-// Intensity Shuttle USB3 prototype capture driver, v0.3
-// Can download 8-bit and 10-bit UYVY/v210 frames from HDMI, quite stable
+// Intensity Shuttle USB3 capture driver, v0.7.8
+// Can download 8-bit and 10-bit UYVY/v210-ish frames from HDMI, quite stable
// (can do captures for hours at a time with no drops), except during startup
// 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 <assert.h>
#include <errno.h>
#include <libusb.h>
+#include <unistd.h>
#include <netinet/in.h>
+#include <pthread.h>
#include <sched.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
-#ifdef __SSE4_1__
+#if HAS_MULTIVERSIONING
#include <immintrin.h>
#endif
-#include "bmusb.h"
+#include "bmusb/bmusb.h"
#include <algorithm>
#include <atomic>
+#include <chrono>
#include <condition_variable>
#include <cstddef>
#include <cstdint>
#include <memory>
#include <mutex>
#include <stack>
+#include <string>
#include <thread>
using namespace std;
+using namespace std::chrono;
using namespace std::placeholders;
-#define WIDTH 1280
-#define HEIGHT 750 /* 30 lines ancillary data? */
-//#define WIDTH 1920
-//#define HEIGHT 1125 /* ??? lines ancillary data? */
+#define USB_VENDOR_BLACKMAGIC 0x1edb
+#define MIN_WIDTH 640
#define HEADER_SIZE 44
//#define HEADER_SIZE 0
#define AUDIO_HEADER_SIZE 4
-//#define FRAME_SIZE (WIDTH * HEIGHT * 2 + HEADER_SIZE) // UYVY
-//#define FRAME_SIZE (WIDTH * HEIGHT * 2 * 4 / 3 + HEADER_SIZE) // v210
-#define FRAME_SIZE (8 << 20)
+#define FRAME_SIZE (8 << 20) // 8 MB.
+#define USB_VIDEO_TRANSFER_SIZE (128 << 10) // 128 kB.
+
+namespace bmusb {
+
+card_connected_callback_t BMUSBCapture::card_connected_callback = nullptr;
+bool BMUSBCapture::hotplug_existing_devices = false;
+
+namespace {
FILE *audiofp;
thread usb_thread;
atomic<bool> should_quit;
-FrameAllocator::~FrameAllocator() {}
+int v210_stride(int width)
+{
+ return (width + 5) / 6 * 4 * sizeof(uint32_t);
+}
-#define NUM_QUEUED_FRAMES 16
-class MallocFrameAllocator : public FrameAllocator {
-public:
- MallocFrameAllocator(size_t frame_size);
- Frame alloc_frame() override;
- void release_frame(Frame frame) override;
+int find_xfer_size_for_width(PixelFormat pixel_format, int width)
+{
+ // Video seems to require isochronous packets scaled with the width;
+ // seemingly six lines is about right, rounded up to the required 1kB
+ // multiple.
+ // Note that for 10-bit input, you'll need to increase size accordingly.
+ int stride;
+ if (pixel_format == PixelFormat_10BitYCbCr) {
+ stride = v210_stride(width);
+ } else {
+ stride = width * sizeof(uint16_t);
+ }
+ int size = stride * 6;
+ if (size % 1024 != 0) {
+ size &= ~1023;
+ size += 1024;
+ }
+ return size;
+}
-private:
- size_t frame_size;
+void change_xfer_size_for_width(PixelFormat pixel_format, int width, libusb_transfer *xfr)
+{
+ assert(width >= MIN_WIDTH);
+ size_t size = find_xfer_size_for_width(pixel_format, width);
+ int num_iso_pack = xfr->length / size;
+ if (num_iso_pack != xfr->num_iso_packets ||
+ size != xfr->iso_packet_desc[0].length) {
+ xfr->num_iso_packets = num_iso_pack;
+ libusb_set_iso_packet_lengths(xfr, size);
+ }
+}
- mutex freelist_mutex;
- stack<unique_ptr<uint8_t[]>> freelist; // All of size <frame_size>.
+struct VideoFormatEntry {
+ uint16_t normalized_video_format;
+ unsigned width, height, second_field_start;
+ unsigned extra_lines_top, extra_lines_bottom;
+ unsigned frame_rate_nom, frame_rate_den;
+ bool interlaced;
};
-MallocFrameAllocator::MallocFrameAllocator(size_t frame_size)
+// Get details for the given video format; returns false if detection was incomplete.
+bool decode_video_format(uint16_t video_format, VideoFormat *decoded_video_format)
+{
+ decoded_video_format->id = video_format;
+ decoded_video_format->interlaced = false;
+
+ // TODO: Add these for all formats as we find them.
+ decoded_video_format->extra_lines_top = decoded_video_format->extra_lines_bottom = decoded_video_format->second_field_start = 0;
+
+ if (video_format == 0x0800) {
+ // No video signal. These green pseudo-frames seem to come at about 30.13 Hz.
+ // It's a strange thing, but what can you do.
+ decoded_video_format->width = 720;
+ decoded_video_format->height = 525;
+ decoded_video_format->stride = 720 * 2;
+ decoded_video_format->extra_lines_top = 0;
+ decoded_video_format->extra_lines_bottom = 0;
+ decoded_video_format->frame_rate_nom = 3013;
+ decoded_video_format->frame_rate_den = 100;
+ decoded_video_format->has_signal = false;
+ return true;
+ }
+ if ((video_format & 0xe000) != 0xe000) {
+ printf("Video format 0x%04x does not appear to be a video format. Assuming 60 Hz.\n",
+ video_format);
+ decoded_video_format->width = 0;
+ decoded_video_format->height = 0;
+ decoded_video_format->stride = 0;
+ decoded_video_format->extra_lines_top = 0;
+ decoded_video_format->extra_lines_bottom = 0;
+ decoded_video_format->frame_rate_nom = 60;
+ decoded_video_format->frame_rate_den = 1;
+ decoded_video_format->has_signal = false;
+ return false;
+ }
+
+ decoded_video_format->has_signal = true;
+
+ // NTSC (480i59.94, I suppose). A special case, see below.
+ if ((video_format & ~0x0800) == 0xe101 ||
+ (video_format & ~0x0800) == 0xe1c1 ||
+ (video_format & ~0x0800) == 0xe001) {
+ decoded_video_format->width = 720;
+ decoded_video_format->height = 480;
+ if (video_format & 0x0800) {
+ decoded_video_format->stride = 720 * 2;
+ } else {
+ decoded_video_format->stride = v210_stride(720);
+ }
+ decoded_video_format->extra_lines_top = 17;
+ decoded_video_format->extra_lines_bottom = 28;
+ decoded_video_format->frame_rate_nom = 30000;
+ decoded_video_format->frame_rate_den = 1001;
+ decoded_video_format->second_field_start = 280;
+ decoded_video_format->interlaced = true;
+ return true;
+ }
+
+ // PAL (576i50, I suppose). A special case, see below.
+ if ((video_format & ~0x0800) == 0xe109 ||
+ (video_format & ~0x0800) == 0xe1c9 ||
+ (video_format & ~0x0800) == 0xe009 ||
+ (video_format & ~0x0800) == 0xe3e9 ||
+ (video_format & ~0x0800) == 0xe3e1) {
+ decoded_video_format->width = 720;
+ decoded_video_format->height = 576;
+ if (video_format & 0x0800) {
+ decoded_video_format->stride = 720 * 2;
+ } else {
+ decoded_video_format->stride = v210_stride(720);
+ }
+ decoded_video_format->extra_lines_top = 22;
+ decoded_video_format->extra_lines_bottom = 27;
+ decoded_video_format->frame_rate_nom = 25;
+ decoded_video_format->frame_rate_den = 1;
+ decoded_video_format->second_field_start = 335;
+ decoded_video_format->interlaced = true;
+ return true;
+ }
+
+ // 0x8 seems to be a flag about availability of deep color on the input,
+ // except when it's not (e.g. it's the only difference between NTSC
+ // and PAL). Rather confusing. But we clear it here nevertheless, because
+ // usually it doesn't mean anything. 0x0800 appears to be 8-bit input
+ // (as opposed to 10-bit).
+ //
+ // 0x4 is a flag I've only seen from the D4. I don't know what it is.
+ uint16_t normalized_video_format = video_format & ~0xe80c;
+ constexpr VideoFormatEntry entries[] = {
+ { 0x01f1, 720, 480, 0, 40, 5, 60000, 1001, false }, // 480p59.94 (believed).
+ { 0x0131, 720, 576, 0, 44, 5, 50, 1, false }, // 576p50.
+ { 0x0151, 720, 576, 0, 44, 5, 50, 1, false }, // 576p50.
+ { 0x0011, 720, 576, 0, 44, 5, 50, 1, false }, // 576p50 (5:4).
+ { 0x0143, 1280, 720, 0, 25, 5, 50, 1, false }, // 720p50.
+ { 0x0161, 1280, 720, 0, 25, 5, 50, 1, false }, // 720p50.
+ { 0x0103, 1280, 720, 0, 25, 5, 60, 1, false }, // 720p60.
+ { 0x0125, 1280, 720, 0, 25, 5, 60, 1, false }, // 720p60.
+ { 0x0121, 1280, 720, 0, 25, 5, 60000, 1001, false }, // 720p59.94.
+ { 0x01c3, 1920, 1080, 0, 41, 4, 30, 1, false }, // 1080p30.
+ { 0x0003, 1920, 1080, 583, 20, 25, 30, 1, true }, // 1080i60.
+ { 0x01e1, 1920, 1080, 0, 41, 4, 30000, 1001, false }, // 1080p29.97.
+ { 0x0021, 1920, 1080, 583, 20, 25, 30000, 1001, true }, // 1080i59.94.
+ { 0x0063, 1920, 1080, 0, 41, 4, 25, 1, false }, // 1080p25.
+ { 0x0043, 1920, 1080, 583, 20, 25, 25, 1, true }, // 1080i50.
+ { 0x0083, 1920, 1080, 0, 41, 4, 24, 1, false }, // 1080p24.
+ { 0x00a1, 1920, 1080, 0, 41, 4, 24000, 1001, false }, // 1080p23.98.
+ };
+ for (const VideoFormatEntry &entry : entries) {
+ if (normalized_video_format == entry.normalized_video_format) {
+ decoded_video_format->width = entry.width;
+ decoded_video_format->height = entry.height;
+ if (video_format & 0x0800) {
+ decoded_video_format->stride = entry.width * 2;
+ } else {
+ decoded_video_format->stride = v210_stride(entry.width);
+ }
+ decoded_video_format->second_field_start = entry.second_field_start;
+ decoded_video_format->extra_lines_top = entry.extra_lines_top;
+ decoded_video_format->extra_lines_bottom = entry.extra_lines_bottom;
+ decoded_video_format->frame_rate_nom = entry.frame_rate_nom;
+ decoded_video_format->frame_rate_den = entry.frame_rate_den;
+ decoded_video_format->interlaced = entry.interlaced;
+ return true;
+ }
+ }
+
+ printf("Unknown video format 0x%04x (normalized 0x%04x). Assuming 720p60.\n", video_format, normalized_video_format);
+ decoded_video_format->width = 1280;
+ decoded_video_format->height = 720;
+ decoded_video_format->stride = 1280 * 2;
+ decoded_video_format->frame_rate_nom = 60;
+ decoded_video_format->frame_rate_den = 1;
+ return false;
+}
+
+// There are seemingly no direct indicators of sample rate; you just get
+// one frame's worth and have to guess from that.
+int guess_sample_rate(const VideoFormat &video_format, size_t len, int default_rate)
+{
+ size_t num_samples = len / 3 / 8;
+ size_t num_samples_per_second = num_samples * video_format.frame_rate_nom / video_format.frame_rate_den;
+
+ // See if we match or are very close to any of the mandatory HDMI sample rates.
+ const int candidate_sample_rates[] = { 32000, 44100, 48000 };
+ for (int rate : candidate_sample_rates) {
+ if (abs(int(num_samples_per_second) - rate) <= 100) {
+ return rate;
+ }
+ }
+
+ fprintf(stderr, "%ld samples at %d/%d fps (%ld Hz) matches no known sample rate, keeping capture at %d Hz\n",
+ num_samples, video_format.frame_rate_nom, video_format.frame_rate_den, num_samples_per_second, default_rate);
+ return default_rate;
+}
+
+} // namespace
+
+FrameAllocator::~FrameAllocator() {}
+
+MallocFrameAllocator::MallocFrameAllocator(size_t frame_size, size_t num_queued_frames)
: frame_size(frame_size)
{
- for (int i = 0; i < NUM_QUEUED_FRAMES; ++i) {
+ for (size_t i = 0; i < num_queued_frames; ++i) {
freelist.push(unique_ptr<uint8_t[]>(new uint8_t[frame_size]));
}
}
void MallocFrameAllocator::release_frame(Frame frame)
{
+ if (frame.overflow > 0) {
+ printf("%d bytes overflow after last (malloc) frame\n", int(frame.overflow));
+ }
unique_lock<mutex> lock(freelist_mutex);
freelist.push(unique_ptr<uint8_t[]>(frame.data));
}
void BMUSBCapture::queue_frame(uint16_t format, uint16_t timecode, FrameAllocator::Frame frame, deque<QueuedFrame> *q)
{
+ unique_lock<mutex> lock(queue_lock);
if (!q->empty() && !uint16_less_than_with_wraparound(q->back().timecode, timecode)) {
printf("Blocks going backwards: prev=0x%04x, cur=0x%04x (dropped)\n",
q->back().timecode, timecode);
qf.format = format;
qf.timecode = timecode;
qf.frame = frame;
-
- {
- unique_lock<mutex> lock(queue_lock);
- q->push_back(move(qf));
- }
+ q->push_back(move(qf));
queues_not_empty.notify_one(); // might be spurious
}
void dump_audio_block(uint8_t *audio_start, size_t audio_len)
{
- fwrite(audio_start + AUDIO_HEADER_SIZE, 1, audio_len - AUDIO_HEADER_SIZE, audiofp);
+ if (audiofp != nullptr) {
+ fwrite(audio_start + AUDIO_HEADER_SIZE, 1, audio_len - AUDIO_HEADER_SIZE, audiofp);
+ }
}
void BMUSBCapture::dequeue_thread_func()
{
+ char thread_name[16];
+ snprintf(thread_name, sizeof(thread_name), "bmusb_dequeue_%d", card_index);
+ pthread_setname_np(pthread_self(), thread_name);
+
if (has_dequeue_callbacks) {
dequeue_init_callback();
}
+ size_t last_sample_rate = 48000;
while (!dequeue_thread_should_quit) {
unique_lock<mutex> lock(queue_lock);
queues_not_empty.wait(lock, [this]{ return dequeue_thread_should_quit || (!pending_video_frames.empty() && !pending_audio_frames.empty()); });
+ if (dequeue_thread_should_quit) break;
+
uint16_t video_timecode = pending_video_frames.front().timecode;
uint16_t audio_timecode = pending_audio_frames.front().timecode;
- if (video_timecode < audio_timecode) {
+ AudioFormat audio_format;
+ audio_format.bits_per_sample = 24;
+ audio_format.num_channels = 8;
+ audio_format.sample_rate = last_sample_rate;
+ if (uint16_less_than_with_wraparound(video_timecode, audio_timecode)) {
printf("Video block 0x%04x without corresponding audio block, dropping.\n",
video_timecode);
- video_frame_allocator->release_frame(pending_video_frames.front().frame);
+ QueuedFrame video_frame = pending_video_frames.front();
pending_video_frames.pop_front();
- } else if (audio_timecode < video_timecode) {
- printf("Audio block 0x%04x without corresponding video block, dropping.\n",
+ lock.unlock();
+ video_frame_allocator->release_frame(video_frame.frame);
+ } else if (uint16_less_than_with_wraparound(audio_timecode, video_timecode)) {
+ printf("Audio block 0x%04x without corresponding video block, sending blank frame.\n",
audio_timecode);
- audio_frame_allocator->release_frame(pending_audio_frames.front().frame);
+ QueuedFrame audio_frame = pending_audio_frames.front();
pending_audio_frames.pop_front();
+ lock.unlock();
+ audio_format.id = audio_frame.format;
+
+ // Use the video format of the pending frame.
+ QueuedFrame video_frame = pending_video_frames.front();
+ VideoFormat video_format;
+ decode_video_format(video_frame.format, &video_format);
+
+ frame_callback(audio_timecode,
+ FrameAllocator::Frame(), 0, video_format,
+ audio_frame.frame, AUDIO_HEADER_SIZE, audio_format);
} else {
QueuedFrame video_frame = pending_video_frames.front();
QueuedFrame audio_frame = pending_audio_frames.front();
dump_audio_block(audio_frame.frame.data, audio_frame.data_len);
#endif
- frame_callback(video_timecode,
- video_frame.frame, HEADER_SIZE, video_frame.format,
- audio_frame.frame, AUDIO_HEADER_SIZE, audio_frame.format);
+ VideoFormat video_format;
+ audio_format.id = audio_frame.format;
+ if (decode_video_format(video_frame.format, &video_format)) {
+ if (audio_frame.frame.len != 0) {
+ audio_format.sample_rate = guess_sample_rate(video_format, audio_frame.frame.len, last_sample_rate);
+ last_sample_rate = audio_format.sample_rate;
+ }
+ frame_callback(video_timecode,
+ video_frame.frame, HEADER_SIZE, video_format,
+ audio_frame.frame, AUDIO_HEADER_SIZE, audio_format);
+ } else {
+ video_frame_allocator->release_frame(video_frame.frame);
+ audio_format.sample_rate = last_sample_rate;
+ frame_callback(video_timecode,
+ FrameAllocator::Frame(), 0, video_format,
+ audio_frame.frame, AUDIO_HEADER_SIZE, audio_format);
+ }
}
}
if (has_dequeue_callbacks) {
uint16_t timecode = (start[1] << 8) | start[0];
if (current_video_frame.len > 0) {
+ current_video_frame.received_timestamp = steady_clock::now();
+
+ // If format is 0x0800 (no signal), add a fake (empty) audio
+ // frame to get it out of the queue.
+ // TODO: Figure out if there are other formats that come with
+ // no audio, and treat them the same.
+ if (format == 0x0800) {
+ FrameAllocator::Frame fake_audio_frame = audio_frame_allocator->alloc_frame();
+ if (fake_audio_frame.data == nullptr) {
+ // Oh well, it's just a no-signal frame anyway.
+ printf("Couldn't allocate fake audio frame, also dropping no-signal video frame.\n");
+ current_video_frame.owner->release_frame(current_video_frame);
+ current_video_frame = video_frame_allocator->alloc_frame();
+ return;
+ }
+ queue_frame(format, timecode, fake_audio_frame, &pending_audio_frames);
+ }
//dump_frame();
queue_frame(format, timecode, current_video_frame, &pending_video_frames);
+
+ // Update the assumed frame width. We might be one frame too late on format changes,
+ // but it's much better than asking the user to choose manually.
+ VideoFormat video_format;
+ if (decode_video_format(format, &video_format)) {
+ assumed_frame_width = video_format.width;
+ }
}
//printf("Found frame start, format 0x%04x timecode 0x%04x, previous frame length was %d/%d\n",
// format, timecode,
uint16_t format = (start[3] << 8) | start[2];
uint16_t timecode = (start[1] << 8) | start[0];
if (current_audio_frame.len > 0) {
+ current_audio_frame.received_timestamp = steady_clock::now();
//dump_audio_block();
queue_frame(format, timecode, current_audio_frame, &pending_audio_frames);
}
- //printf("Found audio block start, format 0x%04x timecode 0x%04x, previous block length was %d\n",
- // format, timecode, read_current_audio_block);
+ //printf("Found audio block start, format 0x%04x timecode 0x%04x\n",
+ // format, timecode);
current_audio_frame = audio_frame_allocator->alloc_frame();
}
int bytes = end - start;
if (current_frame->len + bytes > current_frame->size) {
- printf("%d bytes overflow after last %s frame\n",
- int(current_frame->len + bytes - current_frame->size), frame_type_name);
+ current_frame->overflow = current_frame->len + bytes - current_frame->size;
+ current_frame->len = current_frame->size;
+ if (current_frame->overflow > 1048576) {
+ printf("%d bytes overflow after last %s frame\n",
+ int(current_frame->overflow), frame_type_name);
+ current_frame->overflow = 0;
+ }
//dump_frame();
} else {
+ if (current_frame->data_copy != nullptr) {
+ memcpy(current_frame->data_copy + current_frame->len, start, bytes);
+ }
if (current_frame->interleaved) {
uint8_t *data = current_frame->data + current_frame->len / 2;
uint8_t *data2 = current_frame->data2 + current_frame->len / 2;
}
}
-#ifdef __SSE4_1__
-
#if 0
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)
+
+__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);
+
+__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);
+
// 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
// 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 ||
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);
+ size_t bytes_copied;
const __restrict __m256i *in = (const __m256i *)aligned_start;
if (current_frame->interleaved) {
__restrict __m256i *out1 = (__m256i *)(current_frame->data + (current_frame->len + 1) / 2);
++out1;
++out2;
}
- current_frame->len += (uint8_t *)in - aligned_start;
+ bytes_copied = (uint8_t *)in - aligned_start;
} else {
- __m256i *out = (__m256i *)(current_frame->data + current_frame->len);
+ uint8_t *old_end = current_frame->data + current_frame->len;
+ __m256i *out = (__m256i *)old_end;
while (in < (const __m256i *)limit) {
__m256i data = _mm256_load_si256(in);
_mm256_storeu_si256(out, data); // Store as early as possible, even if the data isn't used.
++in;
++out;
}
- current_frame->len = (uint8_t *)out - current_frame->data;
+ bytes_copied = (uint8_t *)out - old_end;
}
-#else
+ if (current_frame->data_copy != nullptr) {
+ // TODO: It would be somewhat more cache-efficient to write this in the
+ // same loop as above. However, it might not be worth the extra complexity.
+ memcpy(current_frame->data_copy + current_frame->len, aligned_start, bytes_copied);
+ }
+ current_frame->len += bytes_copied;
+
+ //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;
+ size_t bytes_copied;
if (current_frame->interleaved) {
__m128i *out1 = (__m128i *)(current_frame->data + (current_frame->len + 1) / 2);
__m128i *out2 = (__m128i *)(current_frame->data2 + current_frame->len / 2);
++out1;
++out2;
}
- current_frame->len += (uint8_t *)in - aligned_start;
+ bytes_copied = (uint8_t *)in - aligned_start;
} else {
- __m128i *out = (__m128i *)(current_frame->data + current_frame->len);
+ uint8_t *old_end = current_frame->data + current_frame->len;
+ __m128i *out = (__m128i *)old_end;
while (in < (const __m128i *)limit) {
__m128i data = _mm_load_si128(in);
_mm_storeu_si128(out, data); // Store as early as possible, even if the data isn't used.
++in;
++out;
}
- current_frame->len = (uint8_t *)out - current_frame->data;
+ bytes_copied = (uint8_t *)out - old_end;
}
-#endif
+ if (current_frame->data_copy != nullptr) {
+ // TODO: It would be somewhat more cache-efficient to write this in the
+ // same loop as above. However, it might not be worth the extra complexity.
+ memcpy(current_frame->data_copy + current_frame->len, aligned_start, bytes_copied);
+ }
+ current_frame->len += bytes_copied;
//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,
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) {
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);
}
assert(xfr->user_data != nullptr);
BMUSBCapture *usb = static_cast<BMUSBCapture *>(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));
} else {
decode_packs(xfr, "\x00\x00\xff\xff", 4, &usb->current_video_frame, "video", bind(&BMUSBCapture::start_new_frame, usb, _1));
+
+ // Update the transfer with the new assumed width, if we're in the process of changing formats.
+ change_xfer_size_for_width(usb->current_pixel_format, usb->assumed_frame_width, xfr);
}
}
if (xfr->type == LIBUSB_TRANSFER_TYPE_CONTROL) {
}
#endif
- if (libusb_submit_transfer(xfr) < 0) {
- fprintf(stderr, "error re-submitting URB\n");
+ int rc = libusb_submit_transfer(xfr);
+ if (rc < 0) {
+ fprintf(stderr, "error re-submitting URB: %s\n", libusb_error_name(rc));
exit(1);
}
}
+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;
if (sched_setscheduler(0, SCHED_RR, ¶m) == -1) {
printf("couldn't set realtime priority for USB thread: %s\n", strerror(errno));
}
+ pthread_setname_np(pthread_self(), "bmusb_usb_drv");
while (!should_quit) {
- int rc = libusb_handle_events(nullptr);
+ timeval sec { 1, 0 };
+ int rc = libusb_handle_events_timeout(nullptr, &sec);
if (rc != LIBUSB_SUCCESS)
break;
}
}
+namespace {
+
+struct USBCardDevice {
+ uint16_t product;
+ uint8_t bus, port;
+ 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;
+}
+
+vector<USBCardDevice> find_all_cards()
+{
+ libusb_device **devices;
+ ssize_t num_devices = libusb_get_device_list(nullptr, &devices);
+ if (num_devices == -1) {
+ fprintf(stderr, "Error finding USB devices\n");
+ exit(1);
+ }
+ vector<USBCardDevice> found_cards;
+ for (ssize_t i = 0; i < num_devices; ++i) {
+ libusb_device_descriptor desc;
+ if (libusb_get_device_descriptor(devices[i], &desc) < 0) {
+ fprintf(stderr, "Error getting device descriptor for device %d\n", int(i));
+ exit(1);
+ }
+
+ uint8_t bus = libusb_get_bus_number(devices[i]);
+ uint8_t port = libusb_get_port_number(devices[i]);
+
+ if (!(desc.idVendor == USB_VENDOR_BLACKMAGIC && desc.idProduct == 0xbd3b) &&
+ !(desc.idVendor == USB_VENDOR_BLACKMAGIC && desc.idProduct == 0xbd4f)) {
+ libusb_unref_device(devices[i]);
+ continue;
+ }
+
+ found_cards.push_back({ desc.idProduct, bus, port, devices[i] });
+ }
+ libusb_free_device_list(devices, 0);
+
+ // Sort the devices to get a consistent ordering.
+ sort(found_cards.begin(), found_cards.end(), [](const USBCardDevice &a, const USBCardDevice &b) {
+ if (a.product != b.product)
+ return a.product < b.product;
+ if (a.bus != b.bus)
+ return a.bus < b.bus;
+ return a.port < b.port;
+ });
+
+ return found_cards;
+}
+
+libusb_device_handle *open_card(int card_index, string *description)
+{
+ vector<USBCardDevice> found_cards = find_all_cards();
+
+ for (size_t i = 0; i < found_cards.size(); ++i) {
+ 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 = tmp_description;
+ }
+ }
+
+ if (size_t(card_index) >= found_cards.size()) {
+ fprintf(stderr, "Could not open card %d (only %d found)\n", card_index, int(found_cards.size()));
+ exit(1);
+ }
+
+ libusb_device_handle *devh;
+ int rc = libusb_open(found_cards[card_index].device, &devh);
+ if (rc < 0) {
+ fprintf(stderr, "Error opening card %d: %s\n", card_index, libusb_error_name(rc));
+ exit(1);
+ }
+
+ for (size_t i = 0; i < found_cards.size(); ++i) {
+ libusb_unref_device(found_cards[i].device);
+ }
+
+ 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;
+}
+
+} // namespace
+
+unsigned BMUSBCapture::num_cards()
+{
+ int rc = libusb_init(nullptr);
+ if (rc < 0) {
+ fprintf(stderr, "Error initializing libusb: %s\n", libusb_error_name(rc));
+ exit(1);
+ }
+
+ vector<USBCardDevice> found_cards = find_all_cards();
+ unsigned ret = found_cards.size();
+ for (size_t i = 0; i < found_cards.size(); ++i) {
+ libusb_unref_device(found_cards[i].device);
+ }
+ return ret;
+}
+
+void BMUSBCapture::set_pixel_format(PixelFormat pixel_format)
+{
+ current_pixel_format = pixel_format;
+ update_capture_mode();
+}
+
void BMUSBCapture::configure_card()
{
if (video_frame_allocator == nullptr) {
- set_video_frame_allocator(new MallocFrameAllocator(FRAME_SIZE)); // 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)); // 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);
exit(1);
}
- //struct libusb_device_handle *devh = libusb_open_device_with_vid_pid(nullptr, 0x1edb, 0xbd3b);
- //struct libusb_device_handle *devh = libusb_open_device_with_vid_pid(nullptr, 0x1edb, 0xbd4f);
- struct libusb_device_handle *devh = libusb_open_device_with_vid_pid(nullptr, vid, pid);
+ 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);
fprintf(stderr, "Error getting configuration: %s\n", libusb_error_name(rc));
exit(1);
}
+
+#if 0
printf("%d interface\n", config->bNumInterfaces);
for (int interface_number = 0; interface_number < config->bNumInterfaces; ++interface_number) {
printf(" interface %d\n", interface_number);
}
}
}
+#endif
rc = libusb_set_configuration(devh, /*configuration=*/1);
if (rc < 0) {
rc = libusb_set_interface_alt_setting(devh, /*interface=*/0, /*alternate_setting=*/1);
if (rc < 0) {
fprintf(stderr, "Error setting alternate 1: %s\n", libusb_error_name(rc));
+ if (rc == LIBUSB_ERROR_NOT_FOUND) {
+ fprintf(stderr, "This is usually because the card came up in USB2 mode.\n");
+ fprintf(stderr, "In particular, this tends to happen if you boot up with the\n");
+ fprintf(stderr, "card plugged in; just unplug and replug it, and it usually works.\n");
+ }
exit(1);
}
rc = libusb_set_interface_alt_setting(devh, /*interface=*/0, /*alternate_setting=*/2);
if (rc < 0) {
- fprintf(stderr, "Error setting alternate 1: %s\n", libusb_error_name(rc));
+ fprintf(stderr, "Error setting alternate 2: %s\n", libusb_error_name(rc));
exit(1);
}
#if 0
// 0x20 - 720p??
// 0x30 - 576p??
+ update_capture_mode();
+
struct ctrl {
int endpoint;
int request;
{ 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?
fprintf(stderr, "Error on control %d: %s\n", ctrls[req].index, libusb_error_name(rc));
exit(1);
}
-
+
+ if (ctrls[req].index == 16 && rc == 4) {
+ printf("Card firmware version: 0x%02x%02x\n", value[2], value[3]);
+ }
+
+#if 0
printf("rc=%d: ep=%d@%d %d -> 0x", rc, ctrls[req].endpoint, ctrls[req].request, ctrls[req].index);
for (int i = 0; i < rc; ++i) {
printf("%02x", value[i]);
}
printf("\n");
+#endif
}
#if 0
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 = 6;
for (int i = 0; i < num_transfers; ++i) {
+ size_t buf_size;
int num_iso_pack, size;
if (e == 3) {
- // Video seems to require isochronous packets scaled with the width;
- // seemingly six lines is about right, rounded up to the required 1kB
- // multiple.
- size = WIDTH * 2 * 6;
- // Note that for 10-bit input, you'll need to increase size accordingly.
- //size = size * 4 / 3;
- if (size % 1024 != 0) {
- size &= ~1023;
- size += 1024;
- }
- num_iso_pack = (2 << 18) / size; // 512 kB.
- printf("Picking %d packets of 0x%x bytes each\n", num_iso_pack, size);
+ // Allocate for minimum width (because that will give us the most
+ // number of packets, so we don't need to reallocate, but we'll
+ // default to 720p for the first frame.
+ size = find_xfer_size_for_width(PixelFormat_8BitYCbCr, MIN_WIDTH);
+ num_iso_pack = USB_VIDEO_TRANSFER_SIZE / size;
+ buf_size = USB_VIDEO_TRANSFER_SIZE;
} else {
size = 0xc0;
num_iso_pack = 80;
+ buf_size = num_iso_pack * size;
}
int num_bytes = num_iso_pack * size;
- uint8_t *buf = new uint8_t[num_bytes];
+ 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) {
}
int ep = LIBUSB_ENDPOINT_IN | e;
- libusb_fill_iso_transfer(xfr, devh, ep, buf, num_bytes,
+ libusb_fill_iso_transfer(xfr, devh, ep, buf, buf_size,
num_iso_pack, cb_xfr, nullptr, 0);
libusb_set_iso_packet_lengths(xfr, size);
xfr->user_data = this;
+
+ if (e == 3) {
+ change_xfer_size_for_width(current_pixel_format, assumed_frame_width, xfr);
+ }
+
iso_xfrs.push_back(xfr);
}
}
void BMUSBCapture::start_bm_capture()
{
- printf("starting capture\n");
int i = 0;
for (libusb_transfer *xfr : iso_xfrs) {
- printf("submitting transfer...\n");
int rc = libusb_submit_transfer(xfr);
++i;
if (rc < 0) {
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, hotplug_existing_devices ? LIBUSB_HOTPLUG_ENUMERATE : 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);
}
void BMUSBCapture::stop_bm_thread()
{
should_quit = true;
+ libusb_interrupt_event_handler(nullptr);
usb_thread.join();
}
+
+map<uint32_t, VideoMode> 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<uint32_t, std::string> 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<uint32_t, std::string> 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()
+{
+ if (devh == nullptr) {
+ return;
+ }
+
+ // Clearing the 0x08000000 bit seems to change the capture format (other source?).
+ uint32_t mode = htonl(0x09000000 | current_video_input | current_audio_input);
+ if (current_pixel_format == PixelFormat_8BitYCbCr) {
+ mode |= htonl(0x20000000);
+ } else {
+ assert(current_pixel_format == PixelFormat_10BitYCbCr);
+ }
+
+ 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);
+ }
+}
+
+} // namespace bmusb
projects / bmusb / blobdiff