1 #include "decklink_capture.h"
3 #include <DeckLinkAPI.h>
4 #include <DeckLinkAPIConfiguration.h>
5 #include <DeckLinkAPIDiscovery.h>
6 #include <DeckLinkAPIModes.h>
19 #include "bmusb/bmusb.h"
21 #define FRAME_SIZE (8 << 20) // 8 MB.
24 using namespace std::placeholders;
25 using namespace bmusb;
29 // TODO: Support stride.
30 void memcpy_interleaved(uint8_t *dest1, uint8_t *dest2, const uint8_t *src, size_t n)
33 uint8_t *dptr1 = dest1;
34 uint8_t *dptr2 = dest2;
36 for (size_t i = 0; i < n; i += 2) {
44 // Returns the number of bytes consumed.
45 size_t memcpy_interleaved_fastpath(uint8_t *dest1, uint8_t *dest2, const uint8_t *src, size_t n)
47 const uint8_t *limit = src + n;
50 // Align end to 32 bytes.
51 limit = (const uint8_t *)(intptr_t(limit) & ~31);
57 // Process [0,31] bytes, such that start gets aligned to 32 bytes.
58 const uint8_t *aligned_src = (const uint8_t *)(intptr_t(src + 31) & ~31);
59 if (aligned_src != src) {
60 size_t n2 = aligned_src - src;
61 memcpy_interleaved(dest1, dest2, src, n2);
71 // Make the length a multiple of 64.
72 if (((limit - src) % 64) != 0) {
75 assert(((limit - src) % 64) == 0);
78 const __m256i * __restrict in = (const __m256i *)src;
79 __m256i * __restrict out1 = (__m256i *)dest1;
80 __m256i * __restrict out2 = (__m256i *)dest2;
82 __m256i shuffle_cw = _mm256_set_epi8(
83 15, 13, 11, 9, 7, 5, 3, 1, 14, 12, 10, 8, 6, 4, 2, 0,
84 15, 13, 11, 9, 7, 5, 3, 1, 14, 12, 10, 8, 6, 4, 2, 0);
85 while (in < (const __m256i *)limit) {
86 // Note: For brevity, comments show lanes as if they were 2x64-bit (they're actually 2x128).
87 __m256i data1 = _mm256_stream_load_si256(in); // AaBbCcDd EeFfGgHh
88 __m256i data2 = _mm256_stream_load_si256(in + 1); // IiJjKkLl MmNnOoPp
90 data1 = _mm256_shuffle_epi8(data1, shuffle_cw); // ABCDabcd EFGHefgh
91 data2 = _mm256_shuffle_epi8(data2, shuffle_cw); // IJKLijkl MNOPmnop
93 data1 = _mm256_permute4x64_epi64(data1, 0b11011000); // ABCDEFGH abcdefgh
94 data2 = _mm256_permute4x64_epi64(data2, 0b11011000); // IJKLMNOP ijklmnop
96 __m256i lo = _mm256_permute2x128_si256(data1, data2, 0b00100000);
97 __m256i hi = _mm256_permute2x128_si256(data1, data2, 0b00110001);
99 _mm256_storeu_si256(out1, lo);
100 _mm256_storeu_si256(out2, hi);
108 const __m128i * __restrict in = (const __m128i *)src;
109 __m128i * __restrict out1 = (__m128i *)dest1;
110 __m128i * __restrict out2 = (__m128i *)dest2;
112 __m128i mask_lower_byte = _mm_set1_epi16(0x00ff);
113 while (in < (const __m128i *)limit) {
114 __m128i data1 = _mm_load_si128(in);
115 __m128i data2 = _mm_load_si128(in + 1);
116 __m128i data1_lo = _mm_and_si128(data1, mask_lower_byte);
117 __m128i data2_lo = _mm_and_si128(data2, mask_lower_byte);
118 __m128i data1_hi = _mm_srli_epi16(data1, 8);
119 __m128i data2_hi = _mm_srli_epi16(data2, 8);
120 __m128i lo = _mm_packus_epi16(data1_lo, data2_lo);
121 _mm_storeu_si128(out1, lo);
122 __m128i hi = _mm_packus_epi16(data1_hi, data2_hi);
123 _mm_storeu_si128(out2, hi);
139 DeckLinkCapture::DeckLinkCapture(IDeckLink *card, int card_index)
140 : card_index(card_index), card(card)
143 const char *model_name;
145 if (card->GetModelName(&model_name) == S_OK) {
146 snprintf(buf, sizeof(buf), "PCI card %d: %s", card_index, model_name);
148 snprintf(buf, sizeof(buf), "PCI card %d: Unknown DeckLink card", card_index);
153 if (card->QueryInterface(IID_IDeckLinkInput, (void**)&input) != S_OK) {
154 fprintf(stderr, "Card %d has no inputs\n", card_index);
158 IDeckLinkAttributes *attr;
159 if (card->QueryInterface(IID_IDeckLinkAttributes, (void**)&attr) != S_OK) {
160 fprintf(stderr, "Card %d has no attributes\n", card_index);
164 // Get the list of available video inputs.
165 int64_t video_input_mask;
166 if (attr->GetInt(BMDDeckLinkVideoInputConnections, &video_input_mask) != S_OK) {
167 fprintf(stderr, "Failed to enumerate video inputs for card %d\n", card_index);
170 const vector<pair<BMDVideoConnection, string>> video_input_types = {
171 { bmdVideoConnectionSDI, "SDI" },
172 { bmdVideoConnectionHDMI, "HDMI" },
173 { bmdVideoConnectionOpticalSDI, "Optical SDI" },
174 { bmdVideoConnectionComponent, "Component" },
175 { bmdVideoConnectionComposite, "Composite" },
176 { bmdVideoConnectionSVideo, "S-Video" }
178 for (const auto &video_input : video_input_types) {
179 if (video_input_mask & video_input.first) {
180 video_inputs.emplace(video_input.first, video_input.second);
184 // And then the available audio inputs.
185 int64_t audio_input_mask;
186 if (attr->GetInt(BMDDeckLinkAudioInputConnections, &audio_input_mask) != S_OK) {
187 fprintf(stderr, "Failed to enumerate audio inputs for card %d\n", card_index);
190 const vector<pair<BMDAudioConnection, string>> audio_input_types = {
191 { bmdAudioConnectionEmbedded, "Embedded" },
192 { bmdAudioConnectionAESEBU, "AES/EBU" },
193 { bmdAudioConnectionAnalog, "Analog" },
194 { bmdAudioConnectionAnalogXLR, "Analog XLR" },
195 { bmdAudioConnectionAnalogRCA, "Analog RCA" },
196 { bmdAudioConnectionMicrophone, "Microphone" },
197 { bmdAudioConnectionHeadphones, "Headphones" }
199 for (const auto &audio_input : audio_input_types) {
200 if (audio_input_mask & audio_input.first) {
201 audio_inputs.emplace(audio_input.first, audio_input.second);
207 /* Set up the video and audio sources. */
208 if (card->QueryInterface(IID_IDeckLinkConfiguration, (void**)&config) != S_OK) {
209 fprintf(stderr, "Failed to get configuration interface for card %d\n", card_index);
213 set_video_input(bmdVideoConnectionHDMI);
214 set_audio_input(bmdAudioConnectionEmbedded);
216 IDeckLinkDisplayModeIterator *mode_it;
217 if (input->GetDisplayModeIterator(&mode_it) != S_OK) {
218 fprintf(stderr, "Failed to enumerate display modes for card %d\n", card_index);
222 for (IDeckLinkDisplayMode *mode_ptr; mode_it->Next(&mode_ptr) == S_OK; mode_ptr->Release()) {
225 const char *mode_name;
226 if (mode_ptr->GetName(&mode_name) != S_OK) {
227 mode.name = "Unknown mode";
229 mode.name = mode_name;
232 mode.autodetect = false;
234 mode.width = mode_ptr->GetWidth();
235 mode.height = mode_ptr->GetHeight();
237 BMDTimeScale frame_rate_num;
238 BMDTimeValue frame_rate_den;
239 if (mode_ptr->GetFrameRate(&frame_rate_den, &frame_rate_num) != S_OK) {
240 fprintf(stderr, "Could not get frame rate for mode '%s' on card %d\n", mode.name.c_str(), card_index);
243 mode.frame_rate_num = frame_rate_num;
244 mode.frame_rate_den = frame_rate_den;
246 // TODO: Respect the TFF/BFF flag.
247 mode.interlaced = (mode_ptr->GetFieldDominance() == bmdLowerFieldFirst || mode_ptr->GetFieldDominance() == bmdUpperFieldFirst);
249 uint32_t id = mode_ptr->GetDisplayMode();
250 video_modes.insert(make_pair(id, mode));
253 set_video_mode_no_restart(bmdModeHD720p5994);
255 if (input->EnableAudioInput(48000, bmdAudioSampleType32bitInteger, 2) != S_OK) {
256 fprintf(stderr, "Failed to enable audio input for card %d\n", card_index);
260 input->SetCallback(this);
263 DeckLinkCapture::~DeckLinkCapture()
265 if (has_dequeue_callbacks) {
266 dequeue_cleanup_callback();
273 HRESULT STDMETHODCALLTYPE DeckLinkCapture::QueryInterface(REFIID, LPVOID *)
275 return E_NOINTERFACE;
278 ULONG STDMETHODCALLTYPE DeckLinkCapture::AddRef(void)
280 return refcount.fetch_add(1) + 1;
283 ULONG STDMETHODCALLTYPE DeckLinkCapture::Release(void)
285 int new_ref = refcount.fetch_sub(1) - 1;
291 HRESULT STDMETHODCALLTYPE DeckLinkCapture::VideoInputFormatChanged(
292 BMDVideoInputFormatChangedEvents,
293 IDeckLinkDisplayMode* display_mode,
294 BMDDetectedVideoInputFormatFlags)
296 if (display_mode->GetFrameRate(&frame_duration, &time_scale) != S_OK) {
297 fprintf(stderr, "Could not get new frame rate\n");
303 HRESULT STDMETHODCALLTYPE DeckLinkCapture::VideoInputFrameArrived(
304 IDeckLinkVideoInputFrame *video_frame,
305 IDeckLinkAudioInputPacket *audio_frame)
308 if (has_dequeue_callbacks) {
309 dequeue_init_callback();
314 FrameAllocator::Frame current_video_frame, current_audio_frame;
315 VideoFormat video_format;
316 AudioFormat audio_format;
318 video_format.frame_rate_nom = time_scale;
319 video_format.frame_rate_den = frame_duration;
322 video_format.has_signal = !(video_frame->GetFlags() & bmdFrameHasNoInputSource);
324 int width = video_frame->GetWidth();
325 int height = video_frame->GetHeight();
326 const int stride = video_frame->GetRowBytes();
327 assert(stride == width * 2);
329 current_video_frame = video_frame_allocator->alloc_frame();
330 if (current_video_frame.data != nullptr) {
331 const uint8_t *frame_bytes;
332 video_frame->GetBytes((void **)&frame_bytes);
334 uint8_t *data = current_video_frame.data;
335 uint8_t *data2 = current_video_frame.data2;
336 size_t num_bytes = width * height * 2;
338 size_t consumed = memcpy_interleaved_fastpath(data, data2, frame_bytes, num_bytes);
339 frame_bytes += consumed;
340 data += consumed / 2;
341 data2 += consumed / 2;
345 current_video_frame.len += consumed;
346 num_bytes -= consumed;
350 memcpy_interleaved(data, data2, frame_bytes, num_bytes);
352 current_video_frame.len += num_bytes;
354 video_format.width = width;
355 video_format.height = height;
360 int num_samples = audio_frame->GetSampleFrameCount();
362 current_audio_frame = audio_frame_allocator->alloc_frame();
363 if (current_audio_frame.data != nullptr) {
364 const uint8_t *frame_bytes;
365 audio_frame->GetBytes((void **)&frame_bytes);
366 current_audio_frame.len = sizeof(int32_t) * 2 * num_samples;
368 memcpy(current_audio_frame.data, frame_bytes, current_audio_frame.len);
370 audio_format.bits_per_sample = 32;
371 audio_format.num_channels = 2;
375 if (current_video_frame.data != nullptr || current_audio_frame.data != nullptr) {
376 // TODO: Put into a queue and put into a dequeue thread, if the
377 // BlackMagic drivers don't already do that for us?
378 frame_callback(timecode,
379 current_video_frame, /*video_offset=*/0, video_format,
380 current_audio_frame, /*audio_offset=*/0, audio_format);
387 void DeckLinkCapture::configure_card()
389 if (video_frame_allocator == nullptr) {
390 owned_video_frame_allocator.reset(new MallocFrameAllocator(FRAME_SIZE, NUM_QUEUED_VIDEO_FRAMES));
391 set_video_frame_allocator(owned_video_frame_allocator.get());
393 if (audio_frame_allocator == nullptr) {
394 owned_audio_frame_allocator.reset(new MallocFrameAllocator(65536, NUM_QUEUED_AUDIO_FRAMES));
395 set_audio_frame_allocator(owned_audio_frame_allocator.get());
399 void DeckLinkCapture::start_bm_capture()
401 if (input->StartStreams() != S_OK) {
402 fprintf(stderr, "StartStreams failed\n");
407 void DeckLinkCapture::stop_dequeue_thread()
409 if (input->StopStreams() != S_OK) {
410 fprintf(stderr, "StopStreams failed\n");
415 void DeckLinkCapture::set_video_mode(uint32_t video_mode_id)
417 if (input->StopStreams() != S_OK) {
418 fprintf(stderr, "StopStreams failed\n");
422 set_video_mode_no_restart(video_mode_id);
424 if (input->StartStreams() != S_OK) {
425 fprintf(stderr, "StartStreams failed\n");
430 void DeckLinkCapture::set_video_mode_no_restart(uint32_t video_mode_id)
432 BMDDisplayModeSupport support;
433 IDeckLinkDisplayMode *display_mode;
434 if (input->DoesSupportVideoMode(video_mode_id, bmdFormat8BitYUV, /*flags=*/0, &support, &display_mode)) {
435 fprintf(stderr, "Failed to query display mode for card %d\n", card_index);
439 if (support == bmdDisplayModeNotSupported) {
440 fprintf(stderr, "Card %d does not support display mode\n", card_index);
444 if (display_mode->GetFrameRate(&frame_duration, &time_scale) != S_OK) {
445 fprintf(stderr, "Could not get frame rate for card %d\n", card_index);
449 if (input->EnableVideoInput(video_mode_id, bmdFormat8BitYUV, 0) != S_OK) {
450 fprintf(stderr, "Failed to set video mode 0x%04x for card %d\n", video_mode_id, card_index);
454 current_video_mode = video_mode_id;
457 void DeckLinkCapture::set_video_input(uint32_t video_input_id)
459 if (config->SetInt(bmdDeckLinkConfigVideoInputConnection, video_input_id) != S_OK) {
460 fprintf(stderr, "Failed to set video input connection for card %d\n", card_index);
464 current_video_input = video_input_id;
467 void DeckLinkCapture::set_audio_input(uint32_t audio_input_id)
469 if (config->SetInt(bmdDeckLinkConfigAudioInputConnection, audio_input_id) != S_OK) {
470 fprintf(stderr, "Failed to set audio input connection for card %d\n", card_index);
474 current_audio_input = audio_input_id;