X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=mixer.cpp;h=6846d83ffdbaef73420d638cc1b282ce11cb02a3;hb=96cb6414f85e0ef4d660b7bd56267303e80fcd05;hp=c21ac95943bcbbaa7a6b97e4def90bf6d70770ea;hpb=142184354968cd1b1918f35772bb6f9f0b1351d6;p=nageru diff --git a/mixer.cpp b/mixer.cpp index c21ac95..6846d83 100644 --- a/mixer.cpp +++ b/mixer.cpp @@ -42,6 +42,7 @@ #include "ffmpeg_capture.h" #include "flags.h" #include "input_mapping.h" +#include "metrics.h" #include "pbo_frame_allocator.h" #include "ref_counted_gl_sync.h" #include "resampling_queue.h" @@ -79,39 +80,42 @@ void insert_new_frame(RefCountedFrame frame, unsigned field_num, bool interlaced } } -void ensure_texture_resolution(PBOFrameAllocator::Userdata *userdata, unsigned field, unsigned width, unsigned height, unsigned v210_width) +void ensure_texture_resolution(PBOFrameAllocator::Userdata *userdata, unsigned field, unsigned width, unsigned height, unsigned cbcr_width, unsigned cbcr_height, unsigned v210_width) { bool first; switch (userdata->pixel_format) { - case bmusb::PixelFormat_10BitYCbCr: + case PixelFormat_10BitYCbCr: first = userdata->tex_v210[field] == 0 || userdata->tex_444[field] == 0; break; - case bmusb::PixelFormat_8BitYCbCr: + case PixelFormat_8BitYCbCr: first = userdata->tex_y[field] == 0 || userdata->tex_cbcr[field] == 0; break; - case bmusb::PixelFormat_8BitBGRA: + case PixelFormat_8BitBGRA: first = userdata->tex_rgba[field] == 0; break; + case PixelFormat_8BitYCbCrPlanar: + first = userdata->tex_y[field] == 0 || userdata->tex_cb[field] == 0 || userdata->tex_cr[field] == 0; + break; default: assert(false); } if (first || width != userdata->last_width[field] || - height != userdata->last_height[field]) { + height != userdata->last_height[field] || + cbcr_width != userdata->last_cbcr_width[field] || + cbcr_height != userdata->last_cbcr_height[field]) { // We changed resolution since last use of this texture, so we need to create // a new object. Note that this each card has its own PBOFrameAllocator, // we don't need to worry about these flip-flopping between resolutions. switch (userdata->pixel_format) { - case bmusb::PixelFormat_10BitYCbCr: + case PixelFormat_10BitYCbCr: glBindTexture(GL_TEXTURE_2D, userdata->tex_444[field]); check_error(); glTexImage2D(GL_TEXTURE_2D, 0, GL_RGB10_A2, width, height, 0, GL_RGBA, GL_UNSIGNED_INT_2_10_10_10_REV, nullptr); check_error(); break; - case bmusb::PixelFormat_8BitYCbCr: { - size_t cbcr_width = width / 2; - + case PixelFormat_8BitYCbCr: { glBindTexture(GL_TEXTURE_2D, userdata->tex_cbcr[field]); check_error(); glTexImage2D(GL_TEXTURE_2D, 0, GL_RG8, cbcr_width, height, 0, GL_RG, GL_UNSIGNED_BYTE, nullptr); @@ -122,7 +126,22 @@ void ensure_texture_resolution(PBOFrameAllocator::Userdata *userdata, unsigned f check_error(); break; } - case bmusb::PixelFormat_8BitBGRA: + case PixelFormat_8BitYCbCrPlanar: { + glBindTexture(GL_TEXTURE_2D, userdata->tex_y[field]); + check_error(); + glTexImage2D(GL_TEXTURE_2D, 0, GL_R8, width, height, 0, GL_RED, GL_UNSIGNED_BYTE, nullptr); + check_error(); + glBindTexture(GL_TEXTURE_2D, userdata->tex_cb[field]); + check_error(); + glTexImage2D(GL_TEXTURE_2D, 0, GL_R8, cbcr_width, cbcr_height, 0, GL_RED, GL_UNSIGNED_BYTE, nullptr); + check_error(); + glBindTexture(GL_TEXTURE_2D, userdata->tex_cr[field]); + check_error(); + glTexImage2D(GL_TEXTURE_2D, 0, GL_R8, cbcr_width, cbcr_height, 0, GL_RED, GL_UNSIGNED_BYTE, nullptr); + check_error(); + break; + } + case PixelFormat_8BitBGRA: glBindTexture(GL_TEXTURE_2D, userdata->tex_rgba[field]); check_error(); if (global_flags.can_disable_srgb_decoder) { // See the comments in tweaked_inputs.h. @@ -135,6 +154,8 @@ void ensure_texture_resolution(PBOFrameAllocator::Userdata *userdata, unsigned f } userdata->last_width[field] = width; userdata->last_height[field] = height; + userdata->last_cbcr_width[field] = cbcr_width; + userdata->last_cbcr_height[field] = cbcr_height; } if (global_flags.ten_bit_input && (first || v210_width != userdata->last_v210_width[field])) { @@ -177,6 +198,13 @@ void upload_texture(GLuint tex, GLuint width, GLuint height, GLuint stride, bool } // namespace +void QueueLengthPolicy::register_metrics(const vector> &labels) +{ + global_metrics.add("input_queue_length_frames", labels, &metric_input_queue_length_frames, Metrics::TYPE_GAUGE); + global_metrics.add("input_queue_safe_length_frames", labels, &metric_input_queue_safe_length_frames, Metrics::TYPE_GAUGE); + global_metrics.add("input_queue_duped_frames", labels, &metric_input_duped_frames); +} + void QueueLengthPolicy::update_policy(unsigned queue_length) { if (queue_length == 0) { // Starvation. @@ -187,6 +215,8 @@ void QueueLengthPolicy::update_policy(unsigned queue_length) } frames_with_at_least_one = 0; been_at_safe_point_since_last_starvation = false; + ++metric_input_duped_frames; + metric_input_queue_safe_length_frames = safe_queue_length; return; } if (queue_length >= safe_queue_length) { @@ -194,6 +224,7 @@ void QueueLengthPolicy::update_policy(unsigned queue_length) } if (++frames_with_at_least_one >= 1000 && safe_queue_length > 1) { --safe_queue_length; + metric_input_queue_safe_length_frames = safe_queue_length; 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; @@ -206,6 +237,7 @@ Mixer::Mixer(const QSurfaceFormat &format, unsigned num_cards) mixer_surface(create_surface(format)), h264_encoder_surface(create_surface(format)), decklink_output_surface(create_surface(format)), + ycbcr_interpretation(global_flags.ycbcr_interpretation), audio_mixer(num_cards) { CHECK(init_movit(MOVIT_SHADER_DIR, MOVIT_DEBUG_OFF)); @@ -367,6 +399,14 @@ Mixer::Mixer(const QSurfaceFormat &format, unsigned num_cards) desired_output_card_index = global_flags.output_card; set_output_card_internal(global_flags.output_card); } + + metric_start_time_seconds = get_timestamp_for_metrics(); + + global_metrics.add("frames_output_total", &metric_frames_output_total); + global_metrics.add("frames_output_dropped", &metric_frames_output_dropped); + global_metrics.add("start_time_seconds", &metric_start_time_seconds, Metrics::TYPE_GAUGE); + global_metrics.add("memory_used_bytes", &metrics_memory_used_bytes); + global_metrics.add("metrics_memory_locked_limit_bytes", &metrics_memory_locked_limit_bytes); } Mixer::~Mixer() @@ -399,13 +439,14 @@ void Mixer::configure_card(unsigned card_index, CaptureInterface *capture, CardT } card->capture.reset(capture); card->is_fake_capture = (card_type == CardType::FAKE_CAPTURE); + card->type = card_type; if (card->output.get() != output) { card->output.reset(output); } - bmusb::PixelFormat pixel_format; + PixelFormat pixel_format; if (card_type == CardType::FFMPEG_INPUT) { - pixel_format = bmusb::PixelFormat_8BitBGRA; + pixel_format = capture->get_current_pixel_format(); } else if (global_flags.ten_bit_input) { pixel_format = PixelFormat_10BitYCbCr; } else { @@ -431,6 +472,39 @@ void Mixer::configure_card(unsigned card_index, CaptureInterface *capture, CardT audio_mixer.reset_resampler(device); audio_mixer.set_display_name(device, card->capture->get_description()); audio_mixer.trigger_state_changed_callback(); + + // Register metrics. + vector> labels; + char card_name[64]; + snprintf(card_name, sizeof(card_name), "%d", card_index); + labels.emplace_back("card", card_name); + + switch (card_type) { + case CardType::LIVE_CARD: + labels.emplace_back("cardtype", "live"); + break; + case CardType::FAKE_CAPTURE: + labels.emplace_back("cardtype", "fake"); + break; + case CardType::FFMPEG_INPUT: + labels.emplace_back("cardtype", "ffmpeg"); + break; + default: + assert(false); + } + card->queue_length_policy.register_metrics(labels); + global_metrics.add("input_dropped_frames_jitter", labels, &card->metric_input_dropped_frames_jitter); + global_metrics.add("input_dropped_frames_error", labels, &card->metric_input_dropped_frames_error); + global_metrics.add("input_dropped_frames_resets", labels, &card->metric_input_resets); + + global_metrics.add("input_has_signal_bool", labels, &card->metric_input_has_signal_bool, Metrics::TYPE_GAUGE); + global_metrics.add("input_is_connected_bool", labels, &card->metric_input_is_connected_bool, Metrics::TYPE_GAUGE); + global_metrics.add("input_interlaced_bool", labels, &card->metric_input_interlaced_bool, Metrics::TYPE_GAUGE); + global_metrics.add("input_width_pixels", labels, &card->metric_input_width_pixels, Metrics::TYPE_GAUGE); + global_metrics.add("input_height_pixels", labels, &card->metric_input_height_pixels, Metrics::TYPE_GAUGE); + global_metrics.add("input_frame_rate_nom", labels, &card->metric_input_frame_rate_nom, Metrics::TYPE_GAUGE); + global_metrics.add("input_frame_rate_den", labels, &card->metric_input_frame_rate_den, Metrics::TYPE_GAUGE); + global_metrics.add("input_sample_rate_hz", labels, &card->metric_input_sample_rate_hz, Metrics::TYPE_GAUGE); } void Mixer::set_output_card_internal(int card_index) @@ -447,7 +521,7 @@ void Mixer::set_output_card_internal(int card_index) // 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(); + CaptureInterface *fake_capture = old_card->capture.get(); lock.unlock(); fake_capture->stop_dequeue_thread(); lock.lock(); @@ -457,7 +531,7 @@ void Mixer::set_output_card_internal(int card_index) } if (card_index != -1) { CaptureCard *card = &cards[card_index]; - bmusb::CaptureInterface *capture = card->capture.get(); + 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. @@ -465,7 +539,7 @@ void Mixer::set_output_card_internal(int card_index) 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); + 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, CardType::FAKE_CAPTURE, card->output.release()); card->queue_length_policy.reset(card_index); card->capture->start_bm_capture(); @@ -496,6 +570,15 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, DeviceSpec device{InputSourceType::CAPTURE_CARD, card_index}; CaptureCard *card = &cards[card_index]; + card->metric_input_has_signal_bool = video_format.has_signal; + card->metric_input_is_connected_bool = video_format.is_connected; + card->metric_input_interlaced_bool = video_format.interlaced; + card->metric_input_width_pixels = video_format.width; + card->metric_input_height_pixels = video_format.height; + card->metric_input_frame_rate_nom = video_format.frame_rate_nom; + card->metric_input_frame_rate_den = video_format.frame_rate_den; + card->metric_input_sample_rate_hz = audio_format.sample_rate; + if (is_mode_scanning[card_index]) { if (video_format.has_signal) { // Found a stable signal, so stop scanning. @@ -545,10 +628,12 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, card_index, card->last_timecode, timecode); audio_mixer.reset_resampler(device); dropped_frames = 0; + ++card->metric_input_resets; } else if (dropped_frames > 0) { // Insert silence as needed. fprintf(stderr, "Card %d dropped %d frame(s) (before timecode 0x%04x), inserting silence.\n", card_index, dropped_frames, timecode); + card->metric_input_dropped_frames_error += dropped_frames; bool success; do { @@ -565,7 +650,28 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, card->last_timecode = timecode; + PBOFrameAllocator::Userdata *userdata = (PBOFrameAllocator::Userdata *)video_frame.userdata; + + size_t cbcr_width, cbcr_height, cbcr_offset, y_offset; size_t expected_length = video_format.stride * (video_format.height + video_format.extra_lines_top + video_format.extra_lines_bottom); + if (userdata != nullptr && userdata->pixel_format == PixelFormat_8BitYCbCrPlanar) { + // The calculation above is wrong for planar Y'CbCr, so just override it. + assert(card->type == CardType::FFMPEG_INPUT); + assert(video_offset == 0); + expected_length = video_frame.len; + + userdata->ycbcr_format = (static_cast(card->capture.get()))->get_current_frame_ycbcr_format(); + cbcr_width = video_format.width / userdata->ycbcr_format.chroma_subsampling_x; + cbcr_height = video_format.height / userdata->ycbcr_format.chroma_subsampling_y; + cbcr_offset = video_format.width * video_format.height; + y_offset = 0; + } else { + // All the other Y'CbCr formats are 4:2:2. + cbcr_width = video_format.width / 2; + cbcr_height = video_format.height; + cbcr_offset = video_offset / 2; + y_offset = video_frame.size / 2 + video_offset / 2; + } if (video_frame.len - video_offset == 0 || video_frame.len - video_offset != expected_length) { if (video_frame.len != 0) { @@ -592,8 +698,6 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, return; } - PBOFrameAllocator::Userdata *userdata = (PBOFrameAllocator::Userdata *)video_frame.userdata; - unsigned num_fields = video_format.interlaced ? 2 : 1; steady_clock::time_point frame_upload_start; bool interlaced_stride = false; @@ -602,6 +706,7 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, // a deinterlacer to actually get this right. assert(video_format.height % 2 == 0); video_format.height /= 2; + cbcr_height /= 2; assert(frame_length % 2 == 0); frame_length /= 2; num_fields = 2; @@ -618,10 +723,6 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, RefCountedFrame frame(video_frame); // Upload the textures. - const size_t cbcr_width = video_format.width / 2; - const size_t cbcr_offset = video_offset / 2; - const size_t y_offset = video_frame.size / 2 + video_offset / 2; - for (unsigned field = 0; field < num_fields; ++field) { // Put the actual texture upload in a lambda that is executed in the main thread. // It is entirely possible to do this in the same thread (and it might even be @@ -631,7 +732,7 @@ 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 = [this, field, video_format, y_offset, video_offset, cbcr_offset, cbcr_width, interlaced_stride, userdata]() { + auto upload_func = [this, field, video_format, y_offset, video_offset, cbcr_offset, cbcr_width, cbcr_height, interlaced_stride, userdata]() { unsigned field_start_line; if (field == 1) { field_start_line = video_format.second_field_start; @@ -641,30 +742,49 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, // For anything not FRAME_FORMAT_YCBCR_10BIT, v210_width will be nonsensical but not used. size_t v210_width = video_format.stride / sizeof(uint32_t); - ensure_texture_resolution(userdata, field, video_format.width, video_format.height, v210_width); + ensure_texture_resolution(userdata, field, video_format.width, video_format.height, cbcr_width, cbcr_height, v210_width); glBindBuffer(GL_PIXEL_UNPACK_BUFFER, userdata->pbo); check_error(); switch (userdata->pixel_format) { - case bmusb::PixelFormat_10BitYCbCr: { + case PixelFormat_10BitYCbCr: { size_t field_start = video_offset + video_format.stride * field_start_line; upload_texture(userdata->tex_v210[field], v210_width, video_format.height, video_format.stride, interlaced_stride, GL_RGBA, GL_UNSIGNED_INT_2_10_10_10_REV, field_start); v210_converter->convert(userdata->tex_v210[field], userdata->tex_444[field], video_format.width, video_format.height); break; } - case bmusb::PixelFormat_8BitYCbCr: { + case PixelFormat_8BitYCbCr: { size_t field_y_start = y_offset + video_format.width * field_start_line; size_t field_cbcr_start = cbcr_offset + cbcr_width * field_start_line * sizeof(uint16_t); // Make up our own strides, since we are interleaving. upload_texture(userdata->tex_y[field], video_format.width, video_format.height, video_format.width, interlaced_stride, GL_RED, GL_UNSIGNED_BYTE, field_y_start); - upload_texture(userdata->tex_cbcr[field], cbcr_width, video_format.height, cbcr_width * sizeof(uint16_t), interlaced_stride, GL_RG, GL_UNSIGNED_BYTE, field_cbcr_start); + upload_texture(userdata->tex_cbcr[field], cbcr_width, cbcr_height, cbcr_width * sizeof(uint16_t), interlaced_stride, GL_RG, GL_UNSIGNED_BYTE, field_cbcr_start); + break; + } + case PixelFormat_8BitYCbCrPlanar: { + assert(field_start_line == 0); // We don't really support interlaced here. + size_t field_y_start = y_offset; + size_t field_cb_start = cbcr_offset; + size_t field_cr_start = cbcr_offset + cbcr_width * cbcr_height; + + // Make up our own strides, since we are interleaving. + upload_texture(userdata->tex_y[field], video_format.width, video_format.height, video_format.width, interlaced_stride, GL_RED, GL_UNSIGNED_BYTE, field_y_start); + upload_texture(userdata->tex_cb[field], cbcr_width, cbcr_height, cbcr_width, interlaced_stride, GL_RED, GL_UNSIGNED_BYTE, field_cb_start); + upload_texture(userdata->tex_cr[field], cbcr_width, cbcr_height, cbcr_width, interlaced_stride, GL_RED, GL_UNSIGNED_BYTE, field_cr_start); break; } - case bmusb::PixelFormat_8BitBGRA: { + case PixelFormat_8BitBGRA: { size_t field_start = video_offset + video_format.stride * field_start_line; upload_texture(userdata->tex_rgba[field], video_format.width, video_format.height, video_format.stride, interlaced_stride, GL_BGRA, GL_UNSIGNED_INT_8_8_8_8_REV, field_start); + // These could be asked to deliver mipmaps at any time. + glBindTexture(GL_TEXTURE_2D, userdata->tex_rgba[field]); + check_error(); + glGenerateMipmap(GL_TEXTURE_2D); + check_error(); + glBindTexture(GL_TEXTURE_2D, 0); + check_error(); break; } default: @@ -814,6 +934,10 @@ void Mixer::thread_func() now = steady_clock::now(); double elapsed = duration(now - start).count(); + + metric_frames_output_total = frame_num; + metric_frames_output_dropped = stats_dropped_frames; + if (frame_num % 100 == 0) { printf("%d frames (%d dropped) in %.3f seconds = %.1f fps (%.1f ms/frame)", frame_num, stats_dropped_frames, elapsed, frame_num / elapsed, @@ -844,12 +968,16 @@ void Mixer::thread_func() long(limit.rlim_cur / 1048576), float(100.0 * (used.ru_maxrss * 1024.0) / limit.rlim_cur)); } + metrics_memory_locked_limit_bytes = limit.rlim_cur; } else { printf(", using %ld MB memory (not locked)", long(used.ru_maxrss / 1024)); + metrics_memory_locked_limit_bytes = 0.0 / 0.0; } printf("\n"); + + metrics_memory_used_bytes = used.ru_maxrss; } @@ -917,6 +1045,9 @@ void Mixer::trim_queue(CaptureCard *card, unsigned card_index) ++dropped_frames; } + metric_input_queue_length_frames = queue_length; + card->metric_input_dropped_frames_jitter += dropped_frames; + #if 0 if (dropped_frames > 0) { fprintf(stderr, "Card %u dropped %u frame(s) to keep latency down.\n", @@ -1081,6 +1212,17 @@ void Mixer::render_one_frame(int64_t duration) printf("Timecode: '%s'\n", timecode_text.c_str()); } + // Update Y'CbCr settings for all cards. + { + unique_lock lock(card_mutex); + for (unsigned card_index = 0; card_index < num_cards; ++card_index) { + YCbCrInterpretation *interpretation = &ycbcr_interpretation[card_index]; + input_state.ycbcr_coefficients_auto[card_index] = interpretation->ycbcr_coefficients_auto; + input_state.ycbcr_coefficients[card_index] = interpretation->ycbcr_coefficients; + input_state.full_range[card_index] = interpretation->full_range; + } + } + // Get the main chain from the theme, and set its state immediately. Theme::Chain theme_main_chain = theme->get_chain(0, pts(), global_flags.width, global_flags.height, input_state); EffectChain *chain = theme_main_chain.chain; @@ -1282,6 +1424,18 @@ void Mixer::channel_clicked(int preview_num) theme->channel_clicked(preview_num); } +YCbCrInterpretation Mixer::get_input_ycbcr_interpretation(unsigned card_index) const +{ + unique_lock lock(card_mutex); + return ycbcr_interpretation[card_index]; +} + +void Mixer::set_input_ycbcr_interpretation(unsigned card_index, const YCbCrInterpretation &interpretation) +{ + unique_lock lock(card_mutex); + ycbcr_interpretation[card_index] = interpretation; +} + void Mixer::start_mode_scanning(unsigned card_index) { assert(card_index < num_cards); @@ -1299,7 +1453,7 @@ 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 +map Mixer::get_available_output_video_modes() const { assert(desired_output_card_index != -1); unique_lock lock(card_mutex); @@ -1327,10 +1481,12 @@ void Mixer::OutputChannel::output_frame(DisplayFrame frame) } ready_frame = frame; has_ready_frame = true; - } - if (new_frame_ready_callback) { - new_frame_ready_callback(); + // Call the callbacks under the mutex (they should be short), + // so that we don't race against a callback removal. + for (const auto &key_and_callback : new_frame_ready_callbacks) { + key_and_callback.second(); + } } // Reduce the number of callbacks by filtering duplicates. The reason @@ -1397,9 +1553,16 @@ bool Mixer::OutputChannel::get_display_frame(DisplayFrame *frame) return true; } -void Mixer::OutputChannel::set_frame_ready_callback(Mixer::new_frame_ready_callback_t callback) +void Mixer::OutputChannel::add_frame_ready_callback(void *key, Mixer::new_frame_ready_callback_t callback) { - new_frame_ready_callback = callback; + unique_lock lock(frame_mutex); + new_frame_ready_callbacks[key] = callback; +} + +void Mixer::OutputChannel::remove_frame_ready_callback(void *key) +{ + unique_lock lock(frame_mutex); + new_frame_ready_callbacks.erase(key); } void Mixer::OutputChannel::set_transition_names_updated_callback(Mixer::transition_names_updated_callback_t callback)