X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=mixer.cpp;h=1d283e374041ea16da2c880da0584ef5fa614461;hb=a18cfe79e996811cba2acc3d04cbf26120c39c13;hp=9eed47ea8f33414f97e6ec3482f164b7c99da5be;hpb=320956d2f4d88d0a682bada136aaf7d5f7cbdace;p=nageru diff --git a/mixer.cpp b/mixer.cpp index 9eed47e..1d283e3 100644 --- a/mixer.cpp +++ b/mixer.cpp @@ -3,7 +3,6 @@ #include "mixer.h" #include -#include #include #include #include @@ -11,38 +10,44 @@ #include #include #include -#include +#include #include #include #include #include -#include -#include #include #include -#include #include #include +#include #include #include +#include #include #include #include #include +#include "DeckLinkAPI.h" +#include "LinuxCOM.h" +#include "alsa_output.h" #include "bmusb/bmusb.h" #include "bmusb/fake_capture.h" +#include "chroma_subsampler.h" #include "context.h" -#include "db.h" #include "decklink_capture.h" +#include "decklink_output.h" #include "defs.h" #include "disk_space_estimator.h" #include "flags.h" +#include "input_mapping.h" #include "pbo_frame_allocator.h" #include "ref_counted_gl_sync.h" +#include "resampling_queue.h" #include "timebase.h" #include "video_encoder.h" +class IDeckLink; class QOpenGLContext; using namespace movit; @@ -73,25 +78,25 @@ void insert_new_frame(RefCountedFrame frame, unsigned field_num, bool interlaced } // namespace -void QueueLengthPolicy::update_policy(int queue_length) +void QueueLengthPolicy::update_policy(unsigned queue_length) { - if (queue_length < 0) { // Starvation. - if (been_at_safe_point_since_last_starvation && safe_queue_length < 5) { + if (queue_length == 0) { // Starvation. + if (been_at_safe_point_since_last_starvation && safe_queue_length < unsigned(global_flags.max_input_queue_frames)) { ++safe_queue_length; - fprintf(stderr, "Card %u: Starvation, increasing safe limit to %u frames\n", + fprintf(stderr, "Card %u: Starvation, increasing safe limit to %u frame(s)\n", card_index, safe_queue_length); } frames_with_at_least_one = 0; been_at_safe_point_since_last_starvation = false; return; } - if (queue_length > 0) { - if (queue_length >= int(safe_queue_length)) { + if (queue_length >= 1) { + if (queue_length >= safe_queue_length) { been_at_safe_point_since_last_starvation = true; } - if (++frames_with_at_least_one >= 1000 && safe_queue_length > 0) { + if (++frames_with_at_least_one >= 1000 && safe_queue_length > 1) { --safe_queue_length; - fprintf(stderr, "Card %u: Spare frames for more than 1000 frames, reducing safe limit to %u frames\n", + fprintf(stderr, "Card %u: Spare frames for more than 1000 frames, reducing safe limit to %u frame(s)\n", card_index, safe_queue_length); frames_with_at_least_one = 0; } @@ -105,8 +110,8 @@ Mixer::Mixer(const QSurfaceFormat &format, unsigned num_cards) num_cards(num_cards), mixer_surface(create_surface(format)), h264_encoder_surface(create_surface(format)), - audio_mixer(num_cards), - correlation(OUTPUT_FREQUENCY) + decklink_output_surface(create_surface(format)), + audio_mixer(num_cards) { CHECK(init_movit(MOVIT_SHADER_DIR, MOVIT_DEBUG_OFF)); check_error(); @@ -127,15 +132,15 @@ Mixer::Mixer(const QSurfaceFormat &format, unsigned num_cards) inout_format.gamma_curve = GAMMA_sRGB; // Display chain; shows the live output produced by the main chain (its RGBA version). - display_chain.reset(new EffectChain(WIDTH, HEIGHT, resource_pool.get())); + display_chain.reset(new EffectChain(global_flags.width, global_flags.height, resource_pool.get())); check_error(); - display_input = new FlatInput(inout_format, FORMAT_RGB, GL_UNSIGNED_BYTE, WIDTH, HEIGHT); // FIXME: GL_UNSIGNED_BYTE is really wrong. + display_input = new FlatInput(inout_format, FORMAT_RGB, GL_UNSIGNED_BYTE, global_flags.width, global_flags.height); // FIXME: GL_UNSIGNED_BYTE is really wrong. display_chain->add_input(display_input); display_chain->add_output(inout_format, OUTPUT_ALPHA_FORMAT_POSTMULTIPLIED); display_chain->set_dither_bits(0); // Don't bother. display_chain->finalize(); - video_encoder.reset(new VideoEncoder(resource_pool.get(), h264_encoder_surface, global_flags.va_display, WIDTH, HEIGHT, &httpd, global_disk_space_estimator)); + video_encoder.reset(new VideoEncoder(resource_pool.get(), h264_encoder_surface, global_flags.va_display, global_flags.width, global_flags.height, &httpd, global_disk_space_estimator)); // Start listening for clients only once VideoEncoder has written its header, if any. httpd.start(9095); @@ -154,7 +159,10 @@ Mixer::Mixer(const QSurfaceFormat &format, unsigned num_cards) break; } - configure_card(card_index, new DeckLinkCapture(decklink, card_index), /*is_fake_capture=*/false); + DeckLinkCapture *capture = new DeckLinkCapture(decklink, card_index); + DeckLinkOutput *output = new DeckLinkOutput(resource_pool.get(), decklink_output_surface, global_flags.width, global_flags.height, card_index); + output->set_device(decklink); + configure_card(card_index, capture, /*is_fake_capture=*/false, output); ++num_pci_devices; } decklink_iterator->Release(); @@ -163,18 +171,19 @@ Mixer::Mixer(const QSurfaceFormat &format, unsigned num_cards) fprintf(stderr, "DeckLink drivers not found. Probing for USB cards only.\n"); } } + unsigned num_usb_devices = BMUSBCapture::num_cards(); for (unsigned usb_card_index = 0; usb_card_index < num_usb_devices && card_index < num_cards; ++usb_card_index, ++card_index) { BMUSBCapture *capture = new BMUSBCapture(usb_card_index); capture->set_card_disconnected_callback(bind(&Mixer::bm_hotplug_remove, this, card_index)); - configure_card(card_index, capture, /*is_fake_capture=*/false); + configure_card(card_index, capture, /*is_fake_capture=*/false, /*output=*/nullptr); } fprintf(stderr, "Found %u USB card(s).\n", num_usb_devices); unsigned num_fake_cards = 0; for ( ; card_index < num_cards; ++card_index, ++num_fake_cards) { - FakeCapture *capture = new FakeCapture(WIDTH, HEIGHT, FAKE_FPS, OUTPUT_FREQUENCY, card_index, global_flags.fake_cards_audio); - configure_card(card_index, capture, /*is_fake_capture=*/true); + FakeCapture *capture = new FakeCapture(global_flags.width, global_flags.height, FAKE_FPS, OUTPUT_FREQUENCY, card_index, global_flags.fake_cards_audio); + configure_card(card_index, capture, /*is_fake_capture=*/true, /*output=*/nullptr); } if (num_fake_cards > 0) { @@ -184,113 +193,115 @@ Mixer::Mixer(const QSurfaceFormat &format, unsigned num_cards) BMUSBCapture::set_card_connected_callback(bind(&Mixer::bm_hotplug_add, this, _1)); BMUSBCapture::start_bm_thread(); - for (card_index = 0; card_index < num_cards; ++card_index) { + for (unsigned card_index = 0; card_index < num_cards; ++card_index) { cards[card_index].queue_length_policy.reset(card_index); - cards[card_index].capture->start_bm_capture(); } - // Set up stuff for NV12 conversion. - - // Cb/Cr shader. - string cbcr_vert_shader = - "#version 130 \n" - " \n" - "in vec2 position; \n" - "in vec2 texcoord; \n" - "out vec2 tc0; \n" - "uniform vec2 foo_chroma_offset_0; \n" - " \n" - "void main() \n" - "{ \n" - " // The result of glOrtho(0.0, 1.0, 0.0, 1.0, 0.0, 1.0) is: \n" - " // \n" - " // 2.000 0.000 0.000 -1.000 \n" - " // 0.000 2.000 0.000 -1.000 \n" - " // 0.000 0.000 -2.000 -1.000 \n" - " // 0.000 0.000 0.000 1.000 \n" - " gl_Position = vec4(2.0 * position.x - 1.0, 2.0 * position.y - 1.0, -1.0, 1.0); \n" - " vec2 flipped_tc = texcoord; \n" - " tc0 = flipped_tc + foo_chroma_offset_0; \n" - "} \n"; - string cbcr_frag_shader = - "#version 130 \n" - "in vec2 tc0; \n" - "uniform sampler2D cbcr_tex; \n" - "out vec4 FragColor; \n" - "void main() { \n" - " FragColor = texture(cbcr_tex, tc0); \n" - "} \n"; - vector frag_shader_outputs; - cbcr_program_num = resource_pool->compile_glsl_program(cbcr_vert_shader, cbcr_frag_shader, frag_shader_outputs); - - float vertices[] = { - 0.0f, 2.0f, - 0.0f, 0.0f, - 2.0f, 0.0f - }; - cbcr_vbo = generate_vbo(2, GL_FLOAT, sizeof(vertices), vertices); - cbcr_position_attribute_index = glGetAttribLocation(cbcr_program_num, "position"); - cbcr_texcoord_attribute_index = glGetAttribLocation(cbcr_program_num, "texcoord"); - - r128.init(2, OUTPUT_FREQUENCY); - r128.integr_start(); - - // 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, /*frel=*/1.0); + chroma_subsampler.reset(new ChromaSubsampler(resource_pool.get())); if (global_flags.enable_alsa_output) { alsa.reset(new ALSAOutput(OUTPUT_FREQUENCY, /*num_channels=*/2)); } + if (global_flags.output_card != -1) { + desired_output_card_index = global_flags.output_card; + set_output_card_internal(global_flags.output_card); + } } Mixer::~Mixer() { - resource_pool->release_glsl_program(cbcr_program_num); - glDeleteBuffers(1, &cbcr_vbo); BMUSBCapture::stop_bm_thread(); for (unsigned card_index = 0; card_index < num_cards; ++card_index) { { - unique_lock lock(bmusb_mutex); + unique_lock lock(card_mutex); cards[card_index].should_quit = true; // Unblock thread. cards[card_index].new_frames_changed.notify_all(); } cards[card_index].capture->stop_dequeue_thread(); + if (cards[card_index].output) { + cards[card_index].output->end_output(); + cards[card_index].output.reset(); + } } video_encoder.reset(nullptr); } -void Mixer::configure_card(unsigned card_index, CaptureInterface *capture, bool is_fake_capture) +void Mixer::configure_card(unsigned card_index, CaptureInterface *capture, bool is_fake_capture, DeckLinkOutput *output) { printf("Configuring card %d...\n", card_index); CaptureCard *card = &cards[card_index]; if (card->capture != nullptr) { card->capture->stop_dequeue_thread(); - delete card->capture; } - card->capture = capture; + card->capture.reset(capture); card->is_fake_capture = is_fake_capture; + if (card->output.get() != output) { + card->output.reset(output); + } card->capture->set_frame_callback(bind(&Mixer::bm_frame, this, card_index, _1, _2, _3, _4, _5, _6, _7)); if (card->frame_allocator == nullptr) { - card->frame_allocator.reset(new PBOFrameAllocator(8 << 20, WIDTH, HEIGHT)); // 8 MB. + card->frame_allocator.reset(new PBOFrameAllocator(8 << 20, global_flags.width, global_flags.height)); // 8 MB. } card->capture->set_video_frame_allocator(card->frame_allocator.get()); if (card->surface == nullptr) { card->surface = create_surface_with_same_format(mixer_surface); } - while (!card->new_frames.empty()) card->new_frames.pop(); - card->fractional_samples = 0; + while (!card->new_frames.empty()) card->new_frames.pop_front(); card->last_timecode = -1; card->capture->configure_card(); + // NOTE: start_bm_capture() happens in thread_func(). + DeviceSpec device{InputSourceType::CAPTURE_CARD, card_index}; audio_mixer.reset_resampler(device); - audio_mixer.set_name(device, card->capture->get_description()); + audio_mixer.set_display_name(device, card->capture->get_description()); + audio_mixer.trigger_state_changed_callback(); } +void Mixer::set_output_card_internal(int card_index) +{ + // We don't really need to take card_mutex, since we're in the mixer + // thread and don't mess with any queues (which is the only thing that happens + // from other threads), but it's probably the safest in the long run. + unique_lock lock(card_mutex); + if (output_card_index != -1) { + // Switch the old card from output to input. + CaptureCard *old_card = &cards[output_card_index]; + old_card->output->end_output(); + + // Stop the fake card that we put into place. + // This needs to _not_ happen under the mutex, to avoid deadlock + // (delivering the last frame needs to take the mutex). + bmusb::CaptureInterface *fake_capture = old_card->capture.get(); + lock.unlock(); + fake_capture->stop_dequeue_thread(); + lock.lock(); + old_card->capture = move(old_card->parked_capture); + old_card->is_fake_capture = false; + old_card->capture->start_bm_capture(); + } + if (card_index != -1) { + CaptureCard *card = &cards[card_index]; + bmusb::CaptureInterface *capture = card->capture.get(); + // TODO: DeckLinkCapture::stop_dequeue_thread can actually take + // several seconds to complete (blocking on DisableVideoInput); + // see if we can maybe do it asynchronously. + lock.unlock(); + capture->stop_dequeue_thread(); + lock.lock(); + card->parked_capture = move(card->capture); + bmusb::CaptureInterface *fake_capture = new FakeCapture(global_flags.width, global_flags.height, FAKE_FPS, OUTPUT_FREQUENCY, card_index, global_flags.fake_cards_audio); + configure_card(card_index, fake_capture, /*is_fake_capture=*/true, card->output.release()); + card->queue_length_policy.reset(card_index); + card->capture->start_bm_capture(); + desired_output_video_mode = output_video_mode = card->output->pick_video_mode(desired_output_video_mode); + card->output->start_output(desired_output_video_mode, pts_int); + } + output_card_index = card_index; +} namespace { @@ -304,30 +315,6 @@ int unwrap_timecode(uint16_t current_wrapped, int last) } } -float find_peak(const float *samples, size_t num_samples) -{ - float m = fabs(samples[0]); - for (size_t i = 1; i < num_samples; ++i) { - m = max(m, fabs(samples[i])); - } - return m; -} - -void deinterleave_samples(const vector &in, vector *out_l, vector *out_r) -{ - size_t num_samples = in.size() / 2; - out_l->resize(num_samples); - out_r->resize(num_samples); - - const float *inptr = in.data(); - float *lptr = &(*out_l)[0]; - float *rptr = &(*out_r)[0]; - for (size_t i = 0; i < num_samples; ++i) { - *lptr++ = *inptr++; - *rptr++ = *inptr++; - } -} - } // namespace void Mixer::bm_frame(unsigned card_index, uint16_t timecode, @@ -342,7 +329,7 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, // Found a stable signal, so stop scanning. is_mode_scanning[card_index] = false; } else { - static constexpr double switch_time_s = 0.5; // Should be enough time for the signal to stabilize. + static constexpr double switch_time_s = 0.1; // Should be enough time for the signal to stabilize. steady_clock::time_point now = steady_clock::now(); double sec_since_last_switch = duration(steady_clock::now() - last_mode_scan_change[card_index]).count(); if (sec_since_last_switch > switch_time_s) { @@ -391,10 +378,13 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, fprintf(stderr, "Card %d dropped %d frame(s) (before timecode 0x%04x), inserting silence.\n", card_index, dropped_frames, timecode); - audio_mixer.add_silence(device, silence_samples, dropped_frames, frame_length); + bool success; + do { + success = audio_mixer.add_silence(device, silence_samples, dropped_frames, frame_length); + } while (!success); } - audio_mixer.add_audio(device, audio_frame.data + audio_offset, num_samples, audio_format, frame_length); + audio_mixer.add_audio(device, audio_frame.data + audio_offset, num_samples, audio_format, frame_length, audio_frame.received_timestamp); // Done with the audio, so release it. if (audio_frame.owner) { @@ -417,13 +407,13 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, // Still send on the information that we _had_ a frame, even though it's corrupted, // so that pts can go up accordingly. { - unique_lock lock(bmusb_mutex); + unique_lock lock(card_mutex); CaptureCard::NewFrame new_frame; new_frame.frame = RefCountedFrame(FrameAllocator::Frame()); new_frame.length = frame_length; new_frame.interlaced = false; new_frame.dropped_frames = dropped_frames; - card->new_frames.push(move(new_frame)); + card->new_frames.push_back(move(new_frame)); card->new_frames_changed.notify_all(); } return; @@ -433,6 +423,7 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, unsigned num_fields = video_format.interlaced ? 2 : 1; steady_clock::time_point frame_upload_start; + bool interlaced_stride = false; if (video_format.interlaced) { // Send the two fields along as separate frames; the other side will need to add // a deinterlacer to actually get this right. @@ -441,6 +432,9 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, assert(frame_length % 2 == 0); frame_length /= 2; num_fields = 2; + if (video_format.second_field_start == 1) { + interlaced_stride = true; + } frame_upload_start = steady_clock::now(); } userdata->last_interlaced = video_format.interlaced; @@ -464,8 +458,13 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, // Note that this means we must hold on to the actual frame data in // until the upload command is run, but we hold on to much longer than that // (in fact, all the way until we no longer use the texture in rendering). - auto upload_func = [field, video_format, y_offset, cbcr_offset, cbcr_width, userdata]() { - unsigned field_start_line = (field == 1) ? video_format.second_field_start : video_format.extra_lines_top + field * (video_format.height + 22); + auto upload_func = [field, video_format, y_offset, cbcr_offset, cbcr_width, interlaced_stride, userdata]() { + unsigned field_start_line; + if (field == 1) { + field_start_line = video_format.second_field_start; + } else { + field_start_line = video_format.extra_lines_top; + } if (userdata->tex_y[field] == 0 || userdata->tex_cbcr[field] == 0 || @@ -503,16 +502,32 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, glBindTexture(GL_TEXTURE_2D, userdata->tex_cbcr[field]); check_error(); + if (interlaced_stride) { + glPixelStorei(GL_UNPACK_ROW_LENGTH, cbcr_width * 2); + check_error(); + } else { + glPixelStorei(GL_UNPACK_ROW_LENGTH, 0); + check_error(); + } glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, cbcr_width, video_format.height, GL_RG, GL_UNSIGNED_BYTE, BUFFER_OFFSET(field_cbcr_start)); check_error(); glBindTexture(GL_TEXTURE_2D, userdata->tex_y[field]); check_error(); + if (interlaced_stride) { + glPixelStorei(GL_UNPACK_ROW_LENGTH, video_format.width * 2); + check_error(); + } else { + glPixelStorei(GL_UNPACK_ROW_LENGTH, 0); + check_error(); + } glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, video_format.width, video_format.height, GL_RED, GL_UNSIGNED_BYTE, BUFFER_OFFSET(field_y_start)); check_error(); glBindTexture(GL_TEXTURE_2D, 0); check_error(); glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0); check_error(); + glPixelStorei(GL_UNPACK_ROW_LENGTH, 0); + check_error(); }; if (field == 1) { @@ -528,7 +543,7 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, } { - unique_lock lock(bmusb_mutex); + unique_lock lock(card_mutex); CaptureCard::NewFrame new_frame; new_frame.frame = frame; new_frame.length = frame_length; @@ -536,7 +551,8 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, new_frame.interlaced = video_format.interlaced; new_frame.upload_func = upload_func; new_frame.dropped_frames = dropped_frames; - card->new_frames.push(move(new_frame)); + new_frame.received_timestamp = video_frame.received_timestamp; // Ignore the audio timestamp. + card->new_frames.push_back(move(new_frame)); card->new_frames_changed.notify_all(); } } @@ -555,6 +571,8 @@ void Mixer::bm_hotplug_remove(unsigned card_index) void Mixer::thread_func() { + pthread_setname_np(pthread_self(), "Mixer_OpenGL"); + eglBindAPI(EGL_OPENGL_API); QOpenGLContext *context = create_context(mixer_surface); if (!make_current(context, mixer_surface)) { @@ -562,6 +580,14 @@ void Mixer::thread_func() exit(1); } + // Start the actual capture. (We don't want to do it before we're actually ready + // to process output frames.) + for (unsigned card_index = 0; card_index < num_cards; ++card_index) { + if (int(card_index) != output_card_index) { + cards[card_index].capture->start_bm_capture(); + } + } + steady_clock::time_point start, now; start = steady_clock::now(); @@ -569,17 +595,34 @@ void Mixer::thread_func() int stats_dropped_frames = 0; while (!should_quit) { + if (desired_output_card_index != output_card_index) { + set_output_card_internal(desired_output_card_index); + } + if (output_card_index != -1 && + desired_output_video_mode != output_video_mode) { + DeckLinkOutput *output = cards[output_card_index].output.get(); + output->end_output(); + desired_output_video_mode = output_video_mode = output->pick_video_mode(desired_output_video_mode); + output->start_output(desired_output_video_mode, pts_int); + } + CaptureCard::NewFrame new_frames[MAX_VIDEO_CARDS]; bool has_new_frame[MAX_VIDEO_CARDS] = { false }; - int num_samples[MAX_VIDEO_CARDS] = { 0 }; - unsigned master_card_index = theme->map_signal(master_clock_channel); - assert(master_card_index < num_cards); + bool master_card_is_output; + unsigned master_card_index; + if (output_card_index != -1) { + master_card_is_output = true; + master_card_index = output_card_index; + } else { + master_card_is_output = false; + master_card_index = theme->map_signal(master_clock_channel); + assert(master_card_index < num_cards); + } - get_one_frame_from_each_card(master_card_index, new_frames, has_new_frame, num_samples); - schedule_audio_resampling_tasks(new_frames[master_card_index].dropped_frames, num_samples[master_card_index], new_frames[master_card_index].length); - stats_dropped_frames += new_frames[master_card_index].dropped_frames; - send_audio_level_callback(); + OutputFrameInfo output_frame_info = get_one_frame_from_each_card(master_card_index, master_card_is_output, new_frames, has_new_frame); + schedule_audio_resampling_tasks(output_frame_info.dropped_frames, output_frame_info.num_samples, output_frame_info.frame_duration, output_frame_info.is_preroll, output_frame_info.frame_timestamp); + stats_dropped_frames += output_frame_info.dropped_frames; handle_hotplugged_cards(); @@ -598,7 +641,7 @@ void Mixer::thread_func() // If the first card is reporting a corrupted or otherwise dropped frame, // just increase the pts (skipping over this frame) and don't try to compute anything new. - if (new_frames[master_card_index].frame->len == 0) { + if (!master_card_is_output && new_frames[master_card_index].frame->len == 0) { ++stats_dropped_frames; pts_int += new_frames[master_card_index].length; continue; @@ -620,7 +663,7 @@ void Mixer::thread_func() } } - int64_t frame_duration = new_frames[master_card_index].length; + int64_t frame_duration = output_frame_info.frame_duration; render_one_frame(frame_duration); ++frame; pts_int += frame_duration; @@ -648,10 +691,15 @@ void Mixer::thread_func() assert(false); } - printf(", using %ld / %ld MB lockable memory (%.1f%%)", - long(used.ru_maxrss / 1024), - long(limit.rlim_cur / 1048576), - float(100.0 * (used.ru_maxrss * 1024.0) / limit.rlim_cur)); + if (limit.rlim_cur == 0) { + printf(", using %ld MB memory (locked)", + long(used.ru_maxrss / 1024)); + } else { + printf(", using %ld / %ld MB lockable memory (%.1f%%)", + long(used.ru_maxrss / 1024), + long(limit.rlim_cur / 1048576), + float(100.0 * (used.ru_maxrss * 1024.0) / limit.rlim_cur)); + } } else { printf(", using %ld MB memory (not locked)", long(used.ru_maxrss / 1024)); @@ -681,15 +729,77 @@ void Mixer::thread_func() resource_pool->clean_context(); } -void Mixer::get_one_frame_from_each_card(unsigned master_card_index, CaptureCard::NewFrame new_frames[MAX_VIDEO_CARDS], bool has_new_frame[MAX_VIDEO_CARDS], int num_samples[MAX_VIDEO_CARDS]) +bool Mixer::input_card_is_master_clock(unsigned card_index, unsigned master_card_index) const +{ + if (output_card_index != -1) { + // The output card (ie., cards[output_card_index].output) is the master clock, + // so no input card (ie., cards[card_index].capture) is. + return false; + } + return (card_index == master_card_index); +} + +void Mixer::trim_queue(CaptureCard *card, unsigned card_index) +{ + // Count the number of frames in the queue, including any frames + // we dropped. It's hard to know exactly how we should deal with + // dropped (corrupted) input frames; they don't help our goal of + // avoiding starvation, but they still add to the problem of latency. + // Since dropped frames is going to mean a bump in the signal anyway, + // we err on the side of having more stable latency instead. + unsigned queue_length = 0; + for (const CaptureCard::NewFrame &frame : card->new_frames) { + queue_length += frame.dropped_frames + 1; + } + card->queue_length_policy.update_policy(queue_length); + + // If needed, drop frames until the queue is below the safe limit. + // We prefer to drop from the head, because all else being equal, + // we'd like more recent frames (less latency). + unsigned dropped_frames = 0; + while (queue_length > card->queue_length_policy.get_safe_queue_length()) { + assert(!card->new_frames.empty()); + assert(queue_length > card->new_frames.front().dropped_frames); + queue_length -= card->new_frames.front().dropped_frames; + + if (queue_length <= card->queue_length_policy.get_safe_queue_length()) { + // No need to drop anything. + break; + } + + card->new_frames.pop_front(); + card->new_frames_changed.notify_all(); + --queue_length; + ++dropped_frames; + } + + if (dropped_frames > 0) { + fprintf(stderr, "Card %u dropped %u frame(s) to keep latency down.\n", + card_index, dropped_frames); + } +} + + +Mixer::OutputFrameInfo Mixer::get_one_frame_from_each_card(unsigned master_card_index, bool master_card_is_output, CaptureCard::NewFrame new_frames[MAX_VIDEO_CARDS], bool has_new_frame[MAX_VIDEO_CARDS]) { + OutputFrameInfo output_frame_info; start: - // The first card is the master timer, so wait for it to have a new frame. - // TODO: Add a timeout. - unique_lock lock(bmusb_mutex); - cards[master_card_index].new_frames_changed.wait(lock, [this, master_card_index]{ return !cards[master_card_index].new_frames.empty() || cards[master_card_index].capture->get_disconnected(); }); + unique_lock lock(card_mutex, defer_lock); + if (master_card_is_output) { + // Clocked to the output, so wait for it to be ready for the next frame. + cards[master_card_index].output->wait_for_frame(pts_int, &output_frame_info.dropped_frames, &output_frame_info.frame_duration, &output_frame_info.is_preroll, &output_frame_info.frame_timestamp); + lock.lock(); + } else { + // Wait for the master card to have a new frame. + // TODO: Add a timeout. + output_frame_info.is_preroll = false; + lock.lock(); + cards[master_card_index].new_frames_changed.wait(lock, [this, master_card_index]{ return !cards[master_card_index].new_frames.empty() || cards[master_card_index].capture->get_disconnected(); }); + } - if (cards[master_card_index].new_frames.empty()) { + if (master_card_is_output) { + handle_hotplugged_cards(); + } else if (cards[master_card_index].new_frames.empty()) { // We were woken up, but not due to a new frame. Deal with it // and then restart. assert(cards[master_card_index].capture->get_disconnected()); @@ -697,37 +807,43 @@ start: goto start; } + if (!master_card_is_output) { + output_frame_info.frame_timestamp = + cards[master_card_index].new_frames.front().received_timestamp; + } + for (unsigned card_index = 0; card_index < num_cards; ++card_index) { CaptureCard *card = &cards[card_index]; - if (card->new_frames.empty()) { - assert(card_index != master_card_index); - card->queue_length_policy.update_policy(-1); - continue; - } - new_frames[card_index] = move(card->new_frames.front()); - has_new_frame[card_index] = true; - card->new_frames.pop(); - card->new_frames_changed.notify_all(); - - int num_samples_times_timebase = OUTPUT_FREQUENCY * new_frames[card_index].length + card->fractional_samples; - num_samples[card_index] = num_samples_times_timebase / TIMEBASE; - card->fractional_samples = num_samples_times_timebase % TIMEBASE; - assert(num_samples[card_index] >= 0); - - if (card_index == master_card_index) { + if (input_card_is_master_clock(card_index, master_card_index)) { // We don't use the queue length policy for the master card, // but we will if it stops being the master. Thus, clear out // the policy in case we switch in the future. card->queue_length_policy.reset(card_index); + assert(!card->new_frames.empty()); } else { - // If we have excess frames compared to the policy for this card, - // drop frames from the head. - card->queue_length_policy.update_policy(card->new_frames.size()); - while (card->new_frames.size() > card->queue_length_policy.get_safe_queue_length()) { - card->new_frames.pop(); - } + trim_queue(card, card_index); + } + if (!card->new_frames.empty()) { + new_frames[card_index] = move(card->new_frames.front()); + has_new_frame[card_index] = true; + card->new_frames.pop_front(); + card->new_frames_changed.notify_all(); } } + + if (!master_card_is_output) { + output_frame_info.dropped_frames = new_frames[master_card_index].dropped_frames; + output_frame_info.frame_duration = new_frames[master_card_index].length; + } + + // This might get off by a fractional sample when changing master card + // between ones with different frame rates, but that's fine. + int num_samples_times_timebase = OUTPUT_FREQUENCY * output_frame_info.frame_duration + fractional_samples; + output_frame_info.num_samples = num_samples_times_timebase / TIMEBASE; + fractional_samples = num_samples_times_timebase % TIMEBASE; + assert(output_frame_info.num_samples >= 0); + + return output_frame_info; } void Mixer::handle_hotplugged_cards() @@ -737,8 +853,8 @@ void Mixer::handle_hotplugged_cards() CaptureCard *card = &cards[card_index]; if (card->capture->get_disconnected()) { fprintf(stderr, "Card %u went away, replacing with a fake card.\n", card_index); - FakeCapture *capture = new FakeCapture(WIDTH, HEIGHT, FAKE_FPS, OUTPUT_FREQUENCY, card_index, global_flags.fake_cards_audio); - configure_card(card_index, capture, /*is_fake_capture=*/true); + FakeCapture *capture = new FakeCapture(global_flags.width, global_flags.height, FAKE_FPS, OUTPUT_FREQUENCY, card_index, global_flags.fake_cards_audio); + configure_card(card_index, capture, /*is_fake_capture=*/true, /*output=*/nullptr); card->queue_length_policy.reset(card_index); card->capture->start_bm_capture(); } @@ -755,7 +871,7 @@ void Mixer::handle_hotplugged_cards() int free_card_index = -1; for (unsigned card_index = 0; card_index < num_cards; ++card_index) { if (cards[card_index].is_fake_capture) { - free_card_index = int(card_index); + free_card_index = card_index; break; } } @@ -768,7 +884,7 @@ void Mixer::handle_hotplugged_cards() fprintf(stderr, "New card plugged in, choosing slot %d.\n", free_card_index); CaptureCard *card = &cards[free_card_index]; BMUSBCapture *capture = new BMUSBCapture(free_card_index, new_dev); - configure_card(free_card_index, capture, /*is_fake_capture=*/false); + configure_card(free_card_index, capture, /*is_fake_capture=*/false, /*output=*/nullptr); card->queue_length_policy.reset(free_card_index); capture->set_card_disconnected_callback(bind(&Mixer::bm_hotplug_remove, this, free_card_index)); capture->start_bm_capture(); @@ -777,7 +893,7 @@ void Mixer::handle_hotplugged_cards() } -void Mixer::schedule_audio_resampling_tasks(unsigned dropped_frames, int num_samples_per_frame, int length_per_frame) +void Mixer::schedule_audio_resampling_tasks(unsigned dropped_frames, int num_samples_per_frame, int length_per_frame, bool is_preroll, steady_clock::time_point frame_timestamp) { // Resample the audio as needed, including from previously dropped frames. assert(num_cards > 0); @@ -798,8 +914,8 @@ void Mixer::schedule_audio_resampling_tasks(unsigned dropped_frames, int num_sam // since dropped frames are expected to be rare, and it might be // better to just wait until we have a slightly more normal situation). unique_lock lock(audio_mutex); - bool adjust_rate = !dropped_frame; - audio_task_queue.push(AudioTask{pts_int, num_samples_per_frame, adjust_rate}); + bool adjust_rate = !dropped_frame && !is_preroll; + audio_task_queue.push(AudioTask{pts_int, num_samples_per_frame, adjust_rate, frame_timestamp}); audio_task_queue_changed.notify_one(); } if (dropped_frame) { @@ -814,7 +930,7 @@ void Mixer::schedule_audio_resampling_tasks(unsigned dropped_frames, int num_sam void Mixer::render_one_frame(int64_t duration) { // Get the main chain from the theme, and set its state immediately. - Theme::Chain theme_main_chain = theme->get_chain(0, pts(), WIDTH, HEIGHT, input_state); + Theme::Chain theme_main_chain = theme->get_chain(0, pts(), global_flags.width, global_flags.height, input_state); EffectChain *chain = theme_main_chain.chain; theme_main_chain.setup_chain(); //theme_main_chain.chain->enable_phase_timing(true); @@ -824,14 +940,17 @@ void Mixer::render_one_frame(int64_t duration) assert(got_frame); // Render main chain. - GLuint cbcr_full_tex = resource_pool->create_2d_texture(GL_RG8, WIDTH, HEIGHT); - GLuint rgba_tex = resource_pool->create_2d_texture(GL_RGB565, WIDTH, HEIGHT); // Saves texture bandwidth, although dithering gets messed up. + GLuint cbcr_full_tex = resource_pool->create_2d_texture(GL_RG8, global_flags.width, global_flags.height); + GLuint rgba_tex = resource_pool->create_2d_texture(GL_RGB565, global_flags.width, global_flags.height); // Saves texture bandwidth, although dithering gets messed up. GLuint fbo = resource_pool->create_fbo(y_tex, cbcr_full_tex, rgba_tex); check_error(); - chain->render_to_fbo(fbo, WIDTH, HEIGHT); + chain->render_to_fbo(fbo, global_flags.width, global_flags.height); resource_pool->release_fbo(fbo); - subsample_chroma(cbcr_full_tex, cbcr_tex); + chroma_subsampler->subsample_chroma(cbcr_full_tex, global_flags.width, global_flags.height, cbcr_tex); + if (output_card_index != -1) { + cards[output_card_index].output->send_frame(y_tex, cbcr_full_tex, theme_main_chain.input_frames, pts_int, duration); + } resource_pool->release_2d_texture(cbcr_full_tex); // Set the right state for rgba_tex. @@ -841,7 +960,7 @@ void Mixer::render_one_frame(int64_t duration) glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE); - const int64_t av_delay = TIMEBASE / 10; // Corresponds to the fixed delay in resampling_queue.h. TODO: Make less hard-coded. + const int64_t av_delay = lrint(global_flags.audio_queue_length_ms * 0.001 * TIMEBASE); // Corresponds to the delay in ResamplingQueue. RefCountedGLsync fence = video_encoder->end_frame(pts_int + av_delay, duration, theme_main_chain.input_frames); // The live frame just shows the RGBA texture we just rendered. @@ -859,7 +978,7 @@ void Mixer::render_one_frame(int64_t duration) // Set up preview and any additional channels. for (int i = 1; i < theme->get_num_channels() + 2; ++i) { DisplayFrame display_frame; - Theme::Chain chain = theme->get_chain(i, pts(), WIDTH, HEIGHT, input_state); // FIXME: dimensions + Theme::Chain chain = theme->get_chain(i, pts(), global_flags.width, global_flags.height, input_state); // FIXME: dimensions display_frame.chain = chain.chain; display_frame.setup_chain = chain.setup_chain; display_frame.ready_fence = fence; @@ -869,27 +988,10 @@ void Mixer::render_one_frame(int64_t duration) } } -void Mixer::send_audio_level_callback() -{ - if (audio_level_callback == nullptr) { - return; - } - - unique_lock lock(audio_measure_mutex); - double loudness_s = r128.loudness_S(); - double loudness_i = r128.integrated(); - double loudness_range_low = r128.range_min(); - double loudness_range_high = r128.range_max(); - - audio_level_callback(loudness_s, to_db(peak), - loudness_i, loudness_range_low, loudness_range_high, - audio_mixer.get_gain_staging_db(), - audio_mixer.get_final_makeup_gain_db(), - correlation.get_correlation()); -} - void Mixer::audio_thread_func() { + pthread_setname_np(pthread_self(), "Mixer_Audio"); + while (!should_quit) { AudioTask task; @@ -905,110 +1007,21 @@ void Mixer::audio_thread_func() ResamplingQueue::RateAdjustmentPolicy rate_adjustment_policy = task.adjust_rate ? ResamplingQueue::ADJUST_RATE : ResamplingQueue::DO_NOT_ADJUST_RATE; - process_audio_one_frame(task.pts_int, task.num_samples, rate_adjustment_policy); - } -} - -void Mixer::process_audio_one_frame(int64_t frame_pts_int, int num_samples, ResamplingQueue::RateAdjustmentPolicy rate_adjustment_policy) -{ - vector samples_out = audio_mixer.get_output(double(frame_pts_int) / TIMEBASE, num_samples, rate_adjustment_policy); - - // 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()); - { - unique_lock lock(audio_measure_mutex); - - 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)); - peak_resampler.out_data = nullptr; + vector samples_out = audio_mixer.get_output( + task.frame_timestamp, + task.num_samples, + rate_adjustment_policy); + + // Send the samples to the sound card, then add them to the output. + if (alsa) { + alsa->write(samples_out); } + if (output_card_index != -1) { + const int64_t av_delay = lrint(global_flags.audio_queue_length_ms * 0.001 * TIMEBASE); // Corresponds to the delay in ResamplingQueue. + cards[output_card_index].output->send_audio(task.pts_int + av_delay, samples_out); + } + video_encoder->add_audio(task.pts_int, move(samples_out)); } - - // Find R128 levels and L/R correlation. - vector left, right; - deinterleave_samples(samples_out, &left, &right); - float *ptrs[] = { left.data(), right.data() }; - { - unique_lock lock(audio_measure_mutex); - r128.process(left.size(), ptrs); - audio_mixer.set_current_loudness(r128.loudness_M()); - correlation.process_samples(samples_out); - } - - // Send the samples to the sound card. - if (alsa) { - alsa->write(samples_out); - } - - // And finally add them to the output. - video_encoder->add_audio(frame_pts_int, move(samples_out)); -} - -void Mixer::subsample_chroma(GLuint src_tex, GLuint dst_tex) -{ - GLuint vao; - glGenVertexArrays(1, &vao); - check_error(); - - glBindVertexArray(vao); - check_error(); - - // Extract Cb/Cr. - GLuint fbo = resource_pool->create_fbo(dst_tex); - glBindFramebuffer(GL_FRAMEBUFFER, fbo); - glViewport(0, 0, WIDTH/2, HEIGHT/2); - check_error(); - - glUseProgram(cbcr_program_num); - check_error(); - - glActiveTexture(GL_TEXTURE0); - check_error(); - glBindTexture(GL_TEXTURE_2D, src_tex); - check_error(); - glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR); - check_error(); - glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); - check_error(); - glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE); - check_error(); - - float chroma_offset_0[] = { -0.5f / WIDTH, 0.0f }; - set_uniform_vec2(cbcr_program_num, "foo", "chroma_offset_0", chroma_offset_0); - - glBindBuffer(GL_ARRAY_BUFFER, cbcr_vbo); - check_error(); - - for (GLint attr_index : { cbcr_position_attribute_index, cbcr_texcoord_attribute_index }) { - glEnableVertexAttribArray(attr_index); - check_error(); - glVertexAttribPointer(attr_index, 2, GL_FLOAT, GL_FALSE, 0, BUFFER_OFFSET(0)); - check_error(); - } - - glDrawArrays(GL_TRIANGLES, 0, 3); - check_error(); - - for (GLint attr_index : { cbcr_position_attribute_index, cbcr_texcoord_attribute_index }) { - glDisableVertexAttribArray(attr_index); - check_error(); - } - - glUseProgram(0); - check_error(); - glBindFramebuffer(GL_FRAMEBUFFER, 0); - check_error(); - - resource_pool->release_fbo(fbo); - glDeleteVertexArrays(1, &vao); } void Mixer::release_display_frame(DisplayFrame *frame) @@ -1045,16 +1058,6 @@ void Mixer::channel_clicked(int preview_num) theme->channel_clicked(preview_num); } -void Mixer::reset_meters() -{ - unique_lock lock(audio_measure_mutex); - peak_resampler.reset(); - peak = 0.0f; - r128.reset(); - r128.integr_start(); - correlation.reset(); -} - void Mixer::start_mode_scanning(unsigned card_index) { assert(card_index < num_cards); @@ -1072,6 +1075,13 @@ void Mixer::start_mode_scanning(unsigned card_index) last_mode_scan_change[card_index] = steady_clock::now(); } +map Mixer::get_available_output_video_modes() const +{ + assert(desired_output_card_index != -1); + unique_lock lock(card_mutex); + return cards[desired_output_card_index].output->get_available_video_modes(); +} + Mixer::OutputChannel::~OutputChannel() { if (has_current_frame) {