X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=quicksync_encoder.cpp;h=7d211798a504d07e91137e77a77815732ebad3a1;hb=1462715cd71d8f61b9e53b31c34d591d150f2df3;hp=1db37bc8d1654ff97b6b672720de4128989597eb;hpb=dcbd18f44037f097b26aed93fde11392906d86ea;p=nageru diff --git a/quicksync_encoder.cpp b/quicksync_encoder.cpp index 1db37bc..7d21179 100644 --- a/quicksync_encoder.cpp +++ b/quicksync_encoder.cpp @@ -7,16 +7,6 @@ #include #include #include -extern "C" { -#include -#include -#include -#include -#include -#include -#include -#include -} #include #include #include @@ -38,10 +28,10 @@ extern "C" { #include #include +#include "audio_encoder.h" #include "context.h" #include "defs.h" #include "flags.h" -#include "httpd.h" #include "mux.h" #include "timebase.h" #include "x264_encoder.h" @@ -201,20 +191,14 @@ FrameReorderer::Frame FrameReorderer::get_first_frame() return storage; } -class QuickSyncEncoderImpl : public KeyFrameSignalReceiver { +class QuickSyncEncoderImpl { public: - QuickSyncEncoderImpl(QSurface *surface, const string &va_display, int width, int height, HTTPD *httpd); + QuickSyncEncoderImpl(const std::string &filename, QSurface *surface, const string &va_display, int width, int height, Mux *stream_mux, AudioEncoder *stream_audio_encoder); ~QuickSyncEncoderImpl(); void add_audio(int64_t pts, vector audio); bool begin_frame(GLuint *y_tex, GLuint *cbcr_tex); RefCountedGLsync end_frame(int64_t pts, int64_t duration, const vector &input_frames); void shutdown(); - 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 { @@ -234,29 +218,13 @@ private: return int64_t(ip_period - 1) * (TIMEBASE / MAX_FPS); } + void open_output_file(const std::string &filename); 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, int64_t duration, 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, int64_t duration); void storage_task_thread(); - void encode_audio(const vector &audio, - vector *audio_queue, - int64_t audio_pts, - AVCodecContext *ctx, - AVAudioResampleContext *resampler, - const vector &muxes); - void encode_audio_one_frame(const float *audio, - size_t num_samples, // In each channel. - int64_t audio_pts, - AVCodecContext *ctx, - AVAudioResampleContext *resampler, - const vector &muxes); - void encode_last_audio(vector *audio_queue, - int64_t audio_pts, - AVCodecContext *ctx, - AVAudioResampleContext *resampler, - const vector &muxes); void encode_remaining_audio(); void storage_task_enqueue(storage_task task); void save_codeddata(storage_task task); @@ -281,10 +249,6 @@ private: 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; @@ -306,30 +270,14 @@ private: map pending_video_frames; // under frame_queue_mutex map> pending_audio_frames; // under frame_queue_mutex - int64_t last_audio_pts = 0; // The first pts after all audio we've encoded. QSurface *surface; - 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 audio_queue_file; - vector audio_queue_stream; + unique_ptr file_audio_encoder; + AudioEncoder *stream_audio_encoder; - unique_ptr stream_mux; // To HTTP. + Mux* stream_mux; // To HTTP. unique_ptr file_mux; // To local disk. - // While Mux object is constructing, 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; - - AVFrame *audio_frame = nullptr; - HTTPD *httpd; unique_ptr reorderer; unique_ptr x264_encoder; // nullptr if not using x264. @@ -1702,131 +1650,13 @@ void QuickSyncEncoderImpl::save_codeddata(storage_task task) pending_audio_frames.erase(it); } - 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() }); - } - last_audio_pts = audio_pts + audio.size() * TIMEBASE / (OUTPUT_FREQUENCY * 2); + file_audio_encoder->encode_audio(audio, audio_pts + global_delay()); + stream_audio_encoder->encode_audio(audio, audio_pts + global_delay()); if (audio_pts == task.pts) break; } } -void QuickSyncEncoderImpl::encode_audio( - const vector &audio, - vector *audio_queue, - int64_t audio_pts, - AVCodecContext *ctx, - AVAudioResampleContext *resampler, - const vector &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, 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, - adjusted_audio_pts, - ctx, - resampler, - muxes); - } - audio_queue->erase(audio_queue->begin(), audio_queue->begin() + sample_num); -} - -void QuickSyncEncoderImpl::encode_audio_one_frame( - const float *audio, - size_t num_samples, - int64_t audio_pts, - AVCodecContext *ctx, - AVAudioResampleContext *resampler, - const vector &muxes) -{ - audio_frame->pts = audio_pts + global_delay(); - 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; - - 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 (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; - 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 = 0; - for (Mux *mux : muxes) { - mux->add_packet(pkt, pkt.pts, pkt.dts); - } - } - - av_freep(&audio_frame->data[0]); - - av_frame_unref(audio_frame); - av_free_packet(&pkt); -} - -void QuickSyncEncoderImpl::encode_last_audio( - vector *audio_queue, - int64_t audio_pts, - AVCodecContext *ctx, - AVAudioResampleContext *resampler, - const vector &muxes) -{ - if (!audio_queue->empty()) { - // Last frame can be whatever size we want. - assert(audio_queue->size() % 2 == 0); - encode_audio_one_frame(&(*audio_queue)[0], audio_queue->size() / 2, audio_pts, ctx, resampler, muxes); - audio_queue->clear(); - } - - if (ctx->codec->capabilities & AV_CODEC_CAP_DELAY) { - // Collect any delayed frames. - for ( ;; ) { - int got_output = 0; - AVPacket pkt; - av_init_packet(&pkt); - pkt.data = nullptr; - pkt.size = 0; - avcodec_encode_audio2(ctx, &pkt, nullptr, &got_output); - if (!got_output) break; - - pkt.stream_index = 1; - pkt.flags = 0; - for (Mux *mux : muxes) { - mux->add_packet(pkt, pkt.pts, pkt.dts); - } - av_free_packet(&pkt); - } - } -} // this is weird. but it seems to put a new frame onto the queue void QuickSyncEncoderImpl::storage_task_enqueue(storage_task task) @@ -1898,67 +1728,18 @@ int QuickSyncEncoderImpl::deinit_va() namespace { -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) { - 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 = 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 -QuickSyncEncoderImpl::QuickSyncEncoderImpl(QSurface *surface, const string &va_display, int width, int height, HTTPD *httpd) - : current_storage_frame(0), surface(surface), httpd(httpd), frame_width(width), frame_height(height) +QuickSyncEncoderImpl::QuickSyncEncoderImpl(const std::string &filename, QSurface *surface, const string &va_display, int width, int height, Mux *stream_mux, AudioEncoder *stream_audio_encoder) + : current_storage_frame(0), surface(surface), stream_audio_encoder(stream_audio_encoder), stream_mux(stream_mux), frame_width(width), frame_height(height) { - init_audio_encoder(AUDIO_OUTPUT_CODEC_NAME, DEFAULT_AUDIO_OUTPUT_BIT_RATE, &context_audio_file, &resampler_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, &resampler_audio_stream); - } + file_audio_encoder.reset(new AudioEncoder(AUDIO_OUTPUT_CODEC_NAME, DEFAULT_AUDIO_OUTPUT_BIT_RATE)); + open_output_file(filename); + file_audio_encoder->add_mux(file_mux.get()); 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 || @@ -1966,7 +1747,7 @@ QuickSyncEncoderImpl::QuickSyncEncoderImpl(QSurface *surface, const string &va_d 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())); + x264_encoder.reset(new X264Encoder(stream_mux)); } init_va(va_display); @@ -1997,12 +1778,6 @@ QuickSyncEncoderImpl::QuickSyncEncoderImpl(QSurface *surface, const string &va_d QuickSyncEncoderImpl::~QuickSyncEncoderImpl() { shutdown(); - av_frame_free(&audio_frame); - 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 QuickSyncEncoderImpl::begin_frame(GLuint *y_tex, GLuint *cbcr_tex) @@ -2156,6 +1931,7 @@ void QuickSyncEncoderImpl::shutdown() release_encode(); deinit_va(); + file_mux.reset(); is_shutdown = true; } @@ -2174,77 +1950,7 @@ void QuickSyncEncoderImpl::open_output_file(const std::string &filename) exit(1); } - 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 QuickSyncEncoderImpl::close_output_file() -{ - file_mux.reset(); -} - -void QuickSyncEncoderImpl::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, &QuickSyncEncoderImpl::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 QuickSyncEncoderImpl::close_output_stream() -{ - stream_mux.reset(); -} - -int QuickSyncEncoderImpl::write_packet_thunk(void *opaque, uint8_t *buf, int buf_size) -{ - QuickSyncEncoderImpl *h264_encoder = (QuickSyncEncoderImpl *)opaque; - return h264_encoder->write_packet(buf, buf_size); -} - -int QuickSyncEncoderImpl::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; + file_mux.reset(new Mux(avctx, frame_width, frame_height, Mux::CODEC_H264, file_audio_encoder->get_codec(), TIMEBASE, DEFAULT_AUDIO_OUTPUT_BIT_RATE, nullptr)); } void QuickSyncEncoderImpl::encode_thread_func() @@ -2334,23 +2040,16 @@ void QuickSyncEncoderImpl::encode_remaining_audio() int64_t audio_pts = pending_frame.first; vector audio = move(pending_frame.second); - 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() }); + file_audio_encoder->encode_audio(audio, audio_pts + global_delay()); + if (stream_audio_encoder) { + stream_audio_encoder->encode_audio(audio, audio_pts + global_delay()); } - last_audio_pts = audio_pts + audio.size() * TIMEBASE / (OUTPUT_FREQUENCY * 2); } pending_audio_frames.clear(); // Encode any leftover audio in the queues, and also any delayed frames. - if (context_audio_stream) { - encode_last_audio(&audio_queue_file, last_audio_pts, context_audio_file, resampler_audio_file, { file_mux.get() }); - encode_last_audio(&audio_queue_stream, last_audio_pts, context_audio_stream, resampler_audio_stream, { stream_mux.get() }); - } else { - encode_last_audio(&audio_queue_file, last_audio_pts, context_audio_file, resampler_audio_file, { stream_mux.get(), file_mux.get() }); - } + // Note: stream_audio_encoder is not owned by us, so don't call encode_last_audio(). + file_audio_encoder->encode_last_audio(); } void QuickSyncEncoderImpl::add_packet_for_uncompressed_frame(int64_t pts, int64_t duration, const uint8_t *data) @@ -2472,8 +2171,8 @@ void QuickSyncEncoderImpl::encode_frame(QuickSyncEncoderImpl::PendingFrame frame } // Proxy object. -QuickSyncEncoder::QuickSyncEncoder(QSurface *surface, const string &va_display, int width, int height, HTTPD *httpd) - : impl(new QuickSyncEncoderImpl(surface, va_display, width, height, httpd)) {} +QuickSyncEncoder::QuickSyncEncoder(const std::string &filename, QSurface *surface, const string &va_display, int width, int height, Mux *stream_mux, AudioEncoder *stream_audio_encoder) + : impl(new QuickSyncEncoderImpl(filename, surface, va_display, width, height, stream_mux, stream_audio_encoder)) {} // Must be defined here because unique_ptr<> destructor needs to know the impl. QuickSyncEncoder::~QuickSyncEncoder() {} @@ -2497,13 +2196,3 @@ void QuickSyncEncoder::shutdown() { impl->shutdown(); } - -void QuickSyncEncoder::open_output_file(const std::string &filename) -{ - impl->open_output_file(filename); -} - -void QuickSyncEncoder::close_output_file() -{ - impl->close_output_file(); -}