Do resampling in AudioDecoder rather than Player.
[dcpomatic.git] / src / lib / ffmpeg_decoder.cc
index 164785c8368bb0fbffea1a28d468247df4eadf47..baec57f3d24e19473df2dd74788caa7a411e1841 100644 (file)
@@ -40,6 +40,7 @@
 #include "audio_decoder.h"
 #include "compose.hpp"
 #include "subtitle_content.h"
+#include "audio_content.h"
 #include <dcp/subtitle_string.h>
 #include <sub/ssa_reader.h>
 #include <sub/subtitle.h>
@@ -73,11 +74,14 @@ using std::map;
 using boost::shared_ptr;
 using boost::is_any_of;
 using boost::split;
+using boost::optional;
+using boost::dynamic_pointer_cast;
 using dcp::Size;
 
 FFmpegDecoder::FFmpegDecoder (shared_ptr<const FFmpegContent> c, shared_ptr<Log> log)
        : FFmpeg (c)
        , _log (log)
+       , _have_current_subtitle (false)
 {
        if (c->video) {
                video.reset (new VideoDecoder (this, c, log));
@@ -91,14 +95,7 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr<const FFmpegContent> c, shared_ptr<Log>
        }
 
        if (c->subtitle) {
-               subtitle.reset (
-                       new SubtitleDecoder (
-                               this,
-                               c->subtitle,
-                               bind (&FFmpegDecoder::image_subtitles_during, this, _1, _2),
-                               bind (&FFmpegDecoder::text_subtitles_during, this, _1, _2)
-                               )
-                       );
+               subtitle.reset (new SubtitleDecoder (this, c->subtitle, log));
        }
 }
 
@@ -121,7 +118,7 @@ FFmpegDecoder::flush ()
 }
 
 bool
-FFmpegDecoder::pass (PassReason reason, bool accurate)
+FFmpegDecoder::pass ()
 {
        int r = av_read_frame (_format_context, &_packet);
 
@@ -134,7 +131,7 @@ FFmpegDecoder::pass (PassReason reason, bool accurate)
                        /* Maybe we should fail here, but for now we'll just finish off instead */
                        char buf[256];
                        av_strerror (r, buf, sizeof(buf));
-                       LOG_ERROR (N_("error on av_read_frame (%1) (%2)"), buf, r);
+                       LOG_ERROR (N_("error on av_read_frame (%1) (%2)"), &buf[0], r);
                }
 
                flush ();
@@ -144,11 +141,11 @@ FFmpegDecoder::pass (PassReason reason, bool accurate)
        int const si = _packet.stream_index;
        shared_ptr<const FFmpegContent> fc = _ffmpeg_content;
 
-       if (_video_stream && si == _video_stream.get() && !video->ignore() && (accurate || reason != PASS_REASON_SUBTITLE)) {
+       if (_video_stream && si == _video_stream.get() && !video->ignore()) {
                decode_video_packet ();
        } else if (fc->subtitle_stream() && fc->subtitle_stream()->uses_index (_format_context, si)) {
                decode_subtitle_packet ();
-       } else if (accurate || reason != PASS_REASON_SUBTITLE) {
+       } else {
                decode_audio_packet ();
        }
 
@@ -303,17 +300,7 @@ FFmpegDecoder::bytes_per_audio_sample (shared_ptr<FFmpegAudioStream> stream) con
 void
 FFmpegDecoder::seek (ContentTime time, bool accurate)
 {
-       if (video) {
-               video->seek (time, accurate);
-       }
-
-       if (audio) {
-               audio->seek (time, accurate);
-       }
-
-       if (subtitle) {
-               subtitle->seek (time, accurate);
-       }
+       Decoder::seek (time, accurate);
 
        /* If we are doing an `accurate' seek, we need to use pre-roll, as
           we don't really know what the seek will give us.
@@ -326,7 +313,18 @@ FFmpegDecoder::seek (ContentTime time, bool accurate)
           http://www.mjbshaw.com/2012/04/seeking-in-ffmpeg-know-your-timestamp.html
        */
 
-       DCPOMATIC_ASSERT (_video_stream);
+       optional<int> stream;
+
+       if (_video_stream) {
+               stream = _video_stream;
+       } else {
+               shared_ptr<FFmpegAudioStream> s = dynamic_pointer_cast<FFmpegAudioStream> (_ffmpeg_content->audio->stream ());
+               if (s) {
+                       stream = s->index (_format_context);
+               }
+       }
+
+       DCPOMATIC_ASSERT (stream);
 
        ContentTime u = time - _pts_offset;
        if (u < ContentTime ()) {
@@ -334,18 +332,24 @@ FFmpegDecoder::seek (ContentTime time, bool accurate)
        }
        av_seek_frame (
                _format_context,
-               _video_stream.get(),
-               u.seconds() / av_q2d (_format_context->streams[_video_stream.get()]->time_base),
+               stream.get(),
+               u.seconds() / av_q2d (_format_context->streams[stream.get()]->time_base),
                AVSEEK_FLAG_BACKWARD
                );
 
-       avcodec_flush_buffers (video_codec_context());
+       if (video_codec_context ()) {
+               avcodec_flush_buffers (video_codec_context());
+       }
 
-       /* XXX: should be flushing audio buffers? */
+       BOOST_FOREACH (shared_ptr<FFmpegAudioStream> i, ffmpeg_content()->ffmpeg_audio_streams()) {
+               avcodec_flush_buffers (i->stream(_format_context)->codec);
+       }
 
        if (subtitle_codec_context ()) {
                avcodec_flush_buffers (subtitle_codec_context ());
        }
+
+       _have_current_subtitle = false;
 }
 
 void
@@ -376,7 +380,7 @@ FFmpegDecoder::decode_audio_packet ()
                if (decode_result < 0) {
                        /* avcodec_decode_audio4 can sometimes return an error even though it has decoded
                           some valid data; for example dca_subframe_footer can return AVERROR_INVALIDDATA
-                          if it overreads the auxiliary data.  ffplay carries on if frame_finished is true,
+                          if it overreads the auxiliary data.  ffplay carries on if frame_finished is true,
                           even in the face of such an error, so I think we should too.
 
                           Returning from the method here caused mantis #352.
@@ -400,13 +404,18 @@ FFmpegDecoder::decode_audio_packet ()
                        if (ct < ContentTime ()) {
                                /* Discard audio data that comes before time 0 */
                                Frame const remove = min (int64_t (data->frames()), (-ct).frames_ceil(double((*stream)->frame_rate ())));
-                               data->move (remove, 0, data->frames() - remove);
+                               data->move (data->frames() - remove, remove, 0);
                                data->set_frames (data->frames() - remove);
                                ct += ContentTime::from_frames (remove, (*stream)->frame_rate ());
                        }
 
-                       if (data->frames() > 0) {
-                               audio->give (*stream, data, ct);
+                       if (ct < ContentTime()) {
+                               LOG_WARNING ("Crazy timestamp %1", to_string (ct));
+                       }
+
+                       /* Give this data provided there is some, and its time is sane */
+                       if (ct >= ContentTime() && data->frames() > 0) {
+                               audio->emit (*stream, data, ct);
                        }
                }
 
@@ -451,7 +460,7 @@ FFmpegDecoder::decode_video_packet ()
 
                if (i->second != AV_NOPTS_VALUE) {
                        double const pts = i->second * av_q2d (_format_context->streams[_video_stream.get()]->time_base) + _pts_offset.seconds ();
-                       video->give (
+                       video->emit (
                                shared_ptr<ImageProxy> (new RawImageProxy (image)),
                                llrint (pts * _ffmpeg_content->active_video_frame_rate ())
                                );
@@ -472,10 +481,18 @@ FFmpegDecoder::decode_subtitle_packet ()
                return;
        }
 
+       /* Stop any current subtitle, either at the time it was supposed to stop, or now if now is sooner */
+       if (_have_current_subtitle) {
+               if (_current_subtitle_to) {
+                       subtitle->emit_stop (min(*_current_subtitle_to, subtitle_period(sub).from + _pts_offset));
+               } else {
+                       subtitle->emit_stop (subtitle_period(sub).from + _pts_offset);
+               }
+               _have_current_subtitle = false;
+       }
+
        if (sub.num_rects <= 0) {
-               /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
-                  indicate that the previous subtitle should stop.  We can ignore it here.
-               */
+               /* Nothing new in this subtitle */
                return;
        }
 
@@ -483,14 +500,11 @@ FFmpegDecoder::decode_subtitle_packet ()
           source that we may have chopped off for the DCP).
        */
        FFmpegSubtitlePeriod sub_period = subtitle_period (sub);
-       ContentTimePeriod period;
-       period.from = sub_period.from + _pts_offset;
+       ContentTime from;
+       from = sub_period.from + _pts_offset;
+       _have_current_subtitle = true;
        if (sub_period.to) {
-               /* We already know the subtitle period `to' time */
-               period.to = sub_period.to.get() + _pts_offset;
-       } else {
-               /* We have to look up the `to' time in the stream's records */
-               period.to = ffmpeg_content()->subtitle_stream()->find_subtitle_to (subtitle_id (sub));
+               _current_subtitle_to = *sub_period.to + _pts_offset;
        }
 
        for (unsigned int i = 0; i < sub.num_rects; ++i) {
@@ -500,13 +514,13 @@ FFmpegDecoder::decode_subtitle_packet ()
                case SUBTITLE_NONE:
                        break;
                case SUBTITLE_BITMAP:
-                       decode_bitmap_subtitle (rect, period);
+                       decode_bitmap_subtitle (rect, from);
                        break;
                case SUBTITLE_TEXT:
                        cout << "XXX: SUBTITLE_TEXT " << rect->text << "\n";
                        break;
                case SUBTITLE_ASS:
-                       decode_ass_subtitle (rect->ass, period);
+                       decode_ass_subtitle (rect->ass, from);
                        break;
                }
        }
@@ -514,20 +528,8 @@ FFmpegDecoder::decode_subtitle_packet ()
        avsubtitle_free (&sub);
 }
 
-list<ContentTimePeriod>
-FFmpegDecoder::image_subtitles_during (ContentTimePeriod p, bool starting) const
-{
-       return _ffmpeg_content->image_subtitles_during (p, starting);
-}
-
-list<ContentTimePeriod>
-FFmpegDecoder::text_subtitles_during (ContentTimePeriod p, bool starting) const
-{
-       return _ffmpeg_content->text_subtitles_during (p, starting);
-}
-
 void
-FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTimePeriod period)
+FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime from)
 {
        /* Note RGBA is expressed little-endian, so the first byte in the word is R, second
           G, third B, fourth A.
@@ -587,19 +589,20 @@ FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTimeP
                out_p += image->stride()[0] / sizeof (uint32_t);
        }
 
-       dcp::Size const vs = _ffmpeg_content->video->size ();
+       int const target_width = subtitle_codec_context()->width;
+       int const target_height = subtitle_codec_context()->height;
        dcpomatic::Rect<double> const scaled_rect (
-               static_cast<double> (rect->x) / vs.width,
-               static_cast<double> (rect->y) / vs.height,
-               static_cast<double> (rect->w) / vs.width,
-               static_cast<double> (rect->h) / vs.height
+               static_cast<double> (rect->x) / target_width,
+               static_cast<double> (rect->y) / target_height,
+               static_cast<double> (rect->w) / target_width,
+               static_cast<double> (rect->h) / target_height
                );
 
-       subtitle->give_image (period, image, scaled_rect);
+       subtitle->emit_image_start (from, image, scaled_rect);
 }
 
 void
-FFmpegDecoder::decode_ass_subtitle (string ass, ContentTimePeriod period)
+FFmpegDecoder::decode_ass_subtitle (string ass, ContentTime from)
 {
        /* We have no styles and no Format: line, so I'm assuming that FFmpeg
           produces a single format of Dialogue: lines...
@@ -615,6 +618,6 @@ FFmpegDecoder::decode_ass_subtitle (string ass, ContentTimePeriod period)
        list<sub::RawSubtitle> raw = sub::SSAReader::parse_line (base, bits[9]);
 
        BOOST_FOREACH (sub::Subtitle const & i, sub::collect<list<sub::Subtitle> > (raw)) {
-               subtitle->give_text (period, i);
+               subtitle->emit_text_start (from, i);
        }
 }