1 #include "decklink_capture.h"
3 #include <DeckLinkAPI.h>
4 #include <DeckLinkAPIConfiguration.h>
5 #include <DeckLinkAPIDiscovery.h>
6 #include <DeckLinkAPIModes.h>
20 #include "bmusb/bmusb.h"
22 #define FRAME_SIZE (8 << 20) // 8 MB.
25 using namespace std::chrono;
26 using namespace std::placeholders;
27 using namespace bmusb;
31 // TODO: Support stride.
32 void memcpy_interleaved(uint8_t *dest1, uint8_t *dest2, const uint8_t *src, size_t n)
35 uint8_t *dptr1 = dest1;
36 uint8_t *dptr2 = dest2;
38 for (size_t i = 0; i < n; i += 2) {
46 // Returns the number of bytes consumed.
47 size_t memcpy_interleaved_fastpath(uint8_t *dest1, uint8_t *dest2, const uint8_t *src, size_t n)
49 const uint8_t *limit = src + n;
52 // Align end to 32 bytes.
53 limit = (const uint8_t *)(intptr_t(limit) & ~31);
59 // Process [0,31] bytes, such that start gets aligned to 32 bytes.
60 const uint8_t *aligned_src = (const uint8_t *)(intptr_t(src + 31) & ~31);
61 if (aligned_src != src) {
62 size_t n2 = aligned_src - src;
63 memcpy_interleaved(dest1, dest2, src, n2);
73 // Make the length a multiple of 64.
74 if (((limit - src) % 64) != 0) {
77 assert(((limit - src) % 64) == 0);
80 const __m256i * __restrict in = (const __m256i *)src;
81 __m256i * __restrict out1 = (__m256i *)dest1;
82 __m256i * __restrict out2 = (__m256i *)dest2;
84 __m256i shuffle_cw = _mm256_set_epi8(
85 15, 13, 11, 9, 7, 5, 3, 1, 14, 12, 10, 8, 6, 4, 2, 0,
86 15, 13, 11, 9, 7, 5, 3, 1, 14, 12, 10, 8, 6, 4, 2, 0);
87 while (in < (const __m256i *)limit) {
88 // Note: For brevity, comments show lanes as if they were 2x64-bit (they're actually 2x128).
89 __m256i data1 = _mm256_stream_load_si256(in); // AaBbCcDd EeFfGgHh
90 __m256i data2 = _mm256_stream_load_si256(in + 1); // IiJjKkLl MmNnOoPp
92 data1 = _mm256_shuffle_epi8(data1, shuffle_cw); // ABCDabcd EFGHefgh
93 data2 = _mm256_shuffle_epi8(data2, shuffle_cw); // IJKLijkl MNOPmnop
95 data1 = _mm256_permute4x64_epi64(data1, 0b11011000); // ABCDEFGH abcdefgh
96 data2 = _mm256_permute4x64_epi64(data2, 0b11011000); // IJKLMNOP ijklmnop
98 __m256i lo = _mm256_permute2x128_si256(data1, data2, 0b00100000);
99 __m256i hi = _mm256_permute2x128_si256(data1, data2, 0b00110001);
101 _mm256_storeu_si256(out1, lo);
102 _mm256_storeu_si256(out2, hi);
110 const __m128i * __restrict in = (const __m128i *)src;
111 __m128i * __restrict out1 = (__m128i *)dest1;
112 __m128i * __restrict out2 = (__m128i *)dest2;
114 __m128i mask_lower_byte = _mm_set1_epi16(0x00ff);
115 while (in < (const __m128i *)limit) {
116 __m128i data1 = _mm_load_si128(in);
117 __m128i data2 = _mm_load_si128(in + 1);
118 __m128i data1_lo = _mm_and_si128(data1, mask_lower_byte);
119 __m128i data2_lo = _mm_and_si128(data2, mask_lower_byte);
120 __m128i data1_hi = _mm_srli_epi16(data1, 8);
121 __m128i data2_hi = _mm_srli_epi16(data2, 8);
122 __m128i lo = _mm_packus_epi16(data1_lo, data2_lo);
123 _mm_storeu_si128(out1, lo);
124 __m128i hi = _mm_packus_epi16(data1_hi, data2_hi);
125 _mm_storeu_si128(out2, hi);
141 DeckLinkCapture::DeckLinkCapture(IDeckLink *card, int card_index)
142 : card_index(card_index), card(card)
145 const char *model_name;
147 if (card->GetModelName(&model_name) == S_OK) {
148 snprintf(buf, sizeof(buf), "PCI card %d: %s", card_index, model_name);
150 snprintf(buf, sizeof(buf), "PCI card %d: Unknown DeckLink card", card_index);
155 if (card->QueryInterface(IID_IDeckLinkInput, (void**)&input) != S_OK) {
156 fprintf(stderr, "Card %d has no inputs\n", card_index);
160 IDeckLinkAttributes *attr;
161 if (card->QueryInterface(IID_IDeckLinkAttributes, (void**)&attr) != S_OK) {
162 fprintf(stderr, "Card %d has no attributes\n", card_index);
166 // Get the list of available video inputs.
167 int64_t video_input_mask;
168 if (attr->GetInt(BMDDeckLinkVideoInputConnections, &video_input_mask) != S_OK) {
169 fprintf(stderr, "Failed to enumerate video inputs for card %d\n", card_index);
172 const vector<pair<BMDVideoConnection, string>> video_input_types = {
173 { bmdVideoConnectionSDI, "SDI" },
174 { bmdVideoConnectionHDMI, "HDMI" },
175 { bmdVideoConnectionOpticalSDI, "Optical SDI" },
176 { bmdVideoConnectionComponent, "Component" },
177 { bmdVideoConnectionComposite, "Composite" },
178 { bmdVideoConnectionSVideo, "S-Video" }
180 for (const auto &video_input : video_input_types) {
181 if (video_input_mask & video_input.first) {
182 video_inputs.emplace(video_input.first, video_input.second);
186 // And then the available audio inputs.
187 int64_t audio_input_mask;
188 if (attr->GetInt(BMDDeckLinkAudioInputConnections, &audio_input_mask) != S_OK) {
189 fprintf(stderr, "Failed to enumerate audio inputs for card %d\n", card_index);
192 const vector<pair<BMDAudioConnection, string>> audio_input_types = {
193 { bmdAudioConnectionEmbedded, "Embedded" },
194 { bmdAudioConnectionAESEBU, "AES/EBU" },
195 { bmdAudioConnectionAnalog, "Analog" },
196 { bmdAudioConnectionAnalogXLR, "Analog XLR" },
197 { bmdAudioConnectionAnalogRCA, "Analog RCA" },
198 { bmdAudioConnectionMicrophone, "Microphone" },
199 { bmdAudioConnectionHeadphones, "Headphones" }
201 for (const auto &audio_input : audio_input_types) {
202 if (audio_input_mask & audio_input.first) {
203 audio_inputs.emplace(audio_input.first, audio_input.second);
209 /* Set up the video and audio sources. */
210 if (card->QueryInterface(IID_IDeckLinkConfiguration, (void**)&config) != S_OK) {
211 fprintf(stderr, "Failed to get configuration interface for card %d\n", card_index);
215 set_video_input(bmdVideoConnectionHDMI);
216 set_audio_input(bmdAudioConnectionEmbedded);
218 IDeckLinkDisplayModeIterator *mode_it;
219 if (input->GetDisplayModeIterator(&mode_it) != S_OK) {
220 fprintf(stderr, "Failed to enumerate display modes for card %d\n", card_index);
224 for (IDeckLinkDisplayMode *mode_ptr; mode_it->Next(&mode_ptr) == S_OK; mode_ptr->Release()) {
227 const char *mode_name;
228 if (mode_ptr->GetName(&mode_name) != S_OK) {
229 mode.name = "Unknown mode";
231 mode.name = mode_name;
234 mode.autodetect = false;
236 mode.width = mode_ptr->GetWidth();
237 mode.height = mode_ptr->GetHeight();
239 BMDTimeScale frame_rate_num;
240 BMDTimeValue frame_rate_den;
241 if (mode_ptr->GetFrameRate(&frame_rate_den, &frame_rate_num) != S_OK) {
242 fprintf(stderr, "Could not get frame rate for mode '%s' on card %d\n", mode.name.c_str(), card_index);
245 mode.frame_rate_num = frame_rate_num;
246 mode.frame_rate_den = frame_rate_den;
248 // TODO: Respect the TFF/BFF flag.
249 mode.interlaced = (mode_ptr->GetFieldDominance() == bmdLowerFieldFirst || mode_ptr->GetFieldDominance() == bmdUpperFieldFirst);
251 uint32_t id = mode_ptr->GetDisplayMode();
252 video_modes.insert(make_pair(id, mode));
255 set_video_mode_no_restart(bmdModeHD720p5994);
257 if (input->EnableAudioInput(48000, bmdAudioSampleType32bitInteger, 2) != S_OK) {
258 fprintf(stderr, "Failed to enable audio input for card %d\n", card_index);
262 input->SetCallback(this);
265 DeckLinkCapture::~DeckLinkCapture()
267 if (has_dequeue_callbacks) {
268 dequeue_cleanup_callback();
275 HRESULT STDMETHODCALLTYPE DeckLinkCapture::QueryInterface(REFIID, LPVOID *)
277 return E_NOINTERFACE;
280 ULONG STDMETHODCALLTYPE DeckLinkCapture::AddRef(void)
282 return refcount.fetch_add(1) + 1;
285 ULONG STDMETHODCALLTYPE DeckLinkCapture::Release(void)
287 int new_ref = refcount.fetch_sub(1) - 1;
293 HRESULT STDMETHODCALLTYPE DeckLinkCapture::VideoInputFormatChanged(
294 BMDVideoInputFormatChangedEvents,
295 IDeckLinkDisplayMode* display_mode,
296 BMDDetectedVideoInputFormatFlags)
298 if (display_mode->GetFrameRate(&frame_duration, &time_scale) != S_OK) {
299 fprintf(stderr, "Could not get new frame rate\n");
305 HRESULT STDMETHODCALLTYPE DeckLinkCapture::VideoInputFrameArrived(
306 IDeckLinkVideoInputFrame *video_frame,
307 IDeckLinkAudioInputPacket *audio_frame)
310 if (has_dequeue_callbacks) {
311 dequeue_init_callback();
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;
324 video_format.has_signal = !(video_frame->GetFlags() & bmdFrameHasNoInputSource);
326 int width = video_frame->GetWidth();
327 int height = video_frame->GetHeight();
328 const int stride = video_frame->GetRowBytes();
329 assert(stride == width * 2);
331 current_video_frame = video_frame_allocator->alloc_frame();
332 if (current_video_frame.data != nullptr) {
333 const uint8_t *frame_bytes;
334 video_frame->GetBytes((void **)&frame_bytes);
336 uint8_t *data = current_video_frame.data;
337 uint8_t *data2 = current_video_frame.data2;
338 size_t num_bytes = width * height * 2;
340 size_t consumed = memcpy_interleaved_fastpath(data, data2, frame_bytes, num_bytes);
341 frame_bytes += consumed;
342 data += consumed / 2;
343 data2 += consumed / 2;
347 current_video_frame.len += consumed;
348 num_bytes -= consumed;
352 memcpy_interleaved(data, data2, frame_bytes, num_bytes);
354 current_video_frame.len += num_bytes;
356 video_format.width = width;
357 video_format.height = height;
359 current_video_frame.received_timestamp = steady_clock::now();
364 int num_samples = audio_frame->GetSampleFrameCount();
366 current_audio_frame = audio_frame_allocator->alloc_frame();
367 if (current_audio_frame.data != nullptr) {
368 const uint8_t *frame_bytes;
369 audio_frame->GetBytes((void **)&frame_bytes);
370 current_audio_frame.len = sizeof(int32_t) * 2 * num_samples;
372 memcpy(current_audio_frame.data, frame_bytes, current_audio_frame.len);
374 audio_format.bits_per_sample = 32;
375 audio_format.num_channels = 2;
377 current_audio_frame.received_timestamp = steady_clock::now();
381 if (current_video_frame.data != nullptr || current_audio_frame.data != nullptr) {
382 // TODO: Put into a queue and put into a dequeue thread, if the
383 // BlackMagic drivers don't already do that for us?
384 frame_callback(timecode,
385 current_video_frame, /*video_offset=*/0, video_format,
386 current_audio_frame, /*audio_offset=*/0, audio_format);
393 void DeckLinkCapture::configure_card()
395 if (video_frame_allocator == nullptr) {
396 owned_video_frame_allocator.reset(new MallocFrameAllocator(FRAME_SIZE, NUM_QUEUED_VIDEO_FRAMES));
397 set_video_frame_allocator(owned_video_frame_allocator.get());
399 if (audio_frame_allocator == nullptr) {
400 owned_audio_frame_allocator.reset(new MallocFrameAllocator(65536, NUM_QUEUED_AUDIO_FRAMES));
401 set_audio_frame_allocator(owned_audio_frame_allocator.get());
405 void DeckLinkCapture::start_bm_capture()
407 if (input->StartStreams() != S_OK) {
408 fprintf(stderr, "StartStreams failed\n");
413 void DeckLinkCapture::stop_dequeue_thread()
415 if (input->StopStreams() != S_OK) {
416 fprintf(stderr, "StopStreams failed\n");
421 void DeckLinkCapture::set_video_mode(uint32_t video_mode_id)
423 if (input->StopStreams() != S_OK) {
424 fprintf(stderr, "StopStreams failed\n");
428 set_video_mode_no_restart(video_mode_id);
430 if (input->StartStreams() != S_OK) {
431 fprintf(stderr, "StartStreams failed\n");
436 void DeckLinkCapture::set_video_mode_no_restart(uint32_t video_mode_id)
438 BMDDisplayModeSupport support;
439 IDeckLinkDisplayMode *display_mode;
440 if (input->DoesSupportVideoMode(video_mode_id, bmdFormat8BitYUV, /*flags=*/0, &support, &display_mode)) {
441 fprintf(stderr, "Failed to query display mode for card %d\n", card_index);
445 if (support == bmdDisplayModeNotSupported) {
446 fprintf(stderr, "Card %d does not support display mode\n", card_index);
450 if (display_mode->GetFrameRate(&frame_duration, &time_scale) != S_OK) {
451 fprintf(stderr, "Could not get frame rate for card %d\n", card_index);
455 if (input->EnableVideoInput(video_mode_id, bmdFormat8BitYUV, 0) != S_OK) {
456 fprintf(stderr, "Failed to set video mode 0x%04x for card %d\n", video_mode_id, card_index);
460 current_video_mode = video_mode_id;
463 void DeckLinkCapture::set_video_input(uint32_t video_input_id)
465 if (config->SetInt(bmdDeckLinkConfigVideoInputConnection, video_input_id) != S_OK) {
466 fprintf(stderr, "Failed to set video input connection for card %d\n", card_index);
470 current_video_input = video_input_id;
473 void DeckLinkCapture::set_audio_input(uint32_t audio_input_id)
475 if (config->SetInt(bmdDeckLinkConfigAudioInputConnection, audio_input_id) != S_OK) {
476 fprintf(stderr, "Failed to set audio input connection for card %d\n", card_index);
480 current_audio_input = audio_input_id;