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"
18 using namespace bmusb;
19 using namespace movit;
21 using namespace std::placeholders;
23 Mixer *global_mixer = nullptr;
25 int write_packet(void *opaque, uint8_t *buf, int buf_size, AVIODataMarkerType type, int64_t time)
27 static bool seen_sync_markers = false;
28 static string stream_mux_header;
29 HTTPD *httpd = (HTTPD *)opaque;
31 if (type == AVIO_DATA_MARKER_SYNC_POINT || type == AVIO_DATA_MARKER_BOUNDARY_POINT) {
32 seen_sync_markers = true;
33 } else if (type == AVIO_DATA_MARKER_UNKNOWN && !seen_sync_markers) {
34 // We don't know if this is a keyframe or not (the muxer could
35 // avoid marking it), so we just have to make the best of it.
36 type = AVIO_DATA_MARKER_SYNC_POINT;
39 if (type == AVIO_DATA_MARKER_HEADER) {
40 stream_mux_header.append((char *)buf, buf_size);
41 httpd->set_header(stream_mux_header);
43 httpd->add_data((char *)buf, buf_size, type == AVIO_DATA_MARKER_SYNC_POINT);
48 unique_ptr<Mux> create_mux(HTTPD *httpd, AVOutputFormat *oformat, X264Encoder *x264_encoder, AudioEncoder *audio_encoder)
50 AVFormatContext *avctx = avformat_alloc_context();
51 avctx->oformat = oformat;
53 uint8_t *buf = (uint8_t *)av_malloc(MUX_BUFFER_SIZE);
54 avctx->pb = avio_alloc_context(buf, MUX_BUFFER_SIZE, 1, httpd, nullptr, nullptr, nullptr);
55 avctx->pb->write_data_type = &write_packet;
56 avctx->pb->ignore_boundary_point = 1;
58 string video_extradata = x264_encoder->get_global_headers();
61 int time_base = global_flags.stream_coarse_timebase ? COARSE_TIMEBASE : TIMEBASE;
62 mux.reset(new Mux(avctx, global_flags.width, global_flags.height, Mux::CODEC_H264, video_extradata, audio_encoder->get_codec_parameters().get(), time_base,
63 /*write_callback=*/nullptr, Mux::WRITE_FOREGROUND, {}));
67 void video_frame_callback(FFmpegCapture *video, X264Encoder *x264_encoder, AudioEncoder *audio_encoder,
68 int64_t video_pts, AVRational video_timebase,
69 int64_t audio_pts, AVRational audio_timebase,
71 FrameAllocator::Frame video_frame, size_t video_offset, VideoFormat video_format,
72 FrameAllocator::Frame audio_frame, size_t audio_offset, AudioFormat audio_format)
74 if (video_pts >= 0 && video_frame.len > 0) {
75 video_pts = av_rescale_q(video_pts, video_timebase, AVRational{ 1, TIMEBASE });
76 int64_t frame_duration = TIMEBASE * video_format.frame_rate_nom / video_format.frame_rate_den;
77 x264_encoder->add_frame(video_pts, frame_duration, video->get_current_frame_ycbcr_format().luma_coefficients, video_frame.data + video_offset, ReceivedTimestamps());
79 if (audio_frame.len > 0) {
80 // FFmpegCapture takes care of this for us.
81 assert(audio_format.num_channels == 2);
82 assert(audio_format.sample_rate == OUTPUT_FREQUENCY);
84 // TODO: Reduce some duplication against AudioMixer here.
85 size_t num_samples = audio_frame.len / (audio_format.bits_per_sample / 8);
86 vector<float> float_samples;
87 float_samples.resize(num_samples);
88 if (audio_format.bits_per_sample == 16) {
89 const int16_t *src = (const int16_t *)audio_frame.data;
90 float *dst = &float_samples[0];
91 for (size_t i = 0; i < num_samples; ++i) {
92 *dst++ = le16toh(*src++) * (1.0f / 32768.0f);
94 } else if (audio_format.bits_per_sample == 32) {
95 const int32_t *src = (const int32_t *)audio_frame.data;
96 float *dst = &float_samples[0];
97 for (size_t i = 0; i < num_samples; ++i) {
98 *dst++ = le32toh(*src++) * (1.0f / 2147483648.0f);
103 audio_pts = av_rescale_q(audio_pts, audio_timebase, AVRational{ 1, TIMEBASE });
104 audio_encoder->encode_audio(float_samples, audio_pts);
107 if (video_frame.owner) {
108 video_frame.owner->release_frame(video_frame);
110 if (audio_frame.owner) {
111 audio_frame.owner->release_frame(audio_frame);
115 void audio_frame_callback(Mux *mux, const AVPacket *pkt, AVRational timebase)
117 mux->add_packet(*pkt, pkt->pts, pkt->dts == AV_NOPTS_VALUE ? pkt->pts : pkt->dts, timebase);
120 int main(int argc, char *argv[])
122 parse_flags(PROGRAM_KAERU, argc, argv);
123 if (optind + 1 != argc) {
124 usage(PROGRAM_KAERU);
129 avformat_network_init();
133 AVOutputFormat *oformat = av_guess_format(global_flags.stream_mux_name.c_str(), nullptr, nullptr);
134 assert(oformat != nullptr);
136 unique_ptr<AudioEncoder> audio_encoder;
137 if (global_flags.transcode_audio) {
138 if (global_flags.stream_audio_codec_name.empty()) {
139 audio_encoder.reset(new AudioEncoder(AUDIO_OUTPUT_CODEC_NAME, DEFAULT_AUDIO_OUTPUT_BIT_RATE, oformat));
141 audio_encoder.reset(new AudioEncoder(global_flags.stream_audio_codec_name, global_flags.stream_audio_codec_bitrate, oformat));
145 X264Encoder x264_encoder(oformat);
146 unique_ptr<Mux> http_mux = create_mux(&httpd, oformat, &x264_encoder, audio_encoder.get());
147 if (global_flags.transcode_audio) {
148 audio_encoder->add_mux(http_mux.get());
150 x264_encoder.add_mux(http_mux.get());
152 FFmpegCapture video(argv[optind], global_flags.width, global_flags.height);
153 video.set_pixel_format(FFmpegCapture::PixelFormat_NV12);
154 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));
155 if (!global_flags.transcode_audio) {
156 video.set_audio_callback(bind(audio_frame_callback, http_mux.get(), _1, _2));
158 video.configure_card();
159 video.start_bm_capture();
160 video.change_rate(2.0); // Be sure never to really fall behind, but also don't dump huge amounts of stuff onto x264.