#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 "util.h"
#include "log.h"
#include "decoder.h"
-#include "filter.h"
#include "delay_line.h"
-#include "ffmpeg_compatibility.h"
#include "subtitle.h"
+#include "filter_graph.h"
using std::string;
using std::stringstream;
using std::min;
+using std::pair;
+using std::list;
using boost::shared_ptr;
+using boost::optional;
/** @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.
*/
-Decoder::Decoder (boost::shared_ptr<Film> f, boost::shared_ptr<const Options> o, Job* j, bool minimal, bool ignore_length)
+Decoder::Decoder (boost::shared_ptr<Film> f, boost::shared_ptr<const Options> o, Job* j)
: _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)
+ , _audio_frame (0)
, _delay_line (0)
- , _delay_in_bytes (0)
- , _audio_frames_processed (0)
+ , _delay_in_frames (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 */
+/** Start off a decode processing run. This should only be called once on
+ * a given Decoder object.
+ */
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;
+ if (_audio_stream) {
+ _delay_in_frames = _film->audio_delay() * _audio_stream.get().sample_rate() / 1000;
+ _delay_line = new DelayLine (_audio_stream.get().channels(), _delay_in_frames);
+ }
}
/** 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 (_delay_in_frames < 0 && _opt->decode_audio && _audio_stream) {
+ shared_ptr<AudioBuffers> b (new AudioBuffers (_audio_stream.get().channels(), -_delay_in_frames));
+ b->make_silent ();
+ emit_audio (b);
}
- /* If we cut the decode off, the audio may be short; push some silence
- in to get it to the right length.
- */
+ if (_opt->decode_audio && _audio_stream) {
+
+ /* Ensure that our video and audio emissions are the same length */
+
+ int64_t audio_short_by_frames = video_frames_to_audio_frames (_video_frame, _audio_stream.get().sample_rate(), frames_per_second()) - _audio_frame;
+
+ _film->log()->log (
+ String::compose (
+ "Decoder has emitted %1 video frames (which equals %2 audio frames) and %3 audio frames",
+ _video_frame,
+ video_frames_to_audio_frames (_video_frame, _audio_stream.get().sample_rate(), frames_per_second()),
+ _audio_frame
+ )
+ );
+
+ if (audio_short_by_frames < 0) {
+
+ _film->log()->log (String::compose ("Emitted %1 too many audio frames", -audio_short_by_frames));
+
+ /* We have emitted more audio than video. Emit enough black video frames so that we reverse this */
+ int const black_video_frames = ceil (-audio_short_by_frames * frames_per_second() / _audio_stream.get().sample_rate());
- 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;
+ _film->log()->log (String::compose ("Emitting %1 frames of black video", black_video_frames));
- _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)
- );
+ shared_ptr<Image> black (new CompactImage (pixel_format(), native_size()));
+ black->make_black ();
+ for (int i = 0; i < black_video_frames; ++i) {
+ emit_video (black, shared_ptr<Subtitle> ());
+ }
+
+ /* Now recompute our check value */
+ audio_short_by_frames = video_frames_to_audio_frames (_video_frame, _audio_stream.get().sample_rate(), frames_per_second()) - _audio_frame;
+ }
- 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;
+ if (audio_short_by_frames > 0) {
+ _film->log()->log (String::compose ("Emitted %1 too few audio frames", audio_short_by_frames));
+ shared_ptr<AudioBuffers> b (new AudioBuffers (_audio_stream.get().channels(), audio_short_by_frames));
+ b->make_silent ();
+ emit_audio (b);
}
}
}
while (pass () == false) {
if (_job && _film->dcp_length()) {
- _job->set_progress (float (_video_frame) / _film->dcp_length().get());
+ _job->set_progress (float (_video_frame) / _film->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)
-{
- /* Push into the delay line */
- size = _delay_line->feed (data, size);
-
- emit_audio (data, size);
-}
-
-void
-Decoder::emit_audio (uint8_t* data, int size)
+Decoder::process_audio (shared_ptr<AudioBuffers> audio)
{
- /* 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:
- {
- 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);
- }
-
/* 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) {
+ for (int i = 0; i < audio->channels(); ++i) {
+ for (int j = 0; j < audio->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);
+ _delay_line->feed (audio);
+ emit_audio (audio);
}
/** Called by subclasses to tell the world that some video data is ready.
void
Decoder::process_video (AVFrame* frame)
{
- if (_minimal) {
- ++_video_frame;
- return;
- }
-
- /* 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 LIBAVFILTER_VERSION_MAJOR == 2 && LIBAVFILTER_VERSION_MINOR >= 53 && LIBAVFILTER_VERSION_MINOR <= 61
+ shared_ptr<FilterGraph> graph;
- if (av_vsrc_buffer_add_frame (_buffer_src_context, frame, 0) < 0) {
- throw DecodeError ("could not push buffer into filter chain.");
+ list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
+ while (i != _filter_graphs.end() && !(*i)->can_process (Size (frame->width, frame->height), (AVPixelFormat) frame->format)) {
+ ++i;
}
-#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.");
+ if (i == _filter_graphs.end ()) {
+ graph.reset (new FilterGraph (_film, this, _opt->apply_crop, Size (frame->width, frame->height), (AVPixelFormat) frame->format));
+ _filter_graphs.push_back (graph);
+ _film->log()->log (String::compose ("New graph for %1x%2, pixel format %3", frame->width, frame->height, frame->format));
+ } else {
+ graph = *i;
}
-#else
+ list<shared_ptr<Image> > images = graph->process (frame);
- 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");
+ for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) {
+ shared_ptr<Subtitle> sub;
+ if (_timed_subtitle && _timed_subtitle->displayed_at (double (video_frame()) / _film->frames_per_second())) {
+ sub = _timed_subtitle->subtitle ();
}
-
- 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;
- }
+ emit_video (*i, sub);
}
}
-
-/** Set up a video filtering chain to include cropping and any filters that are specified
- * by the Film.
- */
void
-Decoder::setup_video_filters ()
+Decoder::repeat_last_video ()
{
- 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");
+ if (!_last_image) {
+ _last_image.reset (new CompactImage (pixel_format(), native_size()));
+ _last_image->make_black ();
}
- 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;
+ emit_video (_last_image, _last_subtitle);
+}
- _log->log ("Using filter chain `" + filters + "'");
+void
+Decoder::emit_video (shared_ptr<Image> image, shared_ptr<Subtitle> sub)
+{
+ TIMING ("Decoder emits %1", _video_frame);
+ Video (image, _video_frame, sub);
+ ++_video_frame;
-#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.");
- }
-#endif
-
- if (avfilter_graph_config (graph, 0) < 0) {
- throw DecodeError ("could not configure filter graph.");
- }
+ _last_image = image;
+ _last_subtitle = sub;
+}
- /* XXX: leaking `inputs' / `outputs' ? */
+void
+Decoder::emit_audio (shared_ptr<AudioBuffers> audio)
+{
+ Audio (audio, _audio_frame);
+ _audio_frame += audio->frames ();
}
void
}
}
+void
+Decoder::set_audio_stream (optional<AudioStream> s)
+{
+ _audio_stream = s;
+}
-int
-Decoder::bytes_per_audio_sample () const
+void
+Decoder::set_subtitle_stream (optional<SubtitleStream> s)
{
- return av_get_bytes_per_sample (audio_sample_format ());
+ _subtitle_stream = s;
}