase: They are built around an allocated array of variable size that gets reallocated every time an element is added to it although the maximum number of Seek entries is a small compile-time constant, so that one should rather include the array in the SeekHead structure itself; and said structure should be contained in the MatroskaMuxContext instead of being allocated separately. The earlier code reserved space for a SeekHead with 10 entries, although we currently write at most 6. Reducing said number implied that every Matroska/Webm file will be 84 bytes smaller and required to adapt several FATE tests; furthermore, the reserved amount overestimated the amount needed for for the SeekHead's length field and how many bytes need to be reserved to write a EBML Void element, bringing the total reduction to 89 bytes. This also fixes a potential segfault: If !mkv->is_live and if the AVIOContext is initially unseekable when writing the header, the SeekHead is already written when writing the header and this used to free the SeekHead-related structures that have been allocated. But if the AVIOContext happens to be seekable when writing the trailer, it will be attempted to write the SeekHead again which will lead to segfaults because the corresponding structures have already been freed. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
mkv_seekhead seekhead;
mkv_cues cues;
mkv_track *tracks;
AVPacket cur_audio_pkt;
unsigned nb_attachments;
int have_video;
int reserve_cues_space;
int cluster_size_limit;
int64_t cues_pos;
int64_t cluster_time_limit;
int is_dash;
int dash_track_number;
int is_live;
int write_crc;
uint32_t chapter_id_offset;
int wrote_chapters;
int allow_raw_vfw;
int default_mode;
uint32_t segment_uid[4];
} MatroskaMuxContext;
/** 2 bytes * 7 for EBML IDs, 7 1-byte EBML lengths, 6 1-byte uint,
* 8 byte for "matroska" doctype string */
#define MAX_EBML_HEADER_SIZE 35
/** 2 bytes * 3 for EBML IDs, 3 1-byte EBML lengths, 8 bytes for 64 bit
* offset, 4 bytes for target EBML ID */
#define MAX_SEEKENTRY_SIZE 21
/** 4 * (1-byte EBML ID, 1-byte EBML size, 8-byte uint max) */
#define MAX_CUETRACKPOS_SIZE 40
/** Seek preroll value for opus */
#define OPUS_SEEK_PREROLL 80000000
static int ebml_id_size(uint32_t id)
{
return (av_log2(id) + 7U) / 8;
}
static void put_ebml_id(AVIOContext *pb, uint32_t id)
{
int i = ebml_id_size(id);
while (i--)
avio_w8(pb, (uint8_t)(id >> (i * 8)));
}
/**
* Write an EBML size meaning "unknown size".
*
* @param bytes The number of bytes the size should occupy (maximum: 8).
*/
static void put_ebml_size_unknown(AVIOContext *pb, int bytes)
{
av_assert0(bytes <= 8);
avio_w8(pb, 0x1ff >> bytes);
ffio_fill(pb, 0xff, bytes - 1);
}
/**
* Returns how many bytes are needed to represent a number
* as EBML variable length integer.
*/
static int ebml_num_size(uint64_t num)
{
int bytes = 0;
do {
bytes++;
} while (num >>= 7);
return bytes;
}
/**
* Calculate how many bytes are needed to represent the length field
* of an EBML element whose payload has a given length.
*/
static int ebml_length_size(uint64_t length)
{
return ebml_num_size(length + 1);
}
/**
* Write a number as EBML variable length integer on `bytes` bytes.
* `bytes` is taken literally without checking.
*/
static void put_ebml_num(AVIOContext *pb, uint64_t num, int bytes)
{
num |= 1ULL << bytes * 7;
for (int i = bytes - 1; i >= 0; i--)
avio_w8(pb, (uint8_t)(num >> i * 8));
}
/**
* Write a length as EBML variable length integer.
*
* @param bytes The number of bytes that need to be used to write the number.
* If zero, the minimal number of bytes will be used.
*/
static void put_ebml_length(AVIOContext *pb, uint64_t length, int bytes)
{
int needed_bytes = ebml_length_size(length);
// sizes larger than this are currently undefined in EBML
av_assert0(length < (1ULL << 56) - 1);
if (bytes == 0)
bytes = needed_bytes;
// The bytes needed to write the given size must not exceed
// the bytes that we ought to use.
av_assert0(bytes >= needed_bytes);
put_ebml_num(pb, length, bytes);
}
/**
* Write a (random) UID with fixed size to make the output more deterministic
*/
static void put_ebml_uid(AVIOContext *pb, uint32_t elementid, uint64_t uid)
{
put_ebml_id(pb, elementid);
put_ebml_length(pb, 8, 0);
avio_wb64(pb, uid);
}
static void put_ebml_uint(AVIOContext *pb, uint32_t elementid, uint64_t val)
{
int i, bytes = 1;
uint64_t tmp = val;
while (tmp >>= 8)
bytes++;
put_ebml_id(pb, elementid);
put_ebml_length(pb, bytes, 0);
for (i = bytes - 1; i >= 0; i--)
avio_w8(pb, (uint8_t)(val >> i * 8));
}
static void put_ebml_sint(AVIOContext *pb, uint32_t elementid, int64_t val)
{
int i, bytes = 1;
uint64_t tmp = 2*(val < 0 ? val^-1 : val);
while (tmp >>= 8)
bytes++;
put_ebml_id(pb, elementid);
put_ebml_length(pb, bytes, 0);
for (i = bytes - 1; i >= 0; i--)
avio_w8(pb, (uint8_t)(val >> i * 8));
}
static void put_ebml_float(AVIOContext *pb, uint32_t elementid, double val)
{
put_ebml_id(pb, elementid);
put_ebml_length(pb, 8, 0);
avio_wb64(pb, av_double2int(val));
}
static void put_ebml_binary(AVIOContext *pb, uint32_t elementid,
const void *buf, int size)
{
put_ebml_id(pb, elementid);
put_ebml_length(pb, size, 0);
avio_write(pb, buf, size);
}
static void put_ebml_string(AVIOContext *pb, uint32_t elementid,
const char *str)
{
put_ebml_binary(pb, elementid, str, strlen(str));
}
/**
* Write a void element of a given size. Useful for reserving space in
* the file to be written to later.
*
* @param size The number of bytes to reserve, which must be at least 2.
*/
static void put_ebml_void(AVIOContext *pb, int size)
{
av_assert0(size >= 2);
put_ebml_id(pb, EBML_ID_VOID);
// we need to subtract the length needed to store the size from the
// size we need to reserve so 2 cases, we use 8 bytes to store the
// size if possible, 1 byte otherwise
if (size < 10) {
size -= 2;
put_ebml_length(pb, size, 0);
} else {
size -= 9;
put_ebml_length(pb, size, 8);
}
ffio_fill(pb, 0, size);
}
static ebml_master start_ebml_master(AVIOContext *pb, uint32_t elementid,
uint64_t expectedsize)
{
int bytes = expectedsize ? ebml_length_size(expectedsize) : 8;
put_ebml_id(pb, elementid);
put_ebml_size_unknown(pb, bytes);
return (ebml_master) { avio_tell(pb), bytes };
}
static void end_ebml_master(AVIOContext *pb, ebml_master master)
{
int64_t pos = avio_tell(pb);
if (avio_seek(pb, master.pos - master.sizebytes, SEEK_SET) < 0)
return;
put_ebml_length(pb, pos - master.pos, master.sizebytes);
avio_seek(pb, pos, SEEK_SET);
}
static int start_ebml_master_crc32(AVIOContext **dyn_cp, MatroskaMuxContext *mkv)
{
int ret;
avformat/matroskaenc: Stop reallocating of Cluster buffer The Matroska muxer uses a dynamic buffer to buffer the content of Clusters before eventually writing them. Up until now, each time a Cluster was written, the dynamic buffer was closed, i.e. freed; now it is only reset, saving allocations of the AVIOContext itself, its opaque as well as most of the reallocations of the buffer. This is advantageous performance-wise, in particular on systems where reallocations are slow (namely Windows). The following table shows the decicyles for writing a frame on Linux (Ubuntu 19.10) and Windows (7) on an x64 Haswell (to /dev/null on Linux, to stdout which is discarded on Windows (the default values of the size and duration of clusters for seekable output have been explicitly set in this case); in all tests, writing CRC-32 values has been disabled in all tests; calls to the muxer's write_packet function in write_packet() in libavformat/mux.c have been timed; each of the following tests has been repeated 50 times): | Windows before | Windows after | Linux before | Linux after _________________________________________________________________ A | 979437 | 192304 | 259500 | 183320 B | 715936 | 155648 | 152786 | 130879 C | 265115 | 56034 | 78496 | 53243 D | 386224 | 80307 | 128894 | 75354 E | 21732 | 10695 | 11320 | 9801 (A is a 10.2 mb/s file with a GOP length of 2s, amounting to an average Cluster size of about 2.5 MiB; the average Cluster size of B is 1.1 MiB; for C it is 2.35 MiB, for D it is 0.46 MiB; for E - a file with just a single audio track of 158kb/s resulting in a Cluster size of about 100 kB, the relative gains were the smallest, probably because of the small Cluster size.) Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
if (!*dyn_cp && (ret = avio_open_dyn_buf(dyn_cp)) < 0)
return ret;
if (mkv->write_crc)
put_ebml_void(*dyn_cp, 6); /* Reserve space for CRC32 so position/size calculations using avio_tell() take it into account */
return 0;
}
static void end_ebml_master_crc32(AVIOContext *pb, AVIOContext **dyn_cp,
MatroskaMuxContext *mkv, uint32_t id,
avformat/matroskaenc: Stop reallocating of Cluster buffer The Matroska muxer uses a dynamic buffer to buffer the content of Clusters before eventually writing them. Up until now, each time a Cluster was written, the dynamic buffer was closed, i.e. freed; now it is only reset, saving allocations of the AVIOContext itself, its opaque as well as most of the reallocations of the buffer. This is advantageous performance-wise, in particular on systems where reallocations are slow (namely Windows). The following table shows the decicyles for writing a frame on Linux (Ubuntu 19.10) and Windows (7) on an x64 Haswell (to /dev/null on Linux, to stdout which is discarded on Windows (the default values of the size and duration of clusters for seekable output have been explicitly set in this case); in all tests, writing CRC-32 values has been disabled in all tests; calls to the muxer's write_packet function in write_packet() in libavformat/mux.c have been timed; each of the following tests has been repeated 50 times): | Windows before | Windows after | Linux before | Linux after _________________________________________________________________ A | 979437 | 192304 | 259500 | 183320 B | 715936 | 155648 | 152786 | 130879 C | 265115 | 56034 | 78496 | 53243 D | 386224 | 80307 | 128894 | 75354 E | 21732 | 10695 | 11320 | 9801 (A is a 10.2 mb/s file with a GOP length of 2s, amounting to an average Cluster size of about 2.5 MiB; the average Cluster size of B is 1.1 MiB; for C it is 2.35 MiB, for D it is 0.46 MiB; for E - a file with just a single audio track of 158kb/s resulting in a Cluster size of about 100 kB, the relative gains were the smallest, probably because of the small Cluster size.) Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
int length_size, int keep_buffer)
{
uint8_t *buf, crc[4];
int size, skip = 0;
put_ebml_id(pb, id);
avformat/matroskaenc: Stop reallocating of Cluster buffer The Matroska muxer uses a dynamic buffer to buffer the content of Clusters before eventually writing them. Up until now, each time a Cluster was written, the dynamic buffer was closed, i.e. freed; now it is only reset, saving allocations of the AVIOContext itself, its opaque as well as most of the reallocations of the buffer. This is advantageous performance-wise, in particular on systems where reallocations are slow (namely Windows). The following table shows the decicyles for writing a frame on Linux (Ubuntu 19.10) and Windows (7) on an x64 Haswell (to /dev/null on Linux, to stdout which is discarded on Windows (the default values of the size and duration of clusters for seekable output have been explicitly set in this case); in all tests, writing CRC-32 values has been disabled in all tests; calls to the muxer's write_packet function in write_packet() in libavformat/mux.c have been timed; each of the following tests has been repeated 50 times): | Windows before | Windows after | Linux before | Linux after _________________________________________________________________ A | 979437 | 192304 | 259500 | 183320 B | 715936 | 155648 | 152786 | 130879 C | 265115 | 56034 | 78496 | 53243 D | 386224 | 80307 | 128894 | 75354 E | 21732 | 10695 | 11320 | 9801 (A is a 10.2 mb/s file with a GOP length of 2s, amounting to an average Cluster size of about 2.5 MiB; the average Cluster size of B is 1.1 MiB; for C it is 2.35 MiB, for D it is 0.46 MiB; for E - a file with just a single audio track of 158kb/s resulting in a Cluster size of about 100 kB, the relative gains were the smallest, probably because of the small Cluster size.) Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
size = avio_get_dyn_buf(*dyn_cp, &buf);
put_ebml_length(pb, size, length_size);
if (mkv->write_crc) {
skip = 6; /* Skip reserved 6-byte long void element from the dynamic buffer. */
AV_WL32(crc, av_crc(av_crc_get_table(AV_CRC_32_IEEE_LE), UINT32_MAX, buf + skip, size - skip) ^ UINT32_MAX);
put_ebml_binary(pb, EBML_ID_CRC32, crc, sizeof(crc));
}
avio_write(pb, buf + skip, size - skip);
avformat/matroskaenc: Stop reallocating of Cluster buffer The Matroska muxer uses a dynamic buffer to buffer the content of Clusters before eventually writing them. Up until now, each time a Cluster was written, the dynamic buffer was closed, i.e. freed; now it is only reset, saving allocations of the AVIOContext itself, its opaque as well as most of the reallocations of the buffer. This is advantageous performance-wise, in particular on systems where reallocations are slow (namely Windows). The following table shows the decicyles for writing a frame on Linux (Ubuntu 19.10) and Windows (7) on an x64 Haswell (to /dev/null on Linux, to stdout which is discarded on Windows (the default values of the size and duration of clusters for seekable output have been explicitly set in this case); in all tests, writing CRC-32 values has been disabled in all tests; calls to the muxer's write_packet function in write_packet() in libavformat/mux.c have been timed; each of the following tests has been repeated 50 times): | Windows before | Windows after | Linux before | Linux after _________________________________________________________________ A | 979437 | 192304 | 259500 | 183320 B | 715936 | 155648 | 152786 | 130879 C | 265115 | 56034 | 78496 | 53243 D | 386224 | 80307 | 128894 | 75354 E | 21732 | 10695 | 11320 | 9801 (A is a 10.2 mb/s file with a GOP length of 2s, amounting to an average Cluster size of about 2.5 MiB; the average Cluster size of B is 1.1 MiB; for C it is 2.35 MiB, for D it is 0.46 MiB; for E - a file with just a single audio track of 158kb/s resulting in a Cluster size of about 100 kB, the relative gains were the smallest, probably because of the small Cluster size.) Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
if (keep_buffer) {
ffio_reset_dyn_buf(*dyn_cp);
} else {
ffio_free_dyn_buf(dyn_cp);
}
}
avformat/matroskaenc: Regression fix for invalid MKV headers The following three commits created a regression by writing initially invalid mkv headers: 650e17d88b63b5aca6e0a43483e89e64b0f7d2dd avformat/matroskaenc: write a CRC32 element on Tags 3bcadf822711720ff0f8d14db71ae47cdf97e652 avformat/matroskaenc: write a CRC32 element on Info ee888cfbe777cd2916a3548c750e433ab8f8e6a5 avformat/matroskaenc: postpone writing the Tracks master Symptoms: - You can no longer playback a file that is still processed by ffmpeg, e.g. VLC fails playback - You can no longer stream a file to a client while if is still being processed - Various diagnosing tools show header errors or incomplete headers (e.g. ffprobe, mediainfo, mkvalidator) Note: The symptoms do not apply to completed files or ffmpeg runs that were interrupted with 'q' Cause: The mentioned commits made changes in a way that some header elements are only partially written in mkv_write_header, leaving the header in an invalid state. Only in mkv_write_trailer, these elements are finished correctly, but that does only occur at the end of the process. Regression: Before these commits were applied, mkv headers have always been valid, even before completion of ffmpeg. This has worked reliably over many versions of ffmpeg, to it was an obvious regression. Bugtracker: This issue has been recorded as #5977 which is resolved by this patch Patch: The patch adds a new function 'end_ebml_master_crc32_preliminary' that preliminarily finishes the ebml element without destroying the buffer. The buffer can be used to update the ebml element later during mkv_write_trailer. But most important: mkv_write_header finishes with a valid mkv header again. Signed-off-by: James Almer <jamrial@gmail.com>
8 years ago
/**
* Complete ebml master without destroying the buffer, allowing for later updates
avformat/matroskaenc: Regression fix for invalid MKV headers The following three commits created a regression by writing initially invalid mkv headers: 650e17d88b63b5aca6e0a43483e89e64b0f7d2dd avformat/matroskaenc: write a CRC32 element on Tags 3bcadf822711720ff0f8d14db71ae47cdf97e652 avformat/matroskaenc: write a CRC32 element on Info ee888cfbe777cd2916a3548c750e433ab8f8e6a5 avformat/matroskaenc: postpone writing the Tracks master Symptoms: - You can no longer playback a file that is still processed by ffmpeg, e.g. VLC fails playback - You can no longer stream a file to a client while if is still being processed - Various diagnosing tools show header errors or incomplete headers (e.g. ffprobe, mediainfo, mkvalidator) Note: The symptoms do not apply to completed files or ffmpeg runs that were interrupted with 'q' Cause: The mentioned commits made changes in a way that some header elements are only partially written in mkv_write_header, leaving the header in an invalid state. Only in mkv_write_trailer, these elements are finished correctly, but that does only occur at the end of the process. Regression: Before these commits were applied, mkv headers have always been valid, even before completion of ffmpeg. This has worked reliably over many versions of ffmpeg, to it was an obvious regression. Bugtracker: This issue has been recorded as #5977 which is resolved by this patch Patch: The patch adds a new function 'end_ebml_master_crc32_preliminary' that preliminarily finishes the ebml element without destroying the buffer. The buffer can be used to update the ebml element later during mkv_write_trailer. But most important: mkv_write_header finishes with a valid mkv header again. Signed-off-by: James Almer <jamrial@gmail.com>
8 years ago
*/
static void end_ebml_master_crc32_preliminary(AVIOContext *pb, AVIOContext *dyn_cp,
uint32_t id, int64_t *pos)
avformat/matroskaenc: Regression fix for invalid MKV headers The following three commits created a regression by writing initially invalid mkv headers: 650e17d88b63b5aca6e0a43483e89e64b0f7d2dd avformat/matroskaenc: write a CRC32 element on Tags 3bcadf822711720ff0f8d14db71ae47cdf97e652 avformat/matroskaenc: write a CRC32 element on Info ee888cfbe777cd2916a3548c750e433ab8f8e6a5 avformat/matroskaenc: postpone writing the Tracks master Symptoms: - You can no longer playback a file that is still processed by ffmpeg, e.g. VLC fails playback - You can no longer stream a file to a client while if is still being processed - Various diagnosing tools show header errors or incomplete headers (e.g. ffprobe, mediainfo, mkvalidator) Note: The symptoms do not apply to completed files or ffmpeg runs that were interrupted with 'q' Cause: The mentioned commits made changes in a way that some header elements are only partially written in mkv_write_header, leaving the header in an invalid state. Only in mkv_write_trailer, these elements are finished correctly, but that does only occur at the end of the process. Regression: Before these commits were applied, mkv headers have always been valid, even before completion of ffmpeg. This has worked reliably over many versions of ffmpeg, to it was an obvious regression. Bugtracker: This issue has been recorded as #5977 which is resolved by this patch Patch: The patch adds a new function 'end_ebml_master_crc32_preliminary' that preliminarily finishes the ebml element without destroying the buffer. The buffer can be used to update the ebml element later during mkv_write_trailer. But most important: mkv_write_header finishes with a valid mkv header again. Signed-off-by: James Almer <jamrial@gmail.com>
8 years ago
{
uint8_t *buf;
int size = avio_get_dyn_buf(dyn_cp, &buf);
avformat/matroskaenc: Regression fix for invalid MKV headers The following three commits created a regression by writing initially invalid mkv headers: 650e17d88b63b5aca6e0a43483e89e64b0f7d2dd avformat/matroskaenc: write a CRC32 element on Tags 3bcadf822711720ff0f8d14db71ae47cdf97e652 avformat/matroskaenc: write a CRC32 element on Info ee888cfbe777cd2916a3548c750e433ab8f8e6a5 avformat/matroskaenc: postpone writing the Tracks master Symptoms: - You can no longer playback a file that is still processed by ffmpeg, e.g. VLC fails playback - You can no longer stream a file to a client while if is still being processed - Various diagnosing tools show header errors or incomplete headers (e.g. ffprobe, mediainfo, mkvalidator) Note: The symptoms do not apply to completed files or ffmpeg runs that were interrupted with 'q' Cause: The mentioned commits made changes in a way that some header elements are only partially written in mkv_write_header, leaving the header in an invalid state. Only in mkv_write_trailer, these elements are finished correctly, but that does only occur at the end of the process. Regression: Before these commits were applied, mkv headers have always been valid, even before completion of ffmpeg. This has worked reliably over many versions of ffmpeg, to it was an obvious regression. Bugtracker: This issue has been recorded as #5977 which is resolved by this patch Patch: The patch adds a new function 'end_ebml_master_crc32_preliminary' that preliminarily finishes the ebml element without destroying the buffer. The buffer can be used to update the ebml element later during mkv_write_trailer. But most important: mkv_write_header finishes with a valid mkv header again. Signed-off-by: James Almer <jamrial@gmail.com>
8 years ago
*pos = avio_tell(pb);
put_ebml_id(pb, id);
put_ebml_length(pb, size, 0);
avio_write(pb, buf, size);
avformat/matroskaenc: Regression fix for invalid MKV headers The following three commits created a regression by writing initially invalid mkv headers: 650e17d88b63b5aca6e0a43483e89e64b0f7d2dd avformat/matroskaenc: write a CRC32 element on Tags 3bcadf822711720ff0f8d14db71ae47cdf97e652 avformat/matroskaenc: write a CRC32 element on Info ee888cfbe777cd2916a3548c750e433ab8f8e6a5 avformat/matroskaenc: postpone writing the Tracks master Symptoms: - You can no longer playback a file that is still processed by ffmpeg, e.g. VLC fails playback - You can no longer stream a file to a client while if is still being processed - Various diagnosing tools show header errors or incomplete headers (e.g. ffprobe, mediainfo, mkvalidator) Note: The symptoms do not apply to completed files or ffmpeg runs that were interrupted with 'q' Cause: The mentioned commits made changes in a way that some header elements are only partially written in mkv_write_header, leaving the header in an invalid state. Only in mkv_write_trailer, these elements are finished correctly, but that does only occur at the end of the process. Regression: Before these commits were applied, mkv headers have always been valid, even before completion of ffmpeg. This has worked reliably over many versions of ffmpeg, to it was an obvious regression. Bugtracker: This issue has been recorded as #5977 which is resolved by this patch Patch: The patch adds a new function 'end_ebml_master_crc32_preliminary' that preliminarily finishes the ebml element without destroying the buffer. The buffer can be used to update the ebml element later during mkv_write_trailer. But most important: mkv_write_header finishes with a valid mkv header again. Signed-off-by: James Almer <jamrial@gmail.com>
8 years ago
}
static void put_xiph_size(AVIOContext *pb, int size)
{
ffio_fill(pb, 255, size / 255);
avio_w8(pb, size % 255);
}
/**
* Free the members allocated in the mux context.
*/
static void mkv_deinit(AVFormatContext *s)
{
MatroskaMuxContext *mkv = s->priv_data;
av_packet_unref(&mkv->cur_audio_pkt);
ffio_free_dyn_buf(&mkv->cluster_bc);
ffio_free_dyn_buf(&mkv->info_bc);
ffio_free_dyn_buf(&mkv->tracks_bc);
ffio_free_dyn_buf(&mkv->tags_bc);
av_freep(&mkv->cues.entries);
av_freep(&mkv->tracks);
}
/**
avformat/matroskaenc: Avoid allocations for SeekHead Up until e7ddafd5, the Matroska muxer wrote two SeekHeads: One at the beginning referencing the main level 1 elements (i.e. not the Clusters) and one at the end, referencing the Clusters. This second SeekHead was useless and has therefore been removed. Yet the SeekHead-related functions and structures are still geared towards this usecase: They are built around an allocated array of variable size that gets reallocated every time an element is added to it although the maximum number of Seek entries is a small compile-time constant, so that one should rather include the array in the SeekHead structure itself; and said structure should be contained in the MatroskaMuxContext instead of being allocated separately. The earlier code reserved space for a SeekHead with 10 entries, although we currently write at most 6. Reducing said number implied that every Matroska/Webm file will be 84 bytes smaller and required to adapt several FATE tests; furthermore, the reserved amount overestimated the amount needed for for the SeekHead's length field and how many bytes need to be reserved to write a EBML Void element, bringing the total reduction to 89 bytes. This also fixes a potential segfault: If !mkv->is_live and if the AVIOContext is initially unseekable when writing the header, the SeekHead is already written when writing the header and this used to free the SeekHead-related structures that have been allocated. But if the AVIOContext happens to be seekable when writing the trailer, it will be attempted to write the SeekHead again which will lead to segfaults because the corresponding structures have already been freed. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
* Initialize the SeekHead element to be ready to index level 1 Matroska
* elements. Enough space to write MAX_SEEKHEAD_ENTRIES SeekHead entries
* will be reserved at the current file location.
*/
avformat/matroskaenc: Avoid allocations for SeekHead Up until e7ddafd5, the Matroska muxer wrote two SeekHeads: One at the beginning referencing the main level 1 elements (i.e. not the Clusters) and one at the end, referencing the Clusters. This second SeekHead was useless and has therefore been removed. Yet the SeekHead-related functions and structures are still geared towards this usecase: They are built around an allocated array of variable size that gets reallocated every time an element is added to it although the maximum number of Seek entries is a small compile-time constant, so that one should rather include the array in the SeekHead structure itself; and said structure should be contained in the MatroskaMuxContext instead of being allocated separately. The earlier code reserved space for a SeekHead with 10 entries, although we currently write at most 6. Reducing said number implied that every Matroska/Webm file will be 84 bytes smaller and required to adapt several FATE tests; furthermore, the reserved amount overestimated the amount needed for for the SeekHead's length field and how many bytes need to be reserved to write a EBML Void element, bringing the total reduction to 89 bytes. This also fixes a potential segfault: If !mkv->is_live and if the AVIOContext is initially unseekable when writing the header, the SeekHead is already written when writing the header and this used to free the SeekHead-related structures that have been allocated. But if the AVIOContext happens to be seekable when writing the trailer, it will be attempted to write the SeekHead again which will lead to segfaults because the corresponding structures have already been freed. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
static void mkv_start_seekhead(MatroskaMuxContext *mkv, AVIOContext *pb)
{
avformat/matroskaenc: Avoid allocations for SeekHead Up until e7ddafd5, the Matroska muxer wrote two SeekHeads: One at the beginning referencing the main level 1 elements (i.e. not the Clusters) and one at the end, referencing the Clusters. This second SeekHead was useless and has therefore been removed. Yet the SeekHead-related functions and structures are still geared towards this usecase: They are built around an allocated array of variable size that gets reallocated every time an element is added to it although the maximum number of Seek entries is a small compile-time constant, so that one should rather include the array in the SeekHead structure itself; and said structure should be contained in the MatroskaMuxContext instead of being allocated separately. The earlier code reserved space for a SeekHead with 10 entries, although we currently write at most 6. Reducing said number implied that every Matroska/Webm file will be 84 bytes smaller and required to adapt several FATE tests; furthermore, the reserved amount overestimated the amount needed for for the SeekHead's length field and how many bytes need to be reserved to write a EBML Void element, bringing the total reduction to 89 bytes. This also fixes a potential segfault: If !mkv->is_live and if the AVIOContext is initially unseekable when writing the header, the SeekHead is already written when writing the header and this used to free the SeekHead-related structures that have been allocated. But if the AVIOContext happens to be seekable when writing the trailer, it will be attempted to write the SeekHead again which will lead to segfaults because the corresponding structures have already been freed. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
mkv->seekhead.filepos = avio_tell(pb);
// 21 bytes max for a Seek entry, 6 bytes max for the SeekHead ID
// and size, 6 bytes for a CRC32 element, and 2 bytes to guarantee
// that an EBML void element will fit afterwards
mkv->seekhead.reserved_size = MAX_SEEKHEAD_ENTRIES * MAX_SEEKENTRY_SIZE + 14;
put_ebml_void(pb, mkv->seekhead.reserved_size);
}
avformat/matroskaenc: Avoid allocations for SeekHead Up until e7ddafd5, the Matroska muxer wrote two SeekHeads: One at the beginning referencing the main level 1 elements (i.e. not the Clusters) and one at the end, referencing the Clusters. This second SeekHead was useless and has therefore been removed. Yet the SeekHead-related functions and structures are still geared towards this usecase: They are built around an allocated array of variable size that gets reallocated every time an element is added to it although the maximum number of Seek entries is a small compile-time constant, so that one should rather include the array in the SeekHead structure itself; and said structure should be contained in the MatroskaMuxContext instead of being allocated separately. The earlier code reserved space for a SeekHead with 10 entries, although we currently write at most 6. Reducing said number implied that every Matroska/Webm file will be 84 bytes smaller and required to adapt several FATE tests; furthermore, the reserved amount overestimated the amount needed for for the SeekHead's length field and how many bytes need to be reserved to write a EBML Void element, bringing the total reduction to 89 bytes. This also fixes a potential segfault: If !mkv->is_live and if the AVIOContext is initially unseekable when writing the header, the SeekHead is already written when writing the header and this used to free the SeekHead-related structures that have been allocated. But if the AVIOContext happens to be seekable when writing the trailer, it will be attempted to write the SeekHead again which will lead to segfaults because the corresponding structures have already been freed. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
static void mkv_add_seekhead_entry(MatroskaMuxContext *mkv, uint32_t elementid,
uint64_t filepos)
{
avformat/matroskaenc: Avoid allocations for SeekHead Up until e7ddafd5, the Matroska muxer wrote two SeekHeads: One at the beginning referencing the main level 1 elements (i.e. not the Clusters) and one at the end, referencing the Clusters. This second SeekHead was useless and has therefore been removed. Yet the SeekHead-related functions and structures are still geared towards this usecase: They are built around an allocated array of variable size that gets reallocated every time an element is added to it although the maximum number of Seek entries is a small compile-time constant, so that one should rather include the array in the SeekHead structure itself; and said structure should be contained in the MatroskaMuxContext instead of being allocated separately. The earlier code reserved space for a SeekHead with 10 entries, although we currently write at most 6. Reducing said number implied that every Matroska/Webm file will be 84 bytes smaller and required to adapt several FATE tests; furthermore, the reserved amount overestimated the amount needed for for the SeekHead's length field and how many bytes need to be reserved to write a EBML Void element, bringing the total reduction to 89 bytes. This also fixes a potential segfault: If !mkv->is_live and if the AVIOContext is initially unseekable when writing the header, the SeekHead is already written when writing the header and this used to free the SeekHead-related structures that have been allocated. But if the AVIOContext happens to be seekable when writing the trailer, it will be attempted to write the SeekHead again which will lead to segfaults because the corresponding structures have already been freed. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
mkv_seekhead *seekhead = &mkv->seekhead;
avformat/matroskaenc: Avoid allocations for SeekHead Up until e7ddafd5, the Matroska muxer wrote two SeekHeads: One at the beginning referencing the main level 1 elements (i.e. not the Clusters) and one at the end, referencing the Clusters. This second SeekHead was useless and has therefore been removed. Yet the SeekHead-related functions and structures are still geared towards this usecase: They are built around an allocated array of variable size that gets reallocated every time an element is added to it although the maximum number of Seek entries is a small compile-time constant, so that one should rather include the array in the SeekHead structure itself; and said structure should be contained in the MatroskaMuxContext instead of being allocated separately. The earlier code reserved space for a SeekHead with 10 entries, although we currently write at most 6. Reducing said number implied that every Matroska/Webm file will be 84 bytes smaller and required to adapt several FATE tests; furthermore, the reserved amount overestimated the amount needed for for the SeekHead's length field and how many bytes need to be reserved to write a EBML Void element, bringing the total reduction to 89 bytes. This also fixes a potential segfault: If !mkv->is_live and if the AVIOContext is initially unseekable when writing the header, the SeekHead is already written when writing the header and this used to free the SeekHead-related structures that have been allocated. But if the AVIOContext happens to be seekable when writing the trailer, it will be attempted to write the SeekHead again which will lead to segfaults because the corresponding structures have already been freed. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
av_assert1(seekhead->num_entries < MAX_SEEKHEAD_ENTRIES);
seekhead->entries[seekhead->num_entries].elementid = elementid;
avformat/matroskaenc: Avoid allocations for SeekHead Up until e7ddafd5, the Matroska muxer wrote two SeekHeads: One at the beginning referencing the main level 1 elements (i.e. not the Clusters) and one at the end, referencing the Clusters. This second SeekHead was useless and has therefore been removed. Yet the SeekHead-related functions and structures are still geared towards this usecase: They are built around an allocated array of variable size that gets reallocated every time an element is added to it although the maximum number of Seek entries is a small compile-time constant, so that one should rather include the array in the SeekHead structure itself; and said structure should be contained in the MatroskaMuxContext instead of being allocated separately. The earlier code reserved space for a SeekHead with 10 entries, although we currently write at most 6. Reducing said number implied that every Matroska/Webm file will be 84 bytes smaller and required to adapt several FATE tests; furthermore, the reserved amount overestimated the amount needed for for the SeekHead's length field and how many bytes need to be reserved to write a EBML Void element, bringing the total reduction to 89 bytes. This also fixes a potential segfault: If !mkv->is_live and if the AVIOContext is initially unseekable when writing the header, the SeekHead is already written when writing the header and this used to free the SeekHead-related structures that have been allocated. But if the AVIOContext happens to be seekable when writing the trailer, it will be attempted to write the SeekHead again which will lead to segfaults because the corresponding structures have already been freed. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
seekhead->entries[seekhead->num_entries++].segmentpos = filepos - mkv->segment_offset;
}
/**
* Write the SeekHead to the file at the location reserved for it
* and seek to destpos afterwards. When error_on_seek_failure
* is not set, failure to seek to the position designated for the
* SeekHead is not considered an error and it is presumed that
* destpos is the current position; failure to seek to destpos
* afterwards is always an error.
*
* @return 0 on success, < 0 on error.
*/
static int mkv_write_seekhead(AVIOContext *pb, MatroskaMuxContext *mkv,
int error_on_seek_failure, int64_t destpos)
{
avformat/matroskaenc: Stop reallocating of Cluster buffer The Matroska muxer uses a dynamic buffer to buffer the content of Clusters before eventually writing them. Up until now, each time a Cluster was written, the dynamic buffer was closed, i.e. freed; now it is only reset, saving allocations of the AVIOContext itself, its opaque as well as most of the reallocations of the buffer. This is advantageous performance-wise, in particular on systems where reallocations are slow (namely Windows). The following table shows the decicyles for writing a frame on Linux (Ubuntu 19.10) and Windows (7) on an x64 Haswell (to /dev/null on Linux, to stdout which is discarded on Windows (the default values of the size and duration of clusters for seekable output have been explicitly set in this case); in all tests, writing CRC-32 values has been disabled in all tests; calls to the muxer's write_packet function in write_packet() in libavformat/mux.c have been timed; each of the following tests has been repeated 50 times): | Windows before | Windows after | Linux before | Linux after _________________________________________________________________ A | 979437 | 192304 | 259500 | 183320 B | 715936 | 155648 | 152786 | 130879 C | 265115 | 56034 | 78496 | 53243 D | 386224 | 80307 | 128894 | 75354 E | 21732 | 10695 | 11320 | 9801 (A is a 10.2 mb/s file with a GOP length of 2s, amounting to an average Cluster size of about 2.5 MiB; the average Cluster size of B is 1.1 MiB; for C it is 2.35 MiB, for D it is 0.46 MiB; for E - a file with just a single audio track of 158kb/s resulting in a Cluster size of about 100 kB, the relative gains were the smallest, probably because of the small Cluster size.) Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
AVIOContext *dyn_cp = NULL;
avformat/matroskaenc: Avoid allocations for SeekHead Up until e7ddafd5, the Matroska muxer wrote two SeekHeads: One at the beginning referencing the main level 1 elements (i.e. not the Clusters) and one at the end, referencing the Clusters. This second SeekHead was useless and has therefore been removed. Yet the SeekHead-related functions and structures are still geared towards this usecase: They are built around an allocated array of variable size that gets reallocated every time an element is added to it although the maximum number of Seek entries is a small compile-time constant, so that one should rather include the array in the SeekHead structure itself; and said structure should be contained in the MatroskaMuxContext instead of being allocated separately. The earlier code reserved space for a SeekHead with 10 entries, although we currently write at most 6. Reducing said number implied that every Matroska/Webm file will be 84 bytes smaller and required to adapt several FATE tests; furthermore, the reserved amount overestimated the amount needed for for the SeekHead's length field and how many bytes need to be reserved to write a EBML Void element, bringing the total reduction to 89 bytes. This also fixes a potential segfault: If !mkv->is_live and if the AVIOContext is initially unseekable when writing the header, the SeekHead is already written when writing the header and this used to free the SeekHead-related structures that have been allocated. But if the AVIOContext happens to be seekable when writing the trailer, it will be attempted to write the SeekHead again which will lead to segfaults because the corresponding structures have already been freed. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
mkv_seekhead *seekhead = &mkv->seekhead;
int64_t remaining, ret64;
int i, ret;
if ((ret64 = avio_seek(pb, seekhead->filepos, SEEK_SET)) < 0)
return error_on_seek_failure ? ret64 : 0;
ret = start_ebml_master_crc32(&dyn_cp, mkv);
if (ret < 0)
return ret;
for (i = 0; i < seekhead->num_entries; i++) {
mkv_seekhead_entry *entry = &seekhead->entries[i];
ebml_master seekentry = start_ebml_master(dyn_cp, MATROSKA_ID_SEEKENTRY,
MAX_SEEKENTRY_SIZE);
put_ebml_id(dyn_cp, MATROSKA_ID_SEEKID);
put_ebml_length(dyn_cp, ebml_id_size(entry->elementid), 0);
put_ebml_id(dyn_cp, entry->elementid);
put_ebml_uint(dyn_cp, MATROSKA_ID_SEEKPOSITION, entry->segmentpos);
end_ebml_master(dyn_cp, seekentry);
}
avformat/matroskaenc: Stop reallocating of Cluster buffer The Matroska muxer uses a dynamic buffer to buffer the content of Clusters before eventually writing them. Up until now, each time a Cluster was written, the dynamic buffer was closed, i.e. freed; now it is only reset, saving allocations of the AVIOContext itself, its opaque as well as most of the reallocations of the buffer. This is advantageous performance-wise, in particular on systems where reallocations are slow (namely Windows). The following table shows the decicyles for writing a frame on Linux (Ubuntu 19.10) and Windows (7) on an x64 Haswell (to /dev/null on Linux, to stdout which is discarded on Windows (the default values of the size and duration of clusters for seekable output have been explicitly set in this case); in all tests, writing CRC-32 values has been disabled in all tests; calls to the muxer's write_packet function in write_packet() in libavformat/mux.c have been timed; each of the following tests has been repeated 50 times): | Windows before | Windows after | Linux before | Linux after _________________________________________________________________ A | 979437 | 192304 | 259500 | 183320 B | 715936 | 155648 | 152786 | 130879 C | 265115 | 56034 | 78496 | 53243 D | 386224 | 80307 | 128894 | 75354 E | 21732 | 10695 | 11320 | 9801 (A is a 10.2 mb/s file with a GOP length of 2s, amounting to an average Cluster size of about 2.5 MiB; the average Cluster size of B is 1.1 MiB; for C it is 2.35 MiB, for D it is 0.46 MiB; for E - a file with just a single audio track of 158kb/s resulting in a Cluster size of about 100 kB, the relative gains were the smallest, probably because of the small Cluster size.) Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
end_ebml_master_crc32(pb, &dyn_cp, mkv, MATROSKA_ID_SEEKHEAD, 0, 0);
avformat/matroskaenc: Avoid allocations for SeekHead Up until e7ddafd5, the Matroska muxer wrote two SeekHeads: One at the beginning referencing the main level 1 elements (i.e. not the Clusters) and one at the end, referencing the Clusters. This second SeekHead was useless and has therefore been removed. Yet the SeekHead-related functions and structures are still geared towards this usecase: They are built around an allocated array of variable size that gets reallocated every time an element is added to it although the maximum number of Seek entries is a small compile-time constant, so that one should rather include the array in the SeekHead structure itself; and said structure should be contained in the MatroskaMuxContext instead of being allocated separately. The earlier code reserved space for a SeekHead with 10 entries, although we currently write at most 6. Reducing said number implied that every Matroska/Webm file will be 84 bytes smaller and required to adapt several FATE tests; furthermore, the reserved amount overestimated the amount needed for for the SeekHead's length field and how many bytes need to be reserved to write a EBML Void element, bringing the total reduction to 89 bytes. This also fixes a potential segfault: If !mkv->is_live and if the AVIOContext is initially unseekable when writing the header, the SeekHead is already written when writing the header and this used to free the SeekHead-related structures that have been allocated. But if the AVIOContext happens to be seekable when writing the trailer, it will be attempted to write the SeekHead again which will lead to segfaults because the corresponding structures have already been freed. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
remaining = seekhead->filepos + seekhead->reserved_size - avio_tell(pb);
put_ebml_void(pb, remaining);
avformat/matroskaenc: Avoid allocations for SeekHead Up until e7ddafd5, the Matroska muxer wrote two SeekHeads: One at the beginning referencing the main level 1 elements (i.e. not the Clusters) and one at the end, referencing the Clusters. This second SeekHead was useless and has therefore been removed. Yet the SeekHead-related functions and structures are still geared towards this usecase: They are built around an allocated array of variable size that gets reallocated every time an element is added to it although the maximum number of Seek entries is a small compile-time constant, so that one should rather include the array in the SeekHead structure itself; and said structure should be contained in the MatroskaMuxContext instead of being allocated separately. The earlier code reserved space for a SeekHead with 10 entries, although we currently write at most 6. Reducing said number implied that every Matroska/Webm file will be 84 bytes smaller and required to adapt several FATE tests; furthermore, the reserved amount overestimated the amount needed for for the SeekHead's length field and how many bytes need to be reserved to write a EBML Void element, bringing the total reduction to 89 bytes. This also fixes a potential segfault: If !mkv->is_live and if the AVIOContext is initially unseekable when writing the header, the SeekHead is already written when writing the header and this used to free the SeekHead-related structures that have been allocated. But if the AVIOContext happens to be seekable when writing the trailer, it will be attempted to write the SeekHead again which will lead to segfaults because the corresponding structures have already been freed. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
if ((ret64 = avio_seek(pb, destpos, SEEK_SET)) < 0)
return ret64;
return 0;
}
static int mkv_add_cuepoint(MatroskaMuxContext *mkv, int stream, int64_t ts,
int64_t cluster_pos, int64_t relative_pos, int64_t duration)
{
mkv_cues *cues = &mkv->cues;
mkv_cuepoint *entries = cues->entries;
if (ts < 0)
return 0;
entries = av_realloc_array(entries, cues->num_entries + 1, sizeof(mkv_cuepoint));
if (!entries)
return AVERROR(ENOMEM);
cues->entries = entries;
cues->entries[cues->num_entries].pts = ts;
cues->entries[cues->num_entries].stream_idx = stream;
cues->entries[cues->num_entries].cluster_pos = cluster_pos - mkv->segment_offset;
cues->entries[cues->num_entries].relative_pos = relative_pos;
cues->entries[cues->num_entries++].duration = duration;
return 0;
}
static int mkv_assemble_cues(AVStream **streams, AVIOContext *dyn_cp,
mkv_cues *cues, mkv_track *tracks, int num_tracks)
{
AVIOContext *cuepoint;
int ret;
ret = avio_open_dyn_buf(&cuepoint);
if (ret < 0)
return ret;
for (mkv_cuepoint *entry = cues->entries, *end = entry + cues->num_entries;
entry < end;) {
uint64_t pts = entry->pts;
uint8_t *buf;
int size;
put_ebml_uint(cuepoint, MATROSKA_ID_CUETIME, pts);
// put all the entries from different tracks that have the exact same
// timestamp into the same CuePoint
for (int j = 0; j < num_tracks; j++)
tracks[j].has_cue = 0;
do {
ebml_master track_positions;
int idx = entry->stream_idx;
av_assert0(idx >= 0 && idx < num_tracks);
if (tracks[idx].has_cue && streams[idx]->codecpar->codec_type != AVMEDIA_TYPE_SUBTITLE)
continue;
tracks[idx].has_cue = 1;
track_positions = start_ebml_master(cuepoint, MATROSKA_ID_CUETRACKPOSITION, MAX_CUETRACKPOS_SIZE);
put_ebml_uint(cuepoint, MATROSKA_ID_CUETRACK , tracks[idx].track_num);
put_ebml_uint(cuepoint, MATROSKA_ID_CUECLUSTERPOSITION , entry->cluster_pos);
put_ebml_uint(cuepoint, MATROSKA_ID_CUERELATIVEPOSITION, entry->relative_pos);
if (entry->duration != -1)
put_ebml_uint(cuepoint, MATROSKA_ID_CUEDURATION , entry->duration);
end_ebml_master(cuepoint, track_positions);
} while (++entry < end && entry->pts == pts);
size = avio_get_dyn_buf(cuepoint, &buf);
put_ebml_binary(dyn_cp, MATROSKA_ID_POINTENTRY, buf, size);
ffio_reset_dyn_buf(cuepoint);
}
ffio_free_dyn_buf(&cuepoint);
return 0;
}
static int put_xiph_codecpriv(AVFormatContext *s, AVIOContext *pb,
const AVCodecParameters *par)
{
const uint8_t *header_start[3];
int header_len[3];
int first_header_size;
int j;
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.
11 years ago
if (par->codec_id == AV_CODEC_ID_VORBIS)
first_header_size = 30;
else
first_header_size = 42;
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.
11 years ago
if (avpriv_split_xiph_headers(par->extradata, par->extradata_size,
first_header_size, header_start, header_len) < 0) {
av_log(s, AV_LOG_ERROR, "Extradata corrupt.\n");
return -1;
}
avio_w8(pb, 2); // number packets - 1
for (j = 0; j < 2; j++) {
put_xiph_size(pb, header_len[j]);
}
for (j = 0; j < 3; j++)
avio_write(pb, header_start[j], header_len[j]);
return 0;
}
static int put_wv_codecpriv(AVIOContext *pb, const AVCodecParameters *par)
{
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.
11 years ago
if (par->extradata && par->extradata_size == 2)
avio_write(pb, par->extradata, 2);
else
avio_wl16(pb, 0x410); // fallback to the most recent version
return 0;
}
static int put_flac_codecpriv(AVFormatContext *s, AVIOContext *pb,
const AVCodecParameters *par)
{
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.
11 years ago
int write_comment = (par->channel_layout &&
!(par->channel_layout & ~0x3ffffULL) &&
!ff_flac_is_native_layout(par->channel_layout));
int ret = ff_flac_write_header(pb, par->extradata, par->extradata_size,
!write_comment);
if (ret < 0)
return ret;
if (write_comment) {
const char *vendor = (s->flags & AVFMT_FLAG_BITEXACT) ?
"Lavf" : LIBAVFORMAT_IDENT;
AVDictionary *dict = NULL;
uint8_t buf[32], *data, *p;
int64_t len;
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.
11 years ago
snprintf(buf, sizeof(buf), "0x%"PRIx64, par->channel_layout);
av_dict_set(&dict, "WAVEFORMATEXTENSIBLE_CHANNEL_MASK", buf, 0);
len = ff_vorbiscomment_length(dict, vendor, NULL, 0);
if (len >= (1 << 24) - 4) {
av_dict_free(&dict);
return AVERROR(EINVAL);
}
data = av_malloc(len + 4);
if (!data) {
av_dict_free(&dict);
return AVERROR(ENOMEM);
}
data[0] = 0x84;
AV_WB24(data + 1, len);
p = data + 4;
ff_vorbiscomment_write(&p, &dict, vendor, NULL, 0);
avio_write(pb, data, len + 4);
av_freep(&data);
av_dict_free(&dict);
}
return 0;
}
static int get_aac_sample_rates(AVFormatContext *s, const uint8_t *extradata,
int extradata_size, int *sample_rate,
int *output_sample_rate)
{
MPEG4AudioConfig mp4ac;
int ret;
ret = avpriv_mpeg4audio_get_config2(&mp4ac, extradata, extradata_size, 1, s);
/* Don't abort if the failure is because of missing extradata. Assume in that
* case a bitstream filter will provide the muxer with the extradata in the
* first packet.
* Abort however if s->pb is not seekable, as we would not be able to seek back
* to write the sample rate elements once the extradata shows up, anyway. */
if (ret < 0 && (extradata_size || !(s->pb->seekable & AVIO_SEEKABLE_NORMAL))) {
av_log(s, AV_LOG_ERROR,
"Error parsing AAC extradata, unable to determine samplerate.\n");
return AVERROR(EINVAL);
}
if (ret < 0) {
/* This will only happen when this function is called while writing the
* header and no extradata is available. The space for this element has
* to be reserved for when this function is called again after the
* extradata shows up in the first packet, as there's no way to know if
* output_sample_rate will be different than sample_rate or not. */
*output_sample_rate = *sample_rate;
} else {
*sample_rate = mp4ac.sample_rate;
*output_sample_rate = mp4ac.ext_sample_rate;
}
return 0;
}
static int mkv_write_native_codecprivate(AVFormatContext *s, AVIOContext *pb,
const AVCodecParameters *par,
AVIOContext *dyn_cp)
{
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.
11 years ago
switch (par->codec_id) {
case AV_CODEC_ID_VORBIS:
case AV_CODEC_ID_THEORA:
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.
11 years ago
return put_xiph_codecpriv(s, dyn_cp, par);
case AV_CODEC_ID_FLAC:
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.
11 years ago
return put_flac_codecpriv(s, dyn_cp, par);
case AV_CODEC_ID_WAVPACK:
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.
11 years ago
return put_wv_codecpriv(dyn_cp, par);
case AV_CODEC_ID_H264:
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.
11 years ago
return ff_isom_write_avcc(dyn_cp, par->extradata,
par->extradata_size);
case AV_CODEC_ID_HEVC:
return ff_isom_write_hvcc(dyn_cp, par->extradata,
par->extradata_size, 0);
case AV_CODEC_ID_AV1:
if (par->extradata_size)
return ff_isom_write_av1c(dyn_cp, par->extradata,
par->extradata_size);
else
put_ebml_void(pb, 4 + 3);
break;
case AV_CODEC_ID_ALAC:
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.
11 years ago
if (par->extradata_size < 36) {
av_log(s, AV_LOG_ERROR,
"Invalid extradata found, ALAC expects a 36-byte "
"QuickTime atom.");
return AVERROR_INVALIDDATA;
} else
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.
11 years ago
avio_write(dyn_cp, par->extradata + 12,
par->extradata_size - 12);
break;
case AV_CODEC_ID_AAC:
if (par->extradata_size)
avio_write(dyn_cp, par->extradata, par->extradata_size);
else
put_ebml_void(pb, MAX_PCE_SIZE + 2 + 4);
break;
default:
if (par->codec_id == AV_CODEC_ID_PRORES &&
ff_codec_get_id(ff_codec_movvideo_tags, par->codec_tag) == AV_CODEC_ID_PRORES) {
avio_wl32(dyn_cp, par->codec_tag);
} else if (par->extradata_size && par->codec_id != AV_CODEC_ID_TTA)
avio_write(dyn_cp, par->extradata, par->extradata_size);
}
return 0;
}
static int mkv_write_codecprivate(AVFormatContext *s, AVIOContext *pb,
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.
11 years ago
AVCodecParameters *par,
int native_id, int qt_id)
{
AVIOContext *dyn_cp;
uint8_t *codecpriv;
int ret, codecpriv_size;
ret = avio_open_dyn_buf(&dyn_cp);
if (ret < 0)
return ret;
if (native_id) {
ret = mkv_write_native_codecprivate(s, pb, par, dyn_cp);
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.
11 years ago
} else if (par->codec_type == AVMEDIA_TYPE_VIDEO) {
if (qt_id) {
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.
11 years ago
if (!par->codec_tag)
par->codec_tag = ff_codec_get_tag(ff_codec_movvideo_tags,
par->codec_id);
if ( ff_codec_get_id(ff_codec_movvideo_tags, par->codec_tag) == par->codec_id
&& (!par->extradata_size || ff_codec_get_id(ff_codec_movvideo_tags, AV_RL32(par->extradata + 4)) != par->codec_id)
) {
int i;
avio_wb32(dyn_cp, 0x5a + par->extradata_size);
avio_wl32(dyn_cp, par->codec_tag);
for(i = 0; i < 0x5a - 8; i++)
avio_w8(dyn_cp, 0);
}
avio_write(dyn_cp, par->extradata, par->extradata_size);
} else {
if (!ff_codec_get_tag(ff_codec_bmp_tags, par->codec_id))
av_log(s, AV_LOG_WARNING, "codec %s is not supported by this format\n",
avcodec_get_name(par->codec_id));
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.
11 years ago
if (!par->codec_tag)
par->codec_tag = ff_codec_get_tag(ff_codec_bmp_tags,
par->codec_id);
if (!par->codec_tag && par->codec_id != AV_CODEC_ID_RAWVIDEO) {
av_log(s, AV_LOG_ERROR, "No bmp codec tag found for codec %s\n",
avcodec_get_name(par->codec_id));
ret = AVERROR(EINVAL);
}
ff_put_bmp_header(dyn_cp, par, 0, 0);
}
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.
11 years ago
} else if (par->codec_type == AVMEDIA_TYPE_AUDIO) {
unsigned int tag;
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.
11 years ago
tag = ff_codec_get_tag(ff_codec_wav_tags, par->codec_id);
if (!tag) {
av_log(s, AV_LOG_ERROR, "No wav codec tag found for codec %s\n",
avcodec_get_name(par->codec_id));
ret = AVERROR(EINVAL);
}
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.
11 years ago
if (!par->codec_tag)
par->codec_tag = tag;
ff_put_wav_header(s, dyn_cp, par, FF_PUT_WAV_HEADER_FORCE_WAVEFORMATEX);
}
codecpriv_size = avio_get_dyn_buf(dyn_cp, &codecpriv);
if (codecpriv_size)
put_ebml_binary(pb, MATROSKA_ID_CODECPRIVATE, codecpriv,
codecpriv_size);
ffio_free_dyn_buf(&dyn_cp);
return ret;
}
static int mkv_write_video_color(AVIOContext *pb, const AVCodecParameters *par,
const AVStream *st)
{
AVIOContext *dyn_cp;
uint8_t *colorinfo_ptr;
int side_data_size = 0;
int ret, colorinfo_size;
const uint8_t *side_data;
ret = avio_open_dyn_buf(&dyn_cp);
if (ret < 0)
return ret;
if (par->color_trc != AVCOL_TRC_UNSPECIFIED &&
par->color_trc < AVCOL_TRC_NB) {
put_ebml_uint(dyn_cp, MATROSKA_ID_VIDEOCOLORTRANSFERCHARACTERISTICS,
par->color_trc);
}
if (par->color_space != AVCOL_SPC_UNSPECIFIED &&
par->color_space < AVCOL_SPC_NB) {
put_ebml_uint(dyn_cp, MATROSKA_ID_VIDEOCOLORMATRIXCOEFF, par->color_space);
}
if (par->color_primaries != AVCOL_PRI_UNSPECIFIED &&
par->color_primaries < AVCOL_PRI_NB) {
put_ebml_uint(dyn_cp, MATROSKA_ID_VIDEOCOLORPRIMARIES, par->color_primaries);
}
if (par->color_range != AVCOL_RANGE_UNSPECIFIED &&
par->color_range < AVCOL_RANGE_NB) {
put_ebml_uint(dyn_cp, MATROSKA_ID_VIDEOCOLORRANGE, par->color_range);
}
if (par->chroma_location != AVCHROMA_LOC_UNSPECIFIED &&
par->chroma_location <= AVCHROMA_LOC_TOP) {
int xpos, ypos;
avcodec_enum_to_chroma_pos(&xpos, &ypos, par->chroma_location);
put_ebml_uint(dyn_cp, MATROSKA_ID_VIDEOCOLORCHROMASITINGHORZ, (xpos >> 7) + 1);
put_ebml_uint(dyn_cp, MATROSKA_ID_VIDEOCOLORCHROMASITINGVERT, (ypos >> 7) + 1);
}
side_data = av_stream_get_side_data(st, AV_PKT_DATA_CONTENT_LIGHT_LEVEL,
&side_data_size);
if (side_data_size) {
const AVContentLightMetadata *metadata =
(const AVContentLightMetadata*)side_data;
put_ebml_uint(dyn_cp, MATROSKA_ID_VIDEOCOLORMAXCLL, metadata->MaxCLL);
put_ebml_uint(dyn_cp, MATROSKA_ID_VIDEOCOLORMAXFALL, metadata->MaxFALL);
}
side_data = av_stream_get_side_data(st, AV_PKT_DATA_MASTERING_DISPLAY_METADATA,
&side_data_size);
if (side_data_size == sizeof(AVMasteringDisplayMetadata)) {
ebml_master meta_element = start_ebml_master(
dyn_cp, MATROSKA_ID_VIDEOCOLORMASTERINGMETA, 10 * (2 + 1 + 8));
const AVMasteringDisplayMetadata *metadata =
(const AVMasteringDisplayMetadata*)side_data;
if (metadata->has_primaries) {
put_ebml_float(dyn_cp, MATROSKA_ID_VIDEOCOLOR_RX,
av_q2d(metadata->display_primaries[0][0]));
put_ebml_float(dyn_cp, MATROSKA_ID_VIDEOCOLOR_RY,
av_q2d(metadata->display_primaries[0][1]));
put_ebml_float(dyn_cp, MATROSKA_ID_VIDEOCOLOR_GX,
av_q2d(metadata->display_primaries[1][0]));
put_ebml_float(dyn_cp, MATROSKA_ID_VIDEOCOLOR_GY,
av_q2d(metadata->display_primaries[1][1]));
put_ebml_float(dyn_cp, MATROSKA_ID_VIDEOCOLOR_BX,
av_q2d(metadata->display_primaries[2][0]));
put_ebml_float(dyn_cp, MATROSKA_ID_VIDEOCOLOR_BY,
av_q2d(metadata->display_primaries[2][1]));
put_ebml_float(dyn_cp, MATROSKA_ID_VIDEOCOLOR_WHITEX,
av_q2d(metadata->white_point[0]));
put_ebml_float(dyn_cp, MATROSKA_ID_VIDEOCOLOR_WHITEY,
av_q2d(metadata->white_point[1]));
}
if (metadata->has_luminance) {
put_ebml_float(dyn_cp, MATROSKA_ID_VIDEOCOLOR_LUMINANCEMAX,
av_q2d(metadata->max_luminance));
put_ebml_float(dyn_cp, MATROSKA_ID_VIDEOCOLOR_LUMINANCEMIN,
av_q2d(metadata->min_luminance));
}
end_ebml_master(dyn_cp, meta_element);
}
colorinfo_size = avio_get_dyn_buf(dyn_cp, &colorinfo_ptr);
if (colorinfo_size) {
ebml_master colorinfo = start_ebml_master(pb, MATROSKA_ID_VIDEOCOLOR, colorinfo_size);
avio_write(pb, colorinfo_ptr, colorinfo_size);
end_ebml_master(pb, colorinfo);
}
ffio_free_dyn_buf(&dyn_cp);
return 0;
}
static int mkv_write_video_projection(AVFormatContext *s, AVIOContext *pb,
const AVStream *st)
{
ebml_master projection;
int side_data_size = 0;
uint8_t private[20];
const AVSphericalMapping *spherical =
(const AVSphericalMapping *)av_stream_get_side_data(st, AV_PKT_DATA_SPHERICAL,
&side_data_size);
if (!side_data_size)
return 0;
if (spherical->projection != AV_SPHERICAL_EQUIRECTANGULAR &&
spherical->projection != AV_SPHERICAL_EQUIRECTANGULAR_TILE &&
spherical->projection != AV_SPHERICAL_CUBEMAP) {
av_log(s, AV_LOG_WARNING, "Unknown projection type\n");
return 0;
}
// Maximally 4 8-byte elements with id-length 2 + 1 byte length field
// and the private data of the AV_SPHERICAL_EQUIRECTANGULAR_TILE case
projection = start_ebml_master(pb, MATROSKA_ID_VIDEOPROJECTION,
4 * (2 + 1 + 8) + (2 + 1 + 20));
switch (spherical->projection) {
case AV_SPHERICAL_EQUIRECTANGULAR:
put_ebml_uint(pb, MATROSKA_ID_VIDEOPROJECTIONTYPE,
MATROSKA_VIDEO_PROJECTION_TYPE_EQUIRECTANGULAR);
break;
case AV_SPHERICAL_EQUIRECTANGULAR_TILE:
put_ebml_uint(pb, MATROSKA_ID_VIDEOPROJECTIONTYPE,
MATROSKA_VIDEO_PROJECTION_TYPE_EQUIRECTANGULAR);
AV_WB32(private, 0); // version + flags
AV_WB32(private + 4, spherical->bound_top);
AV_WB32(private + 8, spherical->bound_bottom);
AV_WB32(private + 12, spherical->bound_left);
AV_WB32(private + 16, spherical->bound_right);
put_ebml_binary(pb, MATROSKA_ID_VIDEOPROJECTIONPRIVATE,
private, 20);
break;
case AV_SPHERICAL_CUBEMAP:
put_ebml_uint(pb, MATROSKA_ID_VIDEOPROJECTIONTYPE,
MATROSKA_VIDEO_PROJECTION_TYPE_CUBEMAP);
AV_WB32(private, 0); // version + flags
AV_WB32(private + 4, 0); // layout
AV_WB32(private + 8, spherical->padding);
put_ebml_binary(pb, MATROSKA_ID_VIDEOPROJECTIONPRIVATE,
private, 12);
break;
default:
av_assert0(0);
}
if (spherical->yaw)
put_ebml_float(pb, MATROSKA_ID_VIDEOPROJECTIONPOSEYAW,
(double) spherical->yaw / (1 << 16));
if (spherical->pitch)
put_ebml_float(pb, MATROSKA_ID_VIDEOPROJECTIONPOSEPITCH,
(double) spherical->pitch / (1 << 16));
if (spherical->roll)
put_ebml_float(pb, MATROSKA_ID_VIDEOPROJECTIONPOSEROLL,
(double) spherical->roll / (1 << 16));
end_ebml_master(pb, projection);
return 0;
}
static void mkv_write_field_order(AVIOContext *pb, int mode,
enum AVFieldOrder field_order)
{
switch (field_order) {
case AV_FIELD_UNKNOWN:
break;
case AV_FIELD_PROGRESSIVE:
put_ebml_uint(pb, MATROSKA_ID_VIDEOFLAGINTERLACED,
MATROSKA_VIDEO_INTERLACE_FLAG_PROGRESSIVE);
break;
case AV_FIELD_TT:
case AV_FIELD_BB:
case AV_FIELD_TB:
case AV_FIELD_BT:
put_ebml_uint(pb, MATROSKA_ID_VIDEOFLAGINTERLACED,
MATROSKA_VIDEO_INTERLACE_FLAG_INTERLACED);
if (mode != MODE_WEBM) {
switch (field_order) {
case AV_FIELD_TT:
put_ebml_uint(pb, MATROSKA_ID_VIDEOFIELDORDER,
MATROSKA_VIDEO_FIELDORDER_TT);
break;
case AV_FIELD_BB:
put_ebml_uint(pb, MATROSKA_ID_VIDEOFIELDORDER,
MATROSKA_VIDEO_FIELDORDER_BB);
break;
case AV_FIELD_TB:
put_ebml_uint(pb, MATROSKA_ID_VIDEOFIELDORDER,
MATROSKA_VIDEO_FIELDORDER_TB);
break;
case AV_FIELD_BT:
put_ebml_uint(pb, MATROSKA_ID_VIDEOFIELDORDER,
MATROSKA_VIDEO_FIELDORDER_BT);
break;
}
}
}
}
static int mkv_write_stereo_mode(AVFormatContext *s, AVIOContext *pb,
AVStream *st, int mode, int *h_width, int *h_height)
{
int i;
int ret = 0;
const AVDictionaryEntry *tag;
MatroskaVideoStereoModeType format = MATROSKA_VIDEO_STEREOMODE_TYPE_NB;
*h_width = 1;
*h_height = 1;
// convert metadata into proper side data and add it to the stream
if ((tag = av_dict_get(st->metadata, "stereo_mode", NULL, 0)) ||
(tag = av_dict_get( s->metadata, "stereo_mode", NULL, 0))) {
int stereo_mode = atoi(tag->value);
for (i=0; i<MATROSKA_VIDEO_STEREOMODE_TYPE_NB; i++)
if (!strcmp(tag->value, ff_matroska_video_stereo_mode[i])){
stereo_mode = i;
break;
}
if (stereo_mode < MATROSKA_VIDEO_STEREOMODE_TYPE_NB &&
stereo_mode != 10 && stereo_mode != 12) {
int ret = ff_mkv_stereo3d_conv(st, stereo_mode);
if (ret < 0)
return ret;
}
}
// iterate to find the stereo3d side data
for (i = 0; i < st->nb_side_data; i++) {
AVPacketSideData sd = st->side_data[i];
if (sd.type == AV_PKT_DATA_STEREO3D) {
AVStereo3D *stereo = (AVStereo3D *)sd.data;
switch (stereo->type) {
case AV_STEREO3D_2D:
format = MATROSKA_VIDEO_STEREOMODE_TYPE_MONO;
break;
case AV_STEREO3D_SIDEBYSIDE:
format = (stereo->flags & AV_STEREO3D_FLAG_INVERT)
? MATROSKA_VIDEO_STEREOMODE_TYPE_RIGHT_LEFT
: MATROSKA_VIDEO_STEREOMODE_TYPE_LEFT_RIGHT;
*h_width = 2;
break;
case AV_STEREO3D_TOPBOTTOM:
format = MATROSKA_VIDEO_STEREOMODE_TYPE_TOP_BOTTOM;
if (stereo->flags & AV_STEREO3D_FLAG_INVERT)
format--;
*h_height = 2;
break;
case AV_STEREO3D_CHECKERBOARD:
format = MATROSKA_VIDEO_STEREOMODE_TYPE_CHECKERBOARD_LR;
if (stereo->flags & AV_STEREO3D_FLAG_INVERT)
format--;
break;
case AV_STEREO3D_LINES:
format = MATROSKA_VIDEO_STEREOMODE_TYPE_ROW_INTERLEAVED_LR;
if (stereo->flags & AV_STEREO3D_FLAG_INVERT)
format--;
*h_height = 2;
break;
case AV_STEREO3D_COLUMNS:
format = MATROSKA_VIDEO_STEREOMODE_TYPE_COL_INTERLEAVED_LR;
if (stereo->flags & AV_STEREO3D_FLAG_INVERT)
format--;
*h_width = 2;
break;
case AV_STEREO3D_FRAMESEQUENCE:
format = MATROSKA_VIDEO_STEREOMODE_TYPE_BOTH_EYES_BLOCK_LR;
if (stereo->flags & AV_STEREO3D_FLAG_INVERT)
format++;
break;
}
break;
}
}
if (format == MATROSKA_VIDEO_STEREOMODE_TYPE_NB)
return ret;
// if webm, do not write unsupported modes
if ((mode == MODE_WEBM &&
format > MATROSKA_VIDEO_STEREOMODE_TYPE_TOP_BOTTOM &&
format != MATROSKA_VIDEO_STEREOMODE_TYPE_RIGHT_LEFT)
|| format >= MATROSKA_VIDEO_STEREOMODE_TYPE_NB) {
av_log(s, AV_LOG_ERROR,
"The specified stereo mode is not valid.\n");
format = MATROSKA_VIDEO_STEREOMODE_TYPE_NB;
return AVERROR(EINVAL);
}
// write StereoMode if format is valid
put_ebml_uint(pb, MATROSKA_ID_VIDEOSTEREOMODE, format);
return ret;
}
static int mkv_write_track(AVFormatContext *s, MatroskaMuxContext *mkv,
AVStream *st, mkv_track *track, AVIOContext *pb,
int is_default)
{
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.
11 years ago
AVCodecParameters *par = st->codecpar;
ebml_master subinfo, track_master;
int native_id = 0;
int qt_id = 0;
int bit_depth;
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.
11 years ago
int sample_rate = par->sample_rate;
int output_sample_rate = 0;
int display_width_div = 1;
int display_height_div = 1;
int j, ret;
const AVDictionaryEntry *tag;
if (par->codec_type == AVMEDIA_TYPE_ATTACHMENT)
return 0;
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.
11 years ago
if (par->codec_id == AV_CODEC_ID_AAC) {
ret = get_aac_sample_rates(s, par->extradata, par->extradata_size, &sample_rate,
&output_sample_rate);
if (ret < 0)
return ret;
}
track_master = start_ebml_master(pb, MATROSKA_ID_TRACKENTRY, 0);
put_ebml_uint(pb, MATROSKA_ID_TRACKNUMBER, track->track_num);
put_ebml_uid (pb, MATROSKA_ID_TRACKUID, track->uid);
put_ebml_uint(pb, MATROSKA_ID_TRACKFLAGLACING, 0); // no lacing (yet)
if ((tag = av_dict_get(st->metadata, "title", NULL, 0)))
put_ebml_string(pb, MATROSKA_ID_TRACKNAME, tag->value);
tag = av_dict_get(st->metadata, "language", NULL, 0);
put_ebml_string(pb, MATROSKA_ID_TRACKLANGUAGE,
tag && tag->value ? tag->value : "und");
// The default value for TRACKFLAGDEFAULT is 1, so add element
// if we need to clear it.
if (!is_default)
put_ebml_uint(pb, MATROSKA_ID_TRACKFLAGDEFAULT, 0);
if (st->disposition & AV_DISPOSITION_FORCED)
put_ebml_uint(pb, MATROSKA_ID_TRACKFLAGFORCED, 1);
if (mkv->mode == MODE_WEBM) {
const char *codec_id;
if (par->codec_type != AVMEDIA_TYPE_SUBTITLE) {
for (j = 0; ff_webm_codec_tags[j].id != AV_CODEC_ID_NONE; j++) {
if (ff_webm_codec_tags[j].id == par->codec_id) {
codec_id = ff_webm_codec_tags[j].str;
native_id = 1;
break;
}
}
} else if (par->codec_id == AV_CODEC_ID_WEBVTT) {
if (st->disposition & AV_DISPOSITION_CAPTIONS) {
codec_id = "D_WEBVTT/CAPTIONS";
native_id = MATROSKA_TRACK_TYPE_SUBTITLE;
} else if (st->disposition & AV_DISPOSITION_DESCRIPTIONS) {
codec_id = "D_WEBVTT/DESCRIPTIONS";
native_id = MATROSKA_TRACK_TYPE_METADATA;
} else if (st->disposition & AV_DISPOSITION_METADATA) {
codec_id = "D_WEBVTT/METADATA";
native_id = MATROSKA_TRACK_TYPE_METADATA;
} else {
codec_id = "D_WEBVTT/SUBTITLES";
native_id = MATROSKA_TRACK_TYPE_SUBTITLE;
}
}
if (!native_id) {
av_log(s, AV_LOG_ERROR,
"Only VP8 or VP9 or AV1 video and Vorbis or Opus audio and WebVTT subtitles are supported for WebM.\n");
return AVERROR(EINVAL);
}
put_ebml_string(pb, MATROSKA_ID_CODECID, codec_id);
} else {
// look for a codec ID string specific to mkv to use,
// if none are found, use AVI codes
if (par->codec_id != AV_CODEC_ID_RAWVIDEO || par->codec_tag) {
for (j = 0; ff_mkv_codec_tags[j].id != AV_CODEC_ID_NONE; j++) {
if (ff_mkv_codec_tags[j].id == par->codec_id && par->codec_id != AV_CODEC_ID_FFV1) {
put_ebml_string(pb, MATROSKA_ID_CODECID, ff_mkv_codec_tags[j].str);
native_id = 1;
break;
}
}
} else {
if (mkv->allow_raw_vfw) {
native_id = 0;
} else {
av_log(s, AV_LOG_ERROR, "Raw RGB is not supported Natively in Matroska, you can use AVI or NUT or\n"
"If you would like to store it anyway using VFW mode, enable allow_raw_vfw (-allow_raw_vfw 1)\n");
return AVERROR(EINVAL);
}
}
}
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.
11 years ago
switch (par->codec_type) {
case AVMEDIA_TYPE_VIDEO:
mkv->have_video = 1;
put_ebml_uint(pb, MATROSKA_ID_TRACKTYPE, MATROSKA_TRACK_TYPE_VIDEO);
if( st->avg_frame_rate.num > 0 && st->avg_frame_rate.den > 0
&& av_cmp_q(av_inv_q(st->avg_frame_rate), st->time_base) > 0)
put_ebml_uint(pb, MATROSKA_ID_TRACKDEFAULTDURATION, 1000000000LL * st->avg_frame_rate.den / st->avg_frame_rate.num);
else if( st->r_frame_rate.num > 0 && st->r_frame_rate.den > 0
&& av_cmp_q(av_inv_q(st->r_frame_rate), st->time_base) > 0)
put_ebml_uint(pb, MATROSKA_ID_TRACKDEFAULTDURATION, 1000000000LL * st->r_frame_rate.den / st->r_frame_rate.num);
if (!native_id &&
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.
11 years ago
ff_codec_get_tag(ff_codec_movvideo_tags, par->codec_id) &&
((!ff_codec_get_tag(ff_codec_bmp_tags, par->codec_id) && par->codec_id != AV_CODEC_ID_RAWVIDEO) ||
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.
11 years ago
par->codec_id == AV_CODEC_ID_SVQ1 ||
par->codec_id == AV_CODEC_ID_SVQ3 ||
par->codec_id == AV_CODEC_ID_CINEPAK))
qt_id = 1;
if (qt_id)
put_ebml_string(pb, MATROSKA_ID_CODECID, "V_QUICKTIME");
else if (!native_id) {
// if there is no mkv-specific codec ID, use VFW mode
put_ebml_string(pb, MATROSKA_ID_CODECID, "V_MS/VFW/FOURCC");
track->write_dts = 1;
s->internal->avoid_negative_ts_use_pts = 0;
}
subinfo = start_ebml_master(pb, MATROSKA_ID_TRACKVIDEO, 0);
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.
11 years ago
put_ebml_uint (pb, MATROSKA_ID_VIDEOPIXELWIDTH , par->width);
put_ebml_uint (pb, MATROSKA_ID_VIDEOPIXELHEIGHT, par->height);
mkv_write_field_order(pb, mkv->mode, par->field_order);
// check both side data and metadata for stereo information,
// write the result to the bitstream if any is found
ret = mkv_write_stereo_mode(s, pb, st, mkv->mode,
&display_width_div,
&display_height_div);
if (ret < 0)
return ret;
if (((tag = av_dict_get(st->metadata, "alpha_mode", NULL, 0)) && atoi(tag->value)) ||
((tag = av_dict_get( s->metadata, "alpha_mode", NULL, 0)) && atoi(tag->value)) ||
(par->format == AV_PIX_FMT_YUVA420P)) {
put_ebml_uint(pb, MATROSKA_ID_VIDEOALPHAMODE, 1);
}
// write DisplayWidth and DisplayHeight, they contain the size of
// a single source view and/or the display aspect ratio
if (st->sample_aspect_ratio.num) {
int64_t d_width = av_rescale(par->width, st->sample_aspect_ratio.num, st->sample_aspect_ratio.den);
if (d_width > INT_MAX) {
av_log(s, AV_LOG_ERROR, "Overflow in display width\n");
return AVERROR(EINVAL);
}
if (d_width != par->width || display_width_div != 1 || display_height_div != 1) {
if (mkv->mode == MODE_WEBM || display_width_div != 1 || display_height_div != 1) {
put_ebml_uint(pb, MATROSKA_ID_VIDEODISPLAYWIDTH , d_width / display_width_div);
put_ebml_uint(pb, MATROSKA_ID_VIDEODISPLAYHEIGHT, par->height / display_height_div);
} else {
AVRational display_aspect_ratio;
av_reduce(&display_aspect_ratio.num, &display_aspect_ratio.den,
par->width * (int64_t)st->sample_aspect_ratio.num,
par->height * (int64_t)st->sample_aspect_ratio.den,
1024 * 1024);
put_ebml_uint(pb, MATROSKA_ID_VIDEODISPLAYWIDTH, display_aspect_ratio.num);
put_ebml_uint(pb, MATROSKA_ID_VIDEODISPLAYHEIGHT, display_aspect_ratio.den);
put_ebml_uint(pb, MATROSKA_ID_VIDEODISPLAYUNIT, MATROSKA_VIDEO_DISPLAYUNIT_DAR);
}
}
} else if (display_width_div != 1 || display_height_div != 1) {
put_ebml_uint(pb, MATROSKA_ID_VIDEODISPLAYWIDTH , par->width / display_width_div);
put_ebml_uint(pb, MATROSKA_ID_VIDEODISPLAYHEIGHT, par->height / display_height_div);
} else if (mkv->mode != MODE_WEBM)
put_ebml_uint(pb, MATROSKA_ID_VIDEODISPLAYUNIT, MATROSKA_VIDEO_DISPLAYUNIT_UNKNOWN);
if (par->codec_id == AV_CODEC_ID_RAWVIDEO) {
uint32_t color_space = av_le2ne32(par->codec_tag);
put_ebml_binary(pb, MATROSKA_ID_VIDEOCOLORSPACE, &color_space, sizeof(color_space));
}
ret = mkv_write_video_color(pb, par, st);
if (ret < 0)
return ret;
ret = mkv_write_video_projection(s, pb, st);
if (ret < 0)
return ret;
end_ebml_master(pb, subinfo);
break;
case AVMEDIA_TYPE_AUDIO:
if (par->initial_padding && par->codec_id == AV_CODEC_ID_OPUS) {
int64_t codecdelay = av_rescale_q(par->initial_padding,
(AVRational){ 1, 48000 },
(AVRational){ 1, 1000000000 });
if (codecdelay < 0) {
av_log(s, AV_LOG_ERROR, "Initial padding is invalid\n");
return AVERROR(EINVAL);
}
// track->ts_offset = av_rescale_q(par->initial_padding,
// (AVRational){ 1, par->sample_rate },
// st->time_base);
put_ebml_uint(pb, MATROSKA_ID_CODECDELAY, codecdelay);
}
if (par->codec_id == AV_CODEC_ID_OPUS)
put_ebml_uint(pb, MATROSKA_ID_SEEKPREROLL, OPUS_SEEK_PREROLL);
put_ebml_uint(pb, MATROSKA_ID_TRACKTYPE, MATROSKA_TRACK_TYPE_AUDIO);
if (!native_id)
// no mkv-specific ID, use ACM mode
put_ebml_string(pb, MATROSKA_ID_CODECID, "A_MS/ACM");
subinfo = start_ebml_master(pb, MATROSKA_ID_TRACKAUDIO, 6 + 4 * 9);
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.
11 years ago
put_ebml_uint (pb, MATROSKA_ID_AUDIOCHANNELS , par->channels);
track->sample_rate_offset = avio_tell(pb);
put_ebml_float (pb, MATROSKA_ID_AUDIOSAMPLINGFREQ, sample_rate);
if (output_sample_rate)
put_ebml_float(pb, MATROSKA_ID_AUDIOOUTSAMPLINGFREQ, output_sample_rate);
bit_depth = av_get_bits_per_sample(par->codec_id);
if (!bit_depth && par->codec_id != AV_CODEC_ID_ADPCM_G726) {
if (par->bits_per_raw_sample)
bit_depth = par->bits_per_raw_sample;
else
bit_depth = av_get_bytes_per_sample(par->format) << 3;
}
if (!bit_depth)
bit_depth = par->bits_per_coded_sample;
if (bit_depth)
put_ebml_uint(pb, MATROSKA_ID_AUDIOBITDEPTH, bit_depth);
end_ebml_master(pb, subinfo);
break;
case AVMEDIA_TYPE_SUBTITLE:
if (!native_id) {
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.
11 years ago
av_log(s, AV_LOG_ERROR, "Subtitle codec %d is not supported.\n", par->codec_id);
return AVERROR(ENOSYS);
}
if (mkv->mode != MODE_WEBM || par->codec_id != AV_CODEC_ID_WEBVTT)
native_id = MATROSKA_TRACK_TYPE_SUBTITLE;
put_ebml_uint(pb, MATROSKA_ID_TRACKTYPE, native_id);
break;
default:
av_log(s, AV_LOG_ERROR, "Only audio, video, and subtitles are supported for Matroska.\n");
return AVERROR(EINVAL);
}
if (mkv->mode != MODE_WEBM || par->codec_id != AV_CODEC_ID_WEBVTT) {
track->codecpriv_offset = avio_tell(pb);
ret = mkv_write_codecprivate(s, pb, par, native_id, qt_id);
if (ret < 0)
return ret;
}
end_ebml_master(pb, track_master);
return 0;
}
static int mkv_write_tracks(AVFormatContext *s)
{
MatroskaMuxContext *mkv = s->priv_data;
AVIOContext *pb = s->pb;
int i, ret, video_default_idx, audio_default_idx, subtitle_default_idx;
if (mkv->nb_attachments == s->nb_streams)
return 0;
avformat/matroskaenc: Avoid allocations for SeekHead Up until e7ddafd5, the Matroska muxer wrote two SeekHeads: One at the beginning referencing the main level 1 elements (i.e. not the Clusters) and one at the end, referencing the Clusters. This second SeekHead was useless and has therefore been removed. Yet the SeekHead-related functions and structures are still geared towards this usecase: They are built around an allocated array of variable size that gets reallocated every time an element is added to it although the maximum number of Seek entries is a small compile-time constant, so that one should rather include the array in the SeekHead structure itself; and said structure should be contained in the MatroskaMuxContext instead of being allocated separately. The earlier code reserved space for a SeekHead with 10 entries, although we currently write at most 6. Reducing said number implied that every Matroska/Webm file will be 84 bytes smaller and required to adapt several FATE tests; furthermore, the reserved amount overestimated the amount needed for for the SeekHead's length field and how many bytes need to be reserved to write a EBML Void element, bringing the total reduction to 89 bytes. This also fixes a potential segfault: If !mkv->is_live and if the AVIOContext is initially unseekable when writing the header, the SeekHead is already written when writing the header and this used to free the SeekHead-related structures that have been allocated. But if the AVIOContext happens to be seekable when writing the trailer, it will be attempted to write the SeekHead again which will lead to segfaults because the corresponding structures have already been freed. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
mkv_add_seekhead_entry(mkv, MATROSKA_ID_TRACKS, avio_tell(pb));
ret = start_ebml_master_crc32(&mkv->tracks_bc, mkv);
if (ret < 0)
return ret;
if (mkv->default_mode != DEFAULT_MODE_PASSTHROUGH) {
int video_idx, audio_idx, subtitle_idx;
video_idx = video_default_idx =
audio_idx = audio_default_idx =
subtitle_idx = subtitle_default_idx = -1;
for (i = s->nb_streams - 1; i >= 0; i--) {
AVStream *st = s->streams[i];
switch (st->codecpar->codec_type) {
#define CASE(type, variable) \
case AVMEDIA_TYPE_ ## type: \
variable ## _idx = i; \
if (st->disposition & AV_DISPOSITION_DEFAULT) \
variable ## _default_idx = i; \
break;
CASE(VIDEO, video)
CASE(AUDIO, audio)
CASE(SUBTITLE, subtitle)
#undef CASE
}
}
video_default_idx = FFMAX(video_default_idx, video_idx);
audio_default_idx = FFMAX(audio_default_idx, audio_idx);
if (mkv->default_mode != DEFAULT_MODE_INFER_NO_SUBS)
subtitle_default_idx = FFMAX(subtitle_default_idx, subtitle_idx);
}
for (i = 0; i < s->nb_streams; i++) {
AVStream *st = s->streams[i];
int is_default = mkv->default_mode == DEFAULT_MODE_PASSTHROUGH ?
st->disposition & AV_DISPOSITION_DEFAULT :
i == video_default_idx || i == audio_default_idx ||
i == subtitle_default_idx;
ret = mkv_write_track(s, mkv, st, &mkv->tracks[i],
mkv->tracks_bc, is_default);
if (ret < 0)
return ret;
}
if ((pb->seekable & AVIO_SEEKABLE_NORMAL) && !mkv->is_live)
end_ebml_master_crc32_preliminary(pb, mkv->tracks_bc,
MATROSKA_ID_TRACKS, &mkv->tracks_pos);
else
avformat/matroskaenc: Stop reallocating of Cluster buffer The Matroska muxer uses a dynamic buffer to buffer the content of Clusters before eventually writing them. Up until now, each time a Cluster was written, the dynamic buffer was closed, i.e. freed; now it is only reset, saving allocations of the AVIOContext itself, its opaque as well as most of the reallocations of the buffer. This is advantageous performance-wise, in particular on systems where reallocations are slow (namely Windows). The following table shows the decicyles for writing a frame on Linux (Ubuntu 19.10) and Windows (7) on an x64 Haswell (to /dev/null on Linux, to stdout which is discarded on Windows (the default values of the size and duration of clusters for seekable output have been explicitly set in this case); in all tests, writing CRC-32 values has been disabled in all tests; calls to the muxer's write_packet function in write_packet() in libavformat/mux.c have been timed; each of the following tests has been repeated 50 times): | Windows before | Windows after | Linux before | Linux after _________________________________________________________________ A | 979437 | 192304 | 259500 | 183320 B | 715936 | 155648 | 152786 | 130879 C | 265115 | 56034 | 78496 | 53243 D | 386224 | 80307 | 128894 | 75354 E | 21732 | 10695 | 11320 | 9801 (A is a 10.2 mb/s file with a GOP length of 2s, amounting to an average Cluster size of about 2.5 MiB; the average Cluster size of B is 1.1 MiB; for C it is 2.35 MiB, for D it is 0.46 MiB; for E - a file with just a single audio track of 158kb/s resulting in a Cluster size of about 100 kB, the relative gains were the smallest, probably because of the small Cluster size.) Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
end_ebml_master_crc32(pb, &mkv->tracks_bc, mkv, MATROSKA_ID_TRACKS, 0, 0);
return 0;
}
static int mkv_write_chapters(AVFormatContext *s)
{
MatroskaMuxContext *mkv = s->priv_data;
avformat/matroskaenc: Stop reallocating of Cluster buffer The Matroska muxer uses a dynamic buffer to buffer the content of Clusters before eventually writing them. Up until now, each time a Cluster was written, the dynamic buffer was closed, i.e. freed; now it is only reset, saving allocations of the AVIOContext itself, its opaque as well as most of the reallocations of the buffer. This is advantageous performance-wise, in particular on systems where reallocations are slow (namely Windows). The following table shows the decicyles for writing a frame on Linux (Ubuntu 19.10) and Windows (7) on an x64 Haswell (to /dev/null on Linux, to stdout which is discarded on Windows (the default values of the size and duration of clusters for seekable output have been explicitly set in this case); in all tests, writing CRC-32 values has been disabled in all tests; calls to the muxer's write_packet function in write_packet() in libavformat/mux.c have been timed; each of the following tests has been repeated 50 times): | Windows before | Windows after | Linux before | Linux after _________________________________________________________________ A | 979437 | 192304 | 259500 | 183320 B | 715936 | 155648 | 152786 | 130879 C | 265115 | 56034 | 78496 | 53243 D | 386224 | 80307 | 128894 | 75354 E | 21732 | 10695 | 11320 | 9801 (A is a 10.2 mb/s file with a GOP length of 2s, amounting to an average Cluster size of about 2.5 MiB; the average Cluster size of B is 1.1 MiB; for C it is 2.35 MiB, for D it is 0.46 MiB; for E - a file with just a single audio track of 158kb/s resulting in a Cluster size of about 100 kB, the relative gains were the smallest, probably because of the small Cluster size.) Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
AVIOContext *dyn_cp = NULL, *pb = s->pb;
ebml_master editionentry;
AVRational scale = {1, 1E9};
int i, ret;
if (!s->nb_chapters || mkv->wrote_chapters)
return 0;
avformat/matroskaenc: Avoid allocations for SeekHead Up until e7ddafd5, the Matroska muxer wrote two SeekHeads: One at the beginning referencing the main level 1 elements (i.e. not the Clusters) and one at the end, referencing the Clusters. This second SeekHead was useless and has therefore been removed. Yet the SeekHead-related functions and structures are still geared towards this usecase: They are built around an allocated array of variable size that gets reallocated every time an element is added to it although the maximum number of Seek entries is a small compile-time constant, so that one should rather include the array in the SeekHead structure itself; and said structure should be contained in the MatroskaMuxContext instead of being allocated separately. The earlier code reserved space for a SeekHead with 10 entries, although we currently write at most 6. Reducing said number implied that every Matroska/Webm file will be 84 bytes smaller and required to adapt several FATE tests; furthermore, the reserved amount overestimated the amount needed for for the SeekHead's length field and how many bytes need to be reserved to write a EBML Void element, bringing the total reduction to 89 bytes. This also fixes a potential segfault: If !mkv->is_live and if the AVIOContext is initially unseekable when writing the header, the SeekHead is already written when writing the header and this used to free the SeekHead-related structures that have been allocated. But if the AVIOContext happens to be seekable when writing the trailer, it will be attempted to write the SeekHead again which will lead to segfaults because the corresponding structures have already been freed. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
mkv_add_seekhead_entry(mkv, MATROSKA_ID_CHAPTERS, avio_tell(pb));
ret = start_ebml_master_crc32(&dyn_cp, mkv);
if (ret < 0)
return ret;
editionentry = start_ebml_master(dyn_cp, MATROSKA_ID_EDITIONENTRY, 0);
if (mkv->mode != MODE_WEBM)
put_ebml_uint(dyn_cp, MATROSKA_ID_EDITIONFLAGDEFAULT, 1);
for (i = 0; i < s->nb_chapters; i++) {
ebml_master chapteratom, chapterdisplay;
const AVChapter *c = s->chapters[i];
int64_t chapterstart = av_rescale_q(c->start, c->time_base, scale);
int64_t chapterend = av_rescale_q(c->end, c->time_base, scale);
const AVDictionaryEntry *t;
if (chapterstart < 0 || chapterstart > chapterend || chapterend < 0) {
av_log(s, AV_LOG_ERROR,
"Invalid chapter start (%"PRId64") or end (%"PRId64").\n",
chapterstart, chapterend);
return AVERROR_INVALIDDATA;
}
chapteratom = start_ebml_master(dyn_cp, MATROSKA_ID_CHAPTERATOM, 0);
put_ebml_uint(dyn_cp, MATROSKA_ID_CHAPTERUID,
(uint32_t)c->id + (uint64_t)mkv->chapter_id_offset);
put_ebml_uint(dyn_cp, MATROSKA_ID_CHAPTERTIMESTART, chapterstart);
put_ebml_uint(dyn_cp, MATROSKA_ID_CHAPTERTIMEEND, chapterend);
if ((t = av_dict_get(c->metadata, "title", NULL, 0))) {
chapterdisplay = start_ebml_master(dyn_cp, MATROSKA_ID_CHAPTERDISPLAY, 0);
put_ebml_string(dyn_cp, MATROSKA_ID_CHAPSTRING, t->value);
put_ebml_string(dyn_cp, MATROSKA_ID_CHAPLANG , "und");
end_ebml_master(dyn_cp, chapterdisplay);
}
end_ebml_master(dyn_cp, chapteratom);
}
end_ebml_master(dyn_cp, editionentry);
avformat/matroskaenc: Stop reallocating of Cluster buffer The Matroska muxer uses a dynamic buffer to buffer the content of Clusters before eventually writing them. Up until now, each time a Cluster was written, the dynamic buffer was closed, i.e. freed; now it is only reset, saving allocations of the AVIOContext itself, its opaque as well as most of the reallocations of the buffer. This is advantageous performance-wise, in particular on systems where reallocations are slow (namely Windows). The following table shows the decicyles for writing a frame on Linux (Ubuntu 19.10) and Windows (7) on an x64 Haswell (to /dev/null on Linux, to stdout which is discarded on Windows (the default values of the size and duration of clusters for seekable output have been explicitly set in this case); in all tests, writing CRC-32 values has been disabled in all tests; calls to the muxer's write_packet function in write_packet() in libavformat/mux.c have been timed; each of the following tests has been repeated 50 times): | Windows before | Windows after | Linux before | Linux after _________________________________________________________________ A | 979437 | 192304 | 259500 | 183320 B | 715936 | 155648 | 152786 | 130879 C | 265115 | 56034 | 78496 | 53243 D | 386224 | 80307 | 128894 | 75354 E | 21732 | 10695 | 11320 | 9801 (A is a 10.2 mb/s file with a GOP length of 2s, amounting to an average Cluster size of about 2.5 MiB; the average Cluster size of B is 1.1 MiB; for C it is 2.35 MiB, for D it is 0.46 MiB; for E - a file with just a single audio track of 158kb/s resulting in a Cluster size of about 100 kB, the relative gains were the smallest, probably because of the small Cluster size.) Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
end_ebml_master_crc32(pb, &dyn_cp, mkv, MATROSKA_ID_CHAPTERS, 0, 0);
mkv->wrote_chapters = 1;
return 0;
}
static int mkv_write_simpletag(AVIOContext *pb, const AVDictionaryEntry *t)
{
uint8_t *key = av_strdup(t->key);
uint8_t *p = key;
const uint8_t *lang = NULL;
ebml_master tag;
if (!key)
return AVERROR(ENOMEM);
if ((p = strrchr(p, '-')) &&
(lang = ff_convert_lang_to(p + 1, AV_LANG_ISO639_2_BIBL)))
*p = 0;
p = key;
while (*p) {
if (*p == ' ')
*p = '_';
else if (*p >= 'a' && *p <= 'z')
*p -= 'a' - 'A';
p++;
}
tag = start_ebml_master(pb, MATROSKA_ID_SIMPLETAG, 0);
put_ebml_string(pb, MATROSKA_ID_TAGNAME, key);
if (lang)
put_ebml_string(pb, MATROSKA_ID_TAGLANG, lang);
put_ebml_string(pb, MATROSKA_ID_TAGSTRING, t->value);
end_ebml_master(pb, tag);
av_freep(&key);
return 0;
}
static int mkv_write_tag_targets(AVFormatContext *s, uint32_t elementid,
uint64_t uid, ebml_master *tag)
{
AVIOContext *pb;
MatroskaMuxContext *mkv = s->priv_data;
ebml_master targets;
int ret;
if (!mkv->tags_bc) {
avformat/matroskaenc: Avoid allocations for SeekHead Up until e7ddafd5, the Matroska muxer wrote two SeekHeads: One at the beginning referencing the main level 1 elements (i.e. not the Clusters) and one at the end, referencing the Clusters. This second SeekHead was useless and has therefore been removed. Yet the SeekHead-related functions and structures are still geared towards this usecase: They are built around an allocated array of variable size that gets reallocated every time an element is added to it although the maximum number of Seek entries is a small compile-time constant, so that one should rather include the array in the SeekHead structure itself; and said structure should be contained in the MatroskaMuxContext instead of being allocated separately. The earlier code reserved space for a SeekHead with 10 entries, although we currently write at most 6. Reducing said number implied that every Matroska/Webm file will be 84 bytes smaller and required to adapt several FATE tests; furthermore, the reserved amount overestimated the amount needed for for the SeekHead's length field and how many bytes need to be reserved to write a EBML Void element, bringing the total reduction to 89 bytes. This also fixes a potential segfault: If !mkv->is_live and if the AVIOContext is initially unseekable when writing the header, the SeekHead is already written when writing the header and this used to free the SeekHead-related structures that have been allocated. But if the AVIOContext happens to be seekable when writing the trailer, it will be attempted to write the SeekHead again which will lead to segfaults because the corresponding structures have already been freed. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
mkv_add_seekhead_entry(mkv, MATROSKA_ID_TAGS, avio_tell(s->pb));
ret = start_ebml_master_crc32(&mkv->tags_bc, mkv);
if (ret < 0)
return ret;
}
pb = mkv->tags_bc;
*tag = start_ebml_master(pb, MATROSKA_ID_TAG, 0);
targets = start_ebml_master(pb, MATROSKA_ID_TAGTARGETS, 4 + 1 + 8);
if (elementid)
put_ebml_uid(pb, elementid, uid);
end_ebml_master(pb, targets);
return 0;
}
static int mkv_check_tag_name(const char *name, uint32_t elementid)
{
return av_strcasecmp(name, "title") &&
av_strcasecmp(name, "stereo_mode") &&
av_strcasecmp(name, "creation_time") &&
av_strcasecmp(name, "encoding_tool") &&
av_strcasecmp(name, "duration") &&
(elementid != MATROSKA_ID_TAGTARGETS_TRACKUID ||
av_strcasecmp(name, "language")) &&
(elementid != MATROSKA_ID_TAGTARGETS_ATTACHUID ||
(av_strcasecmp(name, "filename") &&
av_strcasecmp(name, "mimetype")));
}
static int mkv_write_tag(AVFormatContext *s, const AVDictionary *m,
uint32_t elementid, uint64_t uid, ebml_master *tag)
{
MatroskaMuxContext *mkv = s->priv_data;
const AVDictionaryEntry *t = NULL;
ebml_master tag2;
int ret;
ret = mkv_write_tag_targets(s, elementid, uid, tag ? tag : &tag2);
if (ret < 0)
return ret;
while ((t = av_dict_get(m, "", t, AV_DICT_IGNORE_SUFFIX))) {
if (mkv_check_tag_name(t->key, elementid)) {
ret = mkv_write_simpletag(mkv->tags_bc, t);
if (ret < 0)
return ret;
}
}
if (!tag)
end_ebml_master(mkv->tags_bc, tag2);
return 0;
}
static int mkv_check_tag(const AVDictionary *m, uint32_t elementid)
{
const AVDictionaryEntry *t = NULL;
while ((t = av_dict_get(m, "", t, AV_DICT_IGNORE_SUFFIX)))
if (mkv_check_tag_name(t->key, elementid))
return 1;
return 0;
}
static int mkv_write_tags(AVFormatContext *s)
{
MatroskaMuxContext *mkv = s->priv_data;
ebml_master tag, *tagp;
int i, ret;
ff_metadata_conv_ctx(s, ff_mkv_metadata_conv, NULL);
if (mkv_check_tag(s->metadata, 0)) {
ret = mkv_write_tag(s, s->metadata, 0, 0, NULL);
if (ret < 0)
return ret;
}
tagp = (s->pb->seekable & AVIO_SEEKABLE_NORMAL) && !mkv->is_live ? &tag : NULL;
for (i = 0; i < s->nb_streams; i++) {
const AVStream *st = s->streams[i];
mkv_track *track = &mkv->tracks[i];
if (st->codecpar->codec_type == AVMEDIA_TYPE_ATTACHMENT)
continue;
if (!tagp && !mkv_check_tag(st->metadata, MATROSKA_ID_TAGTARGETS_TRACKUID))
continue;
ret = mkv_write_tag(s, st->metadata, MATROSKA_ID_TAGTARGETS_TRACKUID,
track->uid, tagp);
if (ret < 0)
return ret;
if (tagp) {
AVIOContext *pb = mkv->tags_bc;
ebml_master simpletag;
simpletag = start_ebml_master(pb, MATROSKA_ID_SIMPLETAG,
2 + 1 + 8 + 23);
put_ebml_string(pb, MATROSKA_ID_TAGNAME, "DURATION");
track->duration_offset = avio_tell(pb);
// Reserve space to write duration as a 20-byte string.
// 2 (ebml id) + 1 (data size) + 20 (data)
put_ebml_void(pb, 23);
end_ebml_master(pb, simpletag);
end_ebml_master(pb, tag);
}
}
if (mkv->mode != MODE_WEBM) {
for (i = 0; i < s->nb_chapters; i++) {
AVChapter *ch = s->chapters[i];
if (!mkv_check_tag(ch->metadata, MATROSKA_ID_TAGTARGETS_CHAPTERUID))
continue;
ret = mkv_write_tag(s, ch->metadata, MATROSKA_ID_TAGTARGETS_CHAPTERUID,
(uint32_t)ch->id + (uint64_t)mkv->chapter_id_offset,
NULL);
if (ret < 0)
return ret;
}
}
if (mkv->nb_attachments && mkv->mode != MODE_WEBM) {
for (i = 0; i < s->nb_streams; i++) {
const mkv_track *track = &mkv->tracks[i];
const AVStream *st = s->streams[i];
if (st->codecpar->codec_type != AVMEDIA_TYPE_ATTACHMENT)
continue;
if (!mkv_check_tag(st->metadata, MATROSKA_ID_TAGTARGETS_ATTACHUID))
continue;
ret = mkv_write_tag(s, st->metadata, MATROSKA_ID_TAGTARGETS_ATTACHUID,
track->uid, NULL);
if (ret < 0)
return ret;
}
}
if (mkv->tags_bc) {
if ((s->pb->seekable & AVIO_SEEKABLE_NORMAL) && !mkv->is_live)
end_ebml_master_crc32_preliminary(s->pb, mkv->tags_bc,
MATROSKA_ID_TAGS, &mkv->tags_pos);
else
avformat/matroskaenc: Stop reallocating of Cluster buffer The Matroska muxer uses a dynamic buffer to buffer the content of Clusters before eventually writing them. Up until now, each time a Cluster was written, the dynamic buffer was closed, i.e. freed; now it is only reset, saving allocations of the AVIOContext itself, its opaque as well as most of the reallocations of the buffer. This is advantageous performance-wise, in particular on systems where reallocations are slow (namely Windows). The following table shows the decicyles for writing a frame on Linux (Ubuntu 19.10) and Windows (7) on an x64 Haswell (to /dev/null on Linux, to stdout which is discarded on Windows (the default values of the size and duration of clusters for seekable output have been explicitly set in this case); in all tests, writing CRC-32 values has been disabled in all tests; calls to the muxer's write_packet function in write_packet() in libavformat/mux.c have been timed; each of the following tests has been repeated 50 times): | Windows before | Windows after | Linux before | Linux after _________________________________________________________________ A | 979437 | 192304 | 259500 | 183320 B | 715936 | 155648 | 152786 | 130879 C | 265115 | 56034 | 78496 | 53243 D | 386224 | 80307 | 128894 | 75354 E | 21732 | 10695 | 11320 | 9801 (A is a 10.2 mb/s file with a GOP length of 2s, amounting to an average Cluster size of about 2.5 MiB; the average Cluster size of B is 1.1 MiB; for C it is 2.35 MiB, for D it is 0.46 MiB; for E - a file with just a single audio track of 158kb/s resulting in a Cluster size of about 100 kB, the relative gains were the smallest, probably because of the small Cluster size.) Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
end_ebml_master_crc32(s->pb, &mkv->tags_bc, mkv, MATROSKA_ID_TAGS, 0, 0);
}
return 0;
}
static int mkv_write_attachments(AVFormatContext *s)
{
MatroskaMuxContext *mkv = s->priv_data;
avformat/matroskaenc: Stop reallocating of Cluster buffer The Matroska muxer uses a dynamic buffer to buffer the content of Clusters before eventually writing them. Up until now, each time a Cluster was written, the dynamic buffer was closed, i.e. freed; now it is only reset, saving allocations of the AVIOContext itself, its opaque as well as most of the reallocations of the buffer. This is advantageous performance-wise, in particular on systems where reallocations are slow (namely Windows). The following table shows the decicyles for writing a frame on Linux (Ubuntu 19.10) and Windows (7) on an x64 Haswell (to /dev/null on Linux, to stdout which is discarded on Windows (the default values of the size and duration of clusters for seekable output have been explicitly set in this case); in all tests, writing CRC-32 values has been disabled in all tests; calls to the muxer's write_packet function in write_packet() in libavformat/mux.c have been timed; each of the following tests has been repeated 50 times): | Windows before | Windows after | Linux before | Linux after _________________________________________________________________ A | 979437 | 192304 | 259500 | 183320 B | 715936 | 155648 | 152786 | 130879 C | 265115 | 56034 | 78496 | 53243 D | 386224 | 80307 | 128894 | 75354 E | 21732 | 10695 | 11320 | 9801 (A is a 10.2 mb/s file with a GOP length of 2s, amounting to an average Cluster size of about 2.5 MiB; the average Cluster size of B is 1.1 MiB; for C it is 2.35 MiB, for D it is 0.46 MiB; for E - a file with just a single audio track of 158kb/s resulting in a Cluster size of about 100 kB, the relative gains were the smallest, probably because of the small Cluster size.) Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
AVIOContext *dyn_cp = NULL, *pb = s->pb;
int i, ret;
if (!mkv->nb_attachments)
return 0;
avformat/matroskaenc: Avoid allocations for SeekHead Up until e7ddafd5, the Matroska muxer wrote two SeekHeads: One at the beginning referencing the main level 1 elements (i.e. not the Clusters) and one at the end, referencing the Clusters. This second SeekHead was useless and has therefore been removed. Yet the SeekHead-related functions and structures are still geared towards this usecase: They are built around an allocated array of variable size that gets reallocated every time an element is added to it although the maximum number of Seek entries is a small compile-time constant, so that one should rather include the array in the SeekHead structure itself; and said structure should be contained in the MatroskaMuxContext instead of being allocated separately. The earlier code reserved space for a SeekHead with 10 entries, although we currently write at most 6. Reducing said number implied that every Matroska/Webm file will be 84 bytes smaller and required to adapt several FATE tests; furthermore, the reserved amount overestimated the amount needed for for the SeekHead's length field and how many bytes need to be reserved to write a EBML Void element, bringing the total reduction to 89 bytes. This also fixes a potential segfault: If !mkv->is_live and if the AVIOContext is initially unseekable when writing the header, the SeekHead is already written when writing the header and this used to free the SeekHead-related structures that have been allocated. But if the AVIOContext happens to be seekable when writing the trailer, it will be attempted to write the SeekHead again which will lead to segfaults because the corresponding structures have already been freed. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
mkv_add_seekhead_entry(mkv, MATROSKA_ID_ATTACHMENTS, avio_tell(pb));
ret = start_ebml_master_crc32(&dyn_cp, mkv);
if (ret < 0)
return ret;
for (i = 0; i < s->nb_streams; i++) {
const AVStream *st = s->streams[i];
mkv_track *track = &mkv->tracks[i];
ebml_master attached_file;
const AVDictionaryEntry *t;
const char *mimetype = NULL;
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.
11 years ago
if (st->codecpar->codec_type != AVMEDIA_TYPE_ATTACHMENT)
continue;
attached_file = start_ebml_master(dyn_cp, MATROSKA_ID_ATTACHEDFILE, 0);
if (t = av_dict_get(st->metadata, "title", NULL, 0))
put_ebml_string(dyn_cp, MATROSKA_ID_FILEDESC, t->value);
if (!(t = av_dict_get(st->metadata, "filename", NULL, 0))) {
av_log(s, AV_LOG_ERROR, "Attachment stream %d has no filename tag.\n", i);
return AVERROR(EINVAL);
}
put_ebml_string(dyn_cp, MATROSKA_ID_FILENAME, t->value);
if (t = av_dict_get(st->metadata, "mimetype", NULL, 0))
mimetype = t->value;
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.
11 years ago
else if (st->codecpar->codec_id != AV_CODEC_ID_NONE ) {
const AVCodecDescriptor *desc = avcodec_descriptor_get(st->codecpar->codec_id);
if (desc && desc->mime_types) {
mimetype = desc->mime_types[0];
} else if (st->codecpar->codec_id == AV_CODEC_ID_TEXT)
mimetype = "text/plain";
}
if (!mimetype) {
av_log(s, AV_LOG_ERROR, "Attachment stream %d has no mimetype tag and "
"it cannot be deduced from the codec id.\n", i);
return AVERROR(EINVAL);
}
put_ebml_string(dyn_cp, MATROSKA_ID_FILEMIMETYPE, mimetype);
put_ebml_binary(dyn_cp, MATROSKA_ID_FILEDATA, st->codecpar->extradata, st->codecpar->extradata_size);
put_ebml_uid(dyn_cp, MATROSKA_ID_FILEUID, track->uid);
end_ebml_master(dyn_cp, attached_file);
}
avformat/matroskaenc: Stop reallocating of Cluster buffer The Matroska muxer uses a dynamic buffer to buffer the content of Clusters before eventually writing them. Up until now, each time a Cluster was written, the dynamic buffer was closed, i.e. freed; now it is only reset, saving allocations of the AVIOContext itself, its opaque as well as most of the reallocations of the buffer. This is advantageous performance-wise, in particular on systems where reallocations are slow (namely Windows). The following table shows the decicyles for writing a frame on Linux (Ubuntu 19.10) and Windows (7) on an x64 Haswell (to /dev/null on Linux, to stdout which is discarded on Windows (the default values of the size and duration of clusters for seekable output have been explicitly set in this case); in all tests, writing CRC-32 values has been disabled in all tests; calls to the muxer's write_packet function in write_packet() in libavformat/mux.c have been timed; each of the following tests has been repeated 50 times): | Windows before | Windows after | Linux before | Linux after _________________________________________________________________ A | 979437 | 192304 | 259500 | 183320 B | 715936 | 155648 | 152786 | 130879 C | 265115 | 56034 | 78496 | 53243 D | 386224 | 80307 | 128894 | 75354 E | 21732 | 10695 | 11320 | 9801 (A is a 10.2 mb/s file with a GOP length of 2s, amounting to an average Cluster size of about 2.5 MiB; the average Cluster size of B is 1.1 MiB; for C it is 2.35 MiB, for D it is 0.46 MiB; for E - a file with just a single audio track of 158kb/s resulting in a Cluster size of about 100 kB, the relative gains were the smallest, probably because of the small Cluster size.) Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
end_ebml_master_crc32(pb, &dyn_cp, mkv, MATROSKA_ID_ATTACHMENTS, 0, 0);
return 0;
}
avformat/matroskaenc: Write duration early during mkv_write_header (Rev #3) Rev #2: Fixes doubled header writing, checked FATE running without errors Rev #3: Fixed coding style This commit addresses the following scenario: we are using ffmpeg to transcode or remux mkv (or something else) to mkv. The result is being streamed on-the-fly to an HTML5 client (streaming starts while ffmpeg is still running). The problem here is that the client is unable to detect the duration because the duration is only written to the mkv at the end of the transcoding/remoxing process. In matroskaenc.c, the duration is only written during mkv_write_trailer but not during mkv_write_header. The approach: FFMPEG is currently putting quite some effort to estimate the durations of source streams, but in many cases the source stream durations are still left at 0 and these durations are nowhere mapped to or used for output streams. As much as I would have liked to deduct or estimate output durations based on input stream durations - I realized that this is a hard task (as Nicolas already mentioned in a previous conversation). It would involve changes to the duration calculation/estimation/deduction for input streams and propagating these durations to output streams or the output context in a correct way. So I looked for a simple and small solution with better chances to get accepted. In webmdashenc.c I found that a duration is written during write_header and this duration is taken from the streams' metadata, so I decided for a similar approach. And here's what it does: At first it is checking the duration of the AVFormatContext. In typical cases this value is not set, but: It is set in cases where the user has specified a recording_time or an end_time via the -t or -to parameters. Then it is looking for a DURATION metadata field in the metadata of the output context (AVFormatContext::metadata). This would only exist in case the user has explicitly specified a metadata DURATION value from the command line. Then it is iterating all streams looking for a "DURATION" metadata (this works unless the option "-map_metadata -1" has been specified) and determines the maximum value. The precendence is as follows: 1. Use duration of AVFormatContext - 2. Use explicitly specified metadata duration value - 3. Use maximum (mapped) metadata duration over all streams. To test this: 1. With explicit recording time: ffmpeg -i file:"src.mkv" -loglevel debug -t 01:38:36.000 -y "dest.mkv" 2. Take duration from metadata specified via command line parameters: ffmpeg -i file:"src.mkv" -loglevel debug -map_metadata -1 -metadata Duration="01:14:33.00" -y "dest.mkv" 3. Take duration from mapped input metadata: ffmpeg -i file:"src.mkv" -loglevel debug -y "dest.mkv" Regression risk: Very low IMO because it only affects the header while ffmpeg is still running. When ffmpeg completes the process, the duration is rewritten to the header with the usual value (same like without this commit). Signed-off-by: SoftWorkz <softworkz@hotmail.com> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
9 years ago
static int64_t get_metadata_duration(AVFormatContext *s)
{
const AVDictionaryEntry *duration = av_dict_get(s->metadata, "DURATION",
NULL, 0);
avformat/matroskaenc: Write duration early during mkv_write_header (Rev #3) Rev #2: Fixes doubled header writing, checked FATE running without errors Rev #3: Fixed coding style This commit addresses the following scenario: we are using ffmpeg to transcode or remux mkv (or something else) to mkv. The result is being streamed on-the-fly to an HTML5 client (streaming starts while ffmpeg is still running). The problem here is that the client is unable to detect the duration because the duration is only written to the mkv at the end of the transcoding/remoxing process. In matroskaenc.c, the duration is only written during mkv_write_trailer but not during mkv_write_header. The approach: FFMPEG is currently putting quite some effort to estimate the durations of source streams, but in many cases the source stream durations are still left at 0 and these durations are nowhere mapped to or used for output streams. As much as I would have liked to deduct or estimate output durations based on input stream durations - I realized that this is a hard task (as Nicolas already mentioned in a previous conversation). It would involve changes to the duration calculation/estimation/deduction for input streams and propagating these durations to output streams or the output context in a correct way. So I looked for a simple and small solution with better chances to get accepted. In webmdashenc.c I found that a duration is written during write_header and this duration is taken from the streams' metadata, so I decided for a similar approach. And here's what it does: At first it is checking the duration of the AVFormatContext. In typical cases this value is not set, but: It is set in cases where the user has specified a recording_time or an end_time via the -t or -to parameters. Then it is looking for a DURATION metadata field in the metadata of the output context (AVFormatContext::metadata). This would only exist in case the user has explicitly specified a metadata DURATION value from the command line. Then it is iterating all streams looking for a "DURATION" metadata (this works unless the option "-map_metadata -1" has been specified) and determines the maximum value. The precendence is as follows: 1. Use duration of AVFormatContext - 2. Use explicitly specified metadata duration value - 3. Use maximum (mapped) metadata duration over all streams. To test this: 1. With explicit recording time: ffmpeg -i file:"src.mkv" -loglevel debug -t 01:38:36.000 -y "dest.mkv" 2. Take duration from metadata specified via command line parameters: ffmpeg -i file:"src.mkv" -loglevel debug -map_metadata -1 -metadata Duration="01:14:33.00" -y "dest.mkv" 3. Take duration from mapped input metadata: ffmpeg -i file:"src.mkv" -loglevel debug -y "dest.mkv" Regression risk: Very low IMO because it only affects the header while ffmpeg is still running. When ffmpeg completes the process, the duration is rewritten to the header with the usual value (same like without this commit). Signed-off-by: SoftWorkz <softworkz@hotmail.com> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
9 years ago
int64_t max = 0;
int64_t us;
if (duration && (av_parse_time(&us, duration->value, 1) == 0) && us > 0) {
avformat/matroskaenc: Write duration early during mkv_write_header (Rev #3) Rev #2: Fixes doubled header writing, checked FATE running without errors Rev #3: Fixed coding style This commit addresses the following scenario: we are using ffmpeg to transcode or remux mkv (or something else) to mkv. The result is being streamed on-the-fly to an HTML5 client (streaming starts while ffmpeg is still running). The problem here is that the client is unable to detect the duration because the duration is only written to the mkv at the end of the transcoding/remoxing process. In matroskaenc.c, the duration is only written during mkv_write_trailer but not during mkv_write_header. The approach: FFMPEG is currently putting quite some effort to estimate the durations of source streams, but in many cases the source stream durations are still left at 0 and these durations are nowhere mapped to or used for output streams. As much as I would have liked to deduct or estimate output durations based on input stream durations - I realized that this is a hard task (as Nicolas already mentioned in a previous conversation). It would involve changes to the duration calculation/estimation/deduction for input streams and propagating these durations to output streams or the output context in a correct way. So I looked for a simple and small solution with better chances to get accepted. In webmdashenc.c I found that a duration is written during write_header and this duration is taken from the streams' metadata, so I decided for a similar approach. And here's what it does: At first it is checking the duration of the AVFormatContext. In typical cases this value is not set, but: It is set in cases where the user has specified a recording_time or an end_time via the -t or -to parameters. Then it is looking for a DURATION metadata field in the metadata of the output context (AVFormatContext::metadata). This would only exist in case the user has explicitly specified a metadata DURATION value from the command line. Then it is iterating all streams looking for a "DURATION" metadata (this works unless the option "-map_metadata -1" has been specified) and determines the maximum value. The precendence is as follows: 1. Use duration of AVFormatContext - 2. Use explicitly specified metadata duration value - 3. Use maximum (mapped) metadata duration over all streams. To test this: 1. With explicit recording time: ffmpeg -i file:"src.mkv" -loglevel debug -t 01:38:36.000 -y "dest.mkv" 2. Take duration from metadata specified via command line parameters: ffmpeg -i file:"src.mkv" -loglevel debug -map_metadata -1 -metadata Duration="01:14:33.00" -y "dest.mkv" 3. Take duration from mapped input metadata: ffmpeg -i file:"src.mkv" -loglevel debug -y "dest.mkv" Regression risk: Very low IMO because it only affects the header while ffmpeg is still running. When ffmpeg completes the process, the duration is rewritten to the header with the usual value (same like without this commit). Signed-off-by: SoftWorkz <softworkz@hotmail.com> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
9 years ago
av_log(s, AV_LOG_DEBUG, "get_metadata_duration found duration in context metadata: %" PRId64 "\n", us);
return us;
}
for (unsigned i = 0; i < s->nb_streams; i++) {
avformat/matroskaenc: Write duration early during mkv_write_header (Rev #3) Rev #2: Fixes doubled header writing, checked FATE running without errors Rev #3: Fixed coding style This commit addresses the following scenario: we are using ffmpeg to transcode or remux mkv (or something else) to mkv. The result is being streamed on-the-fly to an HTML5 client (streaming starts while ffmpeg is still running). The problem here is that the client is unable to detect the duration because the duration is only written to the mkv at the end of the transcoding/remoxing process. In matroskaenc.c, the duration is only written during mkv_write_trailer but not during mkv_write_header. The approach: FFMPEG is currently putting quite some effort to estimate the durations of source streams, but in many cases the source stream durations are still left at 0 and these durations are nowhere mapped to or used for output streams. As much as I would have liked to deduct or estimate output durations based on input stream durations - I realized that this is a hard task (as Nicolas already mentioned in a previous conversation). It would involve changes to the duration calculation/estimation/deduction for input streams and propagating these durations to output streams or the output context in a correct way. So I looked for a simple and small solution with better chances to get accepted. In webmdashenc.c I found that a duration is written during write_header and this duration is taken from the streams' metadata, so I decided for a similar approach. And here's what it does: At first it is checking the duration of the AVFormatContext. In typical cases this value is not set, but: It is set in cases where the user has specified a recording_time or an end_time via the -t or -to parameters. Then it is looking for a DURATION metadata field in the metadata of the output context (AVFormatContext::metadata). This would only exist in case the user has explicitly specified a metadata DURATION value from the command line. Then it is iterating all streams looking for a "DURATION" metadata (this works unless the option "-map_metadata -1" has been specified) and determines the maximum value. The precendence is as follows: 1. Use duration of AVFormatContext - 2. Use explicitly specified metadata duration value - 3. Use maximum (mapped) metadata duration over all streams. To test this: 1. With explicit recording time: ffmpeg -i file:"src.mkv" -loglevel debug -t 01:38:36.000 -y "dest.mkv" 2. Take duration from metadata specified via command line parameters: ffmpeg -i file:"src.mkv" -loglevel debug -map_metadata -1 -metadata Duration="01:14:33.00" -y "dest.mkv" 3. Take duration from mapped input metadata: ffmpeg -i file:"src.mkv" -loglevel debug -y "dest.mkv" Regression risk: Very low IMO because it only affects the header while ffmpeg is still running. When ffmpeg completes the process, the duration is rewritten to the header with the usual value (same like without this commit). Signed-off-by: SoftWorkz <softworkz@hotmail.com> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
9 years ago
int64_t us;
duration = av_dict_get(s->streams[i]->metadata, "DURATION", NULL, 0);
avformat/matroskaenc: Write duration early during mkv_write_header (Rev #3) Rev #2: Fixes doubled header writing, checked FATE running without errors Rev #3: Fixed coding style This commit addresses the following scenario: we are using ffmpeg to transcode or remux mkv (or something else) to mkv. The result is being streamed on-the-fly to an HTML5 client (streaming starts while ffmpeg is still running). The problem here is that the client is unable to detect the duration because the duration is only written to the mkv at the end of the transcoding/remoxing process. In matroskaenc.c, the duration is only written during mkv_write_trailer but not during mkv_write_header. The approach: FFMPEG is currently putting quite some effort to estimate the durations of source streams, but in many cases the source stream durations are still left at 0 and these durations are nowhere mapped to or used for output streams. As much as I would have liked to deduct or estimate output durations based on input stream durations - I realized that this is a hard task (as Nicolas already mentioned in a previous conversation). It would involve changes to the duration calculation/estimation/deduction for input streams and propagating these durations to output streams or the output context in a correct way. So I looked for a simple and small solution with better chances to get accepted. In webmdashenc.c I found that a duration is written during write_header and this duration is taken from the streams' metadata, so I decided for a similar approach. And here's what it does: At first it is checking the duration of the AVFormatContext. In typical cases this value is not set, but: It is set in cases where the user has specified a recording_time or an end_time via the -t or -to parameters. Then it is looking for a DURATION metadata field in the metadata of the output context (AVFormatContext::metadata). This would only exist in case the user has explicitly specified a metadata DURATION value from the command line. Then it is iterating all streams looking for a "DURATION" metadata (this works unless the option "-map_metadata -1" has been specified) and determines the maximum value. The precendence is as follows: 1. Use duration of AVFormatContext - 2. Use explicitly specified metadata duration value - 3. Use maximum (mapped) metadata duration over all streams. To test this: 1. With explicit recording time: ffmpeg -i file:"src.mkv" -loglevel debug -t 01:38:36.000 -y "dest.mkv" 2. Take duration from metadata specified via command line parameters: ffmpeg -i file:"src.mkv" -loglevel debug -map_metadata -1 -metadata Duration="01:14:33.00" -y "dest.mkv" 3. Take duration from mapped input metadata: ffmpeg -i file:"src.mkv" -loglevel debug -y "dest.mkv" Regression risk: Very low IMO because it only affects the header while ffmpeg is still running. When ffmpeg completes the process, the duration is rewritten to the header with the usual value (same like without this commit). Signed-off-by: SoftWorkz <softworkz@hotmail.com> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
9 years ago
if (duration && (av_parse_time(&us, duration->value, 1) == 0))
max = FFMAX(max, us);
}
av_log(s, AV_LOG_DEBUG, "get_metadata_duration returned: %" PRId64 "\n", max);
return max;
}
static int mkv_write_header(AVFormatContext *s)
{
MatroskaMuxContext *mkv = s->priv_data;
AVIOContext *pb = s->pb;
ebml_master ebml_header;
const AVDictionaryEntry *tag;
int ret, i, version = 2;
int64_t creation_time;
if (mkv->mode != MODE_WEBM ||
av_dict_get(s->metadata, "stereo_mode", NULL, 0) ||
av_dict_get(s->metadata, "alpha_mode", NULL, 0))
version = 4;
for (i = 0; i < s->nb_streams; i++) {
if (s->streams[i]->codecpar->codec_id == AV_CODEC_ID_OPUS ||
av_dict_get(s->streams[i]->metadata, "stereo_mode", NULL, 0) ||
av_dict_get(s->streams[i]->metadata, "alpha_mode", NULL, 0))
version = 4;
}
ebml_header = start_ebml_master(pb, EBML_ID_HEADER, MAX_EBML_HEADER_SIZE);
put_ebml_uint (pb, EBML_ID_EBMLVERSION , 1);
put_ebml_uint (pb, EBML_ID_EBMLREADVERSION , 1);
put_ebml_uint (pb, EBML_ID_EBMLMAXIDLENGTH , 4);
put_ebml_uint (pb, EBML_ID_EBMLMAXSIZELENGTH , 8);
put_ebml_string(pb, EBML_ID_DOCTYPE , s->oformat->name);
put_ebml_uint (pb, EBML_ID_DOCTYPEVERSION , version);
put_ebml_uint (pb, EBML_ID_DOCTYPEREADVERSION, 2);
end_ebml_master(pb, ebml_header);
mkv->segment = start_ebml_master(pb, MATROSKA_ID_SEGMENT, 0);
mkv->segment_offset = avio_tell(pb);
// we write a seek head at the beginning to point to all other level
avformat/matroskaenc: Avoid allocations for SeekHead Up until e7ddafd5, the Matroska muxer wrote two SeekHeads: One at the beginning referencing the main level 1 elements (i.e. not the Clusters) and one at the end, referencing the Clusters. This second SeekHead was useless and has therefore been removed. Yet the SeekHead-related functions and structures are still geared towards this usecase: They are built around an allocated array of variable size that gets reallocated every time an element is added to it although the maximum number of Seek entries is a small compile-time constant, so that one should rather include the array in the SeekHead structure itself; and said structure should be contained in the MatroskaMuxContext instead of being allocated separately. The earlier code reserved space for a SeekHead with 10 entries, although we currently write at most 6. Reducing said number implied that every Matroska/Webm file will be 84 bytes smaller and required to adapt several FATE tests; furthermore, the reserved amount overestimated the amount needed for for the SeekHead's length field and how many bytes need to be reserved to write a EBML Void element, bringing the total reduction to 89 bytes. This also fixes a potential segfault: If !mkv->is_live and if the AVIOContext is initially unseekable when writing the header, the SeekHead is already written when writing the header and this used to free the SeekHead-related structures that have been allocated. But if the AVIOContext happens to be seekable when writing the trailer, it will be attempted to write the SeekHead again which will lead to segfaults because the corresponding structures have already been freed. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
// one elements (except Clusters).
mkv_start_seekhead(mkv, pb);
avformat/matroskaenc: Avoid allocations for SeekHead Up until e7ddafd5, the Matroska muxer wrote two SeekHeads: One at the beginning referencing the main level 1 elements (i.e. not the Clusters) and one at the end, referencing the Clusters. This second SeekHead was useless and has therefore been removed. Yet the SeekHead-related functions and structures are still geared towards this usecase: They are built around an allocated array of variable size that gets reallocated every time an element is added to it although the maximum number of Seek entries is a small compile-time constant, so that one should rather include the array in the SeekHead structure itself; and said structure should be contained in the MatroskaMuxContext instead of being allocated separately. The earlier code reserved space for a SeekHead with 10 entries, although we currently write at most 6. Reducing said number implied that every Matroska/Webm file will be 84 bytes smaller and required to adapt several FATE tests; furthermore, the reserved amount overestimated the amount needed for for the SeekHead's length field and how many bytes need to be reserved to write a EBML Void element, bringing the total reduction to 89 bytes. This also fixes a potential segfault: If !mkv->is_live and if the AVIOContext is initially unseekable when writing the header, the SeekHead is already written when writing the header and this used to free the SeekHead-related structures that have been allocated. But if the AVIOContext happens to be seekable when writing the trailer, it will be attempted to write the SeekHead again which will lead to segfaults because the corresponding structures have already been freed. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
mkv_add_seekhead_entry(mkv, MATROSKA_ID_INFO, avio_tell(pb));
ret = start_ebml_master_crc32(&mkv->info_bc, mkv);
if (ret < 0)
return ret;
pb = mkv->info_bc;
put_ebml_uint(pb, MATROSKA_ID_TIMECODESCALE, 1000000);
if ((tag = av_dict_get(s->metadata, "title", NULL, 0)))
put_ebml_string(pb, MATROSKA_ID_TITLE, tag->value);
if (!(s->flags & AVFMT_FLAG_BITEXACT)) {
put_ebml_string(pb, MATROSKA_ID_MUXINGAPP, LIBAVFORMAT_IDENT);
if ((tag = av_dict_get(s->metadata, "encoding_tool", NULL, 0)))
put_ebml_string(pb, MATROSKA_ID_WRITINGAPP, tag->value);
else
put_ebml_string(pb, MATROSKA_ID_WRITINGAPP, LIBAVFORMAT_IDENT);
if (mkv->mode != MODE_WEBM)
put_ebml_binary(pb, MATROSKA_ID_SEGMENTUID, mkv->segment_uid, 16);
} else {
const char *ident = "Lavf";
put_ebml_string(pb, MATROSKA_ID_MUXINGAPP , ident);
put_ebml_string(pb, MATROSKA_ID_WRITINGAPP, ident);
}
if (ff_parse_creation_time_metadata(s, &creation_time, 0) > 0) {
// Adjust time so it's relative to 2001-01-01 and convert to nanoseconds.
int64_t date_utc = (creation_time - 978307200000000LL) * 1000;
uint8_t date_utc_buf[8];
AV_WB64(date_utc_buf, date_utc);
put_ebml_binary(pb, MATROSKA_ID_DATEUTC, date_utc_buf, 8);
}
// reserve space for the duration
mkv->duration = 0;
mkv->duration_offset = avio_tell(pb);
if (!mkv->is_live) {
avformat/matroskaenc: Write duration early during mkv_write_header (Rev #3) Rev #2: Fixes doubled header writing, checked FATE running without errors Rev #3: Fixed coding style This commit addresses the following scenario: we are using ffmpeg to transcode or remux mkv (or something else) to mkv. The result is being streamed on-the-fly to an HTML5 client (streaming starts while ffmpeg is still running). The problem here is that the client is unable to detect the duration because the duration is only written to the mkv at the end of the transcoding/remoxing process. In matroskaenc.c, the duration is only written during mkv_write_trailer but not during mkv_write_header. The approach: FFMPEG is currently putting quite some effort to estimate the durations of source streams, but in many cases the source stream durations are still left at 0 and these durations are nowhere mapped to or used for output streams. As much as I would have liked to deduct or estimate output durations based on input stream durations - I realized that this is a hard task (as Nicolas already mentioned in a previous conversation). It would involve changes to the duration calculation/estimation/deduction for input streams and propagating these durations to output streams or the output context in a correct way. So I looked for a simple and small solution with better chances to get accepted. In webmdashenc.c I found that a duration is written during write_header and this duration is taken from the streams' metadata, so I decided for a similar approach. And here's what it does: At first it is checking the duration of the AVFormatContext. In typical cases this value is not set, but: It is set in cases where the user has specified a recording_time or an end_time via the -t or -to parameters. Then it is looking for a DURATION metadata field in the metadata of the output context (AVFormatContext::metadata). This would only exist in case the user has explicitly specified a metadata DURATION value from the command line. Then it is iterating all streams looking for a "DURATION" metadata (this works unless the option "-map_metadata -1" has been specified) and determines the maximum value. The precendence is as follows: 1. Use duration of AVFormatContext - 2. Use explicitly specified metadata duration value - 3. Use maximum (mapped) metadata duration over all streams. To test this: 1. With explicit recording time: ffmpeg -i file:"src.mkv" -loglevel debug -t 01:38:36.000 -y "dest.mkv" 2. Take duration from metadata specified via command line parameters: ffmpeg -i file:"src.mkv" -loglevel debug -map_metadata -1 -metadata Duration="01:14:33.00" -y "dest.mkv" 3. Take duration from mapped input metadata: ffmpeg -i file:"src.mkv" -loglevel debug -y "dest.mkv" Regression risk: Very low IMO because it only affects the header while ffmpeg is still running. When ffmpeg completes the process, the duration is rewritten to the header with the usual value (same like without this commit). Signed-off-by: SoftWorkz <softworkz@hotmail.com> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
9 years ago
int64_t metadata_duration = get_metadata_duration(s);
if (s->duration > 0) {
int64_t scaledDuration = av_rescale(s->duration, 1000, AV_TIME_BASE);
put_ebml_float(pb, MATROSKA_ID_DURATION, scaledDuration);
av_log(s, AV_LOG_DEBUG, "Write early duration from recording time = %" PRIu64 "\n", scaledDuration);
} else if (metadata_duration > 0) {
int64_t scaledDuration = av_rescale(metadata_duration, 1000, AV_TIME_BASE);
put_ebml_float(pb, MATROSKA_ID_DURATION, scaledDuration);
av_log(s, AV_LOG_DEBUG, "Write early duration from metadata = %" PRIu64 "\n", scaledDuration);
} else {
put_ebml_void(pb, 11); // assumes double-precision float to be written
}
}
if ((s->pb->seekable & AVIO_SEEKABLE_NORMAL) && !mkv->is_live)
end_ebml_master_crc32_preliminary(s->pb, mkv->info_bc,
MATROSKA_ID_INFO, &mkv->info_pos);
else
avformat/matroskaenc: Stop reallocating of Cluster buffer The Matroska muxer uses a dynamic buffer to buffer the content of Clusters before eventually writing them. Up until now, each time a Cluster was written, the dynamic buffer was closed, i.e. freed; now it is only reset, saving allocations of the AVIOContext itself, its opaque as well as most of the reallocations of the buffer. This is advantageous performance-wise, in particular on systems where reallocations are slow (namely Windows). The following table shows the decicyles for writing a frame on Linux (Ubuntu 19.10) and Windows (7) on an x64 Haswell (to /dev/null on Linux, to stdout which is discarded on Windows (the default values of the size and duration of clusters for seekable output have been explicitly set in this case); in all tests, writing CRC-32 values has been disabled in all tests; calls to the muxer's write_packet function in write_packet() in libavformat/mux.c have been timed; each of the following tests has been repeated 50 times): | Windows before | Windows after | Linux before | Linux after _________________________________________________________________ A | 979437 | 192304 | 259500 | 183320 B | 715936 | 155648 | 152786 | 130879 C | 265115 | 56034 | 78496 | 53243 D | 386224 | 80307 | 128894 | 75354 E | 21732 | 10695 | 11320 | 9801 (A is a 10.2 mb/s file with a GOP length of 2s, amounting to an average Cluster size of about 2.5 MiB; the average Cluster size of B is 1.1 MiB; for C it is 2.35 MiB, for D it is 0.46 MiB; for E - a file with just a single audio track of 158kb/s resulting in a Cluster size of about 100 kB, the relative gains were the smallest, probably because of the small Cluster size.) Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
end_ebml_master_crc32(s->pb, &mkv->info_bc, mkv, MATROSKA_ID_INFO, 0, 0);
pb = s->pb;
ret = mkv_write_tracks(s);
if (ret < 0)
return ret;
for (i = 0; i < s->nb_chapters; i++)
if (!s->chapters[i]->id) {
mkv->chapter_id_offset = 1;
break;
}
ret = mkv_write_chapters(s);
if (ret < 0)
return ret;
if (mkv->mode != MODE_WEBM) {
ret = mkv_write_attachments(s);
if (ret < 0)
return ret;
}
ret = mkv_write_tags(s);
if (ret < 0)
return ret;
if (!(s->pb->seekable & AVIO_SEEKABLE_NORMAL) && !mkv->is_live) {
ret = mkv_write_seekhead(pb, mkv, 0, avio_tell(pb));
if (ret < 0)
return ret;
}
if (s->metadata_header_padding > 0) {
if (s->metadata_header_padding == 1)
s->metadata_header_padding++;
put_ebml_void(pb, s->metadata_header_padding);
}
if ((pb->seekable & AVIO_SEEKABLE_NORMAL) && mkv->reserve_cues_space) {
mkv->cues_pos = avio_tell(pb);
if (mkv->reserve_cues_space == 1)
mkv->reserve_cues_space++;
put_ebml_void(pb, mkv->reserve_cues_space);
}
av_init_packet(&mkv->cur_audio_pkt);
mkv->cur_audio_pkt.size = 0;
mkv->cluster_pos = -1;
// start a new cluster every 5 MB or 5 sec, or 32k / 1 sec for streaming or
// after 4k and on a keyframe
if (pb->seekable & AVIO_SEEKABLE_NORMAL) {
if (mkv->cluster_time_limit < 0)
mkv->cluster_time_limit = 5000;
if (mkv->cluster_size_limit < 0)
mkv->cluster_size_limit = 5 * 1024 * 1024;
} else {
if (mkv->cluster_time_limit < 0)
mkv->cluster_time_limit = 1000;
if (mkv->cluster_size_limit < 0)
mkv->cluster_size_limit = 32 * 1024;
}
return 0;
}
static int mkv_blockgroup_size(int pkt_size, int track_num_size)
{
int size = pkt_size + track_num_size + 3;
size += ebml_length_size(size);
size += 2; // EBML ID for block and block duration
size += 9; // max size of block duration incl. length field
return size;
}
static int mkv_strip_wavpack(const uint8_t *src, uint8_t **pdst, int *size)
{
uint8_t *dst;
int srclen = *size;
int offset = 0;
int ret;
dst = av_malloc(srclen);
if (!dst)
return AVERROR(ENOMEM);
while (srclen >= WV_HEADER_SIZE) {
WvHeader header;
ret = ff_wv_parse_header(&header, src);
if (ret < 0)
goto fail;
src += WV_HEADER_SIZE;
srclen -= WV_HEADER_SIZE;
if (srclen < header.blocksize) {
ret = AVERROR_INVALIDDATA;
goto fail;
}
if (header.initial) {
AV_WL32(dst + offset, header.samples);
offset += 4;
}
AV_WL32(dst + offset, header.flags);
AV_WL32(dst + offset + 4, header.crc);
offset += 8;
if (!(header.initial && header.final)) {
AV_WL32(dst + offset, header.blocksize);
offset += 4;
}
memcpy(dst + offset, src, header.blocksize);
src += header.blocksize;
srclen -= header.blocksize;
offset += header.blocksize;
}
*pdst = dst;
*size = offset;
return 0;
fail:
av_freep(&dst);
return ret;
}
static int mkv_write_block(AVFormatContext *s, AVIOContext *pb,
uint32_t blockid, const AVPacket *pkt, int keyframe)
{
MatroskaMuxContext *mkv = s->priv_data;
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.
11 years ago
AVCodecParameters *par = s->streams[pkt->stream_index]->codecpar;
mkv_track *track = &mkv->tracks[pkt->stream_index];
uint8_t *data = NULL, *side_data = NULL;
int err = 0, offset = 0, size = pkt->size, side_data_size = 0;
int64_t ts = track->write_dts ? pkt->dts : pkt->pts;
uint64_t additional_id;
int64_t discard_padding = 0;
unsigned track_number = track->track_num;
ebml_master block_group, block_additions, block_more;
ts += track->ts_offset;
/* The following string is identical to the one in mkv_write_vtt_blocks
* so that only one copy needs to exist in binaries. */
av_log(s, AV_LOG_DEBUG,
"Writing block of size %d with pts %" PRId64 ", dts %" PRId64 ", "
"duration %" PRId64 " at relative offset %" PRId64 " in cluster "
"at offset %" PRId64 ". TrackNumber %u, keyframe %d\n",
pkt->size, pkt->pts, pkt->dts, pkt->duration, avio_tell(pb),
mkv->cluster_pos, track_number, keyframe != 0);
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.
11 years ago
if (par->codec_id == AV_CODEC_ID_H264 && par->extradata_size > 0 &&
(AV_RB24(par->extradata) == 1 || AV_RB32(par->extradata) == 1)) {
err = ff_avc_parse_nal_units_buf(pkt->data, &data, &size);
} else if (par->codec_id == AV_CODEC_ID_HEVC && par->extradata_size > 6 &&
(AV_RB24(par->extradata) == 1 || AV_RB32(par->extradata) == 1)) {
/* extradata is Annex B, assume the bitstream is too and convert it */
err = ff_hevc_annexb2mp4_buf(pkt->data, &data, &size, 0, NULL);
} else if (par->codec_id == AV_CODEC_ID_AV1) {
err = ff_av1_filter_obus_buf(pkt->data, &data, &size, &offset);
} else if (par->codec_id == AV_CODEC_ID_WAVPACK) {
err = mkv_strip_wavpack(pkt->data, &data, &size);
} else
data = pkt->data;
if (err < 0) {
av_log(s, AV_LOG_ERROR, "Error when reformatting data of "
"a packet from stream %d.\n", pkt->stream_index);
return err;
}
if (par->codec_id == AV_CODEC_ID_PRORES && size >= 8) {
/* Matroska specification requires to remove the first QuickTime atom
*/
size -= 8;
offset = 8;
}
side_data = av_packet_get_side_data(pkt,
AV_PKT_DATA_SKIP_SAMPLES,
&side_data_size);
if (side_data && side_data_size >= 10) {
discard_padding = av_rescale_q(AV_RL32(side_data + 4),
(AVRational){1, par->sample_rate},
(AVRational){1, 1000000000});
}
side_data = av_packet_get_side_data(pkt,
AV_PKT_DATA_MATROSKA_BLOCKADDITIONAL,
&side_data_size);
if (side_data) {
// Only the Codec-specific BlockMore (id == 1) is currently supported.
if (side_data_size < 8 || (additional_id = AV_RB64(side_data)) != 1) {
side_data_size = 0;
} else {
side_data += 8;
side_data_size -= 8;
}
}
if (side_data_size || discard_padding) {
block_group = start_ebml_master(pb, MATROSKA_ID_BLOCKGROUP, 0);
blockid = MATROSKA_ID_BLOCK;
}
put_ebml_id(pb, blockid);
put_ebml_length(pb, size + track->track_num_size + 3, 0);
put_ebml_num(pb, track_number, track->track_num_size);
avio_wb16(pb, ts - mkv->cluster_pts);
avio_w8(pb, (blockid == MATROSKA_ID_SIMPLEBLOCK && keyframe) ? (1 << 7) : 0);
avio_write(pb, data + offset, size);
if (data != pkt->data)
av_free(data);
if (blockid == MATROSKA_ID_BLOCK && !keyframe)
put_ebml_sint(pb, MATROSKA_ID_BLOCKREFERENCE, track->last_timestamp - ts);
track->last_timestamp = ts;
if (discard_padding)
put_ebml_sint(pb, MATROSKA_ID_DISCARDPADDING, discard_padding);
if (side_data_size) {
block_additions = start_ebml_master(pb, MATROSKA_ID_BLOCKADDITIONS, 0);
block_more = start_ebml_master(pb, MATROSKA_ID_BLOCKMORE, 0);
/* Until dbc50f8a our demuxer used a wrong default value
* of BlockAddID, so we write it unconditionally. */
put_ebml_uint (pb, MATROSKA_ID_BLOCKADDID, additional_id);
put_ebml_binary(pb, MATROSKA_ID_BLOCKADDITIONAL,
side_data, side_data_size);
end_ebml_master(pb, block_more);
end_ebml_master(pb, block_additions);
}
if (side_data_size || discard_padding)
end_ebml_master(pb, block_group);
return 0;
}
static int mkv_write_vtt_blocks(AVFormatContext *s, AVIOContext *pb, const AVPacket *pkt)
{
MatroskaMuxContext *mkv = s->priv_data;
mkv_track *track = &mkv->tracks[pkt->stream_index];
ebml_master blockgroup;
int id_size, settings_size, size;
uint8_t *id, *settings;
int64_t ts = track->write_dts ? pkt->dts : pkt->pts;
const int flags = 0;
id_size = 0;
id = av_packet_get_side_data(pkt, AV_PKT_DATA_WEBVTT_IDENTIFIER,
&id_size);
settings_size = 0;
settings = av_packet_get_side_data(pkt, AV_PKT_DATA_WEBVTT_SETTINGS,
&settings_size);
size = id_size + 1 + settings_size + 1 + pkt->size;
/* The following string is identical to the one in mkv_write_block so that
* only one copy needs to exist in binaries. */
av_log(s, AV_LOG_DEBUG,
"Writing block of size %d with pts %" PRId64 ", dts %" PRId64 ", "
"duration %" PRId64 " at relative offset %" PRId64 " in cluster "
"at offset %" PRId64 ". TrackNumber %u, keyframe %d\n",
size, pkt->pts, pkt->dts, pkt->duration, avio_tell(pb),
mkv->cluster_pos, track->track_num, 1);
blockgroup = start_ebml_master(pb, MATROSKA_ID_BLOCKGROUP,
mkv_blockgroup_size(size, track->track_num_size));
put_ebml_id(pb, MATROSKA_ID_BLOCK);
put_ebml_length(pb, size + track->track_num_size + 3, 0);
put_ebml_num(pb, track->track_num, track->track_num_size);
avio_wb16(pb, ts - mkv->cluster_pts);
avio_w8(pb, flags);
avio_printf(pb, "%.*s\n%.*s\n%.*s", id_size, id, settings_size, settings, pkt->size, pkt->data);
put_ebml_uint(pb, MATROSKA_ID_BLOCKDURATION, pkt->duration);
end_ebml_master(pb, blockgroup);
return pkt->duration;
}
static void mkv_end_cluster(AVFormatContext *s)
{
MatroskaMuxContext *mkv = s->priv_data;
avformat/matroskaenc: Stop reallocating of Cluster buffer The Matroska muxer uses a dynamic buffer to buffer the content of Clusters before eventually writing them. Up until now, each time a Cluster was written, the dynamic buffer was closed, i.e. freed; now it is only reset, saving allocations of the AVIOContext itself, its opaque as well as most of the reallocations of the buffer. This is advantageous performance-wise, in particular on systems where reallocations are slow (namely Windows). The following table shows the decicyles for writing a frame on Linux (Ubuntu 19.10) and Windows (7) on an x64 Haswell (to /dev/null on Linux, to stdout which is discarded on Windows (the default values of the size and duration of clusters for seekable output have been explicitly set in this case); in all tests, writing CRC-32 values has been disabled in all tests; calls to the muxer's write_packet function in write_packet() in libavformat/mux.c have been timed; each of the following tests has been repeated 50 times): | Windows before | Windows after | Linux before | Linux after _________________________________________________________________ A | 979437 | 192304 | 259500 | 183320 B | 715936 | 155648 | 152786 | 130879 C | 265115 | 56034 | 78496 | 53243 D | 386224 | 80307 | 128894 | 75354 E | 21732 | 10695 | 11320 | 9801 (A is a 10.2 mb/s file with a GOP length of 2s, amounting to an average Cluster size of about 2.5 MiB; the average Cluster size of B is 1.1 MiB; for C it is 2.35 MiB, for D it is 0.46 MiB; for E - a file with just a single audio track of 158kb/s resulting in a Cluster size of about 100 kB, the relative gains were the smallest, probably because of the small Cluster size.) Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
end_ebml_master_crc32(s->pb, &mkv->cluster_bc, mkv, MATROSKA_ID_CLUSTER, 0, 1);
if (!mkv->have_video) {
for (unsigned i = 0; i < s->nb_streams; i++)
mkv->tracks[i].has_cue = 0;
}
mkv->cluster_pos = -1;
avio_write_marker(s->pb, AV_NOPTS_VALUE, AVIO_DATA_MARKER_FLUSH_POINT);
}
static int mkv_check_new_extra_data(AVFormatContext *s, const AVPacket *pkt)
{
MatroskaMuxContext *mkv = s->priv_data;
mkv_track *track = &mkv->tracks[pkt->stream_index];
AVCodecParameters *par = s->streams[pkt->stream_index]->codecpar;
uint8_t *side_data;
int side_data_size = 0, ret;
side_data = av_packet_get_side_data(pkt, AV_PKT_DATA_NEW_EXTRADATA,
&side_data_size);
switch (par->codec_id) {
case AV_CODEC_ID_AAC:
if (side_data_size && (s->pb->seekable & AVIO_SEEKABLE_NORMAL) && !mkv->is_live) {
int filler, output_sample_rate = 0;
ret = get_aac_sample_rates(s, side_data, side_data_size, &track->sample_rate,
&output_sample_rate);
if (ret < 0)
return ret;
if (!output_sample_rate)
output_sample_rate = track->sample_rate; // Space is already reserved, so it's this or a void element.
ret = ff_alloc_extradata(par, side_data_size);
if (ret < 0)
return ret;
memcpy(par->extradata, side_data, side_data_size);
avio_seek(mkv->tracks_bc, track->codecpriv_offset, SEEK_SET);
mkv_write_codecprivate(s, mkv->tracks_bc, par, 1, 0);
filler = MAX_PCE_SIZE + 2 + 4 - (avio_tell(mkv->tracks_bc) - track->codecpriv_offset);
if (filler)
put_ebml_void(mkv->tracks_bc, filler);
avio_seek(mkv->tracks_bc, track->sample_rate_offset, SEEK_SET);
put_ebml_float(mkv->tracks_bc, MATROSKA_ID_AUDIOSAMPLINGFREQ, track->sample_rate);
put_ebml_float(mkv->tracks_bc, MATROSKA_ID_AUDIOOUTSAMPLINGFREQ, output_sample_rate);
} else if (!par->extradata_size && !track->sample_rate) {
// No extradata (codecpar or packet side data).
av_log(s, AV_LOG_ERROR, "Error parsing AAC extradata, unable to determine samplerate.\n");
return AVERROR(EINVAL);
}
break;
case AV_CODEC_ID_FLAC:
if (side_data_size && (s->pb->seekable & AVIO_SEEKABLE_NORMAL) && !mkv->is_live) {
uint8_t *old_extradata = par->extradata;
if (side_data_size != par->extradata_size) {
av_log(s, AV_LOG_ERROR, "Invalid FLAC STREAMINFO metadata for output stream %d\n",
pkt->stream_index);
return AVERROR(EINVAL);
}
par->extradata = side_data;
avio_seek(mkv->tracks_bc, track->codecpriv_offset, SEEK_SET);
mkv_write_codecprivate(s, mkv->tracks_bc, par, 1, 0);
par->extradata = old_extradata;
}
break;
// FIXME: Remove the following once libaom starts propagating extradata during init()
// See https://bugs.chromium.org/p/aomedia/issues/detail?id=2012
case AV_CODEC_ID_AV1:
if (side_data_size && (s->pb->seekable & AVIO_SEEKABLE_NORMAL) && !mkv->is_live &&
!par->extradata_size) {
AVIOContext *dyn_cp;
uint8_t *codecpriv;
int codecpriv_size;
ret = avio_open_dyn_buf(&dyn_cp);
if (ret < 0)
return ret;
ff_isom_write_av1c(dyn_cp, side_data, side_data_size);
codecpriv_size = avio_close_dyn_buf(dyn_cp, &codecpriv);
if (!codecpriv_size) {
av_free(codecpriv);
return AVERROR_INVALIDDATA;
}
avio_seek(mkv->tracks_bc, track->codecpriv_offset, SEEK_SET);
// Do not write the OBUs as we don't have space saved for them
put_ebml_binary(mkv->tracks_bc, MATROSKA_ID_CODECPRIVATE, codecpriv, 4);
av_free(codecpriv);
ret = ff_alloc_extradata(par, side_data_size);
if (ret < 0)
return ret;
memcpy(par->extradata, side_data, side_data_size);
} else if (!par->extradata_size)
return AVERROR_INVALIDDATA;
break;
default:
if (side_data_size)
av_log(s, AV_LOG_DEBUG, "Ignoring new extradata in a packet for stream %d.\n", pkt->stream_index);
break;
}
return 0;
}
static int mkv_write_packet_internal(AVFormatContext *s, const AVPacket *pkt)
{
MatroskaMuxContext *mkv = s->priv_data;
AVIOContext *pb;
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.
11 years ago
AVCodecParameters *par = s->streams[pkt->stream_index]->codecpar;
mkv_track *track = &mkv->tracks[pkt->stream_index];
int keyframe = !!(pkt->flags & AV_PKT_FLAG_KEY);
int duration = pkt->duration;
int ret;
int64_t ts = track->write_dts ? pkt->dts : pkt->pts;
int64_t relative_packet_pos;
if (ts == AV_NOPTS_VALUE) {
av_log(s, AV_LOG_ERROR, "Can't write packet with unknown timestamp\n");
return AVERROR(EINVAL);
}
ts += track->ts_offset;
if (mkv->cluster_pos != -1) {
int64_t cluster_time = ts - mkv->cluster_pts;
if ((int16_t)cluster_time != cluster_time) {
mkv_end_cluster(s);
av_log(s, AV_LOG_WARNING, "Starting new cluster due to timestamp\n");
}
}
if (mkv->cluster_pos == -1) {
ret = start_ebml_master_crc32(&mkv->cluster_bc, mkv);
if (ret < 0)
return ret;
mkv->cluster_pos = avio_tell(s->pb);
put_ebml_uint(mkv->cluster_bc, MATROSKA_ID_CLUSTERTIMECODE, FFMAX(0, ts));
mkv->cluster_pts = FFMAX(0, ts);
av_log(s, AV_LOG_DEBUG,
"Starting new cluster with timestamp "
"%" PRId64 " at offset %" PRId64 " bytes\n",
mkv->cluster_pts, mkv->cluster_pos);
}
pb = mkv->cluster_bc;
relative_packet_pos = avio_tell(pb);
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.
11 years ago
if (par->codec_type != AVMEDIA_TYPE_SUBTITLE) {
ret = mkv_write_block(s, pb, MATROSKA_ID_SIMPLEBLOCK, pkt, keyframe);
if (ret < 0)
return ret;
if ((s->pb->seekable & AVIO_SEEKABLE_NORMAL) && keyframe &&
(par->codec_type == AVMEDIA_TYPE_VIDEO || !mkv->have_video && !track->has_cue)) {
ret = mkv_add_cuepoint(mkv, pkt->stream_index, ts,
mkv->cluster_pos, relative_packet_pos, -1);
if (ret < 0)
return ret;
track->has_cue = 1;
}
} else {
if (par->codec_id == AV_CODEC_ID_WEBVTT) {
duration = mkv_write_vtt_blocks(s, pb, pkt);
} else {
ebml_master blockgroup = start_ebml_master(pb, MATROSKA_ID_BLOCKGROUP,
mkv_blockgroup_size(pkt->size,
track->track_num_size));
#if FF_API_CONVERGENCE_DURATION
FF_DISABLE_DEPRECATION_WARNINGS
/* For backward compatibility, prefer convergence_duration. */
if (pkt->convergence_duration > 0) {
duration = pkt->convergence_duration;
}
FF_ENABLE_DEPRECATION_WARNINGS
#endif
/* All subtitle blocks are considered to be keyframes. */
mkv_write_block(s, pb, MATROSKA_ID_BLOCK, pkt, 1);
put_ebml_uint(pb, MATROSKA_ID_BLOCKDURATION, duration);
end_ebml_master(pb, blockgroup);
}
if (s->pb->seekable & AVIO_SEEKABLE_NORMAL) {
ret = mkv_add_cuepoint(mkv, pkt->stream_index, ts,
mkv->cluster_pos, relative_packet_pos, duration);
if (ret < 0)
return ret;
}
}
mkv->duration = FFMAX(mkv->duration, ts + duration);
track->duration = FFMAX(track->duration, ts + duration);
return 0;
}
static int mkv_write_packet(AVFormatContext *s, const AVPacket *pkt)
{
MatroskaMuxContext *mkv = s->priv_data;
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.
11 years ago
int codec_type = s->streams[pkt->stream_index]->codecpar->codec_type;
int keyframe = !!(pkt->flags & AV_PKT_FLAG_KEY);
int cluster_size;
int64_t cluster_time;
int ret;
int start_new_cluster;
ret = mkv_check_new_extra_data(s, pkt);
if (ret < 0)
return ret;
if (mkv->cluster_pos != -1) {
if (mkv->tracks[pkt->stream_index].write_dts)
cluster_time = pkt->dts - mkv->cluster_pts;
else
cluster_time = pkt->pts - mkv->cluster_pts;
cluster_time += mkv->tracks[pkt->stream_index].ts_offset;
cluster_size = avio_tell(mkv->cluster_bc);
if (mkv->is_dash && codec_type == AVMEDIA_TYPE_VIDEO) {
// WebM DASH specification states that the first block of
// every Cluster has to be a key frame. So for DASH video,
// we only create a Cluster on seeing key frames.
start_new_cluster = keyframe;
} else if (mkv->is_dash && codec_type == AVMEDIA_TYPE_AUDIO &&
cluster_time > mkv->cluster_time_limit) {
// For DASH audio, we create a Cluster based on cluster_time_limit.
start_new_cluster = 1;
} else if (!mkv->is_dash &&
(cluster_size > mkv->cluster_size_limit ||
cluster_time > mkv->cluster_time_limit ||
(codec_type == AVMEDIA_TYPE_VIDEO && keyframe &&
cluster_size > 4 * 1024))) {
start_new_cluster = 1;
} else
start_new_cluster = 0;
if (start_new_cluster)
mkv_end_cluster(s);
}
if (!mkv->cluster_pos)
avio_write_marker(s->pb,
av_rescale_q(pkt->dts, s->streams[pkt->stream_index]->time_base, AV_TIME_BASE_Q),
keyframe && (mkv->have_video ? codec_type == AVMEDIA_TYPE_VIDEO : 1) ? AVIO_DATA_MARKER_SYNC_POINT : AVIO_DATA_MARKER_BOUNDARY_POINT);
// check if we have an audio packet cached
if (mkv->cur_audio_pkt.size > 0) {
ret = mkv_write_packet_internal(s, &mkv->cur_audio_pkt);
av_packet_unref(&mkv->cur_audio_pkt);
if (ret < 0) {
av_log(s, AV_LOG_ERROR,
"Could not write cached audio packet ret:%d\n", ret);
return ret;
}
}
// buffer an audio packet to ensure the packet containing the video
// keyframe's timecode is contained in the same cluster for WebM
if (codec_type == AVMEDIA_TYPE_AUDIO) {
if (pkt->size > 0)
ret = av_packet_ref(&mkv->cur_audio_pkt, pkt);
} else
ret = mkv_write_packet_internal(s, pkt);
return ret;
}
static int mkv_write_flush_packet(AVFormatContext *s, AVPacket *pkt)
{
MatroskaMuxContext *mkv = s->priv_data;
if (!pkt) {
if (mkv->cluster_pos != -1) {
mkv_end_cluster(s);
av_log(s, AV_LOG_DEBUG,
"Flushing cluster at offset %" PRIu64 " bytes\n",
avio_tell(s->pb));
}
return 1;
}
return mkv_write_packet(s, pkt);
}
static int mkv_write_trailer(AVFormatContext *s)
{
MatroskaMuxContext *mkv = s->priv_data;
AVIOContext *pb = s->pb;
int ret;
// check if we have an audio packet cached
if (mkv->cur_audio_pkt.size > 0) {
ret = mkv_write_packet_internal(s, &mkv->cur_audio_pkt);
if (ret < 0) {
av_log(s, AV_LOG_ERROR,
"Could not write cached audio packet ret:%d\n", ret);
return ret;
}
}
if (mkv->cluster_bc) {
avformat/matroskaenc: Stop reallocating of Cluster buffer The Matroska muxer uses a dynamic buffer to buffer the content of Clusters before eventually writing them. Up until now, each time a Cluster was written, the dynamic buffer was closed, i.e. freed; now it is only reset, saving allocations of the AVIOContext itself, its opaque as well as most of the reallocations of the buffer. This is advantageous performance-wise, in particular on systems where reallocations are slow (namely Windows). The following table shows the decicyles for writing a frame on Linux (Ubuntu 19.10) and Windows (7) on an x64 Haswell (to /dev/null on Linux, to stdout which is discarded on Windows (the default values of the size and duration of clusters for seekable output have been explicitly set in this case); in all tests, writing CRC-32 values has been disabled in all tests; calls to the muxer's write_packet function in write_packet() in libavformat/mux.c have been timed; each of the following tests has been repeated 50 times): | Windows before | Windows after | Linux before | Linux after _________________________________________________________________ A | 979437 | 192304 | 259500 | 183320 B | 715936 | 155648 | 152786 | 130879 C | 265115 | 56034 | 78496 | 53243 D | 386224 | 80307 | 128894 | 75354 E | 21732 | 10695 | 11320 | 9801 (A is a 10.2 mb/s file with a GOP length of 2s, amounting to an average Cluster size of about 2.5 MiB; the average Cluster size of B is 1.1 MiB; for C it is 2.35 MiB, for D it is 0.46 MiB; for E - a file with just a single audio track of 158kb/s resulting in a Cluster size of about 100 kB, the relative gains were the smallest, probably because of the small Cluster size.) Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
end_ebml_master_crc32(pb, &mkv->cluster_bc, mkv,
MATROSKA_ID_CLUSTER, 0, 0);
}
ret = mkv_write_chapters(s);
if (ret < 0)
return ret;
if ((pb->seekable & AVIO_SEEKABLE_NORMAL) && !mkv->is_live) {
int64_t endpos, ret64;
endpos = avio_tell(pb);
if (mkv->cues.num_entries) {
avformat/matroskaenc: Stop reallocating of Cluster buffer The Matroska muxer uses a dynamic buffer to buffer the content of Clusters before eventually writing them. Up until now, each time a Cluster was written, the dynamic buffer was closed, i.e. freed; now it is only reset, saving allocations of the AVIOContext itself, its opaque as well as most of the reallocations of the buffer. This is advantageous performance-wise, in particular on systems where reallocations are slow (namely Windows). The following table shows the decicyles for writing a frame on Linux (Ubuntu 19.10) and Windows (7) on an x64 Haswell (to /dev/null on Linux, to stdout which is discarded on Windows (the default values of the size and duration of clusters for seekable output have been explicitly set in this case); in all tests, writing CRC-32 values has been disabled in all tests; calls to the muxer's write_packet function in write_packet() in libavformat/mux.c have been timed; each of the following tests has been repeated 50 times): | Windows before | Windows after | Linux before | Linux after _________________________________________________________________ A | 979437 | 192304 | 259500 | 183320 B | 715936 | 155648 | 152786 | 130879 C | 265115 | 56034 | 78496 | 53243 D | 386224 | 80307 | 128894 | 75354 E | 21732 | 10695 | 11320 | 9801 (A is a 10.2 mb/s file with a GOP length of 2s, amounting to an average Cluster size of about 2.5 MiB; the average Cluster size of B is 1.1 MiB; for C it is 2.35 MiB, for D it is 0.46 MiB; for E - a file with just a single audio track of 158kb/s resulting in a Cluster size of about 100 kB, the relative gains were the smallest, probably because of the small Cluster size.) Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
AVIOContext *cues = NULL;
uint64_t size;
int64_t cuespos = endpos;
int length_size = 0;
ret = start_ebml_master_crc32(&cues, mkv);
if (ret < 0)
return ret;
ret = mkv_assemble_cues(s->streams, cues, &mkv->cues,
mkv->tracks, s->nb_streams);
if (ret < 0) {
ffio_free_dyn_buf(&cues);
return ret;
}
if (mkv->reserve_cues_space) {
size = avio_tell(cues);
length_size = ebml_length_size(size);
size += 4 + length_size;
if (mkv->reserve_cues_space < size) {
av_log(s, AV_LOG_WARNING,
"Insufficient space reserved for Cues: "
"%d < %"PRIu64". No Cues will be output.\n",
mkv->reserve_cues_space, size);
mkv->reserve_cues_space = -1;
ffio_free_dyn_buf(&cues);
goto after_cues;
} else {
cuespos = mkv->cues_pos;
if ((ret64 = avio_seek(pb, mkv->cues_pos, SEEK_SET)) < 0) {
ffio_free_dyn_buf(&cues);
return ret64;
}
if (mkv->reserve_cues_space == size + 1) {
/* There is no way to reserve a single byte because
* the minimal size of an EBML Void element is 2
* (1 byte ID, 1 byte length field). This problem
* is solved by writing the Cues' length field on
* one byte more than necessary. */
length_size++;
size++;
}
}
}
mkv_add_seekhead_entry(mkv, MATROSKA_ID_CUES, cuespos);
avformat/matroskaenc: Stop reallocating of Cluster buffer The Matroska muxer uses a dynamic buffer to buffer the content of Clusters before eventually writing them. Up until now, each time a Cluster was written, the dynamic buffer was closed, i.e. freed; now it is only reset, saving allocations of the AVIOContext itself, its opaque as well as most of the reallocations of the buffer. This is advantageous performance-wise, in particular on systems where reallocations are slow (namely Windows). The following table shows the decicyles for writing a frame on Linux (Ubuntu 19.10) and Windows (7) on an x64 Haswell (to /dev/null on Linux, to stdout which is discarded on Windows (the default values of the size and duration of clusters for seekable output have been explicitly set in this case); in all tests, writing CRC-32 values has been disabled in all tests; calls to the muxer's write_packet function in write_packet() in libavformat/mux.c have been timed; each of the following tests has been repeated 50 times): | Windows before | Windows after | Linux before | Linux after _________________________________________________________________ A | 979437 | 192304 | 259500 | 183320 B | 715936 | 155648 | 152786 | 130879 C | 265115 | 56034 | 78496 | 53243 D | 386224 | 80307 | 128894 | 75354 E | 21732 | 10695 | 11320 | 9801 (A is a 10.2 mb/s file with a GOP length of 2s, amounting to an average Cluster size of about 2.5 MiB; the average Cluster size of B is 1.1 MiB; for C it is 2.35 MiB, for D it is 0.46 MiB; for E - a file with just a single audio track of 158kb/s resulting in a Cluster size of about 100 kB, the relative gains were the smallest, probably because of the small Cluster size.) Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
end_ebml_master_crc32(pb, &cues, mkv, MATROSKA_ID_CUES,
length_size, 0);
if (mkv->reserve_cues_space) {
if (size < mkv->reserve_cues_space)
put_ebml_void(pb, mkv->reserve_cues_space - size);
} else
endpos = avio_tell(pb);
}
after_cues:
ret = mkv_write_seekhead(pb, mkv, 1, mkv->info_pos);
if (ret < 0)
return ret;
// update the duration
av_log(s, AV_LOG_DEBUG, "end duration = %" PRIu64 "\n", mkv->duration);
avio_seek(mkv->info_bc, mkv->duration_offset, SEEK_SET);
put_ebml_float(mkv->info_bc, MATROSKA_ID_DURATION, mkv->duration);
avformat/matroskaenc: Stop reallocating of Cluster buffer The Matroska muxer uses a dynamic buffer to buffer the content of Clusters before eventually writing them. Up until now, each time a Cluster was written, the dynamic buffer was closed, i.e. freed; now it is only reset, saving allocations of the AVIOContext itself, its opaque as well as most of the reallocations of the buffer. This is advantageous performance-wise, in particular on systems where reallocations are slow (namely Windows). The following table shows the decicyles for writing a frame on Linux (Ubuntu 19.10) and Windows (7) on an x64 Haswell (to /dev/null on Linux, to stdout which is discarded on Windows (the default values of the size and duration of clusters for seekable output have been explicitly set in this case); in all tests, writing CRC-32 values has been disabled in all tests; calls to the muxer's write_packet function in write_packet() in libavformat/mux.c have been timed; each of the following tests has been repeated 50 times): | Windows before | Windows after | Linux before | Linux after _________________________________________________________________ A | 979437 | 192304 | 259500 | 183320 B | 715936 | 155648 | 152786 | 130879 C | 265115 | 56034 | 78496 | 53243 D | 386224 | 80307 | 128894 | 75354 E | 21732 | 10695 | 11320 | 9801 (A is a 10.2 mb/s file with a GOP length of 2s, amounting to an average Cluster size of about 2.5 MiB; the average Cluster size of B is 1.1 MiB; for C it is 2.35 MiB, for D it is 0.46 MiB; for E - a file with just a single audio track of 158kb/s resulting in a Cluster size of about 100 kB, the relative gains were the smallest, probably because of the small Cluster size.) Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
end_ebml_master_crc32(pb, &mkv->info_bc, mkv, MATROSKA_ID_INFO, 0, 0);
if (mkv->tracks_bc) {
// write Tracks master
avio_seek(pb, mkv->tracks_pos, SEEK_SET);
end_ebml_master_crc32(pb, &mkv->tracks_bc, mkv,
MATROSKA_ID_TRACKS, 0, 0);
}
// update stream durations
if (mkv->tags_bc) {
int i;
for (i = 0; i < s->nb_streams; ++i) {
const AVStream *st = s->streams[i];
const mkv_track *track = &mkv->tracks[i];
if (track->duration_offset > 0) {
double duration_sec = track->duration * av_q2d(st->time_base);
char duration_string[20] = "";
av_log(s, AV_LOG_DEBUG, "stream %d end duration = %" PRIu64 "\n", i,
track->duration);
avio_seek(mkv->tags_bc, track->duration_offset, SEEK_SET);
snprintf(duration_string, 20, "%02d:%02d:%012.9f",
(int) duration_sec / 3600, ((int) duration_sec / 60) % 60,
fmod(duration_sec, 60));
put_ebml_binary(mkv->tags_bc, MATROSKA_ID_TAGSTRING, duration_string, 20);
}
}
avio_seek(pb, mkv->tags_pos, SEEK_SET);
avformat/matroskaenc: Stop reallocating of Cluster buffer The Matroska muxer uses a dynamic buffer to buffer the content of Clusters before eventually writing them. Up until now, each time a Cluster was written, the dynamic buffer was closed, i.e. freed; now it is only reset, saving allocations of the AVIOContext itself, its opaque as well as most of the reallocations of the buffer. This is advantageous performance-wise, in particular on systems where reallocations are slow (namely Windows). The following table shows the decicyles for writing a frame on Linux (Ubuntu 19.10) and Windows (7) on an x64 Haswell (to /dev/null on Linux, to stdout which is discarded on Windows (the default values of the size and duration of clusters for seekable output have been explicitly set in this case); in all tests, writing CRC-32 values has been disabled in all tests; calls to the muxer's write_packet function in write_packet() in libavformat/mux.c have been timed; each of the following tests has been repeated 50 times): | Windows before | Windows after | Linux before | Linux after _________________________________________________________________ A | 979437 | 192304 | 259500 | 183320 B | 715936 | 155648 | 152786 | 130879 C | 265115 | 56034 | 78496 | 53243 D | 386224 | 80307 | 128894 | 75354 E | 21732 | 10695 | 11320 | 9801 (A is a 10.2 mb/s file with a GOP length of 2s, amounting to an average Cluster size of about 2.5 MiB; the average Cluster size of B is 1.1 MiB; for C it is 2.35 MiB, for D it is 0.46 MiB; for E - a file with just a single audio track of 158kb/s resulting in a Cluster size of about 100 kB, the relative gains were the smallest, probably because of the small Cluster size.) Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
5 years ago
end_ebml_master_crc32(pb, &mkv->tags_bc, mkv, MATROSKA_ID_TAGS, 0, 0);
}
avio_seek(pb, endpos, SEEK_SET);
}
if (!mkv->is_live)
end_ebml_master(pb, mkv->segment);
return mkv->reserve_cues_space < 0 ? AVERROR(EINVAL) : 0;
}
static int mkv_query_codec(enum AVCodecID codec_id, int std_compliance)
{
int i;
for (i = 0; ff_mkv_codec_tags[i].id != AV_CODEC_ID_NONE; i++)
if (ff_mkv_codec_tags[i].id == codec_id)
return 1;
if (std_compliance < FF_COMPLIANCE_NORMAL) {
enum AVMediaType type = avcodec_get_type(codec_id);
// mkv theoretically supports any video/audio through VFW/ACM
if (type == AVMEDIA_TYPE_VIDEO || type == AVMEDIA_TYPE_AUDIO)
return 1;
}
return 0;
}
static int webm_query_codec(enum AVCodecID codec_id, int std_compliance)
{
int i;
for (i = 0; ff_webm_codec_tags[i].id != AV_CODEC_ID_NONE; i++)
if (ff_webm_codec_tags[i].id == codec_id)
return 1;
return 0;
}
static uint64_t mkv_get_uid(const mkv_track *tracks, int i, AVLFG *c)
{
while (1) {
uint64_t uid;
int k;
uid = (uint64_t)av_lfg_get(c) << 32;
uid |= av_lfg_get(c);
if (!uid)
continue;
for (k = 0; k < i; k++) {
if (tracks[k].uid == uid)
break;
}
if (k == i)
return uid;
}
}
static int mkv_init(struct AVFormatContext *s)
{
MatroskaMuxContext *mkv = s->priv_data;
AVLFG c;
unsigned nb_tracks = 0;
int i;
for (i = 0; i < s->nb_streams; i++) {
if (s->streams[i]->codecpar->codec_id == AV_CODEC_ID_ATRAC3 ||
s->streams[i]->codecpar->codec_id == AV_CODEC_ID_COOK ||
s->streams[i]->codecpar->codec_id == AV_CODEC_ID_RA_288 ||
s->streams[i]->codecpar->codec_id == AV_CODEC_ID_SIPR ||
s->streams[i]->codecpar->codec_id == AV_CODEC_ID_RV10 ||
s->streams[i]->codecpar->codec_id == AV_CODEC_ID_RV20) {
av_log(s, AV_LOG_ERROR,
"The Matroska muxer does not yet support muxing %s\n",
avcodec_get_name(s->streams[i]->codecpar->codec_id));
return AVERROR_PATCHWELCOME;
}
}
if (s->avoid_negative_ts < 0) {
s->avoid_negative_ts = 1;
s->internal->avoid_negative_ts_use_pts = 1;
}
if (!strcmp(s->oformat->name, "webm")) {
mkv->mode = MODE_WEBM;
mkv->write_crc = 0;
} else
mkv->mode = MODE_MATROSKAv2;
mkv->tracks = av_mallocz_array(s->nb_streams, sizeof(*mkv->tracks));
if (!mkv->tracks)
return AVERROR(ENOMEM);
if (!(s->flags & AVFMT_FLAG_BITEXACT)) {
av_lfg_init(&c, av_get_random_seed());
// Calculate the SegmentUID now in order not to waste our random seed.
for (i = 0; i < 4; i++)
mkv->segment_uid[i] = av_lfg_get(&c);
}
for (i = 0; i < s->nb_streams; i++) {
AVStream *st = s->streams[i];
mkv_track *track = &mkv->tracks[i];
if (s->flags & AVFMT_FLAG_BITEXACT) {
track->uid = i + 1;
} else {
track->uid = mkv_get_uid(mkv->tracks, i, &c);
}
// ms precision is the de-facto standard timescale for mkv files
avpriv_set_pts_info(st, 64, 1, 1000);
if (st->codecpar->codec_type == AVMEDIA_TYPE_ATTACHMENT) {
if (mkv->mode == MODE_WEBM) {
av_log(s, AV_LOG_WARNING, "Stream %d will be ignored "
"as WebM doesn't support attachments.\n", i);
}
mkv->nb_attachments++;
continue;
}
nb_tracks++;
track->track_num = mkv->is_dash ? mkv->dash_track_number : nb_tracks;
track->track_num_size = ebml_num_size(track->track_num);
}
if (mkv->is_dash && nb_tracks != 1)
return AVERROR(EINVAL);
return 0;
}
static int mkv_check_bitstream(struct AVFormatContext *s, const AVPacket *pkt)
{
int ret = 1;
AVStream *st = s->streams[pkt->stream_index];
if (st->codecpar->codec_id == AV_CODEC_ID_AAC) {
if (pkt->size > 2 && (AV_RB16(pkt->data) & 0xfff0) == 0xfff0)
ret = ff_stream_add_bitstream_filter(st, "aac_adtstoasc", NULL);
} else if (st->codecpar->codec_id == AV_CODEC_ID_VP9) {
ret = ff_stream_add_bitstream_filter(st, "vp9_superframe", NULL);
}
return ret;
}
static const AVCodecTag additional_audio_tags[] = {
{ AV_CODEC_ID_ALAC, 0XFFFFFFFF },
{ AV_CODEC_ID_MLP, 0xFFFFFFFF },
{ AV_CODEC_ID_OPUS, 0xFFFFFFFF },
{ AV_CODEC_ID_PCM_S16BE, 0xFFFFFFFF },
{ AV_CODEC_ID_PCM_S24BE, 0xFFFFFFFF },
{ AV_CODEC_ID_PCM_S32BE, 0xFFFFFFFF },
{ AV_CODEC_ID_QDMC, 0xFFFFFFFF },
{ AV_CODEC_ID_QDM2, 0xFFFFFFFF },
{ AV_CODEC_ID_RA_144, 0xFFFFFFFF },
{ AV_CODEC_ID_RA_288, 0xFFFFFFFF },
{ AV_CODEC_ID_COOK, 0xFFFFFFFF },
{ AV_CODEC_ID_TRUEHD, 0xFFFFFFFF },
{ AV_CODEC_ID_NONE, 0xFFFFFFFF }
};
static const AVCodecTag additional_video_tags[] = {
{ AV_CODEC_ID_RV10, 0xFFFFFFFF },
{ AV_CODEC_ID_RV20, 0xFFFFFFFF },
{ AV_CODEC_ID_RV30, 0xFFFFFFFF },
{ AV_CODEC_ID_NONE, 0xFFFFFFFF }
};
static const AVCodecTag additional_subtitle_tags[] = {
{ AV_CODEC_ID_DVB_SUBTITLE, 0xFFFFFFFF },
{ AV_CODEC_ID_DVD_SUBTITLE, 0xFFFFFFFF },
{ AV_CODEC_ID_HDMV_PGS_SUBTITLE, 0xFFFFFFFF },
{ AV_CODEC_ID_NONE, 0xFFFFFFFF }
};
#define OFFSET(x) offsetof(MatroskaMuxContext, x)
#define FLAGS AV_OPT_FLAG_ENCODING_PARAM
static const AVOption options[] = {
{ "reserve_index_space", "Reserve a given amount of space (in bytes) at the beginning of the file for the index (cues).", OFFSET(reserve_cues_space), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, INT_MAX, FLAGS },
{ "cluster_size_limit", "Store at most the provided amount of bytes in a cluster. ", OFFSET(cluster_size_limit), AV_OPT_TYPE_INT , { .i64 = -1 }, -1, INT_MAX, FLAGS },
{ "cluster_time_limit", "Store at most the provided number of milliseconds in a cluster.", OFFSET(cluster_time_limit), AV_OPT_TYPE_INT64, { .i64 = -1 }, -1, INT64_MAX, FLAGS },
{ "dash", "Create a WebM file conforming to WebM DASH specification", OFFSET(is_dash), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, FLAGS },
{ "dash_track_number", "Track number for the DASH stream", OFFSET(dash_track_number), AV_OPT_TYPE_INT, { .i64 = 1 }, 1, INT_MAX, FLAGS },
{ "live", "Write files assuming it is a live stream.", OFFSET(is_live), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, FLAGS },
{ "allow_raw_vfw", "allow RAW VFW mode", OFFSET(allow_raw_vfw), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, FLAGS },
{ "write_crc32", "write a CRC32 element inside every Level 1 element", OFFSET(write_crc), AV_OPT_TYPE_BOOL, { .i64 = 1 }, 0, 1, FLAGS },
{ "default_mode", "Controls how a track's FlagDefault is inferred", OFFSET(default_mode), AV_OPT_TYPE_INT, { .i64 = DEFAULT_MODE_INFER }, DEFAULT_MODE_INFER, DEFAULT_MODE_PASSTHROUGH, FLAGS, "default_mode" },
{ "infer", "For each track type, mark the first track of disposition default as default; if none exists, mark the first track as default.", 0, AV_OPT_TYPE_CONST, { .i64 = DEFAULT_MODE_INFER }, 0, 0, FLAGS, "default_mode" },
{ "infer_no_subs", "For each track type, mark the first track of disposition default as default; for audio and video: if none exists, mark the first track as default.", 0, AV_OPT_TYPE_CONST, { .i64 = DEFAULT_MODE_INFER_NO_SUBS }, 0, 0, FLAGS, "default_mode" },
{ "passthrough", "Use the disposition flag as-is", 0, AV_OPT_TYPE_CONST, { .i64 = DEFAULT_MODE_PASSTHROUGH }, 0, 0, FLAGS, "default_mode" },
{ NULL },
};
#if CONFIG_MATROSKA_MUXER
static const AVClass matroska_class = {
.class_name = "matroska muxer",
.item_name = av_default_item_name,
.option = options,
.version = LIBAVUTIL_VERSION_INT,
};
AVOutputFormat ff_matroska_muxer = {
.name = "matroska",
.long_name = NULL_IF_CONFIG_SMALL("Matroska"),
.mime_type = "video/x-matroska",
.extensions = "mkv",
.priv_data_size = sizeof(MatroskaMuxContext),
.audio_codec = CONFIG_LIBVORBIS_ENCODER ?
AV_CODEC_ID_VORBIS : AV_CODEC_ID_AC3,
.video_codec = CONFIG_LIBX264_ENCODER ?
AV_CODEC_ID_H264 : AV_CODEC_ID_MPEG4,
.init = mkv_init,
.deinit = mkv_deinit,
.write_header = mkv_write_header,
.write_packet = mkv_write_flush_packet,
.write_trailer = mkv_write_trailer,
.flags = AVFMT_GLOBALHEADER | AVFMT_VARIABLE_FPS |
AVFMT_TS_NONSTRICT | AVFMT_ALLOW_FLUSH,
.codec_tag = (const AVCodecTag* const []){
ff_codec_bmp_tags, ff_codec_wav_tags,
additional_audio_tags, additional_video_tags, additional_subtitle_tags, 0
},
subtitles: introduce ASS codec id and use it. Currently, we have a AV_CODEC_ID_SSA, which matches the way the ASS/SSA markup is muxed in a standalone .ass/.ssa file. This means the AVPacket data starts with a "Dialogue:" string, followed by a timing information (start and end of the event as string) and a trailing CRLF after each line. One packet can contain several lines. We'll refer to this layout as "SSA" or "SSA lines". In matroska, this markup is not stored as such: it has no "Dialogue:" prefix, it contains a ReadOrder field, the timing information is not in the payload, and it doesn't contain the trailing CRLF. See [1] for more info. We'll refer to this layout as "ASS". Since we have only one common codec for both formats, the matroska demuxer is constructing an AVPacket following the "SSA lines" format. This causes several problems, so it was decided to change this into clean ASS packets. Some insight about what is changed or unchanged in this commit: CODECS ------ - the decoding process still writes "SSA lines" markup inside the ass fields of the subtitles rectangles (sub->rects[n]->ass), which is still the current common way of representing decoded subtitles markup. It is meant to change later. - new ASS codec id: AV_CODEC_ID_ASS (which is different from the legacy AV_CODEC_ID_SSA) - lavc/assdec: the "ass" decoder is renamed into "ssa" (instead of "ass") for consistency with the codec id and allows to add a real ass decoder. This ass decoder receives clean ASS lines (so it starts with a ReadOrder, is followed by the Layer, etc). We make sure this is decoded properly in a new ass-line rectangle of the decoded subtitles (the ssa decoder OTOH is doing a simple straightforward copy). Using the packet timing instead of data string makes sure the ass-line now contains the appropriate timing. - lavc/assenc: just like the ass decoder, the "ssa" encoder is renamed into "ssa" (instead of "ass") for consistency with the codec id, and allows to add a real "ass" encoder. One important thing about this encoder is that it only supports one ass rectangle: we could have put several dialogue events in the AVPacket (separated by a \0 for instance) but this would have cause trouble for the muxer which needs not only the start time, but also the duration: typically, you have merged events with the same start time (stored in the AVPacket->pts) but a different duration. At the moment, only the matroska do the merge with the SSA-line codec. We will need to make sure all the decoders in the future can't add more than one rectangle (and only one Dialogue line in it obviously). FORMATS ------- - lavf/assenc: the .ass/.ssa muxer can take both SSA and ASS packets. In the case of ASS packets as input, it adds the timing based on the AVPacket pts and duration, and mux it with "Dialogue:", trailing CRLF, etc. - lavf/assdec: unchanged; it currently still only outputs SSA-lines packets. - lavf/mkv: the demuxer can now output ASS packets without the need of any "SSA-lines" reconstruction hack. It will become the default at next libavformat bump, and the SSA support will be dropped from the demuxer. The muxer can take ASS packets since it's muxed normally, and still supports the old SSA packets. All the SSA support and hacks in Matroska code will be dropped at next lavf bump. [1]: http://www.matroska.org/technical/specs/subtitles/ssa.html
12 years ago
.subtitle_codec = AV_CODEC_ID_ASS,
.query_codec = mkv_query_codec,
.check_bitstream = mkv_check_bitstream,
.priv_class = &matroska_class,
};
#endif
#if CONFIG_WEBM_MUXER
static const AVClass webm_class = {
.class_name = "webm muxer",
.item_name = av_default_item_name,
.option = options,
.version = LIBAVUTIL_VERSION_INT,
};
AVOutputFormat ff_webm_muxer = {
.name = "webm",
.long_name = NULL_IF_CONFIG_SMALL("WebM"),
.mime_type = "video/webm",
.extensions = "webm",
.priv_data_size = sizeof(MatroskaMuxContext),
.audio_codec = CONFIG_LIBOPUS_ENCODER ? AV_CODEC_ID_OPUS : AV_CODEC_ID_VORBIS,
.video_codec = CONFIG_LIBVPX_VP9_ENCODER? AV_CODEC_ID_VP9 : AV_CODEC_ID_VP8,
.subtitle_codec = AV_CODEC_ID_WEBVTT,
.init = mkv_init,
.deinit = mkv_deinit,
.write_header = mkv_write_header,
.write_packet = mkv_write_flush_packet,
.write_trailer = mkv_write_trailer,
.query_codec = webm_query_codec,
.check_bitstream = mkv_check_bitstream,
.flags = AVFMT_GLOBALHEADER | AVFMT_VARIABLE_FPS |
AVFMT_TS_NONSTRICT | AVFMT_ALLOW_FLUSH,
.priv_class = &webm_class,
};
#endif
#if CONFIG_MATROSKA_AUDIO_MUXER
static const AVClass mka_class = {
.class_name = "matroska audio muxer",
.item_name = av_default_item_name,
.option = options,
.version = LIBAVUTIL_VERSION_INT,
};
AVOutputFormat ff_matroska_audio_muxer = {
.name = "matroska",
.long_name = NULL_IF_CONFIG_SMALL("Matroska Audio"),
.mime_type = "audio/x-matroska",
.extensions = "mka",
.priv_data_size = sizeof(MatroskaMuxContext),
.audio_codec = CONFIG_LIBVORBIS_ENCODER ?
AV_CODEC_ID_VORBIS : AV_CODEC_ID_AC3,
.video_codec = AV_CODEC_ID_NONE,
.init = mkv_init,
.deinit = mkv_deinit,
.write_header = mkv_write_header,
.write_packet = mkv_write_flush_packet,
.write_trailer = mkv_write_trailer,
.check_bitstream = mkv_check_bitstream,
.flags = AVFMT_GLOBALHEADER | AVFMT_TS_NONSTRICT |
AVFMT_ALLOW_FLUSH,
.codec_tag = (const AVCodecTag* const []){
ff_codec_wav_tags, additional_audio_tags, 0
},
.priv_class = &mka_class,
};
#endif