1 #ifndef _FFMPEG_CAPTURE_H
2 #define _FFMPEG_CAPTURE_H 1
4 // FFmpegCapture looks much like a capture card, but the frames it spits out
5 // come from a video in real time, looping. Because it decodes the video using
6 // FFmpeg (thus the name), this means it can handle a very wide array of video
7 // formats, and also things like network streaming and V4L capture, but it is
8 // also significantly less integrated and optimized than the regular capture
9 // cards. In particular, the frames are always scaled and converted to 8-bit
10 // RGBA on the CPU before being sent on to the GPU.
12 // Since we don't really know much about the video when building the chains,
13 // there are some limitations. In particular, frames are always assumed to be
14 // sRGB even if the video container says something else. We could probably
15 // try to load the video on startup and pick out the parameters at that point,
16 // but it would require some more plumbing, and it would also fail if the file
17 // changes parameters midway, which is allowed in some formats.
19 // You can get out the audio either as decoded or in raw form (Kaeru uses this).
21 // If there's a subtitle track, you can also get out the last subtitle at the
22 // point of the frame. Note that once we get a video frame, we don't look for
23 // subtitle, so if subtitles and a frame comes at the same time, you might not
24 // see the subtitle until the next frame.
36 #include <movit/effect.h>
37 #include <movit/ycbcr.h>
40 #include <libswresample/swresample.h>
41 #include <libavutil/channel_layout.h>
42 #include <libavutil/pixfmt.h>
43 #include <libavutil/rational.h>
44 #include <libavutil/samplefmt.h>
47 #include "bmusb/bmusb.h"
48 #include "shared/ffmpeg_raii.h"
49 #include "ref_counted_frame.h"
50 #include "quittable_sleeper.h"
52 struct AVFormatContext;
57 class FFmpegCapture : public bmusb::CaptureInterface
60 FFmpegCapture(const std::string &filename, unsigned width, unsigned height);
62 // Takes ownership of the SRT client socket.
63 FFmpegCapture(int srt_sock, const std::string &stream_id);
67 void set_card_index(int card_index)
69 this->card_index = card_index;
72 int get_card_index() const
79 std::lock_guard<std::mutex> lock(queue_mu);
80 command_queue.push_back(QueuedCommand { QueuedCommand::REWIND });
81 producer_thread_should_quit.wakeup();
84 void change_rate(double new_rate)
86 std::lock_guard<std::mutex> lock(queue_mu);
87 command_queue.push_back(QueuedCommand { QueuedCommand::CHANGE_RATE, new_rate });
88 producer_thread_should_quit.wakeup();
91 std::string get_filename() const
93 std::lock_guard<std::mutex> lock(filename_mu);
97 void change_filename(const std::string &new_filename)
99 std::lock_guard<std::mutex> lock(filename_mu);
100 filename = new_filename;
101 should_interrupt = true;
104 // Will stop the stream even if it's hung on blocking I/O.
107 should_interrupt = true;
111 void set_video_frame_allocator(bmusb::FrameAllocator *allocator) override
113 video_frame_allocator = allocator;
114 if (owned_video_frame_allocator.get() != allocator) {
115 owned_video_frame_allocator.reset();
119 bmusb::FrameAllocator *get_video_frame_allocator() override
121 return video_frame_allocator;
124 // Does not take ownership.
125 void set_audio_frame_allocator(bmusb::FrameAllocator *allocator) override
127 audio_frame_allocator = allocator;
128 if (owned_audio_frame_allocator.get() != allocator) {
129 owned_audio_frame_allocator.reset();
133 bmusb::FrameAllocator *get_audio_frame_allocator() override
135 return audio_frame_allocator;
138 // FFmpegCapture-specific overload of set_frame_callback that also gives
139 // the raw original pts from the video. Negative pts means a dummy frame.
140 typedef std::function<void(int64_t video_pts, AVRational video_timebase, int64_t audio_pts, AVRational audio_timebase,
142 bmusb::FrameAllocator::Frame video_frame, size_t video_offset, bmusb::VideoFormat video_format,
143 bmusb::FrameAllocator::Frame audio_frame, size_t audio_offset, bmusb::AudioFormat audio_format)>
145 void set_frame_callback(frame_callback_t callback)
147 frame_callback = callback;
150 void set_frame_callback(bmusb::frame_callback_t callback) override
152 frame_callback = std::bind(
154 std::placeholders::_5,
155 std::placeholders::_6,
156 std::placeholders::_7,
157 std::placeholders::_8,
158 std::placeholders::_9,
159 std::placeholders::_10,
160 std::placeholders::_11);
163 // FFmpegCapture-specific callback that gives the raw audio/video.
164 typedef std::function<void(const AVPacket *pkt, const AVRational timebase)> packet_callback_t;
165 void set_video_callback(packet_callback_t callback)
167 video_callback = callback;
169 void set_audio_callback(packet_callback_t callback)
171 audio_callback = callback;
174 // Used to get precise information about the Y'CbCr format used
175 // for a given frame. Only valid to call during the frame callback,
176 // and only when receiving a frame with pixel format PixelFormat_8BitYCbCrPlanar.
177 movit::YCbCrFormat get_current_frame_ycbcr_format() const
179 return current_frame_ycbcr_format;
182 // Only valid to call during the frame callback.
183 std::string get_last_subtitle() const
185 return last_subtitle;
189 bool get_has_last_subtitle() const
191 return has_last_subtitle;
195 movit::RGBTriplet get_last_neutral_color() const
197 return last_neutral_color;
200 void set_dequeue_thread_callbacks(std::function<void()> init, std::function<void()> cleanup) override
202 dequeue_init_callback = init;
203 dequeue_cleanup_callback = cleanup;
204 has_dequeue_callbacks = true;
207 void set_card_disconnected_callback(bmusb::card_disconnected_callback_t callback)
209 card_disconnected_callback = callback;
212 std::string get_description() const override
217 void configure_card() override;
218 void start_bm_capture() override;
219 void stop_dequeue_thread() override;
220 bool get_disconnected() const override { return disconnected; } // Only if play_once == true.
221 int get_srt_sock() const { return srt_sock; }
223 std::map<uint32_t, bmusb::VideoMode> get_available_video_modes() const override;
224 void set_video_mode(uint32_t video_mode_id) override {} // Ignore.
225 uint32_t get_current_video_mode() const override { return 0; }
227 static constexpr bmusb::PixelFormat PixelFormat_NV12 = static_cast<bmusb::PixelFormat>(100); // In the private range.
228 std::set<bmusb::PixelFormat> get_available_pixel_formats() const override {
229 return std::set<bmusb::PixelFormat>{ bmusb::PixelFormat_8BitBGRA, bmusb::PixelFormat_8BitYCbCrPlanar, PixelFormat_NV12 };
231 void set_pixel_format(bmusb::PixelFormat pixel_format) override {
232 this->pixel_format = pixel_format;
234 bmusb::PixelFormat get_current_pixel_format() const override {
238 std::map<uint32_t, std::string> get_available_video_inputs() const override {
239 return { { 0, "Auto" } }; }
240 void set_video_input(uint32_t video_input_id) override {} // Ignore.
241 uint32_t get_current_video_input() const override { return 0; }
243 std::map<uint32_t, std::string> get_available_audio_inputs() const override {
244 return { { 0, "Embedded" } };
246 void set_audio_input(uint32_t audio_input_id) override {} // Ignore.
247 uint32_t get_current_audio_input() const override { return 0; }
250 void producer_thread_func();
251 void send_disconnected_frame();
252 bool play_video(const std::string &pathname);
253 void internal_rewind();
255 // Returns true if there was an error.
256 bool process_queued_commands(AVFormatContext *format_ctx, const std::string &pathname, timespec last_modified, bool *rewound);
258 // Returns nullptr if no frame was decoded (e.g. EOF).
259 AVFrameWithDeleter decode_frame(AVFormatContext *format_ctx, AVCodecContext *video_codec_ctx, AVCodecContext *audio_codec_ctx,
260 const std::string &pathname, int video_stream_index, int audio_stream_index, int subtitle_stream_index,
261 bmusb::FrameAllocator::Frame *audio_frame, bmusb::AudioFormat *audio_format, int64_t *audio_pts, bool *error);
262 void convert_audio(const AVFrame *audio_avframe, bmusb::FrameAllocator::Frame *audio_frame, bmusb::AudioFormat *audio_format);
264 bmusb::VideoFormat construct_video_format(const AVFrame *frame, AVRational video_timebase);
265 UniqueFrame make_video_frame(const AVFrame *frame, const std::string &pathname, bool *error);
267 static int interrupt_cb_thunk(void *opaque);
271 static int read_srt_thunk(void *opaque, uint8_t *buf, int buf_size);
272 int read_srt(uint8_t *buf, int buf_size);
275 inline unsigned frame_width(const AVFrame *frame) const;
276 inline unsigned frame_height(const AVFrame *frame) const;
278 mutable std::mutex filename_mu;
279 std::string description, filename;
281 uint16_t timecode = 0;
282 unsigned width, height; // 0 means keep input size.
283 bmusb::PixelFormat pixel_format = bmusb::PixelFormat_8BitBGRA;
284 movit::YCbCrFormat current_frame_ycbcr_format;
285 bool running = false;
288 bool play_as_fast_as_possible = false; // Activated iff rate >= 10.0.
289 std::atomic<bool> should_interrupt{false};
290 bool last_frame_was_connected = true;
292 // TODO: Replace with std::optional if we go C++17.
293 bool frame_timeout_valid = false; // If true, will time out any reads after ten seconds.
294 std::chrono::steady_clock::time_point frame_timeout_started; // Only relevant if frame_timeout_valid == true.
296 bool has_dequeue_callbacks = false;
297 std::function<void()> dequeue_init_callback = nullptr;
298 std::function<void()> dequeue_cleanup_callback = nullptr;
300 bmusb::card_disconnected_callback_t card_disconnected_callback = nullptr;
301 bool play_once = false; // End thread after playing. Only for SRT, since the socket is ephemeral.
302 std::atomic<bool> disconnected{false};
304 bmusb::FrameAllocator *video_frame_allocator = nullptr;
305 bmusb::FrameAllocator *audio_frame_allocator = nullptr;
306 std::unique_ptr<bmusb::FrameAllocator> owned_video_frame_allocator;
307 std::unique_ptr<bmusb::FrameAllocator> owned_audio_frame_allocator;
308 frame_callback_t frame_callback = nullptr;
309 packet_callback_t video_callback = nullptr;
310 packet_callback_t audio_callback = nullptr;
312 SwsContextWithDeleter sws_ctx;
313 int sws_last_width = -1, sws_last_height = -1, sws_last_src_format = -1;
314 AVPixelFormat sws_dst_format = AVPixelFormat(-1); // In practice, always initialized.
315 AVRational video_timebase, audio_timebase;
316 bool is_mjpeg = false;
318 QuittableSleeper producer_thread_should_quit;
319 std::thread producer_thread;
321 int64_t pts_origin, last_pts;
322 std::chrono::steady_clock::time_point start, next_frame_start, last_frame;
325 struct QueuedCommand {
326 enum Command { REWIND, CHANGE_RATE } command;
327 double new_rate; // For CHANGE_RATE.
329 std::vector<QueuedCommand> command_queue; // Protected by <queue_mu>.
332 SwrContext *resampler = nullptr;
333 AVSampleFormat last_src_format, last_dst_format;
334 AVChannelLayout last_channel_layout;
335 int last_sample_rate;
337 // Subtitles (no decoding done, really).
338 bool has_last_subtitle = false;
339 std::string last_subtitle;
341 movit::RGBTriplet last_neutral_color{1.0f, 1.0f, 1.0f};
343 // Used for suppressing repeated warnings. Reset when a video starts playing.
344 // -1 is strictly speaking outside the range of the enum, but hopefully, it will be alright.
345 AVColorSpace last_colorspace = static_cast<AVColorSpace>(-1);
346 AVChromaLocation last_chroma_location = static_cast<AVChromaLocation>(-1);
349 #endif // !defined(_FFMPEG_CAPTURE_H)