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");
91 if (config->SetFlag(bmdDeckLinkConfigSMPTELevelAOutput, true) != S_OK) {
92 // This affects at least some no-name SDI->HDMI converters.
93 // Warn, but don't die.
94 fprintf(stderr, "WARNING: Failed to enable SMTPE Level A; resolutions like 1080p60 might have issues.\n");
97 BMDDisplayModeSupport support;
98 IDeckLinkDisplayMode *display_mode;
99 if (output->DoesSupportVideoMode(mode, bmdFormat8BitYUV, bmdVideoOutputFlagDefault,
100 &support, &display_mode) != S_OK) {
101 fprintf(stderr, "Couldn't ask for format support\n");
105 if (support == bmdDisplayModeNotSupported) {
106 fprintf(stderr, "Requested display mode not supported\n");
110 BMDDisplayModeFlags flags = display_mode->GetFlags();
111 if ((flags & bmdDisplayModeColorspaceRec601) && global_flags.ycbcr_rec709_coefficients) {
112 fprintf(stderr, "WARNING: Chosen output mode expects Rec. 601 Y'CbCr coefficients.\n");
113 fprintf(stderr, " Consider --output-ycbcr-coefficients=rec601 (or =auto).\n");
114 } else if ((flags & bmdDisplayModeColorspaceRec709) && !global_flags.ycbcr_rec709_coefficients) {
115 fprintf(stderr, "WARNING: Chosen output mode expects Rec. 709 Y'CbCr coefficients.\n");
116 fprintf(stderr, " Consider --output-ycbcr-coefficients=rec709 (or =auto).\n");
119 BMDTimeValue time_value;
120 BMDTimeScale time_scale;
121 if (display_mode->GetFrameRate(&time_value, &time_scale) != S_OK) {
122 fprintf(stderr, "Couldn't get frame rate\n");
126 frame_duration = time_value * TIMEBASE / time_scale;
128 display_mode->Release();
130 HRESULT result = output->EnableVideoOutput(mode, bmdVideoOutputFlagDefault);
131 if (result != S_OK) {
132 fprintf(stderr, "Couldn't enable output with error 0x%x\n", result);
135 if (output->SetScheduledFrameCompletionCallback(this) != S_OK) {
136 fprintf(stderr, "Couldn't set callback\n");
139 assert(OUTPUT_FREQUENCY == 48000);
140 if (output->EnableAudioOutput(bmdAudioSampleRate48kHz, bmdAudioSampleType32bitInteger, 2, bmdAudioOutputStreamTimestamped) != S_OK) {
141 fprintf(stderr, "Couldn't enable audio output\n");
144 if (output->BeginAudioPreroll() != S_OK) {
145 fprintf(stderr, "Couldn't begin audio preroll\n");
149 present_thread = thread([this]{
150 QOpenGLContext *context = create_context(this->surface);
151 eglBindAPI(EGL_OPENGL_API);
152 if (!make_current(context, this->surface)) {
153 printf("display=%p surface=%p context=%p curr=%p err=%d\n", eglGetCurrentDisplay(), this->surface, context, eglGetCurrentContext(),
157 present_thread_func();
158 delete_context(context);
162 void DeckLinkOutput::end_output()
164 if (!playback_initiated) {
169 frame_queues_changed.notify_all();
170 present_thread.join();
171 playback_initiated = false;
173 output->StopScheduledPlayback(0, nullptr, 0);
174 output->DisableVideoOutput();
175 output->DisableAudioOutput();
177 // Wait until all frames are accounted for, and free them.
179 unique_lock<mutex> lock(frame_queue_mutex);
180 while (!(frame_freelist.empty() && num_frames_in_flight == 0)) {
181 frame_queues_changed.wait(lock, [this]{ return !frame_freelist.empty(); });
182 frame_freelist.pop();
187 void DeckLinkOutput::send_frame(GLuint y_tex, GLuint cbcr_tex, const vector<RefCountedFrame> &input_frames, int64_t pts, int64_t duration)
189 assert(!should_quit);
191 unique_ptr<Frame> frame = move(get_frame());
192 chroma_subsampler->create_uyvy(y_tex, cbcr_tex, width, height, frame->uyvy_tex);
194 // Download the UYVY texture to the PBO.
195 glPixelStorei(GL_PACK_ROW_LENGTH, 0);
198 glBindBuffer(GL_PIXEL_PACK_BUFFER, frame->pbo);
201 glBindTexture(GL_TEXTURE_2D, frame->uyvy_tex);
203 glGetTexImage(GL_TEXTURE_2D, 0, GL_BGRA, GL_UNSIGNED_INT_8_8_8_8_REV, BUFFER_OFFSET(0));
206 glBindTexture(GL_TEXTURE_2D, 0);
208 glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
211 glMemoryBarrier(GL_TEXTURE_UPDATE_BARRIER_BIT | GL_CLIENT_MAPPED_BUFFER_BARRIER_BIT);
214 frame->fence = RefCountedGLsync(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);
216 glFlush(); // Make the DeckLink thread see the fence as soon as possible.
219 frame->input_frames = input_frames;
220 frame->received_ts = find_received_timestamp(input_frames);
222 frame->duration = duration;
225 unique_lock<mutex> lock(frame_queue_mutex);
226 pending_video_frames.push(move(frame));
228 frame_queues_changed.notify_all();
231 void DeckLinkOutput::send_audio(int64_t pts, const std::vector<float> &samples)
233 unique_ptr<int32_t[]> int_samples(new int32_t[samples.size()]);
234 for (size_t i = 0; i < samples.size(); ++i) {
235 int_samples[i] = lrintf(samples[i] * 2147483648.0f);
238 uint32_t frames_written;
239 HRESULT result = output->ScheduleAudioSamples(int_samples.get(), samples.size() / 2,
240 pts, TIMEBASE, &frames_written);
241 if (result != S_OK) {
242 fprintf(stderr, "ScheduleAudioSamples(pts=%ld) failed (result=0x%08x)\n", pts, result);
244 if (frames_written != samples.size() / 2) {
245 fprintf(stderr, "ScheduleAudioSamples() returned short write (%u/%ld)\n", frames_written, samples.size() / 2);
250 void DeckLinkOutput::wait_for_frame(int64_t pts, int *dropped_frames, int64_t *frame_duration, bool *is_preroll, steady_clock::time_point *frame_timestamp)
252 assert(!should_quit);
255 *frame_duration = this->frame_duration;
257 const BMDTimeValue buffer = lrint(*frame_duration * global_flags.output_buffer_frames);
258 const BMDTimeValue max_overshoot = lrint(*frame_duration * global_flags.output_slop_frames);
259 BMDTimeValue target_time = pts - buffer;
261 // While prerolling, we send out frames as quickly as we can.
262 if (target_time < base_pts) {
267 *is_preroll = !playback_started;
269 if (!playback_started) {
270 if (output->EndAudioPreroll() != S_OK) {
271 fprintf(stderr, "Could not end audio preroll\n");
274 if (output->StartScheduledPlayback(base_pts, TIMEBASE, 1.0) != S_OK) {
275 fprintf(stderr, "Could not start playback\n");
278 playback_started = true;
281 BMDTimeValue stream_frame_time;
282 double playback_speed;
283 output->GetScheduledStreamTime(TIMEBASE, &stream_frame_time, &playback_speed);
285 *frame_timestamp = steady_clock::now() +
286 nanoseconds((target_time - stream_frame_time) * 1000000000 / TIMEBASE);
288 // If we're ahead of time, wait for the frame to (approximately) start.
289 if (stream_frame_time < target_time) {
290 this_thread::sleep_until(*frame_timestamp);
294 // If we overshot the previous frame by just a little,
295 // fire off one immediately.
296 if (stream_frame_time < target_time + max_overshoot) {
297 fprintf(stderr, "Warning: Frame was %ld ms late (but not skipping it due to --output-slop-frames).\n",
298 lrint((stream_frame_time - target_time) * 1000.0 / TIMEBASE));
302 // Oops, we missed by more than one frame. Return immediately,
303 // but drop so that we catch up.
304 *dropped_frames = (stream_frame_time - target_time + *frame_duration - 1) / *frame_duration;
305 const int64_t ns_per_frame = this->frame_duration * 1000000000 / TIMEBASE;
306 *frame_timestamp += nanoseconds(*dropped_frames * ns_per_frame);
307 fprintf(stderr, "Dropped %d output frames; skipping.\n", *dropped_frames);
310 uint32_t DeckLinkOutput::pick_video_mode(uint32_t mode) const
312 if (video_modes.count(mode)) {
316 // Prioritize 59.94 > 60 > 29.97. If none of those are found, then pick the highest one.
317 for (const pair<int, int> &desired : vector<pair<int, int>>{ { 60000, 1001 }, { 60, 0 }, { 30000, 1001 } }) {
318 for (const auto &it : video_modes) {
319 if (it.second.frame_rate_num * desired.second == desired.first * it.second.frame_rate_den) {
325 uint32_t best_mode = 0;
326 double best_fps = 0.0;
327 for (const auto &it : video_modes) {
328 double fps = double(it.second.frame_rate_num) / it.second.frame_rate_den;
329 if (fps > best_fps) {
330 best_mode = it.first;
337 HRESULT DeckLinkOutput::ScheduledFrameCompleted(/* in */ IDeckLinkVideoFrame *completedFrame, /* in */ BMDOutputFrameCompletionResult result)
339 Frame *frame = static_cast<Frame *>(completedFrame);
341 case bmdOutputFrameCompleted:
343 case bmdOutputFrameDisplayedLate:
344 fprintf(stderr, "Output frame displayed late (pts=%ld)\n", frame->pts);
345 fprintf(stderr, "Consider increasing --output-buffer-frames if this persists.\n");
347 case bmdOutputFrameDropped:
348 fprintf(stderr, "Output frame was dropped (pts=%ld)\n", frame->pts);
349 fprintf(stderr, "Consider increasing --output-buffer-frames if this persists.\n");
351 case bmdOutputFrameFlushed:
352 fprintf(stderr, "Output frame was flushed (pts=%ld)\n", frame->pts);
355 fprintf(stderr, "Output frame completed with unknown status %d\n", result);
359 static int frameno = 0;
360 print_latency("DeckLink output latency (frame received → output on HDMI):", frame->received_ts, false, &frameno);
363 lock_guard<mutex> lock(frame_queue_mutex);
364 frame_freelist.push(unique_ptr<Frame>(frame));
365 --num_frames_in_flight;
371 HRESULT DeckLinkOutput::ScheduledPlaybackHasStopped()
373 printf("playback stopped!\n");
377 unique_ptr<DeckLinkOutput::Frame> DeckLinkOutput::get_frame()
379 lock_guard<mutex> lock(frame_queue_mutex);
381 if (!frame_freelist.empty()) {
382 unique_ptr<Frame> frame = move(frame_freelist.front());
383 frame_freelist.pop();
387 unique_ptr<Frame> frame(new Frame);
389 frame->uyvy_tex = resource_pool->create_2d_texture(GL_RGBA8, width / 2, height);
391 glGenBuffers(1, &frame->pbo);
393 glBindBuffer(GL_PIXEL_PACK_BUFFER, frame->pbo);
395 glBufferStorage(GL_PIXEL_PACK_BUFFER, width * height * 2, NULL, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
397 frame->uyvy_ptr = (uint8_t *)glMapBufferRange(GL_PIXEL_PACK_BUFFER, 0, width * height * 2, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
399 frame->uyvy_ptr_local.reset(new uint8_t[width * height * 2]);
400 frame->resource_pool = resource_pool;
405 void DeckLinkOutput::present_thread_func()
407 pthread_setname_np(pthread_self(), "DeckLinkOutput");
409 unique_ptr<Frame> frame;
411 unique_lock<mutex> lock(frame_queue_mutex);
412 frame_queues_changed.wait(lock, [this]{
413 return should_quit || !pending_video_frames.empty();
418 frame = move(pending_video_frames.front());
419 pending_video_frames.pop();
420 ++num_frames_in_flight;
423 glWaitSync(frame->fence.get(), /*flags=*/0, GL_TIMEOUT_IGNORED);
425 frame->fence.reset();
427 memcpy(frame->uyvy_ptr_local.get(), frame->uyvy_ptr, width * height * 2);
429 // Release any input frames we needed to render this frame.
430 frame->input_frames.clear();
432 BMDTimeValue pts = frame->pts;
433 BMDTimeValue duration = frame->duration;
434 HRESULT res = output->ScheduleVideoFrame(frame.get(), pts, duration, TIMEBASE);
436 frame.release(); // Owned by the driver now.
438 fprintf(stderr, "Could not schedule video frame! (error=0x%08x)\n", res);
440 lock_guard<mutex> lock(frame_queue_mutex);
441 frame_freelist.push(move(frame));
442 --num_frames_in_flight;
447 HRESULT STDMETHODCALLTYPE DeckLinkOutput::QueryInterface(REFIID, LPVOID *)
449 return E_NOINTERFACE;
452 ULONG STDMETHODCALLTYPE DeckLinkOutput::AddRef()
454 return refcount.fetch_add(1) + 1;
457 ULONG STDMETHODCALLTYPE DeckLinkOutput::Release()
459 int new_ref = refcount.fetch_sub(1) - 1;
465 DeckLinkOutput::Frame::~Frame()
467 glBindBuffer(GL_PIXEL_PACK_BUFFER, pbo);
469 glUnmapBuffer(GL_PIXEL_PACK_BUFFER);
471 glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
473 glDeleteBuffers(1, &pbo);
475 resource_pool->release_2d_texture(uyvy_tex);
479 HRESULT STDMETHODCALLTYPE DeckLinkOutput::Frame::QueryInterface(REFIID, LPVOID *)
481 return E_NOINTERFACE;
484 ULONG STDMETHODCALLTYPE DeckLinkOutput::Frame::AddRef()
486 return refcount.fetch_add(1) + 1;
489 ULONG STDMETHODCALLTYPE DeckLinkOutput::Frame::Release()
491 int new_ref = refcount.fetch_sub(1) - 1;
497 long DeckLinkOutput::Frame::GetWidth()
499 return global_flags.width;
502 long DeckLinkOutput::Frame::GetHeight()
504 return global_flags.height;
507 long DeckLinkOutput::Frame::GetRowBytes()
509 return global_flags.width * 2;
512 BMDPixelFormat DeckLinkOutput::Frame::GetPixelFormat()
514 return bmdFormat8BitYUV;
517 BMDFrameFlags DeckLinkOutput::Frame::GetFlags()
519 return bmdFrameFlagDefault;
522 HRESULT DeckLinkOutput::Frame::GetBytes(/* out */ void **buffer)
524 *buffer = uyvy_ptr_local.get();
528 HRESULT DeckLinkOutput::Frame::GetTimecode(/* in */ BMDTimecodeFormat format, /* out */ IDeckLinkTimecode **timecode)
530 fprintf(stderr, "STUB: GetTimecode()\n");
534 HRESULT DeckLinkOutput::Frame::GetAncillaryData(/* out */ IDeckLinkVideoFrameAncillary **ancillary)
536 fprintf(stderr, "STUB: GetAncillaryData()\n");