cubecl-random 0.8.1

CubeCL Random Number Generation.
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
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111

use cubecl::prelude::*;
use cubecl_core as cubecl;
use cubecl_std::tensor::View;
use std::f32::consts::PI;

use super::{PrngArgs, PrngRuntime, random};

use crate::{RandomFamily, lcg_step, taus_step_0, taus_step_1, taus_step_2, to_unit_interval_open};

#[derive(CubeLaunch, CubeType)]
pub(crate) struct Normal<E: Numeric> {
    mean: E,
    std: E,
}

#[derive(Debug)]
struct NormalFamily;

impl RandomFamily for NormalFamily {
    type Runtime<E: Numeric> = Normal<E>;
}

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

        let mut output_line_0 = Line::empty(line_size);
        let mut output_line_1 = Line::empty(line_size);

        let num_iterations = n_values_per_thread / line_size / 2;
        #[unroll(num_iterations <= 8)]
        for line_index in 0..num_iterations {
            // vectorization
            #[unroll]
            for i in 0..line_size {
                // 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 = Log::log(unit_0) * -2.0;
                let coeff = Sqrt::sqrt(coeff) * 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_line_0[i] = E::cast_from(normal_0);
                output_line_1[i] = E::cast_from(normal_1);
            }

            let iteration_offset = line_index * n_invocations * 2;
            let write_index_0 = write_index_base + iteration_offset;
            let write_index_1 = write_index_0 + n_invocations;

            output[write_index_0] = output_line_0;
            output[write_index_1] = output_line_1;
        }
    }
}

impl<E: Numeric> PrngArgs<E> for Normal<E> {
    type Args = Self;

    fn args<'a, R: Runtime>(self) -> NormalLaunch<'a, E, R> {
        NormalLaunch::new(ScalarArg::new(self.mean), ScalarArg::new(self.std))
    }
}

/// Pseudo-random generator with uniform distribution
pub fn random_normal<R: Runtime, E: Numeric>(
    client: &ComputeClient<R::Server>,
    mean: E,
    std: E,
    out: TensorHandleRef<R>,
) {
    assert_eq!(
        out.elem_size as u32,
        E::elem_size(),
        "Tensor element type must be the same as type E"
    );

    random::<NormalFamily, E, R>(client, Normal { mean, std }, out)
}