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"
21 #include "decklink_util.h"
23 #define FRAME_SIZE (8 << 20) // 8 MB.
26 using namespace std::chrono;
27 using namespace std::placeholders;
28 using namespace bmusb;
32 // TODO: Support stride.
33 void memcpy_interleaved(uint8_t *dest1, uint8_t *dest2, const uint8_t *src, size_t n)
36 uint8_t *dptr1 = dest1;
37 uint8_t *dptr2 = dest2;
39 for (size_t i = 0; i < n; i += 2) {
47 // Returns the number of bytes consumed.
48 size_t memcpy_interleaved_fastpath(uint8_t *dest1, uint8_t *dest2, const uint8_t *src, size_t n)
50 const uint8_t *limit = src + n;
53 // Align end to 32 bytes.
54 limit = (const uint8_t *)(intptr_t(limit) & ~31);
60 // Process [0,31] bytes, such that start gets aligned to 32 bytes.
61 const uint8_t *aligned_src = (const uint8_t *)(intptr_t(src + 31) & ~31);
62 if (aligned_src != src) {
63 size_t n2 = aligned_src - src;
64 memcpy_interleaved(dest1, dest2, src, n2);
74 // Make the length a multiple of 64.
75 if (((limit - src) % 64) != 0) {
78 assert(((limit - src) % 64) == 0);
81 const __m256i * __restrict in = (const __m256i *)src;
82 __m256i * __restrict out1 = (__m256i *)dest1;
83 __m256i * __restrict out2 = (__m256i *)dest2;
85 __m256i shuffle_cw = _mm256_set_epi8(
86 15, 13, 11, 9, 7, 5, 3, 1, 14, 12, 10, 8, 6, 4, 2, 0,
87 15, 13, 11, 9, 7, 5, 3, 1, 14, 12, 10, 8, 6, 4, 2, 0);
88 while (in < (const __m256i *)limit) {
89 // Note: For brevity, comments show lanes as if they were 2x64-bit (they're actually 2x128).
90 __m256i data1 = _mm256_stream_load_si256(in); // AaBbCcDd EeFfGgHh
91 __m256i data2 = _mm256_stream_load_si256(in + 1); // IiJjKkLl MmNnOoPp
93 data1 = _mm256_shuffle_epi8(data1, shuffle_cw); // ABCDabcd EFGHefgh
94 data2 = _mm256_shuffle_epi8(data2, shuffle_cw); // IJKLijkl MNOPmnop
96 data1 = _mm256_permute4x64_epi64(data1, 0b11011000); // ABCDEFGH abcdefgh
97 data2 = _mm256_permute4x64_epi64(data2, 0b11011000); // IJKLMNOP ijklmnop
99 __m256i lo = _mm256_permute2x128_si256(data1, data2, 0b00100000);
100 __m256i hi = _mm256_permute2x128_si256(data1, data2, 0b00110001);
102 _mm256_storeu_si256(out1, lo);
103 _mm256_storeu_si256(out2, hi);
111 const __m128i * __restrict in = (const __m128i *)src;
112 __m128i * __restrict out1 = (__m128i *)dest1;
113 __m128i * __restrict out2 = (__m128i *)dest2;
115 __m128i mask_lower_byte = _mm_set1_epi16(0x00ff);
116 while (in < (const __m128i *)limit) {
117 __m128i data1 = _mm_load_si128(in);
118 __m128i data2 = _mm_load_si128(in + 1);
119 __m128i data1_lo = _mm_and_si128(data1, mask_lower_byte);
120 __m128i data2_lo = _mm_and_si128(data2, mask_lower_byte);
121 __m128i data1_hi = _mm_srli_epi16(data1, 8);
122 __m128i data2_hi = _mm_srli_epi16(data2, 8);
123 __m128i lo = _mm_packus_epi16(data1_lo, data2_lo);
124 _mm_storeu_si128(out1, lo);
125 __m128i hi = _mm_packus_epi16(data1_hi, data2_hi);
126 _mm_storeu_si128(out2, hi);
142 DeckLinkCapture::DeckLinkCapture(IDeckLink *card, int card_index)
143 : card_index(card_index), card(card)
146 const char *model_name;
148 if (card->GetModelName(&model_name) == S_OK) {
149 snprintf(buf, sizeof(buf), "PCI card %d: %s", card_index, model_name);
151 snprintf(buf, sizeof(buf), "PCI card %d: Unknown DeckLink card", card_index);
156 if (card->QueryInterface(IID_IDeckLinkInput, (void**)&input) != S_OK) {
157 fprintf(stderr, "Card %d has no inputs\n", card_index);
161 IDeckLinkAttributes *attr;
162 if (card->QueryInterface(IID_IDeckLinkAttributes, (void**)&attr) != S_OK) {
163 fprintf(stderr, "Card %d has no attributes\n", card_index);
167 // Get the list of available video inputs.
168 int64_t video_input_mask;
169 if (attr->GetInt(BMDDeckLinkVideoInputConnections, &video_input_mask) != S_OK) {
170 fprintf(stderr, "Failed to enumerate video inputs for card %d\n", card_index);
173 const vector<pair<BMDVideoConnection, string>> video_input_types = {
174 { bmdVideoConnectionSDI, "SDI" },
175 { bmdVideoConnectionHDMI, "HDMI" },
176 { bmdVideoConnectionOpticalSDI, "Optical SDI" },
177 { bmdVideoConnectionComponent, "Component" },
178 { bmdVideoConnectionComposite, "Composite" },
179 { bmdVideoConnectionSVideo, "S-Video" }
181 for (const auto &video_input : video_input_types) {
182 if (video_input_mask & video_input.first) {
183 video_inputs.emplace(video_input.first, video_input.second);
187 // And then the available audio inputs.
188 int64_t audio_input_mask;
189 if (attr->GetInt(BMDDeckLinkAudioInputConnections, &audio_input_mask) != S_OK) {
190 fprintf(stderr, "Failed to enumerate audio inputs for card %d\n", card_index);
193 const vector<pair<BMDAudioConnection, string>> audio_input_types = {
194 { bmdAudioConnectionEmbedded, "Embedded" },
195 { bmdAudioConnectionAESEBU, "AES/EBU" },
196 { bmdAudioConnectionAnalog, "Analog" },
197 { bmdAudioConnectionAnalogXLR, "Analog XLR" },
198 { bmdAudioConnectionAnalogRCA, "Analog RCA" },
199 { bmdAudioConnectionMicrophone, "Microphone" },
200 { bmdAudioConnectionHeadphones, "Headphones" }
202 for (const auto &audio_input : audio_input_types) {
203 if (audio_input_mask & audio_input.first) {
204 audio_inputs.emplace(audio_input.first, audio_input.second);
208 // Check if we the card supports input autodetection.
209 if (attr->GetFlag(BMDDeckLinkSupportsInputFormatDetection, &supports_autodetect) != S_OK) {
210 fprintf(stderr, "Warning: Failed to ask card %d whether it supports input format autodetection\n", card_index);
211 supports_autodetect = false;
214 // If there's more than one subdevice on this card, label them.
215 int64_t num_subdevices, subdevice_idx;
216 if (attr->GetInt(BMDDeckLinkNumberOfSubDevices, &num_subdevices) == S_OK && num_subdevices > 1) {
217 if (attr->GetInt(BMDDeckLinkSubDeviceIndex, &subdevice_idx) == S_OK) {
219 snprintf(buf, sizeof(buf), " (subdevice %d)", int(subdevice_idx));
226 /* Set up the video and audio sources. */
227 if (card->QueryInterface(IID_IDeckLinkConfiguration, (void**)&config) != S_OK) {
228 fprintf(stderr, "Failed to get configuration interface for card %d\n", card_index);
232 BMDVideoConnection connection = pick_default_video_connection(card, BMDDeckLinkVideoInputConnections, card_index);
234 set_video_input(connection);
235 set_audio_input(bmdAudioConnectionEmbedded);
237 IDeckLinkDisplayModeIterator *mode_it;
238 if (input->GetDisplayModeIterator(&mode_it) != S_OK) {
239 fprintf(stderr, "Failed to enumerate display modes for card %d\n", card_index);
243 video_modes = summarize_video_modes(mode_it, card_index);
246 set_video_mode_no_restart(bmdModeHD720p5994);
248 input->SetCallback(this);
251 DeckLinkCapture::~DeckLinkCapture()
253 if (has_dequeue_callbacks) {
254 dequeue_cleanup_callback();
261 HRESULT STDMETHODCALLTYPE DeckLinkCapture::QueryInterface(REFIID, LPVOID *)
263 return E_NOINTERFACE;
266 ULONG STDMETHODCALLTYPE DeckLinkCapture::AddRef(void)
268 return refcount.fetch_add(1) + 1;
271 ULONG STDMETHODCALLTYPE DeckLinkCapture::Release(void)
273 int new_ref = refcount.fetch_sub(1) - 1;
279 HRESULT STDMETHODCALLTYPE DeckLinkCapture::VideoInputFormatChanged(
280 BMDVideoInputFormatChangedEvents,
281 IDeckLinkDisplayMode* display_mode,
282 BMDDetectedVideoInputFormatFlags format_flags)
284 if (format_flags & bmdDetectedVideoInputRGB444) {
285 fprintf(stderr, "WARNING: Input detected as 4:4:4 RGB, but Nageru can't consume that yet.\n");
286 fprintf(stderr, "Doing hardware conversion to 4:2:2 Y'CbCr.\n");
288 if (supports_autodetect && display_mode->GetDisplayMode() != current_video_mode) {
289 set_video_mode(display_mode->GetDisplayMode());
291 if (display_mode->GetFrameRate(&frame_duration, &time_scale) != S_OK) {
292 fprintf(stderr, "Could not get new frame rate\n");
295 field_dominance = display_mode->GetFieldDominance();
299 HRESULT STDMETHODCALLTYPE DeckLinkCapture::VideoInputFrameArrived(
300 IDeckLinkVideoInputFrame *video_frame,
301 IDeckLinkAudioInputPacket *audio_frame)
304 if (has_dequeue_callbacks) {
305 dequeue_init_callback();
310 FrameAllocator::Frame current_video_frame, current_audio_frame;
311 VideoFormat video_format;
312 AudioFormat audio_format;
314 video_format.frame_rate_nom = time_scale;
315 video_format.frame_rate_den = frame_duration;
316 // TODO: Respect the TFF/BFF flag.
317 video_format.interlaced = (field_dominance == bmdLowerFieldFirst || field_dominance == bmdUpperFieldFirst);
318 video_format.second_field_start = 1;
321 video_format.has_signal = !(video_frame->GetFlags() & bmdFrameHasNoInputSource);
323 int width = video_frame->GetWidth();
324 int height = video_frame->GetHeight();
325 const int stride = video_frame->GetRowBytes();
326 assert(stride == width * 2);
328 current_video_frame = video_frame_allocator->alloc_frame();
329 if (current_video_frame.data != nullptr) {
330 const uint8_t *frame_bytes;
331 video_frame->GetBytes((void **)&frame_bytes);
333 uint8_t *data = current_video_frame.data;
334 uint8_t *data2 = current_video_frame.data2;
335 size_t num_bytes = width * height * 2;
337 size_t consumed = memcpy_interleaved_fastpath(data, data2, frame_bytes, num_bytes);
338 frame_bytes += consumed;
339 data += consumed / 2;
340 data2 += consumed / 2;
344 current_video_frame.len += consumed;
345 num_bytes -= consumed;
349 memcpy_interleaved(data, data2, frame_bytes, num_bytes);
351 current_video_frame.len += num_bytes;
353 video_format.width = width;
354 video_format.height = height;
356 current_video_frame.received_timestamp = steady_clock::now();
361 int num_samples = audio_frame->GetSampleFrameCount();
363 current_audio_frame = audio_frame_allocator->alloc_frame();
364 if (current_audio_frame.data != nullptr) {
365 const uint8_t *frame_bytes;
366 audio_frame->GetBytes((void **)&frame_bytes);
367 current_audio_frame.len = sizeof(int32_t) * 2 * num_samples;
369 memcpy(current_audio_frame.data, frame_bytes, current_audio_frame.len);
371 audio_format.bits_per_sample = 32;
372 audio_format.num_channels = 2;
374 current_audio_frame.received_timestamp = steady_clock::now();
378 if (current_video_frame.data != nullptr || current_audio_frame.data != nullptr) {
379 // TODO: Put into a queue and put into a dequeue thread, if the
380 // BlackMagic drivers don't already do that for us?
381 frame_callback(timecode,
382 current_video_frame, /*video_offset=*/0, video_format,
383 current_audio_frame, /*audio_offset=*/0, audio_format);
390 void DeckLinkCapture::configure_card()
392 if (video_frame_allocator == nullptr) {
393 owned_video_frame_allocator.reset(new MallocFrameAllocator(FRAME_SIZE, NUM_QUEUED_VIDEO_FRAMES));
394 set_video_frame_allocator(owned_video_frame_allocator.get());
396 if (audio_frame_allocator == nullptr) {
397 owned_audio_frame_allocator.reset(new MallocFrameAllocator(65536, NUM_QUEUED_AUDIO_FRAMES));
398 set_audio_frame_allocator(owned_audio_frame_allocator.get());
402 void DeckLinkCapture::start_bm_capture()
407 if (input->EnableVideoInput(current_video_mode, bmdFormat8BitYUV, supports_autodetect ? bmdVideoInputEnableFormatDetection : 0) != S_OK) {
408 fprintf(stderr, "Failed to set video mode 0x%04x for card %d\n", current_video_mode, card_index);
411 if (input->EnableAudioInput(48000, bmdAudioSampleType32bitInteger, 2) != S_OK) {
412 fprintf(stderr, "Failed to enable audio input for card %d\n", card_index);
416 if (input->StartStreams() != S_OK) {
417 fprintf(stderr, "StartStreams failed\n");
423 void DeckLinkCapture::stop_dequeue_thread()
428 HRESULT result = input->StopStreams();
429 if (result != S_OK) {
430 fprintf(stderr, "StopStreams failed with error 0x%x\n", result);
433 if (input->DisableVideoInput() != S_OK) {
434 fprintf(stderr, "Failed to disable video input for card %d\n", card_index);
437 if (input->DisableAudioInput() != S_OK) {
438 fprintf(stderr, "Failed to disable audio input for card %d\n", card_index);
444 void DeckLinkCapture::set_video_mode(uint32_t video_mode_id)
446 if (input->PauseStreams() != S_OK) {
447 fprintf(stderr, "PauseStreams failed\n");
450 if (input->FlushStreams() != S_OK) {
451 fprintf(stderr, "FlushStreams failed\n");
455 set_video_mode_no_restart(video_mode_id);
457 if (input->StartStreams() != S_OK) {
458 fprintf(stderr, "StartStreams failed\n");
463 void DeckLinkCapture::set_video_mode_no_restart(uint32_t video_mode_id)
465 BMDDisplayModeSupport support;
466 IDeckLinkDisplayMode *display_mode;
467 if (input->DoesSupportVideoMode(video_mode_id, bmdFormat8BitYUV, /*flags=*/0, &support, &display_mode)) {
468 fprintf(stderr, "Failed to query display mode for card %d\n", card_index);
472 if (support == bmdDisplayModeNotSupported) {
473 fprintf(stderr, "Card %d does not support display mode\n", card_index);
477 if (display_mode->GetFrameRate(&frame_duration, &time_scale) != S_OK) {
478 fprintf(stderr, "Could not get frame rate for card %d\n", card_index);
482 field_dominance = display_mode->GetFieldDominance();
485 if (input->EnableVideoInput(video_mode_id, bmdFormat8BitYUV, supports_autodetect ? bmdVideoInputEnableFormatDetection : 0) != S_OK) {
486 fprintf(stderr, "Failed to set video mode 0x%04x for card %d\n", video_mode_id, card_index);
491 current_video_mode = video_mode_id;
494 void DeckLinkCapture::set_video_input(uint32_t video_input_id)
496 if (config->SetInt(bmdDeckLinkConfigVideoInputConnection, video_input_id) != S_OK) {
497 fprintf(stderr, "Failed to set video input connection for card %d\n", card_index);
501 current_video_input = video_input_id;
504 void DeckLinkCapture::set_audio_input(uint32_t audio_input_id)
506 if (config->SetInt(bmdDeckLinkConfigAudioInputConnection, audio_input_id) != S_OK) {
507 fprintf(stderr, "Failed to set audio input connection for card %d\n", card_index);
511 current_audio_input = audio_input_id;