#include <libavcodec/avcodec.h>
#include <libavformat/avformat.h>
}
-#include "film.h"
#include "filter.h"
#include "exceptions.h"
#include "image.h"
#include "util.h"
#include "log.h"
#include "ffmpeg_decoder.h"
+#include "ffmpeg_audio_stream.h"
+#include "ffmpeg_subtitle_stream.h"
#include "filter_graph.h"
#include "audio_buffers.h"
#include "ffmpeg_content.h"
-#include "image_proxy.h"
+#include "raw_image_proxy.h"
+#include "film.h"
+#include "timer.h"
#include "i18n.h"
-#define LOG_GENERAL(...) film->log()->log (String::compose (__VA_ARGS__), Log::TYPE_GENERAL);
-#define LOG_ERROR(...) film->log()->log (String::compose (__VA_ARGS__), Log::TYPE_ERROR);
-#define LOG_WARNING(...) film->log()->log (__VA_ARGS__, Log::TYPE_WARNING);
+#define LOG_GENERAL(...) _video_content->film()->log()->log (String::compose (__VA_ARGS__), Log::TYPE_GENERAL);
+#define LOG_ERROR(...) _video_content->film()->log()->log (String::compose (__VA_ARGS__), Log::TYPE_ERROR);
+#define LOG_WARNING(...) _video_content->film()->log()->log (__VA_ARGS__, Log::TYPE_WARNING);
using std::cout;
using std::string;
using std::list;
using std::min;
using std::pair;
+using std::make_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, c)
- , AudioDecoder (f, c)
- , SubtitleDecoder (f)
+FFmpegDecoder::FFmpegDecoder (shared_ptr<const FFmpegContent> c, shared_ptr<Log> log)
+ : VideoDecoder (c)
+ , AudioDecoder (c)
+ , SubtitleDecoder (c)
, FFmpeg (c)
- , _subtitle_codec_context (0)
- , _subtitle_codec (0)
- , _decode_video (video)
- , _decode_audio (audio)
- , _pts_offset (0)
- , _just_sought (false)
+ , _log (log)
{
- setup_subtitle ();
-
/* Audio and video frame PTS values may not start with 0. We want
to fiddle them so that:
Then we remove big initial gaps in PTS and we allow our
insertion of black frames to work.
- We will do:
- audio_pts_to_use = audio_pts_from_ffmpeg + pts_offset;
- video_pts_to_use = video_pts_from_ffmpeg + pts_offset;
+ We will do pts_to_use = pts_from_ffmpeg + pts_offset;
*/
- bool const have_video = video && c->first_video();
- bool const have_audio = audio && c->audio_stream() && c->audio_stream()->first_audio;
+ bool const have_video = c->first_video();
+ bool const have_audio = c->audio_stream () && c->audio_stream()->first_audio;
/* First, make one of them start at 0 */
/* 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 ()
-{
- boost::mutex::scoped_lock lm (_mutex);
-
- if (_subtitle_codec_context) {
- avcodec_close (_subtitle_codec_context);
+ ContentTime first_video = c->first_video().get() + _pts_offset;
+ ContentTime const old_first_video = first_video;
+ _pts_offset += first_video.round_up (c->video_frame_rate ()) - old_first_video;
}
}
/* 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) {
+ if (_ffmpeg_content->audio_stream()) {
decode_audio_packet ();
+ AudioDecoder::flush ();
}
-
- /* Stop us being asked for any more data */
- _video_position = _ffmpeg_content->video_length_after_3d_combine ();
- _audio_position = _ffmpeg_content->audio_length ();
}
-void
+bool
FFmpegDecoder::pass ()
{
int r = av_read_frame (_format_context, &_packet);
/* Maybe we should fail here, but for now we'll just finish off instead */
char buf[256];
av_strerror (r, buf, sizeof(buf));
- shared_ptr<const Film> film = _film.lock ();
- assert (film);
LOG_ERROR (N_("error on av_read_frame (%1) (%2)"), buf, r);
}
flush ();
- return;
+ return true;
}
- shared_ptr<const Film> film = _film.lock ();
- assert (film);
-
int const si = _packet.stream_index;
-
- if (si == _video_stream && _decode_video) {
+
+ if (si == _video_stream) {
decode_video_packet ();
- } else if (_ffmpeg_content->audio_stream() && _ffmpeg_content->audio_stream()->uses_index (_format_context, si) && _decode_audio) {
+ } else if (_ffmpeg_content->audio_stream() && _ffmpeg_content->audio_stream()->uses_index (_format_context, si)) {
decode_audio_packet ();
- } else if (_ffmpeg_content->subtitle_stream() && _ffmpeg_content->subtitle_stream()->uses_index (_format_context, si) && film->with_subtitles ()) {
+ } else if (_ffmpeg_content->subtitle_stream() && _ffmpeg_content->subtitle_stream()->uses_index (_format_context, si)) {
decode_subtitle_packet ();
}
av_free_packet (&_packet);
+ return false;
}
/** @param data pointer to array of pointers to buffers.
}
void
-FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate)
+FFmpegDecoder::seek (ContentTime time, bool accurate)
{
- double const time_base = av_q2d (_format_context->streams[_video_stream]->time_base);
-
- /* 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.
+ VideoDecoder::seek (time, accurate);
+ AudioDecoder::seek (time, accurate);
+
+ /* If we are doing an `accurate' seek, we need to use pre-roll, as
+ we don't really know what the seek will give us.
*/
- int initial = frame;
- if (accurate) {
- initial -= 5;
- }
+ ContentTime pre_roll = accurate ? ContentTime::from_seconds (2) : ContentTime (0);
+ time -= pre_roll;
- if (initial < 0) {
- initial = 0;
- }
-
- /* Initial seek time in the stream's timebase */
- int64_t const initial_vt = ((initial / _ffmpeg_content->original_video_frame_rate()) - _pts_offset) / time_base;
-
- av_seek_frame (_format_context, _video_stream, initial_vt, AVSEEK_FLAG_BACKWARD);
-
- avcodec_flush_buffers (video_codec_context());
- if (_subtitle_codec_context) {
- avcodec_flush_buffers (_subtitle_codec_context);
- }
-
- /* This !accurate is piling hack upon hack; setting _just_sought to true
- even with accurate == true defeats our attempt to align the start
- of the video and audio. Here we disable that defeat when accurate == true
- i.e. when we are making a DCP rather than just previewing one.
- Ewww. This should be gone in 2.0.
+ /* XXX: it seems debatable whether PTS should be used here...
+ http://www.mjbshaw.com/2012/04/seeking-in-ffmpeg-know-your-timestamp.html
*/
- if (!accurate) {
- _just_sought = true;
- }
-
- _video_position = frame;
- if (frame == 0 || !accurate) {
- /* We're already there, or we're as close as we need to be */
- return;
- }
+ ContentTime const u = time - _pts_offset;
+ int64_t s = u.seconds() / av_q2d (_format_context->streams[_video_stream]->time_base);
- while (true) {
- int r = av_read_frame (_format_context, &_packet);
- if (r < 0) {
- return;
- }
+ if (_ffmpeg_content->audio_stream ()) {
+ s = min (
+ s, int64_t (u.seconds() / av_q2d (_ffmpeg_content->audio_stream()->stream(_format_context)->time_base))
+ );
+ }
- if (_packet.stream_index != _video_stream) {
- av_free_packet (&_packet);
- continue;
- }
-
- int finished = 0;
- r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet);
- if (r >= 0 && finished) {
- _video_position = rint (
- (av_frame_get_best_effort_timestamp (_frame) * time_base + _pts_offset) * _ffmpeg_content->original_video_frame_rate()
- );
+ av_seek_frame (_format_context, _video_stream, s, 0);
- if (_video_position >= (frame - 1)) {
- av_free_packet (&_packet);
- break;
- }
- }
-
- av_free_packet (&_packet);
+ 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 ());
}
-
- /* _video_position should be the next thing to be emitted, which will the one after the thing
- we just saw.
- */
- _video_position++;
}
void
int frame_finished;
int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, ©_packet);
+
if (decode_result < 0) {
- shared_ptr<const Film> film = _film.lock ();
- assert (film);
LOG_ERROR ("avcodec_decode_audio4 failed (%1)", decode_result);
return;
}
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 */
- int64_t frames = pts * _ffmpeg_content->content_audio_frame_rate ();
- while (frames > 0) {
- int64_t const this_time = min (frames, (int64_t) _ffmpeg_content->content_audio_frame_rate() / 2);
-
- shared_ptr<AudioBuffers> silence (
- new AudioBuffers (_ffmpeg_content->audio_channels(), this_time)
- );
-
- silence->make_silent ();
- audio (silence, _audio_position);
- frames -= this_time;
- }
- }
- }
+ ContentTime const ct = ContentTime::from_seconds (
+ av_frame_get_best_effort_timestamp (_frame) *
+ av_q2d (_ffmpeg_content->audio_stream()->stream (_format_context)->time_base))
+ + _pts_offset;
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), _audio_position);
+
+ audio (deinterleave_audio (_frame->data, data_size), ct);
}
copy_packet.data += 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;
}
if (i == _filter_graphs.end ()) {
- 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);
-
LOG_GENERAL (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format);
} else {
graph = *i;
list<pair<shared_ptr<Image>, int64_t> > images = graph->process (_frame);
- shared_ptr<const Film> film = _film.lock ();
- assert (film);
-
for (list<pair<shared_ptr<Image>, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) {
shared_ptr<Image> image = i->first;
if (i->second != AV_NOPTS_VALUE) {
-
- double const pts = i->second * 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->original_video_frame_rate ());
- _just_sought = false;
- }
-
- double const next = _video_position / _ffmpeg_content->original_video_frame_rate();
- double const one_frame = 1 / _ffmpeg_content->original_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.
- */
-
- /* XXX: I think this should be a copy of the last frame... */
- boost::shared_ptr<Image> black (
- new Image (
- static_cast<AVPixelFormat> (_frame->format),
- libdcp::Size (video_codec_context()->width, video_codec_context()->height),
- true
- )
- );
-
- shared_ptr<const Film> film = _film.lock ();
- assert (film);
-
- black->make_black ();
- video (shared_ptr<ImageProxy> (new RawImageProxy (image, film->log())), 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 (shared_ptr<ImageProxy> (new RawImageProxy (image, film->log())), false, _video_position);
- }
-
+ double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _pts_offset.seconds ();
+ video (
+ shared_ptr<ImageProxy> (new RawImageProxy (image, _video_content->film()->log())),
+ rint (pts * _ffmpeg_content->video_frame_rate ())
+ );
} else {
LOG_WARNING ("Dropping frame without PTS");
}
return true;
}
-
-
-void
-FFmpegDecoder::setup_subtitle ()
-{
- boost::mutex::scoped_lock lm (_mutex);
-
- if (!_ffmpeg_content->subtitle_stream()) {
- return;
- }
-
- _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 (N_("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
-{
- 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) {
+ if (avcodec_decode_subtitle2 (subtitle_codec_context(), &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) {
return;
}
indicate that the previous subtitle should stop.
*/
if (sub.num_rects <= 0) {
- subtitle (shared_ptr<Image> (), dcpomatic::Rect<double> (), 0, 0);
+ image_subtitle (ContentTimePeriod (), shared_ptr<Image> (), dcpomatic::Rect<double> ());
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
+ /* Subtitle PTS (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 */
- 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;
+ ContentTimePeriod period = subtitle_period (sub) + _pts_offset;
AVSubtitleRect const * rect = sub.rects[0];
if (rect->type != SUBTITLE_BITMAP) {
- throw DecodeError (_("non-bitmap subtitles not yet supported"));
+ /* XXX */
+ // throw DecodeError (_("non-bitmap subtitles not yet supported"));
+ return;
}
/* 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, libdcp::Size (rect->w, rect->h), true));
+ 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];
out_p += image->stride()[0] / sizeof (uint32_t);
}
- libdcp::Size const vs = _ffmpeg_content->video_size ();
+ dcp::Size const vs = _ffmpeg_content->video_size ();
- subtitle (
+ image_subtitle (
+ period,
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);
}
+
+list<ContentTimePeriod>
+FFmpegDecoder::subtitles_during (ContentTimePeriod p, bool starting) const
+{
+ return _ffmpeg_content->subtitles_during (p, starting);
+}
projects / dcpomatic.git / blobdiff