X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=mixer.cpp;h=e928ed9945319b38eec6585e3b8b11fb3deb8a8d;hb=3585603bd8c584eabf5b4b001f43cbf2c61f1b98;hp=9df2201299b6fb72d142c0741536b83b1dc1c880;hpb=4c29d493f13f39fc6f6b07677a4744b8586a3c1d;p=nageru diff --git a/mixer.cpp b/mixer.cpp index 9df2201..e928ed9 100644 --- a/mixer.cpp +++ b/mixer.cpp @@ -10,12 +10,12 @@ #include #include #include +#include #include #include #include #include #include -#include #include #include #include @@ -59,6 +59,22 @@ void convert_fixed24_to_fp32(float *dst, size_t out_channels, const uint8_t *src } } +void insert_new_frame(RefCountedFrame frame, unsigned field_num, bool interlaced, unsigned card_index, InputState *input_state) +{ + if (interlaced) { + for (unsigned frame_num = FRAME_HISTORY_LENGTH; frame_num --> 1; ) { // :-) + input_state->buffered_frames[card_index][frame_num] = + input_state->buffered_frames[card_index][frame_num - 1]; + } + input_state->buffered_frames[card_index][0] = { frame, field_num }; + } else { + for (unsigned frame_num = 0; frame_num < FRAME_HISTORY_LENGTH; ++frame_num) { + input_state->buffered_frames[card_index][frame_num] = { frame, field_num }; + } + } +} + + } // namespace Mixer::Mixer(const QSurfaceFormat &format, unsigned num_cards) @@ -130,8 +146,6 @@ Mixer::Mixer(const QSurfaceFormat &format, unsigned num_cards) cards[card_index].usb->start_bm_capture(); } - //chain->enable_phase_timing(true); - // Set up stuff for NV12 conversion. // Cb/Cr shader. @@ -143,7 +157,8 @@ Mixer::Mixer(const QSurfaceFormat &format, unsigned num_cards) "void main() { \n" " gl_FragColor = texture2D(cbcr_tex, tc0); \n" "} \n"; - cbcr_program_num = resource_pool->compile_glsl_program(cbcr_vert_shader, cbcr_frag_shader); + vector frag_shader_outputs; + cbcr_program_num = resource_pool->compile_glsl_program(cbcr_vert_shader, cbcr_frag_shader, frag_shader_outputs); r128.init(2, OUTPUT_FREQUENCY); r128.integr_start(); @@ -218,10 +233,10 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, { CaptureCard *card = &cards[card_index]; - unsigned width, height, frame_rate_nom, frame_rate_den, extra_lines_top, extra_lines_bottom; + unsigned width, height, second_field_start, frame_rate_nom, frame_rate_den, extra_lines_top, extra_lines_bottom; bool interlaced; - decode_video_format(video_format, &width, &height, &extra_lines_top, &extra_lines_bottom, + decode_video_format(video_format, &width, &height, &second_field_start, &extra_lines_top, &extra_lines_bottom, &frame_rate_nom, &frame_rate_den, &interlaced); // Ignore return value for now. int64_t frame_length = TIMEBASE * frame_rate_den / frame_rate_nom; @@ -267,8 +282,7 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, // 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(silence_samples * 2); + vector silence(silence_samples * 2, 0.0f); for (int i = 0; i < dropped_frames; ++i) { card->resampling_queue->add_input_samples(local_pts / double(TIMEBASE), silence.data(), silence_samples); // Note that if the format changed in the meantime, we have @@ -316,6 +330,7 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, card->new_data_ready = true; card->new_frame = RefCountedFrame(FrameAllocator::Frame()); card->new_frame_length = frame_length; + card->new_frame_interlaced = false; card->new_data_ready_fence = nullptr; card->dropped_frames = dropped_frames; card->new_data_ready_changed.notify_all(); @@ -325,58 +340,105 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, PBOFrameAllocator::Userdata *userdata = (PBOFrameAllocator::Userdata *)video_frame.userdata; + unsigned num_fields = interlaced ? 2 : 1; + timespec frame_upload_start; + if (interlaced) { + // NOTE: This isn't deinterlacing. This is just sending the two fields along + // as separate frames without considering anything like the half-field offset. + // We'll need to add a proper deinterlacer on the receiving side to get this right. + assert(height % 2 == 0); + height /= 2; + assert(frame_length % 2 == 0); + frame_length /= 2; + num_fields = 2; + clock_gettime(CLOCK_MONOTONIC, &frame_upload_start); + } + RefCountedFrame new_frame(video_frame); + // Upload the textures. size_t cbcr_width = width / 2; size_t cbcr_offset = video_offset / 2; size_t y_offset = video_frame.size / 2 + video_offset / 2; - if (width != userdata->last_width || height != userdata->last_height) { - // We changed resolution since last use of this texture, so we need to create - // a new object. Note that this each card has its own PBOFrameAllocator, - // we don't need to worry about these flip-flopping between resolutions. - glBindTexture(GL_TEXTURE_2D, userdata->tex_cbcr); - check_error(); - glTexImage2D(GL_TEXTURE_2D, 0, GL_RG8, cbcr_width, height, 0, GL_RG, GL_UNSIGNED_BYTE, nullptr); + for (unsigned field = 0; field < num_fields; ++field) { + unsigned field_start_line = (field == 1) ? second_field_start : extra_lines_top + field * (height + 22); + + if (userdata->tex_y[field] == 0 || + userdata->tex_cbcr[field] == 0 || + width != userdata->last_width[field] || + height != userdata->last_height[field]) { + // We changed resolution since last use of this texture, so we need to create + // a new object. Note that this each card has its own PBOFrameAllocator, + // we don't need to worry about these flip-flopping between resolutions. + glBindTexture(GL_TEXTURE_2D, userdata->tex_cbcr[field]); + check_error(); + glTexImage2D(GL_TEXTURE_2D, 0, GL_RG8, cbcr_width, height, 0, GL_RG, GL_UNSIGNED_BYTE, nullptr); + check_error(); + glBindTexture(GL_TEXTURE_2D, userdata->tex_y[field]); + check_error(); + glTexImage2D(GL_TEXTURE_2D, 0, GL_R8, width, height, 0, GL_RED, GL_UNSIGNED_BYTE, nullptr); + check_error(); + userdata->last_width[field] = width; + userdata->last_height[field] = height; + } + + GLuint pbo = userdata->pbo; check_error(); - glBindTexture(GL_TEXTURE_2D, userdata->tex_y); + glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, pbo); check_error(); - glTexImage2D(GL_TEXTURE_2D, 0, GL_R8, width, height, 0, GL_RED, GL_UNSIGNED_BYTE, nullptr); + glMemoryBarrier(GL_CLIENT_MAPPED_BUFFER_BARRIER_BIT); check_error(); - userdata->last_width = width; - userdata->last_height = height; - } - GLuint pbo = userdata->pbo; - check_error(); - glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, pbo); - check_error(); - glFlushMappedBufferRange(GL_PIXEL_UNPACK_BUFFER, 0, video_frame.size); - check_error(); - //glMemoryBarrier(GL_CLIENT_MAPPED_BUFFER_BARRIER_BIT); - //check_error(); + glBindTexture(GL_TEXTURE_2D, userdata->tex_cbcr[field]); + check_error(); + glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, cbcr_width, height, GL_RG, GL_UNSIGNED_BYTE, BUFFER_OFFSET(cbcr_offset + cbcr_width * field_start_line * sizeof(uint16_t))); + check_error(); + glBindTexture(GL_TEXTURE_2D, userdata->tex_y[field]); + check_error(); + glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, width, height, GL_RED, GL_UNSIGNED_BYTE, BUFFER_OFFSET(y_offset + width * field_start_line)); + check_error(); + glBindTexture(GL_TEXTURE_2D, 0); + check_error(); + GLsync fence = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0); + check_error(); + assert(fence != nullptr); + + if (field == 1) { + // Don't upload the second field as fast as we can; wait until + // the field time has approximately passed. (Otherwise, we could + // get timing jitter against the other sources, and possibly also + // against the video display, although the latter is not as critical.) + // This requires our system clock to be reasonably close to the + // video clock, but that's not an unreasonable assumption. + timespec second_field_start; + second_field_start.tv_nsec = frame_upload_start.tv_nsec + + frame_length * 1000000000 / TIMEBASE; + second_field_start.tv_sec = frame_upload_start.tv_sec + + second_field_start.tv_nsec / 1000000000; + second_field_start.tv_nsec %= 1000000000; + + while (clock_nanosleep(CLOCK_MONOTONIC, TIMER_ABSTIME, + &second_field_start, nullptr) == -1 && + errno == EINTR) ; + } - glBindTexture(GL_TEXTURE_2D, userdata->tex_cbcr); - check_error(); - glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, cbcr_width, height, GL_RG, GL_UNSIGNED_BYTE, BUFFER_OFFSET(cbcr_offset + cbcr_width * extra_lines_top * sizeof(uint16_t))); - check_error(); - glBindTexture(GL_TEXTURE_2D, userdata->tex_y); - check_error(); - glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, width, height, GL_RED, GL_UNSIGNED_BYTE, BUFFER_OFFSET(y_offset + width * extra_lines_top)); - check_error(); - glBindTexture(GL_TEXTURE_2D, 0); - check_error(); - GLsync fence = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0); - check_error(); - assert(fence != nullptr); + { + unique_lock lock(bmusb_mutex); + card->new_data_ready = true; + card->new_frame = new_frame; + card->new_frame_length = frame_length; + card->new_frame_field = field; + card->new_frame_interlaced = interlaced; + card->new_data_ready_fence = fence; + card->dropped_frames = dropped_frames; + card->new_data_ready_changed.notify_all(); - { - unique_lock lock(bmusb_mutex); - card->new_data_ready = true; - card->new_frame = RefCountedFrame(video_frame); - card->new_frame_length = frame_length; - card->new_data_ready_fence = fence; - card->dropped_frames = dropped_frames; - card->new_data_ready_changed.notify_all(); + if (field != num_fields - 1) { + // Wait until the previous frame was consumed. + card->new_data_ready_changed.wait(lock, [card]{ return !card->new_data_ready || card->should_quit; }); + if (card->should_quit) return; + } + } } } @@ -412,6 +474,8 @@ 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_frame_length = card->new_frame_length; + card_copy[card_index].new_frame_field = card->new_frame_field; + card_copy[card_index].new_frame_interlaced = card->new_frame_interlaced; 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; @@ -442,6 +506,7 @@ void Mixer::thread_func() } if (audio_level_callback != nullptr) { + unique_lock lock(r128_mutex); double loudness_s = r128.loudness_S(); double loudness_i = r128.integrated(); double loudness_range_low = r128.range_min(); @@ -476,7 +541,7 @@ void Mixer::thread_func() continue; assert(card->new_frame != nullptr); - bmusb_current_rendering_frame[card_index] = card->new_frame; + insert_new_frame(card->new_frame, card->new_frame_field, card->new_frame_interlaced, card_index, &input_state); check_error(); // The new texture might still be uploaded, @@ -487,14 +552,13 @@ void Mixer::thread_func() 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, userdata->last_width, userdata->last_height); } // 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(); + Theme::Chain theme_main_chain = theme->get_chain(0, pts(), WIDTH, HEIGHT, input_state); + EffectChain *chain = theme_main_chain.chain; + theme_main_chain.setup_chain(); + //theme_main_chain.chain->enable_phase_timing(true); GLuint y_tex, cbcr_tex; bool got_frame = h264_encoder->begin_frame(&y_tex, &cbcr_tex); @@ -521,15 +585,8 @@ void Mixer::thread_func() RefCountedGLsync fence(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0); check_error(); - // Make sure the H.264 gets a reference to all the - // input frames needed, so that they are not released back - // until the rendering is done. - vector input_frames; - for (unsigned card_index = 0; card_index < num_cards; ++card_index) { - input_frames.push_back(bmusb_current_rendering_frame[card_index]); - } const int64_t av_delay = TIMEBASE / 10; // Corresponds to the fixed delay in resampling_queue.h. TODO: Make less hard-coded. - h264_encoder->end_frame(fence, pts_int + av_delay, input_frames); + h264_encoder->end_frame(fence, pts_int + av_delay, theme_main_chain.input_frames); ++frame; pts_int += card_copy[0].new_frame_length; @@ -548,15 +605,11 @@ void Mixer::thread_func() // 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; + Theme::Chain chain = theme->get_chain(i, pts(), WIDTH, HEIGHT, input_state); // FIXME: dimensions + display_frame.chain = chain.chain; + display_frame.setup_chain = chain.setup_chain; 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.input_frames = chain.input_frames; display_frame.temp_textures = {}; output_channel[i].output_frame(display_frame); } @@ -696,7 +749,10 @@ void Mixer::process_audio_one_frame(int64_t frame_pts_int, int num_samples) vector left, right; deinterleave_samples(samples_out, &left, &right); float *ptrs[] = { left.data(), right.data() }; - r128.process(left.size(), ptrs); + { + unique_lock lock(r128_mutex); + r128.process(left.size(), ptrs); + } // Send the samples to the sound card. if (alsa) {