X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=bmusb.cpp;h=f8ab2e61af3f8af8e66830db911199f5041338f5;hb=e614cf402e0a7f754a0b88a68b11c766ab9baebe;hp=3e31c01a771a1cbdd62471ee1f86fbd81b32338c;hpb=862d8ccf3a3d48602427eb2bfb319b2fc7181fe0;p=bmusb diff --git a/bmusb.cpp b/bmusb.cpp index 3e31c01..f8ab2e6 100644 --- a/bmusb.cpp +++ b/bmusb.cpp @@ -4,16 +4,21 @@ // 576p60/720p60/1080i60 works, 1080p60 does not work (firmware limitation) // Audio comes out as 8-channel 24-bit raw audio. +#if (defined(__i386__) || defined(__x86_64__)) && defined(__GNUC__) +#define HAS_MULTIVERSIONING 1 +#endif + #include #include #include +#include #include #include #include #include #include #include -#ifdef __SSE4_1__ +#if HAS_MULTIVERSIONING #include #endif #include "bmusb.h" @@ -34,6 +39,7 @@ using namespace std; using namespace std::placeholders; +#define USB_VENDOR_BLACKMAGIC 0x1edb #define MIN_WIDTH 640 #define HEADER_SIZE 44 //#define HEADER_SIZE 0 @@ -42,6 +48,8 @@ using namespace std::placeholders; #define FRAME_SIZE (8 << 20) // 8 MB. #define USB_VIDEO_TRANSFER_SIZE (128 << 10) // 128 kB. +card_connected_callback_t BMUSBCapture::card_connected_callback = nullptr; + namespace { FILE *audiofp; @@ -356,8 +364,6 @@ void add_to_frame(FrameAllocator::Frame *current_frame, const char *frame_type_n } } -#ifdef __SSE4_1__ - #if 0 void avx2_dump(const char *name, __m256i n) { @@ -401,6 +407,18 @@ void avx2_dump(const char *name, __m256i n) } #endif +#ifndef HAS_MULTIVERSIONING + +const uint8_t *add_to_frame_fastpath(FrameAllocator::Frame *current_frame, const uint8_t *start, const uint8_t *limit, const char sync_char) +{ + // No fast path possible unless we have multiversioning. + return start; +} + +#else // defined(HAS_MULTIVERSIONING) + +const uint8_t *add_to_frame_fastpath_core(FrameAllocator::Frame *current_frame, const uint8_t *aligned_start, const uint8_t *limit, const char sync_char); + // Does a memcpy and memchr in one to reduce processing time. // Note that the benefit is somewhat limited if your L3 cache is small, // as you'll (unfortunately) spend most of the time loading the data @@ -410,6 +428,14 @@ void avx2_dump(const char *name, __m256i n) // up until the first instance of "sync_char" (usually a bit before, actually). // This is fine, since 0x00 bytes shouldn't really show up in normal picture // data, and what we really need this for is the 00 00 ff ff marker in video data. +__attribute__((target("default"))) +const uint8_t *add_to_frame_fastpath(FrameAllocator::Frame *current_frame, const uint8_t *start, const uint8_t *limit, const char sync_char) +{ + // No fast path possible unless we have SSE 4.1 or higher. + return start; +} + +__attribute__((target("sse4.1", "avx2"))) const uint8_t *add_to_frame_fastpath(FrameAllocator::Frame *current_frame, const uint8_t *start, const uint8_t *limit, const char sync_char) { if (current_frame->data == nullptr || @@ -453,7 +479,12 @@ const uint8_t *add_to_frame_fastpath(FrameAllocator::Frame *current_frame, const assert(((limit - aligned_start) % 64) == 0); } -#if __AVX2__ + return add_to_frame_fastpath_core(current_frame, aligned_start, limit, sync_char); +} + +__attribute__((target("avx2"))) +const uint8_t *add_to_frame_fastpath_core(FrameAllocator::Frame *current_frame, const uint8_t *aligned_start, const uint8_t *limit, const char sync_char) +{ const __m256i needle = _mm256_set1_epi8(sync_char); const __restrict __m256i *in = (const __m256i *)aligned_start; @@ -512,7 +543,14 @@ const uint8_t *add_to_frame_fastpath(FrameAllocator::Frame *current_frame, const } current_frame->len = (uint8_t *)out - current_frame->data; } -#else + + //printf("managed to fastpath %ld/%ld bytes\n", (const uint8_t *)in - (const uint8_t *)aligned_start, orig_bytes); + return (const uint8_t *)in; +} + +__attribute__((target("sse4.1"))) +const uint8_t *add_to_frame_fastpath_core(FrameAllocator::Frame *current_frame, const uint8_t *aligned_start, const uint8_t *limit, const char sync_char) +{ const __m128i needle = _mm_set1_epi8(sync_char); const __m128i *in = (const __m128i *)aligned_start; @@ -562,13 +600,12 @@ const uint8_t *add_to_frame_fastpath(FrameAllocator::Frame *current_frame, const } current_frame->len = (uint8_t *)out - current_frame->data; } -#endif //printf("managed to fastpath %ld/%ld bytes\n", (const uint8_t *)in - (const uint8_t *)aligned_start, orig_bytes); - return (const uint8_t *)in; } -#endif + +#endif // defined(HAS_MULTIVERSIONING) void decode_packs(const libusb_transfer *xfr, const char *sync_pattern, @@ -590,11 +627,9 @@ void decode_packs(const libusb_transfer *xfr, const uint8_t *start = xfr->buffer + offset; const uint8_t *limit = start + pack->actual_length; while (start < limit) { // Usually runs only one iteration. -#ifdef __SSE4_1__ start = add_to_frame_fastpath(current_frame, start, limit, sync_pattern[0]); if (start == limit) break; assert(start < limit); -#endif const unsigned char* start_next_frame = (const unsigned char *)memmem(start, limit - start, sync_pattern, sync_length); if (start_next_frame == nullptr) { @@ -616,8 +651,9 @@ void decode_packs(const libusb_transfer *xfr, void BMUSBCapture::cb_xfr(struct libusb_transfer *xfr) { - if (xfr->status != LIBUSB_TRANSFER_COMPLETED) { - fprintf(stderr, "transfer status %d\n", xfr->status); + if (xfr->status != LIBUSB_TRANSFER_COMPLETED && + xfr->status != LIBUSB_TRANSFER_NO_DEVICE) { + fprintf(stderr, "error: transfer status %d\n", xfr->status); libusb_free_transfer(xfr); exit(3); } @@ -625,6 +661,18 @@ void BMUSBCapture::cb_xfr(struct libusb_transfer *xfr) assert(xfr->user_data != nullptr); BMUSBCapture *usb = static_cast(xfr->user_data); + if (xfr->status == LIBUSB_TRANSFER_NO_DEVICE) { + if (!usb->disconnected) { + fprintf(stderr, "Device went away, stopping transfers.\n"); + usb->disconnected = true; + if (usb->card_disconnected_callback) { + usb->card_disconnected_callback(); + } + } + // Don't reschedule the transfer; the loop will stop by itself. + return; + } + if (xfr->type == LIBUSB_TRANSFER_TYPE_ISOCHRONOUS) { if (xfr->endpoint == 0x84) { decode_packs(xfr, "DeckLinkAudioResyncT", 20, &usb->current_audio_frame, "audio", bind(&BMUSBCapture::start_new_audio_block, usb, _1)); @@ -682,6 +730,26 @@ void BMUSBCapture::cb_xfr(struct libusb_transfer *xfr) } } +int BMUSBCapture::cb_hotplug(libusb_context *ctx, libusb_device *dev, libusb_hotplug_event event, void *user_data) +{ + if (card_connected_callback != nullptr) { + libusb_device_descriptor desc; + if (libusb_get_device_descriptor(dev, &desc) < 0) { + fprintf(stderr, "Error getting device descriptor for hotplugged device %p, killing hotplug\n", dev); + libusb_unref_device(dev); + return 1; + } + + if ((desc.idVendor == USB_VENDOR_BLACKMAGIC && desc.idProduct == 0xbd3b) || + (desc.idVendor == USB_VENDOR_BLACKMAGIC && desc.idProduct == 0xbd4f)) { + card_connected_callback(dev); // Callback takes ownership. + return 0; + } + } + libusb_unref_device(dev); + return 0; +} + void BMUSBCapture::usb_thread_func() { sched_param param; @@ -703,8 +771,30 @@ struct USBCardDevice { libusb_device *device; }; +const char *get_product_name(uint16_t product) +{ + if (product == 0xbd3b) { + return "Intensity Shuttle"; + } else if (product == 0xbd4f) { + return "UltraStudio SDI"; + } else { + assert(false); + return nullptr; + } +} + +string get_card_description(int id, uint8_t bus, uint8_t port, uint16_t product) +{ + const char *product_name = get_product_name(product); + + char buf[256]; + snprintf(buf, sizeof(buf), "USB card %d: Bus %03u Device %03u %s", + id, bus, port, product_name); + return buf; +} + libusb_device_handle *open_card(int card_index, string *description) -{ +{ libusb_device **devices; ssize_t num_devices = libusb_get_device_list(nullptr, &devices); if (num_devices == -1) { @@ -722,8 +812,8 @@ libusb_device_handle *open_card(int card_index, string *description) uint8_t bus = libusb_get_bus_number(devices[i]); uint8_t port = libusb_get_port_number(devices[i]); - if (!(desc.idVendor == 0x1edb && desc.idProduct == 0xbd3b) && - !(desc.idVendor == 0x1edb && desc.idProduct == 0xbd4f)) { + if (!(desc.idVendor == USB_VENDOR_BLACKMAGIC && desc.idProduct == 0xbd3b) && + !(desc.idVendor == USB_VENDOR_BLACKMAGIC && desc.idProduct == 0xbd4f)) { libusb_unref_device(devices[i]); continue; } @@ -742,22 +832,11 @@ libusb_device_handle *open_card(int card_index, string *description) }); for (size_t i = 0; i < found_cards.size(); ++i) { - const char *product_name = nullptr; - if (found_cards[i].product == 0xbd3b) { - product_name = "Intensity Shuttle"; - } else if (found_cards[i].product == 0xbd4f) { - product_name = "UltraStudio SDI"; - } else { - assert(false); - } - - char buf[256]; - snprintf(buf, sizeof(buf), "Card %d: Bus %03u Device %03u %s", - int(i), found_cards[i].bus, found_cards[i].port, product_name); + string tmp_description = get_card_description(i, found_cards[i].bus, found_cards[i].port, found_cards[i].product); + fprintf(stderr, "%s\n", tmp_description.c_str()); if (i == size_t(card_index)) { - *description = buf; + *description = tmp_description; } - fprintf(stderr, "%s\n", buf); } if (size_t(card_index) >= found_cards.size()) { @@ -779,13 +858,38 @@ libusb_device_handle *open_card(int card_index, string *description) return devh; } +libusb_device_handle *open_card(unsigned card_index, libusb_device *dev, string *description) +{ + uint8_t bus = libusb_get_bus_number(dev); + uint8_t port = libusb_get_port_number(dev); + + libusb_device_descriptor desc; + if (libusb_get_device_descriptor(dev, &desc) < 0) { + fprintf(stderr, "Error getting device descriptor for device %p\n", dev); + exit(1); + } + + *description = get_card_description(card_index, bus, port, desc.idProduct); + + libusb_device_handle *devh; + int rc = libusb_open(dev, &devh); + if (rc < 0) { + fprintf(stderr, "Error opening card %p: %s\n", dev, libusb_error_name(rc)); + exit(1); + } + + return devh; +} + void BMUSBCapture::configure_card() { if (video_frame_allocator == nullptr) { - set_video_frame_allocator(new MallocFrameAllocator(FRAME_SIZE, NUM_QUEUED_VIDEO_FRAMES)); // FIXME: leak. + 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) { - set_audio_frame_allocator(new MallocFrameAllocator(65536, NUM_QUEUED_AUDIO_FRAMES)); // FIXME: leak. + owned_audio_frame_allocator.reset(new MallocFrameAllocator(65536, NUM_QUEUED_AUDIO_FRAMES)); + set_audio_frame_allocator(owned_audio_frame_allocator.get()); } dequeue_thread_should_quit = false; dequeue_thread = thread(&BMUSBCapture::dequeue_thread_func, this); @@ -799,7 +903,12 @@ void BMUSBCapture::configure_card() exit(1); } - libusb_device_handle *devh = open_card(card_index, &description); + if (dev == nullptr) { + devh = open_card(card_index, &description); + } else { + devh = open_card(card_index, dev, &description); + libusb_unref_device(dev); + } if (!devh) { fprintf(stderr, "Error finding USB device\n"); exit(1); @@ -919,6 +1028,8 @@ void BMUSBCapture::configure_card() // 0x20 - 720p?? // 0x30 - 576p?? + update_capture_mode(); + struct ctrl { int endpoint; int request; @@ -929,14 +1040,6 @@ void BMUSBCapture::configure_card() { LIBUSB_ENDPOINT_IN, 214, 16, 0 }, { LIBUSB_ENDPOINT_IN, 214, 0, 0 }, - // seems to capture on HDMI, clearing the 0x20000000 bit seems to activate 10-bit - // capture (v210). - // clearing the 0x08000000 bit seems to change the capture format (other source?) - // 0x10000000 = analog audio instead of embedded audio, it seems - // 0x3a000000 = component video? (analog audio) - // 0x3c000000 = composite video? (analog audio) - // 0x3e000000 = s-video? (analog audio) - { LIBUSB_ENDPOINT_OUT, 215, 0, 0x29000000 }, //{ LIBUSB_ENDPOINT_OUT, 215, 0, 0x80000100 }, //{ LIBUSB_ENDPOINT_OUT, 215, 0, 0x09000000 }, { LIBUSB_ENDPOINT_OUT, 215, 24, 0x73c60001 }, // latch for frame start? @@ -1027,12 +1130,12 @@ void BMUSBCapture::configure_card() xfr->user_data = this; //libusb_submit_transfer(xfr); - audiofp = fopen("audio.raw", "wb"); + //audiofp = fopen("audio.raw", "wb"); // set up isochronous transfers for audio and video for (int e = 3; e <= 4; ++e) { //int num_transfers = (e == 3) ? 6 : 6; - int num_transfers = 10; + int num_transfers = 6; for (int i = 0; i < num_transfers; ++i) { size_t buf_size; int num_iso_pack, size; @@ -1048,8 +1151,23 @@ void BMUSBCapture::configure_card() num_iso_pack = 80; buf_size = num_iso_pack * size; } - assert(size_t(num_iso_pack * size) <= buf_size); - uint8_t *buf = new uint8_t[buf_size]; + int num_bytes = num_iso_pack * size; + assert(size_t(num_bytes) <= buf_size); +#if LIBUSB_API_VERSION >= 0x01000105 + uint8_t *buf = libusb_dev_mem_alloc(devh, num_bytes); +#else + uint8_t *buf = nullptr; +#endif + if (buf == nullptr) { + fprintf(stderr, "Failed to allocate persistent DMA memory "); +#if LIBUSB_API_VERSION >= 0x01000105 + fprintf(stderr, "(probably too old kernel; use 4.6.0 or newer).\n"); +#else + fprintf(stderr, "(compiled against too old libusb-1.0).\n"); +#endif + fprintf(stderr, "Will go slower, and likely fail due to memory fragmentation after a few hours.\n"); + buf = new uint8_t[num_bytes]; + } xfr = libusb_alloc_transfer(num_iso_pack); if (!xfr) { @@ -1106,6 +1224,18 @@ void BMUSBCapture::stop_dequeue_thread() void BMUSBCapture::start_bm_thread() { + // Devices leaving are discovered by seeing the isochronous packets + // coming back with errors, so only care about devices joining. + if (card_connected_callback != nullptr) { + if (libusb_hotplug_register_callback( + nullptr, LIBUSB_HOTPLUG_EVENT_DEVICE_ARRIVED, LIBUSB_HOTPLUG_NO_FLAGS, + USB_VENDOR_BLACKMAGIC, LIBUSB_HOTPLUG_MATCH_ANY, LIBUSB_HOTPLUG_MATCH_ANY, + &BMUSBCapture::cb_hotplug, nullptr, nullptr) < 0) { + fprintf(stderr, "libusb_hotplug_register_callback() failed\n"); + exit(1); + } + } + should_quit = false; usb_thread = thread(&BMUSBCapture::usb_thread_func); } @@ -1230,3 +1360,68 @@ bool decode_video_format(uint16_t video_format, VideoFormat *decoded_video_forma decoded_video_format->frame_rate_den = 1; return false; } + +map BMUSBCapture::get_available_video_modes() const +{ + // The USB3 cards autodetect, and seem to have no provision for forcing modes. + VideoMode auto_mode; + auto_mode.name = "Autodetect"; + auto_mode.autodetect = true; + return {{ 0, auto_mode }}; +} + +uint32_t BMUSBCapture::get_current_video_mode() const +{ + return 0; // Matches get_available_video_modes(). +} + +void BMUSBCapture::set_video_mode(uint32_t video_mode_id) +{ + assert(video_mode_id == 0); // Matches get_available_video_modes(). +} + +std::map BMUSBCapture::get_available_video_inputs() const +{ + return { + { 0x00000000, "HDMI/SDI" }, + { 0x02000000, "Component" }, + { 0x04000000, "Composite" }, + { 0x06000000, "S-video" } + }; +} + +void BMUSBCapture::set_video_input(uint32_t video_input_id) +{ + assert((video_input_id & ~0x06000000) == 0); + current_video_input = video_input_id; + update_capture_mode(); +} + +std::map BMUSBCapture::get_available_audio_inputs() const +{ + return { + { 0x00000000, "Embedded" }, + { 0x10000000, "Analog" } + }; +} + +void BMUSBCapture::set_audio_input(uint32_t audio_input_id) +{ + assert((audio_input_id & ~0x10000000) == 0); + current_audio_input = audio_input_id; + update_capture_mode(); +} + +void BMUSBCapture::update_capture_mode() +{ + // clearing the 0x20000000 bit seems to activate 10-bit capture (v210). + // clearing the 0x08000000 bit seems to change the capture format (other source?) + uint32_t mode = htonl(0x29000000 | current_video_input | current_audio_input); + + int rc = libusb_control_transfer(devh, LIBUSB_REQUEST_TYPE_VENDOR | LIBUSB_ENDPOINT_OUT, + /*request=*/215, /*value=*/0, /*index=*/0, (unsigned char *)&mode, sizeof(mode), /*timeout=*/0); + if (rc < 0) { + fprintf(stderr, "Error on setting mode: %s\n", libusb_error_name(rc)); + exit(1); + } +}