-/* -*- c-basic-offset: 8; default-tab-width: 8; -*- */
-
/*
Copyright (C) 2012 Carl Hetherington <cth@carlh.net>
#include <iostream>
#include <stdint.h>
#include <boost/lexical_cast.hpp>
+#include <sndfile.h>
extern "C" {
#include <libavcodec/avcodec.h>
#include <libavformat/avformat.h>
-#include <libswscale/swscale.h>
-#include <libpostproc/postprocess.h>
}
-#include <sndfile.h>
#include "film.h"
#include "filter.h"
#include "exceptions.h"
#include "log.h"
#include "ffmpeg_decoder.h"
#include "filter_graph.h"
-#include "subtitle.h"
#include "audio_buffers.h"
#include "i18n.h"
using boost::dynamic_pointer_cast;
using libdcp::Size;
-boost::mutex FFmpegDecoder::_mutex;
-
-FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegContent> c, bool video, bool audio, bool subtitles)
+FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegContent> c, bool video, bool audio)
: Decoder (f)
- , VideoDecoder (f, c)
- , AudioDecoder (f, c)
- , _ffmpeg_content (c)
- , _format_context (0)
- , _video_stream (-1)
- , _frame (0)
- , _video_codec_context (0)
- , _video_codec (0)
- , _audio_codec_context (0)
- , _audio_codec (0)
+ , VideoDecoder (f)
+ , AudioDecoder (f)
+ , SubtitleDecoder (f)
+ , FFmpeg (c)
, _subtitle_codec_context (0)
, _subtitle_codec (0)
, _decode_video (video)
, _decode_audio (audio)
- , _decode_subtitles (subtitles)
+ , _pts_offset (0)
+ , _just_sought (false)
{
- setup_general ();
- setup_video ();
- setup_audio ();
setup_subtitle ();
-}
-
-FFmpegDecoder::~FFmpegDecoder ()
-{
- boost::mutex::scoped_lock lm (_mutex);
-
- if (_audio_codec_context) {
- avcodec_close (_audio_codec_context);
- }
-
- if (_video_codec_context) {
- avcodec_close (_video_codec_context);
- }
-
- if (_subtitle_codec_context) {
- avcodec_close (_subtitle_codec_context);
- }
-
- av_free (_frame);
-
- avformat_close_input (&_format_context);
-}
-
-void
-FFmpegDecoder::setup_general ()
-{
- av_register_all ();
-
- if (avformat_open_input (&_format_context, _ffmpeg_content->file().string().c_str(), 0, 0) < 0) {
- throw OpenFileError (_ffmpeg_content->file().string ());
- }
-
- if (avformat_find_stream_info (_format_context, 0) < 0) {
- throw DecodeError (_("could not find stream information"));
- }
-
- /* Find video, audio and subtitle streams */
-
- for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
- AVStream* s = _format_context->streams[i];
- if (s->codec->codec_type == AVMEDIA_TYPE_VIDEO) {
- _video_stream = i;
- } else if (s->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
-
- /* This is a hack; sometimes it seems that _audio_codec_context->channel_layout isn't set up,
- so bodge it here. No idea why we should have to do this.
- */
-
- if (s->codec->channel_layout == 0) {
- s->codec->channel_layout = av_get_default_channel_layout (s->codec->channels);
- }
-
- _audio_streams.push_back (
- shared_ptr<FFmpegAudioStream> (
- new FFmpegAudioStream (stream_name (s), i, s->codec->sample_rate, s->codec->channels)
- )
- );
-
- } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) {
- _subtitle_streams.push_back (shared_ptr<FFmpegSubtitleStream> (new FFmpegSubtitleStream (stream_name (s), i)));
- }
- }
-
- if (_video_stream < 0) {
- throw DecodeError (N_("could not find video stream"));
- }
-
- _frame = avcodec_alloc_frame ();
- if (_frame == 0) {
- throw DecodeError (N_("could not allocate frame"));
- }
-}
-
-void
-FFmpegDecoder::setup_video ()
-{
- boost::mutex::scoped_lock lm (_mutex);
-
- _video_codec_context = _format_context->streams[_video_stream]->codec;
- _video_codec = avcodec_find_decoder (_video_codec_context->codec_id);
- if (_video_codec == 0) {
- throw DecodeError (_("could not find video decoder"));
- }
-
- if (avcodec_open2 (_video_codec_context, _video_codec, 0) < 0) {
- throw DecodeError (N_("could not open video decoder"));
+ if (video && audio && c->audio_stream() && c->first_video() && c->audio_stream()->first_audio) {
+ _pts_offset = compute_pts_offset (c->first_video().get(), c->audio_stream()->first_audio.get(), c->video_frame_rate());
}
}
-void
-FFmpegDecoder::setup_audio ()
+double
+FFmpegDecoder::compute_pts_offset (double first_video, double first_audio, float video_frame_rate)
{
- boost::mutex::scoped_lock lm (_mutex);
+ double const old_first_video = first_video;
- if (!_ffmpeg_content->audio_stream ()) {
- return;
- }
-
- _audio_codec_context = _format_context->streams[_ffmpeg_content->audio_stream()->id]->codec;
- _audio_codec = avcodec_find_decoder (_audio_codec_context->codec_id);
-
- if (_audio_codec == 0) {
- throw DecodeError (_("could not find audio decoder"));
+ /* Round the first video to a frame boundary */
+ if (fabs (rint (first_video * video_frame_rate) - first_video * video_frame_rate) > 1e-6) {
+ first_video = ceil (first_video * video_frame_rate) / video_frame_rate;
}
- if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) {
- throw DecodeError (N_("could not open audio decoder"));
- }
+ /* Compute the required offset (also removing any common start delay) */
+ return first_video - old_first_video - min (first_video, first_audio);
}
-void
-FFmpegDecoder::setup_subtitle ()
+FFmpegDecoder::~FFmpegDecoder ()
{
boost::mutex::scoped_lock lm (_mutex);
-
- if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) {
- return;
- }
- _subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec;
- _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
-
- if (_subtitle_codec == 0) {
- throw DecodeError (_("could not find subtitle decoder"));
- }
-
- if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
- throw DecodeError (N_("could not open subtitle decoder"));
+ if (_subtitle_codec_context) {
+ avcodec_close (_subtitle_codec_context);
}
-}
-
+}
void
FFmpegDecoder::pass ()
/* XXX: should we reset _packet.data and size after each *_decode_* call? */
if (_decode_video) {
- while (decode_video_packet ());
+ while (decode_video_packet ()) {}
}
if (_ffmpeg_content->audio_stream() && _decode_audio) {
}
/* Stop us being asked for any more data */
- _next_video = _next_audio = _ffmpeg_content->length ();
+ _video_position = _ffmpeg_content->video_length ();
+ _audio_position = _ffmpeg_content->audio_length ();
return;
}
decode_video_packet ();
} else if (_ffmpeg_content->audio_stream() && _packet.stream_index == _ffmpeg_content->audio_stream()->id && _decode_audio) {
decode_audio_packet ();
- } else if (_ffmpeg_content->subtitle_stream() && _packet.stream_index == _ffmpeg_content->subtitle_stream()->id && _decode_subtitles) {
-
- int got_subtitle;
- AVSubtitle sub;
- if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) && got_subtitle) {
- /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
- indicate that the previous subtitle should stop.
- */
- if (sub.num_rects > 0) {
- shared_ptr<TimedSubtitle> ts;
- try {
- subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub)));
- } catch (...) {
- /* some problem with the subtitle; we probably didn't understand it */
- }
- } else {
- subtitle (shared_ptr<TimedSubtitle> ());
- }
- avsubtitle_free (&sub);
- }
- } else {
- cout << "[ffmpeg] other packet.\n";
+ } else if (_ffmpeg_content->subtitle_stream() && _packet.stream_index == _ffmpeg_content->subtitle_stream()->id) {
+ decode_subtitle_packet ();
}
av_free_packet (&_packet);
return audio;
}
-float
-FFmpegDecoder::video_frame_rate () const
-{
- AVStream* s = _format_context->streams[_video_stream];
-
- if (s->avg_frame_rate.num && s->avg_frame_rate.den) {
- return av_q2d (s->avg_frame_rate);
- }
-
- return av_q2d (s->r_frame_rate);
-}
-
AVSampleFormat
FFmpegDecoder::audio_sample_format () const
{
- if (_audio_codec_context == 0) {
+ if (!_ffmpeg_content->audio_stream()) {
return (AVSampleFormat) 0;
}
- return _audio_codec_context->sample_fmt;
-}
-
-libdcp::Size
-FFmpegDecoder::video_size () const
-{
- return libdcp::Size (_video_codec_context->width, _video_codec_context->height);
-}
-
-string
-FFmpegDecoder::stream_name (AVStream* s) const
-{
- stringstream n;
-
- if (s->metadata) {
- AVDictionaryEntry const * lang = av_dict_get (s->metadata, N_("language"), 0, 0);
- if (lang) {
- n << lang->value;
- }
-
- AVDictionaryEntry const * title = av_dict_get (s->metadata, N_("title"), 0, 0);
- if (title) {
- if (!n.str().empty()) {
- n << N_(" ");
- }
- n << title->value;
- }
- }
-
- if (n.str().empty()) {
- n << N_("unknown");
- }
-
- return n.str ();
+ return audio_codec_context()->sample_fmt;
}
int
}
void
-FFmpegDecoder::seek (Time t)
+FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate)
{
- do_seek (t, false, false);
- VideoDecoder::seek (t);
-}
+ double const time_base = av_q2d (_format_context->streams[_video_stream]->time_base);
-void
-FFmpegDecoder::seek_back ()
-{
- if (next() < (2.5 * TIME_HZ / video_frame_rate())) {
- return;
+ /* If we are doing an accurate seek, our initial shot will be 5 frames (5 being
+ a number plucked from the air) earlier than we want to end up. The loop below
+ will hopefully then step through to where we want to be.
+ */
+ int initial = frame;
+ if (accurate) {
+ initial -= 5;
}
-
- do_seek (next() - 2.5 * TIME_HZ / video_frame_rate(), true, true);
- VideoDecoder::seek_back ();
-}
-void
-FFmpegDecoder::seek_forward ()
-{
- if (next() >= (_ffmpeg_content->length() - 0.5 * TIME_HZ / video_frame_rate())) {
- return;
- }
+ /* Initial seek time in the stream's timebase */
+ int64_t const initial_vt = initial / (_ffmpeg_content->video_frame_rate() * time_base);
+ /* Wanted final seek time in the stream's timebase */
+ int64_t const final_vt = frame / (_ffmpeg_content->video_frame_rate() * time_base);
- do_seek (next() - 0.5 * TIME_HZ / video_frame_rate(), true, true);
- VideoDecoder::seek_forward ();
-}
+ av_seek_frame (_format_context, _video_stream, initial_vt, AVSEEK_FLAG_BACKWARD);
-void
-FFmpegDecoder::do_seek (Time t, bool backwards, bool accurate)
-{
- int64_t const vt = t / (av_q2d (_format_context->streams[_video_stream]->time_base) * TIME_HZ);
- av_seek_frame (_format_context, _video_stream, vt, backwards ? AVSEEK_FLAG_BACKWARD : 0);
-
- avcodec_flush_buffers (_video_codec_context);
+ avcodec_flush_buffers (video_codec_context());
if (_subtitle_codec_context) {
avcodec_flush_buffers (_subtitle_codec_context);
}
+ _just_sought = true;
+
+ if (frame == 0) {
+ /* We're already there; from here on we can only seek non-zero amounts */
+ return;
+ }
+
if (accurate) {
while (1) {
int r = av_read_frame (_format_context, &_packet);
if (_packet.stream_index == _video_stream) {
int finished = 0;
- int const r = avcodec_decode_video2 (_video_codec_context, _frame, &finished, &_packet);
+ int const r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet);
if (r >= 0 && finished) {
int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
- if (bet > vt) {
+ if (bet >= final_vt) {
+ _video_position = rint (
+ (bet * time_base + _pts_offset) * _ffmpeg_content->video_frame_rate()
+ );
+ av_free_packet (&_packet);
break;
}
}
av_free_packet (&_packet);
}
}
-
- return;
-}
-
-/** @return Length (in video frames) according to our content's header */
-ContentVideoFrame
-FFmpegDecoder::video_length () const
-{
- return (double(_format_context->duration) / AV_TIME_BASE) * video_frame_rate();
}
void
while (copy_packet.size > 0) {
int frame_finished;
- int const decode_result = avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, ©_packet);
+ int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, ©_packet);
if (decode_result >= 0) {
if (frame_finished) {
-
- /* Where we are in the source, in seconds */
- double const source_pts_seconds = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
- * av_frame_get_best_effort_timestamp(_frame);
-
- int const data_size = av_samples_get_buffer_size (
- 0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1
- );
-
- assert (_audio_codec_context->channels == _ffmpeg_content->audio_channels());
- audio (deinterleave_audio (_frame->data, data_size), source_pts_seconds);
+
+ if (_audio_position == 0) {
+ /* Where we are in the source, in seconds */
+ double const pts = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
+ * av_frame_get_best_effort_timestamp(_frame) - _pts_offset;
+
+ if (pts > 0) {
+ /* Emit some silence */
+ shared_ptr<AudioBuffers> silence (
+ new AudioBuffers (
+ _ffmpeg_content->audio_channels(),
+ pts * _ffmpeg_content->content_audio_frame_rate()
+ )
+ );
+
+ silence->make_silent ();
+ audio (silence, _audio_position);
+ }
+ }
+
+ copy_packet.data += decode_result;
+ copy_packet.size -= decode_result;
}
-
- copy_packet.data += decode_result;
- copy_packet.size -= decode_result;
}
}
}
FFmpegDecoder::decode_video_packet ()
{
int frame_finished;
- if (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
+ if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
return false;
}
-
+
boost::mutex::scoped_lock lm (_filter_graphs_mutex);
shared_ptr<FilterGraph> graph;
int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
if (bet != AV_NOPTS_VALUE) {
- /* XXX: may need to insert extra frames / remove frames here ...
- (as per old Matcher)
- */
- Time const t = bet * av_q2d (_format_context->streams[_video_stream]->time_base) * TIME_HZ;
- video (image, false, t);
+
+ double const pts = bet * av_q2d (_format_context->streams[_video_stream]->time_base) - _pts_offset;
+
+ if (_just_sought) {
+ /* We just did a seek, so disable any attempts to correct for where we
+ are / should be.
+ */
+ _video_position = rint (pts * _ffmpeg_content->video_frame_rate ());
+ _just_sought = false;
+ }
+
+ double const next = _video_position / _ffmpeg_content->video_frame_rate();
+ double const one_frame = 1 / _ffmpeg_content->video_frame_rate ();
+ double delta = pts - next;
+
+ while (delta > one_frame) {
+ /* This PTS is more than one frame forward in time of where we think we should be; emit
+ a black frame.
+ */
+ boost::shared_ptr<Image> black (
+ new Image (
+ static_cast<AVPixelFormat> (_frame->format),
+ libdcp::Size (video_codec_context()->width, video_codec_context()->height),
+ true
+ )
+ );
+
+ black->make_black ();
+ video (image, false, _video_position);
+ delta -= one_frame;
+ }
+
+ if (delta > -one_frame) {
+ /* This PTS is within a frame of being right; emit this (otherwise it will be dropped) */
+ video (image, false, _video_position);
+ }
+
} else {
shared_ptr<const Film> film = _film.lock ();
assert (film);
return true;
}
-Time
-FFmpegDecoder::next () const
+
+void
+FFmpegDecoder::setup_subtitle ()
{
- if (_decode_video && _decode_audio && _audio_codec_context) {
- return min (_next_video, _next_audio);
+ boost::mutex::scoped_lock lm (_mutex);
+
+ if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) {
+ return;
}
- if (_decode_audio && _audio_codec_context) {
- return _next_audio;
- }
+ _subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec;
+ _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
- return _next_video;
+ if (_subtitle_codec == 0) {
+ throw DecodeError (_("could not find subtitle decoder"));
+ }
+
+ if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
+ throw DecodeError (N_("could not open subtitle decoder"));
+ }
}
bool
FFmpegDecoder::done () const
{
- return (!_decode_audio || !_audio_codec_context || audio_done()) && (!_decode_video || video_done());
+ bool const vd = !_decode_video || (_video_position >= _ffmpeg_content->video_length());
+ bool const ad = !_decode_audio || !_ffmpeg_content->audio_stream() || (_audio_position >= _ffmpeg_content->audio_length());
+ return vd && ad;
}
+void
+FFmpegDecoder::decode_subtitle_packet ()
+{
+ int got_subtitle;
+ AVSubtitle sub;
+ if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) {
+ return;
+ }
+
+ /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
+ indicate that the previous subtitle should stop.
+ */
+ if (sub.num_rects <= 0) {
+ subtitle (shared_ptr<Image> (), dcpomatic::Rect<double> (), 0, 0);
+ return;
+ } else if (sub.num_rects > 1) {
+ throw DecodeError (_("multi-part subtitles not yet supported"));
+ }
+
+ /* Subtitle PTS in seconds (within the source, not taking into account any of the
+ source that we may have chopped off for the DCP)
+ */
+ double const packet_time = static_cast<double> (sub.pts) / AV_TIME_BASE;
+
+ /* hence start time for this sub */
+ Time const from = (packet_time + (double (sub.start_display_time) / 1e3)) * TIME_HZ;
+ Time const to = (packet_time + (double (sub.end_display_time) / 1e3)) * TIME_HZ;
+
+ AVSubtitleRect const * rect = sub.rects[0];
+
+ if (rect->type != SUBTITLE_BITMAP) {
+ throw DecodeError (_("non-bitmap subtitles not yet supported"));
+ }
+
+ shared_ptr<Image> image (new Image (PIX_FMT_RGBA, libdcp::Size (rect->w, rect->h), true));
+
+ /* Start of the first line in the subtitle */
+ uint8_t* sub_p = rect->pict.data[0];
+ /* sub_p looks up into a RGB palette which is here */
+ uint32_t const * palette = (uint32_t *) rect->pict.data[1];
+ /* Start of the output data */
+ uint32_t* out_p = (uint32_t *) image->data()[0];
+
+ for (int y = 0; y < rect->h; ++y) {
+ uint8_t* sub_line_p = sub_p;
+ uint32_t* out_line_p = out_p;
+ for (int x = 0; x < rect->w; ++x) {
+ *out_line_p++ = palette[*sub_line_p++];
+ }
+ sub_p += rect->pict.linesize[0];
+ out_p += image->stride()[0] / sizeof (uint32_t);
+ }
+
+ libdcp::Size const vs = _ffmpeg_content->video_size ();
+
+ subtitle (
+ image,
+ dcpomatic::Rect<double> (
+ static_cast<double> (rect->x) / vs.width,
+ static_cast<double> (rect->y) / vs.height,
+ static_cast<double> (rect->w) / vs.width,
+ static_cast<double> (rect->h) / vs.height
+ ),
+ from,
+ to
+ );
+
+
+ avsubtitle_free (&sub);
+}