nnl 0.1.6

A high-performance neural network library for Rust with CPU and GPU support
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

#version 450

// Tanh activation compute shader
// Computes: result[i] = tanh(input[i])

layout(local_size_x = 64, local_size_y = 1, local_size_z = 1) in;

layout(set = 0, binding = 0) buffer InputBuffer {
    float input_data[];
};

layout(set = 0, binding = 1) buffer OutputBuffer {
    float result[];
};

void main() {
    uint index = gl_GlobalInvocationID.x;

    // Bounds checking
    if (index >= input_data.length() || index >= result.length()) {
        return;
    }

    // Perform tanh activation entirely on GPU
    // tanh(x) = (e^x - e^(-x)) / (e^x + e^(-x))
    // Use numerically stable implementation to avoid overflow
    float x = input_data[index];

    if (abs(x) > 10.0) {
        // For large values, tanh approaches ±1
        result[index] = sign(x);
    } else {
        // Standard tanh computation for moderate values
        float exp_2x = exp(2.0 * x);
        result[index] = (exp_2x - 1.0) / (exp_2x + 1.0);
    }
}