#include <iostream>
#include <stdint.h>
-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 <boost/lexical_cast.hpp>
#include "film.h"
#include "format.h"
#include "job.h"
-#include "film_state.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 "filter_graph.h"
-using namespace std;
-using namespace boost;
+using std::string;
+using std::stringstream;
+using std::min;
+using std::list;
+using boost::shared_ptr;
-/** @param s FilmState of the Film.
+/** @param f Film.
* @param o Options.
* @param j Job that we are running within, or 0
- * @param l Log to use.
* @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<const FilmState> s, boost::shared_ptr<const Options> o, Job* j, Log* l, bool minimal, bool ignore_length)
- : _fs (s)
+Decoder::Decoder (boost::shared_ptr<Film> f, boost::shared_ptr<const Options> o, Job* j, bool minimal)
+ : _film (f)
, _opt (o)
, _job (j)
- , _log (l)
, _minimal (minimal)
- , _ignore_length (ignore_length)
- , _video_frame (0)
- , _buffer_src_context (0)
- , _buffer_sink_context (0)
-#if HAVE_SWRESAMPLE
- , _swr_context (0)
-#endif
- , _have_setup_video_filters (false)
+ , _video_frame_index (0)
, _delay_line (0)
, _delay_in_bytes (0)
, _audio_frames_processed (0)
{
- if (_opt->decode_video_frequency != 0 && _fs->length == 0) {
- throw DecodeError ("cannot do a partial decode if length == 0");
- }
+
}
Decoder::~Decoder ()
void
Decoder::process_begin ()
{
- if (_fs->audio_sample_rate != dcp_audio_sample_rate (_fs->audio_sample_rate)) {
-#if HAVE_SWRESAMPLE
- _swr_context = swr_alloc_set_opts (
- 0,
- audio_channel_layout(),
- audio_sample_format(),
- dcp_audio_sample_rate (_fs->audio_sample_rate),
- audio_channel_layout(),
- audio_sample_format(),
- _fs->audio_sample_rate,
- 0, 0
- );
-
- swr_init (_swr_context);
-#else
- throw DecodeError ("Cannot resample audio as libswresample is not present");
-#endif
- } else {
-#if HAVE_SWRESAMPLE
- _swr_context = 0;
-#endif
- }
-
- _delay_in_bytes = _fs->audio_delay * _fs->audio_sample_rate * _fs->audio_channels * _fs->bytes_per_sample() / 1000;
+ _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);
void
Decoder::process_end ()
{
-#if HAVE_SWRESAMPLE
- if (_swr_context) {
-
- int mop = 0;
- while (1) {
- uint8_t buffer[256 * _fs->bytes_per_sample() * _fs->audio_channels];
- uint8_t* out[1] = {
- buffer
- };
-
- int const frames = swr_convert (_swr_context, out, 256, 0, 0);
-
- if (frames < 0) {
- throw DecodeError ("could not run sample-rate converter");
- }
-
- if (frames == 0) {
- break;
- }
-
- mop += frames;
- int available = _delay_line->feed (buffer, frames * _fs->audio_channels * _fs->bytes_per_sample());
- Audio (buffer, available);
- }
-
- swr_free (&_swr_context);
- }
-#endif
-
if (_delay_in_bytes < 0) {
uint8_t remainder[-_delay_in_bytes];
_delay_line->get_remaining (remainder);
- _audio_frames_processed += _delay_in_bytes / (_fs->audio_channels * _fs->bytes_per_sample());
- Audio (remainder, _delay_in_bytes);
+ _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 audio_short_by_frames =
- ((int64_t) decoding_frames() * dcp_audio_sample_rate (_fs->audio_sample_rate) / _fs->frames_per_second)
- - _audio_frames_processed;
+ int64_t const video_length_in_audio_frames = ((int64_t) video_frame_index() * 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 ("Source length is %1 (%2 audio frames); %3 frames of audio processed.",
+ video_frame_index(),
+ video_length_in_audio_frames,
+ _audio_frames_processed)
+ );
+
+ if (audio_short_by_frames >= 0 && _opt->decode_audio) {
+
+ _film->log()->log (String::compose ("Source length is %1; %2 frames of audio processed.", video_frame_index(), _audio_frames_processed));
+ _film->log()->log (String::compose ("Adding %1 frames of silence to the end.", audio_short_by_frames));
- if (audio_short_by_frames >= 0) {
- int bytes = audio_short_by_frames * _fs->audio_channels * _fs->bytes_per_sample();
+ /* 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();
- int const silence_size = 64 * 1024;
+ 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) {
- int const t = min (bytes, silence_size);
- Audio (silence, t);
+ int64_t const t = min (bytes, silence_size);
+ emit_audio (silence, t);
bytes -= t;
}
}
{
process_begin ();
- if (_job && _ignore_length) {
+ if (_job && !_film->dcp_length()) {
_job->set_progress_unknown ();
}
while (pass () == false) {
- if (_job && !_ignore_length) {
- _job->set_progress (float (_video_frame) / decoding_frames ());
+ if (_job && _film->dcp_length()) {
+ SourceFrame const p = _video_frame_index - _film->dcp_trim_start();
+ _job->set_progress (float (p) / _film->dcp_length().get());
}
}
process_end ();
}
-/** @return Number of frames that we will be decoding */
-int
-Decoder::decoding_frames () const
-{
- if (_opt->num_frames > 0) {
- return _opt->num_frames;
- }
-
- return _fs->length;
-}
-
-/** 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 (_opt->num_frames != 0 && _video_frame >= _opt->num_frames) {
- return true;
- }
-
- return do_pass ();
-}
-
/** Called by subclasses to tell the world that some audio data is ready
- * @param data Interleaved audio data, in FilmState::audio_sample_format.
+ * @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)
{
- /* Here's samples per channel */
- int const samples = size / _fs->bytes_per_sample();
+ /* Push into the delay line */
+ size = _delay_line->feed (data, size);
-#if HAVE_SWRESAMPLE
- /* And here's frames (where 1 frame is a collection of samples, 1 for each channel,
- so for 5.1 a frame would be 6 samples)
- */
- int const frames = samples / _fs->audio_channels;
-#endif
+ emit_audio (data, size);
+}
- /* Maybe apply gain */
- if (_fs->audio_gain != 0) {
- float const linear_gain = pow (10, _fs->audio_gain / 20);
- uint8_t* p = data;
- switch (_fs->audio_sample_format) {
- case AV_SAMPLE_FMT_S16:
- for (int i = 0; i < samples; ++i) {
- /* XXX: assumes little-endian; also we should probably be dithering here */
-
- /* unsigned sample */
- int const ou = p[0] | (p[1] << 8);
-
- /* signed sample */
- int const os = ou >= 0x8000 ? (- 0x10000 + ou) : ou;
-
- /* signed sample with altered gain */
- int const gs = int (os * linear_gain);
-
- /* unsigned sample with altered gain */
- int const gu = gs > 0 ? gs : (0x10000 + gs);
-
- /* write it back */
- p[0] = gu & 0xff;
- p[1] = (gu & 0xff00) >> 8;
- p += 2;
+void
+Decoder::emit_audio (uint8_t* data, int size)
+{
+ if (size == 0) {
+ return;
+ }
+
+ assert (_film->audio_channels());
+ assert (bytes_per_audio_sample());
+
+ /* Deinterleave and convert to float */
+
+ assert ((size % (bytes_per_audio_sample() * 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 (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;
}
- break;
- default:
- assert (false);
}
}
- /* This is a buffer we might use if we are sample-rate converting;
- it will need freeing if so.
- */
- uint8_t* out_buffer = 0;
-
- /* Maybe sample-rate convert */
-#if HAVE_SWRESAMPLE
- if (_swr_context) {
-
- uint8_t const * in[2] = {
- data,
- 0
- };
-
- /* Compute the resampled frame count and add 32 for luck */
- int const out_buffer_size_frames = ceil (frames * float (dcp_audio_sample_rate (_fs->audio_sample_rate)) / _fs->audio_sample_rate) + 32;
- int const out_buffer_size_bytes = out_buffer_size_frames * _fs->audio_channels * _fs->bytes_per_sample();
- out_buffer = new uint8_t[out_buffer_size_bytes];
-
- uint8_t* out[2] = {
- out_buffer,
- 0
- };
-
- /* Resample audio */
- int out_frames = swr_convert (_swr_context, out, out_buffer_size_frames, in, frames);
- if (out_frames < 0) {
- throw DecodeError ("could not run sample-rate converter");
+ 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;
- /* And point our variables at the resampled audio */
- data = out_buffer;
- size = out_frames * _fs->audio_channels * _fs->bytes_per_sample();
+ default:
+ assert (false);
}
-#endif
-
- /* Update the number of audio frames we've pushed to the encoder */
- _audio_frames_processed += size / (_fs->audio_channels * _fs->bytes_per_sample ());
- /* Push into the delay line and then tell the world what we've got */
- int available = _delay_line->feed (data, size);
- Audio (data, available);
+ /* 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 ();
- /* Delete the sample-rate conversion buffer, if it exists */
- delete[] out_buffer;
+ Audio (audio);
}
/** Called by subclasses to tell the world that some video data is ready.
void
Decoder::process_video (AVFrame* frame)
{
+ assert (_film->length());
+
if (_minimal) {
- ++_video_frame;
+ ++_video_frame_index;
return;
}
- /* Use FilmState::length here as our one may be wrong */
-
- int gap = 0;
- if (_opt->decode_video_frequency != 0) {
- gap = _fs->length / _opt->decode_video_frequency;
- }
+ /* Use Film::length here as our one may be wrong */
- if (_opt->decode_video_frequency != 0 && gap != 0 && (_video_frame % gap) != 0) {
- ++_video_frame;
+ if (_opt->decode_video_skip != 0 && (_video_frame_index % _opt->decode_video_skip) != 0) {
+ ++_video_frame_index;
return;
}
-#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.");
+ if (_film->dcp_trim_start() > _video_frame_index || (_film->length().get() - _film->dcp_trim_end()) < _video_frame_index) {
+ ++_video_frame_index;
+ return;
}
-#elif LIBAVFILTER_VERSION_MAJOR == 2 && LIBAVFILTER_VERSION_MINOR == 15
+ shared_ptr<FilterGraph> graph;
- 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.");
+ 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;
}
-#else
-
- if (av_buffersrc_write_frame (_buffer_src_context, frame) < 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;
}
-#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
+ list<shared_ptr<Image> > images = graph->process (frame);
-#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_index()) / _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 ();
- }
-
- Video (image, _video_frame);
- ++_video_frame;
- }
+ TIMING ("Decoder emits %1", _video_frame_index);
+ Video (*i, _video_frame_index, sub);
+ ++_video_frame_index;
+ _last_image = *i;
+ _last_subtitle = 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 = _fs->cropped_size (native_size ());
- fs << crop_string (Position (_fs->crop.left, _fs->crop.top), size_after_crop);
- } else {
- size_after_crop = native_size ();
- fs << crop_string (Position (0, 0), size_after_crop);
+ if (!_last_image) {
+ _last_image.reset (new CompactImage (pixel_format(), native_size()));
+ _last_image->make_black ();
}
-
- string filters = Filter::ffmpeg_strings (_fs->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");
- }
-
- enum PixelFormat pixel_formats[] = { pixel_format(), PIX_FMT_NONE };
- if (avfilter_graph_create_filter (&_buffer_sink_context, buffer_sink, "out", 0, pixel_formats, 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 + "'");
+ Video (_last_image, _video_frame_index, _last_subtitle);
+ ++_video_frame_index;
+}
-#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
+void
+Decoder::process_subtitle (shared_ptr<TimedSubtitle> s)
+{
+ _timed_subtitle = s;
- if (avfilter_graph_config (graph, 0) < 0) {
- throw DecodeError ("could not configure filter graph.");
+ 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));
}
-
- /* XXX: leaking `inputs' / `outputs' ? */
}
+
+int
+Decoder::bytes_per_audio_sample () const
+{
+ return av_get_bytes_per_sample (audio_sample_format ());
+}