/*
- Copyright (C) 2017-2018 Carl Hetherington <cth@carlh.net>
+ Copyright (C) 2017-2021 Carl Hetherington <cth@carlh.net>
This file is part of DCP-o-matic.
*/
+
+#include "compose.hpp"
+#include "cross.h"
#include "ffmpeg_encoder.h"
+#include "ffmpeg_wrapper.h"
#include "film.h"
+#include "image.h"
#include "job.h"
+#include "log.h"
#include "player.h"
#include "player_video.h"
-#include "log.h"
-#include "image.h"
-#include "cross.h"
-#include "butler.h"
-#include "compose.hpp"
+extern "C" {
+#include <libavutil/channel_layout.h>
+}
#include <iostream>
#include "i18n.h"
-using std::string;
-using std::runtime_error;
+
using std::cout;
-using std::pair;
-using boost::shared_ptr;
+using std::make_shared;
+using std::shared_ptr;
+using std::string;
using boost::bind;
-using boost::weak_ptr;
+using namespace dcpomatic;
+#if BOOST_VERSION >= 106100
+using namespace boost::placeholders;
+#endif
+
int FFmpegFileEncoder::_video_stream_index = 0;
-int FFmpegFileEncoder::_audio_stream_index = 1;
+int FFmpegFileEncoder::_audio_stream_index_base = 1;
+
-static AVPixelFormat
-force_pixel_format (AVPixelFormat, AVPixelFormat out)
+class ExportAudioStream
{
- return out;
-}
+public:
+ ExportAudioStream (string codec_name, int channels, int frame_rate, AVSampleFormat sample_format, AVFormatContext* format_context, int stream_index)
+ : _format_context (format_context)
+ , _stream_index (stream_index)
+ {
+ _codec = avcodec_find_encoder_by_name (codec_name.c_str());
+ if (!_codec) {
+ throw EncodeError (String::compose("avcodec_find_encoder_by_name failed for %1", codec_name));
+ }
+
+ _codec_context = avcodec_alloc_context3 (_codec);
+ if (!_codec_context) {
+ throw std::bad_alloc ();
+ }
+
+ /* XXX: configurable */
+ _codec_context->bit_rate = channels * 128 * 1024;
+ _codec_context->sample_fmt = sample_format;
+ _codec_context->sample_rate = frame_rate;
+ _codec_context->channel_layout = av_get_default_channel_layout (channels);
+ _codec_context->channels = channels;
+
+ int r = avcodec_open2 (_codec_context, _codec, 0);
+ if (r < 0) {
+ throw EncodeError (N_("avcodec_open2"), N_("ExportAudioStream::ExportAudioStream"), r);
+ }
+
+ _stream = avformat_new_stream (format_context, _codec);
+ if (!_stream) {
+ throw EncodeError (N_("avformat_new_stream"), N_("ExportAudioStream::ExportAudioStream"));
+ }
+
+ _stream->id = stream_index;
+ _stream->disposition |= AV_DISPOSITION_DEFAULT;
+ r = avcodec_parameters_from_context (_stream->codecpar, _codec_context);
+ if (r < 0) {
+ throw EncodeError (N_("avcodec_parameters_from_context"), N_("ExportAudioStream::ExportAudioStream"), r);
+ }
+ }
+
+ ~ExportAudioStream ()
+ {
+ avcodec_close (_codec_context);
+ }
+
+ ExportAudioStream (ExportAudioStream const&) = delete;
+ ExportAudioStream& operator= (ExportAudioStream const&) = delete;
+
+ int frame_size () const {
+ return _codec_context->frame_size;
+ }
+
+ bool flush ()
+ {
+ int r = avcodec_send_frame (_codec_context, nullptr);
+ if (r < 0 && r != AVERROR_EOF) {
+ /* We get EOF if we've already flushed the stream once */
+ throw EncodeError (N_("avcodec_send_frame"), N_("ExportAudioStream::flush"), r);
+ }
+
+ ffmpeg::Packet packet;
+ r = avcodec_receive_packet (_codec_context, packet.get());
+ if (r == AVERROR_EOF) {
+ return true;
+ } else if (r < 0) {
+ throw EncodeError (N_("avcodec_receive_packet"), N_("ExportAudioStream::flush"), r);
+ }
+
+ packet->stream_index = _stream_index;
+ av_interleaved_write_frame (_format_context, packet.get());
+ return false;
+ }
+
+ void write (int size, int channel_offset, int channels, float* const* data, int64_t sample_offset)
+ {
+ DCPOMATIC_ASSERT (size);
+
+ auto frame = av_frame_alloc ();
+ DCPOMATIC_ASSERT (frame);
+
+ int line_size;
+ int const buffer_size = av_samples_get_buffer_size (&line_size, channels, size, _codec_context->sample_fmt, 0);
+ DCPOMATIC_ASSERT (buffer_size >= 0);
+
+ auto samples = av_malloc (buffer_size);
+ DCPOMATIC_ASSERT (samples);
+
+ frame->nb_samples = size;
+ frame->format = _codec_context->sample_fmt;
+ frame->channels = channels;
+ int r = avcodec_fill_audio_frame (frame, channels, _codec_context->sample_fmt, (const uint8_t *) samples, buffer_size, 0);
+ DCPOMATIC_ASSERT (r >= 0);
+
+ switch (_codec_context->sample_fmt) {
+ case AV_SAMPLE_FMT_S16:
+ {
+ int16_t* q = reinterpret_cast<int16_t*> (samples);
+ for (int i = 0; i < size; ++i) {
+ for (int j = 0; j < channels; ++j) {
+ *q++ = data[j + channel_offset][i] * 32767;
+ }
+ }
+ break;
+ }
+ case AV_SAMPLE_FMT_S32:
+ {
+ int32_t* q = reinterpret_cast<int32_t*> (samples);
+ for (int i = 0; i < size; ++i) {
+ for (int j = 0; j < channels; ++j) {
+ *q++ = data[j + channel_offset][i] * 2147483647;
+ }
+ }
+ break;
+ }
+ case AV_SAMPLE_FMT_FLTP:
+ {
+ for (int i = 0; i < channels; ++i) {
+ memcpy (reinterpret_cast<float*>(static_cast<uint8_t*>(samples) + i * line_size), data[i + channel_offset], sizeof(float) * size);
+ }
+ break;
+ }
+ default:
+ DCPOMATIC_ASSERT (false);
+ }
+
+ DCPOMATIC_ASSERT (_codec_context->time_base.num == 1);
+ frame->pts = sample_offset * _codec_context->time_base.den / _codec_context->sample_rate;
+
+ r = avcodec_send_frame (_codec_context, frame);
+ av_free (samples);
+ av_frame_free (&frame);
+ if (r < 0) {
+ throw EncodeError (N_("avcodec_send_frame"), N_("ExportAudioStream::write"), r);
+ }
+
+ ffmpeg::Packet packet;
+ r = avcodec_receive_packet (_codec_context, packet.get());
+ if (r < 0 && r != AVERROR(EAGAIN)) {
+ throw EncodeError (N_("avcodec_receive_packet"), N_("ExportAudioStream::write"), r);
+ } else if (r >= 0) {
+ packet->stream_index = _stream_index;
+ av_interleaved_write_frame (_format_context, packet.get());
+ }
+ }
+
+private:
+ AVFormatContext* _format_context;
+ AVCodec const * _codec;
+ AVCodecContext* _codec_context;
+ AVStream* _stream;
+ int _stream_index;
+};
+
FFmpegFileEncoder::FFmpegFileEncoder (
dcp::Size video_frame_size,
int video_frame_rate,
int audio_frame_rate,
int channels,
- shared_ptr<Log> log,
ExportFormat format,
+ bool audio_stream_per_channel,
int x264_crf,
boost::filesystem::path output
)
- : _video_options (0)
+ : _audio_stream_per_channel (audio_stream_per_channel)
, _audio_channels (channels)
, _output (output)
, _video_frame_size (video_frame_size)
, _video_frame_rate (video_frame_rate)
, _audio_frame_rate (audio_frame_rate)
- , _log (log)
{
+ _pixel_format = pixel_format (format);
+
switch (format) {
- case EXPORT_FORMAT_PRORES:
- _pixel_format = AV_PIX_FMT_YUV422P10;
+ case ExportFormat::PRORES_HQ:
_sample_format = AV_SAMPLE_FMT_S16;
_video_codec_name = "prores_ks";
_audio_codec_name = "pcm_s16le";
av_dict_set (&_video_options, "profile", "3", 0);
av_dict_set (&_video_options, "threads", "auto", 0);
break;
- case EXPORT_FORMAT_H264:
- _pixel_format = AV_PIX_FMT_YUV420P;
+ case ExportFormat::H264_AAC:
_sample_format = AV_SAMPLE_FMT_FLTP;
_video_codec_name = "libx264";
_audio_codec_name = "aac";
av_dict_set_int (&_video_options, "crf", x264_crf, 0);
break;
+ default:
+ DCPOMATIC_ASSERT (false);
+ }
+
+ int r = avformat_alloc_output_context2 (&_format_context, 0, 0, _output.string().c_str());
+ if (!_format_context) {
+ throw EncodeError (N_("avformat_alloc_output_context2"), "FFmpegFileEncoder::FFmpegFileEncoder", r);
}
setup_video ();
setup_audio ();
- avformat_alloc_output_context2 (&_format_context, 0, 0, _output.string().c_str());
- if (!_format_context) {
- throw runtime_error ("could not allocate FFmpeg format context");
+ r = avio_open_boost (&_format_context->pb, _output, AVIO_FLAG_WRITE);
+ if (r < 0) {
+ throw EncodeError (String::compose(_("Could not open output file %1 (%2)"), _output.string(), r));
}
- _video_stream = avformat_new_stream (_format_context, _video_codec);
- if (!_video_stream) {
- throw runtime_error ("could not create FFmpeg output video stream");
- }
+ AVDictionary* options = nullptr;
- _audio_stream = avformat_new_stream (_format_context, _audio_codec);
- if (!_audio_stream) {
- throw runtime_error ("could not create FFmpeg output audio stream");
+ r = avformat_write_header (_format_context, &options);
+ if (r < 0) {
+ throw EncodeError (N_("avformat_write_header"), N_("FFmpegFileEncoder::FFmpegFileEncoder"), r);
}
- _video_stream->id = _video_stream_index;
- _video_stream->codec = _video_codec_context;
+ _pending_audio = make_shared<AudioBuffers>(channels, 0);
+}
- _audio_stream->id = _audio_stream_index;
- _audio_stream->codec = _audio_codec_context;
- if (avcodec_open2 (_video_codec_context, _video_codec, &_video_options) < 0) {
- throw runtime_error ("could not open FFmpeg video codec");
- }
-
- int r = avcodec_open2 (_audio_codec_context, _audio_codec, 0);
- if (r < 0) {
- char buffer[256];
- av_strerror (r, buffer, sizeof(buffer));
- throw runtime_error (String::compose ("could not open FFmpeg audio codec (%1)", buffer));
- }
+FFmpegFileEncoder::~FFmpegFileEncoder ()
+{
+ _audio_streams.clear ();
+ avcodec_close (_video_codec_context);
+ avformat_free_context (_format_context);
+}
- if (avio_open_boost (&_format_context->pb, _output, AVIO_FLAG_WRITE) < 0) {
- throw runtime_error ("could not open FFmpeg output file");
- }
- if (avformat_write_header (_format_context, 0) < 0) {
- throw runtime_error ("could not write header to FFmpeg output file");
+AVPixelFormat
+FFmpegFileEncoder::pixel_format (ExportFormat format)
+{
+ switch (format) {
+ case ExportFormat::PRORES_HQ:
+ return AV_PIX_FMT_YUV422P10;
+ case ExportFormat::H264_AAC:
+ return AV_PIX_FMT_YUV420P;
+ default:
+ DCPOMATIC_ASSERT (false);
}
- _pending_audio.reset (new AudioBuffers(channels, 0));
+ return AV_PIX_FMT_YUV422P10;
}
+
void
FFmpegFileEncoder::setup_video ()
{
_video_codec = avcodec_find_encoder_by_name (_video_codec_name.c_str());
if (!_video_codec) {
- throw runtime_error (String::compose ("could not find FFmpeg encoder %1", _video_codec_name));
+ throw EncodeError (String::compose("avcodec_find_encoder_by_name failed for %1", _video_codec_name));
}
_video_codec_context = avcodec_alloc_context3 (_video_codec);
if (!_video_codec_context) {
- throw runtime_error ("could not allocate FFmpeg video context");
+ throw std::bad_alloc ();
}
- avcodec_get_context_defaults3 (_video_codec_context, _video_codec);
-
/* Variable quantisation */
_video_codec_context->global_quality = 0;
_video_codec_context->width = _video_frame_size.width;
_video_codec_context->time_base = (AVRational) { 1, _video_frame_rate };
_video_codec_context->pix_fmt = _pixel_format;
_video_codec_context->flags |= AV_CODEC_FLAG_QSCALE | AV_CODEC_FLAG_GLOBAL_HEADER;
-}
-void
-FFmpegFileEncoder::setup_audio ()
-{
- _audio_codec = avcodec_find_encoder_by_name (_audio_codec_name.c_str());
- if (!_audio_codec) {
- throw runtime_error (String::compose ("could not find FFmpeg encoder %1", _audio_codec_name));
+ if (avcodec_open2 (_video_codec_context, _video_codec, &_video_options) < 0) {
+ throw EncodeError (N_("avcodec_open"), N_("FFmpegFileEncoder::setup_video"));
}
- _audio_codec_context = avcodec_alloc_context3 (_audio_codec);
- if (!_audio_codec_context) {
- throw runtime_error ("could not allocate FFmpeg audio context");
+ _video_stream = avformat_new_stream (_format_context, _video_codec);
+ if (!_video_stream) {
+ throw EncodeError (N_("avformat_new_stream"), N_("FFmpegFileEncoder::setup_video"));
}
- avcodec_get_context_defaults3 (_audio_codec_context, _audio_codec);
+ _video_stream->id = _video_stream_index;
+ int r = avcodec_parameters_from_context (_video_stream->codecpar, _video_codec_context);
+ if (r < 0) {
+ throw EncodeError (N_("avcodec_parameters_from_context"), N_("FFmpegFileEncoder::setup_video"), r);
+ }
+}
+
- /* XXX: configurable */
- _audio_codec_context->bit_rate = 256 * 1024;
- _audio_codec_context->sample_fmt = _sample_format;
- _audio_codec_context->sample_rate = _audio_frame_rate;
- _audio_codec_context->channel_layout = av_get_default_channel_layout (_audio_channels);
- _audio_codec_context->channels = _audio_channels;
+void
+FFmpegFileEncoder::setup_audio ()
+{
+ int const streams = _audio_stream_per_channel ? _audio_channels : 1;
+ int const channels_per_stream = _audio_stream_per_channel ? 1 : _audio_channels;
+
+ for (int i = 0; i < streams; ++i) {
+ _audio_streams.push_back(
+ make_shared<ExportAudioStream>(
+ _audio_codec_name, channels_per_stream, _audio_frame_rate, _sample_format, _format_context, _audio_stream_index_base + i
+ )
+ );
+ }
}
+
void
FFmpegFileEncoder::flush ()
{
bool flushed_audio = false;
while (!flushed_video || !flushed_audio) {
- AVPacket packet;
- av_init_packet (&packet);
- packet.data = 0;
- packet.size = 0;
-
- int got_packet;
- avcodec_encode_video2 (_video_codec_context, &packet, 0, &got_packet);
- if (got_packet) {
- packet.stream_index = 0;
- av_interleaved_write_frame (_format_context, &packet);
- } else {
- flushed_video = true;
+ int r = avcodec_send_frame (_video_codec_context, nullptr);
+ if (r < 0 && r != AVERROR_EOF) {
+ /* We get EOF if we've already flushed the stream once */
+ throw EncodeError (N_("avcodec_send_frame"), N_("FFmpegFileEncoder::flush"), r);
}
- av_packet_unref (&packet);
-
- av_init_packet (&packet);
- packet.data = 0;
- packet.size = 0;
- avcodec_encode_audio2 (_audio_codec_context, &packet, 0, &got_packet);
- if (got_packet) {
- packet.stream_index = 0;
- av_interleaved_write_frame (_format_context, &packet);
+ ffmpeg::Packet packet;
+ r = avcodec_receive_packet (_video_codec_context, packet.get());
+ if (r == AVERROR_EOF) {
+ flushed_video = true;
+ } else if (r < 0) {
+ throw EncodeError (N_("avcodec_receive_packet"), N_("FFmpegFileEncoder::flush"), r);
} else {
- flushed_audio = true;
+ packet->stream_index = _video_stream_index;
+ av_interleaved_write_frame (_format_context, packet.get());
+ }
+
+ flushed_audio = true;
+ for (auto i: _audio_streams) {
+ if (!i->flush()) {
+ flushed_audio = false;
+ }
}
- av_packet_unref (&packet);
}
av_write_trailer (_format_context);
-
- avcodec_close (_video_codec_context);
- avcodec_close (_audio_codec_context);
- avio_close (_format_context->pb);
- avformat_free_context (_format_context);
}
+
void
FFmpegFileEncoder::video (shared_ptr<PlayerVideo> video, DCPTime time)
{
- shared_ptr<Image> image = video->image (
- boost::optional<dcp::NoteHandler>(bind(&Log::dcp_log, _log.get(), _1, _2)),
- bind (&force_pixel_format, _1, _pixel_format),
- true,
+ /* All our output formats are video range at the moment */
+ auto image = video->image (
+ bind (&PlayerVideo::force, _pixel_format),
+ VideoRange::VIDEO,
false
);
- AVFrame* frame = av_frame_alloc ();
+ auto frame = av_frame_alloc ();
DCPOMATIC_ASSERT (frame);
- _pending_images[image->data()[0]] = image;
+ {
+ boost::mutex::scoped_lock lm (_pending_images_mutex);
+ _pending_images[image->data()[0]] = image;
+ }
+
for (int i = 0; i < 3; ++i) {
- AVBufferRef* buffer = av_buffer_create(image->data()[i], image->stride()[i] * image->size().height, &buffer_free, this, 0);
+ auto buffer = av_buffer_create(image->data()[i], image->stride()[i] * image->size().height, &buffer_free, this, 0);
frame->buf[i] = av_buffer_ref (buffer);
frame->data[i] = buffer->data;
frame->linesize[i] = image->stride()[i];
frame->width = image->size().width;
frame->height = image->size().height;
frame->format = _pixel_format;
- frame->pts = time.seconds() / av_q2d (_video_stream->time_base);
-
- AVPacket packet;
- av_init_packet (&packet);
- packet.data = 0;
- packet.size = 0;
+ DCPOMATIC_ASSERT (_video_stream->time_base.num == 1);
+ frame->pts = time.get() * _video_stream->time_base.den / DCPTime::HZ;
- int got_packet;
- if (avcodec_encode_video2 (_video_codec_context, &packet, frame, &got_packet) < 0) {
- throw EncodeError ("FFmpeg video encode failed");
+ int r = avcodec_send_frame (_video_codec_context, frame);
+ av_frame_free (&frame);
+ if (r < 0) {
+ throw EncodeError (N_("avcodec_send_frame"), N_("FFmpegFileEncoder::video"), r);
}
- if (got_packet && packet.size) {
- packet.stream_index = _video_stream_index;
- av_interleaved_write_frame (_format_context, &packet);
- av_packet_unref (&packet);
+ ffmpeg::Packet packet;
+ r = avcodec_receive_packet (_video_codec_context, packet.get());
+ if (r < 0 && r != AVERROR(EAGAIN)) {
+ throw EncodeError (N_("avcodec_receive_packet"), N_("FFmpegFileEncoder::video"), r);
+ } else if (r >= 0) {
+ packet->stream_index = _video_stream_index;
+ av_interleaved_write_frame (_format_context, packet.get());
}
-
- av_frame_free (&frame);
-
}
+
/** Called when the player gives us some audio */
void
FFmpegFileEncoder::audio (shared_ptr<AudioBuffers> audio)
{
_pending_audio->append (audio);
- int frame_size = _audio_codec_context->frame_size;
+ DCPOMATIC_ASSERT (!_audio_streams.empty());
+ int frame_size = _audio_streams[0]->frame_size();
if (frame_size == 0) {
/* codec has AV_CODEC_CAP_VARIABLE_FRAME_SIZE */
frame_size = _audio_frame_rate / _video_frame_rate;
}
}
+
void
FFmpegFileEncoder::audio_frame (int size)
{
- DCPOMATIC_ASSERT (size);
-
- AVFrame* frame = av_frame_alloc ();
- DCPOMATIC_ASSERT (frame);
-
- int const channels = _pending_audio->channels();
- DCPOMATIC_ASSERT (channels);
-
- int const buffer_size = av_samples_get_buffer_size (0, channels, size, _audio_codec_context->sample_fmt, 0);
- DCPOMATIC_ASSERT (buffer_size >= 0);
-
- void* samples = av_malloc (buffer_size);
- DCPOMATIC_ASSERT (samples);
-
- frame->nb_samples = size;
- int r = avcodec_fill_audio_frame (frame, channels, _audio_codec_context->sample_fmt, (const uint8_t *) samples, buffer_size, 0);
- DCPOMATIC_ASSERT (r >= 0);
-
- float** p = _pending_audio->data ();
- switch (_audio_codec_context->sample_fmt) {
- case AV_SAMPLE_FMT_S16:
- {
- int16_t* q = reinterpret_cast<int16_t*> (samples);
- for (int i = 0; i < size; ++i) {
- for (int j = 0; j < channels; ++j) {
- *q++ = p[j][i] * 32767;
- }
+ if (_audio_stream_per_channel) {
+ int offset = 0;
+ for (auto i: _audio_streams) {
+ i->write (size, offset, 1, _pending_audio->data(), _audio_frames);
+ ++offset;
}
- break;
- }
- case AV_SAMPLE_FMT_FLTP:
- {
- float* q = reinterpret_cast<float*> (samples);
- for (int i = 0; i < channels; ++i) {
- memcpy (q, p[i], sizeof(float) * size);
- q += size;
- }
- break;
- }
- default:
- DCPOMATIC_ASSERT (false);
+ } else {
+ DCPOMATIC_ASSERT (!_audio_streams.empty());
+ DCPOMATIC_ASSERT (_pending_audio->channels());
+ _audio_streams[0]->write (size, 0, _pending_audio->channels(), _pending_audio->data(), _audio_frames);
}
- AVPacket packet;
- av_init_packet (&packet);
- packet.data = 0;
- packet.size = 0;
-
- int got_packet;
- if (avcodec_encode_audio2 (_audio_codec_context, &packet, frame, &got_packet) < 0) {
- throw EncodeError ("FFmpeg audio encode failed");
- }
-
- if (got_packet && packet.size) {
- packet.stream_index = _audio_stream_index;
- av_interleaved_write_frame (_format_context, &packet);
- av_packet_unref (&packet);
- }
-
- av_free (samples);
- av_frame_free (&frame);
-
_pending_audio->trim_start (size);
+ _audio_frames += size;
}
+
void
FFmpegFileEncoder::subtitle (PlayerText, DCPTimePeriod)
{
}
+
void
FFmpegFileEncoder::buffer_free (void* opaque, uint8_t* data)
{
reinterpret_cast<FFmpegFileEncoder*>(opaque)->buffer_free2(data);
}
+
void
FFmpegFileEncoder::buffer_free2 (uint8_t* data)
{
- _pending_images.erase (data);
+ boost::mutex::scoped_lock lm (_pending_images_mutex);
+ if (_pending_images.find(data) != _pending_images.end()) {
+ _pending_images.erase (data);
+ }
}
projects / dcpomatic.git / blobdiff