simgear/screen/: added support for video encoding.

Uses ffmpeg libraries to provide video encoding of an osg::GraphicsContext
(which could be for Flightgear's main window) to file.

The video codec is specified as a string at runtime. Success depends on what is
supported by the system's ffmpeg installation.

Handling of resize doesn't always work. We finish current stream and create a
new encoder and continue. Works ok with mpeg2, but with more modern codecs vlc
usually fails to handle the change.

We call the encoder on a separate thread to avoid latency.

If SG_FFMPEG is not defined, we don't make calls to ffmpeg libraries and
VideoEncoder class's constructor always throws an exception.
This commit is contained in:
Julian Smith 2021-11-14 22:58:25 +00:00
parent 0e13e48123
commit 5627b135b4
4 changed files with 831 additions and 1 deletions

View File

@ -5,6 +5,7 @@ set(HEADERS
extensions.hxx
screen-dump.hxx
tr.h
video-encoder.hxx
)
@ -12,6 +13,7 @@ set(SOURCES
extensions.cxx
screen-dump.cxx
tr.cxx
video-encoder.cxx
)
simgear_scene_component(screen screen "${SOURCES}" "${HEADERS}")
simgear_scene_component(screen screen "${SOURCES}" "${HEADERS}")

View File

@ -0,0 +1,531 @@
#pragma once
#ifndef SG_VIDEO_ENCODER_STANDALONE
#include <simgear/debug/logstream.hxx>
#endif
extern "C"
{
#include <libavcodec/avcodec.h>
#include <libavformat/avformat.h>
#include <libavutil/opt.h>
#include <libswscale/swscale.h>
}
#include <assert.h>
#include <iostream>
#include <sstream>
std::ostream& operator<< (std::ostream& out, const AVRational& r)
{
return out << "(" << r.num << "/" << r.den << ")";
}
/* Convenience exception type. Use like:
throw ExceptionStream << "something failed e=" << e;
*/
struct ExceptionStream : std::exception
{
mutable std::ostringstream m_buffer;
mutable std::string m_buffer2;
ExceptionStream()
{
}
ExceptionStream(const ExceptionStream& rhs)
{
m_buffer << rhs.m_buffer.str();
}
template<typename T>
const ExceptionStream& operator<< (const T& t) const
{
m_buffer << t;
return *this;
}
const char* what() const noexcept
{
if (m_buffer2 == "")
{
m_buffer2 = m_buffer.str();
}
return m_buffer2.c_str();
}
void prefix(const std::string& s)
{
what();
m_buffer2 = s + m_buffer2;
}
};
static const char* select(const char** names, int names_num, double select)
{
int i = select * names_num;
if (i == names_num) i -= 1;
return names[i];
}
/* Video encoder which uses ffmpeg libraries. */
struct FfmpegEncoder
{
double m_quality = 0;
double m_speed = 0;
int m_bitrate = 0;
struct SwsContext* m_sws_context = nullptr;
AVFrame* m_frame_yuv = nullptr;
AVCodec* m_codec = nullptr;
AVCodecContext* m_codec_context = nullptr;
AVStream* m_stream = nullptr;
AVPacket* m_packet = nullptr;
AVFormatContext* m_format_context = nullptr;
/* These are only used in exception text. */
std::string m_path;
std::string m_codec_name;
double m_t = 0;
int64_t m_t_int_prev = 0;
bool m_have_written_header = false;
/* Constructor.
Args:
path
Name of output file. Container type is inferred from suffix using
avformat_alloc_output_context2()). List of supported containers can
be found with 'ffmpeg -formats'.
codec_name
Name of codec, passed to avcodec_find_encoder_by_name(). List of
supported codecs can be found with 'ffmpeg -codecs'.
quality
Encoding quality in range 0..1 or -1 to use codec's default.
speed:
Encoding speed in range 0..1 or -1 to use codec's default.
bitrate
Target bitratae in bits/sec or -1 to use codex's default.
Throws exception if we cannot set up encoding, e.g. unrecognised
codec. Other configuration errors may be detected only when encode() is
called.
*/
FfmpegEncoder(
const std::string& path,
const std::string& codec_name,
double quality,
double speed,
int bitrate
)
{
assert(quality == -1 || (quality >= 0 && quality <= 1));
assert(speed == -1 || (speed >= 0 && speed <= 1));
m_path = path;
m_codec_name = codec_name;
m_quality = quality;
m_speed = speed;
m_bitrate = bitrate;
try
{
m_codec = avcodec_find_encoder_by_name(codec_name.c_str());
if (!m_codec)
{
throw ExceptionStream() << "avcodec_find_encoder_by_name() failed to find codec_name='"
<< codec_name << "'";
}
assert(m_codec);
avformat_alloc_output_context2(
&m_format_context,
nullptr /*oformat*/,
nullptr /*format_name*/,
path.c_str()
);
if (!m_format_context)
{
throw ExceptionStream() << "avformat_alloc_output_context2() failed to recognise path='"
<< path << "'";
}
assert(m_format_context);
if (!(m_format_context->oformat->flags & AVFMT_NOFILE))
{
int e = avio_open(&m_format_context->pb, path.c_str(), AVIO_FLAG_WRITE);
if (e < 0)
{
throw ExceptionStream() << "avio_open() failed, e=" << e;
}
assert(e >= 0);
}
m_stream = avformat_new_stream(m_format_context, nullptr);
if (!m_stream)
{
throw ExceptionStream() << "avformat_new_stream() returned null.";
}
assert(m_stream);
m_stream->id = m_format_context->nb_streams - 1;
}
catch (ExceptionStream& e)
{
e.prefix("Video encoding failed (path=" + m_path + " codec=" + m_codec_name + "): ");
clearall();
throw;
}
}
/* Used by constructor if throwing because error occured, and by
destructor. */
void clearall()
{
if (m_codec_context)
{
eof();
}
if (m_format_context && m_have_written_header)
{
av_write_trailer(m_format_context);
}
clear();
if (m_format_context)
{
avformat_free_context(m_format_context); /* Also frees m_stream. */
m_format_context = nullptr;
}
}
/* Destructor flushes any remaining encoded video and cleans up. */
~FfmpegEncoder()
{
clearall();
}
/* Clear state that depends on frame size. */
void clear()
{
av_frame_free(&m_frame_yuv);
sws_freeContext(m_sws_context);
m_sws_context = nullptr;
avcodec_free_context(&m_codec_context);
av_packet_free(&m_packet);
/* m_codec doesn't need freeing. */
}
/* Set state that depends on frame size. Throws if error occurs. */
void set(int width, int height)
{
try
{
int e;
/* Create YUV frame. */
m_frame_yuv = av_frame_alloc();
m_frame_yuv->format = AV_PIX_FMT_YUV420P;
m_frame_yuv->width = width;
m_frame_yuv->height = height;
e = av_frame_get_buffer(m_frame_yuv, 0);
if (e < 0)
{
throw ExceptionStream() << "av_frame_get_buffer() failed: " << e;
}
assert(e >= 0);
/* Create RGB => YUV converter. */
m_sws_context = sws_getContext(
width, height, AV_PIX_FMT_RGB24,
width, height, (enum AVPixelFormat) m_frame_yuv->format,
SWS_BILINEAR,
nullptr /*srcFilter*/,
nullptr /*dstFilter*/,
nullptr /*param*/
);
if (!m_sws_context)
{
throw ExceptionStream() << "sws_getContext() failed for " << width << "x" << height;
}
assert(m_sws_context);
/* Create codec context. */
m_codec_context = avcodec_alloc_context3(m_codec);
if (!m_codec_context)
{
throw ExceptionStream() << "avcodec_alloc_context3() failed";
}
m_codec_context->codec_id = m_codec->id;
if (m_bitrate > 0)
{
m_codec_context->bit_rate = m_bitrate;
}
SG_LOG(SG_GENERAL, SG_DEBUG, "m_codec_context->bit_rate=" << m_codec_context->bit_rate);
/* Resolution must be a multiple of two. */
m_codec_context->width = width / 2 * 2;
m_codec_context->height = height / 2 * 2;
m_codec_context->time_base = (AVRational){ 1, 60 };
//m_codec_context->gop_size = 12; /* emit one intra m_frame every twelve frames at most */
m_codec_context->pix_fmt = AV_PIX_FMT_YUV420P;
/* Some formats want m_stream headers to be separate. */
if (m_format_context->oformat->flags & AVFMT_GLOBALHEADER)
{
m_codec_context->flags |= AV_CODEC_FLAG_GLOBAL_HEADER;
}
/* Set dictionary entries. */
AVDictionary* dictionary = nullptr;
if (m_codec->id == AV_CODEC_ID_H264)
{
if (m_quality != -1)
{
/* -12-51, default 23.0. */
std::string q = std::to_string( (1-m_quality) * (51+12) - 12);
SG_LOG(SG_GENERAL, SG_ALERT, "crf m_quality=" << m_quality << " => " << q);
e = av_dict_set(&dictionary, "crf", q.c_str(), 0 /*flags*/); /* -12-51, default 23.0. */
}
}
if (m_codec->id == AV_CODEC_ID_H265)
{
/* Reduce verbose output. */
e = av_dict_set(&dictionary, "x265-params", "log-level=error", 0 /*flags*/);
assert(e >= 0);
if (m_quality != -1)
{
/* 0-51, default 28.0. */
std::string q = std::to_string( (1-m_quality) * 51);
SG_LOG(SG_GENERAL, SG_DEBUG, "crf m_quality=" << m_quality << " => " << q);
e = av_dict_set(&dictionary, "crf", q.c_str(), 0 /*flags*/);
assert(e >= 0);
}
}
if (m_codec->id == AV_CODEC_ID_THEORA)
{
SG_LOG(SG_GENERAL, SG_DEBUG, "AV_CODEC_ID_THEORA m_quality=" << m_quality);
if (m_quality != -1)
{
/* Enable constant quality mode. */
e = av_dict_set(&dictionary, "flags", "qscale", 0 /*flags*/);
assert(e >= 0);
/* Quality scaling is a little obscure in
https://ffmpeg.org/ffmpeg-codecs.html#libtheora, but this
appears to work with our m_quality's 0..1 range: */
std::string q = std::to_string(m_quality * FF_QP2LAMBDA * 10);
e = av_dict_set(&dictionary, "global_quality", q.c_str(), 0 /*flags*/);
assert(e >= 0);
}
}
if (m_codec->id == AV_CODEC_ID_H264 || m_codec->id == AV_CODEC_ID_H265)
{
if (m_speed != -1)
{
/* Set preset to string derived from m_speed. */
static const char* speeds[] = {
"veryslow",
"slower",
"slow",
"medium",
"fast",
"faster",
"veryfast",
"superfast",
"ultrafast",
};
const char* speed = select(speeds, sizeof(speeds) / sizeof(speeds[0]), m_speed);
SG_LOG(SG_GENERAL, SG_DEBUG, "preset: " << m_speed <<" => " << speed);
e = av_dict_set(&dictionary, "preset", speed, 0 /*flags*/);
assert(e >= 0);
}
}
/* Show dictionary. */
SG_LOG(SG_GENERAL, SG_ALERT, "dictionary " << av_dict_count(dictionary) << ":");
for (AVDictionaryEntry* t = av_dict_get(dictionary, "", nullptr, AV_DICT_IGNORE_SUFFIX);
t;
t = av_dict_get(dictionary, "", t, AV_DICT_IGNORE_SUFFIX)
)
{
SG_LOG(SG_GENERAL, SG_ALERT, " " << t->key << "=" << t->value);
}
e = avcodec_open2(m_codec_context, m_codec, &dictionary);
if (e < 0)
{
throw ExceptionStream() << "avcodec_open2() failed: " << e;
}
assert(e >= 0);
e = avcodec_parameters_from_context(m_stream->codecpar, m_codec_context);
if (e < 0)
{
throw ExceptionStream() << "avcodec_parameters_from_context() failed: " << e;
}
assert(e >= 0);
/* Send header. */
SG_LOG(SG_GENERAL, SG_DEBUG, "m_stream->time_base=" << m_stream->time_base);
/* This appears to override m_stream->time_base to be 1/90,000. */
e = avformat_write_header(m_format_context, &dictionary);
if (e < 0)
{
throw ExceptionStream() << "avformat_write_header() failed: " << e;
}
m_have_written_header = true;
/* Create packet. */
m_packet = av_packet_alloc();
if (!m_packet)
{
throw ExceptionStream() << "av_packet_alloc() failed";
}
}
catch (ExceptionStream& e)
{
e.prefix("Video encoding failed (path=" + m_path + " codec=" + m_codec_name + "): ");
clearall();
throw;
}
}
/* Sends new frame to encoder. Data must be RGB, 8 bits per channel. Will
throw if error occurs. */
void encode(int width, int height, int stride, void* input_raw, double dt)
{
if (!m_format_context)
{
throw ExceptionStream() << "Cannot encode after earlier error - m_format_context is null";
}
assert(dt > 0);
bool restart = false;
if (m_frame_yuv)
{
if (m_frame_yuv->width != width) restart = true;
if (m_frame_yuv->height != height) restart = true;
}
else
{
/* This is the first frame. */
restart = true;
}
if (restart)
{
if (m_frame_yuv)
{
/* Drain any remaining compressed data. */
eof();
}
/* Set up new pipeline. */
clear();
set(width, height);
}
int e = av_frame_make_writable(m_frame_yuv);
assert(e >= 0);
/* Convert raw_input (RGB24) to m_frame_yuv (YUV420P).
We also need to flip the image vertically so set m_input_linesize[0] to
-ve and make m_input_data[0] point to the last line in input_raw. */
int stride2 = -stride;
const uint8_t* input_raw2 = (const uint8_t*) input_raw + stride * (height-1);
sws_scale(
m_sws_context,
//(const uint8_t * const *)
&input_raw2,
&stride2,
0 /*srcSliceY*/,
height,
m_frame_yuv->data,
m_frame_yuv->linesize
);
/* Send m_frame_yuv to encoder. */
m_t += dt;
int64_t t_int = (int64_t) (m_t / m_codec_context->time_base.num * m_codec_context->time_base.den);
int64_t dt_int = t_int - m_t_int_prev;
m_t_int_prev = t_int;
SG_LOG(SG_GENERAL, SG_DEBUG, ""
<< " avcodec_send_frame()"
<< " dt=" << dt
<< " m_t=" << m_t
<< " m_codec_context->time_base=" << m_codec_context->time_base
<< " m_stream->time_base=" << m_stream->time_base
<< " dt_int=" << dt_int
<< " t_int=" << t_int
);
m_frame_yuv->pts = t_int;
e = avcodec_send_frame(m_codec_context, m_frame_yuv);
SG_LOG(SG_GENERAL, SG_DEBUG, "m_stream->time_base=" << m_stream->time_base);
assert(e >= 0);
/* Process any available encoded video data. */
drain();
}
/* End of video at current size. Not necessarily end of output - we create
new encoder when input size changes. */
void eof()
{
/* Send eof to m_codec_context and read final encoded data. */
assert(m_codec_context);
int e = avcodec_send_frame(m_codec_context, nullptr);
assert(e >= 0);
drain();
clear();
}
/* Read all available compressed data and send to m_format_context. Returns
0 on success or 1 at EOF. */
int drain()
{
if (!m_codec_context || !m_packet || !m_stream || !m_format_context)
{
return 1;
}
for(;;)
{
int e = avcodec_receive_packet(m_codec_context, m_packet);
if (e == AVERROR(EAGAIN))
{
SG_LOG(SG_GENERAL, SG_DEBUG, "AVERROR(EAGAIN)");
return 0;
}
if (e == AVERROR_EOF)
{
SG_LOG(SG_GENERAL, SG_DEBUG, "AVERROR_EOF");
return 1;
}
assert(e >= 0);
/* rescale output packet timestamp values from codec to m_stream timebase. */
av_packet_rescale_ts(m_packet, m_codec_context->time_base, m_stream->time_base);
SG_LOG(SG_GENERAL, SG_DEBUG, ""
<< " m_codec_context->time_base=" << m_codec_context->time_base
<< " m_stream->time_base=" << m_stream->time_base
<< " m_packet->pts=" << m_packet->pts
<< " m_packet->dts=" << m_packet->dts
);
m_packet->stream_index = m_stream->index;
/* Write the compressed data. */
e = av_interleaved_write_frame(m_format_context, m_packet);
assert(e >= 0);
/* packet is now blank (av_interleaved_write_frame() takes ownership of
* its contents and resets packet), so that no unreferencing is necessary.
* This would be different if one used av_write_frame(). */
}
}
};

View File

@ -0,0 +1,236 @@
#include "video-encoder.hxx"
#ifdef SG_FFMPEG
/* Video encoding is supported. */
#include "video-encoder-internal.hxx"
#include <simgear/debug/logstream.hxx>
#include <osg/Image>
#include <condition_variable>
#include <mutex>
#include <thread>
namespace simgear
{
/* Support for streaming video of gc's pixels to file. */
struct VideoEncoderInternal : osg::GraphicsOperation
{
osg::ref_ptr<osg::Image> m_image;
FfmpegEncoder m_ffmpeg_encoder;
double m_dt = 0;
int m_encoder_busy = 0; /**/
std::string m_exception;
std::mutex m_mutex;
std::condition_variable m_condition_variable;
std::thread m_thread;
/* See FfmpegEncoder constructor for API. */
VideoEncoderInternal(
const std::string& path,
const std::string& codec,
double quality,
double speed,
int bitrate
)
:
osg::GraphicsOperation("VideoEncoderOperation", false /*keep*/),
m_image(new osg::Image),
m_ffmpeg_encoder(path, codec, quality, speed, bitrate),
m_thread(
[this]
{
this->thread_fn();
}
)
{
}
void thread_fn()
{
/* We repeatedly wait for a new frame to be available, signalled by
m_encoder_busy being +1. */
SG_LOG(SG_GENERAL, SG_DEBUG, "thread_fn() starting");
for(;;)
{
std::unique_lock<std::mutex> lock(m_mutex);
if (m_encoder_busy == -1)
{
/* This is us being told to quit. */
break;
}
else if (m_encoder_busy == 1)
{
/* New frame needs encoding. */
try
{
m_ffmpeg_encoder.encode(
m_image->s(),
m_image->t(),
3 * m_image->s() /*stride*/,
m_image->data(),
m_dt
);
}
catch (std::exception& e)
{
SG_LOG(SG_GENERAL, SG_ALERT, "Caught exception: " << e.what());
m_exception = e.what();
m_encoder_busy = -1;
m_condition_variable.notify_all();
break;
}
m_encoder_busy = 0;
m_condition_variable.notify_all();
}
else
{
m_condition_variable.wait(lock);
}
}
SG_LOG(SG_GENERAL, SG_DEBUG, "thread_fn() returning");
}
void operator() (osg::GraphicsContext* gc)
{
/* Called by OSG when <gc> is ready. We wait for any existing frame
encode to finish, then encode <gs>'s frame. */
std::unique_lock<std::mutex> lock(m_mutex);
for(;;)
{
if (m_encoder_busy == -1)
{
/* Encoding failed or finished. */
SG_LOG(SG_GENERAL, SG_ALERT, "operator(): m_encoder_busy=" << m_encoder_busy);
break;
}
else if (m_encoder_busy == 0)
{
m_image->readPixels(
0,
0,
gc->getTraits()->width,
gc->getTraits()->height,
GL_RGB,
GL_UNSIGNED_BYTE
);
m_encoder_busy = 1;
m_condition_variable.notify_all();
break;
}
else
{
m_condition_variable.wait(lock);
}
}
}
void encode(double dt, osg::GraphicsContext* gc)
{
std::unique_lock<std::mutex> lock(m_mutex);
if (!m_exception.empty())
{
throw std::runtime_error(m_exception);
}
assert(dt != 0);
m_dt = dt;
gc->add(this);
}
~VideoEncoderInternal()
{
{
/* Set m_encoder_busy to -1 to stop our thread. */
std::unique_lock<std::mutex> lock(m_mutex);
m_encoder_busy = -1;
m_condition_variable.notify_all();
}
m_thread.join();
}
};
static void av_log(void* /*avcl*/, int level, const char* format, va_list va)
{
if (level < 0) return;
sgDebugPriority sglevel;
if (level < 20) sglevel = SG_ALERT;
if (level < 28) sglevel = SG_WARN;
if (level < 36) sglevel = SG_INFO;
if (level < 44) sglevel = SG_DEBUG;
if (level < 54) sglevel = SG_BULK;
else sglevel = SG_BULK;
char* message = nullptr;
vasprintf(&message, format, va);
SG_LOG(SG_VIEW, sglevel, "level=" << level << ": " << message);
free(message);
}
VideoEncoder::VideoEncoder(
const std::string& path,
const std::string& codec,
double quality,
double speed,
int bitrate
)
{
av_log_set_callback(av_log);
m_internal = new VideoEncoderInternal(path, codec, quality, speed, bitrate);
}
VideoEncoder::~VideoEncoder()
{
}
void VideoEncoder::encode(double dt, osg::GraphicsContext* gc)
{
if (dt == 0)
{
SG_LOG(SG_GENERAL, SG_ALERT, "Ignoring frame because dt is zero");
return;
}
m_internal->encode(dt, gc);
}
} /* namespace simgear. */
#else
/* SG_FFMPEG not defined - video encoding is not supported. */
namespace simgear
{
struct VideoEncoderInternal : osg::GraphicsOperation
{
};
VideoEncoder::VideoEncoder(
const std::string& path,
const std::string& codec,
double quality,
double speed,
int bitrate
)
{
throw std::runtime_error("Video encoding is not available in this build of Flightgear");
}
VideoEncoder::~VideoEncoder()
{
}
void VideoEncoder::encode(double dt, osg::GraphicsContext* gc)
{
throw std::runtime_error("Video encoding is not available in this build of Flightgear");
}
} /* namespace simgear. */
#endif

View File

@ -0,0 +1,61 @@
#pragma once
#include <memory>
#include <string>
#include <osg/GraphicsContext>
namespace simgear
{
/* Compressed video encoder.
Generated video contains information about frame times, and also copes with
changes to the width and/or height of the frames.
So replay will replicate variable frame rates and window resizing. */
struct VideoEncoder
{
/* Constructor; sets things up to write compressed video to file <path>.
Args:
path
Name of output file. Container type is inferred from suffix using
avformat_alloc_output_context2()). List of supported containers can
be found with 'ffmpeg -formats'.
codec_name
Name of codec, passed to avcodec_find_encoder_by_name(). List of
supported codecs can be found with 'ffmpeg -codecs'.
quality
Encoding quality in range 0..1 or -1 to use codec's default.
speed:
Encoding speed in range 0..1 or -1 to use codec's default.
bitrate
Target bitratae in bits/sec or -1 to use codex's default.
Throws exception if we cannot set up encoding, e.g. unrecognised
codec. Other configuration errors may be detected only when encode() is
called.
*/
VideoEncoder(
const std::string& path,
const std::string& codec,
double quality,
double speed,
int bitrate
);
/* Appends gc's current bitmap to compressed video. Works by scheduling a
callback with gc->add(). <dt> must be non-zero.
Throws exception if error has occurred previously - for example sometimes a
configuration doesn't fail until we start sending frames. */
void encode(double dt, osg::GraphicsContext* gc);
~VideoEncoder();
private:
osg::ref_ptr<struct VideoEncoderInternal> m_internal;
};
}