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.
42 #include <movit/effect.h>
43 #include <movit/ycbcr.h>
46 #include <libavutil/frame.h>
47 #include <libswresample/swresample.h>
48 #include <libavutil/channel_layout.h>
49 #include <libavutil/pixfmt.h>
50 #include <libavutil/rational.h>
51 #include <libavutil/samplefmt.h>
54 #include "bmusb/bmusb.h"
55 #include "shared/ffmpeg_raii.h"
56 #include "ref_counted_frame.h"
57 #include "quittable_sleeper.h"
59 struct AVFormatContext;
64 class FFmpegCapture : public bmusb::CaptureInterface
67 FFmpegCapture(const std::string &filename, unsigned width, unsigned height);
69 // Takes ownership of the SRT client socket.
70 FFmpegCapture(int srt_sock, const std::string &stream_id);
74 void set_card_index(int card_index)
76 this->card_index = card_index;
79 int get_card_index() const
86 std::lock_guard<std::mutex> lock(queue_mu);
87 command_queue.push_back(QueuedCommand { QueuedCommand::REWIND });
88 producer_thread_should_quit.wakeup();
91 void change_rate(double new_rate)
93 std::lock_guard<std::mutex> lock(queue_mu);
94 command_queue.push_back(QueuedCommand { QueuedCommand::CHANGE_RATE, new_rate });
95 producer_thread_should_quit.wakeup();
98 std::string get_filename() const
100 std::lock_guard<std::mutex> lock(filename_mu);
104 void change_filename(const std::string &new_filename)
106 std::lock_guard<std::mutex> lock(filename_mu);
107 filename = new_filename;
108 should_interrupt = true;
111 // Will stop the stream even if it's hung on blocking I/O.
114 should_interrupt = true;
118 void set_video_frame_allocator(bmusb::FrameAllocator *allocator) override
120 video_frame_allocator = allocator;
121 if (owned_video_frame_allocator.get() != allocator) {
122 owned_video_frame_allocator.reset();
126 bmusb::FrameAllocator *get_video_frame_allocator() override
128 return video_frame_allocator;
131 // Does not take ownership.
132 void set_audio_frame_allocator(bmusb::FrameAllocator *allocator) override
134 audio_frame_allocator = allocator;
135 if (owned_audio_frame_allocator.get() != allocator) {
136 owned_audio_frame_allocator.reset();
140 bmusb::FrameAllocator *get_audio_frame_allocator() override
142 return audio_frame_allocator;
145 // FFmpegCapture-specific overload of set_frame_callback that also gives
146 // the raw original pts from the video. Negative pts means a dummy frame.
147 typedef std::function<void(int64_t video_pts, AVRational video_timebase, int64_t audio_pts, AVRational audio_timebase,
149 bmusb::FrameAllocator::Frame video_frame, size_t video_offset, bmusb::VideoFormat video_format,
150 bmusb::FrameAllocator::Frame audio_frame, size_t audio_offset, bmusb::AudioFormat audio_format)>
152 void set_frame_callback(frame_callback_t callback)
154 frame_callback = callback;
157 void set_frame_callback(bmusb::frame_callback_t callback) override
159 frame_callback = std::bind(
161 std::placeholders::_5,
162 std::placeholders::_6,
163 std::placeholders::_7,
164 std::placeholders::_8,
165 std::placeholders::_9,
166 std::placeholders::_10,
167 std::placeholders::_11);
170 // FFmpegCapture-specific callback that gives the raw audio/video.
171 typedef std::function<void(const AVPacket *pkt, const AVRational timebase)> packet_callback_t;
172 void set_video_callback(packet_callback_t callback)
174 video_callback = callback;
176 void set_audio_callback(packet_callback_t callback)
178 audio_callback = callback;
181 // Used to get precise information about the Y'CbCr format used
182 // for a given frame. Only valid to call during the frame callback,
183 // and only when receiving a frame with pixel format PixelFormat_8BitYCbCrPlanar.
184 movit::YCbCrFormat get_current_frame_ycbcr_format() const
186 return current_frame_ycbcr_format;
189 // Only valid to call during the frame callback.
190 std::string get_last_subtitle() const
192 return last_subtitle;
196 bool get_has_last_subtitle() const
198 return has_last_subtitle;
202 movit::RGBTriplet get_last_neutral_color() const
204 return last_neutral_color;
207 void set_dequeue_thread_callbacks(std::function<void()> init, std::function<void()> cleanup) override
209 dequeue_init_callback = init;
210 dequeue_cleanup_callback = cleanup;
211 has_dequeue_callbacks = true;
214 void set_card_disconnected_callback(bmusb::card_disconnected_callback_t callback)
216 card_disconnected_callback = callback;
219 std::string get_description() const override
224 void configure_card() override;
225 void start_bm_capture() override;
226 void stop_dequeue_thread() override;
227 bool get_disconnected() const override { return disconnected; } // Only if play_once == true.
228 int get_srt_sock() const { return srt_sock; }
230 std::map<uint32_t, bmusb::VideoMode> get_available_video_modes() const override;
231 void set_video_mode(uint32_t video_mode_id) override {} // Ignore.
232 uint32_t get_current_video_mode() const override { return 0; }
234 static constexpr bmusb::PixelFormat PixelFormat_NV12 = static_cast<bmusb::PixelFormat>(100); // In the private range.
235 std::set<bmusb::PixelFormat> get_available_pixel_formats() const override {
236 return std::set<bmusb::PixelFormat>{ bmusb::PixelFormat_8BitBGRA, bmusb::PixelFormat_8BitYCbCrPlanar, PixelFormat_NV12 };
238 void set_pixel_format(bmusb::PixelFormat pixel_format) override {
239 this->pixel_format = pixel_format;
241 bmusb::PixelFormat get_current_pixel_format() const override {
245 std::map<uint32_t, std::string> get_available_video_inputs() const override {
246 return { { 0, "Auto" } }; }
247 void set_video_input(uint32_t video_input_id) override {} // Ignore.
248 uint32_t get_current_video_input() const override { return 0; }
250 std::map<uint32_t, std::string> get_available_audio_inputs() const override {
251 return { { 0, "Embedded" } };
253 void set_audio_input(uint32_t audio_input_id) override {} // Ignore.
254 uint32_t get_current_audio_input() const override { return 0; }
257 void producer_thread_func();
258 void send_disconnected_frame();
259 bool play_video(const std::string &pathname);
260 void internal_rewind();
262 // Returns true if there was an error.
263 bool process_queued_commands(AVFormatContext *format_ctx, const std::string &pathname, timespec last_modified, bool *rewound);
265 // Returns nullptr if no frame was decoded (e.g. EOF).
266 AVFrameWithDeleter decode_frame(AVFormatContext *format_ctx, AVCodecContext *video_codec_ctx, AVCodecContext *audio_codec_ctx,
267 const std::string &pathname, int video_stream_index, int audio_stream_index, int subtitle_stream_index,
268 bmusb::FrameAllocator::Frame *audio_frame, bmusb::AudioFormat *audio_format, int64_t *audio_pts, bool *error);
269 void convert_audio(const AVFrame *audio_avframe, bmusb::FrameAllocator::Frame *audio_frame, bmusb::AudioFormat *audio_format);
271 bmusb::VideoFormat construct_video_format(const AVFrame *frame, AVRational video_timebase);
272 UniqueFrame make_video_frame(const AVFrame *frame, const std::string &pathname, bool *error);
274 static int interrupt_cb_thunk(void *opaque);
278 static int read_srt_thunk(void *opaque, uint8_t *buf, int buf_size);
279 int read_srt(uint8_t *buf, int buf_size);
282 inline unsigned frame_width(const AVFrame *frame) const;
283 inline unsigned frame_height(const AVFrame *frame) const;
285 mutable std::mutex filename_mu;
286 std::string description, filename;
288 uint16_t timecode = 0;
289 unsigned width, height; // 0 means keep input size.
290 bmusb::PixelFormat pixel_format = bmusb::PixelFormat_8BitBGRA;
291 movit::YCbCrFormat current_frame_ycbcr_format;
292 bool running = false;
295 bool play_as_fast_as_possible = false; // Activated iff rate >= 10.0.
296 std::atomic<bool> should_interrupt{false};
297 bool last_frame_was_connected = true;
299 // TODO: Replace with std::optional if we go C++17.
300 bool frame_timeout_valid = false; // If true, will time out any reads after ten seconds.
301 std::chrono::steady_clock::time_point frame_timeout_started; // Only relevant if frame_timeout_valid == true.
303 bool has_dequeue_callbacks = false;
304 std::function<void()> dequeue_init_callback = nullptr;
305 std::function<void()> dequeue_cleanup_callback = nullptr;
307 bmusb::card_disconnected_callback_t card_disconnected_callback = nullptr;
308 bool play_once = false; // End thread after playing. Only for SRT, since the socket is ephemeral.
309 std::atomic<bool> disconnected{false};
311 bmusb::FrameAllocator *video_frame_allocator = nullptr;
312 bmusb::FrameAllocator *audio_frame_allocator = nullptr;
313 std::unique_ptr<bmusb::FrameAllocator> owned_video_frame_allocator;
314 std::unique_ptr<bmusb::FrameAllocator> owned_audio_frame_allocator;
315 frame_callback_t frame_callback = nullptr;
316 packet_callback_t video_callback = nullptr;
317 packet_callback_t audio_callback = nullptr;
319 SwsContextWithDeleter sws_ctx;
320 int sws_last_width = -1, sws_last_height = -1, sws_last_src_format = -1;
321 AVPixelFormat sws_dst_format = AVPixelFormat(-1); // In practice, always initialized.
322 AVRational video_timebase, audio_timebase;
323 bool is_mjpeg = false;
325 QuittableSleeper producer_thread_should_quit;
326 std::thread producer_thread;
328 int64_t pts_origin, last_pts;
329 std::chrono::steady_clock::time_point start, next_frame_start, last_frame;
332 struct QueuedCommand {
333 enum Command { REWIND, CHANGE_RATE } command;
334 double new_rate; // For CHANGE_RATE.
336 std::vector<QueuedCommand> command_queue; // Protected by <queue_mu>.
339 SwrContext *resampler = nullptr;
340 AVSampleFormat last_src_format, last_dst_format;
341 AVChannelLayout last_channel_layout;
342 int last_sample_rate;
344 // Subtitles (no decoding done, really).
345 bool has_last_subtitle = false;
346 std::string last_subtitle;
348 movit::RGBTriplet last_neutral_color{1.0f, 1.0f, 1.0f};
350 // Used for suppressing repeated warnings. Reset when a video starts playing.
351 // -1 is strictly speaking outside the range of the enum, but hopefully, it will be alright.
352 AVColorSpace last_colorspace = static_cast<AVColorSpace>(-1);
353 AVChromaLocation last_chroma_location = static_cast<AVChromaLocation>(-1);
356 #endif // !defined(_FFMPEG_CAPTURE_H)