1 // Kaeru (換える), a simple transcoder intended for use with Nageru.
3 #include "audio_encoder.h"
4 #include "basic_stats.h"
7 #include "ffmpeg_capture.h"
9 #include "shared/mux.h"
10 #include "quittable_sleeper.h"
11 #include "shared/timebase.h"
12 #include "x264_encoder.h"
21 using namespace bmusb;
22 using namespace movit;
24 using namespace std::chrono;
25 using namespace std::placeholders;
27 Mixer *global_mixer = nullptr;
28 X264Encoder *global_x264_encoder = nullptr;
30 BasicStats *global_basic_stats = nullptr;
31 QuittableSleeper should_quit;
32 MuxMetrics stream_mux_metrics;
36 int write_packet(void *opaque, uint8_t *buf, int buf_size, AVIODataMarkerType type, int64_t time)
38 static bool seen_sync_markers = false;
39 static string stream_mux_header;
40 HTTPD *httpd = (HTTPD *)opaque;
42 if (type == AVIO_DATA_MARKER_SYNC_POINT || type == AVIO_DATA_MARKER_BOUNDARY_POINT) {
43 seen_sync_markers = true;
44 } else if (type == AVIO_DATA_MARKER_UNKNOWN && !seen_sync_markers) {
45 // We don't know if this is a keyframe or not (the muxer could
46 // avoid marking it), so we just have to make the best of it.
47 type = AVIO_DATA_MARKER_SYNC_POINT;
50 if (type == AVIO_DATA_MARKER_HEADER) {
51 stream_mux_header.append((char *)buf, buf_size);
52 httpd->set_header(HTTPD::MAIN_STREAM, stream_mux_header);
54 httpd->add_data(HTTPD::MAIN_STREAM, (char *)buf, buf_size, type == AVIO_DATA_MARKER_SYNC_POINT, time, AVRational{ AV_TIME_BASE, 1 });
61 unique_ptr<Mux> create_mux(HTTPD *httpd, AVOutputFormat *oformat, X264Encoder *x264_encoder, AudioEncoder *audio_encoder)
63 AVFormatContext *avctx = avformat_alloc_context();
64 avctx->oformat = oformat;
66 uint8_t *buf = (uint8_t *)av_malloc(MUX_BUFFER_SIZE);
67 avctx->pb = avio_alloc_context(buf, MUX_BUFFER_SIZE, 1, httpd, nullptr, nullptr, nullptr);
68 avctx->pb->write_data_type = &write_packet;
69 avctx->pb->ignore_boundary_point = 1;
70 avctx->flags = AVFMT_FLAG_CUSTOM_IO;
72 string video_extradata = x264_encoder->get_global_headers();
75 mux.reset(new Mux(avctx, global_flags.width, global_flags.height, Mux::CODEC_H264, video_extradata, audio_encoder->get_codec_parameters().get(),
76 get_color_space(global_flags.ycbcr_rec709_coefficients), Mux::WITH_AUDIO, COARSE_TIMEBASE,
77 /*write_callback=*/nullptr, Mux::WRITE_FOREGROUND, { &stream_mux_metrics }));
78 stream_mux_metrics.init({{ "destination", "http" }});
82 void video_frame_callback(FFmpegCapture *video, X264Encoder *x264_encoder, AudioEncoder *audio_encoder,
83 int64_t video_pts, AVRational video_timebase,
84 int64_t audio_pts, AVRational audio_timebase,
86 FrameAllocator::Frame video_frame, size_t video_offset, VideoFormat video_format,
87 FrameAllocator::Frame audio_frame, size_t audio_offset, AudioFormat audio_format)
89 if (video_pts >= 0 && video_frame.len > 0) {
90 ReceivedTimestamps ts;
91 ts.ts.push_back(steady_clock::now());
93 video_pts = av_rescale_q(video_pts, video_timebase, AVRational{ 1, TIMEBASE });
94 int64_t frame_duration = TIMEBASE * video_format.frame_rate_den / video_format.frame_rate_nom;
95 x264_encoder->add_frame(video_pts, frame_duration, video->get_current_frame_ycbcr_format().luma_coefficients, video_frame.data + video_offset, ts);
96 global_basic_stats->update(frame_num++, /*dropped_frames=*/0);
98 if (audio_frame.len > 0) {
99 // FFmpegCapture takes care of this for us.
100 assert(audio_format.num_channels == 2);
101 assert(audio_format.sample_rate == OUTPUT_FREQUENCY);
103 // TODO: Reduce some duplication against AudioMixer here.
104 size_t num_samples = audio_frame.len / (audio_format.bits_per_sample / 8);
105 vector<float> float_samples;
106 float_samples.resize(num_samples);
107 if (audio_format.bits_per_sample == 16) {
108 const int16_t *src = (const int16_t *)audio_frame.data;
109 float *dst = &float_samples[0];
110 for (size_t i = 0; i < num_samples; ++i) {
111 *dst++ = le16toh(*src++) * (1.0f / 32768.0f);
113 } else if (audio_format.bits_per_sample == 32) {
114 const int32_t *src = (const int32_t *)audio_frame.data;
115 float *dst = &float_samples[0];
116 for (size_t i = 0; i < num_samples; ++i) {
117 *dst++ = le32toh(*src++) * (1.0f / 2147483648.0f);
122 audio_pts = av_rescale_q(audio_pts, audio_timebase, AVRational{ 1, TIMEBASE });
123 audio_encoder->encode_audio(float_samples, audio_pts);
126 if (video_frame.owner) {
127 video_frame.owner->release_frame(video_frame);
129 if (audio_frame.owner) {
130 audio_frame.owner->release_frame(audio_frame);
134 void audio_frame_callback(Mux *mux, const AVPacket *pkt, AVRational timebase)
136 mux->add_packet(*pkt, pkt->pts, pkt->dts == AV_NOPTS_VALUE ? pkt->pts : pkt->dts, timebase, /*stream_index=*/1);
139 void adjust_bitrate(int signal)
141 int new_bitrate = global_flags.x264_bitrate;
142 if (signal == SIGUSR1) {
144 if (new_bitrate > 100000) {
145 fprintf(stderr, "Ignoring SIGUSR1, can't increase bitrate below 100000 kbit/sec (currently at %d kbit/sec)\n",
146 global_flags.x264_bitrate);
148 fprintf(stderr, "Increasing bitrate to %d kbit/sec due to SIGUSR1.\n", new_bitrate);
149 global_flags.x264_bitrate = new_bitrate;
150 global_x264_encoder->change_bitrate(new_bitrate);
152 } else if (signal == SIGUSR2) {
154 if (new_bitrate < 100) {
155 fprintf(stderr, "Ignoring SIGUSR2, can't decrease bitrate below 100 kbit/sec (currently at %d kbit/sec)\n",
156 global_flags.x264_bitrate);
158 fprintf(stderr, "Decreasing bitrate to %d kbit/sec due to SIGUSR2.\n", new_bitrate);
159 global_flags.x264_bitrate = new_bitrate;
160 global_x264_encoder->change_bitrate(new_bitrate);
165 void request_quit(int signal)
170 int main(int argc, char *argv[])
172 parse_flags(PROGRAM_KAERU, argc, argv);
173 if (optind + 1 != argc) {
174 usage(PROGRAM_KAERU);
177 global_flags.num_cards = 1; // For latency metrics.
179 #if LIBAVFORMAT_VERSION_INT < AV_VERSION_INT(58, 9, 100)
182 avformat_network_init();
186 AVOutputFormat *oformat = av_guess_format(global_flags.stream_mux_name.c_str(), nullptr, nullptr);
187 assert(oformat != nullptr);
189 unique_ptr<AudioEncoder> audio_encoder;
190 if (global_flags.stream_audio_codec_name.empty()) {
191 audio_encoder.reset(new AudioEncoder(AUDIO_OUTPUT_CODEC_NAME, DEFAULT_AUDIO_OUTPUT_BIT_RATE, oformat));
193 audio_encoder.reset(new AudioEncoder(global_flags.stream_audio_codec_name, global_flags.stream_audio_codec_bitrate, oformat));
196 unique_ptr<X264Encoder> x264_encoder(new X264Encoder(oformat));
197 unique_ptr<Mux> http_mux = create_mux(&httpd, oformat, x264_encoder.get(), audio_encoder.get());
198 if (global_flags.transcode_audio) {
199 audio_encoder->add_mux(http_mux.get());
201 x264_encoder->add_mux(http_mux.get());
202 global_x264_encoder = x264_encoder.get();
204 FFmpegCapture video(argv[optind], global_flags.width, global_flags.height);
205 video.set_pixel_format(FFmpegCapture::PixelFormat_NV12);
206 video.set_frame_callback(bind(video_frame_callback, &video, x264_encoder.get(), audio_encoder.get(), _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11));
207 if (!global_flags.transcode_audio) {
208 video.set_audio_callback(bind(audio_frame_callback, http_mux.get(), _1, _2));
210 video.configure_card();
211 video.start_bm_capture();
212 video.change_rate(2.0); // Be sure never to really fall behind, but also don't dump huge amounts of stuff onto x264.
214 BasicStats basic_stats(/*verbose=*/false, /*use_opengl=*/false);
215 global_basic_stats = &basic_stats;
216 httpd.start(global_flags.http_port);
218 signal(SIGUSR1, adjust_bitrate);
219 signal(SIGUSR2, adjust_bitrate);
220 signal(SIGINT, request_quit);
222 while (!should_quit.should_quit()) {
223 should_quit.sleep_for(hours(1000));
226 video.stop_dequeue_thread();
227 // Stop the x264 encoder before killing the mux it's writing to.
228 x264_encoder.reset();