#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 "filter.h"
#include "delay_line.h"
#include "ffmpeg_compatibility.h"
+#include "subtitle.h"
using namespace std;
using namespace boost;
, _video_frame (0)
, _buffer_src_context (0)
, _buffer_sink_context (0)
- , _swr_context (0)
, _have_setup_video_filters (false)
, _delay_line (0)
, _delay_in_bytes (0)
, _audio_frames_processed (0)
{
- if (_opt->decode_video_frequency != 0 && _fs->length == 0) {
+ if (_opt->decode_video_frequency != 0 && _fs->length() == 0) {
throw DecodeError ("cannot do a partial decode if length == 0");
}
}
delete _delay_line;
}
+/** Start off a decode processing run */
void
Decoder::process_begin ()
{
- if (_fs->audio_sample_rate != dcp_audio_sample_rate (_fs->audio_sample_rate)) {
- _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 {
- _swr_context = 0;
- }
-
- _delay_in_bytes = _fs->audio_delay * _fs->audio_sample_rate * _fs->audio_channels * _fs->bytes_per_sample() / 1000;
+ _delay_in_bytes = _fs->audio_delay() * _fs->audio_sample_rate() * _fs->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 (_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);
- }
-
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 / (_fs->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.
*/
- int const audio_short_by_frames =
- (decoding_frames() * dcp_audio_sample_rate (_fs->audio_sample_rate) / _fs->frames_per_second)
+ int64_t const audio_short_by_frames =
+ ((int64_t) _fs->dcp_length() * _fs->target_sample_rate() / _fs->frames_per_second())
- _audio_frames_processed;
- int bytes = audio_short_by_frames * _fs->audio_channels * _fs->bytes_per_sample();
+ if (audio_short_by_frames >= 0) {
- int const silence_size = 64 * 1024;
- uint8_t silence[silence_size];
- memset (silence, 0, silence_size);
+ stringstream s;
+ s << "Adding " << audio_short_by_frames << " frames of silence to the end.";
+ _log->log (s.str ());
- while (bytes) {
- int const t = min (bytes, silence_size);
- Audio (silence, t);
- bytes -= t;
+ int64_t bytes = audio_short_by_frames * _fs->audio_channels() * bytes_per_audio_sample();
+
+ int64_t const silence_size = 64 * 1024;
+ 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;
+ }
}
}
while (pass () == false) {
if (_job && !_ignore_length) {
- _job->set_progress (float (_video_frame) / decoding_frames ());
+ _job->set_progress (float (_video_frame) / _fs->dcp_length());
}
}
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.
_have_setup_video_filters = true;
}
- if (_opt->num_frames != 0 && _video_frame >= _opt->num_frames) {
+ if (_video_frame >= _fs->dcp_length()) {
return true;
}
}
/** 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 FilmState::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);
- /* 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;
+ 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 */
+void
+Decoder::emit_audio (uint8_t* data, int size)
+{
+ /* Deinterleave and convert to float */
- /* unsigned sample */
- int const ou = p[0] | (p[1] << 8);
+ int const total_samples = size / bytes_per_audio_sample();
+ int const frames = total_samples / _fs->audio_channels();
+ shared_ptr<AudioBuffers> audio (new AudioBuffers (_fs->audio_channels(), frames));
- /* signed sample */
- int const os = ou >= 0x8000 ? (- 0x10000 + ou) : ou;
+ switch (audio_sample_format()) {
+ case AV_SAMPLE_FMT_S16:
+ {
+ uint8_t* p = data;
+ int sample = 0;
+ int channel = 0;
+ for (int i = 0; i < total_samples; ++i) {
+ /* unsigned sample */
+ int const ou = p[0] | (p[1] << 8);
+ /* signed sample */
+ int const os = ou >= 0x8000 ? (- 0x10000 + ou) : ou;
+ /* float sample */
+ audio->data(channel)[sample] = float(os) / 0x8000;
+
+ ++channel;
+ if (channel == _fs->audio_channels()) {
+ channel = 0;
+ ++sample;
+ }
- /* signed sample with altered gain */
- int const gs = int (os * linear_gain);
+ p += 2;
+ }
+ }
+ break;
+
+ case AV_SAMPLE_FMT_FLTP:
+ {
+ float* p = reinterpret_cast<float*> (data);
+ for (int i = 0; i < _fs->audio_channels(); ++i) {
+ memcpy (audio->data(i), p, frames * sizeof(float));
+ p += frames;
+ }
+ }
+ break;
- /* unsigned sample with altered gain */
- int const gu = gs > 0 ? gs : (0x10000 + gs);
+ default:
+ assert (false);
+ }
- /* write it back */
- p[0] = gu & 0xff;
- p[1] = (gu & 0xff00) >> 8;
- p += 2;
+ /* Maybe apply gain */
+ if (_fs->audio_gain() != 0) {
+ float const linear_gain = pow (10, _fs->audio_gain() / 20);
+ for (int i = 0; i < _fs->audio_channels(); ++i) {
+ for (int j = 0; j < frames; ++j) {
+ audio->data(i)[j] *= linear_gain;
}
- 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 (_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");
- }
-
- /* And point our variables at the resampled audio */
- data = out_buffer;
- size = out_frames * _fs->audio_channels * _fs->bytes_per_sample();
- }
-
/* Update the number of audio frames we've pushed to the encoder */
- _audio_frames_processed += size / (_fs->audio_channels * _fs->bytes_per_sample ());
+ _audio_frames_processed += frames;
- /* Push into the delay line and then tell the world what we've got */
- int available = _delay_line->feed (data, size);
- Audio (data, available);
-
- /* 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.
int gap = 0;
if (_opt->decode_video_frequency != 0) {
- gap = _fs->length / _opt->decode_video_frequency;
+ gap = _fs->length() / _opt->decode_video_frequency;
}
if (_opt->decode_video_frequency != 0 && gap != 0 && (_video_frame % gap) != 0) {
throw DecodeError ("could not push buffer into filter chain.");
}
-#else
+#elif LIBAVFILTER_VERSION_MAJOR == 2 && LIBAVFILTER_VERSION_MINOR == 15
-#if 0
-
AVRational par;
par.num = sample_aspect_ratio_numerator ();
par.den = sample_aspect_ratio_denominator ();
throw DecodeError ("could not push buffer into filter chain.");
}
-#endif
+#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 >= 23 && LIBAVFILTER_VERSION_MINOR <= 61
+#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 == 53
+#if LIBAVFILTER_VERSION_MAJOR == 2 && LIBAVFILTER_VERSION_MINOR >= 15
int r = avfilter_request_frame (_buffer_sink_context->inputs[0]);
if (r < 0) {
image->make_black ();
}
- Video (image, _video_frame);
+ shared_ptr<Subtitle> sub;
+ if (_timed_subtitle && _timed_subtitle->displayed_at (double (last_video_frame()) / rint (_fs->frames_per_second()))) {
+ sub = _timed_subtitle->subtitle ();
+ }
+
+ TIMING ("Decoder emits %1", _video_frame);
+ Video (image, _video_frame, sub);
++_video_frame;
}
}
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);
+ 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);
}
- string filters = Filter::ffmpeg_strings (_fs->filters).first;
+ string filters = Filter::ffmpeg_strings (_fs->filters()).first;
if (!filters.empty ()) {
filters += ",";
}
throw DecodeError ("Could not find buffer src filter");
}
- AVFilter* buffer_sink = avfilter_get_by_name("buffersink");
- if (buffer_sink == 0) {
- throw DecodeError ("Could not create buffer sink filter");
- }
+ AVFilter* buffer_sink = get_sink ();
stringstream a;
a << native_size().width << ":"
<< 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) {
+ 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.");
}
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.");
}
-
+#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 (_opt->apply_crop) {
+ Position const p = _timed_subtitle->subtitle()->position ();
+ _timed_subtitle->subtitle()->set_position (Position (p.x - _fs->crop().left, p.y - _fs->crop().top));
+ }
+}
+
+
+int
+Decoder::bytes_per_audio_sample () const
+{
+ return av_get_bytes_per_sample (audio_sample_format ());
+}