X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=mixer.cpp;h=43e39b6c18e8518b8b0ecee7df91ebf0d2a0b269;hb=d38817a7a12efac0153bf1746d3c3beea3d29d5d;hp=1ac5d53ee03e40484194f8f3039c10233f547f12;hpb=2c84fdec878d1144b2d615dbfb901d23e838f457;p=nageru diff --git a/mixer.cpp b/mixer.cpp index 1ac5d53..43e39b6 100644 --- a/mixer.cpp +++ b/mixer.cpp @@ -1,32 +1,26 @@ #define WIDTH 1280 #define HEIGHT 720 +#define EXTRAHEIGHT 30 #undef Success #include "mixer.h" #include -#include -#include -#include #include -#include -#include #include -#include -#include -#include -#include -#include +#include +#include +#include +#include +#include #include #include #include #include #include #include -#include -#include -#include +#include #include #include #include @@ -34,13 +28,15 @@ #include #include #include +#include #include -#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; @@ -50,64 +46,50 @@ using namespace std::placeholders; Mixer *global_mixer = nullptr; -Mixer::Mixer(const QSurfaceFormat &format) - : mixer_surface(create_surface(format)), - h264_encoder_surface(create_surface(format)) +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, unsigned num_cards) + : httpd("test.ts", WIDTH, HEIGHT), + num_cards(num_cards), + mixer_surface(create_surface(format)), + h264_encoder_surface(create_surface(format)), + compressor(48000.0f) { + httpd.start(9095); + CHECK(init_movit(MOVIT_SHADER_DIR, MOVIT_DEBUG_OFF)); check_error(); + // Since we allow non-bouncing 4:2:2 YCbCrInputs, effective subpixel precision + // will be halved when sampling them, and we need to compensate here. + movit_texel_subpixel_precision /= 2.0; + resource_pool.reset(new ResourcePool); - output_channel[OUTPUT_LIVE].parent = this; - output_channel[OUTPUT_PREVIEW].parent = this; + theme.reset(new Theme("theme.lua", resource_pool.get(), num_cards)); + for (unsigned i = 0; i < NUM_OUTPUTS; ++i) { + output_channel[i].parent = this; + } ImageFormat inout_format; inout_format.color_space = COLORSPACE_sRGB; inout_format.gamma_curve = GAMMA_sRGB; - YCbCrFormat input_ycbcr_format; - input_ycbcr_format.chroma_subsampling_x = 2; - input_ycbcr_format.chroma_subsampling_y = 1; - input_ycbcr_format.cb_x_position = 0.0; - input_ycbcr_format.cr_x_position = 0.0; - input_ycbcr_format.cb_y_position = 0.5; - input_ycbcr_format.cr_y_position = 0.5; - input_ycbcr_format.luma_coefficients = YCBCR_REC_601; - input_ycbcr_format.full_range = false; - - YCbCrFormat output_ycbcr_format; - output_ycbcr_format.chroma_subsampling_x = 1; - output_ycbcr_format.chroma_subsampling_y = 1; - output_ycbcr_format.luma_coefficients = YCBCR_REC_601; - output_ycbcr_format.full_range = false; - - // Main chain. - chain.reset(new EffectChain(WIDTH, HEIGHT, resource_pool.get())); - check_error(); - input[0] = new YCbCrInput(inout_format, input_ycbcr_format, WIDTH, HEIGHT, YCBCR_INPUT_SPLIT_Y_AND_CBCR); - chain->add_input(input[0]); - input[1] = new YCbCrInput(inout_format, input_ycbcr_format, WIDTH, HEIGHT, YCBCR_INPUT_SPLIT_Y_AND_CBCR); - chain->add_input(input[1]); - resample_effect = chain->add_effect(new ResampleEffect(), input[0]); - padding_effect = chain->add_effect(new IntegralPaddingEffect()); - float border_color[] = { 0.0f, 0.0f, 0.0f, 1.0f }; - CHECK(padding_effect->set_vec4("border_color", border_color)); - - resample2_effect = chain->add_effect(new ResampleEffect(), input[1]); - Effect *saturation_effect = chain->add_effect(new SaturationEffect()); - CHECK(saturation_effect->set_float("saturation", 0.3f)); - Effect *wb_effect = chain->add_effect(new WhiteBalanceEffect()); - CHECK(wb_effect->set_float("output_color_temperature", 3500.0)); - padding2_effect = chain->add_effect(new IntegralPaddingEffect()); - - chain->add_effect(new OverlayEffect(), padding_effect, padding2_effect); - - chain->add_output(inout_format, OUTPUT_ALPHA_FORMAT_POSTMULTIPLIED); - chain->add_ycbcr_output(inout_format, OUTPUT_ALPHA_FORMAT_POSTMULTIPLIED, output_ycbcr_format, YCBCR_OUTPUT_SPLIT_Y_AND_CBCR); - chain->set_dither_bits(8); - chain->set_output_origin(OUTPUT_ORIGIN_TOP_LEFT); - chain->finalize(); - // Display chain; shows the live output produced by the main chain (its RGBA version). display_chain.reset(new EffectChain(WIDTH, HEIGHT, resource_pool.get())); check_error(); @@ -117,43 +99,36 @@ Mixer::Mixer(const QSurfaceFormat &format) display_chain->set_dither_bits(0); // Don't bother. display_chain->finalize(); - // Preview chain (always shows just first input for now). - preview_chain.reset(new EffectChain(WIDTH, HEIGHT, resource_pool.get())); - check_error(); - preview_input = new YCbCrInput(inout_format, input_ycbcr_format, WIDTH, HEIGHT, YCBCR_INPUT_SPLIT_Y_AND_CBCR); - preview_chain->add_input(preview_input); - preview_chain->add_output(inout_format, OUTPUT_ALPHA_FORMAT_POSTMULTIPLIED); - preview_chain->set_dither_bits(0); // Don't bother. - preview_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 - - 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(); + h264_encoder.reset(new H264Encoder(h264_encoder_surface, WIDTH, HEIGHT, &httpd)); + + for (unsigned 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(card->surface); + 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(); - for (int card_index = 0; card_index < NUM_CARDS; ++card_index) { + for (unsigned card_index = 0; card_index < num_cards; ++card_index) { cards[card_index].usb->start_bm_capture(); - input[card_index]->set_pixel_data(0, nullptr, 0); - input[card_index]->set_pixel_data(1, nullptr, 0); } //chain->enable_phase_timing(true); @@ -170,45 +145,153 @@ Mixer::Mixer(const QSurfaceFormat &format) " gl_FragColor = texture2D(cbcr_tex, tc0); \n" "} \n"; cbcr_program_num = resource_pool->compile_glsl_program(cbcr_vert_shader, cbcr_frag_shader); + + r128.init(2, 48000); + r128.integr_start(); } Mixer::~Mixer() { resource_pool->release_glsl_program(cbcr_program_num); BMUSBCapture::stop_bm_thread(); + + for (unsigned card_index = 0; card_index < num_cards; ++card_index) { + { + unique_lock 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); + } +} + +float find_peak(const vector &samples) +{ + float m = fabs(samples[0]); + for (size_t i = 1; i < samples.size(); ++i) { + m = std::max(m, fabs(samples[i])); + } + return m; +} + +void deinterleave_samples(const vector &in, vector *out_l, vector *out_r) +{ + size_t num_samples = in.size() / 2; + out_l->resize(num_samples); + out_r->resize(num_samples); + + const float *inptr = in.data(); + float *lptr = &(*out_l)[0]; + float *rptr = &(*out_r)[0]; + for (size_t i = 0; i < num_samples; ++i) { + *lptr++ = *inptr++; + *rptr++ = *inptr++; + } } -void Mixer::bm_frame(int card_index, uint16_t timecode, +} // namespace + +void Mixer::bm_frame(unsigned 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; } + + 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; + + // Convert the audio to stereo fp32 and add it. + size_t num_samples = (audio_frame.len >= audio_offset) ? (audio_frame.len - audio_offset) / 8 / 3 : 0; + vector audio; + audio.resize(num_samples * 2); + convert_fixed24_to_fp32(&audio[0], 2, audio_frame.data + audio_offset, 8, num_samples); + + // Add the audio. + { + unique_lock 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 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 lock(bmusb_mutex); - card->new_data_ready_changed.wait(lock, [card]{ return !card->new_data_ready; }); + unique_lock 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 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(); @@ -222,95 +305,32 @@ 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 lock(bmusb_mutex); + unique_lock 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::place_rectangle(Effect *resample_effect, Effect *padding_effect, float x0, float y0, float x1, float y1) -{ - float srcx0 = 0.0f; - float srcx1 = 1.0f; - float srcy0 = 0.0f; - float srcy1 = 1.0f; - - // Cull. - if (x0 > 1280.0 || x1 < 0.0 || y0 > 720.0 || y1 < 0.0) { - CHECK(resample_effect->set_int("width", 1)); - CHECK(resample_effect->set_int("height", 1)); - CHECK(resample_effect->set_float("zoom_x", 1280.0)); - CHECK(resample_effect->set_float("zoom_y", 720.0)); - CHECK(padding_effect->set_int("left", 2000)); - CHECK(padding_effect->set_int("top", 2000)); - return; - } - // Clip. (TODO: Clip on upper/left sides, too.) - if (x1 > 1280.0) { - srcx1 = (1280.0 - x0) / (x1 - x0); - x1 = 1280.0; - } - if (y1 > 720.0) { - srcy1 = (720.0 - y0) / (y1 - y0); - y1 = 720.0; - } - - float x_subpixel_offset = x0 - floor(x0); - float y_subpixel_offset = y0 - floor(y0); - - // Resampling must be to an integral number of pixels. Round up, - // and then add an extra pixel so we have some leeway for the border. - int width = int(ceil(x1 - x0)) + 1; - int height = int(ceil(y1 - y0)) + 1; - CHECK(resample_effect->set_int("width", width)); - CHECK(resample_effect->set_int("height", height)); - - // Correct the discrepancy with zoom. (This will leave a small - // excess edge of pixels and subpixels, which we'll correct for soon.) - float zoom_x = (x1 - x0) / (width * (srcx1 - srcx0)); - float zoom_y = (y1 - y0) / (height * (srcy1 - srcy0)); - CHECK(resample_effect->set_float("zoom_x", zoom_x)); - CHECK(resample_effect->set_float("zoom_y", zoom_y)); - CHECK(resample_effect->set_float("zoom_center_x", 0.0f)); - CHECK(resample_effect->set_float("zoom_center_y", 0.0f)); - - // Padding must also be to a whole-pixel offset. - CHECK(padding_effect->set_int("left", floor(x0))); - CHECK(padding_effect->set_int("top", floor(y0))); - - // Correct _that_ discrepancy by subpixel offset in the resampling. - CHECK(resample_effect->set_float("left", -x_subpixel_offset / zoom_x)); - CHECK(resample_effect->set_float("top", -y_subpixel_offset / zoom_y)); - - // Finally, adjust the border so it is exactly where we want it. - CHECK(padding_effect->set_float("border_offset_left", x_subpixel_offset)); - CHECK(padding_effect->set_float("border_offset_right", x1 - (floor(x0) + width))); - CHECK(padding_effect->set_float("border_offset_top", y_subpixel_offset)); - CHECK(padding_effect->set_float("border_offset_bottom", y1 - (floor(y0) + height))); -} - void Mixer::thread_func() { eglBindAPI(EGL_OPENGL_API); - QOpenGLContext *context = create_context(); + QOpenGLContext *context = create_context(mixer_surface); if (!make_current(context, mixer_surface)) { printf("oops\n"); exit(1); @@ -319,114 +339,95 @@ void Mixer::thread_func() struct timespec start, now; clock_gettime(CLOCK_MONOTONIC, &start); - while (!should_quit) { - ++frame; - - //int width0 = lrintf(848 * (1.0 + 0.2 * sin(frame * 0.02))); - int width0 = 848; - int height0 = lrintf(width0 * 9.0 / 16.0); - - //float top0 = 96 + 48 * sin(frame * 0.005); - //float left0 = 96 + 48 * cos(frame * 0.006); - float top0 = 48; - float left0 = 16; - float bottom0 = top0 + height0; - float right0 = left0 + width0; - - int width1 = 384; - int height1 = 216; - - float bottom1 = 720 - 48; - float right1 = 1280 - 16; - float top1 = bottom1 - height1; - float left1 = right1 - width1; - - if (current_source == SOURCE_INPUT1) { - top0 = 0.0; - bottom0 = HEIGHT; - left0 = 0.0; - right0 = WIDTH; - - top1 = HEIGHT + 10; - bottom1 = HEIGHT + 20; - left1 = WIDTH + 10; - right1 = WIDTH + 20; - } else if (current_source == SOURCE_INPUT2) { - top1 = 0.0; - bottom1 = HEIGHT; - left1 = 0.0; - right1 = WIDTH; - - top0 = HEIGHT + 10; - bottom0 = HEIGHT + 20; - left0 = WIDTH + 10; - right0 = WIDTH + 20; - } else { - float t = 0.5 + 0.5 * cos(frame * 0.006); - float scale0 = 1.0 + t * (1280.0 / 848.0 - 1.0); - float tx0 = 0.0 + t * (-16.0 * scale0); - float ty0 = 0.0 + t * (-48.0 * scale0); - - top0 = top0 * scale0 + ty0; - bottom0 = bottom0 * scale0 + ty0; - left0 = left0 * scale0 + tx0; - right0 = right0 * scale0 + tx0; - - top1 = top1 * scale0 + ty0; - bottom1 = bottom1 * scale0 + ty0; - left1 = left1 * scale0 + tx0; - right1 = right1 * scale0 + tx0; - } + int frame = 0; + int dropped_frames = 0; - place_rectangle(resample_effect, padding_effect, left0, top0, right0, bottom0); - place_rectangle(resample2_effect, padding2_effect, left1, top1, right1, bottom1); - - CaptureCard card_copy[NUM_CARDS]; + while (!should_quit) { + CaptureCard card_copy[MAX_CARDS]; { - std::unique_lock lock(bmusb_mutex); + unique_lock 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. cards[0].new_data_ready_changed.wait(lock, [this]{ return cards[0].new_data_ready; }); - for (int card_index = 0; card_index < NUM_CARDS; ++card_index) { + for (unsigned card_index = 0; card_index < num_cards; ++card_index) { CaptureCard *card = &cards[card_index]; card_copy[card_index].usb = card->usb; 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].dropped_frames = card->dropped_frames; card->new_data_ready = false; card->new_data_ready_changed.notify_all(); } } - for (int card_index = 0; card_index < NUM_CARDS; ++card_index) { + // Resample the audio as needed, including from previously dropped frames. + for (unsigned frame_num = 0; frame_num < card_copy[0].dropped_frames + 1; ++frame_num) { + process_audio_one_frame(); + if (frame_num != card_copy[0].dropped_frames) { + // For dropped frames, increase the pts. + ++dropped_frames; + pts_int += TIMEBASE / 60; + } + } + + if (audio_level_callback != nullptr) { + double loudness_s = r128.loudness_S(); + double loudness_i = r128.integrated(); + double loudness_range_low = r128.range_min(); + double loudness_range_high = r128.range_max(); + + audio_level_callback(loudness_s, 20.0 * log10(peak), + loudness_i, loudness_range_low, loudness_range_high); + } + + for (unsigned card_index = 1; card_index < num_cards; ++card_index) { + if (card_copy[card_index].new_data_ready && card_copy[card_index].new_frame->len == 0) { + ++card_copy[card_index].dropped_frames; + } + if (card_copy[card_index].dropped_frames > 0) { + printf("Card %u dropped %d frames before this\n", + card_index, int(card_copy[card_index].dropped_frames)); + } + } + + // 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 (unsigned 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); bmusb_current_rendering_frame[card_index] = card->new_frame; check_error(); // 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(); - const PBOFrameAllocator::Userdata *userdata = (const PBOFrameAllocator::Userdata *)card->new_frame->userdata; - input[card_index]->set_texture_num(0, userdata->tex_y); - input[card_index]->set_texture_num(1, userdata->tex_cbcr); - - if (NUM_CARDS == 1) { - // Set to the other one, too. - input[1]->set_texture_num(0, userdata->tex_y); - input[1]->set_texture_num(1, userdata->tex_cbcr); + 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> theme_main_chain = theme->get_chain(0, pts(), WIDTH, HEIGHT); + EffectChain *chain = theme_main_chain.first; + theme_main_chain.second(); + GLuint y_tex, cbcr_tex; bool got_frame = h264_encoder->begin_frame(&y_tex, &cbcr_tex); assert(got_frame); @@ -435,6 +436,7 @@ void Mixer::thread_func() GLuint cbcr_full_tex = resource_pool->create_2d_texture(GL_RG8, WIDTH, HEIGHT); GLuint rgba_tex = resource_pool->create_2d_texture(GL_RGB565, WIDTH, HEIGHT); // Saves texture bandwidth, although dithering gets messed up. GLuint fbo = resource_pool->create_fbo(y_tex, cbcr_full_tex, rgba_tex); + check_error(); chain->render_to_fbo(fbo, WIDTH, HEIGHT); resource_pool->release_fbo(fbo); @@ -455,10 +457,13 @@ void Mixer::thread_func() // input frames needed, so that they are not released back // until the rendering is done. vector input_frames; - for (int card_index = 0; card_index < NUM_CARDS; ++card_index) { + for (unsigned 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. @@ -472,32 +477,33 @@ void Mixer::thread_func() live_frame.temp_textures = { rgba_tex }; output_channel[OUTPUT_LIVE].output_frame(live_frame); - // The preview frame shows the first input. Note that the textures - // are owned by the input frame, not the display frame. - const PBOFrameAllocator::Userdata *input0_userdata = (const PBOFrameAllocator::Userdata *)bmusb_current_rendering_frame[0]->userdata; - GLuint input0_y_tex = input0_userdata->tex_y; - GLuint input0_cbcr_tex = input0_userdata->tex_cbcr; - DisplayFrame preview_frame; - preview_frame.chain = preview_chain.get(); - preview_frame.setup_chain = [this, input0_y_tex, input0_cbcr_tex]{ - preview_input->set_texture_num(0, input0_y_tex); - preview_input->set_texture_num(1, input0_cbcr_tex); - }; - preview_frame.ready_fence = fence; - preview_frame.input_frames = { bmusb_current_rendering_frame[0] }; - preview_frame.temp_textures = {}; - output_channel[OUTPUT_PREVIEW].output_frame(preview_frame); + // Set up preview and any additional channels. + for (int i = 1; i < theme->get_num_channels() + 2; ++i) { + DisplayFrame display_frame; + pair> 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; + + // FIXME: possible to do better? + for (unsigned card_index = 0; card_index < num_cards; ++card_index) { + display_frame.input_frames.push_back(bmusb_current_rendering_frame[card_index]); + } + display_frame.temp_textures = {}; + output_channel[i].output_frame(display_frame); + } clock_gettime(CLOCK_MONOTONIC, &now); 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 @@ -506,8 +512,67 @@ void Mixer::thread_func() frame = 0; start = now; } +#endif check_error(); } + + resource_pool->clean_context(); +} + +void Mixer::process_audio_one_frame() +{ + vector samples_card; + vector samples_out; + for (unsigned card_index = 0; card_index < num_cards; ++card_index) { + samples_card.resize((48000 / 60) * 2); + { + unique_lock lock(cards[card_index].audio_mutex); + if (!cards[card_index].resampler->get_output_samples(pts(), &samples_card[0], 48000 / 60)) { + printf("Card %d reported previous underrun.\n", card_index); + } + } + // TODO: Allow using audio from the other card(s) as well. + if (card_index == 0) { + samples_out = move(samples_card); + } + } + + // Apply a level compressor to get the general level right. + // Basically, if it's over about -40 dBFS, we squeeze it down to that level + // (or more precisely, near it, since we don't use infinite ratio), + // then apply a makeup gain to get it to -12 dBFS. -12 dBFS is, of course, + // entirely arbitrary, but from practical tests with speech, it seems to + // put ut around -23 LUFS, so it's a reasonable starting point for later use. + // + // TODO: Hook this up to a UI, so we can see the effects, and/or turn it off + // to control the gain manually instead. For now, there's only the #if-ed out + // code below. + // + // TODO: Add the actual compressors/limiters (for taking care of transients) + // later in the chain. + float threshold = 0.01f; // -40 dBFS. + float ratio = 20.0f; + float attack_time = 0.1f; + float release_time = 10.0f; + float makeup_gain = pow(10.0f, 28.0f / 20.0f); // +28 dB takes us to -12 dBFS. + compressor.process(samples_out.data(), samples_out.size() / 2, threshold, ratio, attack_time, release_time, makeup_gain); + +#if 0 + printf("level=%f (%+5.2f dBFS) attenuation=%f (%+5.2f dB) end_result=%+5.2f dB\n", + compressor.get_level(), 20.0 * log10(compressor.get_level()), + compressor.get_attenuation(), 20.0 * log10(compressor.get_attenuation()), + 20.0 * log10(compressor.get_level() * compressor.get_attenuation() * makeup_gain)); +#endif + + // Find peak and R128 levels. + peak = std::max(peak, find_peak(samples_out)); + vector left, right; + deinterleave_samples(samples_out, &left, &right); + float *ptrs[] = { left.data(), right.data() }; + r128.process(left.size(), ptrs); + + // Actually add the samples to the output. + h264_encoder->add_audio(pts_int, move(samples_out)); } void Mixer::subsample_chroma(GLuint src_tex, GLuint dst_tex) @@ -576,7 +641,7 @@ 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() @@ -585,9 +650,24 @@ void Mixer::quit() mixer_thread.join(); } -void Mixer::cut(Source source) +void Mixer::transition_clicked(int transition_num) +{ + theme->transition_clicked(transition_num, pts()); +} + +void Mixer::channel_clicked(int preview_num) { - current_source = source; + theme->channel_clicked(preview_num); +} + +Mixer::OutputChannel::~OutputChannel() +{ + if (has_current_frame) { + parent->release_display_frame(¤t_frame); + } + if (has_ready_frame) { + parent->release_display_frame(&ready_frame); + } } void Mixer::OutputChannel::output_frame(DisplayFrame frame) @@ -595,7 +675,7 @@ 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 lock(frame_mutex); + unique_lock lock(frame_mutex); if (has_ready_frame) { parent->release_display_frame(&ready_frame); } @@ -610,7 +690,7 @@ void Mixer::OutputChannel::output_frame(DisplayFrame frame) bool Mixer::OutputChannel::get_display_frame(DisplayFrame *frame) { - std::unique_lock lock(frame_mutex); + unique_lock lock(frame_mutex); if (!has_current_frame && !has_ready_frame) { return false; } @@ -638,9 +718,3 @@ void Mixer::OutputChannel::set_frame_ready_callback(Mixer::new_frame_ready_callb new_frame_ready_callback = callback; has_new_frame_ready_callback = true; } - -void Mixer::OutputChannel::set_size(int width, int height) -{ - this->width = width; - this->height = height; -}