ruRAND 0.1.69

Random number generation for Ruda.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66

use super::*;

#[ruda]
impl PrngRuntime for Normal {
    fn inner_loop<E: Numeric, N: Size>(
        args: Normal,
        write_index_base: usize,
        n_invocations: u32,
        #[comptime] n_values_per_thread: usize,
        state_0: &mut u32,
        state_1: &mut u32,
        state_2: &mut u32,
        state_3: &mut u32,
        output: &mut View<Vector<E, N>, usize, ReadWrite>,
    ) {
        let mean = f32::cast_from(args.mean);
        let std = f32::cast_from(args.std);

        let mut output_vector_0 = Vector::empty();
        let mut output_vector_1 = Vector::empty();

        let num_iterations = n_values_per_thread / N::value() / 2;
        #[unroll(num_iterations <= 8)]
        for vector_index in 0..num_iterations {
            // vectorization
            #[unroll]
            for i in 0..N::value() {
                // First random uniform integer
                *state_0 = taus_step_0(*state_0);
                *state_1 = taus_step_1(*state_1);
                *state_2 = taus_step_2(*state_2);
                *state_3 = lcg_step(*state_3);

                let int_random = *state_0 ^ *state_1 ^ *state_2 ^ *state_3;
                let unit_0 = to_unit_interval_open(int_random);

                // Second random uniform integer
                *state_0 = taus_step_0(*state_0);
                *state_1 = taus_step_1(*state_1);
                *state_2 = taus_step_2(*state_2);
                *state_3 = lcg_step(*state_3);

                let int_random = *state_0 ^ *state_1 ^ *state_2 ^ *state_3;
                let unit_1 = to_unit_interval_open(int_random);

                // Box-Muller transform
                let coeff = unit_0.ln() * -2.0;
                let coeff = coeff.sqrt() * std;
                let trigo_arg = 2.0 * PI * unit_1;

                let normal_0 = f32::cos(trigo_arg) * coeff + mean;
                let normal_1 = f32::sin(trigo_arg) * coeff + mean;

                output_vector_0[i] = E::cast_from(normal_0);
                output_vector_1[i] = E::cast_from(normal_1);
            }

            let iteration_offset = vector_index * n_invocations as usize * 2;
            let write_index_0 = write_index_base + iteration_offset;
            let write_index_1 = write_index_0 + n_invocations as usize;

            output.write_checked(write_index_0, output_vector_0);
            output.write_checked(write_index_1, output_vector_1);
        }
    }
}