1 // Kaeru (換える), a simple transcoder intended for use with Nageru.
2 // This is experimental code, not yet supported.
4 #include "audio_encoder.h"
7 #include "ffmpeg_capture.h"
11 #include "x264_encoder.h"
20 using namespace bmusb;
21 using namespace movit;
23 using namespace std::chrono;
24 using namespace std::placeholders;
26 Mixer *global_mixer = nullptr;
27 X264Encoder *global_x264_encoder = nullptr;
28 MuxMetrics stream_mux_metrics;
30 int write_packet(void *opaque, uint8_t *buf, int buf_size, AVIODataMarkerType type, int64_t time)
32 static bool seen_sync_markers = false;
33 static string stream_mux_header;
34 HTTPD *httpd = (HTTPD *)opaque;
36 if (type == AVIO_DATA_MARKER_SYNC_POINT || type == AVIO_DATA_MARKER_BOUNDARY_POINT) {
37 seen_sync_markers = true;
38 } else if (type == AVIO_DATA_MARKER_UNKNOWN && !seen_sync_markers) {
39 // We don't know if this is a keyframe or not (the muxer could
40 // avoid marking it), so we just have to make the best of it.
41 type = AVIO_DATA_MARKER_SYNC_POINT;
44 if (type == AVIO_DATA_MARKER_HEADER) {
45 stream_mux_header.append((char *)buf, buf_size);
46 httpd->set_header(stream_mux_header);
48 httpd->add_data((char *)buf, buf_size, type == AVIO_DATA_MARKER_SYNC_POINT);
53 unique_ptr<Mux> create_mux(HTTPD *httpd, AVOutputFormat *oformat, X264Encoder *x264_encoder, AudioEncoder *audio_encoder)
55 AVFormatContext *avctx = avformat_alloc_context();
56 avctx->oformat = oformat;
58 uint8_t *buf = (uint8_t *)av_malloc(MUX_BUFFER_SIZE);
59 avctx->pb = avio_alloc_context(buf, MUX_BUFFER_SIZE, 1, httpd, nullptr, nullptr, nullptr);
60 avctx->pb->write_data_type = &write_packet;
61 avctx->pb->ignore_boundary_point = 1;
63 string video_extradata = x264_encoder->get_global_headers();
66 int time_base = global_flags.stream_coarse_timebase ? COARSE_TIMEBASE : TIMEBASE;
67 mux.reset(new Mux(avctx, global_flags.width, global_flags.height, Mux::CODEC_H264, video_extradata, audio_encoder->get_codec_parameters().get(), time_base,
68 /*write_callback=*/nullptr, Mux::WRITE_FOREGROUND, { &stream_mux_metrics }));
69 stream_mux_metrics.init({{ "destination", "http" }});
73 void video_frame_callback(FFmpegCapture *video, X264Encoder *x264_encoder, AudioEncoder *audio_encoder,
74 int64_t video_pts, AVRational video_timebase,
75 int64_t audio_pts, AVRational audio_timebase,
77 FrameAllocator::Frame video_frame, size_t video_offset, VideoFormat video_format,
78 FrameAllocator::Frame audio_frame, size_t audio_offset, AudioFormat audio_format)
80 if (video_pts >= 0 && video_frame.len > 0) {
81 ReceivedTimestamps ts;
82 ts.ts.push_back(steady_clock::now());
84 video_pts = av_rescale_q(video_pts, video_timebase, AVRational{ 1, TIMEBASE });
85 int64_t frame_duration = TIMEBASE * video_format.frame_rate_den / video_format.frame_rate_nom;
86 x264_encoder->add_frame(video_pts, frame_duration, video->get_current_frame_ycbcr_format().luma_coefficients, video_frame.data + video_offset, ts);
88 if (audio_frame.len > 0) {
89 // FFmpegCapture takes care of this for us.
90 assert(audio_format.num_channels == 2);
91 assert(audio_format.sample_rate == OUTPUT_FREQUENCY);
93 // TODO: Reduce some duplication against AudioMixer here.
94 size_t num_samples = audio_frame.len / (audio_format.bits_per_sample / 8);
95 vector<float> float_samples;
96 float_samples.resize(num_samples);
97 if (audio_format.bits_per_sample == 16) {
98 const int16_t *src = (const int16_t *)audio_frame.data;
99 float *dst = &float_samples[0];
100 for (size_t i = 0; i < num_samples; ++i) {
101 *dst++ = le16toh(*src++) * (1.0f / 32768.0f);
103 } else if (audio_format.bits_per_sample == 32) {
104 const int32_t *src = (const int32_t *)audio_frame.data;
105 float *dst = &float_samples[0];
106 for (size_t i = 0; i < num_samples; ++i) {
107 *dst++ = le32toh(*src++) * (1.0f / 2147483648.0f);
112 audio_pts = av_rescale_q(audio_pts, audio_timebase, AVRational{ 1, TIMEBASE });
113 audio_encoder->encode_audio(float_samples, audio_pts);
116 if (video_frame.owner) {
117 video_frame.owner->release_frame(video_frame);
119 if (audio_frame.owner) {
120 audio_frame.owner->release_frame(audio_frame);
124 void audio_frame_callback(Mux *mux, const AVPacket *pkt, AVRational timebase)
126 mux->add_packet(*pkt, pkt->pts, pkt->dts == AV_NOPTS_VALUE ? pkt->pts : pkt->dts, timebase);
129 void adjust_bitrate(int signal)
131 int new_bitrate = global_flags.x264_bitrate;
132 if (signal == SIGUSR1) {
134 if (new_bitrate > 100000) {
135 fprintf(stderr, "Ignoring SIGUSR1, can't increase bitrate below 100000 kbit/sec (currently at %d kbit/sec)\n",
136 global_flags.x264_bitrate);
138 fprintf(stderr, "Increasing bitrate to %d kbit/sec due to SIGUSR1.\n", new_bitrate);
139 global_flags.x264_bitrate = new_bitrate;
140 global_x264_encoder->change_bitrate(new_bitrate);
142 } else if (signal == SIGUSR2) {
144 if (new_bitrate < 100) {
145 fprintf(stderr, "Ignoring SIGUSR1, can't decrease bitrate below 100 kbit/sec (currently at %d kbit/sec)\n",
146 global_flags.x264_bitrate);
148 fprintf(stderr, "Decreasing 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);
155 int main(int argc, char *argv[])
157 parse_flags(PROGRAM_KAERU, argc, argv);
158 if (optind + 1 != argc) {
159 usage(PROGRAM_KAERU);
162 global_flags.num_cards = 1; // For latency metrics.
165 avformat_network_init();
169 AVOutputFormat *oformat = av_guess_format(global_flags.stream_mux_name.c_str(), nullptr, nullptr);
170 assert(oformat != nullptr);
172 unique_ptr<AudioEncoder> audio_encoder;
173 if (global_flags.transcode_audio) {
174 if (global_flags.stream_audio_codec_name.empty()) {
175 audio_encoder.reset(new AudioEncoder(AUDIO_OUTPUT_CODEC_NAME, DEFAULT_AUDIO_OUTPUT_BIT_RATE, oformat));
177 audio_encoder.reset(new AudioEncoder(global_flags.stream_audio_codec_name, global_flags.stream_audio_codec_bitrate, oformat));
181 X264Encoder x264_encoder(oformat);
182 unique_ptr<Mux> http_mux = create_mux(&httpd, oformat, &x264_encoder, audio_encoder.get());
183 if (global_flags.transcode_audio) {
184 audio_encoder->add_mux(http_mux.get());
186 x264_encoder.add_mux(http_mux.get());
187 global_x264_encoder = &x264_encoder;
189 FFmpegCapture video(argv[optind], global_flags.width, global_flags.height);
190 video.set_pixel_format(FFmpegCapture::PixelFormat_NV12);
191 video.set_frame_callback(bind(video_frame_callback, &video, &x264_encoder, audio_encoder.get(), _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11));
192 if (!global_flags.transcode_audio) {
193 video.set_audio_callback(bind(audio_frame_callback, http_mux.get(), _1, _2));
195 video.configure_card();
196 video.start_bm_capture();
197 video.change_rate(2.0); // Be sure never to really fall behind, but also don't dump huge amounts of stuff onto x264.
201 signal(SIGUSR1, adjust_bitrate);
202 signal(SIGUSR2, adjust_bitrate);