#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
-#include <sys/resource.h>
#include <algorithm>
#include <chrono>
#include <condition_variable>
#include "DeckLinkAPI.h"
#include "LinuxCOM.h"
#include "alsa_output.h"
+#include "basic_stats.h"
#include "bmusb/bmusb.h"
#include "bmusb/fake_capture.h"
+#ifdef HAVE_CEF
+#include "cef_capture.h"
+#endif
#include "chroma_subsampler.h"
#include "context.h"
#include "decklink_capture.h"
#include "decklink_output.h"
#include "defs.h"
#include "disk_space_estimator.h"
+#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"
#include "v210_converter.h"
#include "video_encoder.h"
+#undef Status
+#include <google/protobuf/util/json_util.h>
+#include "json.pb.h"
+
class IDeckLink;
class QOpenGLContext;
using namespace bmusb;
Mixer *global_mixer = nullptr;
-bool uses_mlock = false;
namespace {
}
}
-void ensure_texture_resolution(PBOFrameAllocator::Userdata *userdata, unsigned field, unsigned width, unsigned height)
+void ensure_texture_resolution(PBOFrameAllocator::Userdata *userdata, unsigned field, unsigned width, unsigned height, unsigned cbcr_width, unsigned cbcr_height, unsigned v210_width)
{
bool first;
- if (global_flags.ten_bit_input) {
+ switch (userdata->pixel_format) {
+ case PixelFormat_10BitYCbCr:
first = userdata->tex_v210[field] == 0 || userdata->tex_444[field] == 0;
- } else {
+ break;
+ case PixelFormat_8BitYCbCr:
first = userdata->tex_y[field] == 0 || userdata->tex_cbcr[field] == 0;
+ break;
+ 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.
- if (global_flags.ten_bit_input) {
- const size_t v210_width = v210Converter::get_minimum_v210_texture_width(width);
-
- glBindTexture(GL_TEXTURE_2D, userdata->tex_v210[field]);
- check_error();
- glTexImage2D(GL_TEXTURE_2D, 0, GL_RGB10_A2, v210_width, height, 0, GL_RGBA, GL_UNSIGNED_INT_2_10_10_10_REV, nullptr);
- check_error();
+ switch (userdata->pixel_format) {
+ 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();
- } else {
- size_t cbcr_width = width / 2;
-
+ break;
+ 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);
check_error();
glTexImage2D(GL_TEXTURE_2D, 0, GL_R8, width, height, 0, GL_RED, GL_UNSIGNED_BYTE, nullptr);
check_error();
+ break;
+ }
+ 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.
+ glTexImage2D(GL_TEXTURE_2D, 0, GL_SRGB8_ALPHA8, width, height, 0, GL_BGRA, GL_UNSIGNED_INT_8_8_8_8_REV, nullptr);
+ } else {
+ glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA8, width, height, 0, GL_BGRA, GL_UNSIGNED_INT_8_8_8_8_REV, nullptr);
+ }
+ check_error();
+ break;
}
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])) {
+ // Same as above; we need to recreate the texture.
+ glBindTexture(GL_TEXTURE_2D, userdata->tex_v210[field]);
+ check_error();
+ glTexImage2D(GL_TEXTURE_2D, 0, GL_RGB10_A2, v210_width, height, 0, GL_RGBA, GL_UNSIGNED_INT_2_10_10_10_REV, nullptr);
+ check_error();
+ userdata->last_v210_width[field] = v210_width;
}
}
} // namespace
-void QueueLengthPolicy::update_policy(unsigned queue_length)
+void JitterHistory::register_metrics(const vector<pair<string, string>> &labels)
{
- 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 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 >= 1) {
- if (queue_length >= safe_queue_length) {
- been_at_safe_point_since_last_starvation = true;
+ global_metrics.add("input_underestimated_jitter_frames", labels, &metric_input_underestimated_jitter_frames);
+ global_metrics.add("input_estimated_max_jitter_seconds", labels, &metric_input_estimated_max_jitter_seconds, Metrics::TYPE_GAUGE);
+}
+
+void JitterHistory::unregister_metrics(const vector<pair<string, string>> &labels)
+{
+ global_metrics.remove("input_underestimated_jitter_frames", labels);
+ global_metrics.remove("input_estimated_max_jitter_seconds", labels);
+}
+
+void JitterHistory::frame_arrived(steady_clock::time_point now, int64_t frame_duration, size_t dropped_frames)
+{
+ if (expected_timestamp > steady_clock::time_point::min()) {
+ expected_timestamp += dropped_frames * nanoseconds(frame_duration * 1000000000 / TIMEBASE);
+ double jitter_seconds = fabs(duration<double>(expected_timestamp - now).count());
+ history.push_back(orders.insert(jitter_seconds));
+ if (jitter_seconds > estimate_max_jitter()) {
+ ++metric_input_underestimated_jitter_frames;
}
- 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 frame(s)\n",
- card_index, safe_queue_length);
- frames_with_at_least_one = 0;
+
+ metric_input_estimated_max_jitter_seconds = estimate_max_jitter();
+
+ if (history.size() > history_length) {
+ orders.erase(history.front());
+ history.pop_front();
}
+ assert(history.size() <= history_length);
+ }
+ expected_timestamp = now + nanoseconds(frame_duration * 1000000000 / TIMEBASE);
+}
+
+double JitterHistory::estimate_max_jitter() const
+{
+ if (orders.empty()) {
+ return 0.0;
+ }
+ size_t elem_idx = lrint((orders.size() - 1) * percentile);
+ if (percentile <= 0.5) {
+ return *next(orders.begin(), elem_idx) * multiplier;
+ } else {
+ return *prev(orders.end(), orders.size() - elem_idx) * multiplier;
+ }
+}
+
+void QueueLengthPolicy::register_metrics(const vector<pair<string, string>> &labels)
+{
+ global_metrics.add("input_queue_safe_length_frames", labels, &metric_input_queue_safe_length_frames, Metrics::TYPE_GAUGE);
+}
+
+void QueueLengthPolicy::unregister_metrics(const vector<pair<string, string>> &labels)
+{
+ global_metrics.remove("input_queue_safe_length_frames", labels);
+}
+
+void QueueLengthPolicy::update_policy(steady_clock::time_point now,
+ steady_clock::time_point expected_next_frame,
+ int64_t input_frame_duration,
+ int64_t master_frame_duration,
+ double max_input_card_jitter_seconds,
+ double max_master_card_jitter_seconds)
+{
+ double input_frame_duration_seconds = input_frame_duration / double(TIMEBASE);
+ double master_frame_duration_seconds = master_frame_duration / double(TIMEBASE);
+
+ // Figure out when we can expect the next frame for this card, assuming
+ // worst-case jitter (ie., the frame is maximally late).
+ double seconds_until_next_frame = max(duration<double>(expected_next_frame - now).count() + max_input_card_jitter_seconds, 0.0);
+
+ // How many times are the master card expected to tick in that time?
+ // We assume the master clock has worst-case jitter but not any rate
+ // discrepancy, ie., it ticks as early as possible every time, but not
+ // cumulatively.
+ double frames_needed = (seconds_until_next_frame + max_master_card_jitter_seconds) / master_frame_duration_seconds;
+
+ // As a special case, if the master card ticks faster than the input card,
+ // we expect the queue to drain by itself even without dropping. But if
+ // the difference is small (e.g. 60 Hz master and 59.94 input), it would
+ // go slowly enough that the effect wouldn't really be appreciable.
+ // We account for this by looking at the situation five frames ahead,
+ // assuming everything else is the same.
+ double frames_allowed;
+ if (master_frame_duration < input_frame_duration) {
+ frames_allowed = frames_needed + 5 * (input_frame_duration_seconds - master_frame_duration_seconds) / master_frame_duration_seconds;
} else {
- frames_with_at_least_one = 0;
+ frames_allowed = frames_needed;
}
+
+ safe_queue_length = max<int>(floor(frames_allowed), 0);
+ metric_input_queue_safe_length_frames = safe_queue_length;
}
Mixer::Mixer(const QSurfaceFormat &format, unsigned num_cards)
decklink_output_surface(create_surface(format)),
audio_mixer(num_cards)
{
+ memcpy(ycbcr_interpretation, global_flags.ycbcr_interpretation, sizeof(ycbcr_interpretation));
CHECK(init_movit(MOVIT_SHADER_DIR, MOVIT_DEBUG_OFF));
check_error();
+ // This nearly always should be true.
+ global_flags.can_disable_srgb_decoder =
+ epoxy_has_gl_extension("GL_EXT_texture_sRGB_decode") &&
+ epoxy_has_gl_extension("GL_ARB_sampler_objects");
+
// Since we allow non-bouncing 4:2:2 YCbCrInputs, effective subpixel precision
// will be halved when sampling them, and we need to compensate here.
movit_texel_subpixel_precision /= 2.0;
resource_pool.reset(new ResourcePool);
- theme.reset(new Theme(global_flags.theme_filename, global_flags.theme_dirs, resource_pool.get(), num_cards));
for (unsigned i = 0; i < NUM_OUTPUTS; ++i) {
output_channel[i].parent = this;
output_channel[i].channel = i;
inout_format.color_space = COLORSPACE_sRGB;
inout_format.gamma_curve = GAMMA_sRGB;
- // Display chain; shows the live output produced by the main chain (its RGBA version).
+ // Matches the 4:2:0 format created by the main chain.
+ YCbCrFormat ycbcr_format;
+ ycbcr_format.chroma_subsampling_x = 2;
+ ycbcr_format.chroma_subsampling_y = 2;
+ if (global_flags.ycbcr_rec709_coefficients) {
+ ycbcr_format.luma_coefficients = YCBCR_REC_709;
+ } else {
+ ycbcr_format.luma_coefficients = YCBCR_REC_601;
+ }
+ ycbcr_format.full_range = false;
+ ycbcr_format.num_levels = 1 << global_flags.x264_bit_depth;
+ ycbcr_format.cb_x_position = 0.0f;
+ ycbcr_format.cr_x_position = 0.0f;
+ ycbcr_format.cb_y_position = 0.5f;
+ ycbcr_format.cr_y_position = 0.5f;
+
+ // Display chain; shows the live output produced by the main chain (or rather, a copy of it).
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, global_flags.width, global_flags.height); // FIXME: GL_UNSIGNED_BYTE is really wrong.
+ GLenum type = global_flags.x264_bit_depth > 8 ? GL_UNSIGNED_SHORT : GL_UNSIGNED_BYTE;
+ display_input = new YCbCrInput(inout_format, ycbcr_format, global_flags.width, global_flags.height, YCBCR_INPUT_SPLIT_Y_AND_CBCR, type);
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.
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));
+ // Must be instantiated after VideoEncoder has initialized global_flags.use_zerocopy.
+ theme.reset(new Theme(global_flags.theme_filename, global_flags.theme_dirs, resource_pool.get(), num_cards));
+
+ httpd.add_endpoint("/channels", bind(&Mixer::get_channels_json, this), HTTPD::ALLOW_ALL_ORIGINS);
+ for (int channel_idx = 2; channel_idx < theme->get_num_channels(); ++channel_idx) {
+ char url[256];
+ snprintf(url, sizeof(url), "/channels/%d/color", channel_idx);
+ httpd.add_endpoint(url, bind(&Mixer::get_channel_color_http, this, unsigned(channel_idx)), HTTPD::ALLOW_ALL_ORIGINS);
+ }
+
// Start listening for clients only once VideoEncoder has written its header, if any.
- httpd.start(9095);
+ httpd.start(global_flags.http_port);
// First try initializing the then PCI devices, then USB, then
// fill up with fake cards until we have the desired number of cards.
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);
+ if (!output->set_device(decklink)) {
+ delete output;
+ output = nullptr;
+ }
+ configure_card(card_index, capture, CardType::LIVE_CARD, output);
++num_pci_devices;
}
decklink_iterator->Release();
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, /*output=*/nullptr);
+ configure_card(card_index, capture, CardType::LIVE_CARD, /*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(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);
+ configure_card(card_index, capture, CardType::FAKE_CAPTURE, /*output=*/nullptr);
}
if (num_fake_cards > 0) {
fprintf(stderr, "Initialized %u fake cards.\n", num_fake_cards);
}
+ // Initialize all video inputs the theme asked for. Note that these are
+ // all put _after_ the regular cards, which stop at <num_cards> - 1.
+ std::vector<FFmpegCapture *> video_inputs = theme->get_video_inputs();
+ for (unsigned video_card_index = 0; video_card_index < video_inputs.size(); ++card_index, ++video_card_index) {
+ if (card_index >= MAX_VIDEO_CARDS) {
+ fprintf(stderr, "ERROR: Not enough card slots available for the videos the theme requested.\n");
+ exit(1);
+ }
+ configure_card(card_index, video_inputs[video_card_index], CardType::FFMPEG_INPUT, /*output=*/nullptr);
+ video_inputs[video_card_index]->set_card_index(card_index);
+ }
+ num_video_inputs = video_inputs.size();
+
+#ifdef HAVE_CEF
+ // Same, for HTML inputs.
+ std::vector<CEFCapture *> html_inputs = theme->get_html_inputs();
+ for (unsigned html_card_index = 0; html_card_index < html_inputs.size(); ++card_index, ++html_card_index) {
+ if (card_index >= MAX_VIDEO_CARDS) {
+ fprintf(stderr, "ERROR: Not enough card slots available for the HTML inputs the theme requested.\n");
+ exit(1);
+ }
+ configure_card(card_index, html_inputs[html_card_index], CardType::CEF_INPUT, /*output=*/nullptr);
+ html_inputs[html_card_index]->set_card_index(card_index);
+ }
+ num_html_inputs = html_inputs.size();
+#endif
+
BMUSBCapture::set_card_connected_callback(bind(&Mixer::bm_hotplug_add, this, _1));
BMUSBCapture::start_bm_thread();
- for (unsigned card_index = 0; card_index < num_cards; ++card_index) {
+ for (unsigned card_index = 0; card_index < num_cards + num_video_inputs + num_html_inputs; ++card_index) {
cards[card_index].queue_length_policy.reset(card_index);
}
v210_converter->precompile_shader(3840);
v210_converter->precompile_shader(4096);
}
+ if (global_flags.ten_bit_output) {
+ if (!v210Converter::has_hardware_support()) {
+ fprintf(stderr, "ERROR: --ten-bit-output requires support for OpenGL compute shaders\n");
+ fprintf(stderr, " (OpenGL 4.3, or GL_ARB_compute_shader + GL_ARB_shader_image_load_store).\n");
+ exit(1);
+ }
+ }
timecode_renderer.reset(new TimecodeRenderer(resource_pool.get(), global_flags.width, global_flags.height));
display_timecode_in_stream = global_flags.display_timecode_in_stream;
desired_output_card_index = global_flags.output_card;
set_output_card_internal(global_flags.output_card);
}
+
+ output_jitter_history.register_metrics({{ "card", "output" }});
}
Mixer::~Mixer()
{
BMUSBCapture::stop_bm_thread();
- for (unsigned card_index = 0; card_index < num_cards; ++card_index) {
+ for (unsigned card_index = 0; card_index < num_cards + num_video_inputs + num_html_inputs; ++card_index) {
{
unique_lock<mutex> lock(card_mutex);
cards[card_index].should_quit = true; // Unblock thread.
video_encoder.reset(nullptr);
}
-void Mixer::configure_card(unsigned card_index, CaptureInterface *capture, bool is_fake_capture, DeckLinkOutput *output)
+void Mixer::configure_card(unsigned card_index, CaptureInterface *capture, CardType card_type, DeckLinkOutput *output)
{
printf("Configuring card %d...\n", card_index);
card->capture->stop_dequeue_thread();
}
card->capture.reset(capture);
- card->is_fake_capture = is_fake_capture;
+ card->is_fake_capture = (card_type == CardType::FAKE_CAPTURE);
+ card->type = card_type;
if (card->output.get() != output) {
card->output.reset(output);
}
+
+ PixelFormat pixel_format;
+ if (card_type == CardType::FFMPEG_INPUT) {
+ pixel_format = capture->get_current_pixel_format();
+ } else if (card_type == CardType::CEF_INPUT) {
+ pixel_format = PixelFormat_8BitBGRA;
+ } else if (global_flags.ten_bit_input) {
+ pixel_format = PixelFormat_10BitYCbCr;
+ } else {
+ pixel_format = PixelFormat_8BitYCbCr;
+ }
+
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, global_flags.width, global_flags.height)); // 8 MB.
+ card->frame_allocator.reset(new PBOFrameAllocator(pixel_format, 8 << 20, global_flags.width, global_flags.height)); // 8 MB.
}
card->capture->set_video_frame_allocator(card->frame_allocator.get());
if (card->surface == nullptr) {
}
while (!card->new_frames.empty()) card->new_frames.pop_front();
card->last_timecode = -1;
- card->capture->set_pixel_format(global_flags.ten_bit_input ? PixelFormat_10BitYCbCr : PixelFormat_8BitYCbCr);
+ card->capture->set_pixel_format(pixel_format);
card->capture->configure_card();
// NOTE: start_bm_capture() happens in thread_func().
audio_mixer.reset_resampler(device);
audio_mixer.set_display_name(device, card->capture->get_description());
audio_mixer.trigger_state_changed_callback();
+
+ // Unregister old metrics, if any.
+ if (!card->labels.empty()) {
+ const vector<pair<string, string>> &labels = card->labels;
+ card->jitter_history.unregister_metrics(labels);
+ card->queue_length_policy.unregister_metrics(labels);
+ global_metrics.remove("input_received_frames", labels);
+ global_metrics.remove("input_dropped_frames_jitter", labels);
+ global_metrics.remove("input_dropped_frames_error", labels);
+ global_metrics.remove("input_dropped_frames_resets", labels);
+ global_metrics.remove("input_queue_length_frames", labels);
+ global_metrics.remove("input_queue_duped_frames", labels);
+
+ global_metrics.remove("input_has_signal_bool", labels);
+ global_metrics.remove("input_is_connected_bool", labels);
+ global_metrics.remove("input_interlaced_bool", labels);
+ global_metrics.remove("input_width_pixels", labels);
+ global_metrics.remove("input_height_pixels", labels);
+ global_metrics.remove("input_frame_rate_nom", labels);
+ global_metrics.remove("input_frame_rate_den", labels);
+ global_metrics.remove("input_sample_rate_hz", labels);
+ }
+
+ // Register metrics.
+ vector<pair<string, string>> 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;
+ case CardType::CEF_INPUT:
+ labels.emplace_back("cardtype", "cef");
+ break;
+ default:
+ assert(false);
+ }
+ card->jitter_history.register_metrics(labels);
+ card->queue_length_policy.register_metrics(labels);
+ global_metrics.add("input_received_frames", labels, &card->metric_input_received_frames);
+ 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_queue_length_frames", labels, &card->metric_input_queue_length_frames, Metrics::TYPE_GAUGE);
+ global_metrics.add("input_queue_duped_frames", labels, &card->metric_input_duped_frames);
+
+ 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);
+ card->labels = labels;
}
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();
- old_card->capture = move(old_card->parked_capture);
+ old_card->capture = move(old_card->parked_capture); // TODO: reset the metrics
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();
+ 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.
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());
+ 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();
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;
+ output_jitter_history.clear();
}
namespace {
DeviceSpec device{InputSourceType::CAPTURE_CARD, card_index};
CaptureCard *card = &cards[card_index];
+ ++card->metric_input_received_frames;
+ 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.
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 {
} while (!success);
}
- audio_mixer.add_audio(device, audio_frame.data + audio_offset, num_samples, audio_format, frame_length, audio_frame.received_timestamp);
+ if (num_samples > 0) {
+ 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) {
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<FFmpegCapture *>(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) {
new_frame.dropped_frames = dropped_frames;
new_frame.received_timestamp = video_frame.received_timestamp;
card->new_frames.push_back(move(new_frame));
- card->new_frames_changed.notify_all();
+ card->jitter_history.frame_arrived(video_frame.received_timestamp, frame_length, dropped_frames);
}
+ card->new_frames_changed.notify_all();
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;
// 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;
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
// Note that this means we must hold on to the actual frame data in <userdata>
// until the upload command is run, but we hold on to <frame> 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;
field_start_line = video_format.extra_lines_top;
}
- ensure_texture_resolution(userdata, field, video_format.width, video_format.height);
+ // 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, cbcr_width, cbcr_height, v210_width);
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, userdata->pbo);
check_error();
- if (global_flags.ten_bit_input) {
+ switch (userdata->pixel_format) {
+ case PixelFormat_10BitYCbCr: {
size_t field_start = video_offset + video_format.stride * field_start_line;
- upload_texture(userdata->tex_v210[field], video_format.stride / sizeof(uint32_t), video_format.height, video_format.stride, interlaced_stride, GL_RGBA, GL_UNSIGNED_INT_2_10_10_10_REV, field_start);
+ 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);
- } else {
+ break;
+ }
+ 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 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:
+ assert(false);
}
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
new_frame.dropped_frames = dropped_frames;
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();
+ card->jitter_history.frame_arrived(video_frame.received_timestamp, frame_length, dropped_frames);
}
+ card->new_frames_changed.notify_all();
}
}
// 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) {
+ for (unsigned card_index = 0; card_index < num_cards + num_video_inputs + num_html_inputs; ++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();
-
+ BasicStats basic_stats(/*verbose=*/true);
int stats_dropped_frames = 0;
while (!should_quit) {
handle_hotplugged_cards();
- for (unsigned card_index = 0; card_index < num_cards; ++card_index) {
+ for (unsigned card_index = 0; card_index < num_cards + num_video_inputs + num_html_inputs; ++card_index) {
if (card_index == master_card_index || !has_new_frame[card_index]) {
continue;
}
continue;
}
- for (unsigned card_index = 0; card_index < num_cards; ++card_index) {
+ for (unsigned card_index = 0; card_index < num_cards + num_video_inputs + num_html_inputs; ++card_index) {
if (!has_new_frame[card_index] || new_frames[card_index].frame->len == 0)
continue;
++frame_num;
pts_int += frame_duration;
- now = steady_clock::now();
- double elapsed = duration<double>(now - start).count();
- 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,
- 1e3 * elapsed / frame_num);
- // chain->print_phase_timing();
-
- // Check our memory usage, to see if we are close to our mlockall()
- // limit (if at all set).
- rusage used;
- if (getrusage(RUSAGE_SELF, &used) == -1) {
- perror("getrusage(RUSAGE_SELF)");
- assert(false);
- }
-
- if (uses_mlock) {
- rlimit limit;
- if (getrlimit(RLIMIT_MEMLOCK, &limit) == -1) {
- perror("getrlimit(RLIMIT_MEMLOCK)");
- assert(false);
- }
-
- 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));
- }
-
- printf("\n");
- }
-
+ basic_stats.update(frame_num, stats_dropped_frames);
+ // if (frame_num % 100 == 0) chain->print_phase_timing();
if (should_cut.exchange(false)) { // Test and clear.
video_encoder->do_cut(frame_num);
return (card_index == master_card_index);
}
-void Mixer::trim_queue(CaptureCard *card, unsigned card_index)
+void Mixer::trim_queue(CaptureCard *card, size_t safe_queue_length)
{
// Count the number of frames in the queue, including any frames
// we dropped. It's hard to know exactly how we should deal with
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()) {
+ while (queue_length > 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()) {
+ if (queue_length <= safe_queue_length) {
// No need to drop anything.
break;
}
++dropped_frames;
}
+ card->metric_input_dropped_frames_jitter += dropped_frames;
+ card->metric_input_queue_length_frames = queue_length;
+
+#if 0
if (dropped_frames > 0) {
fprintf(stderr, "Card %u dropped %u frame(s) to keep latency down.\n",
card_index, dropped_frames);
}
+#endif
+}
+
+pair<string, string> Mixer::get_channels_json()
+{
+ Channels ret;
+ for (int channel_idx = 2; channel_idx < theme->get_num_channels(); ++channel_idx) {
+ Channel *channel = ret.add_channel();
+ channel->set_index(channel_idx);
+ channel->set_name(theme->get_channel_name(channel_idx));
+ channel->set_color(theme->get_channel_color(channel_idx));
+ }
+ string contents;
+ google::protobuf::util::MessageToJsonString(ret, &contents); // Ignore any errors.
+ return make_pair(contents, "text/json");
}
+pair<string, string> Mixer::get_channel_color_http(unsigned channel_idx)
+{
+ return make_pair(theme->get_channel_color(channel_idx), "text/plain");
+}
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])
{
// and then restart.
assert(cards[master_card_index].capture->get_disconnected());
handle_hotplugged_cards();
+ lock.unlock();
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) {
+ for (unsigned card_index = 0; card_index < num_cards + num_video_inputs + num_html_inputs; ++card_index) {
CaptureCard *card = &cards[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());
+ if (card->new_frames.empty()) { // Starvation.
+ ++card->metric_input_duped_frames;
} else {
- 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();
}
if (!master_card_is_output) {
+ output_frame_info.frame_timestamp = new_frames[master_card_index].received_timestamp;
output_frame_info.dropped_frames = new_frames[master_card_index].dropped_frames;
output_frame_info.frame_duration = new_frames[master_card_index].length;
}
+ if (!output_frame_info.is_preroll) {
+ output_jitter_history.frame_arrived(output_frame_info.frame_timestamp, output_frame_info.frame_duration, output_frame_info.dropped_frames);
+ }
+
+ for (unsigned card_index = 0; card_index < num_cards + num_video_inputs + num_html_inputs; ++card_index) {
+ CaptureCard *card = &cards[card_index];
+ if (has_new_frame[card_index] &&
+ !input_card_is_master_clock(card_index, master_card_index) &&
+ !output_frame_info.is_preroll) {
+ card->queue_length_policy.update_policy(
+ output_frame_info.frame_timestamp,
+ card->jitter_history.get_expected_next_frame(),
+ new_frames[master_card_index].length,
+ output_frame_info.frame_duration,
+ card->jitter_history.estimate_max_jitter(),
+ output_jitter_history.estimate_max_jitter());
+ trim_queue(card, min<int>(global_flags.max_input_queue_frames,
+ card->queue_length_policy.get_safe_queue_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;
if (card->capture->get_disconnected()) {
fprintf(stderr, "Card %u went away, replacing with a fake card.\n", card_index);
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);
+ configure_card(card_index, capture, CardType::FAKE_CAPTURE, /*output=*/nullptr);
card->queue_length_policy.reset(card_index);
card->capture->start_bm_capture();
}
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, /*output=*/nullptr);
+ configure_card(free_card_index, capture, CardType::LIVE_CARD, /*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();
printf("Timecode: '%s'\n", timecode_text.c_str());
}
+ // Update Y'CbCr settings for all cards.
+ {
+ unique_lock<mutex> 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;
theme_main_chain.setup_chain();
//theme_main_chain.chain->enable_phase_timing(true);
- GLuint y_tex, cbcr_tex;
- bool got_frame = video_encoder->begin_frame(&y_tex, &cbcr_tex);
+ // The theme can't (or at least shouldn't!) call connect_signal() on
+ // each FFmpeg or CEF input, so we'll do it here.
+ for (const pair<LiveInputWrapper *, FFmpegCapture *> &conn : theme->get_video_signal_connections()) {
+ conn.first->connect_signal_raw(conn.second->get_card_index(), input_state);
+ }
+#ifdef HAVE_CEF
+ for (const pair<LiveInputWrapper *, CEFCapture *> &conn : theme->get_html_signal_connections()) {
+ conn.first->connect_signal_raw(conn.second->get_card_index(), input_state);
+ }
+#endif
+
+ // If HDMI/SDI output is active and the user has requested auto mode,
+ // its mode overrides the existing Y'CbCr setting for the chain.
+ YCbCrLumaCoefficients ycbcr_output_coefficients;
+ if (global_flags.ycbcr_auto_coefficients && output_card_index != -1) {
+ ycbcr_output_coefficients = cards[output_card_index].output->preferred_ycbcr_coefficients();
+ } else {
+ ycbcr_output_coefficients = global_flags.ycbcr_rec709_coefficients ? YCBCR_REC_709 : YCBCR_REC_601;
+ }
+
+ // TODO: Reduce the duplication against theme.cpp.
+ YCbCrFormat output_ycbcr_format;
+ output_ycbcr_format.chroma_subsampling_x = 1;
+ output_ycbcr_format.chroma_subsampling_y = 1;
+ output_ycbcr_format.luma_coefficients = ycbcr_output_coefficients;
+ output_ycbcr_format.full_range = false;
+ output_ycbcr_format.num_levels = 1 << global_flags.x264_bit_depth;
+ chain->change_ycbcr_output_format(output_ycbcr_format);
+
+ // Render main chain. If we're using zerocopy Quick Sync encoding
+ // (the default case), we take an extra copy of the created outputs,
+ // so that we can display it back to the screen later (it's less memory
+ // bandwidth than writing and reading back an RGBA texture, even at 16-bit).
+ // Ideally, we'd like to avoid taking copies and just use the main textures
+ // for display as well, but they're just views into VA-API memory and must be
+ // unmapped during encoding, so we can't use them for display, unfortunately.
+ GLuint y_tex, cbcr_full_tex, cbcr_tex;
+ GLuint y_copy_tex, cbcr_copy_tex = 0;
+ GLuint y_display_tex, cbcr_display_tex;
+ GLenum y_type = (global_flags.x264_bit_depth > 8) ? GL_R16 : GL_R8;
+ GLenum cbcr_type = (global_flags.x264_bit_depth > 8) ? GL_RG16 : GL_RG8;
+ const bool is_zerocopy = video_encoder->is_zerocopy();
+ if (is_zerocopy) {
+ cbcr_full_tex = resource_pool->create_2d_texture(cbcr_type, global_flags.width, global_flags.height);
+ y_copy_tex = resource_pool->create_2d_texture(y_type, global_flags.width, global_flags.height);
+ cbcr_copy_tex = resource_pool->create_2d_texture(cbcr_type, global_flags.width / 2, global_flags.height / 2);
+
+ y_display_tex = y_copy_tex;
+ cbcr_display_tex = cbcr_copy_tex;
+
+ // y_tex and cbcr_tex will be given by VideoEncoder.
+ } else {
+ cbcr_full_tex = resource_pool->create_2d_texture(cbcr_type, global_flags.width, global_flags.height);
+ y_tex = resource_pool->create_2d_texture(y_type, global_flags.width, global_flags.height);
+ cbcr_tex = resource_pool->create_2d_texture(cbcr_type, global_flags.width / 2, global_flags.height / 2);
+
+ y_display_tex = y_tex;
+ cbcr_display_tex = cbcr_tex;
+ }
+
+ const int64_t av_delay = lrint(global_flags.audio_queue_length_ms * 0.001 * TIMEBASE); // Corresponds to the delay in ResamplingQueue.
+ bool got_frame = video_encoder->begin_frame(pts_int + av_delay, duration, ycbcr_output_coefficients, theme_main_chain.input_frames, &y_tex, &cbcr_tex);
assert(got_frame);
- // Render main chain.
- 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);
+ GLuint fbo;
+ if (is_zerocopy) {
+ fbo = resource_pool->create_fbo(y_tex, cbcr_full_tex, y_copy_tex);
+ } else {
+ fbo = resource_pool->create_fbo(y_tex, cbcr_full_tex);
+ }
check_error();
chain->render_to_fbo(fbo, global_flags.width, global_flags.height);
resource_pool->release_fbo(fbo);
- chroma_subsampler->subsample_chroma(cbcr_full_tex, global_flags.width, global_flags.height, cbcr_tex);
+ if (is_zerocopy) {
+ chroma_subsampler->subsample_chroma(cbcr_full_tex, global_flags.width, global_flags.height, cbcr_tex, cbcr_copy_tex);
+ } else {
+ 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);
+ cards[output_card_index].output->send_frame(y_tex, cbcr_full_tex, ycbcr_output_coefficients, theme_main_chain.input_frames, pts_int, duration);
}
resource_pool->release_2d_texture(cbcr_full_tex);
- // Set the right state for rgba_tex.
+ // Set the right state for the Y' and CbCr textures we use for display.
glBindFramebuffer(GL_FRAMEBUFFER, 0);
- glBindTexture(GL_TEXTURE_2D, rgba_tex);
+ glBindTexture(GL_TEXTURE_2D, y_display_tex);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
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 = 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);
+ glBindTexture(GL_TEXTURE_2D, cbcr_display_tex);
+ glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
+ glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
+ glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
+
+ RefCountedGLsync fence = video_encoder->end_frame();
- // The live frame just shows the RGBA texture we just rendered.
- // It owns rgba_tex now.
+ // The live frame pieces the Y'CbCr texture copies back into RGB and displays them.
+ // It owns y_display_tex and cbcr_display_tex now (whichever textures they are).
DisplayFrame live_frame;
live_frame.chain = display_chain.get();
- live_frame.setup_chain = [this, rgba_tex]{
- display_input->set_texture_num(rgba_tex);
+ live_frame.setup_chain = [this, y_display_tex, cbcr_display_tex]{
+ display_input->set_texture_num(0, y_display_tex);
+ display_input->set_texture_num(1, cbcr_display_tex);
};
live_frame.ready_fence = fence;
live_frame.input_frames = {};
- live_frame.temp_textures = { rgba_tex };
- output_channel[OUTPUT_LIVE].output_frame(live_frame);
+ live_frame.temp_textures = { y_display_tex, cbcr_display_tex };
+ output_channel[OUTPUT_LIVE].output_frame(move(live_frame));
// 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(), global_flags.width, global_flags.height, input_state); // FIXME: dimensions
- display_frame.chain = chain.chain;
- display_frame.setup_chain = chain.setup_chain;
+ display_frame.chain = move(chain.chain);
+ display_frame.setup_chain = move(chain.setup_chain);
display_frame.ready_fence = fence;
- display_frame.input_frames = chain.input_frames;
+ display_frame.input_frames = move(chain.input_frames);
display_frame.temp_textures = {};
- output_channel[i].output_frame(display_frame);
+ output_channel[i].output_frame(move(display_frame));
}
}
theme->channel_clicked(preview_num);
}
+YCbCrInterpretation Mixer::get_input_ycbcr_interpretation(unsigned card_index) const
+{
+ unique_lock<mutex> lock(card_mutex);
+ return ycbcr_interpretation[card_index];
+}
+
+void Mixer::set_input_ycbcr_interpretation(unsigned card_index, const YCbCrInterpretation &interpretation)
+{
+ unique_lock<mutex> lock(card_mutex);
+ ycbcr_interpretation[card_index] = interpretation;
+}
+
void Mixer::start_mode_scanning(unsigned card_index)
{
assert(card_index < num_cards);
last_mode_scan_change[card_index] = steady_clock::now();
}
-map<uint32_t, bmusb::VideoMode> Mixer::get_available_output_video_modes() const
+map<uint32_t, VideoMode> Mixer::get_available_output_video_modes() const
{
assert(desired_output_card_index != -1);
unique_lock<mutex> lock(card_mutex);
}
}
-void Mixer::OutputChannel::output_frame(DisplayFrame frame)
+void Mixer::OutputChannel::output_frame(DisplayFrame &&frame)
{
// Store this frame for display. Remove the ready frame if any
// (it was seemingly never used).
if (has_ready_frame) {
parent->release_display_frame(&ready_frame);
}
- ready_frame = frame;
+ ready_frame = move(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
}
if (has_ready_frame) {
assert(!has_current_frame);
- current_frame = ready_frame;
+ current_frame = move(ready_frame);
ready_frame.ready_fence.reset(); // Drop the refcount.
ready_frame.input_frames.clear(); // Drop the refcounts.
has_current_frame = true;
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)
+{
+ unique_lock<mutex> lock(frame_mutex);
+ new_frame_ready_callbacks[key] = callback;
+}
+
+void Mixer::OutputChannel::remove_frame_ready_callback(void *key)
{
- new_frame_ready_callback = callback;
+ unique_lock<mutex> lock(frame_mutex);
+ new_frame_ready_callbacks.erase(key);
}
void Mixer::OutputChannel::set_transition_names_updated_callback(Mixer::transition_names_updated_callback_t callback)