]> git.sesse.net Git - nageru/blob - mixer.cpp
Remove unused stuff from the mixer.
[nageru] / mixer.cpp
1 #define GL_GLEXT_PROTOTYPES 1
2 #define NO_SDL_GLEXT 1
3 #define NUM_CARDS 2
4
5 #define WIDTH 1280
6 #define HEIGHT 720
7
8 #include <epoxy/gl.h>
9 #include <epoxy/egl.h>
10
11 #undef Success
12
13 #include <assert.h>
14 #include <features.h>
15 #include <math.h>
16 #include <png.h>
17 #include <pngconf.h>
18 #include <setjmp.h>
19 #include <stdio.h>
20 #include <stdlib.h>
21 #include <sys/time.h>
22 #include <time.h>
23 #include <mutex>
24 #include <queue>
25 #include <condition_variable>
26
27 #include <diffusion_effect.h>
28 #include <effect.h>
29 #include <effect_chain.h>
30 #include <flat_input.h>
31 #include <image_format.h>
32 #include <init.h>
33 #include <lift_gamma_gain_effect.h>
34 #include <saturation_effect.h>
35 #include <util.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>
43 #include <ycbcr.h>
44 #include <resource_pool.h>
45 #include <effect_util.h>
46
47 #include <EGL/egl.h>
48
49 #include "h264encode.h"
50 #include "context.h"
51 #include "bmusb.h"
52 #include "pbo_frame_allocator.h"
53
54 using namespace movit;
55 using namespace std;
56 using namespace std::placeholders;
57
58 // shared between all EGL contexts
59 EGLDisplay egl_display;
60 EGLSurface egl_surface;
61 EGLConfig ecfg;
62 EGLint ctxattr[] = {
63         EGL_CONTEXT_CLIENT_VERSION, 2,
64         EGL_CONTEXT_MAJOR_VERSION_KHR, 3,
65         EGL_CONTEXT_MINOR_VERSION_KHR, 1,
66         //EGL_CONTEXT_OPENGL_PROFILE_MASK_KHR, EGL_CONTEXT_OPENGL_CORE_PROFILE_BIT_KHR,
67         EGL_CONTEXT_OPENGL_PROFILE_MASK_KHR, EGL_CONTEXT_OPENGL_COMPATIBILITY_PROFILE_BIT,
68         EGL_NONE
69 };
70
71 EGLConfig pbuffer_ecfg;
72
73 std::mutex bmusb_mutex;  // protects <cards>
74
75 struct CaptureCard {
76         BMUSBCapture *usb;
77
78         // Threading stuff
79         bool thread_initialized;
80         QSurface *surface;
81         QOpenGLContext *context;
82
83         bool new_data_ready;  // Whether new_frame contains anything.
84         PBOFrameAllocator::Frame new_frame;
85         GLsync new_data_ready_fence;  // Whether new_frame is ready for rendering.
86         std::condition_variable new_data_ready_changed;  // Set whenever new_data_ready is changed.
87 };
88 CaptureCard cards[NUM_CARDS];
89
90 void bm_frame(int card_index, uint16_t timecode,
91               FrameAllocator::Frame video_frame, size_t video_offset, uint16_t video_format,
92               FrameAllocator::Frame audio_frame, size_t audio_offset, uint16_t audio_format)
93 {
94         CaptureCard *card = &cards[card_index];
95         if (!card->thread_initialized) {
96                 printf("initializing context for bmusb thread %d\n", card_index);
97                 eglBindAPI(EGL_OPENGL_API);
98                 card->context = create_context();
99                 if (!make_current(card->context, card->surface)) {
100                         printf("failed to create bmusb context\n");
101                         exit(1);
102                 }
103                 card->thread_initialized = true;
104         }       
105
106         if (video_frame.len - video_offset != 1280 * 750 * 2) {
107                 printf("dropping frame with wrong length (%ld)\n", video_frame.len - video_offset);
108                 FILE *fp = fopen("frame.raw", "wb");
109                 fwrite(video_frame.data, video_frame.len, 1, fp);
110                 fclose(fp);
111                 //exit(1);
112                 card->usb->get_video_frame_allocator()->release_frame(video_frame);
113                 card->usb->get_audio_frame_allocator()->release_frame(audio_frame);
114                 return;
115         }
116         {
117                 // Wait until the previous frame was consumed.
118                 std::unique_lock<std::mutex> lock(bmusb_mutex);
119                 card->new_data_ready_changed.wait(lock, [card]{ return !card->new_data_ready; });
120         }
121         GLuint pbo = (GLint)(intptr_t)video_frame.userdata;
122         check_error();
123         glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, pbo);
124         check_error();
125         glFlushMappedBufferRange(GL_PIXEL_UNPACK_BUFFER, 0, video_frame.size);
126         check_error();
127         //glMemoryBarrier(GL_CLIENT_MAPPED_BUFFER_BARRIER_BIT);
128         //check_error();
129         GLsync fence = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);              
130         check_error();
131         assert(fence != nullptr);
132         {
133                 std::unique_lock<std::mutex> lock(bmusb_mutex);
134                 card->new_data_ready = true;
135                 card->new_frame = video_frame;
136                 card->new_data_ready_fence = fence;
137                 card->new_data_ready_changed.notify_all();
138         }
139
140         // Video frame will be released later.
141         card->usb->get_audio_frame_allocator()->release_frame(audio_frame);
142 }
143         
144 void place_rectangle(Effect *resample_effect, Effect *padding_effect, float x0, float y0, float x1, float y1)
145 {
146         float srcx0 = 0.0f;
147         float srcx1 = 1.0f;
148         float srcy0 = 0.0f;
149         float srcy1 = 1.0f;
150
151         // Cull.
152         if (x0 > 1280.0 || x1 < 0.0 || y0 > 720.0 || y1 < 0.0) {
153                 CHECK(resample_effect->set_int("width", 1));
154                 CHECK(resample_effect->set_int("height", 1));
155                 CHECK(resample_effect->set_float("zoom_x", 1280.0));
156                 CHECK(resample_effect->set_float("zoom_y", 720.0));
157                 CHECK(padding_effect->set_int("left", 2000));
158                 CHECK(padding_effect->set_int("top", 2000));
159                 return; 
160         }
161
162         // Clip. (TODO: Clip on upper/left sides, too.)
163         if (x1 > 1280.0) {
164                 srcx1 = (1280.0 - x0) / (x1 - x0);
165                 x1 = 1280.0;
166         }
167         if (y1 > 720.0) {
168                 srcy1 = (720.0 - y0) / (y1 - y0);
169                 y1 = 720.0;
170         }
171
172         float x_subpixel_offset = x0 - floor(x0);
173         float y_subpixel_offset = y0 - floor(y0);
174
175         // Resampling must be to an integral number of pixels. Round up,
176         // and then add an extra pixel so we have some leeway for the border.
177         int width = int(ceil(x1 - x0)) + 1;
178         int height = int(ceil(y1 - y0)) + 1;
179         CHECK(resample_effect->set_int("width", width));
180         CHECK(resample_effect->set_int("height", height));
181
182         // Correct the discrepancy with zoom. (This will leave a small
183         // excess edge of pixels and subpixels, which we'll correct for soon.)
184         float zoom_x = (x1 - x0) / (width * (srcx1 - srcx0));
185         float zoom_y = (y1 - y0) / (height * (srcy1 - srcy0));
186         CHECK(resample_effect->set_float("zoom_x", zoom_x));
187         CHECK(resample_effect->set_float("zoom_y", zoom_y));
188         CHECK(resample_effect->set_float("zoom_center_x", 0.0f));
189         CHECK(resample_effect->set_float("zoom_center_y", 0.0f));
190
191         // Padding must also be to a whole-pixel offset.
192         CHECK(padding_effect->set_int("left", floor(x0)));
193         CHECK(padding_effect->set_int("top", floor(y0)));
194
195         // Correct _that_ discrepancy by subpixel offset in the resampling.
196         CHECK(resample_effect->set_float("left", -x_subpixel_offset / zoom_x));
197         CHECK(resample_effect->set_float("top", -y_subpixel_offset / zoom_y));
198
199         // Finally, adjust the border so it is exactly where we want it.
200         CHECK(padding_effect->set_float("border_offset_left", x_subpixel_offset));
201         CHECK(padding_effect->set_float("border_offset_right", x1 - (floor(x0) + width)));
202         CHECK(padding_effect->set_float("border_offset_top", y_subpixel_offset));
203         CHECK(padding_effect->set_float("border_offset_bottom", y1 - (floor(y0) + height)));
204 }
205         
206 static bool quit = false;
207         
208 void mixer_thread_func(QSurface *surface, QSurface *surface2, QSurface *surface3, QSurface *surface4)
209 {
210         cards[0].surface = surface3;
211 #if NUM_CARDS == 2
212         cards[1].surface = surface4;
213 #endif
214
215         eglBindAPI(EGL_OPENGL_API);
216         QOpenGLContext *context = create_context();
217         if (!make_current(context, surface)) {
218                 printf("oops\n");
219                 exit(1);
220         }
221
222         CHECK(init_movit("/usr/share/movit", MOVIT_DEBUG_ON));
223         check_error();
224
225         EffectChain chain(WIDTH, HEIGHT);
226         check_error();
227
228         ImageFormat inout_format;
229         inout_format.color_space = COLORSPACE_sRGB;
230         inout_format.gamma_curve = GAMMA_sRGB;
231
232         YCbCrFormat ycbcr_format;
233         ycbcr_format.chroma_subsampling_x = 2;
234         ycbcr_format.chroma_subsampling_y = 1;
235         ycbcr_format.cb_x_position = 0.0;
236         ycbcr_format.cr_x_position = 0.0;
237         ycbcr_format.cb_y_position = 0.5;
238         ycbcr_format.cr_y_position = 0.5;
239         ycbcr_format.luma_coefficients = YCBCR_REC_601;
240         ycbcr_format.full_range = false;
241
242         YCbCrInput *input[NUM_CARDS];
243
244         input[0] = new YCbCrInput(inout_format, ycbcr_format, WIDTH, HEIGHT, YCBCR_INPUT_SPLIT_Y_AND_CBCR);
245         chain.add_input(input[0]);
246         input[1] = new YCbCrInput(inout_format, ycbcr_format, WIDTH, HEIGHT, YCBCR_INPUT_SPLIT_Y_AND_CBCR);
247         chain.add_input(input[1]);
248         Effect *resample_effect = chain.add_effect(new ResampleEffect(), input[0]);
249         Effect *padding_effect = chain.add_effect(new IntegralPaddingEffect());
250         float border_color[] = { 0.0f, 0.0f, 0.0f, 1.0f };
251         CHECK(padding_effect->set_vec4("border_color", border_color));
252
253         Effect *resample2_effect = chain.add_effect(new ResampleEffect(), input[1]);
254         Effect *saturation_effect = chain.add_effect(new SaturationEffect());
255         CHECK(saturation_effect->set_float("saturation", 0.3f));
256         Effect *wb_effect = chain.add_effect(new WhiteBalanceEffect());
257         CHECK(wb_effect->set_float("output_color_temperature", 3500.0));
258         Effect *padding2_effect = chain.add_effect(new IntegralPaddingEffect());
259
260         chain.add_effect(new OverlayEffect(), padding_effect, padding2_effect);
261
262         ycbcr_format.chroma_subsampling_x = 1;
263
264         chain.add_ycbcr_output(inout_format, OUTPUT_ALPHA_FORMAT_POSTMULTIPLIED, ycbcr_format, YCBCR_OUTPUT_SPLIT_Y_AND_CBCR);
265         chain.set_dither_bits(8);
266         chain.set_output_origin(OUTPUT_ORIGIN_TOP_LEFT);
267         chain.finalize();
268
269         H264Encoder h264_encoder(surface2, WIDTH, HEIGHT, "test.mp4");
270
271         printf("Configuring first card...\n");
272         cards[0].usb = new BMUSBCapture(0x1edb, 0xbd3b);  // 0xbd4f
273         //cards[0].usb = new BMUSBCapture(0x1edb, 0xbd4f);
274         cards[0].usb->set_frame_callback(std::bind(bm_frame, 0, _1, _2, _3, _4, _5, _6, _7));
275         std::unique_ptr<PBOFrameAllocator> pbo_allocator1(new PBOFrameAllocator(1280 * 750 * 2 + 44));
276         cards[0].usb->set_video_frame_allocator(pbo_allocator1.get());
277         cards[0].usb->configure_card();
278
279         std::unique_ptr<PBOFrameAllocator> pbo_allocator2(new PBOFrameAllocator(1280 * 750 * 2 + 44));
280         if (NUM_CARDS == 2) {
281                 printf("Configuring second card...\n");
282                 cards[1].usb = new BMUSBCapture(0x1edb, 0xbd4f);
283                 cards[1].usb->set_frame_callback(std::bind(bm_frame, 1, _1, _2, _3, _4, _5, _6, _7));
284                 cards[1].usb->set_video_frame_allocator(pbo_allocator2.get());
285                 cards[1].usb->configure_card();
286         }
287
288         BMUSBCapture::start_bm_thread();
289
290         for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
291                 cards[card_index].usb->start_bm_capture();
292         }
293
294         int frame = 0;
295 #if _POSIX_C_SOURCE >= 199309L
296         struct timespec start, now;
297         clock_gettime(CLOCK_MONOTONIC, &start);
298 #else
299         struct timeval start, now;
300         gettimeofday(&start, NULL);
301 #endif
302
303         PBOFrameAllocator::Frame bmusb_current_rendering_frame[NUM_CARDS];
304         for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
305                 bmusb_current_rendering_frame[card_index] =
306                         cards[card_index].usb->get_video_frame_allocator()->alloc_frame();
307                 GLint input_tex_pbo = (GLint)(intptr_t)bmusb_current_rendering_frame[card_index].userdata;
308                 input[card_index]->set_pixel_data(0, nullptr, input_tex_pbo);
309                 input[card_index]->set_pixel_data(1, nullptr, input_tex_pbo);
310                 check_error();
311         }
312
313         //chain.enable_phase_timing(true);
314
315         // Set up stuff for NV12 conversion.
316         ResourcePool *resource_pool = chain.get_resource_pool();
317         GLuint chroma_tex = resource_pool->create_2d_texture(GL_RG8, WIDTH, HEIGHT);
318
319         // Cb/Cr shader.
320         string cbcr_vert_shader = read_file("vs-cbcr.130.vert");
321         string cbcr_frag_shader =
322                 "#version 130 \n"
323                 "in vec2 tc0; \n"
324                 "uniform sampler2D cbcr_tex; \n"
325                 "void main() { \n"
326                 "    gl_FragColor = texture2D(cbcr_tex, tc0); \n"
327                 "} \n";
328         GLuint cbcr_program_num = resource_pool->compile_glsl_program(cbcr_vert_shader, cbcr_frag_shader);
329
330         GLuint vao;
331         glGenVertexArrays(1, &vao);
332         check_error();
333
334         while (!quit) {
335                 ++frame;
336
337                 //int width0 = lrintf(848 * (1.0 + 0.2 * sin(frame * 0.02)));
338                 int width0 = 848;
339                 int height0 = lrintf(width0 * 9.0 / 16.0);
340
341                 //float top0 = 96 + 48 * sin(frame * 0.005);
342                 //float left0 = 96 + 48 * cos(frame * 0.006);
343                 float top0 = 48;
344                 float left0 = 16;
345                 float bottom0 = top0 + height0;
346                 float right0 = left0 + width0;
347
348                 int width1 = 384;
349                 int height1 = 216;
350         
351                 float bottom1 = 720 - 48;
352                 float right1 = 1280 - 16;
353                 float top1 = bottom1 - height1;
354                 float left1 = right1 - width1;
355                 
356                 float t = 0.5 + 0.5 * cos(frame * 0.006);
357                 //float t = 0.0;
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);
361
362                 top0 = top0 * scale0 + ty0;
363                 bottom0 = bottom0 * scale0 + ty0;
364                 left0 = left0 * scale0 + tx0;
365                 right0 = right0 * scale0 + tx0;
366
367                 top1 = top1 * scale0 + ty0;
368                 bottom1 = bottom1 * scale0 + ty0;
369                 left1 = left1 * scale0 + tx0;
370                 right1 = right1 * scale0 + tx0;
371
372                 place_rectangle(resample_effect, padding_effect, left0, top0, right0, bottom0);
373                 place_rectangle(resample2_effect, padding2_effect, left1, top1, right1, bottom1);
374
375                 CaptureCard card_copy[NUM_CARDS];
376
377                 {
378                         std::unique_lock<std::mutex> lock(bmusb_mutex);
379
380                         // The first card is the master timer, so wait for it to have a new frame.
381                         // TODO: Make configurable, and with a timeout.
382                         cards[0].new_data_ready_changed.wait(lock, []{ return cards[0].new_data_ready; });
383
384                         for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
385                                 CaptureCard *card = &cards[card_index];
386                                 card_copy[card_index].usb = card->usb;
387                                 card_copy[card_index].new_data_ready = card->new_data_ready;
388                                 card_copy[card_index].new_frame = card->new_frame;
389                                 card_copy[card_index].new_data_ready_fence = card->new_data_ready_fence;
390                                 card->new_data_ready = false;
391                                 card->new_data_ready_changed.notify_all();
392                         }
393                 }
394         
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)
398                                 continue;
399
400                         // FIXME: We could still be rendering from it! 
401                         card->usb->get_video_frame_allocator()->release_frame(bmusb_current_rendering_frame[card_index]);
402                         bmusb_current_rendering_frame[card_index] = card->new_frame;
403
404                         // The new texture might still be uploaded,
405                         // tell the GPU to wait until it's there.
406                         if (card->new_data_ready_fence)
407                                 glWaitSync(card->new_data_ready_fence, /*flags=*/0, GL_TIMEOUT_IGNORED);
408                         check_error();
409                         glDeleteSync(card->new_data_ready_fence);
410                         check_error();
411                         GLint input_tex_pbo = (GLint)(intptr_t)bmusb_current_rendering_frame[card_index].userdata;
412                         input[card_index]->set_pixel_data(0, (unsigned char *)BUFFER_OFFSET((1280 * 750 * 2 + 44) / 2 + 1280 * 25 + 22), input_tex_pbo);
413                         input[card_index]->set_pixel_data(1, (unsigned char *)BUFFER_OFFSET(1280 * 25 + 22), input_tex_pbo);
414
415                         if (NUM_CARDS == 1) {
416                                 // Set to the other one, too.
417                                 input[1]->set_pixel_data(0, (unsigned char *)BUFFER_OFFSET((1280 * 750 * 2 + 44) / 2 + 1280 * 25 + 22), input_tex_pbo);
418                                 input[1]->set_pixel_data(1, (unsigned char *)BUFFER_OFFSET(1280 * 25 + 22), input_tex_pbo);
419                         }
420                 }
421
422                 GLuint y_tex, cbcr_tex;
423                 bool got_frame = h264_encoder.begin_frame(&y_tex, &cbcr_tex);
424                 assert(got_frame);
425
426                 // Render chain.
427                 {
428                         GLuint ycbcr_fbo = resource_pool->create_fbo(y_tex, chroma_tex);
429                         chain.render_to_fbo(ycbcr_fbo, WIDTH, HEIGHT);
430                         resource_pool->release_fbo(ycbcr_fbo);
431                 }
432
433                 // Set up for extraction.
434                 float vertices[] = {
435                         0.0f, 2.0f,
436                         0.0f, 0.0f,
437                         2.0f, 0.0f
438                 };
439
440                 glBindVertexArray(vao);
441                 check_error();
442
443                 // Extract Cb/Cr.
444                 GLuint cbcr_fbo = resource_pool->create_fbo(cbcr_tex);
445                 glBindFramebuffer(GL_FRAMEBUFFER, cbcr_fbo);
446                 glViewport(0, 0, WIDTH/2, HEIGHT/2);
447                 check_error();
448                 GLsync fence;
449                 {
450                         glUseProgram(cbcr_program_num);
451                         check_error();
452
453                         glActiveTexture(GL_TEXTURE0);
454                         check_error();
455                         glBindTexture(GL_TEXTURE_2D, chroma_tex);
456                         check_error();
457                         glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
458                         check_error();
459                         glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
460                         check_error();
461                         glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
462                         check_error();
463
464                         float chroma_offset_0[] = { -0.5f / WIDTH, 0.0f };
465                         set_uniform_vec2(cbcr_program_num, "foo", "chroma_offset_0", chroma_offset_0);
466
467                         GLuint position_vbo = fill_vertex_attribute(cbcr_program_num, "position", 2, GL_FLOAT, sizeof(vertices), vertices);
468                         GLuint texcoord_vbo = fill_vertex_attribute(cbcr_program_num, "texcoord", 2, GL_FLOAT, sizeof(vertices), vertices);  // Same as vertices.
469
470                         glDrawArrays(GL_TRIANGLES, 0, 3);
471                         check_error();
472
473                         cleanup_vertex_attribute(cbcr_program_num, "position", position_vbo);
474                         cleanup_vertex_attribute(cbcr_program_num, "texcoord", texcoord_vbo);
475
476                         glUseProgram(0);
477                         check_error();
478
479                         fence = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);              
480                         check_error();
481
482                         resource_pool->release_fbo(cbcr_fbo);
483                 }
484
485                 h264_encoder.end_frame(fence);
486
487
488 #if 1
489 #if _POSIX_C_SOURCE >= 199309L
490                 clock_gettime(CLOCK_MONOTONIC, &now);
491                 double elapsed = now.tv_sec - start.tv_sec +
492                         1e-9 * (now.tv_nsec - start.tv_nsec);
493 #else
494                 gettimeofday(&now, NULL);
495                 double elapsed = now.tv_sec - start.tv_sec +
496                         1e-6 * (now.tv_usec - start.tv_usec);
497 #endif
498                 if (frame % 100 == 0) {
499                         printf("%d frames in %.3f seconds = %.1f fps (%.1f ms/frame)\n",
500                                 frame, elapsed, frame / elapsed,
501                                 1e3 * elapsed / frame);
502                 //      chain.print_phase_timing();
503                 }
504
505                 // Reset every 100 frames, so that local variations in frame times
506                 // (especially for the first few frames, when the shaders are
507                 // compiled etc.) don't make it hard to measure for the entire
508                 // remaining duration of the program.
509                 if (frame == 10000) {
510                         frame = 0;
511                         start = now;
512                 }
513 #endif
514         }
515         glDeleteVertexArrays(1, &vao);
516         resource_pool->release_glsl_program(cbcr_program_num);
517         resource_pool->release_2d_texture(chroma_tex);
518         BMUSBCapture::stop_bm_thread();
519 }
520
521 std::thread mixer_thread;
522
523 void start_mixer(QSurface *surface, QSurface *surface2, QSurface *surface3, QSurface *surface4)
524 {
525         mixer_thread = std::thread([surface, surface2, surface3, surface4]{
526                 mixer_thread_func(surface, surface2, surface3, surface4);
527         });
528 }
529
530 void mixer_quit()
531 {
532         quit = true;
533         mixer_thread.join();
534 }