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"
15 #include "v210_converter.h"
17 using namespace movit;
19 using namespace std::chrono;
21 DeckLinkOutput::DeckLinkOutput(ResourcePool *resource_pool, QSurface *surface, unsigned width, unsigned height, unsigned card_index)
22 : resource_pool(resource_pool), surface(surface), width(width), height(height), card_index(card_index)
24 chroma_subsampler.reset(new ChromaSubsampler(resource_pool));
27 void DeckLinkOutput::set_device(IDeckLink *decklink)
29 if (decklink->QueryInterface(IID_IDeckLinkOutput, (void**)&output) != S_OK) {
30 fprintf(stderr, "Card %u has no outputs\n", card_index);
34 IDeckLinkDisplayModeIterator *mode_it;
35 if (output->GetDisplayModeIterator(&mode_it) != S_OK) {
36 fprintf(stderr, "Failed to enumerate output display modes for card %u\n", card_index);
42 for (const auto &it : summarize_video_modes(mode_it, card_index)) {
43 if (it.second.width != width || it.second.height != height) {
47 // We could support interlaced modes, but let's stay out of it for now,
48 // since we don't have interlaced stream output.
49 if (it.second.interlaced) {
53 video_modes.insert(it);
58 // HDMI or SDI generally mean “both HDMI and SDI at the same time” on DeckLink cards
59 // that support both; pick_default_video_connection() will generally pick one of those
60 // if they exist. We're not very likely to need analog outputs, so we don't need a way
61 // to change beyond that.
62 video_connection = pick_default_video_connection(decklink, BMDDeckLinkVideoOutputConnections, card_index);
65 void DeckLinkOutput::start_output(uint32_t mode, int64_t base_pts)
68 assert(!playback_initiated);
71 playback_initiated = true;
72 playback_started = false;
73 this->base_pts = base_pts;
75 IDeckLinkConfiguration *config = nullptr;
76 if (output->QueryInterface(IID_IDeckLinkConfiguration, (void**)&config) != S_OK) {
77 fprintf(stderr, "Failed to get configuration interface for output card\n");
80 if (config->SetFlag(bmdDeckLinkConfigLowLatencyVideoOutput, true) != S_OK) {
81 fprintf(stderr, "Failed to set low latency output\n");
84 if (config->SetInt(bmdDeckLinkConfigVideoOutputConnection, video_connection) != S_OK) {
85 fprintf(stderr, "Failed to set video output connection for card %u\n", card_index);
88 if (config->SetFlag(bmdDeckLinkConfigUse1080pNotPsF, true) != S_OK) {
89 fprintf(stderr, "Failed to set PsF flag for card\n");
92 if (config->SetFlag(bmdDeckLinkConfigSMPTELevelAOutput, true) != S_OK) {
93 // This affects at least some no-name SDI->HDMI converters.
94 // Warn, but don't die.
95 fprintf(stderr, "WARNING: Failed to enable SMTPE Level A; resolutions like 1080p60 might have issues.\n");
98 BMDDisplayModeSupport support;
99 IDeckLinkDisplayMode *display_mode;
100 BMDPixelFormat pixel_format = global_flags.ten_bit_output ? bmdFormat10BitYUV : bmdFormat8BitYUV;
101 if (output->DoesSupportVideoMode(mode, pixel_format, bmdVideoOutputFlagDefault,
102 &support, &display_mode) != S_OK) {
103 fprintf(stderr, "Couldn't ask for format support\n");
107 if (support == bmdDisplayModeNotSupported) {
108 fprintf(stderr, "Requested display mode not supported\n");
112 current_mode_flags = display_mode->GetFlags();
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, YCbCrLumaCoefficients output_ycbcr_coefficients, const vector<RefCountedFrame> &input_frames, int64_t pts, int64_t duration)
184 assert(!should_quit);
186 if ((current_mode_flags & bmdDisplayModeColorspaceRec601) && output_ycbcr_coefficients == YCBCR_REC_709) {
187 if (!last_frame_had_mode_mismatch) {
188 fprintf(stderr, "WARNING: Chosen output mode expects Rec. 601 Y'CbCr coefficients.\n");
189 fprintf(stderr, " Consider --output-ycbcr-coefficients=rec601 (or =auto).\n");
191 last_frame_had_mode_mismatch = true;
192 } else if ((current_mode_flags & bmdDisplayModeColorspaceRec709) && output_ycbcr_coefficients == YCBCR_REC_601) {
193 if (!last_frame_had_mode_mismatch) {
194 fprintf(stderr, "WARNING: Chosen output mode expects Rec. 709 Y'CbCr coefficients.\n");
195 fprintf(stderr, " Consider --output-ycbcr-coefficients=rec709 (or =auto).\n");
197 last_frame_had_mode_mismatch = true;
199 last_frame_had_mode_mismatch = false;
202 unique_ptr<Frame> frame = move(get_frame());
203 if (global_flags.ten_bit_output) {
204 chroma_subsampler->create_v210(y_tex, cbcr_tex, width, height, frame->uyvy_tex);
206 chroma_subsampler->create_uyvy(y_tex, cbcr_tex, width, height, frame->uyvy_tex);
209 // Download the UYVY texture to the PBO.
210 glPixelStorei(GL_PACK_ROW_LENGTH, 0);
213 glBindBuffer(GL_PIXEL_PACK_BUFFER, frame->pbo);
216 if (global_flags.ten_bit_output) {
217 glBindTexture(GL_TEXTURE_2D, frame->uyvy_tex);
219 glGetTexImage(GL_TEXTURE_2D, 0, GL_RGBA, GL_UNSIGNED_INT_2_10_10_10_REV, BUFFER_OFFSET(0));
222 glBindTexture(GL_TEXTURE_2D, frame->uyvy_tex);
224 glGetTexImage(GL_TEXTURE_2D, 0, GL_BGRA, GL_UNSIGNED_INT_8_8_8_8_REV, BUFFER_OFFSET(0));
228 glBindTexture(GL_TEXTURE_2D, 0);
230 glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
233 glMemoryBarrier(GL_TEXTURE_UPDATE_BARRIER_BIT | GL_CLIENT_MAPPED_BUFFER_BARRIER_BIT);
236 frame->fence = RefCountedGLsync(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);
238 glFlush(); // Make the DeckLink thread see the fence as soon as possible.
241 frame->input_frames = input_frames;
242 frame->received_ts = find_received_timestamp(input_frames);
244 frame->duration = duration;
247 unique_lock<mutex> lock(frame_queue_mutex);
248 pending_video_frames.push(move(frame));
250 frame_queues_changed.notify_all();
253 void DeckLinkOutput::send_audio(int64_t pts, const std::vector<float> &samples)
255 unique_ptr<int32_t[]> int_samples(new int32_t[samples.size()]);
256 for (size_t i = 0; i < samples.size(); ++i) {
257 int_samples[i] = lrintf(samples[i] * 2147483648.0f);
260 uint32_t frames_written;
261 HRESULT result = output->ScheduleAudioSamples(int_samples.get(), samples.size() / 2,
262 pts, TIMEBASE, &frames_written);
263 if (result != S_OK) {
264 fprintf(stderr, "ScheduleAudioSamples(pts=%ld) failed (result=0x%08x)\n", pts, result);
266 if (frames_written != samples.size() / 2) {
267 fprintf(stderr, "ScheduleAudioSamples() returned short write (%u/%ld)\n", frames_written, samples.size() / 2);
272 void DeckLinkOutput::wait_for_frame(int64_t pts, int *dropped_frames, int64_t *frame_duration, bool *is_preroll, steady_clock::time_point *frame_timestamp)
274 assert(!should_quit);
277 *frame_duration = this->frame_duration;
279 const BMDTimeValue buffer = lrint(*frame_duration * global_flags.output_buffer_frames);
280 const BMDTimeValue max_overshoot = lrint(*frame_duration * global_flags.output_slop_frames);
281 BMDTimeValue target_time = pts - buffer;
283 // While prerolling, we send out frames as quickly as we can.
284 if (target_time < base_pts) {
289 *is_preroll = !playback_started;
291 if (!playback_started) {
292 if (output->EndAudioPreroll() != S_OK) {
293 fprintf(stderr, "Could not end audio preroll\n");
296 if (output->StartScheduledPlayback(base_pts, TIMEBASE, 1.0) != S_OK) {
297 fprintf(stderr, "Could not start playback\n");
300 playback_started = true;
303 BMDTimeValue stream_frame_time;
304 double playback_speed;
305 output->GetScheduledStreamTime(TIMEBASE, &stream_frame_time, &playback_speed);
307 *frame_timestamp = steady_clock::now() +
308 nanoseconds((target_time - stream_frame_time) * 1000000000 / TIMEBASE);
310 // If we're ahead of time, wait for the frame to (approximately) start.
311 if (stream_frame_time < target_time) {
312 this_thread::sleep_until(*frame_timestamp);
316 // If we overshot the previous frame by just a little,
317 // fire off one immediately.
318 if (stream_frame_time < target_time + max_overshoot) {
319 fprintf(stderr, "Warning: Frame was %ld ms late (but not skipping it due to --output-slop-frames).\n",
320 lrint((stream_frame_time - target_time) * 1000.0 / TIMEBASE));
324 // Oops, we missed by more than one frame. Return immediately,
325 // but drop so that we catch up.
326 *dropped_frames = (stream_frame_time - target_time + *frame_duration - 1) / *frame_duration;
327 const int64_t ns_per_frame = this->frame_duration * 1000000000 / TIMEBASE;
328 *frame_timestamp += nanoseconds(*dropped_frames * ns_per_frame);
329 fprintf(stderr, "Dropped %d output frames; skipping.\n", *dropped_frames);
332 uint32_t DeckLinkOutput::pick_video_mode(uint32_t mode) const
334 if (video_modes.count(mode)) {
338 // Prioritize 59.94 > 60 > 29.97. If none of those are found, then pick the highest one.
339 for (const pair<int, int> &desired : vector<pair<int, int>>{ { 60000, 1001 }, { 60, 0 }, { 30000, 1001 } }) {
340 for (const auto &it : video_modes) {
341 if (it.second.frame_rate_num * desired.second == desired.first * it.second.frame_rate_den) {
347 uint32_t best_mode = 0;
348 double best_fps = 0.0;
349 for (const auto &it : video_modes) {
350 double fps = double(it.second.frame_rate_num) / it.second.frame_rate_den;
351 if (fps > best_fps) {
352 best_mode = it.first;
359 YCbCrLumaCoefficients DeckLinkOutput::preferred_ycbcr_coefficients() const
361 if (current_mode_flags & bmdDisplayModeColorspaceRec601) {
362 return YCBCR_REC_601;
364 // Don't bother checking bmdDisplayModeColorspaceRec709;
365 // if none is set, 709 is a good default anyway.
366 return YCBCR_REC_709;
370 HRESULT DeckLinkOutput::ScheduledFrameCompleted(/* in */ IDeckLinkVideoFrame *completedFrame, /* in */ BMDOutputFrameCompletionResult result)
372 Frame *frame = static_cast<Frame *>(completedFrame);
374 case bmdOutputFrameCompleted:
376 case bmdOutputFrameDisplayedLate:
377 fprintf(stderr, "Output frame displayed late (pts=%ld)\n", frame->pts);
378 fprintf(stderr, "Consider increasing --output-buffer-frames if this persists.\n");
380 case bmdOutputFrameDropped:
381 fprintf(stderr, "Output frame was dropped (pts=%ld)\n", frame->pts);
382 fprintf(stderr, "Consider increasing --output-buffer-frames if this persists.\n");
384 case bmdOutputFrameFlushed:
385 fprintf(stderr, "Output frame was flushed (pts=%ld)\n", frame->pts);
388 fprintf(stderr, "Output frame completed with unknown status %d\n", result);
392 static int frameno = 0;
393 print_latency("DeckLink output latency (frame received → output on HDMI):", frame->received_ts, false, &frameno);
396 lock_guard<mutex> lock(frame_queue_mutex);
397 frame_freelist.push(unique_ptr<Frame>(frame));
398 --num_frames_in_flight;
404 HRESULT DeckLinkOutput::ScheduledPlaybackHasStopped()
406 printf("playback stopped!\n");
410 unique_ptr<DeckLinkOutput::Frame> DeckLinkOutput::get_frame()
412 lock_guard<mutex> lock(frame_queue_mutex);
414 if (!frame_freelist.empty()) {
415 unique_ptr<Frame> frame = move(frame_freelist.front());
416 frame_freelist.pop();
420 unique_ptr<Frame> frame(new Frame);
423 if (global_flags.ten_bit_output) {
424 stride = v210Converter::get_v210_stride(width);
425 GLint v210_width = stride / sizeof(uint32_t);
426 frame->uyvy_tex = resource_pool->create_2d_texture(GL_RGB10_A2, v210_width, height);
428 // We need valid texture state, or NVIDIA won't allow us to write to the texture.
429 glBindTexture(GL_TEXTURE_2D, frame->uyvy_tex);
431 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
435 frame->uyvy_tex = resource_pool->create_2d_texture(GL_RGBA8, width / 2, height);
438 glGenBuffers(1, &frame->pbo);
440 glBindBuffer(GL_PIXEL_PACK_BUFFER, frame->pbo);
442 glBufferStorage(GL_PIXEL_PACK_BUFFER, stride * height, NULL, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
444 frame->uyvy_ptr = (uint8_t *)glMapBufferRange(GL_PIXEL_PACK_BUFFER, 0, stride * height, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
446 frame->uyvy_ptr_local.reset(new uint8_t[stride * height]);
447 frame->resource_pool = resource_pool;
452 void DeckLinkOutput::present_thread_func()
454 pthread_setname_np(pthread_self(), "DeckLinkOutput");
456 unique_ptr<Frame> frame;
458 unique_lock<mutex> lock(frame_queue_mutex);
459 frame_queues_changed.wait(lock, [this]{
460 return should_quit || !pending_video_frames.empty();
465 frame = move(pending_video_frames.front());
466 pending_video_frames.pop();
467 ++num_frames_in_flight;
470 glClientWaitSync(frame->fence.get(), /*flags=*/0, GL_TIMEOUT_IGNORED);
472 frame->fence.reset();
474 if (global_flags.ten_bit_output) {
475 memcpy(frame->uyvy_ptr_local.get(), frame->uyvy_ptr, v210Converter::get_v210_stride(width) * height);
477 memcpy(frame->uyvy_ptr_local.get(), frame->uyvy_ptr, width * height * 2);
480 // Release any input frames we needed to render this frame.
481 frame->input_frames.clear();
483 BMDTimeValue pts = frame->pts;
484 BMDTimeValue duration = frame->duration;
485 HRESULT res = output->ScheduleVideoFrame(frame.get(), pts, duration, TIMEBASE);
487 frame.release(); // Owned by the driver now.
489 fprintf(stderr, "Could not schedule video frame! (error=0x%08x)\n", res);
491 lock_guard<mutex> lock(frame_queue_mutex);
492 frame_freelist.push(move(frame));
493 --num_frames_in_flight;
498 HRESULT STDMETHODCALLTYPE DeckLinkOutput::QueryInterface(REFIID, LPVOID *)
500 return E_NOINTERFACE;
503 ULONG STDMETHODCALLTYPE DeckLinkOutput::AddRef()
505 return refcount.fetch_add(1) + 1;
508 ULONG STDMETHODCALLTYPE DeckLinkOutput::Release()
510 int new_ref = refcount.fetch_sub(1) - 1;
516 DeckLinkOutput::Frame::~Frame()
518 glBindBuffer(GL_PIXEL_PACK_BUFFER, pbo);
520 glUnmapBuffer(GL_PIXEL_PACK_BUFFER);
522 glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
524 glDeleteBuffers(1, &pbo);
526 resource_pool->release_2d_texture(uyvy_tex);
530 HRESULT STDMETHODCALLTYPE DeckLinkOutput::Frame::QueryInterface(REFIID, LPVOID *)
532 return E_NOINTERFACE;
535 ULONG STDMETHODCALLTYPE DeckLinkOutput::Frame::AddRef()
537 return refcount.fetch_add(1) + 1;
540 ULONG STDMETHODCALLTYPE DeckLinkOutput::Frame::Release()
542 int new_ref = refcount.fetch_sub(1) - 1;
548 long DeckLinkOutput::Frame::GetWidth()
550 return global_flags.width;
553 long DeckLinkOutput::Frame::GetHeight()
555 return global_flags.height;
558 long DeckLinkOutput::Frame::GetRowBytes()
560 if (global_flags.ten_bit_output) {
561 return v210Converter::get_v210_stride(global_flags.width);
563 return global_flags.width * 2;
567 BMDPixelFormat DeckLinkOutput::Frame::GetPixelFormat()
569 if (global_flags.ten_bit_output) {
570 return bmdFormat10BitYUV;
572 return bmdFormat8BitYUV;
576 BMDFrameFlags DeckLinkOutput::Frame::GetFlags()
578 return bmdFrameFlagDefault;
581 HRESULT DeckLinkOutput::Frame::GetBytes(/* out */ void **buffer)
583 *buffer = uyvy_ptr_local.get();
587 HRESULT DeckLinkOutput::Frame::GetTimecode(/* in */ BMDTimecodeFormat format, /* out */ IDeckLinkTimecode **timecode)
589 fprintf(stderr, "STUB: GetTimecode()\n");
593 HRESULT DeckLinkOutput::Frame::GetAncillaryData(/* out */ IDeckLinkVideoFrameAncillary **ancillary)
595 fprintf(stderr, "STUB: GetAncillaryData()\n");