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);
57 void DeckLinkOutput::start_output(uint32_t mode, int64_t base_pts)
62 playback_started = false;
63 this->base_pts = base_pts;
65 IDeckLinkConfiguration *config = nullptr;
66 if (output->QueryInterface(IID_IDeckLinkConfiguration, (void**)&config) != S_OK) {
67 fprintf(stderr, "Failed to get configuration interface for output card\n");
70 if (config->SetFlag(bmdDeckLinkConfigLowLatencyVideoOutput, true) != S_OK) {
71 fprintf(stderr, "Failed to set low latency output\n");
74 // HDMI or SDI generally mean “both HDMI and SDI at the same time” on DeckLink cards.
75 // We're not very likely to need analog outputs.
76 if (config->SetInt(bmdDeckLinkConfigVideoOutputConnection, bmdVideoConnectionHDMI) != S_OK) {
77 fprintf(stderr, "Failed to set video output connection for card %u\n", card_index);
80 if (config->SetFlag(bmdDeckLinkConfigUse1080pNotPsF, true) != S_OK) {
81 fprintf(stderr, "Failed to set PsF flag for card\n");
85 BMDDisplayModeSupport support;
86 IDeckLinkDisplayMode *display_mode;
87 if (output->DoesSupportVideoMode(mode, bmdFormat8BitYUV, bmdVideoOutputFlagDefault,
88 &support, &display_mode) != S_OK) {
89 fprintf(stderr, "Couldn't ask for format support\n");
93 if (support == bmdDisplayModeNotSupported) {
94 fprintf(stderr, "Requested display mode not supported\n");
98 BMDDisplayModeFlags flags = display_mode->GetFlags();
99 if ((flags & bmdDisplayModeColorspaceRec601) && global_flags.ycbcr_rec709_coefficients) {
100 fprintf(stderr, "WARNING: Chosen output mode expects Rec. 601 Y'CbCr coefficients.\n");
101 fprintf(stderr, " Consider --output-ycbcr-coefficients=rec601 (or =auto).\n");
102 } else if ((flags & bmdDisplayModeColorspaceRec709) && !global_flags.ycbcr_rec709_coefficients) {
103 fprintf(stderr, "WARNING: Chosen output mode expects Rec. 709 Y'CbCr coefficients.\n");
104 fprintf(stderr, " Consider --output-ycbcr-coefficients=rec709 (or =auto).\n");
107 BMDTimeValue time_value;
108 BMDTimeScale time_scale;
109 if (display_mode->GetFrameRate(&time_value, &time_scale) != S_OK) {
110 fprintf(stderr, "Couldn't get frame rate\n");
114 frame_duration = time_value * TIMEBASE / time_scale;
116 display_mode->Release();
118 HRESULT result = output->EnableVideoOutput(mode, bmdVideoOutputFlagDefault);
119 if (result != S_OK) {
120 fprintf(stderr, "Couldn't enable output with error 0x%x\n", result);
123 if (output->SetScheduledFrameCompletionCallback(this) != S_OK) {
124 fprintf(stderr, "Couldn't set callback\n");
127 assert(OUTPUT_FREQUENCY == 48000);
128 if (output->EnableAudioOutput(bmdAudioSampleRate48kHz, bmdAudioSampleType32bitInteger, 2, bmdAudioOutputStreamTimestamped) != S_OK) {
129 fprintf(stderr, "Couldn't enable audio output\n");
132 if (output->BeginAudioPreroll() != S_OK) {
133 fprintf(stderr, "Couldn't begin audio preroll\n");
137 present_thread = thread([this]{
138 QOpenGLContext *context = create_context(this->surface);
139 eglBindAPI(EGL_OPENGL_API);
140 if (!make_current(context, this->surface)) {
141 printf("display=%p surface=%p context=%p curr=%p err=%d\n", eglGetCurrentDisplay(), this->surface, context, eglGetCurrentContext(),
145 present_thread_func();
146 delete_context(context);
150 void DeckLinkOutput::end_output()
153 frame_queues_changed.notify_all();
154 present_thread.join();
156 output->StopScheduledPlayback(0, nullptr, 0);
157 output->DisableVideoOutput();
158 output->DisableAudioOutput();
160 // Wait until all frames are accounted for, and free them.
162 unique_lock<mutex> lock(frame_queue_mutex);
163 while (!(frame_freelist.empty() && num_frames_in_flight == 0)) {
164 frame_queues_changed.wait(lock, [this]{ return !frame_freelist.empty(); });
165 frame_freelist.pop();
170 void DeckLinkOutput::send_frame(GLuint y_tex, GLuint cbcr_tex, const vector<RefCountedFrame> &input_frames, int64_t pts, int64_t duration)
172 unique_ptr<Frame> frame = move(get_frame());
173 chroma_subsampler->create_uyvy(y_tex, cbcr_tex, width, height, frame->uyvy_tex);
175 // Download the UYVY texture to the PBO.
176 glPixelStorei(GL_PACK_ROW_LENGTH, 0);
179 glBindBuffer(GL_PIXEL_PACK_BUFFER, frame->pbo);
182 glBindTexture(GL_TEXTURE_2D, frame->uyvy_tex);
184 glGetTexImage(GL_TEXTURE_2D, 0, GL_BGRA, GL_UNSIGNED_INT_8_8_8_8_REV, BUFFER_OFFSET(0));
187 glBindTexture(GL_TEXTURE_2D, 0);
189 glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
192 glMemoryBarrier(GL_TEXTURE_UPDATE_BARRIER_BIT | GL_CLIENT_MAPPED_BUFFER_BARRIER_BIT);
195 frame->fence = RefCountedGLsync(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);
197 glFlush(); // Make the DeckLink thread see the fence as soon as possible.
200 frame->input_frames = input_frames;
201 frame->received_ts = find_received_timestamp(input_frames);
203 frame->duration = duration;
206 unique_lock<mutex> lock(frame_queue_mutex);
207 pending_video_frames.push(move(frame));
209 frame_queues_changed.notify_all();
212 void DeckLinkOutput::send_audio(int64_t pts, const std::vector<float> &samples)
214 unique_ptr<int32_t[]> int_samples(new int32_t[samples.size()]);
215 for (size_t i = 0; i < samples.size(); ++i) {
216 int_samples[i] = lrintf(samples[i] * 2147483648.0f);
219 uint32_t frames_written;
220 HRESULT result = output->ScheduleAudioSamples(int_samples.get(), samples.size() / 2,
221 pts, TIMEBASE, &frames_written);
222 if (result != S_OK) {
223 fprintf(stderr, "ScheduleAudioSamples(pts=%ld) failed (result=0x%08x)\n", pts, result);
225 if (frames_written != samples.size() / 2) {
226 fprintf(stderr, "ScheduleAudioSamples() returned short write (%u/%ld)\n", frames_written, samples.size() / 2);
231 void DeckLinkOutput::wait_for_frame(int64_t pts, int *dropped_frames, int64_t *frame_duration)
234 *frame_duration = this->frame_duration;
236 const BMDTimeValue buffer = lrint(*frame_duration * global_flags.output_buffer_frames);
237 const BMDTimeValue max_overshoot = lrint(*frame_duration * global_flags.output_slop_frames);
238 BMDTimeValue target_time = pts - buffer;
240 // While prerolling, we send out frames as quickly as we can.
241 if (target_time < base_pts) {
245 if (!playback_started) {
246 if (output->EndAudioPreroll() != S_OK) {
247 fprintf(stderr, "Could not end audio preroll\n");
250 if (output->StartScheduledPlayback(base_pts, TIMEBASE, 1.0) != S_OK) {
251 fprintf(stderr, "Could not start playback\n");
254 playback_started = true;
257 BMDTimeValue stream_frame_time;
258 double playback_speed;
259 output->GetScheduledStreamTime(TIMEBASE, &stream_frame_time, &playback_speed);
261 // If we're ahead of time, wait for the frame to (approximately) start.
262 if (stream_frame_time < target_time) {
263 steady_clock::time_point t = steady_clock::now() +
264 nanoseconds((target_time - stream_frame_time) * 1000000000 / TIMEBASE);
265 this_thread::sleep_until(t);
269 // If we overshot the previous frame by just a little,
270 // fire off one immediately.
271 if (stream_frame_time < target_time + max_overshoot) {
272 fprintf(stderr, "Warning: Frame was %ld ms late (but not skipping it due to --output-slop-frames).\n",
273 lrint((stream_frame_time - target_time) * 1000.0 / TIMEBASE));
277 // Oops, we missed by more than one frame. Return immediately,
278 // but drop so that we catch up.
279 *dropped_frames = (stream_frame_time - target_time + *frame_duration - 1) / *frame_duration;
280 fprintf(stderr, "Dropped %d output frames; skipping.\n", *dropped_frames);
283 HRESULT DeckLinkOutput::ScheduledFrameCompleted(/* in */ IDeckLinkVideoFrame *completedFrame, /* in */ BMDOutputFrameCompletionResult result)
285 Frame *frame = static_cast<Frame *>(completedFrame);
287 case bmdOutputFrameCompleted:
289 case bmdOutputFrameDisplayedLate:
290 fprintf(stderr, "Output frame displayed late (pts=%ld)\n", frame->pts);
291 fprintf(stderr, "Consider increasing --output-buffer-frames if this persists.\n");
293 case bmdOutputFrameDropped:
294 fprintf(stderr, "Output frame was dropped (pts=%ld)\n", frame->pts);
295 fprintf(stderr, "Consider increasing --output-buffer-frames if this persists.\n");
297 case bmdOutputFrameFlushed:
298 fprintf(stderr, "Output frame was flushed (pts=%ld)\n", frame->pts);
301 fprintf(stderr, "Output frame completed with unknown status %d\n", result);
306 print_latency("DeckLink output latency (frame received → output on HDMI):", frame->received_ts, false, &hei);
309 lock_guard<mutex> lock(frame_queue_mutex);
310 frame_freelist.push(unique_ptr<Frame>(frame));
311 --num_frames_in_flight;
317 HRESULT DeckLinkOutput::ScheduledPlaybackHasStopped()
319 printf("playback stopped!\n");
323 unique_ptr<DeckLinkOutput::Frame> DeckLinkOutput::get_frame()
325 lock_guard<mutex> lock(frame_queue_mutex);
327 if (!frame_freelist.empty()) {
328 unique_ptr<Frame> frame = move(frame_freelist.front());
329 frame_freelist.pop();
333 unique_ptr<Frame> frame(new Frame);
335 frame->uyvy_tex = resource_pool->create_2d_texture(GL_RGBA8, width / 2, height);
337 glGenBuffers(1, &frame->pbo);
339 glBindBuffer(GL_PIXEL_PACK_BUFFER, frame->pbo);
341 glBufferStorage(GL_PIXEL_PACK_BUFFER, width * height * 2, NULL, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
343 frame->uyvy_ptr = (uint8_t *)glMapBufferRange(GL_PIXEL_PACK_BUFFER, 0, width * height * 2, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
345 frame->uyvy_ptr_local.reset(new uint8_t[width * height * 2]);
346 frame->resource_pool = resource_pool;
351 void DeckLinkOutput::present_thread_func()
354 unique_ptr<Frame> frame;
356 unique_lock<mutex> lock(frame_queue_mutex);
357 frame_queues_changed.wait(lock, [this]{
358 return should_quit || !pending_video_frames.empty();
363 frame = move(pending_video_frames.front());
364 pending_video_frames.pop();
365 ++num_frames_in_flight;
368 glWaitSync(frame->fence.get(), /*flags=*/0, GL_TIMEOUT_IGNORED);
370 frame->fence.reset();
372 memcpy(frame->uyvy_ptr_local.get(), frame->uyvy_ptr, width * height * 2);
374 // Release any input frames we needed to render this frame.
375 frame->input_frames.clear();
377 BMDTimeValue pts = frame->pts;
378 BMDTimeValue duration = frame->duration;
379 HRESULT res = output->ScheduleVideoFrame(frame.get(), pts, duration, TIMEBASE);
381 frame.release(); // Owned by the driver now.
383 fprintf(stderr, "Could not schedule video frame! (error=0x%08x)\n", res);
385 lock_guard<mutex> lock(frame_queue_mutex);
386 frame_freelist.push(move(frame));
387 --num_frames_in_flight;
392 HRESULT STDMETHODCALLTYPE DeckLinkOutput::QueryInterface(REFIID, LPVOID *)
394 return E_NOINTERFACE;
397 ULONG STDMETHODCALLTYPE DeckLinkOutput::AddRef()
399 return refcount.fetch_add(1) + 1;
402 ULONG STDMETHODCALLTYPE DeckLinkOutput::Release()
404 int new_ref = refcount.fetch_sub(1) - 1;
410 DeckLinkOutput::Frame::~Frame()
412 glBindBuffer(GL_PIXEL_PACK_BUFFER, pbo);
414 glUnmapBuffer(GL_PIXEL_PACK_BUFFER);
416 glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
418 glDeleteBuffers(1, &pbo);
420 resource_pool->release_2d_texture(uyvy_tex);
424 HRESULT STDMETHODCALLTYPE DeckLinkOutput::Frame::QueryInterface(REFIID, LPVOID *)
426 return E_NOINTERFACE;
429 ULONG STDMETHODCALLTYPE DeckLinkOutput::Frame::AddRef()
431 return refcount.fetch_add(1) + 1;
434 ULONG STDMETHODCALLTYPE DeckLinkOutput::Frame::Release()
436 int new_ref = refcount.fetch_sub(1) - 1;
442 long DeckLinkOutput::Frame::GetWidth()
444 return global_flags.width;
447 long DeckLinkOutput::Frame::GetHeight()
449 return global_flags.height;
452 long DeckLinkOutput::Frame::GetRowBytes()
454 return global_flags.width * 2;
457 BMDPixelFormat DeckLinkOutput::Frame::GetPixelFormat()
459 return bmdFormat8BitYUV;
462 BMDFrameFlags DeckLinkOutput::Frame::GetFlags()
464 return bmdFrameFlagDefault;
467 HRESULT DeckLinkOutput::Frame::GetBytes(/* out */ void **buffer)
469 *buffer = uyvy_ptr_local.get();
473 HRESULT DeckLinkOutput::Frame::GetTimecode(/* in */ BMDTimecodeFormat format, /* out */ IDeckLinkTimecode **timecode)
475 fprintf(stderr, "STUB: GetTimecode()\n");
479 HRESULT DeckLinkOutput::Frame::GetAncillaryData(/* out */ IDeckLinkVideoFrameAncillary **ancillary)
481 fprintf(stderr, "STUB: GetAncillaryData()\n");