]> git.sesse.net Git - bmusb/blobdiff - bmusb.cpp
Use even smaller transfers since evidently the kernel has problems allocating 512...
[bmusb] / bmusb.cpp
index c3914ac83ca29482a3e2ec503c8266daa3337a95..8639b938969bb0169640e8266c53afcdfb60afea 100644 (file)
--- a/bmusb.cpp
+++ b/bmusb.cpp
@@ -4,29 +4,31 @@
 // 576p60/720p60/1080i60 works, 1080p60 does not work (firmware limitation)
 // Audio comes out as 8-channel 24-bit raw audio.
 
+#include <assert.h>
+#include <errno.h>
+#include <libusb.h>
+#include <netinet/in.h>
+#include <sched.h>
+#include <stdint.h>
 #include <stdio.h>
 #include <stdlib.h>
-#include <libusb.h>
-#include <arpa/inet.h>
-#include <unistd.h>
 #include <string.h>
-#include <fcntl.h>
-#include <stdint.h>
-#include <assert.h>
-#ifdef __SSE2__
+#ifdef __SSE4_1__
 #include <immintrin.h>
 #endif
+#include "bmusb.h"
+
 #include <algorithm>
+#include <atomic>
+#include <condition_variable>
+#include <cstddef>
+#include <cstdint>
+#include <deque>
 #include <functional>
 #include <memory>
-#include <deque>
-#include <utility>
 #include <mutex>
-#include <condition_variable>
-#include <thread>
 #include <stack>
-#include <atomic>
-#include "bmusb.h"
+#include <thread>
 
 using namespace std;
 using namespace std::placeholders;
@@ -45,12 +47,21 @@ using namespace std::placeholders;
 
 FILE *audiofp;
 
+thread usb_thread;
+atomic<bool> should_quit;
+
 FrameAllocator::~FrameAllocator() {}
 
-#define NUM_QUEUED_FRAMES 8
+// Audio is more important than video, and also much cheaper.
+// By having many more audio frames available, hopefully if something
+// starts to drop, we'll have CPU load go down (from not having to
+// process as much video) before we have to drop audio.
+#define NUM_QUEUED_VIDEO_FRAMES 16
+#define NUM_QUEUED_AUDIO_FRAMES 64
+
 class MallocFrameAllocator : public FrameAllocator {
 public:
-       MallocFrameAllocator(size_t frame_size);
+       MallocFrameAllocator(size_t frame_size, size_t num_queued_frames);
        Frame alloc_frame() override;
        void release_frame(Frame frame) override;
 
@@ -61,10 +72,10 @@ private:
        stack<unique_ptr<uint8_t[]>> freelist;  // All of size <frame_size>.
 };
 
-MallocFrameAllocator::MallocFrameAllocator(size_t frame_size)
+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]));
        }
 }
@@ -88,6 +99,9 @@ FrameAllocator::Frame MallocFrameAllocator::alloc_frame()
 
 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));
 }
@@ -139,24 +153,31 @@ 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);
 }
 
-void BMUSBCapture::dequeue_thread()
+void BMUSBCapture::dequeue_thread_func()
 {
-       for ( ;; ) {
+       if (has_dequeue_callbacks) {
+               dequeue_init_callback();
+       }
+       while (!dequeue_thread_should_quit) {
                unique_lock<mutex> lock(queue_lock);
-               queues_not_empty.wait(lock, [this]{ return !pending_video_frames.empty() && !pending_audio_frames.empty(); });
+               queues_not_empty.wait(lock, [this]{ return dequeue_thread_should_quit || (!pending_video_frames.empty() && !pending_audio_frames.empty()); });
 
                uint16_t video_timecode = pending_video_frames.front().timecode;
                uint16_t audio_timecode = pending_audio_frames.front().timecode;
-               if (video_timecode < audio_timecode) {
+               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);
                        pending_video_frames.pop_front();
-               } else if (audio_timecode < video_timecode) {
-                       printf("Audio block 0x%04x without corresponding video block, dropping.\n",
+               } 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();
+                       frame_callback(audio_timecode,
+                                      FrameAllocator::Frame(), 0, 0x0000,
+                                      audio_frame.frame, AUDIO_HEADER_SIZE, audio_frame.format);
                } else {
                        QueuedFrame video_frame = pending_video_frames.front();
                        QueuedFrame audio_frame = pending_audio_frames.front();
@@ -176,6 +197,9 @@ void BMUSBCapture::dequeue_thread()
                                       audio_frame.frame, AUDIO_HEADER_SIZE, audio_frame.format);
                }
        }
+       if (has_dequeue_callbacks) {
+               dequeue_cleanup_callback();
+       }
 }
 
 void BMUSBCapture::start_new_frame(const uint8_t *start)
@@ -184,6 +208,21 @@ void BMUSBCapture::start_new_frame(const uint8_t *start)
        uint16_t timecode = (start[1] << 8) | start[0];
 
        if (current_video_frame.len > 0) {
+               // 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);
        }
@@ -252,8 +291,13 @@ void add_to_frame(FrameAllocator::Frame *current_frame, const char *frame_type_n
 
        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->interleaved) {
@@ -278,7 +322,50 @@ void add_to_frame(FrameAllocator::Frame *current_frame, const char *frame_type_n
        }
 }
 
-#ifdef __SSE2__
+#ifdef __SSE4_1__
+
+#if 0
+void avx2_dump(const char *name, __m256i n)
+{
+       printf("%-10s:", name);
+       printf(" %02x", _mm256_extract_epi8(n, 0));
+       printf(" %02x", _mm256_extract_epi8(n, 1));
+       printf(" %02x", _mm256_extract_epi8(n, 2));
+       printf(" %02x", _mm256_extract_epi8(n, 3));
+       printf(" %02x", _mm256_extract_epi8(n, 4));
+       printf(" %02x", _mm256_extract_epi8(n, 5));
+       printf(" %02x", _mm256_extract_epi8(n, 6));
+       printf(" %02x", _mm256_extract_epi8(n, 7));
+       printf(" ");
+       printf(" %02x", _mm256_extract_epi8(n, 8));
+       printf(" %02x", _mm256_extract_epi8(n, 9));
+       printf(" %02x", _mm256_extract_epi8(n, 10));
+       printf(" %02x", _mm256_extract_epi8(n, 11));
+       printf(" %02x", _mm256_extract_epi8(n, 12));
+       printf(" %02x", _mm256_extract_epi8(n, 13));
+       printf(" %02x", _mm256_extract_epi8(n, 14));
+       printf(" %02x", _mm256_extract_epi8(n, 15));
+       printf(" ");
+       printf(" %02x", _mm256_extract_epi8(n, 16));
+       printf(" %02x", _mm256_extract_epi8(n, 17));
+       printf(" %02x", _mm256_extract_epi8(n, 18));
+       printf(" %02x", _mm256_extract_epi8(n, 19));
+       printf(" %02x", _mm256_extract_epi8(n, 20));
+       printf(" %02x", _mm256_extract_epi8(n, 21));
+       printf(" %02x", _mm256_extract_epi8(n, 22));
+       printf(" %02x", _mm256_extract_epi8(n, 23));
+       printf(" ");
+       printf(" %02x", _mm256_extract_epi8(n, 24));
+       printf(" %02x", _mm256_extract_epi8(n, 25));
+       printf(" %02x", _mm256_extract_epi8(n, 26));
+       printf(" %02x", _mm256_extract_epi8(n, 27));
+       printf(" %02x", _mm256_extract_epi8(n, 28));
+       printf(" %02x", _mm256_extract_epi8(n, 29));
+       printf(" %02x", _mm256_extract_epi8(n, 30));
+       printf(" %02x", _mm256_extract_epi8(n, 31));
+       printf("\n");
+}
+#endif
 
 // Does a memcpy and memchr in one to reduce processing time.
 // Note that the benefit is somewhat limited if your L3 cache is small,
@@ -335,32 +422,39 @@ const uint8_t *add_to_frame_fastpath(FrameAllocator::Frame *current_frame, const
 #if __AVX2__
        const __m256i needle = _mm256_set1_epi8(sync_char);
 
-       const __m256i *in = (const __m256i *)aligned_start;
+       const __restrict __m256i *in = (const __m256i *)aligned_start;
        if (current_frame->interleaved) {
-               __m256i *out1 = (__m256i *)(current_frame->data + (current_frame->len + 1) / 2);
-               __m256i *out2 = (__m256i *)(current_frame->data2 + current_frame->len / 2);
+               __restrict __m256i *out1 = (__m256i *)(current_frame->data + (current_frame->len + 1) / 2);
+               __restrict __m256i *out2 = (__m256i *)(current_frame->data2 + current_frame->len / 2);
                if (current_frame->len % 2 == 1) {
                        swap(out1, out2);
                }
 
-               __m256i mask_lower_byte = _mm256_set1_epi16(0x00ff);
+               __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) {
-                       __m256i data1 = _mm256_load_si256(in);
-                       __m256i data2 = _mm256_load_si256(in + 1);
-                       __m256i data1_lo = _mm256_and_si256(data1, mask_lower_byte);
-                       __m256i data2_lo = _mm256_and_si256(data2, mask_lower_byte);
-                       __m256i data1_hi = _mm256_srli_epi16(data1, 8);
-                       __m256i data2_hi = _mm256_srli_epi16(data2, 8);
-                       __m256i lo = _mm256_packus_epi16(data1_lo, data2_lo);
-                       lo = _mm256_permute4x64_epi64(lo, 0b11011000);
-                       _mm256_storeu_si256(out1, lo);  // Store as early as possible, even if the data isn't used.
-                       __m256i hi = _mm256_packus_epi16(data1_hi, data2_hi);
-                       hi = _mm256_permute4x64_epi64(hi, 0b11011000);
-                       _mm256_storeu_si256(out2, hi);
+                       // 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
+
                        __m256i found1 = _mm256_cmpeq_epi8(data1, needle);
                        __m256i found2 = _mm256_cmpeq_epi8(data2, needle);
-                       if (!_mm256_testz_si256(found1, found1) ||
-                           !_mm256_testz_si256(found2, found2)) {
+                       __m256i found = _mm256_or_si256(found1, found2);
+
+                       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);  // Store as early as possible, even if the data isn't used.
+                       _mm256_storeu_si256(out2, hi);
+
+                       if (!_mm256_testz_si256(found, found)) {
                                break;
                        }
 
@@ -462,7 +556,7 @@ 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 __SSE2__
+#ifdef __SSE4_1__
                        start = add_to_frame_fastpath(current_frame, start, limit, sync_pattern[0]);
                        if (start == limit) break;
                        assert(start < limit);
@@ -544,16 +638,15 @@ void BMUSBCapture::cb_xfr(struct libusb_transfer *xfr)
        }
 #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);
        }
 }
 
 void BMUSBCapture::usb_thread_func()
 {
-       printf("usb thread started\n");
-
        sched_param param;
        memset(&param, 0, sizeof(param));
        param.sched_priority = 1;
@@ -567,19 +660,93 @@ void BMUSBCapture::usb_thread_func()
        }
 }
 
-void BMUSBCapture::start_bm_capture()
+struct USBCardDevice {
+       uint16_t product;
+       uint8_t bus, port;
+       libusb_device *device;
+};
+
+libusb_device_handle *open_card(int card_index)
+{      
+       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 == 0x1edb && desc.idProduct == 0xbd3b) &&
+                   !(desc.idVendor == 0x1edb && 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;
+       });
+
+       for (size_t i = 0; i < found_cards.size(); ++i) {
+               fprintf(stderr, "Card %d: Bus %03u Device %03u ", int(i), found_cards[i].bus, found_cards[i].port);
+               if (found_cards[i].product == 0xbd3b) {
+                       fprintf(stderr, "Intensity Shuttle\n");
+               } else if (found_cards[i].product == 0xbd4f) {
+                       fprintf(stderr, "UltraStudio SDI\n");
+               } else {
+                       assert(false);
+               }
+       }
+
+       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;
+}
+
+void BMUSBCapture::configure_card()
 {
        if (video_frame_allocator == nullptr) {
-               set_video_frame_allocator(new MallocFrameAllocator(FRAME_SIZE));  // FIXME: leak.
+               set_video_frame_allocator(new MallocFrameAllocator(FRAME_SIZE, NUM_QUEUED_VIDEO_FRAMES));  // FIXME: leak.
        }
        if (audio_frame_allocator == nullptr) {
-               set_audio_frame_allocator(new MallocFrameAllocator(65536));  // FIXME: leak.
+               set_audio_frame_allocator(new MallocFrameAllocator(65536, NUM_QUEUED_AUDIO_FRAMES));  // FIXME: leak.
        }
-       thread(&BMUSBCapture::dequeue_thread, this).detach();
+       dequeue_thread_should_quit = false;
+       dequeue_thread = thread(&BMUSBCapture::dequeue_thread_func, this);
 
        int rc;
        struct libusb_transfer *xfr;
-       vector<libusb_transfer *> iso_xfrs;
 
        rc = libusb_init(nullptr);
        if (rc < 0) {
@@ -587,7 +754,7 @@ void BMUSBCapture::start_bm_capture()
                exit(1);
        }
 
-       struct libusb_device_handle *devh = libusb_open_device_with_vid_pid(nullptr, 0x1edb, 0xbd3b);
+       libusb_device_handle *devh = open_card(card_index);
        if (!devh) {
                fprintf(stderr, "Error finding USB device\n");
                exit(1);
@@ -628,6 +795,9 @@ void BMUSBCapture::start_bm_capture()
        // Alternate setting 1 is output, alternate setting 2 is input.
        // Card is reset when switching alternates, so the driver uses
        // this “double switch” when it wants to reset.
+       //
+       // There's also alternate settings 3 and 4, which seem to be
+       // like 1 and 2 except they advertise less bandwidth needed.
        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));
@@ -635,7 +805,7 @@ void BMUSBCapture::start_bm_capture()
        }
        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
@@ -686,8 +856,9 @@ void BMUSBCapture::start_bm_capture()
        //    36 can be set to 0 with no apparent effect (all of this tested on both video and audio),
        //    but the driver sets it to 0x8036802a at some point.
        //
-       //    0 is written to during firmware upgrade. Probably best to stay out of it unless you know
-       //    what you're doing.
+       //    all of this is on request 214/215. other requests (192, 219,
+       //    222, 223, 224) are used for firmware upgrade. Probably best to
+       //    stay out of it unless you know what you're doing.
        //
        //
        // register 16:
@@ -709,59 +880,6 @@ void BMUSBCapture::start_bm_capture()
        static const ctrl ctrls[] = {
                { LIBUSB_ENDPOINT_IN,  214, 16, 0 },
                { LIBUSB_ENDPOINT_IN,  214,  0, 0 },
-               { LIBUSB_ENDPOINT_IN,  214,  0, 0 },
-               { LIBUSB_ENDPOINT_IN,  214,  4, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 16, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 20, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 24, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 28, 0 },
-               { LIBUSB_ENDPOINT_IN,  215, 32, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 36, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 40, 0 },
-               { LIBUSB_ENDPOINT_IN,  216, 44, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 48, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 52, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 40, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 40, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 40, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 24, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 40, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 40, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 40, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 24, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 40, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 40, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },  // packet 354
-               { LIBUSB_ENDPOINT_IN,  214, 24, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 12, 0 },
-               { LIBUSB_ENDPOINT_IN,  214, 40, 0 },
-               // more...
-               //{ LIBUSB_ENDPOINT_OUT, 215,  0, 0x80000100 },
-               //{ LIBUSB_ENDPOINT_OUT, 215,  0, 0x09000000 },  // wow, some kind of mode
 
                // seems to capture on HDMI, clearing the 0x20000000 bit seems to activate 10-bit
                // capture (v210).
@@ -771,46 +889,10 @@ void BMUSBCapture::start_bm_capture()
                // 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, 28, 0xffffffff },
-               //{ LIBUSB_ENDPOINT_OUT, 215, 32, 0xffffffff },
-               //{ LIBUSB_ENDPOINT_OUT, 215, 28, 0x40404040 },
-               //{ LIBUSB_ENDPOINT_OUT, 215, 32, 0x40404040 },
-               //{ LIBUSB_ENDPOINT_OUT, 215, 36, 0x8036802a },
                { LIBUSB_ENDPOINT_OUT, 215, 24, 0x73c60001 },  // latch for frame start?
-               //{ LIBUSB_ENDPOINT_OUT, 215, 24, 0x13370001 },  // latch for frame start?
                { LIBUSB_ENDPOINT_IN,  214, 24, 0 },  // 
-               //{ LIBUSB_ENDPOINT_OUT, 215,  4, 0x00000000 },  // appears to have no e fect
-               //{ LIBUSB_ENDPOINT_OUT, 215,  8, 0x00000000 },  // appears to have no effect
-               //{ LIBUSB_ENDPOINT_OUT, 215, 20, 0x00000000 },  // appears to have no effect
-               //{ LIBUSB_ENDPOINT_OUT, 215, 28, 0x00000000 },  // appears to have no effect
-               //{ LIBUSB_ENDPOINT_OUT, 215, 32, 0x00000000 },  // appears to have no effect
-               //{ LIBUSB_ENDPOINT_OUT, 215, 36, 0x00000000 },  // appears to have no effect
-#if 0
-               { LIBUSB_ENDPOINT_OUT, 215,  0 },
-               { LIBUSB_ENDPOINT_OUT, 215,  0 },
-               { LIBUSB_ENDPOINT_OUT, 215, 28 },
-               { LIBUSB_ENDPOINT_OUT, 215, 32 },
-               { LIBUSB_ENDPOINT_OUT, 215, 36 },
-               { LIBUSB_ENDPOINT_OUT, 215, 24 },
-               { LIBUSB_ENDPOINT_OUT, 215, 24 },
-               { LIBUSB_ENDPOINT_OUT, 215, 24 },
-               { LIBUSB_ENDPOINT_OUT, 215, 24 },
-               { LIBUSB_ENDPOINT_OUT, 215, 24 },
-               { LIBUSB_ENDPOINT_OUT, 215, 24 },
-               { LIBUSB_ENDPOINT_OUT, 215, 24 },
-               { LIBUSB_ENDPOINT_OUT, 215, 24 },
-               { LIBUSB_ENDPOINT_OUT, 215, 24 },
-               { LIBUSB_ENDPOINT_OUT, 215, 24 },
-               { LIBUSB_ENDPOINT_OUT, 215,  0 },
-               { LIBUSB_ENDPOINT_OUT, 215, 24 },
-               { LIBUSB_ENDPOINT_OUT, 215, 24 },
-               { LIBUSB_ENDPOINT_OUT, 215, 24 },
-               { LIBUSB_ENDPOINT_OUT, 215, 24 },
-               { LIBUSB_ENDPOINT_OUT, 215, 24 },
-               { LIBUSB_ENDPOINT_OUT, 215, 24 },
-#endif
        };
 
        for (unsigned req = 0; req < sizeof(ctrls) / sizeof(ctrls[0]); ++req) {
@@ -896,7 +978,7 @@ void BMUSBCapture::start_bm_capture()
        // 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;
+               int num_transfers = 10;
                for (int i = 0; i < num_transfers; ++i) {
                        int num_iso_pack, size;
                        if (e == 3) {
@@ -910,7 +992,7 @@ void BMUSBCapture::start_bm_capture()
                                        size &= ~1023;
                                        size += 1024;
                                }
-                               num_iso_pack = (2 << 20) / size;  // 2 MB.
+                               num_iso_pack = (2 << 16) / size;  // 128 kB.
                                printf("Picking %d packets of 0x%x bytes each\n", num_iso_pack, size);
                        } else {
                                size = 0xc0;
@@ -933,24 +1015,24 @@ void BMUSBCapture::start_bm_capture()
                        iso_xfrs.push_back(xfr);
                }
        }
+}
 
-       {
-               int i = 0;
-               for (libusb_transfer *xfr : iso_xfrs) {
-                       rc = libusb_submit_transfer(xfr);
-                       ++i;
-                       if (rc < 0) {
-                               //printf("num_bytes=%d\n", num_bytes);
-                               fprintf(stderr, "Error submitting iso to endpoint 0x%02x, number %d: %s\n",
-                                       xfr->endpoint, i, libusb_error_name(rc));
-                               exit(1);
-                       }
+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) {
+                       //printf("num_bytes=%d\n", num_bytes);
+                       fprintf(stderr, "Error submitting iso to endpoint 0x%02x, number %d: %s\n",
+                               xfr->endpoint, i, libusb_error_name(rc));
+                       exit(1);
                }
        }
 
-       should_quit = false;
-       usb_thread = thread(&BMUSBCapture::usb_thread_func, this);
-
 
 #if 0
        libusb_release_interface(devh, 0);
@@ -962,7 +1044,20 @@ out:
 #endif
 }
 
-void BMUSBCapture::stop_bm_capture()
+void BMUSBCapture::stop_dequeue_thread()
+{
+       dequeue_thread_should_quit = true;
+       queues_not_empty.notify_all();
+       dequeue_thread.join();
+}
+
+void BMUSBCapture::start_bm_thread()
+{
+       should_quit = false;
+       usb_thread = thread(&BMUSBCapture::usb_thread_func);
+}
+
+void BMUSBCapture::stop_bm_thread()
 {
        should_quit = true;
        usb_thread.join();