X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=decklink_capture.cpp;h=81566cb1a5ad20d85a3be25a8b01e2c24a62906e;hb=6379b0adf2cee84b46d5a0d8e2ee5e6a7d1615cf;hp=e7f8eba0a16531c7341619be0cb2e8f7541b583a;hpb=3e8c584c42ae8a53e2cf335ff6c986f1b87f3f5e;p=nageru diff --git a/decklink_capture.cpp b/decklink_capture.cpp index e7f8eba..81566cb 100644 --- a/decklink_capture.cpp +++ b/decklink_capture.cpp @@ -4,7 +4,11 @@ #include #include #include +#include #include +#ifdef __SSE2__ +#include +#endif #include #include @@ -12,13 +16,15 @@ #include #include "bmusb/bmusb.h" +#define FRAME_SIZE (8 << 20) // 8 MB. + using namespace std; using namespace std::placeholders; +using namespace bmusb; namespace { // TODO: Support stride. -// TODO: Support AVX2 (adapt from bmusb). void memcpy_interleaved(uint8_t *dest1, uint8_t *dest2, const uint8_t *src, size_t n) { assert(n % 2 == 0); @@ -31,17 +37,113 @@ void memcpy_interleaved(uint8_t *dest1, uint8_t *dest2, const uint8_t *src, size } } +#ifdef __SSE2__ + +// Returns the number of bytes consumed. +size_t memcpy_interleaved_fastpath(uint8_t *dest1, uint8_t *dest2, const uint8_t *src, size_t n) +{ + const uint8_t *limit = src + n; + size_t consumed = 0; + + // Align end to 32 bytes. + limit = (const uint8_t *)(intptr_t(limit) & ~31); + + if (src >= limit) { + return 0; + } + + // Process [0,31] bytes, such that start gets aligned to 32 bytes. + const uint8_t *aligned_src = (const uint8_t *)(intptr_t(src + 31) & ~31); + if (aligned_src != src) { + size_t n2 = aligned_src - src; + memcpy_interleaved(dest1, dest2, src, n2); + dest1 += n2 / 2; + dest2 += n2 / 2; + if (n2 % 2) { + swap(dest1, dest2); + } + src = aligned_src; + consumed += n2; + } + + // Make the length a multiple of 64. + if (((limit - src) % 64) != 0) { + limit -= 32; + } + assert(((limit - src) % 64) == 0); + +#if __AVX2__ + const __m256i * __restrict in = (const __m256i *)src; + __m256i * __restrict out1 = (__m256i *)dest1; + __m256i * __restrict out2 = (__m256i *)dest2; + + __m256i shuffle_cw = _mm256_set_epi8( + 15, 13, 11, 9, 7, 5, 3, 1, 14, 12, 10, 8, 6, 4, 2, 0, + 15, 13, 11, 9, 7, 5, 3, 1, 14, 12, 10, 8, 6, 4, 2, 0); + while (in < (const __m256i *)limit) { + // Note: For brevity, comments show lanes as if they were 2x64-bit (they're actually 2x128). + __m256i data1 = _mm256_stream_load_si256(in); // AaBbCcDd EeFfGgHh + __m256i data2 = _mm256_stream_load_si256(in + 1); // IiJjKkLl MmNnOoPp + + data1 = _mm256_shuffle_epi8(data1, shuffle_cw); // ABCDabcd EFGHefgh + data2 = _mm256_shuffle_epi8(data2, shuffle_cw); // IJKLijkl MNOPmnop + + data1 = _mm256_permute4x64_epi64(data1, 0b11011000); // ABCDEFGH abcdefgh + data2 = _mm256_permute4x64_epi64(data2, 0b11011000); // IJKLMNOP ijklmnop + + __m256i lo = _mm256_permute2x128_si256(data1, data2, 0b00100000); + __m256i hi = _mm256_permute2x128_si256(data1, data2, 0b00110001); + + _mm256_storeu_si256(out1, lo); + _mm256_storeu_si256(out2, hi); + + in += 2; + ++out1; + ++out2; + consumed += 64; + } +#else + const __m128i * __restrict in = (const __m128i *)src; + __m128i * __restrict out1 = (__m128i *)dest1; + __m128i * __restrict out2 = (__m128i *)dest2; + + __m128i mask_lower_byte = _mm_set1_epi16(0x00ff); + while (in < (const __m128i *)limit) { + __m128i data1 = _mm_load_si128(in); + __m128i data2 = _mm_load_si128(in + 1); + __m128i data1_lo = _mm_and_si128(data1, mask_lower_byte); + __m128i data2_lo = _mm_and_si128(data2, mask_lower_byte); + __m128i data1_hi = _mm_srli_epi16(data1, 8); + __m128i data2_hi = _mm_srli_epi16(data2, 8); + __m128i lo = _mm_packus_epi16(data1_lo, data2_lo); + _mm_storeu_si128(out1, lo); + __m128i hi = _mm_packus_epi16(data1_hi, data2_hi); + _mm_storeu_si128(out2, hi); + + in += 2; + ++out1; + ++out2; + consumed += 32; + } +#endif + + return consumed; +} + +#endif // __SSE2__ + } // namespace DeckLinkCapture::DeckLinkCapture(IDeckLink *card, int card_index) + : card_index(card_index), card(card) { { const char *model_name; char buf[256]; if (card->GetModelName(&model_name) == S_OK) { - snprintf(buf, sizeof(buf), "Card %d: %s", card_index, model_name); + snprintf(buf, sizeof(buf), "PCI card %d: %s", card_index, model_name); } else { - snprintf(buf, sizeof(buf), "Card %d: Unknown DeckLink card", card_index); + snprintf(buf, sizeof(buf), "PCI card %d: Unknown DeckLink card", card_index); } description = buf; } @@ -51,47 +153,104 @@ DeckLinkCapture::DeckLinkCapture(IDeckLink *card, int card_index) exit(1); } - /* Set up the video and audio sources. */ - IDeckLinkConfiguration *config; - if (card->QueryInterface(IID_IDeckLinkConfiguration, (void**)&config) != S_OK) { - fprintf(stderr, "Failed to get configuration interface for card %d\n", card_index); + IDeckLinkAttributes *attr; + if (card->QueryInterface(IID_IDeckLinkAttributes, (void**)&attr) != S_OK) { + fprintf(stderr, "Card %d has no attributes\n", card_index); exit(1); } - if (config->SetInt(bmdDeckLinkConfigVideoInputConnection, bmdVideoConnectionHDMI) != S_OK) { - fprintf(stderr, "Failed to set video input connection for card %d\n", card_index); + // Get the list of available video inputs. + int64_t video_input_mask; + if (attr->GetInt(BMDDeckLinkVideoInputConnections, &video_input_mask) != S_OK) { + fprintf(stderr, "Failed to enumerate video inputs for card %d\n", card_index); exit(1); } - - if (config->SetInt(bmdDeckLinkConfigAudioInputConnection, bmdAudioConnectionEmbedded) != S_OK) { - fprintf(stderr, "Failed to set video input connection for card %d\n", card_index); - exit(1); + const vector> video_input_types = { + { bmdVideoConnectionSDI, "SDI" }, + { bmdVideoConnectionHDMI, "HDMI" }, + { bmdVideoConnectionOpticalSDI, "Optical SDI" }, + { bmdVideoConnectionComponent, "Component" }, + { bmdVideoConnectionComposite, "Composite" }, + { bmdVideoConnectionSVideo, "S-Video" } + }; + for (const auto &video_input : video_input_types) { + if (video_input_mask & video_input.first) { + video_inputs.emplace(video_input.first, video_input.second); + } } - // TODO: Make the user mode selectable. - BMDDisplayModeSupport support; - IDeckLinkDisplayMode *display_mode; - if (input->DoesSupportVideoMode(bmdModeHD720p5994, bmdFormat8BitYUV, /*flags=*/0, &support, &display_mode)) { - fprintf(stderr, "Failed to query display mode for card %d\n", card_index); + // And then the available audio inputs. + int64_t audio_input_mask; + if (attr->GetInt(BMDDeckLinkAudioInputConnections, &audio_input_mask) != S_OK) { + fprintf(stderr, "Failed to enumerate audio inputs for card %d\n", card_index); exit(1); } + const vector> audio_input_types = { + { bmdAudioConnectionEmbedded, "Embedded" }, + { bmdAudioConnectionAESEBU, "AES/EBU" }, + { bmdAudioConnectionAnalog, "Analog" }, + { bmdAudioConnectionAnalogXLR, "Analog XLR" }, + { bmdAudioConnectionAnalogRCA, "Analog RCA" }, + { bmdAudioConnectionMicrophone, "Microphone" }, + { bmdAudioConnectionHeadphones, "Headphones" } + }; + for (const auto &audio_input : audio_input_types) { + if (audio_input_mask & audio_input.first) { + audio_inputs.emplace(audio_input.first, audio_input.second); + } + } - if (support == bmdDisplayModeNotSupported) { - fprintf(stderr, "Card %d does not support display mode\n", card_index); + attr->Release(); + + /* Set up the video and audio sources. */ + if (card->QueryInterface(IID_IDeckLinkConfiguration, (void**)&config) != S_OK) { + fprintf(stderr, "Failed to get configuration interface for card %d\n", card_index); exit(1); } - if (display_mode->GetFrameRate(&frame_duration, &time_scale) != S_OK) { - fprintf(stderr, "Could not get frame rate for card %d\n", card_index); + set_video_input(bmdVideoConnectionHDMI); + set_audio_input(bmdAudioConnectionEmbedded); + + IDeckLinkDisplayModeIterator *mode_it; + if (input->GetDisplayModeIterator(&mode_it) != S_OK) { + fprintf(stderr, "Failed to enumerate display modes for card %d\n", card_index); exit(1); } - if (input->EnableVideoInput(bmdModeHD720p5994, bmdFormat8BitYUV, 0) != S_OK) { - fprintf(stderr, "Failed to set 720p59.94 connection for card %d\n", card_index); - exit(1); + for (IDeckLinkDisplayMode *mode_ptr; mode_it->Next(&mode_ptr) == S_OK; mode_ptr->Release()) { + VideoMode mode; + + const char *mode_name; + if (mode_ptr->GetName(&mode_name) != S_OK) { + mode.name = "Unknown mode"; + } else { + mode.name = mode_name; + } + + mode.autodetect = false; + + mode.width = mode_ptr->GetWidth(); + mode.height = mode_ptr->GetHeight(); + + BMDTimeScale frame_rate_num; + BMDTimeValue frame_rate_den; + if (mode_ptr->GetFrameRate(&frame_rate_den, &frame_rate_num) != S_OK) { + fprintf(stderr, "Could not get frame rate for mode '%s' on card %d\n", mode.name.c_str(), card_index); + exit(1); + } + mode.frame_rate_num = frame_rate_num; + mode.frame_rate_den = frame_rate_den; + + // TODO: Respect the TFF/BFF flag. + mode.interlaced = (mode_ptr->GetFieldDominance() == bmdLowerFieldFirst || mode_ptr->GetFieldDominance() == bmdUpperFieldFirst); + + uint32_t id = mode_ptr->GetDisplayMode(); + video_modes.insert(make_pair(id, mode)); } - if (input->EnableAudioInput(48000, bmdAudioSampleType16bitInteger, 2) != S_OK) { + set_video_mode_no_restart(bmdModeHD720p5994); + + if (input->EnableAudioInput(48000, bmdAudioSampleType32bitInteger, 2) != S_OK) { fprintf(stderr, "Failed to enable audio input for card %d\n", card_index); exit(1); } @@ -104,6 +263,9 @@ DeckLinkCapture::~DeckLinkCapture() if (has_dequeue_callbacks) { dequeue_cleanup_callback(); } + input->Release(); + config->Release(); + card->Release(); } HRESULT STDMETHODCALLTYPE DeckLinkCapture::QueryInterface(REFIID, LPVOID *) @@ -149,6 +311,10 @@ HRESULT STDMETHODCALLTYPE DeckLinkCapture::VideoInputFrameArrived( FrameAllocator::Frame current_video_frame, current_audio_frame; VideoFormat video_format; + AudioFormat audio_format; + + video_format.frame_rate_nom = time_scale; + video_format.frame_rate_den = frame_duration; if (video_frame) { video_format.has_signal = !(video_frame->GetFlags() & bmdFrameHasNoInputSource); @@ -163,14 +329,44 @@ HRESULT STDMETHODCALLTYPE DeckLinkCapture::VideoInputFrameArrived( const uint8_t *frame_bytes; video_frame->GetBytes((void **)&frame_bytes); - memcpy_interleaved(current_video_frame.data, current_video_frame.data2, - frame_bytes, width * height * 2); - current_video_frame.len += width * height * 2; + uint8_t *data = current_video_frame.data; + uint8_t *data2 = current_video_frame.data2; + size_t num_bytes = width * height * 2; +#ifdef __SSE2__ + size_t consumed = memcpy_interleaved_fastpath(data, data2, frame_bytes, num_bytes); + frame_bytes += consumed; + data += consumed / 2; + data2 += consumed / 2; + if (num_bytes % 2) { + swap(data, data2); + } + current_video_frame.len += consumed; + num_bytes -= consumed; +#endif + + if (num_bytes > 0) { + memcpy_interleaved(data, data2, frame_bytes, num_bytes); + } + current_video_frame.len += num_bytes; video_format.width = width; video_format.height = height; - video_format.frame_rate_nom = time_scale; - video_format.frame_rate_den = frame_duration; + } + } + + if (audio_frame) { + int num_samples = audio_frame->GetSampleFrameCount(); + + current_audio_frame = audio_frame_allocator->alloc_frame(); + if (current_audio_frame.data != nullptr) { + const uint8_t *frame_bytes; + audio_frame->GetBytes((void **)&frame_bytes); + current_audio_frame.len = sizeof(int32_t) * 2 * num_samples; + + memcpy(current_audio_frame.data, frame_bytes, current_audio_frame.len); + + audio_format.bits_per_sample = 32; + audio_format.num_channels = 2; } } @@ -179,13 +375,25 @@ HRESULT STDMETHODCALLTYPE DeckLinkCapture::VideoInputFrameArrived( // BlackMagic drivers don't already do that for us? frame_callback(timecode, current_video_frame, /*video_offset=*/0, video_format, - current_audio_frame, /*audio_offset=*/0, 0x0000); + current_audio_frame, /*audio_offset=*/0, audio_format); } timecode++; return S_OK; } +void DeckLinkCapture::configure_card() +{ + if (video_frame_allocator == nullptr) { + owned_video_frame_allocator.reset(new MallocFrameAllocator(FRAME_SIZE, NUM_QUEUED_VIDEO_FRAMES)); + set_video_frame_allocator(owned_video_frame_allocator.get()); + } + if (audio_frame_allocator == nullptr) { + owned_audio_frame_allocator.reset(new MallocFrameAllocator(65536, NUM_QUEUED_AUDIO_FRAMES)); + set_audio_frame_allocator(owned_audio_frame_allocator.get()); + } +} + void DeckLinkCapture::start_bm_capture() { if (input->StartStreams() != S_OK) { @@ -202,3 +410,64 @@ void DeckLinkCapture::stop_dequeue_thread() } } +void DeckLinkCapture::set_video_mode(uint32_t video_mode_id) +{ + if (input->StopStreams() != S_OK) { + fprintf(stderr, "StopStreams failed\n"); + exit(1); + } + + set_video_mode_no_restart(video_mode_id); + + if (input->StartStreams() != S_OK) { + fprintf(stderr, "StartStreams failed\n"); + exit(1); + } +} + +void DeckLinkCapture::set_video_mode_no_restart(uint32_t video_mode_id) +{ + BMDDisplayModeSupport support; + IDeckLinkDisplayMode *display_mode; + if (input->DoesSupportVideoMode(video_mode_id, bmdFormat8BitYUV, /*flags=*/0, &support, &display_mode)) { + fprintf(stderr, "Failed to query display mode for card %d\n", card_index); + exit(1); + } + + if (support == bmdDisplayModeNotSupported) { + fprintf(stderr, "Card %d does not support display mode\n", card_index); + exit(1); + } + + if (display_mode->GetFrameRate(&frame_duration, &time_scale) != S_OK) { + fprintf(stderr, "Could not get frame rate for card %d\n", card_index); + exit(1); + } + + if (input->EnableVideoInput(video_mode_id, bmdFormat8BitYUV, 0) != S_OK) { + fprintf(stderr, "Failed to set video mode 0x%04x for card %d\n", video_mode_id, card_index); + exit(1); + } + + current_video_mode = video_mode_id; +} + +void DeckLinkCapture::set_video_input(uint32_t video_input_id) +{ + if (config->SetInt(bmdDeckLinkConfigVideoInputConnection, video_input_id) != S_OK) { + fprintf(stderr, "Failed to set video input connection for card %d\n", card_index); + exit(1); + } + + current_video_input = video_input_id; +} + +void DeckLinkCapture::set_audio_input(uint32_t audio_input_id) +{ + if (config->SetInt(bmdDeckLinkConfigAudioInputConnection, audio_input_id) != S_OK) { + fprintf(stderr, "Failed to set audio input connection for card %d\n", card_index); + exit(1); + } + + current_audio_input = audio_input_id; +}