]> git.sesse.net Git - nageru/blobdiff - mixer.cpp
Re-run IWYU, again with lots of manual cleanup.
[nageru] / mixer.cpp
index 55707490baa706fd842dd13a8bb6b46c51c95984..981c0cb4e38b699b7d6628b07d72ba58873aa2c0 100644 (file)
--- a/mixer.cpp
+++ b/mixer.cpp
@@ -7,27 +7,20 @@
 #include "mixer.h"
 
 #include <assert.h>
-#include <effect.h>
-#include <effect_chain.h>
-#include <effect_util.h>
 #include <epoxy/egl.h>
-#include <features.h>
-#include <image_format.h>
 #include <init.h>
-#include <overlay_effect.h>
-#include <padding_effect.h>
-#include <resample_effect.h>
-#include <resource_pool.h>
-#include <saturation_effect.h>
+#include <movit/effect_chain.h>
+#include <movit/effect_util.h>
+#include <movit/flat_input.h>
+#include <movit/image_format.h>
+#include <movit/resource_pool.h>
 #include <stdint.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <sys/time.h>
 #include <time.h>
 #include <util.h>
-#include <white_balance_effect.h>
-#include <ycbcr.h>
-#include <ycbcr_input.h>
+#include <algorithm>
 #include <cmath>
 #include <condition_variable>
 #include <cstddef>
@@ -35,6 +28,7 @@
 #include <mutex>
 #include <string>
 #include <thread>
+#include <utility>
 #include <vector>
 
 #include "bmusb/bmusb.h"
@@ -70,8 +64,9 @@ void convert_fixed24_to_fp32(float *dst, size_t out_channels, const uint8_t *src
 
 }  // namespace
 
-Mixer::Mixer(const QSurfaceFormat &format)
+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))
 {
@@ -80,8 +75,12 @@ Mixer::Mixer(const QSurfaceFormat &format)
        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);
-       theme.reset(new Theme("theme.lua", resource_pool.get()));
+       theme.reset(new Theme("theme.lua", resource_pool.get(), num_cards));
        output_channel[OUTPUT_LIVE].parent = this;
        output_channel[OUTPUT_PREVIEW].parent = this;
        output_channel[OUTPUT_INPUT0].parent = this;
@@ -102,7 +101,7 @@ Mixer::Mixer(const QSurfaceFormat &format)
 
        h264_encoder.reset(new H264Encoder(h264_encoder_surface, WIDTH, HEIGHT, &httpd));
 
-       for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
+       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);
@@ -128,7 +127,7 @@ Mixer::Mixer(const QSurfaceFormat &format)
 
        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();
        }
 
@@ -147,7 +146,8 @@ Mixer::Mixer(const QSurfaceFormat &format)
                "} \n";
        cbcr_program_num = resource_pool->compile_glsl_program(cbcr_vert_shader, cbcr_frag_shader);
 
-       r128_state = ebur128_init(2, 48000, EBUR128_MODE_TRUE_PEAK | EBUR128_MODE_M | EBUR128_MODE_S | EBUR128_MODE_I | EBUR128_MODE_LRA);
+       r128.init(2, 48000);
+       r128.integr_start();
 }
 
 Mixer::~Mixer()
@@ -155,7 +155,7 @@ 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) {
+       for (unsigned card_index = 0; card_index < num_cards; ++card_index) {
                {
                        unique_lock<mutex> lock(bmusb_mutex);
                        cards[card_index].should_quit = true;  // Unblock thread.
@@ -163,8 +163,6 @@ Mixer::~Mixer()
                }
                cards[card_index].usb->stop_dequeue_thread();
        }
-
-       ebur128_destroy(&r128_state);
 }
 
 namespace {
@@ -179,9 +177,33 @@ int unwrap_timecode(uint16_t current_wrapped, int last)
        }
 }
 
+float find_peak(const vector<float> &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<float> &in, vector<float> *out_l, vector<float> *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++;
+       }
+}
+
 }  // namespace
 
-void Mixer::bm_frame(int card_index, uint16_t timecode,
+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)
 {
@@ -200,12 +222,6 @@ void Mixer::bm_frame(int card_index, uint16_t timecode,
                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) {
@@ -214,6 +230,12 @@ void Mixer::bm_frame(int card_index, uint16_t timecode,
        }
        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<float> 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<mutex> lock(card->audio_mutex);
@@ -321,7 +343,7 @@ void Mixer::thread_func()
        int dropped_frames = 0;
 
        while (!should_quit) {
-               CaptureCard card_copy[NUM_CARDS];
+               CaptureCard card_copy[MAX_CARDS];
 
                {
                        unique_lock<mutex> lock(bmusb_mutex);
@@ -330,7 +352,7 @@ void Mixer::thread_func()
                        // 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;
@@ -347,7 +369,7 @@ void Mixer::thread_func()
                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) {
+                       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);
@@ -356,7 +378,11 @@ void Mixer::thread_func()
                                        }
                                }
                                if (card_index == 0) {
-                                       ebur128_add_frames_float(r128_state, samples_out.data(), samples_out.size() / 2);
+                                       vector<float> left, right;
+                                       peak = std::max(peak, find_peak(samples_out));
+                                       deinterleave_samples(samples_out, &left, &right);
+                                       float *ptrs[] = { left.data(), right.data() };
+                                       r128.process(left.size(), ptrs);
                                        h264_encoder->add_audio(pts_int, move(samples_out));
                                }
                        }
@@ -368,22 +394,25 @@ void Mixer::thread_func()
                }
 
                if (audio_level_callback != nullptr) {
-                       double loudness_s, loudness_i, peak_level_l, peak_level_r;
-                       double lra;
-                       ebur128_loudness_shortterm(r128_state, &loudness_s);
-                       ebur128_loudness_global(r128_state, &loudness_i);
-                       ebur128_loudness_range(r128_state, &lra);
-                       ebur128_true_peak(r128_state, 0, &peak_level_l);
-                       ebur128_true_peak(r128_state, 1, &peak_level_r);
-
-                       // FIXME: This is wrong. We need proper support from libebur128 for this.
-                       double loudness_range_low = loudness_i - 0.5 * lra;
-                       double loudness_range_high = loudness_i + 0.5 * lra;
-
-                       audio_level_callback(loudness_s, 20.0 * log10(max(peak_level_l, peak_level_r)),
+                       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) {
@@ -392,7 +421,7 @@ void Mixer::thread_func()
                        continue;
                }
 
-               for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
+               for (unsigned card_index = 0; card_index < num_cards; ++card_index) {
                        CaptureCard *card = &card_copy[card_index];
                        if (!card->new_data_ready || card->new_frame->len == 0)
                                continue;
@@ -426,6 +455,7 @@ 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);
 
@@ -446,7 +476,7 @@ void Mixer::thread_func()
                // input frames needed, so that they are not released back
                // until the rendering is done.
                vector<RefCountedFrame> 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]);
                }
                const int64_t av_delay = TIMEBASE / 10;  // Corresponds to the fixed delay in resampler.h. TODO: Make less hard-coded.
@@ -473,7 +503,11 @@ void Mixer::thread_func()
                        display_frame.chain = chain.first;
                        display_frame.setup_chain = chain.second;
                        display_frame.ready_fence = fence;
-                       display_frame.input_frames = { bmusb_current_rendering_frame[0], bmusb_current_rendering_frame[1] };  // FIXME: possible to do better?
+
+                       // 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);
                }