+
+bool FFmpegCapture::process_queued_commands(AVFormatContext *format_ctx, const std::string &pathname, timespec last_modified, bool *rewound)
+{
+ // Process any queued commands from other threads.
+ vector<QueuedCommand> commands;
+ {
+ lock_guard<mutex> lock(queue_mu);
+ swap(commands, command_queue);
+ }
+ for (const QueuedCommand &cmd : commands) {
+ switch (cmd.command) {
+ case QueuedCommand::REWIND:
+ if (av_seek_frame(format_ctx, /*stream_index=*/-1, /*timestamp=*/0, /*flags=*/0) < 0) {
+ fprintf(stderr, "%s: Rewind failed, stopping play.\n", pathname.c_str());
+ }
+ // If the file has changed since last time, return to get it reloaded.
+ // Note that depending on how you move the file into place, you might
+ // end up corrupting the one you're already playing, so this path
+ // might not trigger.
+ if (changed_since(pathname, last_modified)) {
+ return true;
+ }
+ internal_rewind();
+ if (rewound != nullptr) {
+ *rewound = true;
+ }
+ break;
+
+ case QueuedCommand::CHANGE_RATE:
+ // Change the origin to the last played frame.
+ start = compute_frame_start(last_pts, pts_origin, video_timebase, start, rate);
+ pts_origin = last_pts;
+ rate = cmd.new_rate;
+ break;
+ }
+ }
+ return false;
+}
+
+namespace {
+
+} // namespace
+
+AVFrameWithDeleter FFmpegCapture::decode_frame(AVFormatContext *format_ctx, AVCodecContext *video_codec_ctx, AVCodecContext *audio_codec_ctx,
+ const std::string &pathname, int video_stream_index, int audio_stream_index,
+ FrameAllocator::Frame *audio_frame, AudioFormat *audio_format, int64_t *audio_pts, bool *error)
+{
+ *error = false;
+
+ // Read packets until we have a frame or there are none left.
+ bool frame_finished = false;
+ AVFrameWithDeleter audio_avframe = av_frame_alloc_unique();
+ AVFrameWithDeleter video_avframe = av_frame_alloc_unique();
+ bool eof = false;
+ *audio_pts = -1;
+ do {
+ AVPacket pkt;
+ unique_ptr<AVPacket, decltype(av_packet_unref)*> pkt_cleanup(
+ &pkt, av_packet_unref);
+ av_init_packet(&pkt);
+ pkt.data = nullptr;
+ pkt.size = 0;
+ if (av_read_frame(format_ctx, &pkt) == 0) {
+ if (pkt.stream_index == audio_stream_index && audio_callback != nullptr) {
+ audio_callback(&pkt, format_ctx->streams[audio_stream_index]->time_base);
+ }
+ if (pkt.stream_index == video_stream_index) {
+ if (avcodec_send_packet(video_codec_ctx, &pkt) < 0) {
+ fprintf(stderr, "%s: Cannot send packet to video codec.\n", pathname.c_str());
+ *error = true;
+ return AVFrameWithDeleter(nullptr);
+ }
+ } else if (pkt.stream_index == audio_stream_index) {
+ if (*audio_pts == -1) {
+ *audio_pts = pkt.pts;
+ }
+ if (avcodec_send_packet(audio_codec_ctx, &pkt) < 0) {
+ fprintf(stderr, "%s: Cannot send packet to audio codec.\n", pathname.c_str());
+ *error = true;
+ return AVFrameWithDeleter(nullptr);
+ }
+ }
+ } else {
+ eof = true; // Or error, but ignore that for the time being.
+ }
+
+ // Decode audio, if any.
+ if (*audio_pts != -1) {
+ for ( ;; ) {
+ int err = avcodec_receive_frame(audio_codec_ctx, audio_avframe.get());
+ if (err == 0) {
+ convert_audio(audio_avframe.get(), audio_frame, audio_format);
+ } else if (err == AVERROR(EAGAIN)) {
+ break;
+ } else {
+ fprintf(stderr, "%s: Cannot receive frame from audio codec.\n", pathname.c_str());
+ *error = true;
+ return AVFrameWithDeleter(nullptr);
+ }
+ }
+ }
+
+ // Decode video, if we have a frame.
+ int err = avcodec_receive_frame(video_codec_ctx, video_avframe.get());
+ if (err == 0) {
+ frame_finished = true;
+ break;
+ } else if (err != AVERROR(EAGAIN)) {
+ fprintf(stderr, "%s: Cannot receive frame from video codec.\n", pathname.c_str());
+ *error = true;
+ return AVFrameWithDeleter(nullptr);
+ }
+ } while (!eof);
+
+ if (frame_finished)
+ return video_avframe;
+ else
+ return AVFrameWithDeleter(nullptr);
+}
+
+void FFmpegCapture::convert_audio(const AVFrame *audio_avframe, FrameAllocator::Frame *audio_frame, AudioFormat *audio_format)
+{
+ // Decide on a format. If there already is one in this audio frame,
+ // we're pretty much forced to use it. If not, we try to find an exact match.
+ // If that still doesn't work, we default to 32-bit signed chunked
+ // (float would be nice, but there's really no way to signal that yet).
+ AVSampleFormat dst_format;
+ if (audio_format->bits_per_sample == 0) {
+ switch (audio_avframe->format) {
+ case AV_SAMPLE_FMT_S16:
+ case AV_SAMPLE_FMT_S16P:
+ audio_format->bits_per_sample = 16;
+ dst_format = AV_SAMPLE_FMT_S16;
+ break;
+ case AV_SAMPLE_FMT_S32:
+ case AV_SAMPLE_FMT_S32P:
+ default:
+ audio_format->bits_per_sample = 32;
+ dst_format = AV_SAMPLE_FMT_S32;
+ break;
+ }
+ } else if (audio_format->bits_per_sample == 16) {
+ dst_format = AV_SAMPLE_FMT_S16;
+ } else if (audio_format->bits_per_sample == 32) {
+ dst_format = AV_SAMPLE_FMT_S32;
+ } else {
+ assert(false);
+ }
+ 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 ||
+ channel_layout != last_channel_layout ||
+ av_frame_get_sample_rate(audio_avframe) != last_sample_rate) {
+ avresample_free(&resampler);
+ resampler = avresample_alloc_context();
+ if (resampler == nullptr) {
+ fprintf(stderr, "Allocating resampler failed.\n");
+ exit(1);
+ }
+
+ 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);
+ av_opt_set_int(resampler, "in_sample_fmt", audio_avframe->format, 0);
+ av_opt_set_int(resampler, "out_sample_fmt", dst_format, 0);
+
+ if (avresample_open(resampler) < 0) {
+ fprintf(stderr, "Could not open resample context.\n");
+ exit(1);
+ }
+
+ last_src_format = AVSampleFormat(audio_avframe->format);
+ last_dst_format = dst_format;
+ last_channel_layout = channel_layout;
+ last_sample_rate = av_frame_get_sample_rate(audio_avframe);
+ }
+
+ size_t bytes_per_sample = (audio_format->bits_per_sample / 8) * 2;
+ size_t num_samples_room = (audio_frame->size - audio_frame->len) / bytes_per_sample;
+
+ uint8_t *data = audio_frame->data + audio_frame->len;
+ int out_samples = avresample_convert(resampler, &data, 0, num_samples_room,
+ 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);
+ }
+
+ audio_frame->len += out_samples * bytes_per_sample;
+}
+
+VideoFormat FFmpegCapture::construct_video_format(const AVFrame *frame, AVRational video_timebase)
+{
+ VideoFormat video_format;
+ video_format.width = width;
+ video_format.height = height;
+ if (pixel_format == bmusb::PixelFormat_8BitBGRA) {
+ video_format.stride = width * 4;
+ } else if (pixel_format == FFmpegCapture::PixelFormat_NV12) {
+ video_format.stride = width;
+ } else {
+ assert(pixel_format == bmusb::PixelFormat_8BitYCbCrPlanar);
+ video_format.stride = width;
+ }
+ video_format.frame_rate_nom = video_timebase.den;
+ video_format.frame_rate_den = av_frame_get_pkt_duration(frame) * video_timebase.num;
+ if (video_format.frame_rate_nom == 0 || video_format.frame_rate_den == 0) {
+ // Invalid frame rate.
+ video_format.frame_rate_nom = 60;
+ video_format.frame_rate_den = 1;
+ }
+ video_format.has_signal = true;
+ video_format.is_connected = true;
+ return video_format;
+}
+
+UniqueFrame FFmpegCapture::make_video_frame(const AVFrame *frame, const string &pathname, bool *error)
+{
+ *error = false;
+
+ UniqueFrame video_frame(video_frame_allocator->alloc_frame());
+ if (video_frame->data == nullptr) {
+ return video_frame;
+ }
+
+ if (sws_ctx == nullptr ||
+ sws_last_width != frame->width ||
+ sws_last_height != frame->height ||
+ sws_last_src_format != frame->format) {
+ sws_dst_format = decide_dst_format(AVPixelFormat(frame->format), pixel_format);
+ sws_ctx.reset(
+ sws_getContext(frame->width, frame->height, AVPixelFormat(frame->format),
+ width, height, sws_dst_format,
+ SWS_BICUBIC, nullptr, nullptr, nullptr));
+ sws_last_width = frame->width;
+ sws_last_height = frame->height;
+ sws_last_src_format = frame->format;
+ }
+ if (sws_ctx == nullptr) {
+ fprintf(stderr, "%s: Could not create scaler context\n", pathname.c_str());
+ *error = true;
+ return video_frame;
+ }
+
+ uint8_t *pic_data[4] = { nullptr, nullptr, nullptr, nullptr };
+ int linesizes[4] = { 0, 0, 0, 0 };
+ if (pixel_format == bmusb::PixelFormat_8BitBGRA) {
+ pic_data[0] = video_frame->data;
+ linesizes[0] = width * 4;
+ video_frame->len = (width * 4) * height;
+ } else if (pixel_format == PixelFormat_NV12) {
+ pic_data[0] = video_frame->data;
+ linesizes[0] = width;
+
+ pic_data[1] = pic_data[0] + width * height;
+ linesizes[1] = width;
+
+ video_frame->len = (width * 2) * height;
+
+ const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(sws_dst_format);
+ current_frame_ycbcr_format = decode_ycbcr_format(desc, frame);
+ } else {
+ assert(pixel_format == bmusb::PixelFormat_8BitYCbCrPlanar);
+ const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(sws_dst_format);
+
+ int chroma_width = AV_CEIL_RSHIFT(int(width), desc->log2_chroma_w);
+ int chroma_height = AV_CEIL_RSHIFT(int(height), desc->log2_chroma_h);
+
+ pic_data[0] = video_frame->data;
+ linesizes[0] = width;
+
+ pic_data[1] = pic_data[0] + width * height;
+ linesizes[1] = chroma_width;
+
+ pic_data[2] = pic_data[1] + chroma_width * chroma_height;
+ linesizes[2] = chroma_width;
+
+ video_frame->len = width * height + 2 * chroma_width * chroma_height;
+
+ current_frame_ycbcr_format = decode_ycbcr_format(desc, frame);
+ }
+ sws_scale(sws_ctx.get(), frame->data, frame->linesize, 0, frame->height, pic_data, linesizes);
+
+ return video_frame;
+}