#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"
}
}
-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;
- 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) {
+ 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])) {
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));
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;
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));
+
// Start listening for clients only once VideoEncoder has written its header, if any.
httpd.start(9095);
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);
+ 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();
+
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; ++card_index) {
cards[card_index].queue_length_policy.reset(card_index);
}
desired_output_card_index = global_flags.output_card;
set_output_card_internal(global_flags.output_card);
}
+
+ global_metrics.register_int_metric("num_frames", &metrics_num_frames);
+ global_metrics.register_int_metric("dropped_frames", &metrics_dropped_frames);
+ global_metrics.register_double_metric("uptime", &metrics_uptime);
}
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; ++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 (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().
// 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();
}
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->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) {
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;
}
- // For 8-bit input, v210_width will be nonsensical but not used.
+ // 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();
- 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], 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);
// 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; ++card_index) {
if (int(card_index) != output_card_index) {
cards[card_index].capture->start_bm_capture();
}
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; ++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; ++card_index) {
if (!has_new_frame[card_index] || new_frames[card_index].frame->len == 0)
continue;
now = steady_clock::now();
double elapsed = duration<double>(now - start).count();
+
+ metrics_num_frames = frame_num;
+ metrics_dropped_frames = stats_dropped_frames;
+ metrics_uptime = elapsed;
+
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,
++dropped_frames;
}
+#if 0
if (dropped_frames > 0) {
fprintf(stderr, "Card %u dropped %u frame(s) to keep latency down.\n",
card_index, dropped_frames);
}
+#endif
}
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; ++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,
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);
+ // The theme can't (or at least shouldn't!) call connect_signal() on
+ // each FFmpeg input, so we'll do it here.
+ for (const pair<LiveInputWrapper *, FFmpegCapture *> &conn : theme->get_signal_connections()) {
+ conn.first->connect_signal_raw(conn.second->get_card_index());
+ }
+
// 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;
output_ycbcr_format.num_levels = 1 << global_flags.x264_bit_depth;
chain->change_ycbcr_output_format(output_ycbcr_format);
- const int64_t av_delay = lrint(global_flags.audio_queue_length_ms * 0.001 * TIMEBASE); // Corresponds to the delay in ResamplingQueue.
- GLuint y_tex, cbcr_tex;
- 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. We take an extra copy of the created outputs,
+ // 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 if they're used for zero-copy Quick Sync encoding
- // (the default case), they're just views into VA-API memory and must be
+ // 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 cbcr_full_tex, cbcr_copy_tex, y_copy_tex;
- if (global_flags.x264_bit_depth > 8) {
- cbcr_full_tex = resource_pool->create_2d_texture(GL_RG16, global_flags.width, global_flags.height);
- y_copy_tex = resource_pool->create_2d_texture(GL_R16, global_flags.width, global_flags.height);
- cbcr_copy_tex = resource_pool->create_2d_texture(GL_RG16, global_flags.width / 2, global_flags.height / 2);
+ 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(GL_RG8, global_flags.width, global_flags.height);
- y_copy_tex = resource_pool->create_2d_texture(GL_R8, global_flags.width, global_flags.height);
- cbcr_copy_tex = resource_pool->create_2d_texture(GL_RG8, global_flags.width / 2, global_flags.height / 2);
+ 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);
+
+ 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);
}
- GLuint fbo = resource_pool->create_fbo(y_tex, cbcr_full_tex, y_copy_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, cbcr_copy_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, ycbcr_output_coefficients, theme_main_chain.input_frames, pts_int, duration);
}
resource_pool->release_2d_texture(cbcr_full_tex);
- // Set the right state for the Y' and CbCr copies.
+ // Set the right state for the Y' and CbCr textures we use for display.
glBindFramebuffer(GL_FRAMEBUFFER, 0);
- glBindTexture(GL_TEXTURE_2D, y_copy_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);
- glBindTexture(GL_TEXTURE_2D, cbcr_copy_tex);
+ 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 pieces the Y'CbCr texture copies back into RGB and displays them.
- // It owns y_copy_tex and cbcr_copy_tex now.
+ // 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, y_copy_tex, cbcr_copy_tex]{
- display_input->set_texture_num(0, y_copy_tex);
- display_input->set_texture_num(1, cbcr_copy_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 = { y_copy_tex, cbcr_copy_tex };
+ live_frame.temp_textures = { y_display_tex, cbcr_display_tex };
output_channel[OUTPUT_LIVE].output_frame(live_frame);
// Set up preview and any additional channels.
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);
}
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
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)