numr 0.5.2

High-performance numerical computing with multi-backend GPU acceleration (CPU/CUDA/WebGPU)
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

// F32 scalar operations

const WORKGROUP_SIZE: u32 = 256u;

struct ScalarParams {
    numel: u32,
    scalar: f32,
}

@group(0) @binding(0) var<storage, read_write> scalar_a: array<f32>;
@group(0) @binding(1) var<storage, read_write> scalar_out: array<f32>;
@group(0) @binding(2) var<uniform> scalar_params: ScalarParams;

@compute @workgroup_size(256)
fn add_scalar_f32(@builtin(global_invocation_id) gid: vec3<u32>) {
    let idx = gid.x;
    if (idx < scalar_params.numel) {
        scalar_out[idx] = scalar_a[idx] + scalar_params.scalar;
    }
}

@compute @workgroup_size(256)
fn sub_scalar_f32(@builtin(global_invocation_id) gid: vec3<u32>) {
    let idx = gid.x;
    if (idx < scalar_params.numel) {
        scalar_out[idx] = scalar_a[idx] - scalar_params.scalar;
    }
}

@compute @workgroup_size(256)
fn rsub_scalar_f32(@builtin(global_invocation_id) gid: vec3<u32>) {
    let idx = gid.x;
    if (idx < scalar_params.numel) {
        scalar_out[idx] = scalar_params.scalar - scalar_a[idx];
    }
}

@compute @workgroup_size(256)
fn mul_scalar_f32(@builtin(global_invocation_id) gid: vec3<u32>) {
    let idx = gid.x;
    if (idx < scalar_params.numel) {
        scalar_out[idx] = scalar_a[idx] * scalar_params.scalar;
    }
}

@compute @workgroup_size(256)
fn div_scalar_f32(@builtin(global_invocation_id) gid: vec3<u32>) {
    let idx = gid.x;
    if (idx < scalar_params.numel) {
        scalar_out[idx] = scalar_a[idx] / scalar_params.scalar;
    }
}

@compute @workgroup_size(256)
fn pow_scalar_f32(@builtin(global_invocation_id) gid: vec3<u32>) {
    let idx = gid.x;
    if (idx < scalar_params.numel) {
        scalar_out[idx] = pow(scalar_a[idx], scalar_params.scalar);
    }
}

@compute @workgroup_size(256)
fn leaky_relu_f32(@builtin(global_invocation_id) gid: vec3<u32>) {
    let idx = gid.x;
    if (idx < scalar_params.numel) {
        let x = scalar_a[idx];
        let slope = scalar_params.scalar;
        scalar_out[idx] = max(slope * x, x);
    }
}

@compute @workgroup_size(256)
fn elu_f32(@builtin(global_invocation_id) gid: vec3<u32>) {
    let idx = gid.x;
    if (idx < scalar_params.numel) {
        let x = scalar_a[idx];
        let alpha = scalar_params.scalar;
        scalar_out[idx] = select(alpha * (exp(x) - 1.0), x, x > 0.0);
    }
}