1 #include "decklink_capture.h"
3 #include <DeckLinkAPI.h>
4 #include <DeckLinkAPIConfiguration.h>
5 #include <DeckLinkAPIDiscovery.h>
6 #include <DeckLinkAPIModes.h>
21 #include "bmusb/bmusb.h"
22 #include "decklink_util.h"
24 #define FRAME_SIZE (8 << 20) // 8 MB.
27 using namespace std::chrono;
28 using namespace std::placeholders;
29 using namespace bmusb;
33 // TODO: Support stride.
34 void memcpy_interleaved(uint8_t *dest1, uint8_t *dest2, const uint8_t *src, size_t n)
37 uint8_t *dptr1 = dest1;
38 uint8_t *dptr2 = dest2;
40 for (size_t i = 0; i < n; i += 2) {
48 // Returns the number of bytes consumed.
49 size_t memcpy_interleaved_fastpath(uint8_t *dest1, uint8_t *dest2, const uint8_t *src, size_t n)
51 const uint8_t *limit = src + n;
54 // Align end to 32 bytes.
55 limit = (const uint8_t *)(intptr_t(limit) & ~31);
61 // Process [0,31] bytes, such that start gets aligned to 32 bytes.
62 const uint8_t *aligned_src = (const uint8_t *)(intptr_t(src + 31) & ~31);
63 if (aligned_src != src) {
64 size_t n2 = aligned_src - src;
65 memcpy_interleaved(dest1, dest2, src, n2);
75 // Make the length a multiple of 64.
76 if (((limit - src) % 64) != 0) {
79 assert(((limit - src) % 64) == 0);
82 const __m256i * __restrict in = (const __m256i *)src;
83 __m256i * __restrict out1 = (__m256i *)dest1;
84 __m256i * __restrict out2 = (__m256i *)dest2;
86 __m256i shuffle_cw = _mm256_set_epi8(
87 15, 13, 11, 9, 7, 5, 3, 1, 14, 12, 10, 8, 6, 4, 2, 0,
88 15, 13, 11, 9, 7, 5, 3, 1, 14, 12, 10, 8, 6, 4, 2, 0);
89 while (in < (const __m256i *)limit) {
90 // Note: For brevity, comments show lanes as if they were 2x64-bit (they're actually 2x128).
91 __m256i data1 = _mm256_stream_load_si256(in); // AaBbCcDd EeFfGgHh
92 __m256i data2 = _mm256_stream_load_si256(in + 1); // IiJjKkLl MmNnOoPp
94 data1 = _mm256_shuffle_epi8(data1, shuffle_cw); // ABCDabcd EFGHefgh
95 data2 = _mm256_shuffle_epi8(data2, shuffle_cw); // IJKLijkl MNOPmnop
97 data1 = _mm256_permute4x64_epi64(data1, 0b11011000); // ABCDEFGH abcdefgh
98 data2 = _mm256_permute4x64_epi64(data2, 0b11011000); // IJKLMNOP ijklmnop
100 __m256i lo = _mm256_permute2x128_si256(data1, data2, 0b00100000);
101 __m256i hi = _mm256_permute2x128_si256(data1, data2, 0b00110001);
103 _mm256_storeu_si256(out1, lo);
104 _mm256_storeu_si256(out2, hi);
112 const __m128i * __restrict in = (const __m128i *)src;
113 __m128i * __restrict out1 = (__m128i *)dest1;
114 __m128i * __restrict out2 = (__m128i *)dest2;
116 __m128i mask_lower_byte = _mm_set1_epi16(0x00ff);
117 while (in < (const __m128i *)limit) {
118 __m128i data1 = _mm_load_si128(in);
119 __m128i data2 = _mm_load_si128(in + 1);
120 __m128i data1_lo = _mm_and_si128(data1, mask_lower_byte);
121 __m128i data2_lo = _mm_and_si128(data2, mask_lower_byte);
122 __m128i data1_hi = _mm_srli_epi16(data1, 8);
123 __m128i data2_hi = _mm_srli_epi16(data2, 8);
124 __m128i lo = _mm_packus_epi16(data1_lo, data2_lo);
125 _mm_storeu_si128(out1, lo);
126 __m128i hi = _mm_packus_epi16(data1_hi, data2_hi);
127 _mm_storeu_si128(out2, hi);
143 DeckLinkCapture::DeckLinkCapture(IDeckLink *card, int card_index)
144 : card_index(card_index), card(card)
147 const char *model_name;
149 if (card->GetModelName(&model_name) == S_OK) {
150 snprintf(buf, sizeof(buf), "PCI card %d: %s", card_index, model_name);
152 snprintf(buf, sizeof(buf), "PCI card %d: Unknown DeckLink card", card_index);
157 if (card->QueryInterface(IID_IDeckLinkInput, (void**)&input) != S_OK) {
158 fprintf(stderr, "Card %d has no inputs\n", card_index);
162 IDeckLinkAttributes *attr;
163 if (card->QueryInterface(IID_IDeckLinkAttributes, (void**)&attr) != S_OK) {
164 fprintf(stderr, "Card %d has no attributes\n", card_index);
168 // Get the list of available video inputs.
169 int64_t video_input_mask;
170 if (attr->GetInt(BMDDeckLinkVideoInputConnections, &video_input_mask) != S_OK) {
171 fprintf(stderr, "Failed to enumerate video inputs for card %d\n", card_index);
174 const vector<pair<BMDVideoConnection, string>> video_input_types = {
175 { bmdVideoConnectionSDI, "SDI" },
176 { bmdVideoConnectionHDMI, "HDMI" },
177 { bmdVideoConnectionOpticalSDI, "Optical SDI" },
178 { bmdVideoConnectionComponent, "Component" },
179 { bmdVideoConnectionComposite, "Composite" },
180 { bmdVideoConnectionSVideo, "S-Video" }
182 for (const auto &video_input : video_input_types) {
183 if (video_input_mask & video_input.first) {
184 video_inputs.emplace(video_input.first, video_input.second);
188 // And then the available audio inputs.
189 int64_t audio_input_mask;
190 if (attr->GetInt(BMDDeckLinkAudioInputConnections, &audio_input_mask) != S_OK) {
191 fprintf(stderr, "Failed to enumerate audio inputs for card %d\n", card_index);
194 const vector<pair<BMDAudioConnection, string>> audio_input_types = {
195 { bmdAudioConnectionEmbedded, "Embedded" },
196 { bmdAudioConnectionAESEBU, "AES/EBU" },
197 { bmdAudioConnectionAnalog, "Analog" },
198 { bmdAudioConnectionAnalogXLR, "Analog XLR" },
199 { bmdAudioConnectionAnalogRCA, "Analog RCA" },
200 { bmdAudioConnectionMicrophone, "Microphone" },
201 { bmdAudioConnectionHeadphones, "Headphones" }
203 for (const auto &audio_input : audio_input_types) {
204 if (audio_input_mask & audio_input.first) {
205 audio_inputs.emplace(audio_input.first, audio_input.second);
209 // Check if we the card supports input autodetection.
210 if (attr->GetFlag(BMDDeckLinkSupportsInputFormatDetection, &supports_autodetect) != S_OK) {
211 fprintf(stderr, "Warning: Failed to ask card %d whether it supports input format autodetection\n", card_index);
212 supports_autodetect = false;
215 // If there's more than one subdevice on this card, label them.
216 int64_t num_subdevices, subdevice_idx;
217 if (attr->GetInt(BMDDeckLinkNumberOfSubDevices, &num_subdevices) == S_OK && num_subdevices > 1) {
218 if (attr->GetInt(BMDDeckLinkSubDeviceIndex, &subdevice_idx) == S_OK) {
220 snprintf(buf, sizeof(buf), " (subdevice %d)", int(subdevice_idx));
227 /* Set up the video and audio sources. */
228 if (card->QueryInterface(IID_IDeckLinkConfiguration, (void**)&config) != S_OK) {
229 fprintf(stderr, "Failed to get configuration interface for card %d\n", card_index);
233 BMDVideoConnection connection = pick_default_video_connection(card, BMDDeckLinkVideoInputConnections, card_index);
235 set_video_input(connection);
236 set_audio_input(bmdAudioConnectionEmbedded);
238 IDeckLinkDisplayModeIterator *mode_it;
239 if (input->GetDisplayModeIterator(&mode_it) != S_OK) {
240 fprintf(stderr, "Failed to enumerate display modes for card %d\n", card_index);
244 video_modes = summarize_video_modes(mode_it, card_index);
247 set_video_mode_no_restart(bmdModeHD720p5994);
249 input->SetCallback(this);
252 DeckLinkCapture::~DeckLinkCapture()
254 if (has_dequeue_callbacks) {
255 dequeue_cleanup_callback();
262 HRESULT STDMETHODCALLTYPE DeckLinkCapture::QueryInterface(REFIID, LPVOID *)
264 return E_NOINTERFACE;
267 ULONG STDMETHODCALLTYPE DeckLinkCapture::AddRef(void)
269 return refcount.fetch_add(1) + 1;
272 ULONG STDMETHODCALLTYPE DeckLinkCapture::Release(void)
274 int new_ref = refcount.fetch_sub(1) - 1;
280 HRESULT STDMETHODCALLTYPE DeckLinkCapture::VideoInputFormatChanged(
281 BMDVideoInputFormatChangedEvents,
282 IDeckLinkDisplayMode* display_mode,
283 BMDDetectedVideoInputFormatFlags format_flags)
285 if (format_flags & bmdDetectedVideoInputRGB444) {
286 fprintf(stderr, "WARNING: Input detected as 4:4:4 RGB, but Nageru can't consume that yet.\n");
287 fprintf(stderr, "Doing hardware conversion to 4:2:2 Y'CbCr.\n");
289 if (supports_autodetect && display_mode->GetDisplayMode() != current_video_mode) {
290 set_video_mode(display_mode->GetDisplayMode());
292 if (display_mode->GetFrameRate(&frame_duration, &time_scale) != S_OK) {
293 fprintf(stderr, "Could not get new frame rate\n");
296 field_dominance = display_mode->GetFieldDominance();
300 HRESULT STDMETHODCALLTYPE DeckLinkCapture::VideoInputFrameArrived(
301 IDeckLinkVideoInputFrame *video_frame,
302 IDeckLinkAudioInputPacket *audio_frame)
305 char thread_name[16];
306 snprintf(thread_name, sizeof(thread_name), "DeckLink_C_%d", card_index);
307 pthread_setname_np(pthread_self(), thread_name);
308 if (has_dequeue_callbacks) {
309 dequeue_init_callback();
314 steady_clock::time_point now = steady_clock::now();
316 FrameAllocator::Frame current_video_frame, current_audio_frame;
317 VideoFormat video_format;
318 AudioFormat audio_format;
320 video_format.frame_rate_nom = time_scale;
321 video_format.frame_rate_den = frame_duration;
322 // TODO: Respect the TFF/BFF flag.
323 video_format.interlaced = (field_dominance == bmdLowerFieldFirst || field_dominance == bmdUpperFieldFirst);
324 video_format.second_field_start = 1;
327 video_format.has_signal = !(video_frame->GetFlags() & bmdFrameHasNoInputSource);
329 int width = video_frame->GetWidth();
330 int height = video_frame->GetHeight();
331 const int stride = video_frame->GetRowBytes();
332 assert(stride == width * 2);
334 current_video_frame = video_frame_allocator->alloc_frame();
335 if (current_video_frame.data != nullptr) {
336 const uint8_t *frame_bytes;
337 video_frame->GetBytes((void **)&frame_bytes);
338 size_t num_bytes = stride * height;
340 if (current_video_frame.interleaved) {
341 uint8_t *data = current_video_frame.data;
342 uint8_t *data2 = current_video_frame.data2;
344 size_t consumed = memcpy_interleaved_fastpath(data, data2, frame_bytes, num_bytes);
345 frame_bytes += consumed;
346 data += consumed / 2;
347 data2 += consumed / 2;
351 current_video_frame.len += consumed;
352 num_bytes -= consumed;
356 memcpy_interleaved(data, data2, frame_bytes, num_bytes);
359 memcpy(current_video_frame.data, frame_bytes, num_bytes);
361 current_video_frame.len += num_bytes;
363 video_format.width = width;
364 video_format.height = height;
369 int num_samples = audio_frame->GetSampleFrameCount();
371 current_audio_frame = audio_frame_allocator->alloc_frame();
372 if (current_audio_frame.data != nullptr) {
373 const uint8_t *frame_bytes;
374 audio_frame->GetBytes((void **)&frame_bytes);
375 current_audio_frame.len = sizeof(int32_t) * 2 * num_samples;
377 memcpy(current_audio_frame.data, frame_bytes, current_audio_frame.len);
379 audio_format.bits_per_sample = 32;
380 audio_format.num_channels = 2;
384 current_video_frame.received_timestamp = now;
385 current_audio_frame.received_timestamp = now;
387 if (current_video_frame.data != nullptr || current_audio_frame.data != nullptr) {
388 // TODO: Put into a queue and put into a dequeue thread, if the
389 // BlackMagic drivers don't already do that for us?
390 frame_callback(timecode,
391 current_video_frame, /*video_offset=*/0, video_format,
392 current_audio_frame, /*audio_offset=*/0, audio_format);
399 void DeckLinkCapture::configure_card()
401 if (video_frame_allocator == nullptr) {
402 owned_video_frame_allocator.reset(new MallocFrameAllocator(FRAME_SIZE, NUM_QUEUED_VIDEO_FRAMES));
403 set_video_frame_allocator(owned_video_frame_allocator.get());
405 if (audio_frame_allocator == nullptr) {
406 owned_audio_frame_allocator.reset(new MallocFrameAllocator(65536, NUM_QUEUED_AUDIO_FRAMES));
407 set_audio_frame_allocator(owned_audio_frame_allocator.get());
411 void DeckLinkCapture::start_bm_capture()
416 if (input->EnableVideoInput(current_video_mode, bmdFormat8BitYUV, supports_autodetect ? bmdVideoInputEnableFormatDetection : 0) != S_OK) {
417 fprintf(stderr, "Failed to set video mode 0x%04x for card %d\n", current_video_mode, card_index);
420 if (input->EnableAudioInput(48000, bmdAudioSampleType32bitInteger, 2) != S_OK) {
421 fprintf(stderr, "Failed to enable audio input for card %d\n", card_index);
425 if (input->StartStreams() != S_OK) {
426 fprintf(stderr, "StartStreams failed\n");
432 void DeckLinkCapture::stop_dequeue_thread()
437 HRESULT result = input->StopStreams();
438 if (result != S_OK) {
439 fprintf(stderr, "StopStreams failed with error 0x%x\n", result);
442 if (input->DisableVideoInput() != S_OK) {
443 fprintf(stderr, "Failed to disable video input for card %d\n", card_index);
446 if (input->DisableAudioInput() != S_OK) {
447 fprintf(stderr, "Failed to disable audio input for card %d\n", card_index);
453 void DeckLinkCapture::set_video_mode(uint32_t video_mode_id)
455 if (input->PauseStreams() != S_OK) {
456 fprintf(stderr, "PauseStreams failed\n");
459 if (input->FlushStreams() != S_OK) {
460 fprintf(stderr, "FlushStreams failed\n");
464 set_video_mode_no_restart(video_mode_id);
466 if (input->StartStreams() != S_OK) {
467 fprintf(stderr, "StartStreams failed\n");
472 void DeckLinkCapture::set_video_mode_no_restart(uint32_t video_mode_id)
474 BMDDisplayModeSupport support;
475 IDeckLinkDisplayMode *display_mode;
476 if (input->DoesSupportVideoMode(video_mode_id, bmdFormat8BitYUV, /*flags=*/0, &support, &display_mode)) {
477 fprintf(stderr, "Failed to query display mode for card %d\n", card_index);
481 if (support == bmdDisplayModeNotSupported) {
482 fprintf(stderr, "Card %d does not support display mode\n", card_index);
486 if (display_mode->GetFrameRate(&frame_duration, &time_scale) != S_OK) {
487 fprintf(stderr, "Could not get frame rate for card %d\n", card_index);
491 field_dominance = display_mode->GetFieldDominance();
494 if (input->EnableVideoInput(video_mode_id, bmdFormat8BitYUV, supports_autodetect ? bmdVideoInputEnableFormatDetection : 0) != S_OK) {
495 fprintf(stderr, "Failed to set video mode 0x%04x for card %d\n", video_mode_id, card_index);
500 current_video_mode = video_mode_id;
503 void DeckLinkCapture::set_video_input(uint32_t video_input_id)
505 if (config->SetInt(bmdDeckLinkConfigVideoInputConnection, video_input_id) != S_OK) {
506 fprintf(stderr, "Failed to set video input connection for card %d\n", card_index);
510 current_video_input = video_input_id;
513 void DeckLinkCapture::set_audio_input(uint32_t audio_input_id)
515 if (config->SetInt(bmdDeckLinkConfigAudioInputConnection, audio_input_id) != S_OK) {
516 fprintf(stderr, "Failed to set audio input connection for card %d\n", card_index);
520 current_audio_input = audio_input_id;