cubecl-runtime 0.9.0

Crate that helps creating high performance async runtimes for CubeCL.
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

use std::{thread::sleep, time::Duration};

use cubecl_runtime::{id::KernelId, storage::BytesResource};

use crate::dummy::DummyKernel;

const SLEEP_MS: u64 = 1;

#[derive(Debug)]
pub struct DummyElementwiseAdditionSlowWrong;
#[derive(Debug)]
pub struct DummyElementwiseMultiplication;
#[derive(Debug)]
pub struct DummyElementwiseMultiplicationSlowWrong;

impl DummyKernel for DummyElementwiseAdditionSlowWrong {
    fn compute(&self, inputs: &mut [&mut BytesResource]) {
        // Slow and wrong on purpose, for tests
        let lhs = &inputs[0].read();
        let out = &mut inputs[2].write();

        let size = lhs.len();

        for i in 0..size {
            sleep(Duration::from_millis(SLEEP_MS));
            out[i] = lhs[i]
        }
    }

    fn id(&self) -> KernelId {
        KernelId::new::<Self>()
    }
}

impl DummyKernel for DummyElementwiseMultiplication {
    fn compute(&self, inputs: &mut [&mut BytesResource]) {
        let lhs = &inputs[0].read();
        let rhs = &inputs[1].read();
        let out = &mut inputs[2].write();

        let size = lhs.len();

        for i in 0..size {
            out[i] = lhs[i] * rhs[i];
        }
    }

    fn id(&self) -> KernelId {
        KernelId::new::<Self>()
    }
}

impl DummyKernel for DummyElementwiseMultiplicationSlowWrong {
    fn compute(&self, inputs: &mut [&mut BytesResource]) {
        // Slow and wrong on purpose, for tests
        let lhs = &inputs[0].read();
        let out = &mut inputs[2].write();

        let size = lhs.len();

        for i in 0..size {
            sleep(Duration::from_millis(SLEEP_MS));
            out[i] = lhs[i];
        }
    }

    fn id(&self) -> KernelId {
        KernelId::new::<Self>()
    }
}