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, 1280, 720));
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, 1280, 720));
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 const PBOFrameAllocator::Userdata *userdata = (const PBOFrameAllocator::Userdata *)video_frame.userdata;
205 GLuint pbo = userdata->pbo;
207 glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, pbo);
209 glFlushMappedBufferRange(GL_PIXEL_UNPACK_BUFFER, 0, video_frame.size);
211 //glMemoryBarrier(GL_CLIENT_MAPPED_BUFFER_BARRIER_BIT);
214 // Upload the textures.
215 glBindTexture(GL_TEXTURE_2D, userdata->tex_y);
217 glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, 1280, 720, GL_RED, GL_UNSIGNED_BYTE, BUFFER_OFFSET((1280 * 750 * 2 + 44) / 2 + 1280 * 25 + 22));
219 glBindTexture(GL_TEXTURE_2D, userdata->tex_cbcr);
221 glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, 1280/2, 720, GL_RG, GL_UNSIGNED_BYTE, BUFFER_OFFSET(1280 * 25 + 22));
223 glBindTexture(GL_TEXTURE_2D, 0);
225 GLsync fence = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);
227 assert(fence != nullptr);
229 std::unique_lock<std::mutex> lock(bmusb_mutex);
230 card->new_data_ready = true;
231 card->new_frame = video_frame;
232 card->new_data_ready_fence = fence;
233 card->new_data_ready_changed.notify_all();
236 // Video frame will be released later.
237 card->usb->get_audio_frame_allocator()->release_frame(audio_frame);
240 void Mixer::place_rectangle(Effect *resample_effect, Effect *padding_effect, float x0, float y0, float x1, float y1)
248 if (x0 > 1280.0 || x1 < 0.0 || y0 > 720.0 || y1 < 0.0) {
249 CHECK(resample_effect->set_int("width", 1));
250 CHECK(resample_effect->set_int("height", 1));
251 CHECK(resample_effect->set_float("zoom_x", 1280.0));
252 CHECK(resample_effect->set_float("zoom_y", 720.0));
253 CHECK(padding_effect->set_int("left", 2000));
254 CHECK(padding_effect->set_int("top", 2000));
258 // Clip. (TODO: Clip on upper/left sides, too.)
260 srcx1 = (1280.0 - x0) / (x1 - x0);
264 srcy1 = (720.0 - y0) / (y1 - y0);
268 float x_subpixel_offset = x0 - floor(x0);
269 float y_subpixel_offset = y0 - floor(y0);
271 // Resampling must be to an integral number of pixels. Round up,
272 // and then add an extra pixel so we have some leeway for the border.
273 int width = int(ceil(x1 - x0)) + 1;
274 int height = int(ceil(y1 - y0)) + 1;
275 CHECK(resample_effect->set_int("width", width));
276 CHECK(resample_effect->set_int("height", height));
278 // Correct the discrepancy with zoom. (This will leave a small
279 // excess edge of pixels and subpixels, which we'll correct for soon.)
280 float zoom_x = (x1 - x0) / (width * (srcx1 - srcx0));
281 float zoom_y = (y1 - y0) / (height * (srcy1 - srcy0));
282 CHECK(resample_effect->set_float("zoom_x", zoom_x));
283 CHECK(resample_effect->set_float("zoom_y", zoom_y));
284 CHECK(resample_effect->set_float("zoom_center_x", 0.0f));
285 CHECK(resample_effect->set_float("zoom_center_y", 0.0f));
287 // Padding must also be to a whole-pixel offset.
288 CHECK(padding_effect->set_int("left", floor(x0)));
289 CHECK(padding_effect->set_int("top", floor(y0)));
291 // Correct _that_ discrepancy by subpixel offset in the resampling.
292 CHECK(resample_effect->set_float("left", -x_subpixel_offset / zoom_x));
293 CHECK(resample_effect->set_float("top", -y_subpixel_offset / zoom_y));
295 // Finally, adjust the border so it is exactly where we want it.
296 CHECK(padding_effect->set_float("border_offset_left", x_subpixel_offset));
297 CHECK(padding_effect->set_float("border_offset_right", x1 - (floor(x0) + width)));
298 CHECK(padding_effect->set_float("border_offset_top", y_subpixel_offset));
299 CHECK(padding_effect->set_float("border_offset_bottom", y1 - (floor(y0) + height)));
302 void Mixer::thread_func()
304 eglBindAPI(EGL_OPENGL_API);
305 QOpenGLContext *context = create_context();
306 if (!make_current(context, mixer_surface)) {
311 struct timespec start, now;
312 clock_gettime(CLOCK_MONOTONIC, &start);
314 while (!should_quit) {
317 //int width0 = lrintf(848 * (1.0 + 0.2 * sin(frame * 0.02)));
319 int height0 = lrintf(width0 * 9.0 / 16.0);
321 //float top0 = 96 + 48 * sin(frame * 0.005);
322 //float left0 = 96 + 48 * cos(frame * 0.006);
325 float bottom0 = top0 + height0;
326 float right0 = left0 + width0;
331 float bottom1 = 720 - 48;
332 float right1 = 1280 - 16;
333 float top1 = bottom1 - height1;
334 float left1 = right1 - width1;
336 if (current_source == SOURCE_INPUT1) {
343 bottom1 = HEIGHT + 20;
346 } else if (current_source == SOURCE_INPUT2) {
353 bottom0 = HEIGHT + 20;
357 float t = 0.5 + 0.5 * cos(frame * 0.006);
358 float scale0 = 1.0 + t * (1280.0 / 848.0 - 1.0);
359 float tx0 = 0.0 + t * (-16.0 * scale0);
360 float ty0 = 0.0 + t * (-48.0 * scale0);
362 top0 = top0 * scale0 + ty0;
363 bottom0 = bottom0 * scale0 + ty0;
364 left0 = left0 * scale0 + tx0;
365 right0 = right0 * scale0 + tx0;
367 top1 = top1 * scale0 + ty0;
368 bottom1 = bottom1 * scale0 + ty0;
369 left1 = left1 * scale0 + tx0;
370 right1 = right1 * scale0 + tx0;
373 place_rectangle(resample_effect, padding_effect, left0, top0, right0, bottom0);
374 place_rectangle(resample2_effect, padding2_effect, left1, top1, right1, bottom1);
376 CaptureCard card_copy[NUM_CARDS];
379 std::unique_lock<std::mutex> lock(bmusb_mutex);
381 // The first card is the master timer, so wait for it to have a new frame.
382 // TODO: Make configurable, and with a timeout.
383 cards[0].new_data_ready_changed.wait(lock, [this]{ return cards[0].new_data_ready; });
385 for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
386 CaptureCard *card = &cards[card_index];
387 card_copy[card_index].usb = card->usb;
388 card_copy[card_index].new_data_ready = card->new_data_ready;
389 card_copy[card_index].new_frame = card->new_frame;
390 card_copy[card_index].new_data_ready_fence = card->new_data_ready_fence;
391 card->new_data_ready = false;
392 card->new_data_ready_changed.notify_all();
396 vector<FrameAllocator::Frame> input_frames_to_release;
398 for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
399 CaptureCard *card = &card_copy[card_index];
400 if (!card->new_data_ready)
403 // Now we're done with the previous frame, so we can definitely
404 // release it when this is done rendering. (Actually, we could do
405 // it one frame earlier, but before we have a new one, there's no
406 // knowing when the current one is released.)
407 if (bmusb_current_rendering_frame[card_index].owner != nullptr) {
408 input_frames_to_release.push_back(bmusb_current_rendering_frame[card_index]);
410 bmusb_current_rendering_frame[card_index] = card->new_frame;
413 // The new texture might still be uploaded,
414 // tell the GPU to wait until it's there.
415 if (card->new_data_ready_fence)
416 glWaitSync(card->new_data_ready_fence, /*flags=*/0, GL_TIMEOUT_IGNORED);
418 glDeleteSync(card->new_data_ready_fence);
420 const PBOFrameAllocator::Userdata *userdata = (const PBOFrameAllocator::Userdata *)card->new_frame.userdata;
421 input[card_index]->set_texture_num(0, userdata->tex_y);
422 input[card_index]->set_texture_num(1, userdata->tex_cbcr);
424 if (NUM_CARDS == 1) {
425 // Set to the other one, too.
426 input[1]->set_texture_num(0, userdata->tex_y);
427 input[1]->set_texture_num(1, userdata->tex_cbcr);
430 if (card_index == 0) {
431 preview_input->set_texture_num(0, userdata->tex_y);
432 preview_input->set_texture_num(1, userdata->tex_cbcr);
436 GLuint y_tex, cbcr_tex;
437 bool got_frame = h264_encoder->begin_frame(&y_tex, &cbcr_tex);
440 // Render main chain.
441 GLuint cbcr_full_tex = resource_pool->create_2d_texture(GL_RG8, WIDTH, HEIGHT);
442 GLuint rgba_tex = resource_pool->create_2d_texture(GL_RGB565, WIDTH, HEIGHT); // Saves texture bandwidth, although dithering gets messed up.
443 GLuint fbo = resource_pool->create_fbo(y_tex, cbcr_full_tex, rgba_tex);
444 chain->render_to_fbo(fbo, WIDTH, HEIGHT);
445 resource_pool->release_fbo(fbo);
447 subsample_chroma(cbcr_full_tex, cbcr_tex);
448 resource_pool->release_2d_texture(cbcr_full_tex);
450 // Render preview chain.
451 GLuint preview_rgba_tex = resource_pool->create_2d_texture(GL_RGB565, output_channel[OUTPUT_PREVIEW].width, output_channel[OUTPUT_PREVIEW].height); // Saves texture bandwidth, although dithering gets messed up.
452 fbo = resource_pool->create_fbo(preview_rgba_tex);
453 preview_chain->render_to_fbo(fbo, output_channel[OUTPUT_PREVIEW].width, output_channel[OUTPUT_PREVIEW].height);
454 resource_pool->release_fbo(fbo);
456 RefCountedGLsync fence(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);
458 h264_encoder->end_frame(fence, input_frames_to_release);
460 output_channel[OUTPUT_LIVE].output_frame(rgba_tex, fence);
461 output_channel[OUTPUT_PREVIEW].output_frame(preview_rgba_tex, fence);
463 clock_gettime(CLOCK_MONOTONIC, &now);
464 double elapsed = now.tv_sec - start.tv_sec +
465 1e-9 * (now.tv_nsec - start.tv_nsec);
466 if (frame % 100 == 0) {
467 printf("%d frames in %.3f seconds = %.1f fps (%.1f ms/frame)\n",
468 frame, elapsed, frame / elapsed,
469 1e3 * elapsed / frame);
470 // chain->print_phase_timing();
473 // Reset every 100 frames, so that local variations in frame times
474 // (especially for the first few frames, when the shaders are
475 // compiled etc.) don't make it hard to measure for the entire
476 // remaining duration of the program.
477 if (frame == 10000) {
485 void Mixer::subsample_chroma(GLuint src_tex, GLuint dst_tex)
488 glGenVertexArrays(1, &vao);
497 glBindVertexArray(vao);
501 GLuint fbo = resource_pool->create_fbo(dst_tex);
502 glBindFramebuffer(GL_FRAMEBUFFER, fbo);
503 glViewport(0, 0, WIDTH/2, HEIGHT/2);
506 glUseProgram(cbcr_program_num);
509 glActiveTexture(GL_TEXTURE0);
511 glBindTexture(GL_TEXTURE_2D, src_tex);
513 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
515 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
517 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
520 float chroma_offset_0[] = { -0.5f / WIDTH, 0.0f };
521 set_uniform_vec2(cbcr_program_num, "foo", "chroma_offset_0", chroma_offset_0);
523 GLuint position_vbo = fill_vertex_attribute(cbcr_program_num, "position", 2, GL_FLOAT, sizeof(vertices), vertices);
524 GLuint texcoord_vbo = fill_vertex_attribute(cbcr_program_num, "texcoord", 2, GL_FLOAT, sizeof(vertices), vertices); // Same as vertices.
526 glDrawArrays(GL_TRIANGLES, 0, 3);
529 cleanup_vertex_attribute(cbcr_program_num, "position", position_vbo);
530 cleanup_vertex_attribute(cbcr_program_num, "texcoord", texcoord_vbo);
535 resource_pool->release_fbo(fbo);
536 glDeleteVertexArrays(1, &vao);
539 void Mixer::release_display_frame(DisplayFrame *frame)
541 resource_pool->release_2d_texture(frame->texnum);
543 frame->ready_fence.reset();
548 mixer_thread = std::thread(&Mixer::thread_func, this);
557 void Mixer::cut(Source source)
559 current_source = source;
562 void Mixer::OutputChannel::output_frame(GLuint tex, RefCountedGLsync fence)
564 // Store this frame for display. Remove the ready frame if any
565 // (it was seemingly never used).
567 std::unique_lock<std::mutex> lock(frame_mutex);
568 if (has_ready_frame) {
569 parent->release_display_frame(&ready_frame);
571 ready_frame.texnum = tex;
572 ready_frame.ready_fence = fence;
573 has_ready_frame = true;
576 if (has_new_frame_ready_callback) {
577 new_frame_ready_callback();
581 bool Mixer::OutputChannel::get_display_frame(DisplayFrame *frame)
583 std::unique_lock<std::mutex> lock(frame_mutex);
584 if (!has_current_frame && !has_ready_frame) {
588 if (has_current_frame && has_ready_frame) {
589 // We have a new ready frame. Toss the current one.
590 parent->release_display_frame(¤t_frame);
591 has_current_frame = false;
593 if (has_ready_frame) {
594 assert(!has_current_frame);
595 current_frame = ready_frame;
596 ready_frame.ready_fence.reset(); // Drop the refcount.
597 has_current_frame = true;
598 has_ready_frame = false;
601 *frame = current_frame;
605 void Mixer::OutputChannel::set_frame_ready_callback(Mixer::new_frame_ready_callback_t callback)
607 new_frame_ready_callback = callback;
608 has_new_frame_ready_callback = true;
611 void Mixer::OutputChannel::set_size(int width, int height)
614 this->height = height;