Use a separate AVFrame for each stream when decoding.
[dcpomatic.git] / src / lib / ffmpeg_examiner.cc
index e93a8331269d72ad1b9508cb90da26c0faf327d4..5e53f0974cd61865d8c09287c4b806590ca65d2f 100644 (file)
@@ -1,5 +1,5 @@
 /*
-    Copyright (C) 2013-2015 Carl Hetherington <cth@carlh.net>
+    Copyright (C) 2013-2021 Carl Hetherington <cth@carlh.net>
 
     This file is part of DCP-o-matic.
 
@@ -18,6 +18,7 @@
 
 */
 
+
 #include "dcpomatic_log.h"
 #include "ffmpeg_examiner.h"
 #include "ffmpeg_content.h"
@@ -32,19 +33,22 @@ extern "C" {
 #include <libavformat/avformat.h>
 #include <libavutil/pixfmt.h>
 #include <libavutil/pixdesc.h>
-#include <libavutil/eval.h>
+#include <libavutil/channel_layout.h>
 #include <libavutil/display.h>
+#include <libavutil/eval.h>
 }
 DCPOMATIC_ENABLE_WARNINGS
 #include <iostream>
 
 #include "i18n.h"
 
-using std::string;
+
 using std::cout;
+using std::make_shared;
 using std::max;
-using std::vector;
 using std::shared_ptr;
+using std::string;
+using std::vector;
 using boost::optional;
 using namespace dcpomatic;
 
@@ -65,37 +69,34 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Jo
        /* Find audio and subtitle streams */
 
        for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
-               AVStream* s = _format_context->streams[i];
-DCPOMATIC_DISABLE_WARNINGS
-               if (s->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
+               auto s = _format_context->streams[i];
+               auto codec = _codec_context[i] ? _codec_context[i]->codec : nullptr;
+               if (s->codecpar->codec_type == AVMEDIA_TYPE_AUDIO && codec) {
 
                        /* This is a hack; sometimes it seems that _audio_codec_context->channel_layout isn't set up,
                           so bodge it here.  No idea why we should have to do this.
                        */
 
-                       if (s->codec->channel_layout == 0) {
-                               s->codec->channel_layout = av_get_default_channel_layout (s->codec->channels);
+                       if (s->codecpar->channel_layout == 0) {
+                               s->codecpar->channel_layout = av_get_default_channel_layout (s->codecpar->channels);
                        }
 
                        DCPOMATIC_ASSERT (_format_context->duration != AV_NOPTS_VALUE);
-                       DCPOMATIC_ASSERT (s->codec->codec);
-                       DCPOMATIC_ASSERT (s->codec->codec->name);
+                       DCPOMATIC_ASSERT (codec->name);
 
                        _audio_streams.push_back (
-                               shared_ptr<FFmpegAudioStream> (
-                                       new FFmpegAudioStream (
-                                               stream_name (s),
-                                               s->codec->codec->name,
-                                               s->id,
-                                               s->codec->sample_rate,
-                                               llrint ((double (_format_context->duration) / AV_TIME_BASE) * s->codec->sample_rate),
-                                               s->codec->channels
-                                               )
+                               make_shared<FFmpegAudioStream>(
+                                       stream_name (s),
+                                       codec->name,
+                                       s->id,
+                                       s->codecpar->sample_rate,
+                                       llrint ((double(_format_context->duration) / AV_TIME_BASE) * s->codecpar->sample_rate),
+                                       s->codecpar->channels
                                        )
                                );
 
-               } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) {
-                       _subtitle_streams.push_back (shared_ptr<FFmpegSubtitleStream> (new FFmpegSubtitleStream (subtitle_stream_name (s), s->id)));
+               } else if (s->codecpar->codec_type == AVMEDIA_TYPE_SUBTITLE) {
+                       _subtitle_streams.push_back (make_shared<FFmpegSubtitleStream>(subtitle_stream_name (s), s->id));
                }
        }
 
@@ -118,16 +119,17 @@ DCPOMATIC_DISABLE_WARNINGS
         */
 
        int64_t const len = _file_group.length ();
-       auto context = _format_context->streams[_packet.stream_index]->codec;
-DCPOMATIC_ENABLE_WARNINGS
        /* A string which we build up to describe the top-field-first and repeat-first-frame values for the first few frames.
         * It would be nicer to use something like vector<bool> here but we want to search the array for a pattern later,
         * and a string seems a reasonably neat way to do that.
         */
        string temporal_reference;
        while (true) {
-               int r = av_read_frame (_format_context, &_packet);
+               auto packet = av_packet_alloc ();
+               DCPOMATIC_ASSERT (packet);
+               int r = av_read_frame (_format_context, packet);
                if (r < 0) {
+                       av_packet_free (&packet);
                        break;
                }
 
@@ -139,22 +141,24 @@ DCPOMATIC_ENABLE_WARNINGS
                        }
                }
 
-               if (_video_stream && _packet.stream_index == _video_stream.get()) {
-                       video_packet (context, temporal_reference);
+               auto context = _codec_context[packet->stream_index];
+
+               if (_video_stream && packet->stream_index == _video_stream.get()) {
+                       video_packet (context, temporal_reference, packet);
                }
 
                bool got_all_audio = true;
 
                for (size_t i = 0; i < _audio_streams.size(); ++i) {
-                       if (_audio_streams[i]->uses_index (_format_context, _packet.stream_index)) {
-                               audio_packet (context, _audio_streams[i]);
+                       if (_audio_streams[i]->uses_index(_format_context, packet->stream_index)) {
+                               audio_packet (context, _audio_streams[i], packet);
                        }
                        if (!_audio_streams[i]->first_audio) {
                                got_all_audio = false;
                        }
                }
 
-               av_packet_unref (&_packet);
+               av_packet_free (&packet);
 
                if (_first_video && got_all_audio && temporal_reference.size() >= (PULLDOWN_CHECK_FRAMES * 2)) {
                        /* All done */
@@ -162,20 +166,20 @@ DCPOMATIC_ENABLE_WARNINGS
                }
        }
 
-       _packet.data = nullptr;
-       _packet.size = 0;
-       while (video_packet(context, temporal_reference)) {}
-       /* XXX: I'm not sure this makes any sense: how does _packet.stream_index get the right value here? */
-       for (size_t i = 0; i < _audio_streams.size(); ++i) {
-               if (_audio_streams[i]->uses_index (_format_context, _packet.stream_index)) {
-                       audio_packet (context, _audio_streams[i]);
-               }
+       if (_video_stream) {
+               auto context = _codec_context[_video_stream.get()];
+               while (video_packet(context, temporal_reference, nullptr)) {}
+       }
+
+       for (auto i: _audio_streams) {
+               auto context = _codec_context[i->index(_format_context)];
+               audio_packet(context, i, nullptr);
        }
 
        if (_video_stream) {
                /* This code taken from get_rotation() in ffmpeg:cmdutils.c */
-               AVStream* stream = _format_context->streams[*_video_stream];
-               AVDictionaryEntry* rotate_tag = av_dict_get (stream->metadata, "rotate", 0, 0);
+               auto stream = _format_context->streams[*_video_stream];
+               auto rotate_tag = av_dict_get (stream->metadata, "rotate", 0, 0);
                uint8_t* displaymatrix = av_stream_get_side_data (stream, AV_PKT_DATA_DISPLAYMATRIX, 0);
                _rotation = 0;
 
@@ -209,7 +213,7 @@ DCPOMATIC_ENABLE_WARNINGS
  *  @return true if some video was decoded, otherwise false.
  */
 bool
-FFmpegExaminer::video_packet (AVCodecContext* context, string& temporal_reference)
+FFmpegExaminer::video_packet (AVCodecContext* context, string& temporal_reference, AVPacket* packet)
 {
        DCPOMATIC_ASSERT (_video_stream);
 
@@ -217,24 +221,35 @@ FFmpegExaminer::video_packet (AVCodecContext* context, string& temporal_referenc
                return false;
        }
 
-       int frame_finished;
-DCPOMATIC_DISABLE_WARNINGS
-       if (avcodec_decode_video2 (context, _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
+       int r = avcodec_send_packet (context, packet);
+       if (r < 0 && !(r == AVERROR_EOF && !packet)) {
+               /* We could cope with AVERROR(EAGAIN) and re-send the packet but I think it should never happen.
+                * AVERROR_EOF can happen during flush if we've already sent a flush packet.
+                */
+               throw DecodeError (N_("avcodec_send_packet"), N_("FFmpegExaminer::video_packet"), r);
+       }
+
+       r = avcodec_receive_frame (context, _video_frame);
+       if (r == AVERROR(EAGAIN)) {
+               /* More input is required */
+               return true;
+       } else if (r == AVERROR_EOF) {
+               /* No more output is coming */
                return false;
        }
-DCPOMATIC_ENABLE_WARNINGS
 
        if (!_first_video) {
-               _first_video = frame_time (_format_context->streams[_video_stream.get()]);
+               _first_video = frame_time (_video_frame, _format_context->streams[_video_stream.get()]);
        }
        if (_need_video_length) {
                _video_length = frame_time (
+                       _video_frame,
                        _format_context->streams[_video_stream.get()]
                        ).get_value_or (ContentTime ()).frames_round (video_frame_rate().get ());
        }
        if (temporal_reference.size() < (PULLDOWN_CHECK_FRAMES * 2)) {
-               temporal_reference += (_frame->top_field_first ? "T" : "B");
-               temporal_reference += (_frame->repeat_pict ? "3" : "2");
+               temporal_reference += (_video_frame->top_field_first ? "T" : "B");
+               temporal_reference += (_video_frame->repeat_pict ? "3" : "2");
        }
 
        return true;
@@ -242,31 +257,39 @@ DCPOMATIC_ENABLE_WARNINGS
 
 
 void
-FFmpegExaminer::audio_packet (AVCodecContext* context, shared_ptr<FFmpegAudioStream> stream)
+FFmpegExaminer::audio_packet (AVCodecContext* context, shared_ptr<FFmpegAudioStream> stream, AVPacket* packet)
 {
        if (stream->first_audio) {
                return;
        }
 
-       int frame_finished;
-DCPOMATIC_DISABLE_WARNINGS
-       if (avcodec_decode_audio4 (context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
-DCPOMATIC_ENABLE_WARNINGS
-               stream->first_audio = frame_time (stream->stream (_format_context));
+       int r = avcodec_send_packet (context, packet);
+       if (r < 0 && !(r == AVERROR_EOF && !packet) && r != AVERROR(EAGAIN)) {
+               /* We could cope with AVERROR(EAGAIN) and re-send the packet but I think it should never happen.
+                * AVERROR_EOF can happen during flush if we've already sent a flush packet.
+                * EAGAIN means we need to do avcodec_receive_frame, so just carry on and do that.
+                */
+               throw DecodeError (N_("avcodec_send_packet"), N_("FFmpegExaminer::audio_packet"), r);
        }
+
+       auto frame = audio_frame (stream);
+
+       if (avcodec_receive_frame (context, frame) < 0) {
+               return;
+       }
+
+       stream->first_audio = frame_time (frame, stream->stream(_format_context));
 }
 
 
 optional<ContentTime>
-FFmpegExaminer::frame_time (AVStream* s) const
+FFmpegExaminer::frame_time (AVFrame* frame, AVStream* stream) const
 {
        optional<ContentTime> t;
 
-DCPOMATIC_DISABLE_WARNINGS
-       int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
-DCPOMATIC_ENABLE_WARNINGS
+       int64_t const bet = frame->best_effort_timestamp;
        if (bet != AV_NOPTS_VALUE) {
-               t = ContentTime::from_seconds (bet * av_q2d (s->time_base));
+               t = ContentTime::from_seconds (bet * av_q2d(stream->time_base));
        }
 
        return t;
@@ -280,12 +303,14 @@ FFmpegExaminer::video_frame_rate () const
        return av_q2d(av_guess_frame_rate(_format_context, _format_context->streams[_video_stream.get()], 0));
 }
 
+
 dcp::Size
 FFmpegExaminer::video_size () const
 {
        return dcp::Size (video_codec_context()->width, video_codec_context()->height);
 }
 
+
 /** @return Length according to our content's header */
 Frame
 FFmpegExaminer::video_length () const
@@ -293,22 +318,24 @@ FFmpegExaminer::video_length () const
        return max (Frame (1), _video_length);
 }
 
+
 optional<double>
 FFmpegExaminer::sample_aspect_ratio () const
 {
        DCPOMATIC_ASSERT (_video_stream);
-       AVRational sar = av_guess_sample_aspect_ratio (_format_context, _format_context->streams[_video_stream.get()], 0);
+       auto sar = av_guess_sample_aspect_ratio (_format_context, _format_context->streams[_video_stream.get()], 0);
        if (sar.num == 0) {
                /* I assume this means that we don't know */
-               return optional<double> ();
+               return {};
        }
        return double (sar.num) / sar.den;
 }
 
+
 string
 FFmpegExaminer::subtitle_stream_name (AVStream* s) const
 {
-       string n = stream_name (s);
+       auto n = stream_name (s);
 
        if (n.empty()) {
                n = _("unknown");
@@ -317,18 +344,19 @@ FFmpegExaminer::subtitle_stream_name (AVStream* s) const
        return n;
 }
 
+
 string
 FFmpegExaminer::stream_name (AVStream* s) const
 {
        string n;
 
        if (s->metadata) {
-               AVDictionaryEntry const * lang = av_dict_get (s->metadata, "language", 0, 0);
+               auto const lang = av_dict_get (s->metadata, "language", 0, 0);
                if (lang) {
                        n = lang->value;
                }
 
-               AVDictionaryEntry const * title = av_dict_get (s->metadata, "title", 0, 0);
+               auto const title = av_dict_get (s->metadata, "title", 0, 0);
                if (title) {
                        if (!n.empty()) {
                                n += " ";
@@ -340,18 +368,20 @@ FFmpegExaminer::stream_name (AVStream* s) const
        return n;
 }
 
+
 optional<int>
 FFmpegExaminer::bits_per_pixel () const
 {
        if (video_codec_context()->pix_fmt == -1) {
-               return optional<int>();
+               return {};
        }
 
-       AVPixFmtDescriptor const * d = av_pix_fmt_desc_get (video_codec_context()->pix_fmt);
+       auto const d = av_pix_fmt_desc_get (video_codec_context()->pix_fmt);
        DCPOMATIC_ASSERT (d);
        return av_get_bits_per_pixel (d);
 }
 
+
 bool
 FFmpegExaminer::yuv () const
 {
@@ -433,12 +463,14 @@ FFmpegExaminer::yuv () const
        }
 }
 
+
 bool
 FFmpegExaminer::has_video () const
 {
-       return static_cast<bool> (_video_stream);
+       return static_cast<bool>(_video_stream);
 }
 
+
 VideoRange
 FFmpegExaminer::range () const
 {
@@ -451,3 +483,12 @@ FFmpegExaminer::range () const
                return VideoRange::FULL;
        }
 }
+
+
+PixelQuanta
+FFmpegExaminer::pixel_quanta () const
+{
+       auto const desc = av_pix_fmt_desc_get(video_codec_context()->pix_fmt);
+       return { 1 << desc->log2_chroma_w, 1 << desc->log2_chroma_h };
+}
+