/*
- Copyright (C) 2012 Carl Hetherington <cth@carlh.net>
+ Copyright (C) 2012-2014 Carl Hetherington <cth@carlh.net>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
#include "audio_buffers.h"
#include "exceptions.h"
#include "log.h"
+#include "resampler.h"
+#include "util.h"
+#include "film.h"
#include "i18n.h"
using std::stringstream;
+using std::list;
+using std::pair;
+using std::cout;
+using std::min;
using boost::optional;
using boost::shared_ptr;
-AudioDecoder::AudioDecoder (shared_ptr<const Film> f, shared_ptr<const AudioContent> c)
- : Decoder (f)
- , _next_audio (0)
- , _audio_content (c)
- , _output_audio_frame_rate (_audio_content->output_audio_frame_rate (f))
+AudioDecoder::AudioDecoder (shared_ptr<const AudioContent> content)
+ : _audio_content (content)
{
- if (_audio_content->content_audio_frame_rate() != _output_audio_frame_rate) {
-
- shared_ptr<const Film> film = _film.lock ();
- assert (film);
-
- stringstream s;
- s << String::compose ("Will resample audio from %1 to %2", _audio_content->content_audio_frame_rate(), _output_audio_frame_rate);
- film->log()->log (s.str ());
-
- /* We will be using planar float data when we call the
- resampler. As far as I can see, the audio channel
- layout is not necessary for our purposes; it seems
- only to be used get the number of channels and
- decide if rematrixing is needed. It won't be, since
- input and output layouts are the same.
- */
-
- _swr_context = swr_alloc_set_opts (
- 0,
- av_get_default_channel_layout (MAX_AUDIO_CHANNELS),
- AV_SAMPLE_FMT_FLTP,
- _output_audio_frame_rate,
- av_get_default_channel_layout (MAX_AUDIO_CHANNELS),
- AV_SAMPLE_FMT_FLTP,
- _audio_content->content_audio_frame_rate(),
- 0, 0
- );
-
- swr_init (_swr_context);
- } else {
- _swr_context = 0;
+ if (content->output_audio_frame_rate() != content->content_audio_frame_rate() && content->audio_channels ()) {
+ _resampler.reset (new Resampler (content->content_audio_frame_rate(), content->output_audio_frame_rate(), content->audio_channels ()));
}
+
+ reset_decoded_audio ();
}
-AudioDecoder::~AudioDecoder ()
+void
+AudioDecoder::reset_decoded_audio ()
{
- if (_swr_context) {
- swr_free (&_swr_context);
- }
+ _decoded_audio = ContentAudio (shared_ptr<AudioBuffers> (new AudioBuffers (_audio_content->audio_channels(), 0)), 0);
}
-
-#if 0
-void
-AudioDecoder::process_end ()
+shared_ptr<ContentAudio>
+AudioDecoder::get_audio (AudioFrame frame, AudioFrame length, bool accurate)
{
- if (_swr_context) {
+ shared_ptr<ContentAudio> dec;
- shared_ptr<const Film> film = _film.lock ();
- assert (film);
+ AudioFrame const end = frame + length - 1;
- shared_ptr<AudioBuffers> out (new AudioBuffers (film->audio_mapping().dcp_channels(), 256));
-
- while (1) {
- int const frames = swr_convert (_swr_context, (uint8_t **) out->data(), 256, 0, 0);
-
- if (frames < 0) {
- throw EncodeError (_("could not run sample-rate converter"));
- }
+ if (frame < _decoded_audio.frame || end > (_decoded_audio.frame + length * 4)) {
+ /* Either we have no decoded data, or what we do have is a long way from what we want: seek */
+ seek (ContentTime::from_frames (frame, _audio_content->content_audio_frame_rate()), accurate);
+ }
- if (frames == 0) {
- break;
- }
+ AudioFrame decoded_offset = 0;
+
+ /* Now enough pass() calls will either:
+ * (a) give us what we want, or
+ * (b) hit the end of the decoder.
+ *
+ * If we are being accurate, we want the right frames,
+ * otherwise any frames will do.
+ */
+ if (accurate) {
+ while (!pass() && _decoded_audio.audio->frames() < length) {}
+ /* Use decoded_offset of 0, as we don't really care what frames we return */
+ } else {
+ while (!pass() && (_decoded_audio.frame > frame || (_decoded_audio.frame + _decoded_audio.audio->frames()) < end)) {}
+ decoded_offset = frame - _decoded_audio.frame;
+ }
- out->set_frames (frames);
- _writer->write (out);
- }
+ AudioFrame const amount_left = _decoded_audio.audio->frames() - decoded_offset;
+
+ AudioFrame const to_return = min (amount_left, length);
+ shared_ptr<AudioBuffers> out (new AudioBuffers (_decoded_audio.audio->channels(), to_return));
+ out->copy_from (_decoded_audio.audio.get(), to_return, decoded_offset, 0);
+
+ /* Clean up decoded */
+ _decoded_audio.audio->move (decoded_offset + to_return, 0, amount_left - to_return);
+ _decoded_audio.audio->set_frames (amount_left - to_return);
- }
+ return shared_ptr<ContentAudio> (new ContentAudio (out, frame));
}
-#endif
+/** Called by subclasses when audio data is ready.
+ *
+ * Audio timestamping is made hard by many factors, but perhaps the most entertaining is resampling.
+ * We have to assume that we are feeding continuous data into the resampler, and so we get continuous
+ * data out. Hence we do the timestamping here, post-resampler, just by counting samples.
+ *
+ * The time is passed in here so that after a seek we can set up our _audio_position. The
+ * time is ignored once this has been done.
+ */
void
-AudioDecoder::emit_audio (shared_ptr<const AudioBuffers> data, Time time)
+AudioDecoder::audio (shared_ptr<const AudioBuffers> data, ContentTime time)
{
- /* XXX: map audio to 5.1 */
-
- /* Maybe resample */
- if (_swr_context) {
+ if (_resampler) {
+ data = _resampler->run (data);
+ }
- /* Compute the resampled frames count and add 32 for luck */
- int const max_resampled_frames = ceil ((int64_t) data->frames() * _output_audio_frame_rate / _audio_content->content_audio_frame_rate()) + 32;
+ if (!_audio_position) {
+ _audio_position = time.frames (_audio_content->output_audio_frame_rate ());
+ }
- shared_ptr<AudioBuffers> resampled (new AudioBuffers (MAX_AUDIO_CHANNELS, max_resampled_frames));
+ assert (_audio_position.get() >= (_decoded_audio.frame + _decoded_audio.audio->frames()));
- /* Resample audio */
- int const resampled_frames = swr_convert (
- _swr_context, (uint8_t **) resampled->data(), max_resampled_frames, (uint8_t const **) data->data(), data->frames()
- );
-
- if (resampled_frames < 0) {
- throw EncodeError (_("could not run sample-rate converter"));
- }
+ /* Resize _decoded_audio to fit the new data */
+ int const new_size = _audio_position.get() + data->frames() - _decoded_audio.frame;
+ _decoded_audio.audio->ensure_size (new_size);
+ _decoded_audio.audio->set_frames (new_size);
- resampled->set_frames (resampled_frames);
-
- /* And point our variables at the resampled audio */
- data = resampled;
- }
+ /* Copy new data in */
+ _decoded_audio.audio->copy_from (data.get(), data->frames(), 0, _audio_position.get() - _decoded_audio.frame);
+ _audio_position = _audio_position.get() + data->frames ();
+}
- Audio (data, time);
+/* XXX: called? */
+void
+AudioDecoder::flush ()
+{
+ if (!_resampler) {
+ return;
+ }
- shared_ptr<const Film> film = _film.lock ();
- assert (film);
- _next_audio = time + film->audio_frames_to_time (data->frames());
+ /*
+ shared_ptr<const AudioBuffers> b = _resampler->flush ();
+ if (b) {
+ _pending.push_back (shared_ptr<DecodedAudio> (new DecodedAudio (b, _audio_position.get ())));
+ _audio_position = _audio_position.get() + b->frames ();
+ }
+ */
}
-
+void
+AudioDecoder::seek (ContentTime, bool)
+{
+ _audio_position.reset ();
+ reset_decoded_audio ();
+}