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 #include "v210_converter.h"
26 #define FRAME_SIZE (8 << 20) // 8 MB.
29 using namespace std::chrono;
30 using namespace std::placeholders;
31 using namespace bmusb;
35 // TODO: Support stride.
36 void memcpy_interleaved(uint8_t *dest1, uint8_t *dest2, const uint8_t *src, size_t n)
39 uint8_t *dptr1 = dest1;
40 uint8_t *dptr2 = dest2;
42 for (size_t i = 0; i < n; i += 2) {
50 // Returns the number of bytes consumed.
51 size_t memcpy_interleaved_fastpath(uint8_t *dest1, uint8_t *dest2, const uint8_t *src, size_t n)
53 const uint8_t *limit = src + n;
56 // Align end to 32 bytes.
57 limit = (const uint8_t *)(intptr_t(limit) & ~31);
63 // Process [0,31] bytes, such that start gets aligned to 32 bytes.
64 const uint8_t *aligned_src = (const uint8_t *)(intptr_t(src + 31) & ~31);
65 if (aligned_src != src) {
66 size_t n2 = aligned_src - src;
67 memcpy_interleaved(dest1, dest2, src, n2);
77 // Make the length a multiple of 64.
78 if (((limit - src) % 64) != 0) {
81 assert(((limit - src) % 64) == 0);
84 const __m256i * __restrict in = (const __m256i *)src;
85 __m256i * __restrict out1 = (__m256i *)dest1;
86 __m256i * __restrict out2 = (__m256i *)dest2;
88 __m256i shuffle_cw = _mm256_set_epi8(
89 15, 13, 11, 9, 7, 5, 3, 1, 14, 12, 10, 8, 6, 4, 2, 0,
90 15, 13, 11, 9, 7, 5, 3, 1, 14, 12, 10, 8, 6, 4, 2, 0);
91 while (in < (const __m256i *)limit) {
92 // Note: For brevity, comments show lanes as if they were 2x64-bit (they're actually 2x128).
93 __m256i data1 = _mm256_stream_load_si256(in); // AaBbCcDd EeFfGgHh
94 __m256i data2 = _mm256_stream_load_si256(in + 1); // IiJjKkLl MmNnOoPp
96 data1 = _mm256_shuffle_epi8(data1, shuffle_cw); // ABCDabcd EFGHefgh
97 data2 = _mm256_shuffle_epi8(data2, shuffle_cw); // IJKLijkl MNOPmnop
99 data1 = _mm256_permute4x64_epi64(data1, 0b11011000); // ABCDEFGH abcdefgh
100 data2 = _mm256_permute4x64_epi64(data2, 0b11011000); // IJKLMNOP ijklmnop
102 __m256i lo = _mm256_permute2x128_si256(data1, data2, 0b00100000);
103 __m256i hi = _mm256_permute2x128_si256(data1, data2, 0b00110001);
105 _mm256_storeu_si256(out1, lo);
106 _mm256_storeu_si256(out2, hi);
114 const __m128i * __restrict in = (const __m128i *)src;
115 __m128i * __restrict out1 = (__m128i *)dest1;
116 __m128i * __restrict out2 = (__m128i *)dest2;
118 __m128i mask_lower_byte = _mm_set1_epi16(0x00ff);
119 while (in < (const __m128i *)limit) {
120 __m128i data1 = _mm_load_si128(in);
121 __m128i data2 = _mm_load_si128(in + 1);
122 __m128i data1_lo = _mm_and_si128(data1, mask_lower_byte);
123 __m128i data2_lo = _mm_and_si128(data2, mask_lower_byte);
124 __m128i data1_hi = _mm_srli_epi16(data1, 8);
125 __m128i data2_hi = _mm_srli_epi16(data2, 8);
126 __m128i lo = _mm_packus_epi16(data1_lo, data2_lo);
127 _mm_storeu_si128(out1, lo);
128 __m128i hi = _mm_packus_epi16(data1_hi, data2_hi);
129 _mm_storeu_si128(out2, hi);
143 BMDPixelFormat pixel_format_to_bmd(PixelFormat pixel_format)
145 switch (pixel_format) {
146 case PixelFormat_8BitYCbCr:
147 return bmdFormat8BitYUV;
148 case PixelFormat_10BitYCbCr:
149 return bmdFormat10BitYUV;
157 DeckLinkCapture::DeckLinkCapture(IDeckLink *card, int card_index)
158 : card_index(card_index), card(card)
161 const char *model_name;
163 if (card->GetModelName(&model_name) == S_OK) {
164 snprintf(buf, sizeof(buf), "PCI card %d: %s", card_index, model_name);
166 snprintf(buf, sizeof(buf), "PCI card %d: Unknown DeckLink card", card_index);
171 if (card->QueryInterface(IID_IDeckLinkInput, (void**)&input) != S_OK) {
172 fprintf(stderr, "Card %d has no inputs\n", card_index);
176 IDeckLinkAttributes *attr;
177 if (card->QueryInterface(IID_IDeckLinkAttributes, (void**)&attr) != S_OK) {
178 fprintf(stderr, "Card %d has no attributes\n", card_index);
182 // Get the list of available video inputs.
183 int64_t video_input_mask;
184 if (attr->GetInt(BMDDeckLinkVideoInputConnections, &video_input_mask) != S_OK) {
185 fprintf(stderr, "Failed to enumerate video inputs for card %d\n", card_index);
188 const vector<pair<BMDVideoConnection, string>> video_input_types = {
189 { bmdVideoConnectionSDI, "SDI" },
190 { bmdVideoConnectionHDMI, "HDMI" },
191 { bmdVideoConnectionOpticalSDI, "Optical SDI" },
192 { bmdVideoConnectionComponent, "Component" },
193 { bmdVideoConnectionComposite, "Composite" },
194 { bmdVideoConnectionSVideo, "S-Video" }
196 for (const auto &video_input : video_input_types) {
197 if (video_input_mask & video_input.first) {
198 video_inputs.emplace(video_input.first, video_input.second);
202 // And then the available audio inputs.
203 int64_t audio_input_mask;
204 if (attr->GetInt(BMDDeckLinkAudioInputConnections, &audio_input_mask) != S_OK) {
205 fprintf(stderr, "Failed to enumerate audio inputs for card %d\n", card_index);
208 const vector<pair<BMDAudioConnection, string>> audio_input_types = {
209 { bmdAudioConnectionEmbedded, "Embedded" },
210 { bmdAudioConnectionAESEBU, "AES/EBU" },
211 { bmdAudioConnectionAnalog, "Analog" },
212 { bmdAudioConnectionAnalogXLR, "Analog XLR" },
213 { bmdAudioConnectionAnalogRCA, "Analog RCA" },
214 { bmdAudioConnectionMicrophone, "Microphone" },
215 { bmdAudioConnectionHeadphones, "Headphones" }
217 for (const auto &audio_input : audio_input_types) {
218 if (audio_input_mask & audio_input.first) {
219 audio_inputs.emplace(audio_input.first, audio_input.second);
223 // Check if we the card supports input autodetection.
224 if (attr->GetFlag(BMDDeckLinkSupportsInputFormatDetection, &supports_autodetect) != S_OK) {
225 fprintf(stderr, "Warning: Failed to ask card %d whether it supports input format autodetection\n", card_index);
226 supports_autodetect = false;
229 // If there's more than one subdevice on this card, label them.
230 int64_t num_subdevices, subdevice_idx;
231 if (attr->GetInt(BMDDeckLinkNumberOfSubDevices, &num_subdevices) == S_OK && num_subdevices > 1) {
232 if (attr->GetInt(BMDDeckLinkSubDeviceIndex, &subdevice_idx) == S_OK) {
234 snprintf(buf, sizeof(buf), " (subdevice %d)", int(subdevice_idx));
241 /* Set up the video and audio sources. */
242 if (card->QueryInterface(IID_IDeckLinkConfiguration, (void**)&config) != S_OK) {
243 fprintf(stderr, "Failed to get configuration interface for card %d\n", card_index);
247 BMDVideoConnection connection = pick_default_video_connection(card, BMDDeckLinkVideoInputConnections, card_index);
249 set_video_input(connection);
250 set_audio_input(bmdAudioConnectionEmbedded);
252 IDeckLinkDisplayModeIterator *mode_it;
253 if (input->GetDisplayModeIterator(&mode_it) != S_OK) {
254 fprintf(stderr, "Failed to enumerate display modes for card %d\n", card_index);
258 video_modes = summarize_video_modes(mode_it, card_index);
261 set_video_mode_no_restart(bmdModeHD720p5994);
263 input->SetCallback(this);
266 DeckLinkCapture::~DeckLinkCapture()
268 if (has_dequeue_callbacks) {
269 dequeue_cleanup_callback();
276 HRESULT STDMETHODCALLTYPE DeckLinkCapture::QueryInterface(REFIID, LPVOID *)
278 return E_NOINTERFACE;
281 ULONG STDMETHODCALLTYPE DeckLinkCapture::AddRef(void)
283 return refcount.fetch_add(1) + 1;
286 ULONG STDMETHODCALLTYPE DeckLinkCapture::Release(void)
288 int new_ref = refcount.fetch_sub(1) - 1;
294 HRESULT STDMETHODCALLTYPE DeckLinkCapture::VideoInputFormatChanged(
295 BMDVideoInputFormatChangedEvents,
296 IDeckLinkDisplayMode* display_mode,
297 BMDDetectedVideoInputFormatFlags format_flags)
299 if (format_flags & bmdDetectedVideoInputRGB444) {
300 fprintf(stderr, "WARNING: Input detected as 4:4:4 RGB, but Nageru can't consume that yet.\n");
301 fprintf(stderr, "Doing hardware conversion to 4:2:2 Y'CbCr.\n");
303 if (supports_autodetect && display_mode->GetDisplayMode() != current_video_mode) {
304 set_video_mode(display_mode->GetDisplayMode());
306 if (display_mode->GetFrameRate(&frame_duration, &time_scale) != S_OK) {
307 fprintf(stderr, "Could not get new frame rate\n");
310 field_dominance = display_mode->GetFieldDominance();
314 HRESULT STDMETHODCALLTYPE DeckLinkCapture::VideoInputFrameArrived(
315 IDeckLinkVideoInputFrame *video_frame,
316 IDeckLinkAudioInputPacket *audio_frame)
319 char thread_name[16];
320 snprintf(thread_name, sizeof(thread_name), "DeckLink_C_%d", card_index);
321 pthread_setname_np(pthread_self(), thread_name);
322 if (has_dequeue_callbacks) {
323 dequeue_init_callback();
328 steady_clock::time_point now = steady_clock::now();
330 FrameAllocator::Frame current_video_frame, current_audio_frame;
331 VideoFormat video_format;
332 AudioFormat audio_format;
334 video_format.frame_rate_nom = time_scale;
335 video_format.frame_rate_den = frame_duration;
336 // TODO: Respect the TFF/BFF flag.
337 video_format.interlaced = (field_dominance == bmdLowerFieldFirst || field_dominance == bmdUpperFieldFirst);
338 video_format.second_field_start = 1;
341 video_format.has_signal = !(video_frame->GetFlags() & bmdFrameHasNoInputSource);
343 int width = video_frame->GetWidth();
344 int height = video_frame->GetHeight();
345 const int stride = video_frame->GetRowBytes();
346 const BMDPixelFormat format = video_frame->GetPixelFormat();
347 assert(format == pixel_format_to_bmd(current_pixel_format));
348 if (global_flags.ten_bit_input) {
349 assert(stride == int(v210Converter::get_v210_stride(width)));
351 assert(stride == width * 2);
354 current_video_frame = video_frame_allocator->alloc_frame();
355 if (current_video_frame.data != nullptr) {
356 const uint8_t *frame_bytes;
357 video_frame->GetBytes((void **)&frame_bytes);
358 size_t num_bytes = stride * height;
360 if (current_video_frame.interleaved) {
361 uint8_t *data = current_video_frame.data;
362 uint8_t *data2 = current_video_frame.data2;
364 size_t consumed = memcpy_interleaved_fastpath(data, data2, frame_bytes, num_bytes);
365 frame_bytes += consumed;
366 data += consumed / 2;
367 data2 += consumed / 2;
371 current_video_frame.len += consumed;
372 num_bytes -= consumed;
376 memcpy_interleaved(data, data2, frame_bytes, num_bytes);
379 memcpy(current_video_frame.data, frame_bytes, num_bytes);
381 current_video_frame.len += num_bytes;
383 video_format.width = width;
384 video_format.height = height;
385 video_format.stride = stride;
390 int num_samples = audio_frame->GetSampleFrameCount();
392 current_audio_frame = audio_frame_allocator->alloc_frame();
393 if (current_audio_frame.data != nullptr) {
394 const uint8_t *frame_bytes;
395 audio_frame->GetBytes((void **)&frame_bytes);
396 current_audio_frame.len = sizeof(int32_t) * 2 * num_samples;
398 memcpy(current_audio_frame.data, frame_bytes, current_audio_frame.len);
400 audio_format.bits_per_sample = 32;
401 audio_format.num_channels = 2;
405 current_video_frame.received_timestamp = now;
406 current_audio_frame.received_timestamp = now;
408 if (current_video_frame.data != nullptr || current_audio_frame.data != nullptr) {
409 // TODO: Put into a queue and put into a dequeue thread, if the
410 // BlackMagic drivers don't already do that for us?
411 frame_callback(timecode,
412 current_video_frame, /*video_offset=*/0, video_format,
413 current_audio_frame, /*audio_offset=*/0, audio_format);
420 void DeckLinkCapture::configure_card()
422 if (video_frame_allocator == nullptr) {
423 owned_video_frame_allocator.reset(new MallocFrameAllocator(FRAME_SIZE, NUM_QUEUED_VIDEO_FRAMES));
424 set_video_frame_allocator(owned_video_frame_allocator.get());
426 if (audio_frame_allocator == nullptr) {
427 owned_audio_frame_allocator.reset(new MallocFrameAllocator(65536, NUM_QUEUED_AUDIO_FRAMES));
428 set_audio_frame_allocator(owned_audio_frame_allocator.get());
432 void DeckLinkCapture::start_bm_capture()
437 if (input->EnableVideoInput(current_video_mode, pixel_format_to_bmd(current_pixel_format), supports_autodetect ? bmdVideoInputEnableFormatDetection : 0) != S_OK) {
438 fprintf(stderr, "Failed to set video mode 0x%04x for card %d\n", current_video_mode, card_index);
441 if (input->EnableAudioInput(48000, bmdAudioSampleType32bitInteger, 2) != S_OK) {
442 fprintf(stderr, "Failed to enable audio input for card %d\n", card_index);
446 if (input->StartStreams() != S_OK) {
447 fprintf(stderr, "StartStreams failed\n");
453 void DeckLinkCapture::stop_dequeue_thread()
458 HRESULT result = input->StopStreams();
459 if (result != S_OK) {
460 fprintf(stderr, "StopStreams failed with error 0x%x\n", result);
463 if (input->DisableVideoInput() != S_OK) {
464 fprintf(stderr, "Failed to disable video input for card %d\n", card_index);
467 if (input->DisableAudioInput() != S_OK) {
468 fprintf(stderr, "Failed to disable audio input for card %d\n", card_index);
474 void DeckLinkCapture::set_video_mode(uint32_t video_mode_id)
477 if (input->PauseStreams() != S_OK) {
478 fprintf(stderr, "PauseStreams failed\n");
481 if (input->FlushStreams() != S_OK) {
482 fprintf(stderr, "FlushStreams failed\n");
487 set_video_mode_no_restart(video_mode_id);
490 if (input->StartStreams() != S_OK) {
491 fprintf(stderr, "StartStreams failed\n");
497 void DeckLinkCapture::set_pixel_format(PixelFormat pixel_format)
499 current_pixel_format = pixel_format;
500 set_video_mode(current_video_mode);
503 void DeckLinkCapture::set_video_mode_no_restart(uint32_t video_mode_id)
505 BMDDisplayModeSupport support;
506 IDeckLinkDisplayMode *display_mode;
507 if (input->DoesSupportVideoMode(video_mode_id, pixel_format_to_bmd(current_pixel_format), /*flags=*/0, &support, &display_mode)) {
508 fprintf(stderr, "Failed to query display mode for card %d\n", card_index);
512 if (support == bmdDisplayModeNotSupported) {
513 fprintf(stderr, "Card %d does not support display mode\n", card_index);
517 if (display_mode->GetFrameRate(&frame_duration, &time_scale) != S_OK) {
518 fprintf(stderr, "Could not get frame rate for card %d\n", card_index);
522 field_dominance = display_mode->GetFieldDominance();
525 if (input->EnableVideoInput(video_mode_id, pixel_format_to_bmd(current_pixel_format), supports_autodetect ? bmdVideoInputEnableFormatDetection : 0) != S_OK) {
526 fprintf(stderr, "Failed to set video mode 0x%04x for card %d\n", video_mode_id, card_index);
531 current_video_mode = video_mode_id;
534 void DeckLinkCapture::set_video_input(uint32_t video_input_id)
536 if (config->SetInt(bmdDeckLinkConfigVideoInputConnection, video_input_id) != S_OK) {
537 fprintf(stderr, "Failed to set video input connection for card %d\n", card_index);
541 current_video_input = video_input_id;
544 void DeckLinkCapture::set_audio_input(uint32_t audio_input_id)
546 if (config->SetInt(bmdDeckLinkConfigAudioInputConnection, audio_input_id) != S_OK) {
547 fprintf(stderr, "Failed to set audio input connection for card %d\n", card_index);
551 current_audio_input = audio_input_id;