10 #include <effect_chain.h>
11 #include <effect_util.h>
12 #include <epoxy/egl.h>
14 #include <image_format.h>
16 #include <overlay_effect.h>
17 #include <padding_effect.h>
18 #include <resample_effect.h>
19 #include <resource_pool.h>
20 #include <saturation_effect.h>
27 #include <white_balance_effect.h>
29 #include <ycbcr_input.h>
31 #include <condition_variable>
41 #include "h264encode.h"
42 #include "pbo_frame_allocator.h"
43 #include "ref_counted_gl_sync.h"
47 using namespace movit;
49 using namespace std::placeholders;
51 Mixer *global_mixer = nullptr;
53 Mixer::Mixer(const QSurfaceFormat &format)
54 : mixer_surface(create_surface(format)),
55 h264_encoder_surface(create_surface(format))
57 CHECK(init_movit(MOVIT_SHADER_DIR, MOVIT_DEBUG_OFF));
60 resource_pool.reset(new ResourcePool);
61 output_channel[OUTPUT_LIVE].parent = this;
62 output_channel[OUTPUT_PREVIEW].parent = this;
64 ImageFormat inout_format;
65 inout_format.color_space = COLORSPACE_sRGB;
66 inout_format.gamma_curve = GAMMA_sRGB;
68 YCbCrFormat input_ycbcr_format;
69 input_ycbcr_format.chroma_subsampling_x = 2;
70 input_ycbcr_format.chroma_subsampling_y = 1;
71 input_ycbcr_format.cb_x_position = 0.0;
72 input_ycbcr_format.cr_x_position = 0.0;
73 input_ycbcr_format.cb_y_position = 0.5;
74 input_ycbcr_format.cr_y_position = 0.5;
75 input_ycbcr_format.luma_coefficients = YCBCR_REC_601;
76 input_ycbcr_format.full_range = false;
78 YCbCrFormat output_ycbcr_format;
79 output_ycbcr_format.chroma_subsampling_x = 1;
80 output_ycbcr_format.chroma_subsampling_y = 1;
81 output_ycbcr_format.luma_coefficients = YCBCR_REC_601;
82 output_ycbcr_format.full_range = false;
85 chain.reset(new EffectChain(WIDTH, HEIGHT, resource_pool.get()));
87 input[0] = new YCbCrInput(inout_format, input_ycbcr_format, WIDTH, HEIGHT, YCBCR_INPUT_SPLIT_Y_AND_CBCR);
88 chain->add_input(input[0]);
89 input[1] = new YCbCrInput(inout_format, input_ycbcr_format, WIDTH, HEIGHT, YCBCR_INPUT_SPLIT_Y_AND_CBCR);
90 chain->add_input(input[1]);
91 resample_effect = chain->add_effect(new ResampleEffect(), input[0]);
92 padding_effect = chain->add_effect(new IntegralPaddingEffect());
93 float border_color[] = { 0.0f, 0.0f, 0.0f, 1.0f };
94 CHECK(padding_effect->set_vec4("border_color", border_color));
96 resample2_effect = chain->add_effect(new ResampleEffect(), input[1]);
97 Effect *saturation_effect = chain->add_effect(new SaturationEffect());
98 CHECK(saturation_effect->set_float("saturation", 0.3f));
99 Effect *wb_effect = chain->add_effect(new WhiteBalanceEffect());
100 CHECK(wb_effect->set_float("output_color_temperature", 3500.0));
101 padding2_effect = chain->add_effect(new IntegralPaddingEffect());
103 chain->add_effect(new OverlayEffect(), padding_effect, padding2_effect);
105 chain->add_output(inout_format, OUTPUT_ALPHA_FORMAT_POSTMULTIPLIED);
106 chain->add_ycbcr_output(inout_format, OUTPUT_ALPHA_FORMAT_POSTMULTIPLIED, output_ycbcr_format, YCBCR_OUTPUT_SPLIT_Y_AND_CBCR);
107 chain->set_dither_bits(8);
108 chain->set_output_origin(OUTPUT_ORIGIN_TOP_LEFT);
111 // Preview chain (always shows just first input for now).
112 preview_chain.reset(new EffectChain(WIDTH, HEIGHT, resource_pool.get()));
114 preview_input = new YCbCrInput(inout_format, input_ycbcr_format, WIDTH, HEIGHT, YCBCR_INPUT_SPLIT_Y_AND_CBCR);
115 preview_chain->add_input(preview_input);
116 preview_chain->add_output(inout_format, OUTPUT_ALPHA_FORMAT_POSTMULTIPLIED);
117 preview_chain->set_dither_bits(0); // Don't bother.
118 preview_chain->set_output_origin(OUTPUT_ORIGIN_TOP_LEFT);
119 preview_chain->finalize();
121 h264_encoder.reset(new H264Encoder(h264_encoder_surface, WIDTH, HEIGHT, "test.mp4"));
123 printf("Configuring first card...\n");
124 cards[0].usb = new BMUSBCapture(0x1edb, 0xbd3b); // 0xbd4f
125 cards[0].usb->set_frame_callback(std::bind(&Mixer::bm_frame, this, 0, _1, _2, _3, _4, _5, _6, _7));
126 cards[0].frame_allocator.reset(new PBOFrameAllocator(1280 * 750 * 2 + 44));
127 cards[0].usb->set_video_frame_allocator(cards[0].frame_allocator.get());
128 cards[0].usb->configure_card();
129 cards[0].surface = create_surface(format);
131 cards[1].surface = create_surface(format);
134 if (NUM_CARDS == 2) {
135 printf("Configuring second card...\n");
136 cards[1].usb = new BMUSBCapture(0x1edb, 0xbd4f);
137 cards[1].usb->set_frame_callback(std::bind(&Mixer::bm_frame, this, 1, _1, _2, _3, _4, _5, _6, _7));
138 cards[1].frame_allocator.reset(new PBOFrameAllocator(1280 * 750 * 2 + 44));
139 cards[1].usb->set_video_frame_allocator(cards[1].frame_allocator.get());
140 cards[1].usb->configure_card();
143 BMUSBCapture::start_bm_thread();
145 for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
146 cards[card_index].usb->start_bm_capture();
147 input[card_index]->set_pixel_data(0, nullptr, 0);
148 input[card_index]->set_pixel_data(1, nullptr, 0);
151 //chain->enable_phase_timing(true);
153 // Set up stuff for NV12 conversion.
156 string cbcr_vert_shader = read_file("vs-cbcr.130.vert");
157 string cbcr_frag_shader =
160 "uniform sampler2D cbcr_tex; \n"
162 " gl_FragColor = texture2D(cbcr_tex, tc0); \n"
164 cbcr_program_num = resource_pool->compile_glsl_program(cbcr_vert_shader, cbcr_frag_shader);
169 resource_pool->release_glsl_program(cbcr_program_num);
170 BMUSBCapture::stop_bm_thread();
173 void Mixer::bm_frame(int card_index, uint16_t timecode,
174 FrameAllocator::Frame video_frame, size_t video_offset, uint16_t video_format,
175 FrameAllocator::Frame audio_frame, size_t audio_offset, uint16_t audio_format)
177 CaptureCard *card = &cards[card_index];
178 if (!card->thread_initialized) {
179 printf("initializing context for bmusb thread %d\n", card_index);
180 eglBindAPI(EGL_OPENGL_API);
181 card->context = create_context();
182 if (!make_current(card->context, card->surface)) {
183 printf("failed to create bmusb context\n");
186 card->thread_initialized = true;
189 if (video_frame.len - video_offset != 1280 * 750 * 2) {
190 printf("dropping frame with wrong length (%ld)\n", video_frame.len - video_offset);
191 FILE *fp = fopen("frame.raw", "wb");
192 fwrite(video_frame.data, video_frame.len, 1, fp);
195 card->usb->get_video_frame_allocator()->release_frame(video_frame);
196 card->usb->get_audio_frame_allocator()->release_frame(audio_frame);
200 // Wait until the previous frame was consumed.
201 std::unique_lock<std::mutex> lock(bmusb_mutex);
202 card->new_data_ready_changed.wait(lock, [card]{ return !card->new_data_ready; });
204 GLuint pbo = (GLint)(intptr_t)video_frame.userdata;
206 glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, pbo);
208 glFlushMappedBufferRange(GL_PIXEL_UNPACK_BUFFER, 0, video_frame.size);
210 //glMemoryBarrier(GL_CLIENT_MAPPED_BUFFER_BARRIER_BIT);
212 GLsync fence = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);
214 assert(fence != nullptr);
216 std::unique_lock<std::mutex> lock(bmusb_mutex);
217 card->new_data_ready = true;
218 card->new_frame = video_frame;
219 card->new_data_ready_fence = fence;
220 card->new_data_ready_changed.notify_all();
223 // Video frame will be released later.
224 card->usb->get_audio_frame_allocator()->release_frame(audio_frame);
227 void Mixer::place_rectangle(Effect *resample_effect, Effect *padding_effect, float x0, float y0, float x1, float y1)
235 if (x0 > 1280.0 || x1 < 0.0 || y0 > 720.0 || y1 < 0.0) {
236 CHECK(resample_effect->set_int("width", 1));
237 CHECK(resample_effect->set_int("height", 1));
238 CHECK(resample_effect->set_float("zoom_x", 1280.0));
239 CHECK(resample_effect->set_float("zoom_y", 720.0));
240 CHECK(padding_effect->set_int("left", 2000));
241 CHECK(padding_effect->set_int("top", 2000));
245 // Clip. (TODO: Clip on upper/left sides, too.)
247 srcx1 = (1280.0 - x0) / (x1 - x0);
251 srcy1 = (720.0 - y0) / (y1 - y0);
255 float x_subpixel_offset = x0 - floor(x0);
256 float y_subpixel_offset = y0 - floor(y0);
258 // Resampling must be to an integral number of pixels. Round up,
259 // and then add an extra pixel so we have some leeway for the border.
260 int width = int(ceil(x1 - x0)) + 1;
261 int height = int(ceil(y1 - y0)) + 1;
262 CHECK(resample_effect->set_int("width", width));
263 CHECK(resample_effect->set_int("height", height));
265 // Correct the discrepancy with zoom. (This will leave a small
266 // excess edge of pixels and subpixels, which we'll correct for soon.)
267 float zoom_x = (x1 - x0) / (width * (srcx1 - srcx0));
268 float zoom_y = (y1 - y0) / (height * (srcy1 - srcy0));
269 CHECK(resample_effect->set_float("zoom_x", zoom_x));
270 CHECK(resample_effect->set_float("zoom_y", zoom_y));
271 CHECK(resample_effect->set_float("zoom_center_x", 0.0f));
272 CHECK(resample_effect->set_float("zoom_center_y", 0.0f));
274 // Padding must also be to a whole-pixel offset.
275 CHECK(padding_effect->set_int("left", floor(x0)));
276 CHECK(padding_effect->set_int("top", floor(y0)));
278 // Correct _that_ discrepancy by subpixel offset in the resampling.
279 CHECK(resample_effect->set_float("left", -x_subpixel_offset / zoom_x));
280 CHECK(resample_effect->set_float("top", -y_subpixel_offset / zoom_y));
282 // Finally, adjust the border so it is exactly where we want it.
283 CHECK(padding_effect->set_float("border_offset_left", x_subpixel_offset));
284 CHECK(padding_effect->set_float("border_offset_right", x1 - (floor(x0) + width)));
285 CHECK(padding_effect->set_float("border_offset_top", y_subpixel_offset));
286 CHECK(padding_effect->set_float("border_offset_bottom", y1 - (floor(y0) + height)));
289 void Mixer::thread_func()
291 eglBindAPI(EGL_OPENGL_API);
292 QOpenGLContext *context = create_context();
293 if (!make_current(context, mixer_surface)) {
298 struct timespec start, now;
299 clock_gettime(CLOCK_MONOTONIC, &start);
301 while (!should_quit) {
304 //int width0 = lrintf(848 * (1.0 + 0.2 * sin(frame * 0.02)));
306 int height0 = lrintf(width0 * 9.0 / 16.0);
308 //float top0 = 96 + 48 * sin(frame * 0.005);
309 //float left0 = 96 + 48 * cos(frame * 0.006);
312 float bottom0 = top0 + height0;
313 float right0 = left0 + width0;
318 float bottom1 = 720 - 48;
319 float right1 = 1280 - 16;
320 float top1 = bottom1 - height1;
321 float left1 = right1 - width1;
323 if (current_source == SOURCE_INPUT1) {
330 bottom1 = HEIGHT + 20;
333 } else if (current_source == SOURCE_INPUT2) {
340 bottom0 = HEIGHT + 20;
344 float t = 0.5 + 0.5 * cos(frame * 0.006);
345 float scale0 = 1.0 + t * (1280.0 / 848.0 - 1.0);
346 float tx0 = 0.0 + t * (-16.0 * scale0);
347 float ty0 = 0.0 + t * (-48.0 * scale0);
349 top0 = top0 * scale0 + ty0;
350 bottom0 = bottom0 * scale0 + ty0;
351 left0 = left0 * scale0 + tx0;
352 right0 = right0 * scale0 + tx0;
354 top1 = top1 * scale0 + ty0;
355 bottom1 = bottom1 * scale0 + ty0;
356 left1 = left1 * scale0 + tx0;
357 right1 = right1 * scale0 + tx0;
360 place_rectangle(resample_effect, padding_effect, left0, top0, right0, bottom0);
361 place_rectangle(resample2_effect, padding2_effect, left1, top1, right1, bottom1);
363 CaptureCard card_copy[NUM_CARDS];
366 std::unique_lock<std::mutex> lock(bmusb_mutex);
368 // The first card is the master timer, so wait for it to have a new frame.
369 // TODO: Make configurable, and with a timeout.
370 cards[0].new_data_ready_changed.wait(lock, [this]{ return cards[0].new_data_ready; });
372 for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
373 CaptureCard *card = &cards[card_index];
374 card_copy[card_index].usb = card->usb;
375 card_copy[card_index].new_data_ready = card->new_data_ready;
376 card_copy[card_index].new_frame = card->new_frame;
377 card_copy[card_index].new_data_ready_fence = card->new_data_ready_fence;
378 card->new_data_ready = false;
379 card->new_data_ready_changed.notify_all();
383 vector<FrameAllocator::Frame> input_frames_to_release;
385 for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
386 CaptureCard *card = &card_copy[card_index];
387 if (!card->new_data_ready)
390 // Now we're done with the previous frame, so we can definitely
391 // release it when this is done rendering. (Actually, we could do
392 // it one frame earlier, but before we have a new one, there's no
393 // knowing when the current one is released.)
394 if (bmusb_current_rendering_frame[card_index].owner != nullptr) {
395 input_frames_to_release.push_back(bmusb_current_rendering_frame[card_index]);
397 bmusb_current_rendering_frame[card_index] = card->new_frame;
400 // The new texture might still be uploaded,
401 // tell the GPU to wait until it's there.
402 if (card->new_data_ready_fence)
403 glWaitSync(card->new_data_ready_fence, /*flags=*/0, GL_TIMEOUT_IGNORED);
405 glDeleteSync(card->new_data_ready_fence);
407 GLint input_tex_pbo = (GLint)(intptr_t)card->new_frame.userdata;
408 input[card_index]->set_pixel_data(0, (unsigned char *)BUFFER_OFFSET((1280 * 750 * 2 + 44) / 2 + 1280 * 25 + 22), input_tex_pbo);
409 input[card_index]->set_pixel_data(1, (unsigned char *)BUFFER_OFFSET(1280 * 25 + 22), input_tex_pbo);
411 if (NUM_CARDS == 1) {
412 // Set to the other one, too.
413 input[1]->set_pixel_data(0, (unsigned char *)BUFFER_OFFSET((1280 * 750 * 2 + 44) / 2 + 1280 * 25 + 22), input_tex_pbo);
414 input[1]->set_pixel_data(1, (unsigned char *)BUFFER_OFFSET(1280 * 25 + 22), input_tex_pbo);
417 if (card_index == 0) {
418 preview_input->set_pixel_data(0, (unsigned char *)BUFFER_OFFSET((1280 * 750 * 2 + 44) / 2 + 1280 * 25 + 22), input_tex_pbo);
419 preview_input->set_pixel_data(1, (unsigned char *)BUFFER_OFFSET(1280 * 25 + 22), input_tex_pbo);
423 GLuint y_tex, cbcr_tex;
424 bool got_frame = h264_encoder->begin_frame(&y_tex, &cbcr_tex);
427 // Render main chain.
428 GLuint cbcr_full_tex = resource_pool->create_2d_texture(GL_RG8, WIDTH, HEIGHT);
429 GLuint rgba_tex = resource_pool->create_2d_texture(GL_RGBA8, WIDTH, HEIGHT);
430 GLuint fbo = resource_pool->create_fbo(y_tex, cbcr_full_tex, rgba_tex);
431 chain->render_to_fbo(fbo, WIDTH, HEIGHT);
432 resource_pool->release_fbo(fbo);
434 subsample_chroma(cbcr_full_tex, cbcr_tex);
435 resource_pool->release_2d_texture(cbcr_full_tex);
437 // Render preview chain.
438 GLuint preview_rgba_tex = resource_pool->create_2d_texture(GL_RGBA8, output_channel[OUTPUT_PREVIEW].width, output_channel[OUTPUT_PREVIEW].height);
439 fbo = resource_pool->create_fbo(preview_rgba_tex);
440 preview_chain->render_to_fbo(fbo, output_channel[OUTPUT_PREVIEW].width, output_channel[OUTPUT_PREVIEW].height);
441 resource_pool->release_fbo(fbo);
443 RefCountedGLsync fence(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);
445 h264_encoder->end_frame(fence, input_frames_to_release);
447 output_channel[OUTPUT_LIVE].output_frame(rgba_tex, fence);
448 output_channel[OUTPUT_PREVIEW].output_frame(preview_rgba_tex, fence);
450 clock_gettime(CLOCK_MONOTONIC, &now);
451 double elapsed = now.tv_sec - start.tv_sec +
452 1e-9 * (now.tv_nsec - start.tv_nsec);
453 if (frame % 100 == 0) {
454 printf("%d frames in %.3f seconds = %.1f fps (%.1f ms/frame)\n",
455 frame, elapsed, frame / elapsed,
456 1e3 * elapsed / frame);
457 // chain->print_phase_timing();
460 // Reset every 100 frames, so that local variations in frame times
461 // (especially for the first few frames, when the shaders are
462 // compiled etc.) don't make it hard to measure for the entire
463 // remaining duration of the program.
464 if (frame == 10000) {
472 void Mixer::subsample_chroma(GLuint src_tex, GLuint dst_tex)
475 glGenVertexArrays(1, &vao);
484 glBindVertexArray(vao);
488 GLuint fbo = resource_pool->create_fbo(dst_tex);
489 glBindFramebuffer(GL_FRAMEBUFFER, fbo);
490 glViewport(0, 0, WIDTH/2, HEIGHT/2);
493 glUseProgram(cbcr_program_num);
496 glActiveTexture(GL_TEXTURE0);
498 glBindTexture(GL_TEXTURE_2D, src_tex);
500 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
502 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
504 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
507 float chroma_offset_0[] = { -0.5f / WIDTH, 0.0f };
508 set_uniform_vec2(cbcr_program_num, "foo", "chroma_offset_0", chroma_offset_0);
510 GLuint position_vbo = fill_vertex_attribute(cbcr_program_num, "position", 2, GL_FLOAT, sizeof(vertices), vertices);
511 GLuint texcoord_vbo = fill_vertex_attribute(cbcr_program_num, "texcoord", 2, GL_FLOAT, sizeof(vertices), vertices); // Same as vertices.
513 glDrawArrays(GL_TRIANGLES, 0, 3);
516 cleanup_vertex_attribute(cbcr_program_num, "position", position_vbo);
517 cleanup_vertex_attribute(cbcr_program_num, "texcoord", texcoord_vbo);
522 resource_pool->release_fbo(fbo);
523 glDeleteVertexArrays(1, &vao);
526 void Mixer::release_display_frame(DisplayFrame *frame)
528 resource_pool->release_2d_texture(frame->texnum);
530 frame->ready_fence.reset();
535 mixer_thread = std::thread(&Mixer::thread_func, this);
544 void Mixer::cut(Source source)
546 current_source = source;
549 void Mixer::OutputChannel::output_frame(GLuint tex, RefCountedGLsync fence)
551 // Store this frame for display. Remove the ready frame if any
552 // (it was seemingly never used).
554 std::unique_lock<std::mutex> lock(frame_mutex);
555 if (has_ready_frame) {
556 parent->release_display_frame(&ready_frame);
558 ready_frame.texnum = tex;
559 ready_frame.ready_fence = fence;
560 has_ready_frame = true;
563 if (has_new_frame_ready_callback) {
564 new_frame_ready_callback();
568 bool Mixer::OutputChannel::get_display_frame(DisplayFrame *frame)
570 std::unique_lock<std::mutex> lock(frame_mutex);
571 if (!has_current_frame && !has_ready_frame) {
575 if (has_current_frame && has_ready_frame) {
576 // We have a new ready frame. Toss the current one.
577 parent->release_display_frame(¤t_frame);
578 has_current_frame = false;
580 if (has_ready_frame) {
581 assert(!has_current_frame);
582 current_frame = ready_frame;
583 ready_frame.ready_fence.reset(); // Drop the refcount.
584 has_current_frame = true;
585 has_ready_frame = false;
588 *frame = current_frame;
592 void Mixer::OutputChannel::set_frame_ready_callback(Mixer::new_frame_ready_callback_t callback)
594 new_frame_ready_callback = callback;
595 has_new_frame_ready_callback = true;
598 void Mixer::OutputChannel::set_size(int width, int height)
601 this->height = height;