Some fixes to playback of drop-frame content.
[dcpomatic.git] / src / lib / decoder.cc
index eee288341d728fdc04d72d41157de1563381171a..7f4973737faff0146f971f91c846cc4c8a5dabef 100644 (file)
  *  @brief Parent class for decoders of content.
  */
 
-#include <iostream>
-#include <stdint.h>
-#include <boost/lexical_cast.hpp>
-extern "C" {
-#include <libavfilter/avfiltergraph.h>
-#include <libavfilter/buffersrc.h>
-#if (LIBAVFILTER_VERSION_MAJOR == 2 && LIBAVFILTER_VERSION_MINOR >= 53 && LIBAVFILTER_VERSION_MINOR <= 77) || LIBAVFILTER_VERSION_MAJOR == 3
-#include <libavfilter/avcodec.h>
-#include <libavfilter/buffersink.h>
-#elif LIBAVFILTER_VERSION_MAJOR == 2 && LIBAVFILTER_VERSION_MINOR == 15
-#include <libavfilter/vsrc_buffer.h>
-#endif
-#include <libavformat/avio.h>
-}
 #include "film.h"
-#include "format.h"
-#include "job.h"
-#include "options.h"
-#include "exceptions.h"
-#include "image.h"
-#include "util.h"
-#include "log.h"
 #include "decoder.h"
-#include "filter.h"
-#include "delay_line.h"
-#include "ffmpeg_compatibility.h"
-#include "subtitle.h"
+#include "decoded.h"
+
+#include "i18n.h"
 
-using std::string;
-using std::stringstream;
-using std::min;
 using boost::shared_ptr;
 
 /** @param f Film.
- *  @param o Options.
- *  @param j Job that we are running within, or 0
- *  @param minimal true to do the bare minimum of work; just run through the content.  Useful for acquiring
- *  accurate frame counts as quickly as possible.  This generates no video or audio output.
- *  @param ignore_length Ignore the content's claimed length when computing progress.
+ *  @param o Decode options.
  */
-Decoder::Decoder (boost::shared_ptr<Film> f, boost::shared_ptr<const Options> o, Job* j, bool minimal, bool ignore_length)
+Decoder::Decoder (shared_ptr<const Film> f)
        : _film (f)
-       , _opt (o)
-       , _job (j)
-       , _minimal (minimal)
-       , _ignore_length (ignore_length)
-       , _video_frame (0)
-       , _buffer_src_context (0)
-       , _buffer_sink_context (0)
-       , _have_setup_video_filters (false)
-       , _delay_line (0)
-       , _delay_in_bytes (0)
-       , _audio_frames_processed (0)
-{
-       if (_opt->decode_video_frequency != 0 && !_film->length()) {
-               throw DecodeError ("cannot do a partial decode if length is unknown");
-       }
-}
-
-Decoder::~Decoder ()
-{
-       delete _delay_line;
-}
-
-/** Start off a decode processing run */
-void
-Decoder::process_begin ()
-{
-       _delay_in_bytes = _film->audio_delay() * audio_sample_rate() * audio_channels() * bytes_per_audio_sample() / 1000;
-       delete _delay_line;
-       _delay_line = new DelayLine (_delay_in_bytes);
-
-       _audio_frames_processed = 0;
-}
-
-/** Finish off a decode processing run */
-void
-Decoder::process_end ()
-{
-       if (_delay_in_bytes < 0) {
-               uint8_t remainder[-_delay_in_bytes];
-               _delay_line->get_remaining (remainder);
-               _audio_frames_processed += _delay_in_bytes / (audio_channels() * bytes_per_audio_sample());
-               emit_audio (remainder, -_delay_in_bytes);
-       }
-
-       /* If we cut the decode off, the audio may be short; push some silence
-          in to get it to the right length.
-       */
-
-       int64_t const video_length_in_audio_frames = ((int64_t) last_video_frame() * audio_sample_rate() / frames_per_second());
-       int64_t const audio_short_by_frames = video_length_in_audio_frames - _audio_frames_processed;
-
-       _log->log (
-               String::compose ("DCP length is %1 (%2 audio frames); %3 frames of audio processed.",
-                                last_video_frame(),
-                                video_length_in_audio_frames,
-                                _audio_frames_processed)
-               );
-       
-       if (audio_short_by_frames >= 0 && _opt->decode_audio) {
-
-               _log->log (String::compose ("DCP length is %1; %2 frames of audio processed.", last_video_frame(), _audio_frames_processed));
-               _log->log (String::compose ("Adding %1 frames of silence to the end.", audio_short_by_frames));
-
-               /* XXX: this is slightly questionable; does memset () give silence with all
-                  sample formats?
-               */
-
-               int64_t bytes = audio_short_by_frames * _film->audio_channels() * bytes_per_audio_sample();
-               
-               int64_t const silence_size = 16 * 1024 * _film->audio_channels() * bytes_per_audio_sample();
-               uint8_t silence[silence_size];
-               memset (silence, 0, silence_size);
-               
-               while (bytes) {
-                       int64_t const t = min (bytes, silence_size);
-                       emit_audio (silence, t);
-                       bytes -= t;
-               }
-       }
-}
-
-/** Start decoding */
-void
-Decoder::go ()
-{
-       process_begin ();
-
-       if (_job && !_film->dcp_length()) {
-               _job->set_progress_unknown ();
-       }
-
-       while (pass () == false) {
-               if (_job && _film->dcp_length()) {
-                       _job->set_progress (float (_video_frame) / _film->dcp_length().get());
-               }
-       }
-
-       process_end ();
-}
-
-/** Run one pass.  This may or may not generate any actual video / audio data;
- *  some decoders may require several passes to generate a single frame.
- *  @return true if we have finished processing all data; otherwise false.
- */
-bool
-Decoder::pass ()
 {
-       if (!_have_setup_video_filters) {
-               setup_video_filters ();
-               _have_setup_video_filters = true;
-       }
-       
-       if (!_ignore_length && _video_frame >= _film->dcp_length()) {
-               return true;
-       }
 
-       return do_pass ();
 }
 
-/** Called by subclasses to tell the world that some audio data is ready
- *  @param data Audio data, in Film::audio_sample_format.
- *  @param size Number of bytes of data.
- */
-void
-Decoder::process_audio (uint8_t* data, int size)
+class DecodedSorter
 {
-       /* Push into the delay line */
-       size = _delay_line->feed (data, size);
-
-       emit_audio (data, size);
-}
-
-void
-Decoder::emit_audio (uint8_t* data, int size)
-{
-       /* Deinterleave and convert to float */
-
-       assert ((size % (bytes_per_audio_sample() * _film->audio_channels())) == 0);
-
-       int const total_samples = size / bytes_per_audio_sample();
-       int const frames = total_samples / _film->audio_channels();
-       shared_ptr<AudioBuffers> audio (new AudioBuffers (_film->audio_channels(), frames));
-
-       switch (audio_sample_format()) {
-       case AV_SAMPLE_FMT_S16:
-       {
-               int16_t* p = (int16_t *) data;
-               int sample = 0;
-               int channel = 0;
-               for (int i = 0; i < total_samples; ++i) {
-                       audio->data(channel)[sample] = float(*p++) / (1 << 15);
-
-                       ++channel;
-                       if (channel == _film->audio_channels()) {
-                               channel = 0;
-                               ++sample;
-                       }
-               }
-       }
-       break;
-
-       case AV_SAMPLE_FMT_S32:
+public:
+       bool operator() (shared_ptr<Decoded> a, shared_ptr<Decoded> b)
        {
-               int32_t* p = (int32_t *) data;
-               int sample = 0;
-               int channel = 0;
-               for (int i = 0; i < total_samples; ++i) {
-                       audio->data(channel)[sample] = float(*p++) / (1 << 31);
-
-                       ++channel;
-                       if (channel == _film->audio_channels()) {
-                               channel = 0;
-                               ++sample;
-                       }
-               }
-       }
-
-       case AV_SAMPLE_FMT_FLTP:
-       {
-               float* p = reinterpret_cast<float*> (data);
-               for (int i = 0; i < _film->audio_channels(); ++i) {
-                       memcpy (audio->data(i), p, frames * sizeof(float));
-                       p += frames;
-               }
-       }
-       break;
-
-       default:
-               assert (false);
+               return a->dcp_time < b->dcp_time;
        }
+};
 
-       /* Maybe apply gain */
-       if (_film->audio_gain() != 0) {
-               float const linear_gain = pow (10, _film->audio_gain() / 20);
-               for (int i = 0; i < _film->audio_channels(); ++i) {
-                       for (int j = 0; j < frames; ++j) {
-                               audio->data(i)[j] *= linear_gain;
-                       }
-               }
-       }
-
-       /* Update the number of audio frames we've pushed to the encoder */
-       _audio_frames_processed += audio->frames ();
-
-       Audio (audio);
-}
-
-/** Called by subclasses to tell the world that some video data is ready.
- *  We do some post-processing / filtering then emit it for listeners.
- *  @param frame to decode; caller manages memory.
- */
-void
-Decoder::process_video (AVFrame* frame)
+shared_ptr<Decoded>
+Decoder::peek ()
 {
-       if (_minimal) {
-               ++_video_frame;
-               return;
-       }
+       while (_pending.empty() && !pass ()) {}
 
-       /* Use Film::length here as our one may be wrong */
-
-       int gap = 0;
-       if (_opt->decode_video_frequency != 0) {
-               gap = _film->length().get() / _opt->decode_video_frequency;
-       }
-
-       if (_opt->decode_video_frequency != 0 && gap != 0 && (_video_frame % gap) != 0) {
-               ++_video_frame;
-               return;
+       if (_pending.empty ()) {
+               return shared_ptr<Decoded> ();
        }
 
-#if LIBAVFILTER_VERSION_MAJOR == 2 && LIBAVFILTER_VERSION_MINOR >= 53 && LIBAVFILTER_VERSION_MINOR <= 61
-
-       if (av_vsrc_buffer_add_frame (_buffer_src_context, frame, 0) < 0) {
-               throw DecodeError ("could not push buffer into filter chain.");
-       }
-
-#elif LIBAVFILTER_VERSION_MAJOR == 2 && LIBAVFILTER_VERSION_MINOR == 15
-
-       AVRational par;
-       par.num = sample_aspect_ratio_numerator ();
-       par.den = sample_aspect_ratio_denominator ();
-
-       if (av_vsrc_buffer_add_frame (_buffer_src_context, frame, 0, par) < 0) {
-               throw DecodeError ("could not push buffer into filter chain.");
-       }
-
-#else
-
-       if (av_buffersrc_write_frame (_buffer_src_context, frame) < 0) {
-               throw DecodeError ("could not push buffer into filter chain.");
-       }
-
-#endif 
-       
-#if LIBAVFILTER_VERSION_MAJOR == 2 && LIBAVFILTER_VERSION_MINOR >= 15 && LIBAVFILTER_VERSION_MINOR <= 61       
-       while (avfilter_poll_frame (_buffer_sink_context->inputs[0])) {
-#else
-       while (av_buffersink_read (_buffer_sink_context, 0)) {
-#endif         
-
-#if LIBAVFILTER_VERSION_MAJOR == 2 && LIBAVFILTER_VERSION_MINOR >= 15
-               
-               int r = avfilter_request_frame (_buffer_sink_context->inputs[0]);
-               if (r < 0) {
-                       throw DecodeError ("could not request filtered frame");
-               }
-               
-               AVFilterBufferRef* filter_buffer = _buffer_sink_context->inputs[0]->cur_buf;
-               
-#else
-
-               AVFilterBufferRef* filter_buffer;
-               if (av_buffersink_get_buffer_ref (_buffer_sink_context, &filter_buffer, 0) < 0) {
-                       filter_buffer = 0;
-               }
-
-#endif         
-               
-               if (filter_buffer) {
-                       /* This takes ownership of filter_buffer */
-                       shared_ptr<Image> image (new FilterBufferImage ((PixelFormat) frame->format, filter_buffer));
-
-                       if (_opt->black_after > 0 && _video_frame > _opt->black_after) {
-                               image->make_black ();
-                       }
-
-                       shared_ptr<Subtitle> sub;
-                       if (_timed_subtitle && _timed_subtitle->displayed_at (double (last_video_frame()) / _film->frames_per_second())) {
-                               sub = _timed_subtitle->subtitle ();
-                       }
-
-                       TIMING ("Decoder emits %1", _video_frame);
-                       Video (image, _video_frame, sub);
-                       ++_video_frame;
-               }
-       }
+       return _pending.front ();
 }
 
-
-/** Set up a video filtering chain to include cropping and any filters that are specified
- *  by the Film.
- */
 void
-Decoder::setup_video_filters ()
+Decoder::consume ()
 {
-       stringstream fs;
-       Size size_after_crop;
-       
-       if (_opt->apply_crop) {
-               size_after_crop = _film->cropped_size (native_size ());
-               fs << crop_string (Position (_film->crop().left, _film->crop().top), size_after_crop);
-       } else {
-               size_after_crop = native_size ();
-               fs << crop_string (Position (0, 0), size_after_crop);
-       }
-
-       string filters = Filter::ffmpeg_strings (_film->filters()).first;
-       if (!filters.empty ()) {
-               filters += ",";
-       }
-
-       filters += fs.str ();
-
-       avfilter_register_all ();
-       
-       AVFilterGraph* graph = avfilter_graph_alloc();
-       if (graph == 0) {
-               throw DecodeError ("Could not create filter graph.");
-       }
-
-       AVFilter* buffer_src = avfilter_get_by_name("buffer");
-       if (buffer_src == 0) {
-               throw DecodeError ("Could not find buffer src filter");
-       }
-
-       AVFilter* buffer_sink = get_sink ();
-
-       stringstream a;
-       a << native_size().width << ":"
-         << native_size().height << ":"
-         << pixel_format() << ":"
-         << time_base_numerator() << ":"
-         << time_base_denominator() << ":"
-         << sample_aspect_ratio_numerator() << ":"
-         << sample_aspect_ratio_denominator();
-
-       int r;
-
-       if ((r = avfilter_graph_create_filter (&_buffer_src_context, buffer_src, "in", a.str().c_str(), 0, graph)) < 0) {
-               throw DecodeError ("could not create buffer source");
-       }
-
-       AVBufferSinkParams* sink_params = av_buffersink_params_alloc ();
-       PixelFormat* pixel_fmts = new PixelFormat[2];
-       pixel_fmts[0] = pixel_format ();
-       pixel_fmts[1] = PIX_FMT_NONE;
-       sink_params->pixel_fmts = pixel_fmts;
-       
-       if (avfilter_graph_create_filter (&_buffer_sink_context, buffer_sink, "out", 0, sink_params, graph) < 0) {
-               throw DecodeError ("could not create buffer sink.");
-       }
-
-       AVFilterInOut* outputs = avfilter_inout_alloc ();
-       outputs->name = av_strdup("in");
-       outputs->filter_ctx = _buffer_src_context;
-       outputs->pad_idx = 0;
-       outputs->next = 0;
-
-       AVFilterInOut* inputs = avfilter_inout_alloc ();
-       inputs->name = av_strdup("out");
-       inputs->filter_ctx = _buffer_sink_context;
-       inputs->pad_idx = 0;
-       inputs->next = 0;
-
-       _log->log ("Using filter chain `" + filters + "'");
-
-#if LIBAVFILTER_VERSION_MAJOR == 2 && LIBAVFILTER_VERSION_MINOR == 15
-       if (avfilter_graph_parse (graph, filters.c_str(), inputs, outputs, 0) < 0) {
-               throw DecodeError ("could not set up filter graph.");
-       }
-#else  
-       if (avfilter_graph_parse (graph, filters.c_str(), &inputs, &outputs, 0) < 0) {
-               throw DecodeError ("could not set up filter graph.");
+       if (!_pending.empty ()) {
+               _pending.pop_front ();
        }
-#endif 
-       
-       if (avfilter_graph_config (graph, 0) < 0) {
-               throw DecodeError ("could not configure filter graph.");
-       }
-
-       /* XXX: leaking `inputs' / `outputs' ? */
 }
 
 void
-Decoder::process_subtitle (shared_ptr<TimedSubtitle> s)
-{
-       _timed_subtitle = s;
-       
-       if (_timed_subtitle && _opt->apply_crop) {
-               Position const p = _timed_subtitle->subtitle()->position ();
-               _timed_subtitle->subtitle()->set_position (Position (p.x - _film->crop().left, p.y - _film->crop().top));
-       }
-}
-
-
-int
-Decoder::bytes_per_audio_sample () const
+Decoder::seek (ContentTime, bool)
 {
-       return av_get_bytes_per_sample (audio_sample_format ());
+       _pending.clear ();
 }