]> git.sesse.net Git - nageru/blob - mixer.cpp
Add a wrapper to make fences refcounted, as we will soon be needing to split them...
[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 #include "ref_counted_gl_sync.h"
54
55 using namespace movit;
56 using namespace std;
57 using namespace std::placeholders;
58
59
60 std::mutex bmusb_mutex;  // protects <cards>
61
62 struct CaptureCard {
63         BMUSBCapture *usb;
64
65         // Threading stuff
66         bool thread_initialized;
67         QSurface *surface;
68         QOpenGLContext *context;
69
70         bool new_data_ready;  // Whether new_frame contains anything.
71         PBOFrameAllocator::Frame new_frame;
72         GLsync new_data_ready_fence;  // Whether new_frame is ready for rendering.
73         std::condition_variable new_data_ready_changed;  // Set whenever new_data_ready is changed.
74 };
75 CaptureCard cards[NUM_CARDS];
76
77 void bm_frame(int card_index, uint16_t timecode,
78               FrameAllocator::Frame video_frame, size_t video_offset, uint16_t video_format,
79               FrameAllocator::Frame audio_frame, size_t audio_offset, uint16_t audio_format)
80 {
81         CaptureCard *card = &cards[card_index];
82         if (!card->thread_initialized) {
83                 printf("initializing context for bmusb thread %d\n", card_index);
84                 eglBindAPI(EGL_OPENGL_API);
85                 card->context = create_context();
86                 if (!make_current(card->context, card->surface)) {
87                         printf("failed to create bmusb context\n");
88                         exit(1);
89                 }
90                 card->thread_initialized = true;
91         }       
92
93         if (video_frame.len - video_offset != 1280 * 750 * 2) {
94                 printf("dropping frame with wrong length (%ld)\n", video_frame.len - video_offset);
95                 FILE *fp = fopen("frame.raw", "wb");
96                 fwrite(video_frame.data, video_frame.len, 1, fp);
97                 fclose(fp);
98                 //exit(1);
99                 card->usb->get_video_frame_allocator()->release_frame(video_frame);
100                 card->usb->get_audio_frame_allocator()->release_frame(audio_frame);
101                 return;
102         }
103         {
104                 // Wait until the previous frame was consumed.
105                 std::unique_lock<std::mutex> lock(bmusb_mutex);
106                 card->new_data_ready_changed.wait(lock, [card]{ return !card->new_data_ready; });
107         }
108         GLuint pbo = (GLint)(intptr_t)video_frame.userdata;
109         check_error();
110         glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, pbo);
111         check_error();
112         glFlushMappedBufferRange(GL_PIXEL_UNPACK_BUFFER, 0, video_frame.size);
113         check_error();
114         //glMemoryBarrier(GL_CLIENT_MAPPED_BUFFER_BARRIER_BIT);
115         //check_error();
116         GLsync fence = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);              
117         check_error();
118         assert(fence != nullptr);
119         {
120                 std::unique_lock<std::mutex> lock(bmusb_mutex);
121                 card->new_data_ready = true;
122                 card->new_frame = video_frame;
123                 card->new_data_ready_fence = fence;
124                 card->new_data_ready_changed.notify_all();
125         }
126
127         // Video frame will be released later.
128         card->usb->get_audio_frame_allocator()->release_frame(audio_frame);
129 }
130         
131 void place_rectangle(Effect *resample_effect, Effect *padding_effect, float x0, float y0, float x1, float y1)
132 {
133         float srcx0 = 0.0f;
134         float srcx1 = 1.0f;
135         float srcy0 = 0.0f;
136         float srcy1 = 1.0f;
137
138         // Cull.
139         if (x0 > 1280.0 || x1 < 0.0 || y0 > 720.0 || y1 < 0.0) {
140                 CHECK(resample_effect->set_int("width", 1));
141                 CHECK(resample_effect->set_int("height", 1));
142                 CHECK(resample_effect->set_float("zoom_x", 1280.0));
143                 CHECK(resample_effect->set_float("zoom_y", 720.0));
144                 CHECK(padding_effect->set_int("left", 2000));
145                 CHECK(padding_effect->set_int("top", 2000));
146                 return; 
147         }
148
149         // Clip. (TODO: Clip on upper/left sides, too.)
150         if (x1 > 1280.0) {
151                 srcx1 = (1280.0 - x0) / (x1 - x0);
152                 x1 = 1280.0;
153         }
154         if (y1 > 720.0) {
155                 srcy1 = (720.0 - y0) / (y1 - y0);
156                 y1 = 720.0;
157         }
158
159         float x_subpixel_offset = x0 - floor(x0);
160         float y_subpixel_offset = y0 - floor(y0);
161
162         // Resampling must be to an integral number of pixels. Round up,
163         // and then add an extra pixel so we have some leeway for the border.
164         int width = int(ceil(x1 - x0)) + 1;
165         int height = int(ceil(y1 - y0)) + 1;
166         CHECK(resample_effect->set_int("width", width));
167         CHECK(resample_effect->set_int("height", height));
168
169         // Correct the discrepancy with zoom. (This will leave a small
170         // excess edge of pixels and subpixels, which we'll correct for soon.)
171         float zoom_x = (x1 - x0) / (width * (srcx1 - srcx0));
172         float zoom_y = (y1 - y0) / (height * (srcy1 - srcy0));
173         CHECK(resample_effect->set_float("zoom_x", zoom_x));
174         CHECK(resample_effect->set_float("zoom_y", zoom_y));
175         CHECK(resample_effect->set_float("zoom_center_x", 0.0f));
176         CHECK(resample_effect->set_float("zoom_center_y", 0.0f));
177
178         // Padding must also be to a whole-pixel offset.
179         CHECK(padding_effect->set_int("left", floor(x0)));
180         CHECK(padding_effect->set_int("top", floor(y0)));
181
182         // Correct _that_ discrepancy by subpixel offset in the resampling.
183         CHECK(resample_effect->set_float("left", -x_subpixel_offset / zoom_x));
184         CHECK(resample_effect->set_float("top", -y_subpixel_offset / zoom_y));
185
186         // Finally, adjust the border so it is exactly where we want it.
187         CHECK(padding_effect->set_float("border_offset_left", x_subpixel_offset));
188         CHECK(padding_effect->set_float("border_offset_right", x1 - (floor(x0) + width)));
189         CHECK(padding_effect->set_float("border_offset_top", y_subpixel_offset));
190         CHECK(padding_effect->set_float("border_offset_bottom", y1 - (floor(y0) + height)));
191 }
192         
193 static bool quit = false;
194         
195 void mixer_thread_func(QSurface *surface, QSurface *surface2, QSurface *surface3, QSurface *surface4)
196 {
197         cards[0].surface = surface3;
198 #if NUM_CARDS == 2
199         cards[1].surface = surface4;
200 #endif
201
202         eglBindAPI(EGL_OPENGL_API);
203         QOpenGLContext *context = create_context();
204         if (!make_current(context, surface)) {
205                 printf("oops\n");
206                 exit(1);
207         }
208
209         CHECK(init_movit(MOVIT_SHADER_DIR, MOVIT_DEBUG_OFF));
210         check_error();
211
212         EffectChain chain(WIDTH, HEIGHT);
213         check_error();
214
215         ImageFormat inout_format;
216         inout_format.color_space = COLORSPACE_sRGB;
217         inout_format.gamma_curve = GAMMA_sRGB;
218
219         YCbCrFormat ycbcr_format;
220         ycbcr_format.chroma_subsampling_x = 2;
221         ycbcr_format.chroma_subsampling_y = 1;
222         ycbcr_format.cb_x_position = 0.0;
223         ycbcr_format.cr_x_position = 0.0;
224         ycbcr_format.cb_y_position = 0.5;
225         ycbcr_format.cr_y_position = 0.5;
226         ycbcr_format.luma_coefficients = YCBCR_REC_601;
227         ycbcr_format.full_range = false;
228
229         YCbCrInput *input[NUM_CARDS];
230
231         input[0] = new YCbCrInput(inout_format, ycbcr_format, WIDTH, HEIGHT, YCBCR_INPUT_SPLIT_Y_AND_CBCR);
232         chain.add_input(input[0]);
233         input[1] = new YCbCrInput(inout_format, ycbcr_format, WIDTH, HEIGHT, YCBCR_INPUT_SPLIT_Y_AND_CBCR);
234         chain.add_input(input[1]);
235         Effect *resample_effect = chain.add_effect(new ResampleEffect(), input[0]);
236         Effect *padding_effect = chain.add_effect(new IntegralPaddingEffect());
237         float border_color[] = { 0.0f, 0.0f, 0.0f, 1.0f };
238         CHECK(padding_effect->set_vec4("border_color", border_color));
239
240         Effect *resample2_effect = chain.add_effect(new ResampleEffect(), input[1]);
241         Effect *saturation_effect = chain.add_effect(new SaturationEffect());
242         CHECK(saturation_effect->set_float("saturation", 0.3f));
243         Effect *wb_effect = chain.add_effect(new WhiteBalanceEffect());
244         CHECK(wb_effect->set_float("output_color_temperature", 3500.0));
245         Effect *padding2_effect = chain.add_effect(new IntegralPaddingEffect());
246
247         chain.add_effect(new OverlayEffect(), padding_effect, padding2_effect);
248
249         ycbcr_format.chroma_subsampling_x = 1;
250
251         chain.add_ycbcr_output(inout_format, OUTPUT_ALPHA_FORMAT_POSTMULTIPLIED, ycbcr_format, YCBCR_OUTPUT_SPLIT_Y_AND_CBCR);
252         chain.set_dither_bits(8);
253         chain.set_output_origin(OUTPUT_ORIGIN_TOP_LEFT);
254         chain.finalize();
255
256         H264Encoder h264_encoder(surface2, WIDTH, HEIGHT, "test.mp4");
257
258         printf("Configuring first card...\n");
259         cards[0].usb = new BMUSBCapture(0x1edb, 0xbd3b);  // 0xbd4f
260         //cards[0].usb = new BMUSBCapture(0x1edb, 0xbd4f);
261         cards[0].usb->set_frame_callback(std::bind(bm_frame, 0, _1, _2, _3, _4, _5, _6, _7));
262         std::unique_ptr<PBOFrameAllocator> pbo_allocator1(new PBOFrameAllocator(1280 * 750 * 2 + 44));
263         cards[0].usb->set_video_frame_allocator(pbo_allocator1.get());
264         cards[0].usb->configure_card();
265
266         std::unique_ptr<PBOFrameAllocator> pbo_allocator2(new PBOFrameAllocator(1280 * 750 * 2 + 44));
267         if (NUM_CARDS == 2) {
268                 printf("Configuring second card...\n");
269                 cards[1].usb = new BMUSBCapture(0x1edb, 0xbd4f);
270                 cards[1].usb->set_frame_callback(std::bind(bm_frame, 1, _1, _2, _3, _4, _5, _6, _7));
271                 cards[1].usb->set_video_frame_allocator(pbo_allocator2.get());
272                 cards[1].usb->configure_card();
273         }
274
275         BMUSBCapture::start_bm_thread();
276
277         for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
278                 cards[card_index].usb->start_bm_capture();
279         }
280
281         int frame = 0;
282 #if _POSIX_C_SOURCE >= 199309L
283         struct timespec start, now;
284         clock_gettime(CLOCK_MONOTONIC, &start);
285 #else
286         struct timeval start, now;
287         gettimeofday(&start, NULL);
288 #endif
289
290         PBOFrameAllocator::Frame bmusb_current_rendering_frame[NUM_CARDS];
291         for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
292                 bmusb_current_rendering_frame[card_index] =
293                         cards[card_index].usb->get_video_frame_allocator()->alloc_frame();
294                 GLint input_tex_pbo = (GLint)(intptr_t)bmusb_current_rendering_frame[card_index].userdata;
295                 input[card_index]->set_pixel_data(0, nullptr, input_tex_pbo);
296                 input[card_index]->set_pixel_data(1, nullptr, input_tex_pbo);
297                 check_error();
298         }
299
300         //chain.enable_phase_timing(true);
301
302         // Set up stuff for NV12 conversion.
303         ResourcePool *resource_pool = chain.get_resource_pool();
304         GLuint chroma_tex = resource_pool->create_2d_texture(GL_RG8, WIDTH, HEIGHT);
305
306         // Cb/Cr shader.
307         string cbcr_vert_shader = read_file("vs-cbcr.130.vert");
308         string cbcr_frag_shader =
309                 "#version 130 \n"
310                 "in vec2 tc0; \n"
311                 "uniform sampler2D cbcr_tex; \n"
312                 "void main() { \n"
313                 "    gl_FragColor = texture2D(cbcr_tex, tc0); \n"
314                 "} \n";
315         GLuint cbcr_program_num = resource_pool->compile_glsl_program(cbcr_vert_shader, cbcr_frag_shader);
316
317         GLuint vao;
318         glGenVertexArrays(1, &vao);
319         check_error();
320
321         while (!quit) {
322                 ++frame;
323
324                 //int width0 = lrintf(848 * (1.0 + 0.2 * sin(frame * 0.02)));
325                 int width0 = 848;
326                 int height0 = lrintf(width0 * 9.0 / 16.0);
327
328                 //float top0 = 96 + 48 * sin(frame * 0.005);
329                 //float left0 = 96 + 48 * cos(frame * 0.006);
330                 float top0 = 48;
331                 float left0 = 16;
332                 float bottom0 = top0 + height0;
333                 float right0 = left0 + width0;
334
335                 int width1 = 384;
336                 int height1 = 216;
337         
338                 float bottom1 = 720 - 48;
339                 float right1 = 1280 - 16;
340                 float top1 = bottom1 - height1;
341                 float left1 = right1 - width1;
342                 
343                 float t = 0.5 + 0.5 * cos(frame * 0.006);
344                 //float t = 0.0;
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);
348
349                 top0 = top0 * scale0 + ty0;
350                 bottom0 = bottom0 * scale0 + ty0;
351                 left0 = left0 * scale0 + tx0;
352                 right0 = right0 * scale0 + tx0;
353
354                 top1 = top1 * scale0 + ty0;
355                 bottom1 = bottom1 * scale0 + ty0;
356                 left1 = left1 * scale0 + tx0;
357                 right1 = right1 * scale0 + tx0;
358
359                 place_rectangle(resample_effect, padding_effect, left0, top0, right0, bottom0);
360                 place_rectangle(resample2_effect, padding2_effect, left1, top1, right1, bottom1);
361
362                 CaptureCard card_copy[NUM_CARDS];
363
364                 {
365                         std::unique_lock<std::mutex> lock(bmusb_mutex);
366
367                         // The first card is the master timer, so wait for it to have a new frame.
368                         // TODO: Make configurable, and with a timeout.
369                         cards[0].new_data_ready_changed.wait(lock, []{ return cards[0].new_data_ready; });
370
371                         for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
372                                 CaptureCard *card = &cards[card_index];
373                                 card_copy[card_index].usb = card->usb;
374                                 card_copy[card_index].new_data_ready = card->new_data_ready;
375                                 card_copy[card_index].new_frame = card->new_frame;
376                                 card_copy[card_index].new_data_ready_fence = card->new_data_ready_fence;
377                                 card->new_data_ready = false;
378                                 card->new_data_ready_changed.notify_all();
379                         }
380                 }
381
382                 vector<FrameAllocator::Frame> input_frames_to_release;
383         
384                 for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
385                         CaptureCard *card = &card_copy[card_index];
386                         if (!card->new_data_ready)
387                                 continue;
388
389                         // Now we're done with the previous frame, so we can definitely
390                         // release it when this is done rendering. (Actually, we could do
391                         // it one frame earlier, but before we have a new one, there's no
392                         // knowing when the current one is released.)
393                         input_frames_to_release.push_back(bmusb_current_rendering_frame[card_index]);
394                         bmusb_current_rendering_frame[card_index] = card->new_frame;
395
396                         // The new texture might still be uploaded,
397                         // tell the GPU to wait until it's there.
398                         if (card->new_data_ready_fence)
399                                 glWaitSync(card->new_data_ready_fence, /*flags=*/0, GL_TIMEOUT_IGNORED);
400                         check_error();
401                         glDeleteSync(card->new_data_ready_fence);
402                         check_error();
403                         GLint input_tex_pbo = (GLint)(intptr_t)card->new_frame.userdata;
404                         input[card_index]->set_pixel_data(0, (unsigned char *)BUFFER_OFFSET((1280 * 750 * 2 + 44) / 2 + 1280 * 25 + 22), input_tex_pbo);
405                         input[card_index]->set_pixel_data(1, (unsigned char *)BUFFER_OFFSET(1280 * 25 + 22), input_tex_pbo);
406
407                         if (NUM_CARDS == 1) {
408                                 // Set to the other one, too.
409                                 input[1]->set_pixel_data(0, (unsigned char *)BUFFER_OFFSET((1280 * 750 * 2 + 44) / 2 + 1280 * 25 + 22), input_tex_pbo);
410                                 input[1]->set_pixel_data(1, (unsigned char *)BUFFER_OFFSET(1280 * 25 + 22), input_tex_pbo);
411                         }
412                 }
413
414                 GLuint y_tex, cbcr_tex;
415                 bool got_frame = h264_encoder.begin_frame(&y_tex, &cbcr_tex);
416                 assert(got_frame);
417
418                 // Render chain.
419                 {
420                         GLuint ycbcr_fbo = resource_pool->create_fbo(y_tex, chroma_tex);
421                         chain.render_to_fbo(ycbcr_fbo, WIDTH, HEIGHT);
422                         resource_pool->release_fbo(ycbcr_fbo);
423                 }
424
425                 // Set up for extraction.
426                 float vertices[] = {
427                         0.0f, 2.0f,
428                         0.0f, 0.0f,
429                         2.0f, 0.0f
430                 };
431
432                 glBindVertexArray(vao);
433                 check_error();
434
435                 // Extract Cb/Cr.
436                 GLuint cbcr_fbo = resource_pool->create_fbo(cbcr_tex);
437                 glBindFramebuffer(GL_FRAMEBUFFER, cbcr_fbo);
438                 glViewport(0, 0, WIDTH/2, HEIGHT/2);
439                 check_error();
440
441                 glUseProgram(cbcr_program_num);
442                 check_error();
443
444                 glActiveTexture(GL_TEXTURE0);
445                 check_error();
446                 glBindTexture(GL_TEXTURE_2D, chroma_tex);
447                 check_error();
448                 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
449                 check_error();
450                 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
451                 check_error();
452                 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
453                 check_error();
454
455                 float chroma_offset_0[] = { -0.5f / WIDTH, 0.0f };
456                 set_uniform_vec2(cbcr_program_num, "foo", "chroma_offset_0", chroma_offset_0);
457
458                 GLuint position_vbo = fill_vertex_attribute(cbcr_program_num, "position", 2, GL_FLOAT, sizeof(vertices), vertices);
459                 GLuint texcoord_vbo = fill_vertex_attribute(cbcr_program_num, "texcoord", 2, GL_FLOAT, sizeof(vertices), vertices);  // Same as vertices.
460
461                 glDrawArrays(GL_TRIANGLES, 0, 3);
462                 check_error();
463
464                 cleanup_vertex_attribute(cbcr_program_num, "position", position_vbo);
465                 cleanup_vertex_attribute(cbcr_program_num, "texcoord", texcoord_vbo);
466
467                 glUseProgram(0);
468                 check_error();
469
470                 RefCountedGLsync fence(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);
471                 check_error();
472
473                 resource_pool->release_fbo(cbcr_fbo);
474
475                 h264_encoder.end_frame(fence, input_frames_to_release);
476
477
478 #if 1
479 #if _POSIX_C_SOURCE >= 199309L
480                 clock_gettime(CLOCK_MONOTONIC, &now);
481                 double elapsed = now.tv_sec - start.tv_sec +
482                         1e-9 * (now.tv_nsec - start.tv_nsec);
483 #else
484                 gettimeofday(&now, NULL);
485                 double elapsed = now.tv_sec - start.tv_sec +
486                         1e-6 * (now.tv_usec - start.tv_usec);
487 #endif
488                 if (frame % 100 == 0) {
489                         printf("%d frames in %.3f seconds = %.1f fps (%.1f ms/frame)\n",
490                                 frame, elapsed, frame / elapsed,
491                                 1e3 * elapsed / frame);
492                 //      chain.print_phase_timing();
493                 }
494
495                 // Reset every 100 frames, so that local variations in frame times
496                 // (especially for the first few frames, when the shaders are
497                 // compiled etc.) don't make it hard to measure for the entire
498                 // remaining duration of the program.
499                 if (frame == 10000) {
500                         frame = 0;
501                         start = now;
502                 }
503 #endif
504         }
505         glDeleteVertexArrays(1, &vao);
506         resource_pool->release_glsl_program(cbcr_program_num);
507         resource_pool->release_2d_texture(chroma_tex);
508         BMUSBCapture::stop_bm_thread();
509 }
510
511 std::thread mixer_thread;
512
513 void start_mixer(QSurface *surface, QSurface *surface2, QSurface *surface3, QSurface *surface4)
514 {
515         mixer_thread = std::thread([surface, surface2, surface3, surface4]{
516                 mixer_thread_func(surface, surface2, surface3, surface4);
517         });
518 }
519
520 void mixer_quit()
521 {
522         quit = true;
523         mixer_thread.join();
524 }