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)
68 playback_started = false;
69 this->base_pts = base_pts;
71 IDeckLinkConfiguration *config = nullptr;
72 if (output->QueryInterface(IID_IDeckLinkConfiguration, (void**)&config) != S_OK) {
73 fprintf(stderr, "Failed to get configuration interface for output card\n");
76 if (config->SetFlag(bmdDeckLinkConfigLowLatencyVideoOutput, true) != S_OK) {
77 fprintf(stderr, "Failed to set low latency output\n");
80 if (config->SetInt(bmdDeckLinkConfigVideoOutputConnection, video_connection) != S_OK) {
81 fprintf(stderr, "Failed to set video output connection for card %u\n", card_index);
84 if (config->SetFlag(bmdDeckLinkConfigUse1080pNotPsF, true) != S_OK) {
85 fprintf(stderr, "Failed to set PsF flag for card\n");
89 BMDDisplayModeSupport support;
90 IDeckLinkDisplayMode *display_mode;
91 if (output->DoesSupportVideoMode(mode, bmdFormat8BitYUV, bmdVideoOutputFlagDefault,
92 &support, &display_mode) != S_OK) {
93 fprintf(stderr, "Couldn't ask for format support\n");
97 if (support == bmdDisplayModeNotSupported) {
98 fprintf(stderr, "Requested display mode not supported\n");
102 BMDDisplayModeFlags flags = display_mode->GetFlags();
103 if ((flags & bmdDisplayModeColorspaceRec601) && global_flags.ycbcr_rec709_coefficients) {
104 fprintf(stderr, "WARNING: Chosen output mode expects Rec. 601 Y'CbCr coefficients.\n");
105 fprintf(stderr, " Consider --output-ycbcr-coefficients=rec601 (or =auto).\n");
106 } else if ((flags & bmdDisplayModeColorspaceRec709) && !global_flags.ycbcr_rec709_coefficients) {
107 fprintf(stderr, "WARNING: Chosen output mode expects Rec. 709 Y'CbCr coefficients.\n");
108 fprintf(stderr, " Consider --output-ycbcr-coefficients=rec709 (or =auto).\n");
111 BMDTimeValue time_value;
112 BMDTimeScale time_scale;
113 if (display_mode->GetFrameRate(&time_value, &time_scale) != S_OK) {
114 fprintf(stderr, "Couldn't get frame rate\n");
118 frame_duration = time_value * TIMEBASE / time_scale;
120 display_mode->Release();
122 HRESULT result = output->EnableVideoOutput(mode, bmdVideoOutputFlagDefault);
123 if (result != S_OK) {
124 fprintf(stderr, "Couldn't enable output with error 0x%x\n", result);
127 if (output->SetScheduledFrameCompletionCallback(this) != S_OK) {
128 fprintf(stderr, "Couldn't set callback\n");
131 assert(OUTPUT_FREQUENCY == 48000);
132 if (output->EnableAudioOutput(bmdAudioSampleRate48kHz, bmdAudioSampleType32bitInteger, 2, bmdAudioOutputStreamTimestamped) != S_OK) {
133 fprintf(stderr, "Couldn't enable audio output\n");
136 if (output->BeginAudioPreroll() != S_OK) {
137 fprintf(stderr, "Couldn't begin audio preroll\n");
141 present_thread = thread([this]{
142 QOpenGLContext *context = create_context(this->surface);
143 eglBindAPI(EGL_OPENGL_API);
144 if (!make_current(context, this->surface)) {
145 printf("display=%p surface=%p context=%p curr=%p err=%d\n", eglGetCurrentDisplay(), this->surface, context, eglGetCurrentContext(),
149 present_thread_func();
150 delete_context(context);
154 void DeckLinkOutput::end_output()
157 frame_queues_changed.notify_all();
158 present_thread.join();
160 output->StopScheduledPlayback(0, nullptr, 0);
161 output->DisableVideoOutput();
162 output->DisableAudioOutput();
164 // Wait until all frames are accounted for, and free them.
166 unique_lock<mutex> lock(frame_queue_mutex);
167 while (!(frame_freelist.empty() && num_frames_in_flight == 0)) {
168 frame_queues_changed.wait(lock, [this]{ return !frame_freelist.empty(); });
169 frame_freelist.pop();
174 void DeckLinkOutput::send_frame(GLuint y_tex, GLuint cbcr_tex, const vector<RefCountedFrame> &input_frames, int64_t pts, int64_t duration)
176 unique_ptr<Frame> frame = move(get_frame());
177 chroma_subsampler->create_uyvy(y_tex, cbcr_tex, width, height, frame->uyvy_tex);
179 // Download the UYVY texture to the PBO.
180 glPixelStorei(GL_PACK_ROW_LENGTH, 0);
183 glBindBuffer(GL_PIXEL_PACK_BUFFER, frame->pbo);
186 glBindTexture(GL_TEXTURE_2D, frame->uyvy_tex);
188 glGetTexImage(GL_TEXTURE_2D, 0, GL_BGRA, GL_UNSIGNED_INT_8_8_8_8_REV, BUFFER_OFFSET(0));
191 glBindTexture(GL_TEXTURE_2D, 0);
193 glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
196 glMemoryBarrier(GL_TEXTURE_UPDATE_BARRIER_BIT | GL_CLIENT_MAPPED_BUFFER_BARRIER_BIT);
199 frame->fence = RefCountedGLsync(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);
201 glFlush(); // Make the DeckLink thread see the fence as soon as possible.
204 frame->input_frames = input_frames;
205 frame->received_ts = find_received_timestamp(input_frames);
207 frame->duration = duration;
210 unique_lock<mutex> lock(frame_queue_mutex);
211 pending_video_frames.push(move(frame));
213 frame_queues_changed.notify_all();
216 void DeckLinkOutput::send_audio(int64_t pts, const std::vector<float> &samples)
218 unique_ptr<int32_t[]> int_samples(new int32_t[samples.size()]);
219 for (size_t i = 0; i < samples.size(); ++i) {
220 int_samples[i] = lrintf(samples[i] * 2147483648.0f);
223 uint32_t frames_written;
224 HRESULT result = output->ScheduleAudioSamples(int_samples.get(), samples.size() / 2,
225 pts, TIMEBASE, &frames_written);
226 if (result != S_OK) {
227 fprintf(stderr, "ScheduleAudioSamples(pts=%ld) failed (result=0x%08x)\n", pts, result);
229 if (frames_written != samples.size() / 2) {
230 fprintf(stderr, "ScheduleAudioSamples() returned short write (%u/%ld)\n", frames_written, samples.size() / 2);
235 void DeckLinkOutput::wait_for_frame(int64_t pts, int *dropped_frames, int64_t *frame_duration)
238 *frame_duration = this->frame_duration;
240 const BMDTimeValue buffer = lrint(*frame_duration * global_flags.output_buffer_frames);
241 const BMDTimeValue max_overshoot = lrint(*frame_duration * global_flags.output_slop_frames);
242 BMDTimeValue target_time = pts - buffer;
244 // While prerolling, we send out frames as quickly as we can.
245 if (target_time < base_pts) {
249 if (!playback_started) {
250 if (output->EndAudioPreroll() != S_OK) {
251 fprintf(stderr, "Could not end audio preroll\n");
254 if (output->StartScheduledPlayback(base_pts, TIMEBASE, 1.0) != S_OK) {
255 fprintf(stderr, "Could not start playback\n");
258 playback_started = true;
261 BMDTimeValue stream_frame_time;
262 double playback_speed;
263 output->GetScheduledStreamTime(TIMEBASE, &stream_frame_time, &playback_speed);
265 // If we're ahead of time, wait for the frame to (approximately) start.
266 if (stream_frame_time < target_time) {
267 steady_clock::time_point t = steady_clock::now() +
268 nanoseconds((target_time - stream_frame_time) * 1000000000 / TIMEBASE);
269 this_thread::sleep_until(t);
273 // If we overshot the previous frame by just a little,
274 // fire off one immediately.
275 if (stream_frame_time < target_time + max_overshoot) {
276 fprintf(stderr, "Warning: Frame was %ld ms late (but not skipping it due to --output-slop-frames).\n",
277 lrint((stream_frame_time - target_time) * 1000.0 / TIMEBASE));
281 // Oops, we missed by more than one frame. Return immediately,
282 // but drop so that we catch up.
283 *dropped_frames = (stream_frame_time - target_time + *frame_duration - 1) / *frame_duration;
284 fprintf(stderr, "Dropped %d output frames; skipping.\n", *dropped_frames);
287 HRESULT DeckLinkOutput::ScheduledFrameCompleted(/* in */ IDeckLinkVideoFrame *completedFrame, /* in */ BMDOutputFrameCompletionResult result)
289 Frame *frame = static_cast<Frame *>(completedFrame);
291 case bmdOutputFrameCompleted:
293 case bmdOutputFrameDisplayedLate:
294 fprintf(stderr, "Output frame displayed late (pts=%ld)\n", frame->pts);
295 fprintf(stderr, "Consider increasing --output-buffer-frames if this persists.\n");
297 case bmdOutputFrameDropped:
298 fprintf(stderr, "Output frame was dropped (pts=%ld)\n", frame->pts);
299 fprintf(stderr, "Consider increasing --output-buffer-frames if this persists.\n");
301 case bmdOutputFrameFlushed:
302 fprintf(stderr, "Output frame was flushed (pts=%ld)\n", frame->pts);
305 fprintf(stderr, "Output frame completed with unknown status %d\n", result);
310 print_latency("DeckLink output latency (frame received → output on HDMI):", frame->received_ts, false, &hei);
313 lock_guard<mutex> lock(frame_queue_mutex);
314 frame_freelist.push(unique_ptr<Frame>(frame));
315 --num_frames_in_flight;
321 HRESULT DeckLinkOutput::ScheduledPlaybackHasStopped()
323 printf("playback stopped!\n");
327 unique_ptr<DeckLinkOutput::Frame> DeckLinkOutput::get_frame()
329 lock_guard<mutex> lock(frame_queue_mutex);
331 if (!frame_freelist.empty()) {
332 unique_ptr<Frame> frame = move(frame_freelist.front());
333 frame_freelist.pop();
337 unique_ptr<Frame> frame(new Frame);
339 frame->uyvy_tex = resource_pool->create_2d_texture(GL_RGBA8, width / 2, height);
341 glGenBuffers(1, &frame->pbo);
343 glBindBuffer(GL_PIXEL_PACK_BUFFER, frame->pbo);
345 glBufferStorage(GL_PIXEL_PACK_BUFFER, width * height * 2, NULL, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
347 frame->uyvy_ptr = (uint8_t *)glMapBufferRange(GL_PIXEL_PACK_BUFFER, 0, width * height * 2, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
349 frame->uyvy_ptr_local.reset(new uint8_t[width * height * 2]);
350 frame->resource_pool = resource_pool;
355 void DeckLinkOutput::present_thread_func()
358 unique_ptr<Frame> frame;
360 unique_lock<mutex> lock(frame_queue_mutex);
361 frame_queues_changed.wait(lock, [this]{
362 return should_quit || !pending_video_frames.empty();
367 frame = move(pending_video_frames.front());
368 pending_video_frames.pop();
369 ++num_frames_in_flight;
372 glWaitSync(frame->fence.get(), /*flags=*/0, GL_TIMEOUT_IGNORED);
374 frame->fence.reset();
376 memcpy(frame->uyvy_ptr_local.get(), frame->uyvy_ptr, width * height * 2);
378 // Release any input frames we needed to render this frame.
379 frame->input_frames.clear();
381 BMDTimeValue pts = frame->pts;
382 BMDTimeValue duration = frame->duration;
383 HRESULT res = output->ScheduleVideoFrame(frame.get(), pts, duration, TIMEBASE);
385 frame.release(); // Owned by the driver now.
387 fprintf(stderr, "Could not schedule video frame! (error=0x%08x)\n", res);
389 lock_guard<mutex> lock(frame_queue_mutex);
390 frame_freelist.push(move(frame));
391 --num_frames_in_flight;
396 HRESULT STDMETHODCALLTYPE DeckLinkOutput::QueryInterface(REFIID, LPVOID *)
398 return E_NOINTERFACE;
401 ULONG STDMETHODCALLTYPE DeckLinkOutput::AddRef()
403 return refcount.fetch_add(1) + 1;
406 ULONG STDMETHODCALLTYPE DeckLinkOutput::Release()
408 int new_ref = refcount.fetch_sub(1) - 1;
414 DeckLinkOutput::Frame::~Frame()
416 glBindBuffer(GL_PIXEL_PACK_BUFFER, pbo);
418 glUnmapBuffer(GL_PIXEL_PACK_BUFFER);
420 glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
422 glDeleteBuffers(1, &pbo);
424 resource_pool->release_2d_texture(uyvy_tex);
428 HRESULT STDMETHODCALLTYPE DeckLinkOutput::Frame::QueryInterface(REFIID, LPVOID *)
430 return E_NOINTERFACE;
433 ULONG STDMETHODCALLTYPE DeckLinkOutput::Frame::AddRef()
435 return refcount.fetch_add(1) + 1;
438 ULONG STDMETHODCALLTYPE DeckLinkOutput::Frame::Release()
440 int new_ref = refcount.fetch_sub(1) - 1;
446 long DeckLinkOutput::Frame::GetWidth()
448 return global_flags.width;
451 long DeckLinkOutput::Frame::GetHeight()
453 return global_flags.height;
456 long DeckLinkOutput::Frame::GetRowBytes()
458 return global_flags.width * 2;
461 BMDPixelFormat DeckLinkOutput::Frame::GetPixelFormat()
463 return bmdFormat8BitYUV;
466 BMDFrameFlags DeckLinkOutput::Frame::GetFlags()
468 return bmdFrameFlagDefault;
471 HRESULT DeckLinkOutput::Frame::GetBytes(/* out */ void **buffer)
473 *buffer = uyvy_ptr_local.get();
477 HRESULT DeckLinkOutput::Frame::GetTimecode(/* in */ BMDTimecodeFormat format, /* out */ IDeckLinkTimecode **timecode)
479 fprintf(stderr, "STUB: GetTimecode()\n");
483 HRESULT DeckLinkOutput::Frame::GetAncillaryData(/* out */ IDeckLinkVideoFrameAncillary **ancillary)
485 fprintf(stderr, "STUB: GetAncillaryData()\n");