]> git.sesse.net Git - nageru/blobdiff - ffmpeg_capture.cpp
Reset audio resampler when FFmpeg inputs restart due to errors.
[nageru] / ffmpeg_capture.cpp
index 9a151b60d044c9eb038cdcc7a8c74b5a98e7bbd5..9ae35042d1ce998701e34b9072ca26b2700d047f 100644 (file)
@@ -211,6 +211,8 @@ FFmpegCapture::FFmpegCapture(const string &filename, unsigned width, unsigned he
        // Not really used for anything.
        description = "Video: " + filename;
 
+       last_frame = steady_clock::now();
+
        avformat_network_init();  // In case someone wants this.
 }
 
@@ -288,6 +290,7 @@ void FFmpegCapture::producer_thread_func()
                        producer_thread_should_quit.sleep_for(seconds(1));
                        continue;
                }
+               should_interrupt = false;
                if (!play_video(pathname)) {
                        // Error.
                        fprintf(stderr, "Error when playing %s, sleeping one second and trying again...\n", pathname.c_str());
@@ -313,17 +316,31 @@ void FFmpegCapture::send_disconnected_frame()
                VideoFormat video_format;
                video_format.width = width;
                video_format.height = height;
-               video_format.stride = width * 4;
                video_format.frame_rate_nom = 60;
                video_format.frame_rate_den = 1;
                video_format.is_connected = false;
-
-               video_frame.len = width * height * 4;
+               if (pixel_format == bmusb::PixelFormat_8BitBGRA) {
+                       video_format.stride = width * 4;
+                       video_frame.len = width * height * 4;
+               } else {
+                       video_format.stride = width;
+                       current_frame_ycbcr_format.luma_coefficients = YCBCR_REC_709;
+                       current_frame_ycbcr_format.full_range = true;
+                       current_frame_ycbcr_format.num_levels = 256;
+                       current_frame_ycbcr_format.chroma_subsampling_x = 2;
+                       current_frame_ycbcr_format.chroma_subsampling_y = 2;
+                       current_frame_ycbcr_format.cb_x_position = 0.0f;
+                       current_frame_ycbcr_format.cb_y_position = 0.0f;
+                       current_frame_ycbcr_format.cr_x_position = 0.0f;
+                       current_frame_ycbcr_format.cr_y_position = 0.0f;
+                       video_frame.len = width * height * 2;
+               }
                memset(video_frame.data, 0, video_frame.len);
 
                frame_callback(-1, AVRational{1, TIMEBASE}, -1, AVRational{1, TIMEBASE}, timecode++,
                        video_frame, /*video_offset=*/0, video_format,
                        FrameAllocator::Frame(), /*audio_offset=*/0, AudioFormat());
+               last_frame_was_connected = false;
        }
 }
 
@@ -341,7 +358,10 @@ bool FFmpegCapture::play_video(const string &pathname)
                last_modified = buf.st_mtim;
        }
 
-       auto format_ctx = avformat_open_input_unique(pathname.c_str(), nullptr, nullptr);
+       AVDictionary *opts = nullptr;
+       av_dict_set(&opts, "fflags", "nobuffer", 0);
+
+       auto format_ctx = avformat_open_input_unique(pathname.c_str(), nullptr, &opts, AVIOInterruptCB{ &FFmpegCapture::interrupt_cb_thunk, this });
        if (format_ctx == nullptr) {
                fprintf(stderr, "%s: Error opening file\n", pathname.c_str());
                return false;
@@ -381,8 +401,9 @@ bool FFmpegCapture::play_video(const string &pathname)
                video_codec_ctx.get(), avcodec_close);
 
        // Open audio decoder, if we have audio.
-       AVCodecContextWithDeleter audio_codec_ctx = avcodec_alloc_context3_unique(nullptr);
+       AVCodecContextWithDeleter audio_codec_ctx;
        if (audio_stream_index != -1) {
+               audio_codec_ctx = avcodec_alloc_context3_unique(nullptr);
                const AVCodecParameters *audio_codecpar = format_ctx->streams[audio_stream_index]->codecpar;
                audio_timebase = format_ctx->streams[audio_stream_index]->time_base;
                if (avcodec_parameters_to_context(audio_codec_ctx.get(), audio_codecpar) < 0) {
@@ -405,6 +426,7 @@ bool FFmpegCapture::play_video(const string &pathname)
        internal_rewind();
 
        // Main loop.
+       bool first_frame = true;
        while (!producer_thread_should_quit.should_quit()) {
                if (process_queued_commands(format_ctx.get(), pathname, last_modified, /*rewound=*/nullptr)) {
                        return true;
@@ -453,15 +475,34 @@ bool FFmpegCapture::play_video(const string &pathname)
                                pts_origin = frame->pts;        
                        }
                        next_frame_start = compute_frame_start(frame->pts, pts_origin, video_timebase, start, rate);
+                       if (first_frame && last_frame_was_connected) {
+                               // If reconnect took more than one second, this is probably a live feed,
+                               // and we should reset the resampler. (Or the rate is really, really low,
+                               // in which case a reset on the first frame is fine anyway.)
+                               if (duration<double>(next_frame_start - last_frame).count() >= 1.0) {
+                                       last_frame_was_connected = false;
+                               }
+                       }
                        video_frame->received_timestamp = next_frame_start;
+                       audio_frame->received_timestamp = next_frame_start;
                        bool finished_wakeup = producer_thread_should_quit.sleep_until(next_frame_start);
                        if (finished_wakeup) {
                                if (audio_frame->len > 0) {
                                        assert(audio_pts != -1);
                                }
+                               if (!last_frame_was_connected) {
+                                       // We're recovering from an error (or really slow load, see above).
+                                       // Make sure to get the audio resampler reset. (This is a hack;
+                                       // ideally, the frame callback should just accept a way to signal
+                                       // audio discontinuity.)
+                                       timecode += MAX_FPS * 2 + 1;
+                               }
                                frame_callback(frame->pts, video_timebase, audio_pts, audio_timebase, timecode++,
                                        video_frame.get_and_release(), 0, video_format,
                                        audio_frame.get_and_release(), 0, audio_format);
+                               first_frame = false;
+                               last_frame = steady_clock::now();
+                               last_frame_was_connected = true;
                                break;
                        } else {
                                if (producer_thread_should_quit.should_quit()) break;
@@ -645,10 +686,15 @@ void FFmpegCapture::convert_audio(const AVFrame *audio_avframe, FrameAllocator::
        }
        audio_format->num_channels = 2;
 
+       int64_t channel_layout = audio_avframe->channel_layout;
+       if (channel_layout == 0) {
+               channel_layout = av_get_default_channel_layout(audio_avframe->channels);
+       }
+
        if (resampler == nullptr ||
            audio_avframe->format != last_src_format ||
            dst_format != last_dst_format ||
-           av_frame_get_channel_layout(audio_avframe) != last_channel_layout ||
+           channel_layout != last_channel_layout ||
            av_frame_get_sample_rate(audio_avframe) != last_sample_rate) {
                avresample_free(&resampler);
                resampler = avresample_alloc_context();
@@ -657,7 +703,7 @@ void FFmpegCapture::convert_audio(const AVFrame *audio_avframe, FrameAllocator::
                        exit(1);
                }
 
-               av_opt_set_int(resampler, "in_channel_layout",  av_frame_get_channel_layout(audio_avframe), 0);
+               av_opt_set_int(resampler, "in_channel_layout",  channel_layout,                             0);
                av_opt_set_int(resampler, "out_channel_layout", AV_CH_LAYOUT_STEREO,                        0);
                av_opt_set_int(resampler, "in_sample_rate",     av_frame_get_sample_rate(audio_avframe),    0);
                av_opt_set_int(resampler, "out_sample_rate",    OUTPUT_FREQUENCY,                           0);
@@ -671,7 +717,7 @@ void FFmpegCapture::convert_audio(const AVFrame *audio_avframe, FrameAllocator::
 
                last_src_format = AVSampleFormat(audio_avframe->format);
                last_dst_format = dst_format;
-               last_channel_layout = av_frame_get_channel_layout(audio_avframe);
+               last_channel_layout = channel_layout;
                last_sample_rate = av_frame_get_sample_rate(audio_avframe);
        }
 
@@ -680,7 +726,7 @@ void FFmpegCapture::convert_audio(const AVFrame *audio_avframe, FrameAllocator::
 
        uint8_t *data = audio_frame->data + audio_frame->len;
        int out_samples = avresample_convert(resampler, &data, 0, num_samples_room,
-               audio_avframe->data, audio_avframe->linesize[0], audio_avframe->nb_samples);
+               const_cast<uint8_t **>(audio_avframe->data), audio_avframe->linesize[0], audio_avframe->nb_samples);
        if (out_samples < 0) {
                 fprintf(stderr, "Audio conversion failed.\n");
                 exit(1);
@@ -783,3 +829,13 @@ UniqueFrame FFmpegCapture::make_video_frame(const AVFrame *frame, const string &
 
        return video_frame;
 }
+
+int FFmpegCapture::interrupt_cb_thunk(void *unique)
+{
+       return reinterpret_cast<FFmpegCapture *>(unique)->interrupt_cb();
+}
+
+int FFmpegCapture::interrupt_cb()
+{
+       return should_interrupt.load();
+}