1 #define GL_GLEXT_PROTOTYPES 1
14 #include <effect_chain.h>
15 #include <effect_util.h>
16 #include <epoxy/egl.h>
18 #include <image_format.h>
20 #include <overlay_effect.h>
21 #include <padding_effect.h>
22 #include <resample_effect.h>
23 #include <resource_pool.h>
24 #include <saturation_effect.h>
31 #include <white_balance_effect.h>
33 #include <ycbcr_input.h>
35 #include <condition_variable>
45 #include "h264encode.h"
46 #include "pbo_frame_allocator.h"
47 #include "ref_counted_gl_sync.h"
51 using namespace movit;
53 using namespace std::placeholders;
55 Source current_source = SOURCE_INPUT1;
57 ResourcePool *resource_pool;
59 std::mutex display_frame_mutex;
60 DisplayFrame current_display_frame, ready_display_frame; // protected by <frame_mutex>
61 bool has_current_display_frame = false, has_ready_display_frame = false; // protected by <frame_mutex>
63 std::mutex bmusb_mutex;
68 bool thread_initialized;
70 QOpenGLContext *context;
72 bool new_data_ready; // Whether new_frame contains anything.
73 PBOFrameAllocator::Frame new_frame;
74 GLsync new_data_ready_fence; // Whether new_frame is ready for rendering.
75 std::condition_variable new_data_ready_changed; // Set whenever new_data_ready is changed.
77 CaptureCard cards[NUM_CARDS]; // protected by <bmusb_mutex>
79 new_frame_ready_callback_t new_frame_ready_callback;
80 bool has_new_frame_ready_callback = false;
82 void bm_frame(int card_index, uint16_t timecode,
83 FrameAllocator::Frame video_frame, size_t video_offset, uint16_t video_format,
84 FrameAllocator::Frame audio_frame, size_t audio_offset, uint16_t audio_format)
86 CaptureCard *card = &cards[card_index];
87 if (!card->thread_initialized) {
88 printf("initializing context for bmusb thread %d\n", card_index);
89 eglBindAPI(EGL_OPENGL_API);
90 card->context = create_context();
91 if (!make_current(card->context, card->surface)) {
92 printf("failed to create bmusb context\n");
95 card->thread_initialized = true;
98 if (video_frame.len - video_offset != 1280 * 750 * 2) {
99 printf("dropping frame with wrong length (%ld)\n", video_frame.len - video_offset);
100 FILE *fp = fopen("frame.raw", "wb");
101 fwrite(video_frame.data, video_frame.len, 1, fp);
104 card->usb->get_video_frame_allocator()->release_frame(video_frame);
105 card->usb->get_audio_frame_allocator()->release_frame(audio_frame);
109 // Wait until the previous frame was consumed.
110 std::unique_lock<std::mutex> lock(bmusb_mutex);
111 card->new_data_ready_changed.wait(lock, [card]{ return !card->new_data_ready; });
113 GLuint pbo = (GLint)(intptr_t)video_frame.userdata;
115 glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, pbo);
117 glFlushMappedBufferRange(GL_PIXEL_UNPACK_BUFFER, 0, video_frame.size);
119 //glMemoryBarrier(GL_CLIENT_MAPPED_BUFFER_BARRIER_BIT);
121 GLsync fence = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);
123 assert(fence != nullptr);
125 std::unique_lock<std::mutex> lock(bmusb_mutex);
126 card->new_data_ready = true;
127 card->new_frame = video_frame;
128 card->new_data_ready_fence = fence;
129 card->new_data_ready_changed.notify_all();
132 // Video frame will be released later.
133 card->usb->get_audio_frame_allocator()->release_frame(audio_frame);
136 void place_rectangle(Effect *resample_effect, Effect *padding_effect, float x0, float y0, float x1, float y1)
144 if (x0 > 1280.0 || x1 < 0.0 || y0 > 720.0 || y1 < 0.0) {
145 CHECK(resample_effect->set_int("width", 1));
146 CHECK(resample_effect->set_int("height", 1));
147 CHECK(resample_effect->set_float("zoom_x", 1280.0));
148 CHECK(resample_effect->set_float("zoom_y", 720.0));
149 CHECK(padding_effect->set_int("left", 2000));
150 CHECK(padding_effect->set_int("top", 2000));
154 // Clip. (TODO: Clip on upper/left sides, too.)
156 srcx1 = (1280.0 - x0) / (x1 - x0);
160 srcy1 = (720.0 - y0) / (y1 - y0);
164 float x_subpixel_offset = x0 - floor(x0);
165 float y_subpixel_offset = y0 - floor(y0);
167 // Resampling must be to an integral number of pixels. Round up,
168 // and then add an extra pixel so we have some leeway for the border.
169 int width = int(ceil(x1 - x0)) + 1;
170 int height = int(ceil(y1 - y0)) + 1;
171 CHECK(resample_effect->set_int("width", width));
172 CHECK(resample_effect->set_int("height", height));
174 // Correct the discrepancy with zoom. (This will leave a small
175 // excess edge of pixels and subpixels, which we'll correct for soon.)
176 float zoom_x = (x1 - x0) / (width * (srcx1 - srcx0));
177 float zoom_y = (y1 - y0) / (height * (srcy1 - srcy0));
178 CHECK(resample_effect->set_float("zoom_x", zoom_x));
179 CHECK(resample_effect->set_float("zoom_y", zoom_y));
180 CHECK(resample_effect->set_float("zoom_center_x", 0.0f));
181 CHECK(resample_effect->set_float("zoom_center_y", 0.0f));
183 // Padding must also be to a whole-pixel offset.
184 CHECK(padding_effect->set_int("left", floor(x0)));
185 CHECK(padding_effect->set_int("top", floor(y0)));
187 // Correct _that_ discrepancy by subpixel offset in the resampling.
188 CHECK(resample_effect->set_float("left", -x_subpixel_offset / zoom_x));
189 CHECK(resample_effect->set_float("top", -y_subpixel_offset / zoom_y));
191 // Finally, adjust the border so it is exactly where we want it.
192 CHECK(padding_effect->set_float("border_offset_left", x_subpixel_offset));
193 CHECK(padding_effect->set_float("border_offset_right", x1 - (floor(x0) + width)));
194 CHECK(padding_effect->set_float("border_offset_top", y_subpixel_offset));
195 CHECK(padding_effect->set_float("border_offset_bottom", y1 - (floor(y0) + height)));
198 static bool quit = false;
200 void mixer_thread_func(QSurface *surface, QSurface *surface2, QSurface *surface3, QSurface *surface4)
202 cards[0].surface = surface3;
204 cards[1].surface = surface4;
207 eglBindAPI(EGL_OPENGL_API);
208 QOpenGLContext *context = create_context();
209 if (!make_current(context, surface)) {
214 CHECK(init_movit(MOVIT_SHADER_DIR, MOVIT_DEBUG_OFF));
217 EffectChain chain(WIDTH, HEIGHT);
220 ImageFormat inout_format;
221 inout_format.color_space = COLORSPACE_sRGB;
222 inout_format.gamma_curve = GAMMA_sRGB;
224 YCbCrFormat ycbcr_format;
225 ycbcr_format.chroma_subsampling_x = 2;
226 ycbcr_format.chroma_subsampling_y = 1;
227 ycbcr_format.cb_x_position = 0.0;
228 ycbcr_format.cr_x_position = 0.0;
229 ycbcr_format.cb_y_position = 0.5;
230 ycbcr_format.cr_y_position = 0.5;
231 ycbcr_format.luma_coefficients = YCBCR_REC_601;
232 ycbcr_format.full_range = false;
234 YCbCrInput *input[NUM_CARDS];
236 input[0] = new YCbCrInput(inout_format, ycbcr_format, WIDTH, HEIGHT, YCBCR_INPUT_SPLIT_Y_AND_CBCR);
237 chain.add_input(input[0]);
238 input[1] = new YCbCrInput(inout_format, ycbcr_format, WIDTH, HEIGHT, YCBCR_INPUT_SPLIT_Y_AND_CBCR);
239 chain.add_input(input[1]);
240 Effect *resample_effect = chain.add_effect(new ResampleEffect(), input[0]);
241 Effect *padding_effect = chain.add_effect(new IntegralPaddingEffect());
242 float border_color[] = { 0.0f, 0.0f, 0.0f, 1.0f };
243 CHECK(padding_effect->set_vec4("border_color", border_color));
245 Effect *resample2_effect = chain.add_effect(new ResampleEffect(), input[1]);
246 Effect *saturation_effect = chain.add_effect(new SaturationEffect());
247 CHECK(saturation_effect->set_float("saturation", 0.3f));
248 Effect *wb_effect = chain.add_effect(new WhiteBalanceEffect());
249 CHECK(wb_effect->set_float("output_color_temperature", 3500.0));
250 Effect *padding2_effect = chain.add_effect(new IntegralPaddingEffect());
252 chain.add_effect(new OverlayEffect(), padding_effect, padding2_effect);
254 ycbcr_format.chroma_subsampling_x = 1;
256 chain.add_output(inout_format, OUTPUT_ALPHA_FORMAT_POSTMULTIPLIED);
257 chain.add_ycbcr_output(inout_format, OUTPUT_ALPHA_FORMAT_POSTMULTIPLIED, ycbcr_format, YCBCR_OUTPUT_SPLIT_Y_AND_CBCR);
258 chain.set_dither_bits(8);
259 chain.set_output_origin(OUTPUT_ORIGIN_TOP_LEFT);
262 H264Encoder h264_encoder(surface2, WIDTH, HEIGHT, "test.mp4");
264 printf("Configuring first card...\n");
265 cards[0].usb = new BMUSBCapture(0x1edb, 0xbd3b); // 0xbd4f
266 //cards[0].usb = new BMUSBCapture(0x1edb, 0xbd4f);
267 cards[0].usb->set_frame_callback(std::bind(bm_frame, 0, _1, _2, _3, _4, _5, _6, _7));
268 std::unique_ptr<PBOFrameAllocator> pbo_allocator1(new PBOFrameAllocator(1280 * 750 * 2 + 44));
269 cards[0].usb->set_video_frame_allocator(pbo_allocator1.get());
270 cards[0].usb->configure_card();
272 std::unique_ptr<PBOFrameAllocator> pbo_allocator2(new PBOFrameAllocator(1280 * 750 * 2 + 44));
273 if (NUM_CARDS == 2) {
274 printf("Configuring second card...\n");
275 cards[1].usb = new BMUSBCapture(0x1edb, 0xbd4f);
276 cards[1].usb->set_frame_callback(std::bind(bm_frame, 1, _1, _2, _3, _4, _5, _6, _7));
277 cards[1].usb->set_video_frame_allocator(pbo_allocator2.get());
278 cards[1].usb->configure_card();
281 BMUSBCapture::start_bm_thread();
283 for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
284 cards[card_index].usb->start_bm_capture();
288 #if _POSIX_C_SOURCE >= 199309L
289 struct timespec start, now;
290 clock_gettime(CLOCK_MONOTONIC, &start);
292 struct timeval start, now;
293 gettimeofday(&start, NULL);
296 PBOFrameAllocator::Frame bmusb_current_rendering_frame[NUM_CARDS];
297 for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
298 bmusb_current_rendering_frame[card_index] =
299 cards[card_index].usb->get_video_frame_allocator()->alloc_frame();
300 GLint input_tex_pbo = (GLint)(intptr_t)bmusb_current_rendering_frame[card_index].userdata;
301 input[card_index]->set_pixel_data(0, nullptr, input_tex_pbo);
302 input[card_index]->set_pixel_data(1, nullptr, input_tex_pbo);
305 //chain.enable_phase_timing(true);
307 // Set up stuff for NV12 conversion.
308 resource_pool = chain.get_resource_pool();
309 GLuint chroma_tex = resource_pool->create_2d_texture(GL_RG8, WIDTH, HEIGHT);
312 string cbcr_vert_shader = read_file("vs-cbcr.130.vert");
313 string cbcr_frag_shader =
316 "uniform sampler2D cbcr_tex; \n"
318 " gl_FragColor = texture2D(cbcr_tex, tc0); \n"
320 GLuint cbcr_program_num = resource_pool->compile_glsl_program(cbcr_vert_shader, cbcr_frag_shader);
323 glGenVertexArrays(1, &vao);
329 //int width0 = lrintf(848 * (1.0 + 0.2 * sin(frame * 0.02)));
331 int height0 = lrintf(width0 * 9.0 / 16.0);
333 //float top0 = 96 + 48 * sin(frame * 0.005);
334 //float left0 = 96 + 48 * cos(frame * 0.006);
337 float bottom0 = top0 + height0;
338 float right0 = left0 + width0;
343 float bottom1 = 720 - 48;
344 float right1 = 1280 - 16;
345 float top1 = bottom1 - height1;
346 float left1 = right1 - width1;
348 if (current_source == SOURCE_INPUT1) {
355 bottom1 = HEIGHT + 20;
358 } else if (current_source == SOURCE_INPUT2) {
365 bottom0 = HEIGHT + 20;
369 float t = 0.5 + 0.5 * cos(frame * 0.006);
370 float scale0 = 1.0 + t * (1280.0 / 848.0 - 1.0);
371 float tx0 = 0.0 + t * (-16.0 * scale0);
372 float ty0 = 0.0 + t * (-48.0 * scale0);
374 top0 = top0 * scale0 + ty0;
375 bottom0 = bottom0 * scale0 + ty0;
376 left0 = left0 * scale0 + tx0;
377 right0 = right0 * scale0 + tx0;
379 top1 = top1 * scale0 + ty0;
380 bottom1 = bottom1 * scale0 + ty0;
381 left1 = left1 * scale0 + tx0;
382 right1 = right1 * scale0 + tx0;
385 place_rectangle(resample_effect, padding_effect, left0, top0, right0, bottom0);
386 place_rectangle(resample2_effect, padding2_effect, left1, top1, right1, bottom1);
388 CaptureCard card_copy[NUM_CARDS];
391 std::unique_lock<std::mutex> lock(bmusb_mutex);
393 // The first card is the master timer, so wait for it to have a new frame.
394 // TODO: Make configurable, and with a timeout.
395 cards[0].new_data_ready_changed.wait(lock, []{ return cards[0].new_data_ready; });
397 for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
398 CaptureCard *card = &cards[card_index];
399 card_copy[card_index].usb = card->usb;
400 card_copy[card_index].new_data_ready = card->new_data_ready;
401 card_copy[card_index].new_frame = card->new_frame;
402 card_copy[card_index].new_data_ready_fence = card->new_data_ready_fence;
403 card->new_data_ready = false;
404 card->new_data_ready_changed.notify_all();
408 vector<FrameAllocator::Frame> input_frames_to_release;
410 for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
411 CaptureCard *card = &card_copy[card_index];
412 if (!card->new_data_ready)
415 // Now we're done with the previous frame, so we can definitely
416 // release it when this is done rendering. (Actually, we could do
417 // it one frame earlier, but before we have a new one, there's no
418 // knowing when the current one is released.)
419 input_frames_to_release.push_back(bmusb_current_rendering_frame[card_index]);
420 bmusb_current_rendering_frame[card_index] = card->new_frame;
423 // The new texture might still be uploaded,
424 // tell the GPU to wait until it's there.
425 if (card->new_data_ready_fence)
426 glWaitSync(card->new_data_ready_fence, /*flags=*/0, GL_TIMEOUT_IGNORED);
428 glDeleteSync(card->new_data_ready_fence);
430 GLint input_tex_pbo = (GLint)(intptr_t)card->new_frame.userdata;
431 input[card_index]->set_pixel_data(0, (unsigned char *)BUFFER_OFFSET((1280 * 750 * 2 + 44) / 2 + 1280 * 25 + 22), input_tex_pbo);
432 input[card_index]->set_pixel_data(1, (unsigned char *)BUFFER_OFFSET(1280 * 25 + 22), input_tex_pbo);
434 if (NUM_CARDS == 1) {
435 // Set to the other one, too.
436 input[1]->set_pixel_data(0, (unsigned char *)BUFFER_OFFSET((1280 * 750 * 2 + 44) / 2 + 1280 * 25 + 22), input_tex_pbo);
437 input[1]->set_pixel_data(1, (unsigned char *)BUFFER_OFFSET(1280 * 25 + 22), input_tex_pbo);
441 GLuint y_tex, cbcr_tex;
442 bool got_frame = h264_encoder.begin_frame(&y_tex, &cbcr_tex);
446 GLuint rgba_tex = resource_pool->create_2d_texture(GL_RGBA8, WIDTH, HEIGHT);
447 GLuint ycbcr_fbo = resource_pool->create_fbo(y_tex, chroma_tex, rgba_tex);
448 chain.render_to_fbo(ycbcr_fbo, WIDTH, HEIGHT);
449 resource_pool->release_fbo(ycbcr_fbo);
451 // Set up for extraction.
458 glBindVertexArray(vao);
462 GLuint cbcr_fbo = resource_pool->create_fbo(cbcr_tex);
463 glBindFramebuffer(GL_FRAMEBUFFER, cbcr_fbo);
464 glViewport(0, 0, WIDTH/2, HEIGHT/2);
467 glUseProgram(cbcr_program_num);
470 glActiveTexture(GL_TEXTURE0);
472 glBindTexture(GL_TEXTURE_2D, chroma_tex);
474 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
476 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
478 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
481 float chroma_offset_0[] = { -0.5f / WIDTH, 0.0f };
482 set_uniform_vec2(cbcr_program_num, "foo", "chroma_offset_0", chroma_offset_0);
484 GLuint position_vbo = fill_vertex_attribute(cbcr_program_num, "position", 2, GL_FLOAT, sizeof(vertices), vertices);
485 GLuint texcoord_vbo = fill_vertex_attribute(cbcr_program_num, "texcoord", 2, GL_FLOAT, sizeof(vertices), vertices); // Same as vertices.
487 glDrawArrays(GL_TRIANGLES, 0, 3);
490 RefCountedGLsync fence(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);
493 cleanup_vertex_attribute(cbcr_program_num, "position", position_vbo);
494 cleanup_vertex_attribute(cbcr_program_num, "texcoord", texcoord_vbo);
499 resource_pool->release_fbo(cbcr_fbo);
501 h264_encoder.end_frame(fence, input_frames_to_release);
503 // Store this frame for display. Remove the ready frame if any
504 // (it was seemingly never used).
506 std::unique_lock<std::mutex> lock(display_frame_mutex);
507 if (has_ready_display_frame) {
508 resource_pool->release_2d_texture(ready_display_frame.texnum);
509 ready_display_frame.ready_fence.reset();
511 ready_display_frame.texnum = rgba_tex;
512 ready_display_frame.ready_fence = fence;
513 has_ready_display_frame = true;
516 if (has_new_frame_ready_callback) {
517 new_frame_ready_callback();
521 #if _POSIX_C_SOURCE >= 199309L
522 clock_gettime(CLOCK_MONOTONIC, &now);
523 double elapsed = now.tv_sec - start.tv_sec +
524 1e-9 * (now.tv_nsec - start.tv_nsec);
526 gettimeofday(&now, NULL);
527 double elapsed = now.tv_sec - start.tv_sec +
528 1e-6 * (now.tv_usec - start.tv_usec);
530 if (frame % 100 == 0) {
531 printf("%d frames in %.3f seconds = %.1f fps (%.1f ms/frame)\n",
532 frame, elapsed, frame / elapsed,
533 1e3 * elapsed / frame);
534 // chain.print_phase_timing();
537 // Reset every 100 frames, so that local variations in frame times
538 // (especially for the first few frames, when the shaders are
539 // compiled etc.) don't make it hard to measure for the entire
540 // remaining duration of the program.
541 if (frame == 10000) {
548 glDeleteVertexArrays(1, &vao);
549 resource_pool->release_glsl_program(cbcr_program_num);
550 resource_pool->release_2d_texture(chroma_tex);
551 BMUSBCapture::stop_bm_thread();
554 bool mixer_get_display_frame(DisplayFrame *frame)
556 std::unique_lock<std::mutex> lock(display_frame_mutex);
557 if (!has_current_display_frame && !has_ready_display_frame) {
561 if (has_current_display_frame && has_ready_display_frame) {
562 // We have a new ready frame. Toss the current one.
563 resource_pool->release_2d_texture(current_display_frame.texnum);
564 current_display_frame.ready_fence.reset();
565 has_current_display_frame = false;
567 if (has_ready_display_frame) {
568 assert(!has_current_display_frame);
569 current_display_frame = ready_display_frame;
570 ready_display_frame.ready_fence.reset(); // Drop the refcount.
571 has_current_display_frame = true;
572 has_ready_display_frame = false;
575 *frame = current_display_frame;
579 void set_frame_ready_fallback(new_frame_ready_callback_t callback)
581 new_frame_ready_callback = callback;
582 has_new_frame_ready_callback = true;
585 std::thread mixer_thread;
587 void start_mixer(QSurface *surface, QSurface *surface2, QSurface *surface3, QSurface *surface4)
589 mixer_thread = std::thread([surface, surface2, surface3, surface4]{
590 mixer_thread_func(surface, surface2, surface3, surface4);
600 void mixer_cut(Source source)
602 current_source = source;