]> git.sesse.net Git - nageru/blobdiff - mixer.cpp
Unify muxing between the local file and networking.
[nageru] / mixer.cpp
index 8c7a25bceff483eeb4eff9c846440510fda49950..59f520aacdc12a0dbc70998c13da3df890a72a8c 100644 (file)
--- a/mixer.cpp
+++ b/mixer.cpp
@@ -1,5 +1,6 @@
 #define WIDTH 1280
 #define HEIGHT 720
+#define EXTRAHEIGHT 30
 
 #undef Success
 
 #include <thread>
 #include <vector>
 
-#include "bmusb.h"
+#include "bmusb/bmusb.h"
 #include "context.h"
 #include "h264encode.h"
 #include "pbo_frame_allocator.h"
 #include "ref_counted_gl_sync.h"
+#include "timebase.h"
 
 class QOpenGLContext;
 
@@ -50,10 +52,31 @@ using namespace std::placeholders;
 
 Mixer *global_mixer = nullptr;
 
+namespace {
+
+void convert_fixed24_to_fp32(float *dst, size_t out_channels, const uint8_t *src, size_t in_channels, size_t num_samples)
+{
+       for (size_t i = 0; i < num_samples; ++i) {
+               for (size_t j = 0; j < out_channels; ++j) {
+                       uint32_t s1 = *src++;
+                       uint32_t s2 = *src++;
+                       uint32_t s3 = *src++;
+                       uint32_t s = s1 | (s1 << 8) | (s2 << 16) | (s3 << 24);
+                       dst[i * out_channels + j] = int(s) * (1.0f / 4294967296.0f);
+               }
+               src += 3 * (in_channels - out_channels);
+       }
+}
+
+}  // namespace
+
 Mixer::Mixer(const QSurfaceFormat &format)
-       : mixer_surface(create_surface(format)),
+       : httpd("test.ts", WIDTH, HEIGHT),
+         mixer_surface(create_surface(format)),
          h264_encoder_surface(create_surface(format))
 {
+       httpd.start(9095);
+
        CHECK(init_movit(MOVIT_SHADER_DIR, MOVIT_DEBUG_OFF));
        check_error();
 
@@ -77,26 +100,30 @@ Mixer::Mixer(const QSurfaceFormat &format)
        display_chain->set_dither_bits(0);  // Don't bother.
        display_chain->finalize();
 
-       h264_encoder.reset(new H264Encoder(h264_encoder_surface, WIDTH, HEIGHT, "test.mp4"));
-
-       printf("Configuring first card...\n");
-       cards[0].usb = new BMUSBCapture(0x1edb, 0xbd3b);  // 0xbd4f
-       cards[0].usb->set_frame_callback(std::bind(&Mixer::bm_frame, this, 0, _1, _2, _3, _4, _5, _6, _7));
-       cards[0].frame_allocator.reset(new PBOFrameAllocator(1280 * 750 * 2 + 44, 1280, 720));
-       cards[0].usb->set_video_frame_allocator(cards[0].frame_allocator.get());
-       cards[0].usb->configure_card();
-       cards[0].surface = create_surface(format);
-#if NUM_CARDS == 2
-       cards[1].surface = create_surface(format);
-#endif
+       h264_encoder.reset(new H264Encoder(h264_encoder_surface, WIDTH, HEIGHT, &httpd));
 
-       if (NUM_CARDS == 2) {
-               printf("Configuring second card...\n");
-               cards[1].usb = new BMUSBCapture(0x1edb, 0xbd4f);
-               cards[1].usb->set_frame_callback(std::bind(&Mixer::bm_frame, this, 1, _1, _2, _3, _4, _5, _6, _7));
-               cards[1].frame_allocator.reset(new PBOFrameAllocator(1280 * 750 * 2 + 44, 1280, 720));
-               cards[1].usb->set_video_frame_allocator(cards[1].frame_allocator.get());
-               cards[1].usb->configure_card();
+       for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
+               printf("Configuring card %d...\n", card_index);
+               CaptureCard *card = &cards[card_index];
+               card->usb = new BMUSBCapture(card_index);
+               card->usb->set_frame_callback(bind(&Mixer::bm_frame, this, card_index, _1, _2, _3, _4, _5, _6, _7));
+               card->frame_allocator.reset(new PBOFrameAllocator(WIDTH * (HEIGHT+EXTRAHEIGHT) * 2 + 44, WIDTH, HEIGHT));
+               card->usb->set_video_frame_allocator(card->frame_allocator.get());
+               card->surface = create_surface(format);
+               card->usb->set_dequeue_thread_callbacks(
+                       [card]{
+                               eglBindAPI(EGL_OPENGL_API);
+                               card->context = create_context();
+                               if (!make_current(card->context, card->surface)) {
+                                       printf("failed to create bmusb context\n");
+                                       exit(1);
+                               }
+                       },
+                       [this]{
+                               resource_pool->clean_context();
+                       });
+               card->resampler.reset(new Resampler(48000.0, 48000.0, 2));
+               card->usb->configure_card();
        }
 
        BMUSBCapture::start_bm_thread();
@@ -125,39 +152,120 @@ Mixer::~Mixer()
 {
        resource_pool->release_glsl_program(cbcr_program_num);
        BMUSBCapture::stop_bm_thread();
+
+       for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
+               {
+                       unique_lock<mutex> lock(bmusb_mutex);
+                       cards[card_index].should_quit = true;  // Unblock thread.
+                       cards[card_index].new_data_ready_changed.notify_all();
+               }
+               cards[card_index].usb->stop_dequeue_thread();
+       }
 }
 
+namespace {
+
+int unwrap_timecode(uint16_t current_wrapped, int last)
+{
+       uint16_t last_wrapped = last & 0xffff;
+       if (current_wrapped > last_wrapped) {
+               return (last & ~0xffff) | current_wrapped;
+       } else {
+               return 0x10000 + ((last & ~0xffff) | current_wrapped);
+       }
+}
+
+}  // namespace
+
 void Mixer::bm_frame(int card_index, uint16_t timecode,
                      FrameAllocator::Frame video_frame, size_t video_offset, uint16_t video_format,
                     FrameAllocator::Frame audio_frame, size_t audio_offset, uint16_t audio_format)
 {
        CaptureCard *card = &cards[card_index];
-       if (!card->thread_initialized) {
-               printf("initializing context for bmusb thread %d\n", card_index);
-               eglBindAPI(EGL_OPENGL_API);
-               card->context = create_context();
-               if (!make_current(card->context, card->surface)) {
-                       printf("failed to create bmusb context\n");
-                       exit(1);
+
+       if (audio_frame.len - audio_offset > 30000) {
+               printf("Card %d: Dropping frame with implausible audio length (len=%d, offset=%d) [timecode=0x%04x video_len=%d video_offset=%d video_format=%x)\n",
+                       card_index, int(audio_frame.len), int(audio_offset),
+                       timecode, int(video_frame.len), int(video_offset), video_format);
+               if (video_frame.owner) {
+                       video_frame.owner->release_frame(video_frame);
+               }
+               if (audio_frame.owner) {
+                       audio_frame.owner->release_frame(audio_frame);
                }
-               card->thread_initialized = true;
-       }       
-
-       if (video_frame.len - video_offset != 1280 * 750 * 2) {
-               printf("dropping frame with wrong length (%ld)\n", video_frame.len - video_offset);
-               FILE *fp = fopen("frame.raw", "wb");
-               fwrite(video_frame.data, video_frame.len, 1, fp);
-               fclose(fp);
-               //exit(1);
-               card->usb->get_video_frame_allocator()->release_frame(video_frame);
-               card->usb->get_audio_frame_allocator()->release_frame(audio_frame);
                return;
        }
+
+       // Convert the audio to stereo fp32 and add it.
+       size_t num_samples = (audio_frame.len - audio_offset) / 8 / 3;
+       vector<float> audio;
+       audio.resize(num_samples * 2);
+       convert_fixed24_to_fp32(&audio[0], 2, audio_frame.data + audio_offset, 8, num_samples);
+
+       int unwrapped_timecode = timecode;
+       int dropped_frames = 0;
+       if (card->last_timecode != -1) {
+               unwrapped_timecode = unwrap_timecode(unwrapped_timecode, card->last_timecode);
+               dropped_frames = unwrapped_timecode - card->last_timecode - 1;
+       }
+       card->last_timecode = unwrapped_timecode;
+
+       // Add the audio.
+       {
+               unique_lock<mutex> lock(card->audio_mutex);
+
+               int unwrapped_timecode = timecode;
+               if (dropped_frames > 60 * 2) {
+                       fprintf(stderr, "Card %d lost more than two seconds (or time code jumping around), resetting resampler\n",
+                               card_index);
+                       card->resampler.reset(new Resampler(48000.0, 48000.0, 2));
+               } else if (dropped_frames > 0) {
+                       // Insert silence as needed.
+                       fprintf(stderr, "Card %d dropped %d frame(s) (before timecode 0x%04x), inserting silence.\n",
+                               card_index, dropped_frames, timecode);
+                       vector<float> silence;
+                       silence.resize((48000 / 60) * 2);
+                       for (int i = 0; i < dropped_frames; ++i) {
+                               card->resampler->add_input_samples((unwrapped_timecode - dropped_frames + i) / 60.0, silence.data(), (48000 / 60));
+                       }
+               }
+               card->resampler->add_input_samples(unwrapped_timecode / 60.0, audio.data(), num_samples);
+       }
+
+       // Done with the audio, so release it.
+       if (audio_frame.owner) {
+               audio_frame.owner->release_frame(audio_frame);
+       }
+
        {
                // Wait until the previous frame was consumed.
-               std::unique_lock<std::mutex> lock(bmusb_mutex);
-               card->new_data_ready_changed.wait(lock, [card]{ return !card->new_data_ready; });
+               unique_lock<mutex> lock(bmusb_mutex);
+               card->new_data_ready_changed.wait(lock, [card]{ return !card->new_data_ready || card->should_quit; });
+               if (card->should_quit) return;
+       }
+
+       if (video_frame.len - video_offset != WIDTH * (HEIGHT+EXTRAHEIGHT) * 2) {
+               if (video_frame.len != 0) {
+                       printf("Card %d: Dropping video frame with wrong length (%ld)\n",
+                               card_index, video_frame.len - video_offset);
+               }
+               if (video_frame.owner) {
+                       video_frame.owner->release_frame(video_frame);
+               }
+
+               // Still send on the information that we _had_ a frame, even though it's corrupted,
+               // so that pts can go up accordingly.
+               {
+                       unique_lock<mutex> lock(bmusb_mutex);
+                       card->new_data_ready = true;
+                       card->new_frame = RefCountedFrame(FrameAllocator::Frame());
+                       card->new_data_ready_fence = nullptr;
+                       card->dropped_frames = dropped_frames;
+                       card->new_data_ready_changed.notify_all();
+               }
+               return;
        }
+
        const PBOFrameAllocator::Userdata *userdata = (const PBOFrameAllocator::Userdata *)video_frame.userdata;
        GLuint pbo = userdata->pbo;
        check_error();
@@ -171,27 +279,26 @@ void Mixer::bm_frame(int card_index, uint16_t timecode,
        // Upload the textures.
        glBindTexture(GL_TEXTURE_2D, userdata->tex_y);
        check_error();
-       glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, 1280, 720, GL_RED, GL_UNSIGNED_BYTE, BUFFER_OFFSET((1280 * 750 * 2 + 44) / 2 + 1280 * 25 + 22));
+       glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, WIDTH, HEIGHT, GL_RED, GL_UNSIGNED_BYTE, BUFFER_OFFSET((WIDTH * (HEIGHT+EXTRAHEIGHT) * 2 + 44) / 2 + WIDTH * 25 + 22));
        check_error();
        glBindTexture(GL_TEXTURE_2D, userdata->tex_cbcr);
        check_error();
-       glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, 1280/2, 720, GL_RG, GL_UNSIGNED_BYTE, BUFFER_OFFSET(1280 * 25 + 22));
+       glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, WIDTH/2, HEIGHT, GL_RG, GL_UNSIGNED_BYTE, BUFFER_OFFSET(WIDTH * 25 + 22));
        check_error();
        glBindTexture(GL_TEXTURE_2D, 0);
        check_error();
        GLsync fence = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);              
        check_error();
        assert(fence != nullptr);
+
        {
-               std::unique_lock<std::mutex> lock(bmusb_mutex);
+               unique_lock<mutex> lock(bmusb_mutex);
                card->new_data_ready = true;
                card->new_frame = RefCountedFrame(video_frame);
                card->new_data_ready_fence = fence;
+               card->dropped_frames = dropped_frames;
                card->new_data_ready_changed.notify_all();
        }
-
-       // Video frame will be released when last user of card->new_frame goes out of scope.
-        card->usb->get_audio_frame_allocator()->release_frame(audio_frame);
 }
 
 void Mixer::thread_func()
@@ -206,13 +313,14 @@ void Mixer::thread_func()
        struct timespec start, now;
        clock_gettime(CLOCK_MONOTONIC, &start);
 
-       while (!should_quit) {
-               ++frame;
+       int frame = 0;
+       int dropped_frames = 0;
 
+       while (!should_quit) {
                CaptureCard card_copy[NUM_CARDS];
 
                {
-                       std::unique_lock<std::mutex> lock(bmusb_mutex);
+                       unique_lock<mutex> lock(bmusb_mutex);
 
                        // The first card is the master timer, so wait for it to have a new frame.
                        // TODO: Make configurable, and with a timeout.
@@ -224,14 +332,47 @@ void Mixer::thread_func()
                                card_copy[card_index].new_data_ready = card->new_data_ready;
                                card_copy[card_index].new_frame = card->new_frame;
                                card_copy[card_index].new_data_ready_fence = card->new_data_ready_fence;
+                               card_copy[card_index].new_frame_audio = move(card->new_frame_audio);
+                               card_copy[card_index].dropped_frames = card->dropped_frames;
                                card->new_data_ready = false;
                                card->new_data_ready_changed.notify_all();
                        }
                }
 
+               // Resample the audio as needed, including from previously dropped frames.
+               vector<float> samples_out;
+               // TODO: Allow using audio from the other card(s) as well.
+               for (unsigned frame_num = 0; frame_num < card_copy[0].dropped_frames + 1; ++frame_num) {
+                       for (unsigned card_index = 0; card_index < NUM_CARDS; ++card_index) {
+                               samples_out.resize((48000 / 60) * 2);
+                               {
+                                       unique_lock<mutex> lock(cards[card_index].audio_mutex);
+                                       if (!cards[card_index].resampler->get_output_samples(pts(), &samples_out[0], 48000 / 60)) {
+                                               printf("Card %d reported previous underrun.\n", card_index);
+                                       }
+                               }
+                               if (card_index == 0) {
+                                       h264_encoder->add_audio(pts_int, move(samples_out));
+                               }
+                       }
+                       if (frame_num != card_copy[0].dropped_frames) {
+                               // For dropped frames, increase the pts.
+                               ++dropped_frames;
+                               pts_int += TIMEBASE / 60;
+                       }
+               }
+
+               // If the first card is reporting a corrupted or otherwise dropped frame,
+               // just increase the pts (skipping over this frame) and don't try to compute anything new.
+               if (card_copy[0].new_frame->len == 0) {
+                       ++dropped_frames;
+                       pts_int += TIMEBASE / 60;
+                       continue;
+               }
+
                for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
                        CaptureCard *card = &card_copy[card_index];
-                       if (!card->new_data_ready)
+                       if (!card->new_data_ready || card->new_frame->len == 0)
                                continue;
 
                        assert(card->new_frame != nullptr);
@@ -240,17 +381,18 @@ void Mixer::thread_func()
 
                        // The new texture might still be uploaded,
                        // tell the GPU to wait until it's there.
-                       if (card->new_data_ready_fence)
+                       if (card->new_data_ready_fence) {
                                glWaitSync(card->new_data_ready_fence, /*flags=*/0, GL_TIMEOUT_IGNORED);
-                       check_error();
-                       glDeleteSync(card->new_data_ready_fence);
-                       check_error();
+                               check_error();
+                               glDeleteSync(card->new_data_ready_fence);
+                               check_error();
+                       }
                        const PBOFrameAllocator::Userdata *userdata = (const PBOFrameAllocator::Userdata *)card->new_frame->userdata;
                        theme->set_input_textures(card_index, userdata->tex_y, userdata->tex_cbcr);
                }
 
                // Get the main chain from the theme, and set its state immediately.
-               pair<EffectChain *, function<void()>> theme_main_chain = theme->get_chain(0, frame / 60.0f, WIDTH, HEIGHT);
+               pair<EffectChain *, function<void()>> theme_main_chain = theme->get_chain(0, pts(), WIDTH, HEIGHT);
                EffectChain *chain = theme_main_chain.first;
                theme_main_chain.second();
 
@@ -285,7 +427,10 @@ void Mixer::thread_func()
                for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
                        input_frames.push_back(bmusb_current_rendering_frame[card_index]);
                }
-               h264_encoder->end_frame(fence, input_frames);
+               const int64_t av_delay = TIMEBASE / 10;  // Corresponds to the fixed delay in resampler.h. TODO: Make less hard-coded.
+               h264_encoder->end_frame(fence, pts_int + av_delay, input_frames);
+               ++frame;
+               pts_int += TIMEBASE / 60;
 
                // The live frame just shows the RGBA texture we just rendered.
                // It owns rgba_tex now.
@@ -300,9 +445,9 @@ void Mixer::thread_func()
                output_channel[OUTPUT_LIVE].output_frame(live_frame);
 
                // Set up preview and any additional channels.
-               for (unsigned i = 1; i < theme->get_num_channels() + 2; ++i) {
+               for (int i = 1; i < theme->get_num_channels() + 2; ++i) {
                        DisplayFrame display_frame;
-                       pair<EffectChain *, function<void()>> chain = theme->get_chain(i, frame / 60.0f, WIDTH, HEIGHT);  // FIXME: dimensions
+                       pair<EffectChain *, function<void()>> chain = theme->get_chain(i, pts(), WIDTH, HEIGHT);  // FIXME: dimensions
                        display_frame.chain = chain.first;
                        display_frame.setup_chain = chain.second;
                        display_frame.ready_fence = fence;
@@ -315,12 +460,13 @@ void Mixer::thread_func()
                double elapsed = now.tv_sec - start.tv_sec +
                        1e-9 * (now.tv_nsec - start.tv_nsec);
                if (frame % 100 == 0) {
-                       printf("%d frames in %.3f seconds = %.1f fps (%.1f ms/frame)\n",
-                               frame, elapsed, frame / elapsed,
+                       printf("%d frames (%d dropped) in %.3f seconds = %.1f fps (%.1f ms/frame)\n",
+                               frame, dropped_frames, elapsed, frame / elapsed,
                                1e3 * elapsed / frame);
                //      chain->print_phase_timing();
                }
 
+#if 0
                // Reset every 100 frames, so that local variations in frame times
                // (especially for the first few frames, when the shaders are
                // compiled etc.) don't make it hard to measure for the entire
@@ -329,8 +475,11 @@ void Mixer::thread_func()
                        frame = 0;
                        start = now;
                }
+#endif
                check_error();
        }
+
+       resource_pool->clean_context();
 }
 
 void Mixer::subsample_chroma(GLuint src_tex, GLuint dst_tex)
@@ -399,7 +548,7 @@ void Mixer::release_display_frame(DisplayFrame *frame)
 
 void Mixer::start()
 {
-       mixer_thread = std::thread(&Mixer::thread_func, this);
+       mixer_thread = thread(&Mixer::thread_func, this);
 }
 
 void Mixer::quit()
@@ -410,7 +559,12 @@ void Mixer::quit()
 
 void Mixer::transition_clicked(int transition_num)
 {
-       theme->transition_clicked(transition_num, frame / 60.0);
+       theme->transition_clicked(transition_num, pts());
+}
+
+void Mixer::channel_clicked(int preview_num)
+{
+       theme->channel_clicked(preview_num);
 }
 
 Mixer::OutputChannel::~OutputChannel()
@@ -428,7 +582,7 @@ void Mixer::OutputChannel::output_frame(DisplayFrame frame)
        // Store this frame for display. Remove the ready frame if any
        // (it was seemingly never used).
        {
-               std::unique_lock<std::mutex> lock(frame_mutex);
+               unique_lock<mutex> lock(frame_mutex);
                if (has_ready_frame) {
                        parent->release_display_frame(&ready_frame);
                }
@@ -443,7 +597,7 @@ void Mixer::OutputChannel::output_frame(DisplayFrame frame)
 
 bool Mixer::OutputChannel::get_display_frame(DisplayFrame *frame)
 {
-       std::unique_lock<std::mutex> lock(frame_mutex);
+       unique_lock<mutex> lock(frame_mutex);
        if (!has_current_frame && !has_ready_frame) {
                return false;
        }