#include <thread>
#include "context.h"
+#include "timebase.h"
class QOpenGLContext;
class QSurface;
static int minimal_qp = 0;
static int intra_period = 30;
static int intra_idr_period = 60;
-static int ip_period = 1;
+static int ip_period = 3;
static int rc_mode = -1;
static int rc_default_modes[] = {
VA_RC_VBR,
bitstream_put_ui(bs, 1, 1); /* timing_info_present_flag */
{
bitstream_put_ui(bs, 1, 32); // FPS
- bitstream_put_ui(bs, frame_rate * 2, 32); // FPS
+ bitstream_put_ui(bs, TIMEBASE * 2, 32); // FPS
bitstream_put_ui(bs, 1, 1);
}
bitstream_put_ui(bs, 1, 1); /* nal_hrd_parameters_present_flag */
seq_param.max_num_ref_frames = num_ref_frames;
seq_param.seq_fields.bits.frame_mbs_only_flag = 1;
- seq_param.time_scale = frame_rate * 2;
+ seq_param.time_scale = TIMEBASE * 2;
seq_param.num_units_in_tick = 1; /* Tc = num_units_in_tick / scale */
seq_param.seq_fields.bits.log2_max_pic_order_cnt_lsb_minus4 = Log2MaxPicOrderCntLsb - 4;
seq_param.seq_fields.bits.log2_max_frame_num_minus4 = Log2MaxFrameNum - 4;;
}
vaUnmapBuffer(va_dpy, gl_surfaces[task.display_order % SURFACE_NUM].coded_buf);
- AVPacket pkt;
- memset(&pkt, 0, sizeof(pkt));
- pkt.buf = nullptr;
- pkt.pts = av_rescale_q(task.display_order, AVRational{1, frame_rate}, avstream->time_base);
- pkt.dts = av_rescale_q(task.encode_order, AVRational{1, frame_rate}, avstream->time_base);
- pkt.data = reinterpret_cast<uint8_t *>(&data[0]);
- pkt.size = data.size();
- pkt.stream_index = 0;
- if (task.frame_type == FRAME_IDR || task.frame_type == FRAME_I) {
- pkt.flags = AV_PKT_FLAG_KEY;
- } else {
- pkt.flags = 0;
+ const int64_t pts_dts_delay = (ip_period - 1) * (TIMEBASE / frame_rate); // FIXME: Wrong for variable frame rate.
+ const int64_t av_delay = TIMEBASE / 10; // Corresponds to the fixed delay in resampler.h. TODO: Make less hard-coded.
+ int64_t pts, dts;
+ {
+ {
+ unique_lock<mutex> lock(frame_queue_mutex);
+ assert(timestamps.count(task.display_order));
+ assert(timestamps.count(task.encode_order));
+ pts = timestamps[task.display_order];
+ dts = timestamps[task.encode_order];
+ }
+ // Add video.
+ AVPacket pkt;
+ memset(&pkt, 0, sizeof(pkt));
+ pkt.buf = nullptr;
+ pkt.pts = av_rescale_q(pts + av_delay + pts_dts_delay, AVRational{1, TIMEBASE}, avstream_video->time_base);
+ pkt.dts = av_rescale_q(dts + av_delay, AVRational{1, TIMEBASE}, avstream_video->time_base);
+ pkt.data = reinterpret_cast<uint8_t *>(&data[0]);
+ pkt.size = data.size();
+ pkt.stream_index = 0;
+ if (task.frame_type == FRAME_IDR || task.frame_type == FRAME_I) {
+ pkt.flags = AV_PKT_FLAG_KEY;
+ } else {
+ pkt.flags = 0;
+ }
+ //pkt.duration = 1;
+ av_interleaved_write_frame(avctx, &pkt);
+ }
+ // Encode and add all audio frames up to and including the pts of this video frame.
+ // (They can never be queued to us after the video frame they belong to, only before.)
+ for ( ;; ) {
+ int64_t audio_pts;
+ std::vector<float> audio;
+ {
+ unique_lock<mutex> lock(frame_queue_mutex);
+ if (pending_audio_frames.empty()) break;
+ auto it = pending_audio_frames.begin();
+ if (it->first > int(pts)) break;
+ audio_pts = it->first;
+ audio = move(it->second);
+ pending_audio_frames.erase(it);
+ }
+ AVFrame *frame = avcodec_alloc_frame();
+ frame->nb_samples = audio.size() / 2;
+ frame->format = AV_SAMPLE_FMT_FLT;
+ frame->channel_layout = AV_CH_LAYOUT_STEREO;
+
+ unique_ptr<float[]> planar_samples(new float[audio.size()]);
+ avcodec_fill_audio_frame(frame, 2, AV_SAMPLE_FMT_FLTP, (const uint8_t*)planar_samples.get(), audio.size() * sizeof(float), 0);
+ for (int i = 0; i < frame->nb_samples; ++i) {
+ planar_samples[i] = audio[i * 2 + 0];
+ planar_samples[i + frame->nb_samples] = audio[i * 2 + 1];
+ }
+
+ AVPacket pkt;
+ av_init_packet(&pkt);
+ pkt.data = nullptr;
+ pkt.size = 0;
+ int got_output;
+ avcodec_encode_audio2(avstream_audio->codec, &pkt, frame, &got_output);
+ if (got_output) {
+ pkt.pts = av_rescale_q(audio_pts + pts_dts_delay, AVRational{1, TIMEBASE}, avstream_audio->time_base);
+ pkt.dts = pkt.pts;
+ pkt.stream_index = 1;
+ av_interleaved_write_frame(avctx, &pkt);
+ }
+ // TODO: Delayed frames.
+ avcodec_free_frame(&frame);
+ }
+ {
+ unique_lock<mutex> lock(frame_queue_mutex);
+ timestamps.erase(task.encode_order - (ip_period - 1));
}
- pkt.duration = 1;
- av_interleaved_write_frame(avctx, &pkt);
#if 0
printf("\r "); /* return back to startpoint */
fprintf(stderr, "%s: avio_open2() failed\n", output_filename);
exit(1);
}
- AVCodec *codec = avcodec_find_encoder(AV_CODEC_ID_H264);
- avstream = avformat_new_stream(avctx, codec);
- if (avstream == nullptr) {
+ AVCodec *codec_video = avcodec_find_encoder(AV_CODEC_ID_H264);
+ avstream_video = avformat_new_stream(avctx, codec_video);
+ if (avstream_video == nullptr) {
fprintf(stderr, "%s: avformat_new_stream() failed\n", output_filename);
exit(1);
}
- avstream->time_base = AVRational{1, frame_rate};
- avstream->codec->width = width;
- avstream->codec->height = height;
- avstream->codec->time_base = AVRational{1, frame_rate};
- avstream->codec->ticks_per_frame = 1; // or 2?
+ avstream_video->time_base = AVRational{1, TIMEBASE};
+ avstream_video->codec->width = width;
+ avstream_video->codec->height = height;
+ avstream_video->codec->time_base = AVRational{1, TIMEBASE};
+ avstream_video->codec->ticks_per_frame = 1; // or 2?
+
+ AVCodec *codec_audio = avcodec_find_encoder(AV_CODEC_ID_MP3);
+ avstream_audio = avformat_new_stream(avctx, codec_audio);
+ if (avstream_audio == nullptr) {
+ fprintf(stderr, "%s: avformat_new_stream() failed\n", output_filename);
+ exit(1);
+ }
+ avstream_audio->time_base = AVRational{1, TIMEBASE};
+ avstream_audio->codec->bit_rate = 256000;
+ avstream_audio->codec->sample_rate = 48000;
+ avstream_audio->codec->sample_fmt = AV_SAMPLE_FMT_FLTP;
+ avstream_audio->codec->channels = 2;
+ avstream_audio->codec->channel_layout = AV_CH_LAYOUT_STEREO;
+ avstream_audio->codec->time_base = AVRational{1, TIMEBASE};
+
+ /* open it */
+ if (avcodec_open2(avstream_audio->codec, codec_audio, NULL) < 0) {
+ fprintf(stderr, "Could not open codec\n");
+ exit(1);
+ }
if (avformat_write_header(avctx, NULL) < 0) {
fprintf(stderr, "%s: avformat_write_header() failed\n", output_filename);
return true;
}
-void H264Encoder::end_frame(RefCountedGLsync fence, const std::vector<RefCountedFrame> &input_frames)
+void H264Encoder::add_audio(int64_t pts, std::vector<float> audio)
+{
+ {
+ unique_lock<mutex> lock(frame_queue_mutex);
+ pending_audio_frames[pts] = move(audio);
+ }
+ frame_queue_nonempty.notify_one();
+}
+
+
+void H264Encoder::end_frame(RefCountedGLsync fence, int64_t pts, const std::vector<RefCountedFrame> &input_frames)
{
{
unique_lock<mutex> lock(frame_queue_mutex);
- pending_frames[current_storage_frame++] = PendingFrame{ fence, input_frames };
+ pending_video_frames[current_storage_frame] = PendingFrame{ fence, input_frames };
+ timestamps[current_storage_frame] = pts;
+ ++current_storage_frame;
}
frame_queue_nonempty.notify_one();
}
{
unique_lock<mutex> lock(frame_queue_mutex);
- frame_queue_nonempty.wait(lock, [this]{ return copy_thread_should_quit || pending_frames.count(current_frame_display) != 0; });
+ frame_queue_nonempty.wait(lock, [this]{ return copy_thread_should_quit || pending_video_frames.count(current_frame_display) != 0; });
if (copy_thread_should_quit) return;
- frame = move(pending_frames[current_frame_display]);
- pending_frames.erase(current_frame_display);
+ frame = move(pending_video_frames[current_frame_display]);
+ pending_video_frames.erase(current_frame_display);
}
// Wait for the GPU to be done with the frame.