]> git.sesse.net Git - nageru/blobdiff - ffmpeg_capture.cpp
Update the queue length metric after trimming, not before.
[nageru] / ffmpeg_capture.cpp
index b5768770dae37201ba5dd7064901f6059196dfe5..fb1b223c326f4e789193da4e078c65a48d729ec5 100644 (file)
@@ -200,7 +200,7 @@ YCbCrFormat decode_ycbcr_format(const AVPixFmtDescriptor *desc, const AVFrame *f
 }  // namespace
 
 FFmpegCapture::FFmpegCapture(const string &filename, unsigned width, unsigned height)
-       : filename(filename), width(width), height(height)
+       : filename(filename), width(width), height(height), video_timebase{1, 1}
 {
        // Not really used for anything.
        description = "Video: " + filename;
@@ -351,7 +351,7 @@ bool FFmpegCapture::play_video(const string &pathname)
        }
 
        const AVCodecParameters *codecpar = format_ctx->streams[video_stream_index]->codecpar;
-       AVRational video_timebase = format_ctx->streams[video_stream_index]->time_base;
+       video_timebase = format_ctx->streams[video_stream_index]->time_base;
        AVCodecContextWithDeleter codec_ctx = avcodec_alloc_context3_unique(nullptr);
        if (avcodec_parameters_to_context(codec_ctx.get(), codecpar) < 0) {
                fprintf(stderr, "%s: Cannot fill codec parameters\n", pathname.c_str());
@@ -371,51 +371,18 @@ bool FFmpegCapture::play_video(const string &pathname)
 
        internal_rewind();
 
-       unique_ptr<SwsContext, decltype(sws_freeContext)*> sws_ctx(nullptr, sws_freeContext);
-       int sws_last_width = -1, sws_last_height = -1, sws_last_src_format = -1;
-       AVPixelFormat sws_dst_format = AVPixelFormat(-1);  // In practice, always initialized.
-
        // Main loop.
        while (!producer_thread_should_quit.should_quit()) {
-               if (process_queued_commands(format_ctx.get(), pathname, last_modified)) {
+               if (process_queued_commands(format_ctx.get(), pathname, last_modified, /*rewound=*/nullptr)) {
                        return true;
                }
 
-               // Read packets until we have a frame or there are none left.
-               int frame_finished = 0;
-               AVFrameWithDeleter frame = av_frame_alloc_unique();
-               bool eof = false;
-               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.get(), &pkt) == 0) {
-                               if (pkt.stream_index != video_stream_index) {
-                                       // Ignore audio for now.
-                                       continue;
-                               }
-                               if (avcodec_send_packet(codec_ctx.get(), &pkt) < 0) {
-                                       fprintf(stderr, "%s: Cannot send packet to codec.\n", pathname.c_str());
-                                       return false;
-                               }
-                       } else {
-                               eof = true;  // Or error, but ignore that for the time being.
-                       }
-
-                       int err = avcodec_receive_frame(codec_ctx.get(), frame.get());
-                       if (err == 0) {
-                               frame_finished = true;
-                               break;
-                       } else if (err != AVERROR(EAGAIN)) {
-                               fprintf(stderr, "%s: Cannot receive frame from codec.\n", pathname.c_str());
-                               return false;
-                       }
-               } while (!eof);
-
-               if (!frame_finished) {
+               bool error;
+               AVFrameWithDeleter frame = decode_frame(format_ctx.get(), codec_ctx.get(), pathname, video_stream_index, &error);
+               if (error) {
+                       return false;
+               }
+               if (frame == nullptr) {
                        // EOF. Loop back to the start if we can.
                        if (av_seek_frame(format_ctx.get(), /*stream_index=*/-1, /*timestamp=*/0, /*flags=*/0) < 0) {
                                fprintf(stderr, "%s: Rewind failed, not looping.\n", pathname.c_str());
@@ -432,87 +399,50 @@ bool FFmpegCapture::play_video(const string &pathname)
                        continue;
                }
 
-               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());
+               VideoFormat video_format = construct_video_format(frame.get(), video_timebase);
+               FrameAllocator::Frame video_frame = make_video_frame(frame.get(), pathname, &error);
+               if (error) {
                        return false;
                }
 
-               VideoFormat video_format;
-               video_format.width = width;
-               video_format.height = height;
-               if (pixel_format == bmusb::PixelFormat_8BitBGRA) {
-                       video_format.stride = width * 4;
-               } 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.get()) * 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;
-
-               next_frame_start = compute_frame_start(frame->pts, pts_origin, video_timebase, start, rate);
-               last_pts = frame->pts;
-
-               FrameAllocator::Frame video_frame = video_frame_allocator->alloc_frame();
-               if (video_frame.data != nullptr) {
-                       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] = video_format.stride;
-                               video_frame.len = video_format.stride * height;
-                       } 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.get());
-                       }
-                       sws_scale(sws_ctx.get(), frame->data, frame->linesize, 0, frame->height, pic_data, linesizes);
-                       video_frame.received_timestamp = next_frame_start;
-               }
-
                FrameAllocator::Frame audio_frame;
                AudioFormat audio_format;
                 audio_format.bits_per_sample = 32;
                 audio_format.num_channels = 8;
 
-               producer_thread_should_quit.sleep_until(next_frame_start);
-               frame_callback(timecode++,
-                       video_frame, 0, video_format,
-                       audio_frame, 0, audio_format);
+               for ( ;; ) {
+                       next_frame_start = compute_frame_start(frame->pts, pts_origin, video_timebase, start, rate);
+                       video_frame.received_timestamp = next_frame_start;
+                       bool finished_wakeup = producer_thread_should_quit.sleep_until(next_frame_start);
+                       if (finished_wakeup) {
+                               frame_callback(timecode++,
+                                       video_frame, 0, video_format,
+                                       audio_frame, 0, audio_format);
+                               break;
+                       } else {
+                               if (producer_thread_should_quit.should_quit()) break;
+
+                               bool rewound = false;
+                               if (process_queued_commands(format_ctx.get(), pathname, last_modified, &rewound)) {
+                                       return true;
+                               }
+                               // If we just rewound, drop this frame on the floor and be done.
+                               if (rewound) {
+                                       video_frame_allocator->release_frame(video_frame);
+                                       break;
+                               }
+                               // OK, we didn't, so probably a rate change. Recalculate next_frame_start,
+                               // but if it's now in the past, we'll reset the origin, so that we don't
+                               // generate a huge backlog of frames that we need to run through quickly.
+                               next_frame_start = compute_frame_start(frame->pts, pts_origin, video_timebase, start, rate);
+                               steady_clock::time_point now = steady_clock::now();
+                               if (next_frame_start < now) {
+                                       pts_origin = frame->pts;
+                                       start = next_frame_start = now;
+                               }
+                       }
+               }
+               last_pts = frame->pts;
        }
        return true;
 }
@@ -523,7 +453,7 @@ void FFmpegCapture::internal_rewind()
        start = next_frame_start = steady_clock::now();
 }
 
-bool FFmpegCapture::process_queued_commands(AVFormatContext *format_ctx, const std::string &pathname, timespec last_modified)
+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;
@@ -545,10 +475,14 @@ bool FFmpegCapture::process_queued_commands(AVFormatContext *format_ctx, const s
                                return true;
                        }
                        internal_rewind();
+                       if (rewound != nullptr) {
+                               *rewound = true;
+                       }
                        break;
 
                case QueuedCommand::CHANGE_RATE:
-                       start = next_frame_start;
+                       // 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;
@@ -556,3 +490,131 @@ bool FFmpegCapture::process_queued_commands(AVFormatContext *format_ctx, const s
        }
        return false;
 }
+
+AVFrameWithDeleter FFmpegCapture::decode_frame(AVFormatContext *format_ctx, AVCodecContext *codec_ctx, const std::string &pathname, int video_stream_index, bool *error)
+{
+       *error = false;
+
+       // Read packets until we have a frame or there are none left.
+       bool frame_finished = false;
+       AVFrameWithDeleter frame = av_frame_alloc_unique();
+       bool eof = false;
+       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 != video_stream_index) {
+                               // Ignore audio for now.
+                               continue;
+                       }
+                       if (avcodec_send_packet(codec_ctx, &pkt) < 0) {
+                               fprintf(stderr, "%s: Cannot send packet to codec.\n", pathname.c_str());
+                               *error = true;
+                               return AVFrameWithDeleter(nullptr);
+                       }
+               } else {
+                       eof = true;  // Or error, but ignore that for the time being.
+               }
+
+               int err = avcodec_receive_frame(codec_ctx, frame.get());
+               if (err == 0) {
+                       frame_finished = true;
+                       break;
+               } else if (err != AVERROR(EAGAIN)) {
+                       fprintf(stderr, "%s: Cannot receive frame from codec.\n", pathname.c_str());
+                       *error = true;
+                       return AVFrameWithDeleter(nullptr);
+               }
+       } while (!eof);
+
+       if (frame_finished)
+               return frame;
+       else
+               return AVFrameWithDeleter(nullptr);
+}
+
+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 {
+               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;
+}
+
+FrameAllocator::Frame FFmpegCapture::make_video_frame(const AVFrame *frame, const string &pathname, bool *error)
+{
+       *error = false;
+
+       FrameAllocator::Frame 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 {
+               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;
+}