1 #include <movit/effect_util.h>
2 #include <movit/util.h>
3 #include <movit/resource_pool.h> // Must be above the Xlib includes.
7 #include "chroma_subsampler.h"
8 #include "decklink_output.h"
9 #include "decklink_util.h"
11 #include "print_latency.h"
12 #include "resource_pool.h"
15 using namespace movit;
17 using namespace std::chrono;
19 DeckLinkOutput::DeckLinkOutput(ResourcePool *resource_pool, QSurface *surface, unsigned width, unsigned height, unsigned card_index)
20 : resource_pool(resource_pool), surface(surface), width(width), height(height), card_index(card_index)
22 chroma_subsampler.reset(new ChromaSubsampler(resource_pool));
25 void DeckLinkOutput::set_device(IDeckLink *decklink)
27 if (decklink->QueryInterface(IID_IDeckLinkOutput, (void**)&output) != S_OK) {
28 fprintf(stderr, "Card %u has no outputs\n", card_index);
32 IDeckLinkDisplayModeIterator *mode_it;
33 if (output->GetDisplayModeIterator(&mode_it) != S_OK) {
34 fprintf(stderr, "Failed to enumerate output display modes for card %u\n", card_index);
40 for (const auto &it : summarize_video_modes(mode_it, card_index)) {
41 if (it.second.width != width || it.second.height != height) {
45 // We could support interlaced modes, but let's stay out of it for now,
46 // since we don't have interlaced stream output.
47 if (it.second.interlaced) {
51 video_modes.insert(it);
56 // HDMI or SDI generally mean “both HDMI and SDI at the same time” on DeckLink cards
57 // that support both; pick_default_video_connection() will generally pick one of those
58 // if they exist. We're not very likely to need analog outputs, so we don't need a way
59 // to change beyond that.
60 video_connection = pick_default_video_connection(decklink, BMDDeckLinkVideoOutputConnections, card_index);
63 void DeckLinkOutput::start_output(uint32_t mode, int64_t base_pts)
66 assert(!playback_initiated);
69 playback_initiated = true;
70 playback_started = false;
71 this->base_pts = base_pts;
73 IDeckLinkConfiguration *config = nullptr;
74 if (output->QueryInterface(IID_IDeckLinkConfiguration, (void**)&config) != S_OK) {
75 fprintf(stderr, "Failed to get configuration interface for output card\n");
78 if (config->SetFlag(bmdDeckLinkConfigLowLatencyVideoOutput, true) != S_OK) {
79 fprintf(stderr, "Failed to set low latency output\n");
82 if (config->SetInt(bmdDeckLinkConfigVideoOutputConnection, video_connection) != S_OK) {
83 fprintf(stderr, "Failed to set video output connection for card %u\n", card_index);
86 if (config->SetFlag(bmdDeckLinkConfigUse1080pNotPsF, true) != S_OK) {
87 fprintf(stderr, "Failed to set PsF flag for card\n");
91 BMDDisplayModeSupport support;
92 IDeckLinkDisplayMode *display_mode;
93 if (output->DoesSupportVideoMode(mode, bmdFormat8BitYUV, bmdVideoOutputFlagDefault,
94 &support, &display_mode) != S_OK) {
95 fprintf(stderr, "Couldn't ask for format support\n");
99 if (support == bmdDisplayModeNotSupported) {
100 fprintf(stderr, "Requested display mode not supported\n");
104 BMDDisplayModeFlags flags = display_mode->GetFlags();
105 if ((flags & bmdDisplayModeColorspaceRec601) && global_flags.ycbcr_rec709_coefficients) {
106 fprintf(stderr, "WARNING: Chosen output mode expects Rec. 601 Y'CbCr coefficients.\n");
107 fprintf(stderr, " Consider --output-ycbcr-coefficients=rec601 (or =auto).\n");
108 } else if ((flags & bmdDisplayModeColorspaceRec709) && !global_flags.ycbcr_rec709_coefficients) {
109 fprintf(stderr, "WARNING: Chosen output mode expects Rec. 709 Y'CbCr coefficients.\n");
110 fprintf(stderr, " Consider --output-ycbcr-coefficients=rec709 (or =auto).\n");
113 BMDTimeValue time_value;
114 BMDTimeScale time_scale;
115 if (display_mode->GetFrameRate(&time_value, &time_scale) != S_OK) {
116 fprintf(stderr, "Couldn't get frame rate\n");
120 frame_duration = time_value * TIMEBASE / time_scale;
122 display_mode->Release();
124 HRESULT result = output->EnableVideoOutput(mode, bmdVideoOutputFlagDefault);
125 if (result != S_OK) {
126 fprintf(stderr, "Couldn't enable output with error 0x%x\n", result);
129 if (output->SetScheduledFrameCompletionCallback(this) != S_OK) {
130 fprintf(stderr, "Couldn't set callback\n");
133 assert(OUTPUT_FREQUENCY == 48000);
134 if (output->EnableAudioOutput(bmdAudioSampleRate48kHz, bmdAudioSampleType32bitInteger, 2, bmdAudioOutputStreamTimestamped) != S_OK) {
135 fprintf(stderr, "Couldn't enable audio output\n");
138 if (output->BeginAudioPreroll() != S_OK) {
139 fprintf(stderr, "Couldn't begin audio preroll\n");
143 present_thread = thread([this]{
144 QOpenGLContext *context = create_context(this->surface);
145 eglBindAPI(EGL_OPENGL_API);
146 if (!make_current(context, this->surface)) {
147 printf("display=%p surface=%p context=%p curr=%p err=%d\n", eglGetCurrentDisplay(), this->surface, context, eglGetCurrentContext(),
151 present_thread_func();
152 delete_context(context);
156 void DeckLinkOutput::end_output()
158 if (!playback_initiated) {
163 frame_queues_changed.notify_all();
164 present_thread.join();
165 playback_initiated = false;
167 output->StopScheduledPlayback(0, nullptr, 0);
168 output->DisableVideoOutput();
169 output->DisableAudioOutput();
171 // Wait until all frames are accounted for, and free them.
173 unique_lock<mutex> lock(frame_queue_mutex);
174 while (!(frame_freelist.empty() && num_frames_in_flight == 0)) {
175 frame_queues_changed.wait(lock, [this]{ return !frame_freelist.empty(); });
176 frame_freelist.pop();
181 void DeckLinkOutput::send_frame(GLuint y_tex, GLuint cbcr_tex, const vector<RefCountedFrame> &input_frames, int64_t pts, int64_t duration)
183 unique_ptr<Frame> frame = move(get_frame());
184 chroma_subsampler->create_uyvy(y_tex, cbcr_tex, width, height, frame->uyvy_tex);
186 // Download the UYVY texture to the PBO.
187 glPixelStorei(GL_PACK_ROW_LENGTH, 0);
190 glBindBuffer(GL_PIXEL_PACK_BUFFER, frame->pbo);
193 glBindTexture(GL_TEXTURE_2D, frame->uyvy_tex);
195 glGetTexImage(GL_TEXTURE_2D, 0, GL_BGRA, GL_UNSIGNED_INT_8_8_8_8_REV, BUFFER_OFFSET(0));
198 glBindTexture(GL_TEXTURE_2D, 0);
200 glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
203 glMemoryBarrier(GL_TEXTURE_UPDATE_BARRIER_BIT | GL_CLIENT_MAPPED_BUFFER_BARRIER_BIT);
206 frame->fence = RefCountedGLsync(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);
208 glFlush(); // Make the DeckLink thread see the fence as soon as possible.
211 frame->input_frames = input_frames;
212 frame->received_ts = find_received_timestamp(input_frames);
214 frame->duration = duration;
217 unique_lock<mutex> lock(frame_queue_mutex);
218 pending_video_frames.push(move(frame));
220 frame_queues_changed.notify_all();
223 void DeckLinkOutput::send_audio(int64_t pts, const std::vector<float> &samples)
225 unique_ptr<int32_t[]> int_samples(new int32_t[samples.size()]);
226 for (size_t i = 0; i < samples.size(); ++i) {
227 int_samples[i] = lrintf(samples[i] * 2147483648.0f);
230 uint32_t frames_written;
231 HRESULT result = output->ScheduleAudioSamples(int_samples.get(), samples.size() / 2,
232 pts, TIMEBASE, &frames_written);
233 if (result != S_OK) {
234 fprintf(stderr, "ScheduleAudioSamples(pts=%ld) failed (result=0x%08x)\n", pts, result);
236 if (frames_written != samples.size() / 2) {
237 fprintf(stderr, "ScheduleAudioSamples() returned short write (%u/%ld)\n", frames_written, samples.size() / 2);
242 void DeckLinkOutput::wait_for_frame(int64_t pts, int *dropped_frames, int64_t *frame_duration)
245 *frame_duration = this->frame_duration;
247 const BMDTimeValue buffer = lrint(*frame_duration * global_flags.output_buffer_frames);
248 const BMDTimeValue max_overshoot = lrint(*frame_duration * global_flags.output_slop_frames);
249 BMDTimeValue target_time = pts - buffer;
251 // While prerolling, we send out frames as quickly as we can.
252 if (target_time < base_pts) {
256 if (!playback_started) {
257 if (output->EndAudioPreroll() != S_OK) {
258 fprintf(stderr, "Could not end audio preroll\n");
261 if (output->StartScheduledPlayback(base_pts, TIMEBASE, 1.0) != S_OK) {
262 fprintf(stderr, "Could not start playback\n");
265 playback_started = true;
268 BMDTimeValue stream_frame_time;
269 double playback_speed;
270 output->GetScheduledStreamTime(TIMEBASE, &stream_frame_time, &playback_speed);
272 // If we're ahead of time, wait for the frame to (approximately) start.
273 if (stream_frame_time < target_time) {
274 steady_clock::time_point t = steady_clock::now() +
275 nanoseconds((target_time - stream_frame_time) * 1000000000 / TIMEBASE);
276 this_thread::sleep_until(t);
280 // If we overshot the previous frame by just a little,
281 // fire off one immediately.
282 if (stream_frame_time < target_time + max_overshoot) {
283 fprintf(stderr, "Warning: Frame was %ld ms late (but not skipping it due to --output-slop-frames).\n",
284 lrint((stream_frame_time - target_time) * 1000.0 / TIMEBASE));
288 // Oops, we missed by more than one frame. Return immediately,
289 // but drop so that we catch up.
290 *dropped_frames = (stream_frame_time - target_time + *frame_duration - 1) / *frame_duration;
291 fprintf(stderr, "Dropped %d output frames; skipping.\n", *dropped_frames);
294 HRESULT DeckLinkOutput::ScheduledFrameCompleted(/* in */ IDeckLinkVideoFrame *completedFrame, /* in */ BMDOutputFrameCompletionResult result)
296 Frame *frame = static_cast<Frame *>(completedFrame);
298 case bmdOutputFrameCompleted:
300 case bmdOutputFrameDisplayedLate:
301 fprintf(stderr, "Output frame displayed late (pts=%ld)\n", frame->pts);
302 fprintf(stderr, "Consider increasing --output-buffer-frames if this persists.\n");
304 case bmdOutputFrameDropped:
305 fprintf(stderr, "Output frame was dropped (pts=%ld)\n", frame->pts);
306 fprintf(stderr, "Consider increasing --output-buffer-frames if this persists.\n");
308 case bmdOutputFrameFlushed:
309 fprintf(stderr, "Output frame was flushed (pts=%ld)\n", frame->pts);
312 fprintf(stderr, "Output frame completed with unknown status %d\n", result);
317 print_latency("DeckLink output latency (frame received → output on HDMI):", frame->received_ts, false, &hei);
320 lock_guard<mutex> lock(frame_queue_mutex);
321 frame_freelist.push(unique_ptr<Frame>(frame));
322 --num_frames_in_flight;
328 HRESULT DeckLinkOutput::ScheduledPlaybackHasStopped()
330 printf("playback stopped!\n");
334 unique_ptr<DeckLinkOutput::Frame> DeckLinkOutput::get_frame()
336 lock_guard<mutex> lock(frame_queue_mutex);
338 if (!frame_freelist.empty()) {
339 unique_ptr<Frame> frame = move(frame_freelist.front());
340 frame_freelist.pop();
344 unique_ptr<Frame> frame(new Frame);
346 frame->uyvy_tex = resource_pool->create_2d_texture(GL_RGBA8, width / 2, height);
348 glGenBuffers(1, &frame->pbo);
350 glBindBuffer(GL_PIXEL_PACK_BUFFER, frame->pbo);
352 glBufferStorage(GL_PIXEL_PACK_BUFFER, width * height * 2, NULL, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
354 frame->uyvy_ptr = (uint8_t *)glMapBufferRange(GL_PIXEL_PACK_BUFFER, 0, width * height * 2, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
356 frame->uyvy_ptr_local.reset(new uint8_t[width * height * 2]);
357 frame->resource_pool = resource_pool;
362 void DeckLinkOutput::present_thread_func()
365 unique_ptr<Frame> frame;
367 unique_lock<mutex> lock(frame_queue_mutex);
368 frame_queues_changed.wait(lock, [this]{
369 return should_quit || !pending_video_frames.empty();
374 frame = move(pending_video_frames.front());
375 pending_video_frames.pop();
376 ++num_frames_in_flight;
379 glWaitSync(frame->fence.get(), /*flags=*/0, GL_TIMEOUT_IGNORED);
381 frame->fence.reset();
383 memcpy(frame->uyvy_ptr_local.get(), frame->uyvy_ptr, width * height * 2);
385 // Release any input frames we needed to render this frame.
386 frame->input_frames.clear();
388 BMDTimeValue pts = frame->pts;
389 BMDTimeValue duration = frame->duration;
390 HRESULT res = output->ScheduleVideoFrame(frame.get(), pts, duration, TIMEBASE);
392 frame.release(); // Owned by the driver now.
394 fprintf(stderr, "Could not schedule video frame! (error=0x%08x)\n", res);
396 lock_guard<mutex> lock(frame_queue_mutex);
397 frame_freelist.push(move(frame));
398 --num_frames_in_flight;
403 HRESULT STDMETHODCALLTYPE DeckLinkOutput::QueryInterface(REFIID, LPVOID *)
405 return E_NOINTERFACE;
408 ULONG STDMETHODCALLTYPE DeckLinkOutput::AddRef()
410 return refcount.fetch_add(1) + 1;
413 ULONG STDMETHODCALLTYPE DeckLinkOutput::Release()
415 int new_ref = refcount.fetch_sub(1) - 1;
421 DeckLinkOutput::Frame::~Frame()
423 glBindBuffer(GL_PIXEL_PACK_BUFFER, pbo);
425 glUnmapBuffer(GL_PIXEL_PACK_BUFFER);
427 glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
429 glDeleteBuffers(1, &pbo);
431 resource_pool->release_2d_texture(uyvy_tex);
435 HRESULT STDMETHODCALLTYPE DeckLinkOutput::Frame::QueryInterface(REFIID, LPVOID *)
437 return E_NOINTERFACE;
440 ULONG STDMETHODCALLTYPE DeckLinkOutput::Frame::AddRef()
442 return refcount.fetch_add(1) + 1;
445 ULONG STDMETHODCALLTYPE DeckLinkOutput::Frame::Release()
447 int new_ref = refcount.fetch_sub(1) - 1;
453 long DeckLinkOutput::Frame::GetWidth()
455 return global_flags.width;
458 long DeckLinkOutput::Frame::GetHeight()
460 return global_flags.height;
463 long DeckLinkOutput::Frame::GetRowBytes()
465 return global_flags.width * 2;
468 BMDPixelFormat DeckLinkOutput::Frame::GetPixelFormat()
470 return bmdFormat8BitYUV;
473 BMDFrameFlags DeckLinkOutput::Frame::GetFlags()
475 return bmdFrameFlagDefault;
478 HRESULT DeckLinkOutput::Frame::GetBytes(/* out */ void **buffer)
480 *buffer = uyvy_ptr_local.get();
484 HRESULT DeckLinkOutput::Frame::GetTimecode(/* in */ BMDTimecodeFormat format, /* out */ IDeckLinkTimecode **timecode)
486 fprintf(stderr, "STUB: GetTimecode()\n");
490 HRESULT DeckLinkOutput::Frame::GetAncillaryData(/* out */ IDeckLinkVideoFrameAncillary **ancillary)
492 fprintf(stderr, "STUB: GetAncillaryData()\n");