extern "C" {
#include <libavcodec/avcodec.h>
#include <libavformat/avformat.h>
+#include <libavresample/avresample.h>
#include <libavutil/channel_layout.h>
#include <libavutil/frame.h>
#include <libavutil/rational.h>
#include <libavutil/samplefmt.h>
+#include <libavutil/opt.h>
}
#include <libdrm/drm_fourcc.h>
#include <stdio.h>
#include "defs.h"
#include "flags.h"
#include "httpd.h"
+#include "mux.h"
#include "timebase.h"
+#include "x264encode.h"
using namespace std;
return make_pair(-pts, storage.second); // Re-invert pts (see reorder_frame()).
}
-class H264EncoderImpl {
+class H264EncoderImpl : public KeyFrameSignalReceiver {
public:
H264EncoderImpl(QSurface *surface, const string &va_display, int width, int height, HTTPD *httpd);
~H264EncoderImpl();
void open_output_file(const std::string &filename);
void close_output_file();
+ virtual void signal_keyframe() override {
+ stream_mux_writing_keyframes = true;
+ }
+
private:
struct storage_task {
unsigned long long display_order;
void encode_audio(const vector<float> &audio,
vector<float> *audio_queue,
int64_t audio_pts,
- AVCodecContext *ctx);
+ AVCodecContext *ctx,
+ AVAudioResampleContext *resampler,
+ const vector<Mux *> &muxes);
void encode_audio_one_frame(const float *audio,
size_t num_samples, // In each channel.
int64_t audio_pts,
- AVCodecContext *ctx);
+ AVCodecContext *ctx,
+ AVAudioResampleContext *resampler,
+ const vector<Mux *> &muxes);
void storage_task_enqueue(storage_task task);
void save_codeddata(storage_task task);
int render_packedsequence();
int release_encode();
void update_ReferenceFrames(int frame_type);
int update_RefPicList(int frame_type);
+ void open_output_stream();
+ void close_output_stream();
+ static int write_packet_thunk(void *opaque, uint8_t *buf, int buf_size);
+ int write_packet(uint8_t *buf, int buf_size);
bool is_shutdown = false;
bool use_zerocopy;
map<int64_t, vector<float>> pending_audio_frames; // under frame_queue_mutex
QSurface *surface;
- AVCodecContext *context_audio;
- vector<float> audio_queue;
+ AVCodecContext *context_audio_file;
+ AVCodecContext *context_audio_stream = nullptr; // nullptr = don't code separate audio for stream.
+
+ AVAudioResampleContext *resampler_audio_file = nullptr;
+ AVAudioResampleContext *resampler_audio_stream = nullptr;
+
+ vector<float> audio_queue_file;
+ vector<float> audio_queue_stream;
AVFrame *audio_frame = nullptr;
HTTPD *httpd;
unique_ptr<FrameReorderer> reorderer;
+ unique_ptr<X264Encoder> x264_encoder; // nullptr if not using x264.
Display *x11_display = nullptr;
int frame_width_mbaligned;
int frame_height_mbaligned;
+ unique_ptr<Mux> stream_mux; // To HTTP.
unique_ptr<Mux> file_mux; // To local disk.
+
+ // While Mux object is constructing, <stream_mux_writing_header> is true,
+ // and the header is being collected into stream_mux_header.
+ bool stream_mux_writing_header;
+ string stream_mux_header;
+
+ bool stream_mux_writing_keyframes = false;
};
// Supposedly vaRenderPicture() is supposed to destroy the buffer implicitly,
if (global_flags.uncompressed_video_to_http) {
fprintf(stderr, "Disabling zerocopy H.264 encoding due to --uncompressed_video_to_http.\n");
use_zerocopy = false;
+ } else if (global_flags.x264_video_to_http) {
+ fprintf(stderr, "Disabling zerocopy H.264 encoding due to --x264_video_to_http.\n");
+ use_zerocopy = false;
} else {
use_zerocopy = true;
}
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());
+ if (!global_flags.uncompressed_video_to_http &&
+ !global_flags.x264_video_to_http) {
+ stream_mux->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.
pending_audio_frames.erase(it);
}
- encode_audio(audio, &audio_queue, audio_pts, context_audio);
+ if (context_audio_stream) {
+ encode_audio(audio, &audio_queue_file, audio_pts, context_audio_file, resampler_audio_file, { file_mux.get() });
+ encode_audio(audio, &audio_queue_stream, audio_pts, context_audio_stream, resampler_audio_stream, { stream_mux.get() });
+ } else {
+ encode_audio(audio, &audio_queue_file, audio_pts, context_audio_file, resampler_audio_file, { stream_mux.get(), file_mux.get() });
+ }
if (audio_pts == task.pts) break;
}
const vector<float> &audio,
vector<float> *audio_queue,
int64_t audio_pts,
- AVCodecContext *ctx)
+ AVCodecContext *ctx,
+ AVAudioResampleContext *resampler,
+ const vector<Mux *> &muxes)
{
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);
+ encode_audio_one_frame(&audio[0], audio.size() / 2, audio_pts, ctx, resampler, muxes);
return;
}
+ int64_t sample_offset = audio_queue->size();
+
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) {
+ int64_t adjusted_audio_pts = audio_pts + (int64_t(sample_num) - sample_offset) * TIMEBASE / (OUTPUT_FREQUENCY * 2);
encode_audio_one_frame(&(*audio_queue)[sample_num],
ctx->frame_size,
- audio_pts,
- ctx);
+ adjusted_audio_pts,
+ ctx,
+ resampler,
+ muxes);
}
audio_queue->erase(audio_queue->begin(), audio_queue->begin() + sample_num);
}
const float *audio,
size_t num_samples,
int64_t audio_pts,
- AVCodecContext *ctx)
+ AVCodecContext *ctx,
+ AVAudioResampleContext *resampler,
+ const vector<Mux *> &muxes)
{
audio_frame->nb_samples = num_samples;
audio_frame->channel_layout = AV_CH_LAYOUT_STEREO;
+ audio_frame->format = ctx->sample_fmt;
+ audio_frame->sample_rate = OUTPUT_FREQUENCY;
- unique_ptr<float[]> planar_samples;
- unique_ptr<int32_t[]> int_samples;
+ if (av_samples_alloc(audio_frame->data, nullptr, 2, num_samples, ctx->sample_fmt, 0) < 0) {
+ fprintf(stderr, "Could not allocate %ld samples.\n", num_samples);
+ exit(1);
+ }
- 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 (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] = -2147483647;
- } else {
- int_samples[i] = lrintf(audio[i] * 2147483647.0f);
- }
- }
+ if (avresample_convert(resampler, audio_frame->data, 0, num_samples,
+ (uint8_t **)&audio, 0, num_samples) < 0) {
+ fprintf(stderr, "Audio conversion failed.\n");
+ exit(1);
}
AVPacket pkt;
avcodec_encode_audio2(ctx, &pkt, audio_frame, &got_output);
if (got_output) {
pkt.stream_index = 1;
- pkt.flags = AV_PKT_FLAG_KEY;
- if (file_mux) {
- file_mux->add_packet(pkt, audio_pts + global_delay(), audio_pts + global_delay());
+ pkt.flags = 0;
+ for (Mux *mux : muxes) {
+ mux->add_packet(pkt, audio_pts + global_delay(), audio_pts + global_delay());
}
- httpd->add_packet(pkt, audio_pts + global_delay(), audio_pts + global_delay());
}
+
+ av_freep(&audio_frame->data[0]);
+
// TODO: Delayed frames.
av_frame_unref(audio_frame);
av_free_packet(&pkt);
namespace {
-void init_audio_encoder(const string &codec_name, int bit_rate, AVCodecContext **ctx)
+void init_audio_encoder(const string &codec_name, int bit_rate, AVCodecContext **ctx, AVAudioResampleContext **resampler)
{
AVCodec *codec_audio = avcodec_find_encoder_by_name(codec_name.c_str());
if (codec_audio == nullptr) {
AVCodecContext *context_audio = avcodec_alloc_context3(codec_audio);
context_audio->bit_rate = bit_rate;
context_audio->sample_rate = OUTPUT_FREQUENCY;
-
- // 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->sample_fmt = codec_audio->sample_fmts[0];
context_audio->channels = 2;
context_audio->channel_layout = AV_CH_LAYOUT_STEREO;
context_audio->time_base = AVRational{1, TIMEBASE};
+ context_audio->flags |= CODEC_FLAG_GLOBAL_HEADER;
if (avcodec_open2(context_audio, codec_audio, NULL) < 0) {
fprintf(stderr, "Could not open codec '%s'\n", codec_name.c_str());
exit(1);
}
*ctx = context_audio;
+
+ *resampler = avresample_alloc_context();
+ if (*resampler == nullptr) {
+ fprintf(stderr, "Allocating resampler failed.\n");
+ exit(1);
+ }
+
+ av_opt_set_int(*resampler, "in_channel_layout", AV_CH_LAYOUT_STEREO, 0);
+ av_opt_set_int(*resampler, "out_channel_layout", AV_CH_LAYOUT_STEREO, 0);
+ av_opt_set_int(*resampler, "in_sample_rate", OUTPUT_FREQUENCY, 0);
+ av_opt_set_int(*resampler, "out_sample_rate", OUTPUT_FREQUENCY, 0);
+ av_opt_set_int(*resampler, "in_sample_fmt", AV_SAMPLE_FMT_FLT, 0);
+ av_opt_set_int(*resampler, "out_sample_fmt", context_audio->sample_fmt, 0);
+
+ if (avresample_open(*resampler) < 0) {
+ fprintf(stderr, "Could not open resample context.\n");
+ exit(1);
+ }
}
} // namespace
H264EncoderImpl::H264EncoderImpl(QSurface *surface, const string &va_display, int width, int height, HTTPD *httpd)
- : current_storage_frame(0), surface(surface), httpd(httpd)
+ : current_storage_frame(0), surface(surface), httpd(httpd), frame_width(width), frame_height(height)
{
- init_audio_encoder(AUDIO_OUTPUT_CODEC_NAME, AUDIO_OUTPUT_BIT_RATE, &context_audio);
+ init_audio_encoder(AUDIO_OUTPUT_CODEC_NAME, DEFAULT_AUDIO_OUTPUT_BIT_RATE, &context_audio_file, &resampler_audio_file);
- audio_frame = av_frame_alloc();
+ 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, &resampler_audio_stream);
+ }
- frame_width = width;
- frame_height = height;
frame_width_mbaligned = (frame_width + 15) & (~15);
frame_height_mbaligned = (frame_height + 15) & (~15);
+ open_output_stream();
+
+ audio_frame = av_frame_alloc();
+
//print_input();
- if (global_flags.uncompressed_video_to_http) {
+ if (global_flags.uncompressed_video_to_http ||
+ global_flags.x264_video_to_http) {
reorderer.reset(new FrameReorderer(ip_period - 1, frame_width, frame_height));
}
+ if (global_flags.x264_video_to_http) {
+ x264_encoder.reset(new X264Encoder(stream_mux.get()));
+ }
init_va(va_display);
setup_encode();
{
shutdown();
av_frame_free(&audio_frame);
-
- // TODO: Destroy context.
+ avresample_free(&resampler_audio_file);
+ avresample_free(&resampler_audio_stream);
+ avcodec_free_context(&context_audio_file);
+ avcodec_free_context(&context_audio_stream);
+ close_output_stream();
}
bool H264EncoderImpl::begin_frame(GLuint *y_tex, GLuint *cbcr_tex)
exit(1);
}
- file_mux.reset(new Mux(avctx, frame_width, frame_height, Mux::CODEC_H264, TIMEBASE));
+ file_mux.reset(new Mux(avctx, frame_width, frame_height, Mux::CODEC_H264, context_audio_file->codec, TIMEBASE, DEFAULT_AUDIO_OUTPUT_BIT_RATE, nullptr));
}
void H264EncoderImpl::close_output_file()
file_mux.reset();
}
+void H264EncoderImpl::open_output_stream()
+{
+ AVFormatContext *avctx = avformat_alloc_context();
+ AVOutputFormat *oformat = av_guess_format(global_flags.stream_mux_name.c_str(), nullptr, nullptr);
+ assert(oformat != nullptr);
+ avctx->oformat = oformat;
+
+ string codec_name;
+ int bit_rate;
+
+ if (global_flags.stream_audio_codec_name.empty()) {
+ codec_name = AUDIO_OUTPUT_CODEC_NAME;
+ bit_rate = DEFAULT_AUDIO_OUTPUT_BIT_RATE;
+ } else {
+ codec_name = global_flags.stream_audio_codec_name;
+ bit_rate = global_flags.stream_audio_codec_bitrate;
+ }
+
+ uint8_t *buf = (uint8_t *)av_malloc(MUX_BUFFER_SIZE);
+ avctx->pb = avio_alloc_context(buf, MUX_BUFFER_SIZE, 1, this, nullptr, &H264EncoderImpl::write_packet_thunk, nullptr);
+
+ Mux::Codec video_codec;
+ if (global_flags.uncompressed_video_to_http) {
+ video_codec = Mux::CODEC_NV12;
+ } else {
+ video_codec = Mux::CODEC_H264;
+ }
+
+ avctx->flags = AVFMT_FLAG_CUSTOM_IO;
+ 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);
+ }
+
+ int time_base = global_flags.stream_coarse_timebase ? COARSE_TIMEBASE : TIMEBASE;
+ stream_mux_writing_header = true;
+ stream_mux.reset(new Mux(avctx, frame_width, frame_height, video_codec, codec_audio, time_base, bit_rate, this));
+ stream_mux_writing_header = false;
+ httpd->set_header(stream_mux_header);
+ stream_mux_header.clear();
+}
+
+void H264EncoderImpl::close_output_stream()
+{
+ stream_mux.reset();
+}
+
+int H264EncoderImpl::write_packet_thunk(void *opaque, uint8_t *buf, int buf_size)
+{
+ H264EncoderImpl *h264_encoder = (H264EncoderImpl *)opaque;
+ return h264_encoder->write_packet(buf, buf_size);
+}
+
+int H264EncoderImpl::write_packet(uint8_t *buf, int buf_size)
+{
+ if (stream_mux_writing_header) {
+ stream_mux_header.append((char *)buf, buf_size);
+ } else {
+ httpd->add_data((char *)buf, buf_size, stream_mux_writing_keyframes);
+ stream_mux_writing_keyframes = false;
+ }
+ return buf_size;
+}
+
void H264EncoderImpl::encode_thread_func()
{
int64_t last_dts = -1;
last_dts = dts;
}
- if (global_flags.uncompressed_video_to_http) {
+ if (global_flags.uncompressed_video_to_http ||
+ global_flags.x264_video_to_http) {
// Add frames left in reorderer.
while (!reorderer->empty()) {
pair<int64_t, const uint8_t *> output_frame = reorderer->get_first_frame();
- add_packet_for_uncompressed_frame(output_frame.first, output_frame.second);
+ if (global_flags.uncompressed_video_to_http) {
+ add_packet_for_uncompressed_frame(output_frame.first, output_frame.second);
+ } else {
+ assert(global_flags.x264_video_to_http);
+ x264_encoder->add_frame(output_frame.first, output_frame.second);
+ }
}
}
}
pkt.size = frame_width * frame_height * 2;
pkt.stream_index = 0;
pkt.flags = AV_PKT_FLAG_KEY;
- httpd->add_packet(pkt, pts, pts);
+ stream_mux->add_packet(pkt, pts, pts);
}
namespace {
va_status = vaUnmapBuffer(va_dpy, surf->surface_image.buf);
CHECK_VASTATUS(va_status, "vaUnmapBuffer");
- if (global_flags.uncompressed_video_to_http) {
+ if (global_flags.uncompressed_video_to_http ||
+ global_flags.x264_video_to_http) {
// Add uncompressed video. (Note that pts == dts here.)
// 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);
+ if (global_flags.uncompressed_video_to_http) {
+ add_packet_for_uncompressed_frame(output_frame.first, output_frame.second);
+ } else {
+ assert(global_flags.x264_video_to_http);
+ x264_encoder->add_frame(output_frame.first, output_frame.second);
+ }
}
}
}