This is launch statement.

We use nested for loop. i is stage and it increases until log2(n). For example, let say n is 16, so i is 1 to 4. Another for loop represents round and round cannot be bigger than stage.

This is bitonic kernel. There are several variables.

'''idx''' shows which elements in array will be picked by threads. For example, thread 0 in stage 1 will have (0 / 1) * 1 * 2 + 0 % 1 = 0, and thread 1 will have (1 / 1) * 1 * 2 + 1 % 1 = 2, and thread 2 will have (2 / 1) * 1 * 2 + 2 % 1 = 4 ... 6, 8, 10. because in stage 1, index 0 will be compared with 1, 2 compared with 3, 4 compared with 5 .... another example, if thread 0 in round 1 in stage 2 will have ( 0 / 2)*2*2 + 0%2 = 0, thread 1 will have ( 1/2)*2*2 + 1%2 = 1, thread 2 will have ( 2/2)*2*2 + 2%2 = 4, thread 3 will have (3/2) * 2 * 2 + 3%2 = 5. So threads will have 0, 1, 4, 5, 8, 9 .... because in round 1 in stage 2, index 0 will be compared with 2 and 1 compared with 3, 4 compared with 6 and 5 compared with 7.