X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=mixer.cpp;h=42a02521258e93f34c71518c4de0917135c2fad4;hb=6800acad34e7b16d07da9d2dc7f2ecb93ee24b23;hp=a91b3d8cc843c556dd376946eb05be9988e78e3a;hpb=0d7183f398856c9331e58808d0374e63593334b6;p=nageru diff --git a/mixer.cpp b/mixer.cpp index a91b3d8..42a0252 100644 --- a/mixer.cpp +++ b/mixer.cpp @@ -1,5 +1,3 @@ -#define WIDTH 1280 -#define HEIGHT 720 #define EXTRAHEIGHT 30 #undef Success @@ -66,7 +64,7 @@ void convert_fixed24_to_fp32(float *dst, size_t out_channels, const uint8_t *src } // namespace Mixer::Mixer(const QSurfaceFormat &format, unsigned num_cards) - : httpd("test.ts", WIDTH, HEIGHT), + : httpd(LOCAL_DUMP_FILE_NAME, WIDTH, HEIGHT), num_cards(num_cards), mixer_surface(create_surface(format)), h264_encoder_surface(create_surface(format)), @@ -124,7 +122,7 @@ Mixer::Mixer(const QSurfaceFormat &format, unsigned num_cards) [this]{ resource_pool->clean_context(); }); - card->resampler.reset(new Resampler(OUTPUT_FREQUENCY, OUTPUT_FREQUENCY, 2)); + card->resampling_queue.reset(new ResamplingQueue(OUTPUT_FREQUENCY, OUTPUT_FREQUENCY, 2)); card->usb->configure_card(); } @@ -153,6 +151,12 @@ Mixer::Mixer(const QSurfaceFormat &format, unsigned num_cards) r128.integr_start(); locut.init(FILTER_HPF, 2); + + // hlen=16 is pretty low quality, but we use quite a bit of CPU otherwise, + // and there's a limit to how important the peak meter is. + peak_resampler.setup(OUTPUT_FREQUENCY, OUTPUT_FREQUENCY * 4, /*num_channels=*/2, /*hlen=*/16); + + alsa.reset(new ALSAOutput(OUTPUT_FREQUENCY, /*num_channels=*/2)); } Mixer::~Mixer() @@ -168,6 +172,8 @@ Mixer::~Mixer() } cards[card_index].usb->stop_dequeue_thread(); } + + h264_encoder.reset(nullptr); } namespace { @@ -182,10 +188,10 @@ int unwrap_timecode(uint16_t current_wrapped, int last) } } -float find_peak(const vector &samples) +float find_peak(const float *samples, size_t num_samples) { float m = fabs(samples[0]); - for (size_t i = 1; i < samples.size(); ++i) { + for (size_t i = 1; i < num_samples; ++i) { m = std::max(m, fabs(samples[i])); } return m; @@ -249,7 +255,7 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, if (dropped_frames > FPS * 2) { fprintf(stderr, "Card %d lost more than two seconds (or time code jumping around), resetting resampler\n", card_index); - card->resampler.reset(new Resampler(OUTPUT_FREQUENCY, OUTPUT_FREQUENCY, 2)); + card->resampling_queue.reset(new ResamplingQueue(OUTPUT_FREQUENCY, OUTPUT_FREQUENCY, 2)); } else if (dropped_frames > 0) { // Insert silence as needed. fprintf(stderr, "Card %d dropped %d frame(s) (before timecode 0x%04x), inserting silence.\n", @@ -257,10 +263,10 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, vector silence; silence.resize((OUTPUT_FREQUENCY / FPS) * 2); for (int i = 0; i < dropped_frames; ++i) { - card->resampler->add_input_samples((unwrapped_timecode - dropped_frames + i) / double(FPS), silence.data(), (OUTPUT_FREQUENCY / FPS)); + card->resampling_queue->add_input_samples((unwrapped_timecode - dropped_frames + i) / double(FPS), silence.data(), (OUTPUT_FREQUENCY / FPS)); } } - card->resampler->add_input_samples(unwrapped_timecode / double(FPS), audio.data(), num_samples); + card->resampling_queue->add_input_samples(unwrapped_timecode / double(FPS), audio.data(), num_samples); } // Done with the audio, so release it. @@ -371,7 +377,12 @@ void Mixer::thread_func() // Resample the audio as needed, including from previously dropped frames. for (unsigned frame_num = 0; frame_num < card_copy[0].dropped_frames + 1; ++frame_num) { - process_audio_one_frame(); + { + // Signal to the audio thread to process this frame. + unique_lock lock(audio_mutex); + audio_pts_queue.push(pts_int); + audio_pts_queue_changed.notify_one(); + } if (frame_num != card_copy[0].dropped_frames) { // For dropped frames, increase the pts. ++dropped_frames; @@ -466,7 +477,7 @@ void Mixer::thread_func() for (unsigned card_index = 0; card_index < num_cards; ++card_index) { input_frames.push_back(bmusb_current_rendering_frame[card_index]); } - const int64_t av_delay = TIMEBASE / 10; // Corresponds to the fixed delay in resampler.h. TODO: Make less hard-coded. + const int64_t av_delay = TIMEBASE / 10; // Corresponds to the fixed delay in resampling_queue.h. TODO: Make less hard-coded. h264_encoder->end_frame(fence, pts_int + av_delay, input_frames); ++frame; pts_int += TIMEBASE / FPS; @@ -525,7 +536,23 @@ void Mixer::thread_func() resource_pool->clean_context(); } -void Mixer::process_audio_one_frame() +void Mixer::audio_thread_func() +{ + while (!should_quit) { + int64_t frame_pts_int; + + { + unique_lock lock(audio_mutex); + audio_pts_queue_changed.wait(lock, [this]{ return !audio_pts_queue.empty(); }); + frame_pts_int = audio_pts_queue.front(); + audio_pts_queue.pop(); + } + + process_audio_one_frame(frame_pts_int); + } +} + +void Mixer::process_audio_one_frame(int64_t frame_pts_int) { vector samples_card; vector samples_out; @@ -533,7 +560,7 @@ void Mixer::process_audio_one_frame() samples_card.resize((OUTPUT_FREQUENCY / FPS) * 2); { unique_lock lock(cards[card_index].audio_mutex); - if (!cards[card_index].resampler->get_output_samples(pts(), &samples_card[0], OUTPUT_FREQUENCY / FPS)) { + if (!cards[card_index].resampling_queue->get_output_samples(double(frame_pts_int) / TIMEBASE, &samples_card[0], OUTPUT_FREQUENCY / FPS)) { printf("Card %d reported previous underrun.\n", card_index); } } @@ -543,9 +570,10 @@ void Mixer::process_audio_one_frame() } } - // Cut away everything under 150 Hz; we don't need it for voice, - // and it will reduce headroom and confuse the compressor. - // (In particular, any hums at 50 or 60 Hz should be dampened.) + // Cut away everything under 120 Hz (or whatever the cutoff is); + // we don't need it for voice, and it will reduce headroom + // and confuse the compressor. (In particular, any hums at 50 or 60 Hz + // should be dampened.) locut.render(samples_out.data(), samples_out.size() / 2, locut_cutoff_hz * 2.0 * M_PI / OUTPUT_FREQUENCY, 0.5f); // Apply a level compressor to get the general level right. @@ -574,39 +602,58 @@ void Mixer::process_audio_one_frame() // float limiter_att, compressor_att; - // Then a limiter at +0 dB (so, -14 dBFS) to take out the worst peaks only. - { - float threshold = pow(10.0f, (ref_level_dbfs + 0.0f) / 20.0f); // +0 dB. - float ratio = 1000.0f; // Infinity. - float attack_time = 0.001f; - float release_time = 0.005f; - float makeup_gain = 1.0f; // 0 dB. - limiter.process(samples_out.data(), samples_out.size() / 2, threshold, ratio, attack_time, release_time, makeup_gain); -// limiter_att = limiter.get_attenuation(); - } - - // Finally, the real compressor. - { - float threshold = pow(10.0f, (ref_level_dbfs - 12.0f) / 20.0f); // -12 dB. + // The real compressor. + if (compressor_enabled) { + float threshold = pow(10.0f, compressor_threshold_dbfs / 20.0f); float ratio = 20.0f; float attack_time = 0.005f; float release_time = 0.040f; - float makeup_gain = 2.0f; // +3 dB. + float makeup_gain = 2.0f; // +6 dB. compressor.process(samples_out.data(), samples_out.size() / 2, threshold, ratio, attack_time, release_time, makeup_gain); // compressor_att = compressor.get_attenuation(); } + // Finally a limiter at -4 dB (so, -10 dBFS) to take out the worst peaks only. + // Note that since ratio is not infinite, we could go slightly higher than this. + if (limiter_enabled) { + float threshold = pow(10.0f, limiter_threshold_dbfs / 20.0f); + float ratio = 30.0f; + float attack_time = 0.0f; // Instant. + float release_time = 0.020f; + float makeup_gain = 1.0f; // 0 dB. + limiter.process(samples_out.data(), samples_out.size() / 2, threshold, ratio, attack_time, release_time, makeup_gain); +// limiter_att = limiter.get_attenuation(); + } + // printf("limiter=%+5.1f compressor=%+5.1f\n", 20.0*log10(limiter_att), 20.0*log10(compressor_att)); - // Find peak and R128 levels. - peak = max(peak, find_peak(samples_out)); + // Upsample 4x to find interpolated peak. + peak_resampler.inp_data = samples_out.data(); + peak_resampler.inp_count = samples_out.size() / 2; + + vector interpolated_samples_out; + interpolated_samples_out.resize(samples_out.size()); + while (peak_resampler.inp_count > 0) { // About four iterations. + peak_resampler.out_data = &interpolated_samples_out[0]; + peak_resampler.out_count = interpolated_samples_out.size() / 2; + peak_resampler.process(); + size_t out_stereo_samples = interpolated_samples_out.size() / 2 - peak_resampler.out_count; + peak = max(peak, find_peak(interpolated_samples_out.data(), out_stereo_samples * 2)); + } + + // Find R128 levels. vector left, right; deinterleave_samples(samples_out, &left, &right); float *ptrs[] = { left.data(), right.data() }; r128.process(left.size(), ptrs); - // Actually add the samples to the output. - h264_encoder->add_audio(pts_int, move(samples_out)); + // Send the samples to the sound card. + if (alsa) { + alsa->write(samples_out); + } + + // And finally add them to the output. + h264_encoder->add_audio(frame_pts_int, move(samples_out)); } void Mixer::subsample_chroma(GLuint src_tex, GLuint dst_tex) @@ -676,12 +723,14 @@ void Mixer::release_display_frame(DisplayFrame *frame) void Mixer::start() { mixer_thread = thread(&Mixer::thread_func, this); + audio_thread = thread(&Mixer::audio_thread_func, this); } void Mixer::quit() { should_quit = true; mixer_thread.join(); + audio_thread.join(); } void Mixer::transition_clicked(int transition_num) @@ -696,6 +745,7 @@ void Mixer::channel_clicked(int preview_num) void Mixer::reset_meters() { + peak_resampler.reset(); peak = 0.0f; r128.reset(); r128.integr_start();