bool begin_frame(GLuint *y_tex, GLuint *cbcr_tex);
RefCountedGLsync end_frame(int64_t pts, const vector<RefCountedFrame> &input_frames);
void shutdown();
+ void open_output_file(const std::string &filename);
+ void close_output_file();
private:
struct storage_task {
int64_t pts;
};
+ // So we never get negative dts.
+ int64_t global_delay() const {
+ return int64_t(ip_period - 1) * (TIMEBASE / MAX_FPS);
+ }
+
void encode_thread_func();
void encode_remaining_frames_as_p(int encoding_frame_num, int gop_start_display_frame_num, int64_t last_dts);
void add_packet_for_uncompressed_frame(int64_t pts, const uint8_t *data);
void encode_frame(PendingFrame frame, int encoding_frame_num, int display_frame_num, int gop_start_display_frame_num,
int frame_type, int64_t pts, int64_t dts);
void storage_task_thread();
+ void encode_audio(const vector<float> &audio,
+ vector<float> *audio_queue,
+ int64_t audio_pts,
+ AVCodecContext *ctx,
+ const vector<PacketDestination *> &destinations);
+ void encode_audio_one_frame(const float *audio,
+ size_t num_samples, // In each channel.
+ int64_t audio_pts,
+ AVCodecContext *ctx,
+ const vector<PacketDestination *> &destinations);
void storage_task_enqueue(storage_task task);
void save_codeddata(storage_task task);
int render_packedsequence();
map<int64_t, vector<float>> pending_audio_frames; // under frame_queue_mutex
QSurface *surface;
- AVCodecContext *context_audio;
+ AVCodecContext *context_audio_file;
+ AVCodecContext *context_audio_stream = nullptr; // nullptr = don't code separate audio for stream.
+
+ vector<float> audio_queue_file;
+ vector<float> audio_queue_stream;
+
AVFrame *audio_frame = nullptr;
HTTPD *httpd;
unique_ptr<FrameReorderer> reorderer;
int frame_height;
int frame_width_mbaligned;
int frame_height_mbaligned;
+
+ unique_ptr<Mux> file_mux; // To local disk.
};
// Supposedly vaRenderPicture() is supposed to destroy the buffer implicitly,
string data;
- const int64_t global_delay = int64_t(ip_period - 1) * (TIMEBASE / MAX_FPS); // So we never get negative dts.
-
va_status = vaMapBuffer(va_dpy, gl_surfaces[task.display_order % SURFACE_NUM].coded_buf, (void **)(&buf_list));
CHECK_VASTATUS(va_status, "vaMapBuffer");
while (buf_list != NULL) {
pkt.flags = 0;
}
//pkt.duration = 1;
- httpd->add_packet(pkt, task.pts + global_delay, task.dts + global_delay,
- global_flags.uncompressed_video_to_http ? HTTPD::DESTINATION_FILE_ONLY : HTTPD::DESTINATION_FILE_AND_HTTP);
+ if (file_mux) {
+ file_mux->add_packet(pkt, task.pts + global_delay(), task.dts + global_delay());
+ }
+ if (!global_flags.uncompressed_video_to_http) {
+ httpd->add_packet(pkt, task.pts + global_delay(), task.dts + global_delay());
+ }
}
// Encode and add all audio frames up to and including the pts of this video frame.
for ( ;; ) {
pending_audio_frames.erase(it);
}
- audio_frame->nb_samples = audio.size() / 2;
- audio_frame->format = AV_SAMPLE_FMT_S32;
- audio_frame->channel_layout = AV_CH_LAYOUT_STEREO;
+ if (context_audio_stream) {
+ encode_audio(audio, &audio_queue_file, audio_pts, context_audio_file, { file_mux.get() });
+ encode_audio(audio, &audio_queue_stream, audio_pts, context_audio_stream, { httpd });
+ } else {
+ encode_audio(audio, &audio_queue_file, audio_pts, context_audio_file, { httpd, file_mux.get() });
+ }
+
+ if (audio_pts == task.pts) break;
+ }
+}
+
+void H264EncoderImpl::encode_audio(
+ const vector<float> &audio,
+ vector<float> *audio_queue,
+ int64_t audio_pts,
+ AVCodecContext *ctx,
+ const vector<PacketDestination *> &destinations)
+{
+ if (ctx->frame_size == 0) {
+ // No queueing needed.
+ assert(audio_queue->empty());
+ assert(audio.size() % 2 == 0);
+ encode_audio_one_frame(&audio[0], audio.size() / 2, audio_pts, ctx, destinations);
+ return;
+ }
+
+ audio_queue->insert(audio_queue->end(), audio.begin(), audio.end());
+ size_t sample_num;
+ for (sample_num = 0;
+ sample_num + ctx->frame_size * 2 <= audio_queue->size();
+ sample_num += ctx->frame_size * 2) {
+ encode_audio_one_frame(&(*audio_queue)[sample_num],
+ ctx->frame_size,
+ audio_pts,
+ ctx,
+ destinations);
+ }
+ audio_queue->erase(audio_queue->begin(), audio_queue->begin() + sample_num);
+}
- unique_ptr<int32_t[]> int_samples(new int32_t[audio.size()]);
- int ret = avcodec_fill_audio_frame(audio_frame, 2, AV_SAMPLE_FMT_S32, (const uint8_t*)int_samples.get(), audio.size() * sizeof(int32_t), 1);
+void H264EncoderImpl::encode_audio_one_frame(
+ const float *audio,
+ size_t num_samples,
+ int64_t audio_pts,
+ AVCodecContext *ctx,
+ const vector<PacketDestination *> &destinations)
+{
+ audio_frame->nb_samples = num_samples;
+ audio_frame->channel_layout = AV_CH_LAYOUT_STEREO;
+
+ unique_ptr<float[]> planar_samples;
+ unique_ptr<int32_t[]> int_samples;
+
+ if (ctx->sample_fmt == AV_SAMPLE_FMT_FLTP) {
+ audio_frame->format = AV_SAMPLE_FMT_FLTP;
+ planar_samples.reset(new float[num_samples * 2]);
+ avcodec_fill_audio_frame(audio_frame, 2, AV_SAMPLE_FMT_FLTP, (const uint8_t*)planar_samples.get(), num_samples * 2 * sizeof(float), 0);
+ for (size_t i = 0; i < num_samples; ++i) {
+ planar_samples[i] = audio[i * 2 + 0];
+ planar_samples[i + num_samples] = audio[i * 2 + 1];
+ }
+ } else {
+ assert(ctx->sample_fmt == AV_SAMPLE_FMT_S32);
+ int_samples.reset(new int32_t[num_samples * 2]);
+ int ret = avcodec_fill_audio_frame(audio_frame, 2, AV_SAMPLE_FMT_S32, (const uint8_t*)int_samples.get(), num_samples * 2 * sizeof(int32_t), 1);
if (ret < 0) {
fprintf(stderr, "avcodec_fill_audio_frame() failed with %d\n", ret);
exit(1);
}
- for (int i = 0; i < audio_frame->nb_samples * 2; ++i) {
+ for (size_t i = 0; i < num_samples * 2; ++i) {
if (audio[i] >= 1.0f) {
int_samples[i] = 2147483647;
} else if (audio[i] <= -1.0f) {
int_samples[i] = lrintf(audio[i] * 2147483647.0f);
}
}
+ }
- AVPacket pkt;
- av_init_packet(&pkt);
- pkt.data = nullptr;
- pkt.size = 0;
- int got_output;
- avcodec_encode_audio2(context_audio, &pkt, audio_frame, &got_output);
- if (got_output) {
- pkt.stream_index = 1;
- pkt.flags = AV_PKT_FLAG_KEY;
- httpd->add_packet(pkt, audio_pts + global_delay, audio_pts + global_delay, HTTPD::DESTINATION_FILE_AND_HTTP);
+ AVPacket pkt;
+ av_init_packet(&pkt);
+ pkt.data = nullptr;
+ pkt.size = 0;
+ int got_output = 0;
+ avcodec_encode_audio2(ctx, &pkt, audio_frame, &got_output);
+ if (got_output) {
+ pkt.stream_index = 1;
+ pkt.flags = AV_PKT_FLAG_KEY;
+ for (PacketDestination *dest : destinations) {
+ dest->add_packet(pkt, audio_pts + global_delay(), audio_pts + global_delay());
}
- // TODO: Delayed frames.
- av_frame_unref(audio_frame);
- av_free_packet(&pkt);
- if (audio_pts == task.pts) break;
}
+ // TODO: Delayed frames.
+ av_frame_unref(audio_frame);
+ av_free_packet(&pkt);
}
-
// this is weird. but it seems to put a new frame onto the queue
void H264EncoderImpl::storage_task_enqueue(storage_task task)
{
return 0;
}
+namespace {
-H264EncoderImpl::H264EncoderImpl(QSurface *surface, const string &va_display, int width, int height, HTTPD *httpd)
- : current_storage_frame(0), surface(surface), httpd(httpd)
+void init_audio_encoder(const string &codec_name, int bit_rate, AVCodecContext **ctx)
{
- AVCodec *codec_audio = avcodec_find_encoder(AUDIO_OUTPUT_CODEC);
- context_audio = avcodec_alloc_context3(codec_audio);
- context_audio->bit_rate = AUDIO_OUTPUT_BIT_RATE;
+ AVCodec *codec_audio = avcodec_find_encoder_by_name(codec_name.c_str());
+ if (codec_audio == nullptr) {
+ fprintf(stderr, "ERROR: Could not find codec '%s'\n", codec_name.c_str());
+ exit(1);
+ }
+
+ AVCodecContext *context_audio = avcodec_alloc_context3(codec_audio);
+ context_audio->bit_rate = bit_rate;
context_audio->sample_rate = OUTPUT_FREQUENCY;
- context_audio->sample_fmt = AUDIO_OUTPUT_SAMPLE_FMT;
+
+ // Choose sample format; we currently only support these two
+ // (see encode_audio), so we're a bit picky.
+ const AVSampleFormat *ptr = codec_audio->sample_fmts;
+ for ( ; *ptr != -1; ++ptr) {
+ if (*ptr == AV_SAMPLE_FMT_FLTP || *ptr == AV_SAMPLE_FMT_S32) {
+ context_audio->sample_fmt = *ptr;
+ break;
+ }
+ }
+ if (*ptr == -1) {
+ fprintf(stderr, "ERROR: Audio codec does not support fltp or s32 sample formats\n");
+ exit(1);
+ }
+
context_audio->channels = 2;
context_audio->channel_layout = AV_CH_LAYOUT_STEREO;
context_audio->time_base = AVRational{1, TIMEBASE};
if (avcodec_open2(context_audio, codec_audio, NULL) < 0) {
- fprintf(stderr, "Could not open codec\n");
+ fprintf(stderr, "Could not open codec '%s'\n", codec_name.c_str());
exit(1);
}
+
+ *ctx = context_audio;
+}
+
+} // namespace
+
+H264EncoderImpl::H264EncoderImpl(QSurface *surface, const string &va_display, int width, int height, HTTPD *httpd)
+ : current_storage_frame(0), surface(surface), httpd(httpd)
+{
+ init_audio_encoder(AUDIO_OUTPUT_CODEC_NAME, DEFAULT_AUDIO_OUTPUT_BIT_RATE, &context_audio_file);
+
+ if (!global_flags.stream_audio_codec_name.empty()) {
+ init_audio_encoder(global_flags.stream_audio_codec_name,
+ global_flags.stream_audio_codec_bitrate, &context_audio_stream);
+ }
+
audio_frame = av_frame_alloc();
frame_width = width;
is_shutdown = true;
}
+void H264EncoderImpl::open_output_file(const std::string &filename)
+{
+ AVFormatContext *avctx = avformat_alloc_context();
+ avctx->oformat = av_guess_format(NULL, filename.c_str(), NULL);
+ assert(filename.size() < sizeof(avctx->filename) - 1);
+ strcpy(avctx->filename, filename.c_str());
+
+ string url = "file:" + filename;
+ int ret = avio_open2(&avctx->pb, url.c_str(), AVIO_FLAG_WRITE, &avctx->interrupt_callback, NULL);
+ if (ret < 0) {
+ char tmp[AV_ERROR_MAX_STRING_SIZE];
+ fprintf(stderr, "%s: avio_open2() failed: %s\n", filename.c_str(), av_make_error_string(tmp, sizeof(tmp), ret));
+ exit(1);
+ }
+
+ file_mux.reset(new Mux(avctx, frame_width, frame_height, Mux::CODEC_H264, TIMEBASE, DEFAULT_AUDIO_OUTPUT_BIT_RATE));
+}
+
+void H264EncoderImpl::close_output_file()
+{
+ file_mux.reset();
+}
+
void H264EncoderImpl::encode_thread_func()
{
int64_t last_dts = -1;
pkt.size = frame_width * frame_height * 2;
pkt.stream_index = 0;
pkt.flags = AV_PKT_FLAG_KEY;
- httpd->add_packet(pkt, pts, pts, HTTPD::DESTINATION_HTTP_ONLY);
+ httpd->add_packet(pkt, pts, pts);
}
namespace {
if (global_flags.uncompressed_video_to_http) {
// Add uncompressed video. (Note that pts == dts here.)
- const int64_t global_delay = int64_t(ip_period - 1) * (TIMEBASE / MAX_FPS); // Needs to match audio.
- pair<int64_t, const uint8_t *> output_frame = reorderer->reorder_frame(pts + global_delay, reinterpret_cast<uint8_t *>(surf->y_ptr));
+ // Delay needs to match audio.
+ pair<int64_t, const uint8_t *> output_frame = reorderer->reorder_frame(pts + global_delay(), reinterpret_cast<uint8_t *>(surf->y_ptr));
if (output_frame.second != nullptr) {
add_packet_for_uncompressed_frame(output_frame.first, output_frame.second);
}
impl->shutdown();
}
-// Real class.
+void H264Encoder::open_output_file(const std::string &filename)
+{
+ impl->open_output_file(filename);
+}
+
+void H264Encoder::close_output_file()
+{
+ impl->close_output_file();
+}