Tizen Native API
8.0
|
efl_thread_6.c
//Compile with: //gcc -o efl_thread_6 efl_thread_6.c -g `pkg-config --cflags --libs elementary` #include <Elementary.h> static Evas_Object *win = NULL; static Eina_List *threads; struct info { Evas_Object *obj; int *pix; }; // BEGIN - code running in my custom thread instance // static void mandel(Ecore_Thread *th, int *pix, int w, int h) { double x, xx, y, cx, cy, cox, coy; int iteration, hx, hy, val, r, g, b, rr, gg, bb; int itermax = 10000; double magnify = 0.02; // this mandel calc is run in the worker threads so it's here. it is // just here to calculate something and consume cpu to demonstrate the // ecore thread worker queue. don't pay much attention to the below code magnify += ((double)(rand() % 100) / 100.0) / 4.0; cox = (double)(rand() % 100) / 100.0; coy = (double)(rand() % 100) / 100.0; cox /= (magnify * 3.0); r = rand() % 255; g = rand() % 255; b = rand() % 255; for (hy = 0; hy < h; hy++) { // every line check if thread has been cancelled to return early if (ecore_thread_check(th)) return; for (hx = 0; hx < w; hx++) { cx = (((float)hx) / ((float)w) - 0.5) / (magnify * 3.0); cy = (((float)hy) / ((float)h) - 0.5) / (magnify * 3.0); cx += cox; cy += coy; x = 0.0; y = 0.0; for (iteration = 1; iteration < itermax; iteration++) { xx = (x * x) - (y * y) + cx; y = (2.0 * x * y) + cy; x = xx; if (((x * x) + (y * y)) > 100.0) iteration = 999999; } val = (((x * x) + (y * y)) * 2.55) / 100.0; if (val > 255) val = 255; if (iteration >= 99999) { rr = (r * val) / 255; gg = (g * val) / 255; bb = (b * val) / 255; pix[(hy * w) + hx] = (val << 24) | (rr << 16) | (gg << 8) | (bb); } else pix[(hy * w) + hx] = 0xffffffff; } } } static void th_do(void *data, Ecore_Thread *th) { struct info *inf = data; // CANNOT TOUCH inf->obj here! just inf->pix which is 256x256 @ 32bpp // quick and dirty to consume some cpu - do a mandelbrot calc mandel(th, inf->pix, 256, 256); } // // END - code running in my custom thread instance static void // thread job finished - collect results and put in img obj th_end(void *data, Ecore_Thread *th) { struct info *inf = data; // copy data to object, free calculated data and info struc evas_object_image_data_copy_set(inf->obj, inf->pix); evas_object_show(inf->obj); free(inf->pix); free(inf); threads = eina_list_remove(threads, th); } static void // if the thread is cancelled - free pix, keep obj tho th_cancel(void *data, Ecore_Thread *th EINA_UNUSED) { struct info *inf = data; // just free pixel data and info struct free(inf->pix); free(inf); } static Eina_Bool // animate the objects so you see all the madels move anim(void *data) { Evas_Object *o = data; double t, z; int w, h, v; Evas_Coord x, y; // just calculate some position using the pointer value of the object as // a seed value to make different objects go into different places over time v = ((int)(uintptr_t)o) & 0xff; t = ecore_loop_time_get(); w = 100 + ((v * 100) >> 8); h = 100 + ((v * 100) >> 8); z = (double)(v) / 100.0; x = (w * sin(t)); y = (h * cos(t + z)); // do the actual move evas_object_move(o, 200 + x - 128, 200 + y - 128); // keep looping - return true return EINA_TRUE; } EAPI_MAIN int elm_main(int argc EINA_UNUSED, char **argv EINA_UNUSED) { Evas_Object *o; Ecore_Thread *th; int i; elm_policy_set(ELM_POLICY_QUIT, ELM_POLICY_QUIT_LAST_WINDOW_CLOSED); win = elm_win_util_standard_add("efl-thread-6", "EFL Thread 6"); elm_win_autodel_set(win, EINA_TRUE); // queue up 64 mandel generation thread jobs for (i = 0; i < 64; i++) { struct info *inf; // create ecore thread to do some threaded job inside the worker pool inf = malloc(sizeof(struct info)); if (inf) { o = evas_object_image_filled_add(evas_object_evas_get(win)); evas_object_image_size_set(o, 256, 256); evas_object_image_alpha_set(o, EINA_TRUE); evas_object_resize(o, 256, 256); inf->obj = o; inf->pix = malloc(256 * 256 * sizeof(int)); th = ecore_thread_run(th_do, th_end, th_cancel, inf); threads = eina_list_append(threads, th); // bonus - slide the objects around all the time with an // animator that ticks off every frame. ecore_animator_add(anim, o); } } evas_object_resize(win, 400, 400); evas_object_show(win); elm_run(); // if some threads are still running - cancel them EINA_LIST_FREE(threads, th) ecore_thread_cancel(th); return 0; } ELM_MAIN()