#include "log.h"
#include "ffmpeg_decoder.h"
#include "filter_graph.h"
-#include "subtitle.h"
#include "audio_buffers.h"
+#include "ffmpeg_content.h"
#include "i18n.h"
using std::stringstream;
using std::list;
using std::min;
+using std::pair;
using boost::shared_ptr;
using boost::optional;
using boost::dynamic_pointer_cast;
-using libdcp::Size;
+using dcp::Size;
FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegContent> c, bool video, bool audio)
: Decoder (f)
- , VideoDecoder (f)
- , AudioDecoder (f)
+ , VideoDecoder (f, c)
+ , AudioDecoder (f, c)
+ , SubtitleDecoder (f)
, FFmpeg (c)
, _subtitle_codec_context (0)
, _subtitle_codec (0)
, _decode_video (video)
, _decode_audio (audio)
, _pts_offset (0)
- , _just_sought (false)
{
setup_subtitle ();
- 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());
- }
-}
+ /* Audio and video frame PTS values may not start with 0. We want
+ to fiddle them so that:
-double
-FFmpegDecoder::compute_pts_offset (double first_video, double first_audio, float video_frame_rate)
-{
- double const old_first_video = first_video;
-
- /* 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;
+ 1. One of them starts at time 0.
+ 2. The first video PTS value ends up on a frame boundary.
+
+ Then we remove big initial gaps in PTS and we allow our
+ insertion of black frames to work.
+
+ We will do pts_to_use = pts_from_ffmpeg + pts_offset;
+ */
+
+ bool const have_video = video && c->first_video();
+ bool const have_audio = _decode_audio && c->audio_stream () && c->audio_stream()->first_audio;
+
+ /* First, make one of them start at 0 */
+
+ if (have_audio && have_video) {
+ _pts_offset = - min (c->first_video().get(), c->audio_stream()->first_audio.get());
+ } else if (have_video) {
+ _pts_offset = - c->first_video().get();
+ } else if (have_audio) {
+ _pts_offset = - c->audio_stream()->first_audio.get();
}
- /* Compute the required offset (also removing any common start delay) */
- return first_video - old_first_video - min (first_video, first_audio);
+ /* Now adjust both so that the video pts starts on a frame */
+ if (have_video && have_audio) {
+ double first_video = c->first_video().get() + _pts_offset;
+ double const old_first_video = first_video;
+
+ /* Round the first video up to a frame boundary */
+ if (fabs (rint (first_video * c->video_frame_rate()) - first_video * c->video_frame_rate()) > 1e-6) {
+ first_video = ceil (first_video * c->video_frame_rate()) / c->video_frame_rate ();
+ }
+
+ _pts_offset += first_video - old_first_video;
+ }
}
FFmpegDecoder::~FFmpegDecoder ()
if (_subtitle_codec_context) {
avcodec_close (_subtitle_codec_context);
}
-}
+}
void
+FFmpegDecoder::flush ()
+{
+ /* Get any remaining frames */
+
+ _packet.data = 0;
+ _packet.size = 0;
+
+ /* XXX: should we reset _packet.data and size after each *_decode_* call? */
+
+ if (_decode_video) {
+ while (decode_video_packet ()) {}
+ }
+
+ if (_ffmpeg_content->audio_stream() && _decode_audio) {
+ decode_audio_packet ();
+ AudioDecoder::flush ();
+ }
+}
+
+bool
FFmpegDecoder::pass ()
{
int r = av_read_frame (_format_context, &_packet);
film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r));
}
- /* Get any remaining frames */
-
- _packet.data = 0;
- _packet.size = 0;
-
- /* XXX: should we reset _packet.data and size after each *_decode_* call? */
-
- if (_decode_video) {
- while (decode_video_packet ()) {}
- }
-
- if (_ffmpeg_content->audio_stream() && _decode_audio) {
- decode_audio_packet ();
- }
-
- /* Stop us being asked for any more data */
- _video_position = _ffmpeg_content->video_length ();
- _audio_position = _ffmpeg_content->audio_length ();
- return;
+ flush ();
+ return true;
}
- avcodec_get_frame_defaults (_frame);
+ shared_ptr<const Film> film = _film.lock ();
+ assert (film);
- if (_packet.stream_index == _video_stream && _decode_video) {
+ int const si = _packet.stream_index;
+
+ if (si == _video_stream && _decode_video) {
decode_video_packet ();
- } else if (_ffmpeg_content->audio_stream() && _packet.stream_index == _ffmpeg_content->audio_stream()->id && _decode_audio) {
+ } else if (_ffmpeg_content->audio_stream() && _ffmpeg_content->audio_stream()->uses_index (_format_context, si) && _decode_audio) {
decode_audio_packet ();
- } else if (_ffmpeg_content->subtitle_stream() && _packet.stream_index == _ffmpeg_content->subtitle_stream()->id) {
-#if 0
-
- 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);
- }
-#endif
+ } else if (_ffmpeg_content->subtitle_stream() && _ffmpeg_content->subtitle_stream()->uses_index (_format_context, si) && film->with_subtitles ()) {
+ decode_subtitle_packet ();
}
av_free_packet (&_packet);
+ return false;
}
/** @param data pointer to array of pointers to buffers.
return av_get_bytes_per_sample (audio_sample_format ());
}
+int
+FFmpegDecoder::minimal_run (boost::function<bool (optional<ContentTime>, optional<ContentTime>, int)> finished)
+{
+ int frames_read = 0;
+ optional<ContentTime> last_video;
+ optional<ContentTime> last_audio;
+
+ while (!finished (last_video, last_audio, frames_read)) {
+ int r = av_read_frame (_format_context, &_packet);
+ if (r < 0) {
+ /* We should flush our decoders here, possibly yielding a few more frames,
+ but the consequence of having to do that is too hideous to contemplate.
+ Instead we give up and say that you can't seek too close to the end
+ of a file.
+ */
+ return frames_read;
+ }
+
+ ++frames_read;
+
+ double const time_base = av_q2d (_format_context->streams[_packet.stream_index]->time_base);
+
+ if (_packet.stream_index == _video_stream) {
+
+ avcodec_get_frame_defaults (_frame);
+
+ int finished = 0;
+ r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet);
+ if (r >= 0 && finished) {
+ last_video = rint (
+ (av_frame_get_best_effort_timestamp (_frame) * time_base + _pts_offset) * TIME_HZ
+ );
+ }
+
+ } else if (_ffmpeg_content->audio_stream() && _ffmpeg_content->audio_stream()->uses_index (_format_context, _packet.stream_index)) {
+ AVPacket copy_packet = _packet;
+ while (copy_packet.size > 0) {
+
+ int finished;
+ r = avcodec_decode_audio4 (audio_codec_context(), _frame, &finished, &_packet);
+ if (r >= 0 && finished) {
+ last_audio = rint (
+ (av_frame_get_best_effort_timestamp (_frame) * time_base + _pts_offset) * TIME_HZ
+ );
+ }
+
+ copy_packet.data += r;
+ copy_packet.size -= r;
+ }
+ }
+
+ av_free_packet (&_packet);
+ }
+
+ return frames_read;
+}
+
+bool
+FFmpegDecoder::seek_overrun_finished (ContentTime seek, optional<ContentTime> last_video, optional<ContentTime> last_audio) const
+{
+ return (last_video && last_video.get() >= seek) || (last_audio && last_audio.get() >= seek);
+}
+
+bool
+FFmpegDecoder::seek_final_finished (int n, int done) const
+{
+ return n == done;
+}
+
void
-FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate)
+FFmpegDecoder::seek_and_flush (ContentTime t)
{
- double const time_base = av_q2d (_format_context->streams[_video_stream]->time_base);
- int64_t const vt = frame / (_ffmpeg_content->video_frame_rate() * time_base);
- av_seek_frame (_format_context, _video_stream, vt, AVSEEK_FLAG_BACKWARD);
+ int64_t s = ((double (t) / TIME_HZ) - _pts_offset) /
+ av_q2d (_format_context->streams[_video_stream]->time_base);
+
+ if (_ffmpeg_content->audio_stream ()) {
+ s = min (
+ s, int64_t (
+ ((double (t) / TIME_HZ) - _pts_offset) /
+ av_q2d (_ffmpeg_content->audio_stream()->stream(_format_context)->time_base)
+ )
+ );
+ }
+
+ /* Ridiculous empirical hack */
+ s--;
+
+ av_seek_frame (_format_context, _video_stream, s, AVSEEK_FLAG_BACKWARD);
avcodec_flush_buffers (video_codec_context());
+ if (audio_codec_context ()) {
+ avcodec_flush_buffers (audio_codec_context ());
+ }
if (_subtitle_codec_context) {
avcodec_flush_buffers (_subtitle_codec_context);
}
+}
+
+void
+FFmpegDecoder::seek (ContentTime time, bool accurate)
+{
+ Decoder::seek (time, accurate);
+ if (_decode_audio) {
+ AudioDecoder::seek (time, accurate);
+ }
+
+ /* If we are doing an accurate seek, our initial shot will be 200ms (200 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.
+ */
+
+ ContentTime pre_roll = accurate ? (0.2 * TIME_HZ) : 0;
+ ContentTime initial_seek = time - pre_roll;
+ if (initial_seek < 0) {
+ initial_seek = 0;
+ }
+
+ /* Initial seek time in the video stream's timebase */
- _just_sought = true;
+ seek_and_flush (initial_seek);
- if (frame == 0) {
- /* We're already there; from here on we can only seek non-zero amounts */
+ if (!accurate) {
+ /* That'll do */
return;
}
- if (accurate) {
- while (1) {
- int r = av_read_frame (_format_context, &_packet);
- if (r < 0) {
- return;
- }
-
- avcodec_get_frame_defaults (_frame);
-
- if (_packet.stream_index == _video_stream) {
- int finished = 0;
- 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) {
- _video_position = rint (
- (bet * time_base + _pts_offset) * _ffmpeg_content->video_frame_rate()
- );
- av_free_packet (&_packet);
- break;
- }
- }
- }
-
- av_free_packet (&_packet);
- }
+ int const N = minimal_run (boost::bind (&FFmpegDecoder::seek_overrun_finished, this, time, _1, _2));
+
+ seek_and_flush (initial_seek);
+ if (N > 0) {
+ minimal_run (boost::bind (&FFmpegDecoder::seek_final_finished, this, N - 1, _3));
}
}
*/
AVPacket copy_packet = _packet;
-
+
while (copy_packet.size > 0) {
int frame_finished;
int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, ©_packet);
- if (decode_result >= 0) {
- if (frame_finished) {
-
- 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;
- }
+
+ if (decode_result < 0) {
+ shared_ptr<const Film> film = _film.lock ();
+ assert (film);
+ film->log()->log (String::compose ("avcodec_decode_audio4 failed (%1)", decode_result));
+ return;
+ }
+
+ if (frame_finished) {
+ ContentTime const ct = (
+ av_frame_get_best_effort_timestamp (_frame) *
+ av_q2d (_ffmpeg_content->audio_stream()->stream (_format_context)->time_base)
+ + _pts_offset
+ ) * TIME_HZ;
+
+ int const data_size = av_samples_get_buffer_size (
+ 0, audio_codec_context()->channels, _frame->nb_samples, audio_sample_format (), 1
+ );
+
+ audio (deinterleave_audio (_frame->data, data_size), ct);
}
+
+ copy_packet.data += decode_result;
+ copy_packet.size -= decode_result;
}
}
shared_ptr<FilterGraph> graph;
list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
- while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
+ while (i != _filter_graphs.end() && !(*i)->can_process (dcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
++i;
}
shared_ptr<const Film> film = _film.lock ();
assert (film);
- graph.reset (new FilterGraph (_ffmpeg_content, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
+ graph.reset (new FilterGraph (_ffmpeg_content, dcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
_filter_graphs.push_back (graph);
film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format));
graph = *i;
}
- list<shared_ptr<Image> > images = graph->process (_frame);
+ list<pair<shared_ptr<Image>, int64_t> > images = graph->process (_frame);
string post_process = Filter::ffmpeg_strings (_ffmpeg_content->filters()).second;
- for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) {
+ for (list<pair<shared_ptr<Image>, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) {
- shared_ptr<Image> image = *i;
+ shared_ptr<Image> image = i->first;
if (!post_process.empty ()) {
image = image->post_process (post_process, true);
}
- int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
- if (bet != AV_NOPTS_VALUE) {
-
- 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 SimpleImage (
- 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);
- }
-
+ if (i->second != AV_NOPTS_VALUE) {
+ double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _pts_offset;
+ VideoFrame const f = rint (pts * _ffmpeg_content->video_frame_rate ());
+ video (image, false, f);
} else {
shared_ptr<const Film> film = _film.lock ();
assert (film);
{
boost::mutex::scoped_lock lm (_mutex);
- if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) {
+ if (!_ffmpeg_content->subtitle_stream()) {
return;
}
- _subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec;
+ _subtitle_codec_context = _ffmpeg_content->subtitle_stream()->stream(_format_context)->codec;
+ if (_subtitle_codec_context == 0) {
+ throw DecodeError (N_("could not find subtitle stream"));
+ }
+
_subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
if (_subtitle_codec == 0) {
- throw DecodeError (_("could not find subtitle decoder"));
+ throw DecodeError (N_("could not find subtitle decoder"));
}
if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
}
}
-bool
-FFmpegDecoder::done () const
+void
+FFmpegDecoder::decode_subtitle_packet ()
{
- 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;
-}
+ 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) {
+ image_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) + _pts_offset;
+ /* hence start time for this sub */
+ ContentTime const from = (packet_time + (double (sub.start_display_time) / 1e3)) * TIME_HZ;
+ ContentTime 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"));
+ }
+
+ /* Note RGBA is expressed little-endian, so the first byte in the word is R, second
+ G, third B, fourth A.
+ */
+ shared_ptr<Image> image (new Image (PIX_FMT_RGBA, dcp::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 BGRA palette which is here
+ (i.e. first byte B, second G, third R, fourth A)
+ */
+ 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) {
+ uint32_t const p = palette[*sub_line_p++];
+ *out_line_p++ = ((p & 0xff) << 16) | (p & 0xff00) | ((p & 0xff0000) >> 16) | (p & 0xff000000);
+ }
+ sub_p += rect->pict.linesize[0];
+ out_p += image->stride()[0] / sizeof (uint32_t);
+ }
+
+ dcp::Size const vs = _ffmpeg_content->video_size ();
+
+ image_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);
+}