1 #define GL_GLEXT_PROTOTYPES 1
25 #include <condition_variable>
27 #include <diffusion_effect.h>
29 #include <effect_chain.h>
30 #include <flat_input.h>
31 #include <image_format.h>
33 #include <lift_gamma_gain_effect.h>
34 #include <saturation_effect.h>
36 #include <ycbcr_input.h>
37 #include <vignette_effect.h>
38 #include <resample_effect.h>
39 #include <resize_effect.h>
40 #include <overlay_effect.h>
41 #include <padding_effect.h>
42 #include <white_balance_effect.h>
44 #include <resource_pool.h>
45 #include <effect_util.h>
49 #include "h264encode.h"
52 #include "pbo_frame_allocator.h"
54 #include "ref_counted_gl_sync.h"
56 using namespace movit;
58 using namespace std::placeholders;
60 static float t = 0.0f;
62 ResourcePool *resource_pool;
64 std::mutex display_frame_mutex;
65 DisplayFrame current_display_frame, ready_display_frame; // protected by <frame_mutex>
66 bool has_current_display_frame = false, has_ready_display_frame = false; // protected by <frame_mutex>
68 std::mutex bmusb_mutex;
73 bool thread_initialized;
75 QOpenGLContext *context;
77 bool new_data_ready; // Whether new_frame contains anything.
78 PBOFrameAllocator::Frame new_frame;
79 GLsync new_data_ready_fence; // Whether new_frame is ready for rendering.
80 std::condition_variable new_data_ready_changed; // Set whenever new_data_ready is changed.
82 CaptureCard cards[NUM_CARDS]; // protected by <bmusb_mutex>
84 new_frame_ready_callback_t new_frame_ready_callback;
85 bool has_new_frame_ready_callback = false;
87 void bm_frame(int card_index, uint16_t timecode,
88 FrameAllocator::Frame video_frame, size_t video_offset, uint16_t video_format,
89 FrameAllocator::Frame audio_frame, size_t audio_offset, uint16_t audio_format)
91 CaptureCard *card = &cards[card_index];
92 if (!card->thread_initialized) {
93 printf("initializing context for bmusb thread %d\n", card_index);
94 eglBindAPI(EGL_OPENGL_API);
95 card->context = create_context();
96 if (!make_current(card->context, card->surface)) {
97 printf("failed to create bmusb context\n");
100 card->thread_initialized = true;
103 if (video_frame.len - video_offset != 1280 * 750 * 2) {
104 printf("dropping frame with wrong length (%ld)\n", video_frame.len - video_offset);
105 FILE *fp = fopen("frame.raw", "wb");
106 fwrite(video_frame.data, video_frame.len, 1, fp);
109 card->usb->get_video_frame_allocator()->release_frame(video_frame);
110 card->usb->get_audio_frame_allocator()->release_frame(audio_frame);
114 // Wait until the previous frame was consumed.
115 std::unique_lock<std::mutex> lock(bmusb_mutex);
116 card->new_data_ready_changed.wait(lock, [card]{ return !card->new_data_ready; });
118 GLuint pbo = (GLint)(intptr_t)video_frame.userdata;
120 glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, pbo);
122 glFlushMappedBufferRange(GL_PIXEL_UNPACK_BUFFER, 0, video_frame.size);
124 //glMemoryBarrier(GL_CLIENT_MAPPED_BUFFER_BARRIER_BIT);
126 GLsync fence = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);
128 assert(fence != nullptr);
130 std::unique_lock<std::mutex> lock(bmusb_mutex);
131 card->new_data_ready = true;
132 card->new_frame = video_frame;
133 card->new_data_ready_fence = fence;
134 card->new_data_ready_changed.notify_all();
137 // Video frame will be released later.
138 card->usb->get_audio_frame_allocator()->release_frame(audio_frame);
141 void place_rectangle(Effect *resample_effect, Effect *padding_effect, float x0, float y0, float x1, float y1)
149 if (x0 > 1280.0 || x1 < 0.0 || y0 > 720.0 || y1 < 0.0) {
150 CHECK(resample_effect->set_int("width", 1));
151 CHECK(resample_effect->set_int("height", 1));
152 CHECK(resample_effect->set_float("zoom_x", 1280.0));
153 CHECK(resample_effect->set_float("zoom_y", 720.0));
154 CHECK(padding_effect->set_int("left", 2000));
155 CHECK(padding_effect->set_int("top", 2000));
159 // Clip. (TODO: Clip on upper/left sides, too.)
161 srcx1 = (1280.0 - x0) / (x1 - x0);
165 srcy1 = (720.0 - y0) / (y1 - y0);
169 float x_subpixel_offset = x0 - floor(x0);
170 float y_subpixel_offset = y0 - floor(y0);
172 // Resampling must be to an integral number of pixels. Round up,
173 // and then add an extra pixel so we have some leeway for the border.
174 int width = int(ceil(x1 - x0)) + 1;
175 int height = int(ceil(y1 - y0)) + 1;
176 CHECK(resample_effect->set_int("width", width));
177 CHECK(resample_effect->set_int("height", height));
179 // Correct the discrepancy with zoom. (This will leave a small
180 // excess edge of pixels and subpixels, which we'll correct for soon.)
181 float zoom_x = (x1 - x0) / (width * (srcx1 - srcx0));
182 float zoom_y = (y1 - y0) / (height * (srcy1 - srcy0));
183 CHECK(resample_effect->set_float("zoom_x", zoom_x));
184 CHECK(resample_effect->set_float("zoom_y", zoom_y));
185 CHECK(resample_effect->set_float("zoom_center_x", 0.0f));
186 CHECK(resample_effect->set_float("zoom_center_y", 0.0f));
188 // Padding must also be to a whole-pixel offset.
189 CHECK(padding_effect->set_int("left", floor(x0)));
190 CHECK(padding_effect->set_int("top", floor(y0)));
192 // Correct _that_ discrepancy by subpixel offset in the resampling.
193 CHECK(resample_effect->set_float("left", -x_subpixel_offset / zoom_x));
194 CHECK(resample_effect->set_float("top", -y_subpixel_offset / zoom_y));
196 // Finally, adjust the border so it is exactly where we want it.
197 CHECK(padding_effect->set_float("border_offset_left", x_subpixel_offset));
198 CHECK(padding_effect->set_float("border_offset_right", x1 - (floor(x0) + width)));
199 CHECK(padding_effect->set_float("border_offset_top", y_subpixel_offset));
200 CHECK(padding_effect->set_float("border_offset_bottom", y1 - (floor(y0) + height)));
203 static bool quit = false;
205 void mixer_thread_func(QSurface *surface, QSurface *surface2, QSurface *surface3, QSurface *surface4)
207 cards[0].surface = surface3;
209 cards[1].surface = surface4;
212 eglBindAPI(EGL_OPENGL_API);
213 QOpenGLContext *context = create_context();
214 if (!make_current(context, surface)) {
219 CHECK(init_movit(MOVIT_SHADER_DIR, MOVIT_DEBUG_OFF));
222 EffectChain chain(WIDTH, HEIGHT);
225 ImageFormat inout_format;
226 inout_format.color_space = COLORSPACE_sRGB;
227 inout_format.gamma_curve = GAMMA_sRGB;
229 YCbCrFormat ycbcr_format;
230 ycbcr_format.chroma_subsampling_x = 2;
231 ycbcr_format.chroma_subsampling_y = 1;
232 ycbcr_format.cb_x_position = 0.0;
233 ycbcr_format.cr_x_position = 0.0;
234 ycbcr_format.cb_y_position = 0.5;
235 ycbcr_format.cr_y_position = 0.5;
236 ycbcr_format.luma_coefficients = YCBCR_REC_601;
237 ycbcr_format.full_range = false;
239 YCbCrInput *input[NUM_CARDS];
241 input[0] = new YCbCrInput(inout_format, ycbcr_format, WIDTH, HEIGHT, YCBCR_INPUT_SPLIT_Y_AND_CBCR);
242 chain.add_input(input[0]);
243 input[1] = new YCbCrInput(inout_format, ycbcr_format, WIDTH, HEIGHT, YCBCR_INPUT_SPLIT_Y_AND_CBCR);
244 chain.add_input(input[1]);
245 Effect *resample_effect = chain.add_effect(new ResampleEffect(), input[0]);
246 Effect *padding_effect = chain.add_effect(new IntegralPaddingEffect());
247 float border_color[] = { 0.0f, 0.0f, 0.0f, 1.0f };
248 CHECK(padding_effect->set_vec4("border_color", border_color));
250 Effect *resample2_effect = chain.add_effect(new ResampleEffect(), input[1]);
251 Effect *saturation_effect = chain.add_effect(new SaturationEffect());
252 CHECK(saturation_effect->set_float("saturation", 0.3f));
253 Effect *wb_effect = chain.add_effect(new WhiteBalanceEffect());
254 CHECK(wb_effect->set_float("output_color_temperature", 3500.0));
255 Effect *padding2_effect = chain.add_effect(new IntegralPaddingEffect());
257 chain.add_effect(new OverlayEffect(), padding_effect, padding2_effect);
259 ycbcr_format.chroma_subsampling_x = 1;
261 chain.add_output(inout_format, OUTPUT_ALPHA_FORMAT_POSTMULTIPLIED);
262 chain.add_ycbcr_output(inout_format, OUTPUT_ALPHA_FORMAT_POSTMULTIPLIED, ycbcr_format, YCBCR_OUTPUT_SPLIT_Y_AND_CBCR);
263 chain.set_dither_bits(8);
264 chain.set_output_origin(OUTPUT_ORIGIN_TOP_LEFT);
267 H264Encoder h264_encoder(surface2, WIDTH, HEIGHT, "test.mp4");
269 printf("Configuring first card...\n");
270 cards[0].usb = new BMUSBCapture(0x1edb, 0xbd3b); // 0xbd4f
271 //cards[0].usb = new BMUSBCapture(0x1edb, 0xbd4f);
272 cards[0].usb->set_frame_callback(std::bind(bm_frame, 0, _1, _2, _3, _4, _5, _6, _7));
273 std::unique_ptr<PBOFrameAllocator> pbo_allocator1(new PBOFrameAllocator(1280 * 750 * 2 + 44));
274 cards[0].usb->set_video_frame_allocator(pbo_allocator1.get());
275 cards[0].usb->configure_card();
277 std::unique_ptr<PBOFrameAllocator> pbo_allocator2(new PBOFrameAllocator(1280 * 750 * 2 + 44));
278 if (NUM_CARDS == 2) {
279 printf("Configuring second card...\n");
280 cards[1].usb = new BMUSBCapture(0x1edb, 0xbd4f);
281 cards[1].usb->set_frame_callback(std::bind(bm_frame, 1, _1, _2, _3, _4, _5, _6, _7));
282 cards[1].usb->set_video_frame_allocator(pbo_allocator2.get());
283 cards[1].usb->configure_card();
286 BMUSBCapture::start_bm_thread();
288 for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
289 cards[card_index].usb->start_bm_capture();
293 #if _POSIX_C_SOURCE >= 199309L
294 struct timespec start, now;
295 clock_gettime(CLOCK_MONOTONIC, &start);
297 struct timeval start, now;
298 gettimeofday(&start, NULL);
301 PBOFrameAllocator::Frame bmusb_current_rendering_frame[NUM_CARDS];
302 for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
303 bmusb_current_rendering_frame[card_index] =
304 cards[card_index].usb->get_video_frame_allocator()->alloc_frame();
305 GLint input_tex_pbo = (GLint)(intptr_t)bmusb_current_rendering_frame[card_index].userdata;
306 input[card_index]->set_pixel_data(0, nullptr, input_tex_pbo);
307 input[card_index]->set_pixel_data(1, nullptr, input_tex_pbo);
311 //chain.enable_phase_timing(true);
313 // Set up stuff for NV12 conversion.
314 resource_pool = chain.get_resource_pool();
315 GLuint chroma_tex = resource_pool->create_2d_texture(GL_RG8, WIDTH, HEIGHT);
318 string cbcr_vert_shader = read_file("vs-cbcr.130.vert");
319 string cbcr_frag_shader =
322 "uniform sampler2D cbcr_tex; \n"
324 " gl_FragColor = texture2D(cbcr_tex, tc0); \n"
326 GLuint cbcr_program_num = resource_pool->compile_glsl_program(cbcr_vert_shader, cbcr_frag_shader);
329 glGenVertexArrays(1, &vao);
335 //int width0 = lrintf(848 * (1.0 + 0.2 * sin(frame * 0.02)));
337 int height0 = lrintf(width0 * 9.0 / 16.0);
339 //float top0 = 96 + 48 * sin(frame * 0.005);
340 //float left0 = 96 + 48 * cos(frame * 0.006);
343 float bottom0 = top0 + height0;
344 float right0 = left0 + width0;
349 float bottom1 = 720 - 48;
350 float right1 = 1280 - 16;
351 float top1 = bottom1 - height1;
352 float left1 = right1 - width1;
354 float t = 0.5 + 0.5 * cos(frame * 0.006);
356 float scale0 = 1.0 + t * (1280.0 / 848.0 - 1.0);
357 float tx0 = 0.0 + t * (-16.0 * scale0);
358 float ty0 = 0.0 + t * (-48.0 * scale0);
360 top0 = top0 * scale0 + ty0;
361 bottom0 = bottom0 * scale0 + ty0;
362 left0 = left0 * scale0 + tx0;
363 right0 = right0 * scale0 + tx0;
365 top1 = top1 * scale0 + ty0;
366 bottom1 = bottom1 * scale0 + ty0;
367 left1 = left1 * scale0 + tx0;
368 right1 = right1 * scale0 + tx0;
370 place_rectangle(resample_effect, padding_effect, left0, top0, right0, bottom0);
371 place_rectangle(resample2_effect, padding2_effect, left1, top1, right1, bottom1);
373 CaptureCard card_copy[NUM_CARDS];
376 std::unique_lock<std::mutex> lock(bmusb_mutex);
378 // The first card is the master timer, so wait for it to have a new frame.
379 // TODO: Make configurable, and with a timeout.
380 cards[0].new_data_ready_changed.wait(lock, []{ return cards[0].new_data_ready; });
382 for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
383 CaptureCard *card = &cards[card_index];
384 card_copy[card_index].usb = card->usb;
385 card_copy[card_index].new_data_ready = card->new_data_ready;
386 card_copy[card_index].new_frame = card->new_frame;
387 card_copy[card_index].new_data_ready_fence = card->new_data_ready_fence;
388 card->new_data_ready = false;
389 card->new_data_ready_changed.notify_all();
393 vector<FrameAllocator::Frame> input_frames_to_release;
395 for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
396 CaptureCard *card = &card_copy[card_index];
397 if (!card->new_data_ready)
400 // Now we're done with the previous frame, so we can definitely
401 // release it when this is done rendering. (Actually, we could do
402 // it one frame earlier, but before we have a new one, there's no
403 // knowing when the current one is released.)
404 input_frames_to_release.push_back(bmusb_current_rendering_frame[card_index]);
405 bmusb_current_rendering_frame[card_index] = card->new_frame;
407 // The new texture might still be uploaded,
408 // tell the GPU to wait until it's there.
409 if (card->new_data_ready_fence)
410 glWaitSync(card->new_data_ready_fence, /*flags=*/0, GL_TIMEOUT_IGNORED);
412 glDeleteSync(card->new_data_ready_fence);
414 GLint input_tex_pbo = (GLint)(intptr_t)card->new_frame.userdata;
415 input[card_index]->set_pixel_data(0, (unsigned char *)BUFFER_OFFSET((1280 * 750 * 2 + 44) / 2 + 1280 * 25 + 22), input_tex_pbo);
416 input[card_index]->set_pixel_data(1, (unsigned char *)BUFFER_OFFSET(1280 * 25 + 22), input_tex_pbo);
418 if (NUM_CARDS == 1) {
419 // Set to the other one, too.
420 input[1]->set_pixel_data(0, (unsigned char *)BUFFER_OFFSET((1280 * 750 * 2 + 44) / 2 + 1280 * 25 + 22), input_tex_pbo);
421 input[1]->set_pixel_data(1, (unsigned char *)BUFFER_OFFSET(1280 * 25 + 22), input_tex_pbo);
425 GLuint y_tex, cbcr_tex;
426 bool got_frame = h264_encoder.begin_frame(&y_tex, &cbcr_tex);
430 GLuint rgba_tex = resource_pool->create_2d_texture(GL_RGBA8, WIDTH, HEIGHT);
431 GLuint ycbcr_fbo = resource_pool->create_fbo(y_tex, chroma_tex, rgba_tex);
432 chain.render_to_fbo(ycbcr_fbo, WIDTH, HEIGHT);
433 resource_pool->release_fbo(ycbcr_fbo);
435 // Set up for extraction.
442 glBindVertexArray(vao);
446 GLuint cbcr_fbo = resource_pool->create_fbo(cbcr_tex);
447 glBindFramebuffer(GL_FRAMEBUFFER, cbcr_fbo);
448 glViewport(0, 0, WIDTH/2, HEIGHT/2);
451 glUseProgram(cbcr_program_num);
454 glActiveTexture(GL_TEXTURE0);
456 glBindTexture(GL_TEXTURE_2D, chroma_tex);
458 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
460 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
462 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
465 float chroma_offset_0[] = { -0.5f / WIDTH, 0.0f };
466 set_uniform_vec2(cbcr_program_num, "foo", "chroma_offset_0", chroma_offset_0);
468 GLuint position_vbo = fill_vertex_attribute(cbcr_program_num, "position", 2, GL_FLOAT, sizeof(vertices), vertices);
469 GLuint texcoord_vbo = fill_vertex_attribute(cbcr_program_num, "texcoord", 2, GL_FLOAT, sizeof(vertices), vertices); // Same as vertices.
471 glDrawArrays(GL_TRIANGLES, 0, 3);
474 RefCountedGLsync fence(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);
477 cleanup_vertex_attribute(cbcr_program_num, "position", position_vbo);
478 cleanup_vertex_attribute(cbcr_program_num, "texcoord", texcoord_vbo);
483 resource_pool->release_fbo(cbcr_fbo);
485 h264_encoder.end_frame(fence, input_frames_to_release);
487 // Store this frame for display. Remove the ready frame if any
488 // (it was seemingly never used).
490 std::unique_lock<std::mutex> lock(display_frame_mutex);
491 if (has_ready_display_frame) {
492 resource_pool->release_2d_texture(ready_display_frame.texnum);
493 ready_display_frame.ready_fence.reset();
495 ready_display_frame.texnum = rgba_tex;
496 ready_display_frame.ready_fence = fence;
497 has_ready_display_frame = true;
500 if (has_new_frame_ready_callback) {
501 new_frame_ready_callback();
505 #if _POSIX_C_SOURCE >= 199309L
506 clock_gettime(CLOCK_MONOTONIC, &now);
507 double elapsed = now.tv_sec - start.tv_sec +
508 1e-9 * (now.tv_nsec - start.tv_nsec);
510 gettimeofday(&now, NULL);
511 double elapsed = now.tv_sec - start.tv_sec +
512 1e-6 * (now.tv_usec - start.tv_usec);
514 if (frame % 100 == 0) {
515 printf("%d frames in %.3f seconds = %.1f fps (%.1f ms/frame)\n",
516 frame, elapsed, frame / elapsed,
517 1e3 * elapsed / frame);
518 // chain.print_phase_timing();
521 // Reset every 100 frames, so that local variations in frame times
522 // (especially for the first few frames, when the shaders are
523 // compiled etc.) don't make it hard to measure for the entire
524 // remaining duration of the program.
525 if (frame == 10000) {
531 glDeleteVertexArrays(1, &vao);
532 resource_pool->release_glsl_program(cbcr_program_num);
533 resource_pool->release_2d_texture(chroma_tex);
534 BMUSBCapture::stop_bm_thread();
537 bool mixer_get_display_frame(DisplayFrame *frame)
539 std::unique_lock<std::mutex> lock(display_frame_mutex);
540 if (!has_current_display_frame && !has_ready_display_frame) {
544 if (has_current_display_frame && has_ready_display_frame) {
545 // We have a new ready frame. Toss the current one.
546 resource_pool->release_2d_texture(current_display_frame.texnum);
547 current_display_frame.ready_fence.reset();
548 has_current_display_frame = false;
550 if (has_ready_display_frame) {
551 assert(!has_current_display_frame);
552 current_display_frame = ready_display_frame;
553 ready_display_frame.ready_fence.reset(); // Drop the refcount.
554 has_current_display_frame = true;
555 has_ready_display_frame = false;
558 *frame = current_display_frame;
562 void set_frame_ready_fallback(new_frame_ready_callback_t callback)
564 new_frame_ready_callback = callback;
565 has_new_frame_ready_callback = true;
568 std::thread mixer_thread;
570 void start_mixer(QSurface *surface, QSurface *surface2, QSurface *surface3, QSurface *surface4)
572 mixer_thread = std::thread([surface, surface2, surface3, surface4]{
573 mixer_thread_func(surface, surface2, surface3, surface4);