1 #include <movit/effect_util.h>
2 #include <movit/util.h>
3 #include <movit/resource_pool.h> // Must be above the Xlib includes.
8 #include "chroma_subsampler.h"
9 #include "decklink_output.h"
10 #include "decklink_util.h"
12 #include "print_latency.h"
13 #include "resource_pool.h"
16 using namespace movit;
18 using namespace std::chrono;
20 DeckLinkOutput::DeckLinkOutput(ResourcePool *resource_pool, QSurface *surface, unsigned width, unsigned height, unsigned card_index)
21 : resource_pool(resource_pool), surface(surface), width(width), height(height), card_index(card_index)
23 chroma_subsampler.reset(new ChromaSubsampler(resource_pool));
26 void DeckLinkOutput::set_device(IDeckLink *decklink)
28 if (decklink->QueryInterface(IID_IDeckLinkOutput, (void**)&output) != S_OK) {
29 fprintf(stderr, "Card %u has no outputs\n", card_index);
33 IDeckLinkDisplayModeIterator *mode_it;
34 if (output->GetDisplayModeIterator(&mode_it) != S_OK) {
35 fprintf(stderr, "Failed to enumerate output display modes for card %u\n", card_index);
41 for (const auto &it : summarize_video_modes(mode_it, card_index)) {
42 if (it.second.width != width || it.second.height != height) {
46 // We could support interlaced modes, but let's stay out of it for now,
47 // since we don't have interlaced stream output.
48 if (it.second.interlaced) {
52 video_modes.insert(it);
57 // HDMI or SDI generally mean “both HDMI and SDI at the same time” on DeckLink cards
58 // that support both; pick_default_video_connection() will generally pick one of those
59 // if they exist. We're not very likely to need analog outputs, so we don't need a way
60 // to change beyond that.
61 video_connection = pick_default_video_connection(decklink, BMDDeckLinkVideoOutputConnections, card_index);
64 void DeckLinkOutput::start_output(uint32_t mode, int64_t base_pts)
67 assert(!playback_initiated);
70 playback_initiated = true;
71 playback_started = false;
72 this->base_pts = base_pts;
74 IDeckLinkConfiguration *config = nullptr;
75 if (output->QueryInterface(IID_IDeckLinkConfiguration, (void**)&config) != S_OK) {
76 fprintf(stderr, "Failed to get configuration interface for output card\n");
79 if (config->SetFlag(bmdDeckLinkConfigLowLatencyVideoOutput, true) != S_OK) {
80 fprintf(stderr, "Failed to set low latency output\n");
83 if (config->SetInt(bmdDeckLinkConfigVideoOutputConnection, video_connection) != S_OK) {
84 fprintf(stderr, "Failed to set video output connection for card %u\n", card_index);
87 if (config->SetFlag(bmdDeckLinkConfigUse1080pNotPsF, true) != S_OK) {
88 fprintf(stderr, "Failed to set PsF flag for card\n");
92 BMDDisplayModeSupport support;
93 IDeckLinkDisplayMode *display_mode;
94 if (output->DoesSupportVideoMode(mode, bmdFormat8BitYUV, bmdVideoOutputFlagDefault,
95 &support, &display_mode) != S_OK) {
96 fprintf(stderr, "Couldn't ask for format support\n");
100 if (support == bmdDisplayModeNotSupported) {
101 fprintf(stderr, "Requested display mode not supported\n");
105 BMDDisplayModeFlags flags = display_mode->GetFlags();
106 if ((flags & bmdDisplayModeColorspaceRec601) && global_flags.ycbcr_rec709_coefficients) {
107 fprintf(stderr, "WARNING: Chosen output mode expects Rec. 601 Y'CbCr coefficients.\n");
108 fprintf(stderr, " Consider --output-ycbcr-coefficients=rec601 (or =auto).\n");
109 } else if ((flags & bmdDisplayModeColorspaceRec709) && !global_flags.ycbcr_rec709_coefficients) {
110 fprintf(stderr, "WARNING: Chosen output mode expects Rec. 709 Y'CbCr coefficients.\n");
111 fprintf(stderr, " Consider --output-ycbcr-coefficients=rec709 (or =auto).\n");
114 BMDTimeValue time_value;
115 BMDTimeScale time_scale;
116 if (display_mode->GetFrameRate(&time_value, &time_scale) != S_OK) {
117 fprintf(stderr, "Couldn't get frame rate\n");
121 frame_duration = time_value * TIMEBASE / time_scale;
123 display_mode->Release();
125 HRESULT result = output->EnableVideoOutput(mode, bmdVideoOutputFlagDefault);
126 if (result != S_OK) {
127 fprintf(stderr, "Couldn't enable output with error 0x%x\n", result);
130 if (output->SetScheduledFrameCompletionCallback(this) != S_OK) {
131 fprintf(stderr, "Couldn't set callback\n");
134 assert(OUTPUT_FREQUENCY == 48000);
135 if (output->EnableAudioOutput(bmdAudioSampleRate48kHz, bmdAudioSampleType32bitInteger, 2, bmdAudioOutputStreamTimestamped) != S_OK) {
136 fprintf(stderr, "Couldn't enable audio output\n");
139 if (output->BeginAudioPreroll() != S_OK) {
140 fprintf(stderr, "Couldn't begin audio preroll\n");
144 present_thread = thread([this]{
145 QOpenGLContext *context = create_context(this->surface);
146 eglBindAPI(EGL_OPENGL_API);
147 if (!make_current(context, this->surface)) {
148 printf("display=%p surface=%p context=%p curr=%p err=%d\n", eglGetCurrentDisplay(), this->surface, context, eglGetCurrentContext(),
152 present_thread_func();
153 delete_context(context);
157 void DeckLinkOutput::end_output()
159 if (!playback_initiated) {
164 frame_queues_changed.notify_all();
165 present_thread.join();
166 playback_initiated = false;
168 output->StopScheduledPlayback(0, nullptr, 0);
169 output->DisableVideoOutput();
170 output->DisableAudioOutput();
172 // Wait until all frames are accounted for, and free them.
174 unique_lock<mutex> lock(frame_queue_mutex);
175 while (!(frame_freelist.empty() && num_frames_in_flight == 0)) {
176 frame_queues_changed.wait(lock, [this]{ return !frame_freelist.empty(); });
177 frame_freelist.pop();
182 void DeckLinkOutput::send_frame(GLuint y_tex, GLuint cbcr_tex, const vector<RefCountedFrame> &input_frames, int64_t pts, int64_t duration)
184 assert(!should_quit);
186 unique_ptr<Frame> frame = move(get_frame());
187 chroma_subsampler->create_uyvy(y_tex, cbcr_tex, width, height, frame->uyvy_tex);
189 // Download the UYVY texture to the PBO.
190 glPixelStorei(GL_PACK_ROW_LENGTH, 0);
193 glBindBuffer(GL_PIXEL_PACK_BUFFER, frame->pbo);
196 glBindTexture(GL_TEXTURE_2D, frame->uyvy_tex);
198 glGetTexImage(GL_TEXTURE_2D, 0, GL_BGRA, GL_UNSIGNED_INT_8_8_8_8_REV, BUFFER_OFFSET(0));
201 glBindTexture(GL_TEXTURE_2D, 0);
203 glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
206 glMemoryBarrier(GL_TEXTURE_UPDATE_BARRIER_BIT | GL_CLIENT_MAPPED_BUFFER_BARRIER_BIT);
209 frame->fence = RefCountedGLsync(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);
211 glFlush(); // Make the DeckLink thread see the fence as soon as possible.
214 frame->input_frames = input_frames;
215 frame->received_ts = find_received_timestamp(input_frames);
217 frame->duration = duration;
220 unique_lock<mutex> lock(frame_queue_mutex);
221 pending_video_frames.push(move(frame));
223 frame_queues_changed.notify_all();
226 void DeckLinkOutput::send_audio(int64_t pts, const std::vector<float> &samples)
228 unique_ptr<int32_t[]> int_samples(new int32_t[samples.size()]);
229 for (size_t i = 0; i < samples.size(); ++i) {
230 int_samples[i] = lrintf(samples[i] * 2147483648.0f);
233 uint32_t frames_written;
234 HRESULT result = output->ScheduleAudioSamples(int_samples.get(), samples.size() / 2,
235 pts, TIMEBASE, &frames_written);
236 if (result != S_OK) {
237 fprintf(stderr, "ScheduleAudioSamples(pts=%ld) failed (result=0x%08x)\n", pts, result);
239 if (frames_written != samples.size() / 2) {
240 fprintf(stderr, "ScheduleAudioSamples() returned short write (%u/%ld)\n", frames_written, samples.size() / 2);
245 void DeckLinkOutput::wait_for_frame(int64_t pts, int *dropped_frames, int64_t *frame_duration, bool *is_preroll, steady_clock::time_point *frame_timestamp)
247 assert(!should_quit);
250 *frame_duration = this->frame_duration;
252 const BMDTimeValue buffer = lrint(*frame_duration * global_flags.output_buffer_frames);
253 const BMDTimeValue max_overshoot = lrint(*frame_duration * global_flags.output_slop_frames);
254 BMDTimeValue target_time = pts - buffer;
256 // While prerolling, we send out frames as quickly as we can.
257 if (target_time < base_pts) {
262 *is_preroll = !playback_started;
264 if (!playback_started) {
265 if (output->EndAudioPreroll() != S_OK) {
266 fprintf(stderr, "Could not end audio preroll\n");
269 if (output->StartScheduledPlayback(base_pts, TIMEBASE, 1.0) != S_OK) {
270 fprintf(stderr, "Could not start playback\n");
273 playback_started = true;
276 BMDTimeValue stream_frame_time;
277 double playback_speed;
278 output->GetScheduledStreamTime(TIMEBASE, &stream_frame_time, &playback_speed);
280 *frame_timestamp = steady_clock::now() +
281 nanoseconds((target_time - stream_frame_time) * 1000000000 / TIMEBASE);
283 // If we're ahead of time, wait for the frame to (approximately) start.
284 if (stream_frame_time < target_time) {
285 this_thread::sleep_until(*frame_timestamp);
289 // If we overshot the previous frame by just a little,
290 // fire off one immediately.
291 if (stream_frame_time < target_time + max_overshoot) {
292 fprintf(stderr, "Warning: Frame was %ld ms late (but not skipping it due to --output-slop-frames).\n",
293 lrint((stream_frame_time - target_time) * 1000.0 / TIMEBASE));
297 // Oops, we missed by more than one frame. Return immediately,
298 // but drop so that we catch up.
299 *dropped_frames = (stream_frame_time - target_time + *frame_duration - 1) / *frame_duration;
300 const int64_t ns_per_frame = this->frame_duration * 1000000000 / TIMEBASE;
301 *frame_timestamp += nanoseconds(*dropped_frames * ns_per_frame);
302 fprintf(stderr, "Dropped %d output frames; skipping.\n", *dropped_frames);
305 HRESULT DeckLinkOutput::ScheduledFrameCompleted(/* in */ IDeckLinkVideoFrame *completedFrame, /* in */ BMDOutputFrameCompletionResult result)
307 Frame *frame = static_cast<Frame *>(completedFrame);
309 case bmdOutputFrameCompleted:
311 case bmdOutputFrameDisplayedLate:
312 fprintf(stderr, "Output frame displayed late (pts=%ld)\n", frame->pts);
313 fprintf(stderr, "Consider increasing --output-buffer-frames if this persists.\n");
315 case bmdOutputFrameDropped:
316 fprintf(stderr, "Output frame was dropped (pts=%ld)\n", frame->pts);
317 fprintf(stderr, "Consider increasing --output-buffer-frames if this persists.\n");
319 case bmdOutputFrameFlushed:
320 fprintf(stderr, "Output frame was flushed (pts=%ld)\n", frame->pts);
323 fprintf(stderr, "Output frame completed with unknown status %d\n", result);
328 print_latency("DeckLink output latency (frame received → output on HDMI):", frame->received_ts, false, &hei);
331 lock_guard<mutex> lock(frame_queue_mutex);
332 frame_freelist.push(unique_ptr<Frame>(frame));
333 --num_frames_in_flight;
339 HRESULT DeckLinkOutput::ScheduledPlaybackHasStopped()
341 printf("playback stopped!\n");
345 unique_ptr<DeckLinkOutput::Frame> DeckLinkOutput::get_frame()
347 lock_guard<mutex> lock(frame_queue_mutex);
349 if (!frame_freelist.empty()) {
350 unique_ptr<Frame> frame = move(frame_freelist.front());
351 frame_freelist.pop();
355 unique_ptr<Frame> frame(new Frame);
357 frame->uyvy_tex = resource_pool->create_2d_texture(GL_RGBA8, width / 2, height);
359 glGenBuffers(1, &frame->pbo);
361 glBindBuffer(GL_PIXEL_PACK_BUFFER, frame->pbo);
363 glBufferStorage(GL_PIXEL_PACK_BUFFER, width * height * 2, NULL, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
365 frame->uyvy_ptr = (uint8_t *)glMapBufferRange(GL_PIXEL_PACK_BUFFER, 0, width * height * 2, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
367 frame->uyvy_ptr_local.reset(new uint8_t[width * height * 2]);
368 frame->resource_pool = resource_pool;
373 void DeckLinkOutput::present_thread_func()
375 pthread_setname_np(pthread_self(), "DeckLinkOutput");
377 unique_ptr<Frame> frame;
379 unique_lock<mutex> lock(frame_queue_mutex);
380 frame_queues_changed.wait(lock, [this]{
381 return should_quit || !pending_video_frames.empty();
386 frame = move(pending_video_frames.front());
387 pending_video_frames.pop();
388 ++num_frames_in_flight;
391 glWaitSync(frame->fence.get(), /*flags=*/0, GL_TIMEOUT_IGNORED);
393 frame->fence.reset();
395 memcpy(frame->uyvy_ptr_local.get(), frame->uyvy_ptr, width * height * 2);
397 // Release any input frames we needed to render this frame.
398 frame->input_frames.clear();
400 BMDTimeValue pts = frame->pts;
401 BMDTimeValue duration = frame->duration;
402 HRESULT res = output->ScheduleVideoFrame(frame.get(), pts, duration, TIMEBASE);
404 frame.release(); // Owned by the driver now.
406 fprintf(stderr, "Could not schedule video frame! (error=0x%08x)\n", res);
408 lock_guard<mutex> lock(frame_queue_mutex);
409 frame_freelist.push(move(frame));
410 --num_frames_in_flight;
415 HRESULT STDMETHODCALLTYPE DeckLinkOutput::QueryInterface(REFIID, LPVOID *)
417 return E_NOINTERFACE;
420 ULONG STDMETHODCALLTYPE DeckLinkOutput::AddRef()
422 return refcount.fetch_add(1) + 1;
425 ULONG STDMETHODCALLTYPE DeckLinkOutput::Release()
427 int new_ref = refcount.fetch_sub(1) - 1;
433 DeckLinkOutput::Frame::~Frame()
435 glBindBuffer(GL_PIXEL_PACK_BUFFER, pbo);
437 glUnmapBuffer(GL_PIXEL_PACK_BUFFER);
439 glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
441 glDeleteBuffers(1, &pbo);
443 resource_pool->release_2d_texture(uyvy_tex);
447 HRESULT STDMETHODCALLTYPE DeckLinkOutput::Frame::QueryInterface(REFIID, LPVOID *)
449 return E_NOINTERFACE;
452 ULONG STDMETHODCALLTYPE DeckLinkOutput::Frame::AddRef()
454 return refcount.fetch_add(1) + 1;
457 ULONG STDMETHODCALLTYPE DeckLinkOutput::Frame::Release()
459 int new_ref = refcount.fetch_sub(1) - 1;
465 long DeckLinkOutput::Frame::GetWidth()
467 return global_flags.width;
470 long DeckLinkOutput::Frame::GetHeight()
472 return global_flags.height;
475 long DeckLinkOutput::Frame::GetRowBytes()
477 return global_flags.width * 2;
480 BMDPixelFormat DeckLinkOutput::Frame::GetPixelFormat()
482 return bmdFormat8BitYUV;
485 BMDFrameFlags DeckLinkOutput::Frame::GetFlags()
487 return bmdFrameFlagDefault;
490 HRESULT DeckLinkOutput::Frame::GetBytes(/* out */ void **buffer)
492 *buffer = uyvy_ptr_local.get();
496 HRESULT DeckLinkOutput::Frame::GetTimecode(/* in */ BMDTimecodeFormat format, /* out */ IDeckLinkTimecode **timecode)
498 fprintf(stderr, "STUB: GetTimecode()\n");
502 HRESULT DeckLinkOutput::Frame::GetAncillaryData(/* out */ IDeckLinkVideoFrameAncillary **ancillary)
504 fprintf(stderr, "STUB: GetAncillaryData()\n");