NeuralAmpModeler-rs 3.0.0

An opinionated, high-performance Neural Amp Modeler (NAM) client and core implementation in Rust for Linux/PipeWire and CLAP plugins.
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
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139

// SPDX-License-Identifier: Apache-2.0
// Copyright (c) 2026 Fábio Henrique de Lima Silva (fhl.bsb@gmail.com) All rights reserved.

use super::*;

#[test]
fn test_gating_mode_default() {
    assert_eq!(GatingMode::default(), GatingMode::None);
}

#[test]
fn test_gating_apply_tanh_sigmoid() {
    let config = GatingActivationConfig::new(ActivationType::Tanh, ActivationType::Sigmoid);
    let mut buf = vec![0.5f32, -0.3, 0.8, -0.6];
    let mut golden = buf.clone();

    // Golden: apply activations to each half independently, then multiply.
    ActivationType::Tanh.apply(&mut golden[..2]);
    ActivationType::Sigmoid.apply(&mut golden[2..]);
    golden[0] *= golden[2];
    golden[1] *= golden[3];

    config.apply_gating(&mut buf);

    // Compare only the first half (output channels).
    assert!((buf[0] - golden[0]).abs() < 1e-7);
    assert!((buf[1] - golden[1]).abs() < 1e-7);
}

#[test]
fn test_blending_apply_tanh_sigmoid() {
    let mut config =
        BlendingActivationConfig::new(ActivationType::Tanh, ActivationType::Sigmoid, 2).unwrap();
    let mut buf = [0.5f32, -0.3f32, 0.8f32, -0.6f32];
    let orig = buf;
    let mut golden = [0.0f32; 2];

    // Golden: apply activations, then blend.
    let mut activated = [orig[0], orig[1]];
    ActivationType::Tanh.apply(&mut activated);
    let mut alpha = [orig[2], orig[3]];
    ActivationType::Sigmoid.apply(&mut alpha);
    for i in 0..2 {
        golden[i] = activated[i].mul_add(alpha[i], orig[i] * (1.0 - alpha[i]));
    }

    config.apply_blending(&mut buf);

    // Compare only the first half (output channels).
    assert!((buf[0] - golden[0]).abs() < 1e-7);
    assert!((buf[1] - golden[1]).abs() < 1e-7);
}

#[test]
fn test_gating_channel_pairs_zero_input() {
    let config = GatingActivationConfig::new(ActivationType::Tanh, ActivationType::Sigmoid);
    let mut buf = vec![0.0f32; 8]; // ch=4
    config.apply_gating(&mut buf);
    // tanh(0)=0, sigmoid(0)=0.5, product=0
    for v in buf.iter().take(4) {
        assert!((*v).abs() < 1e-7);
    }
}

#[test]
fn test_blending_channel_pairs_zero_input() {
    let mut config =
        BlendingActivationConfig::new(ActivationType::Tanh, ActivationType::Sigmoid, 4).unwrap();
    let mut buf = vec![0.0f32; 8]; // ch=4
    config.apply_blending(&mut buf);
    // tanh(0)=0, sigmoid(0)=0.5, blended = 0 * 0.5 + 0 * 0.5 = 0
    for v in buf.iter().take(4) {
        assert!((*v).abs() < 1e-7);
    }
}

#[test]
fn test_gating_unity_gate() {
    let config = GatingActivationConfig::new(ActivationType::Tanh, ActivationType::Sigmoid);
    let mut buf = vec![0.5f32, -0.3f32, 10.0f32, 10.0f32]; // ch=2
    let mut expected = [0.5f32, -0.3f32];
    ActivationType::Tanh.apply(&mut expected);
    // sigmoid(10) ≈ 1, so gate multiplier ≈ 1
    config.apply_gating(&mut buf);
    assert!((buf[0] - expected[0]).abs() < 1e-4);
    assert!((buf[1] - expected[1]).abs() < 1e-4);
}

#[test]
fn test_blending_alpha_one() {
    let mut config =
        BlendingActivationConfig::new(ActivationType::Tanh, ActivationType::Sigmoid, 2).unwrap();
    let mut buf = vec![0.5f32, -0.3f32, 10.0f32, 10.0f32]; // ch=2
    let mut expected = [0.5f32, -0.3f32];
    ActivationType::Tanh.apply(&mut expected);
    // sigmoid(10) ≈ 1 → blended ≈ activated_input
    config.apply_blending(&mut buf);
    assert!((buf[0] - expected[0]).abs() < 1e-4);
    assert!((buf[1] - expected[1]).abs() < 1e-4);
}

#[test]
fn test_blending_alpha_zero() {
    let mut config =
        BlendingActivationConfig::new(ActivationType::Tanh, ActivationType::Sigmoid, 2).unwrap();
    let orig = [0.5f32, -0.3f32];
    let mut buf = [orig[0], orig[1], -10.0f32, -10.0f32]; // ch=2, alpha≈0
    let mut golden = orig;

    // Golden with actual sigmoid: alpha ≈ 0 → blended ≈ original.
    // For x ≤ -8, sigmoid(x) = 0 in our approximation.
    let mut alpha = [-10.0f32, -10.0f32];
    ActivationType::Sigmoid.apply(&mut alpha);
    assert!(alpha[0].abs() < 1e-4);

    let mut activated = orig;
    ActivationType::Tanh.apply(&mut activated);
    for i in 0..2 {
        golden[i] = activated[i].mul_add(alpha[i], orig[i] * (1.0 - alpha[i]));
    }

    config.apply_blending(&mut buf);
    assert!((buf[0] - golden[0]).abs() < 1e-4);
    assert!((buf[1] - golden[1]).abs() < 1e-4);
}

#[test]
fn test_blending_scratch_preserved() {
    let mut config =
        BlendingActivationConfig::new(ActivationType::Tanh, ActivationType::Sigmoid, 2).unwrap();
    assert_eq!(config.channels(), 2);

    // Apply twice with same config — scratch must not leak state.
    let mut buf1 = [0.5f32, 0.3f32, 0.8f32, 0.6f32];
    let mut buf2 = [0.5f32, 0.3f32, 0.8f32, 0.6f32];
    config.apply_blending(&mut buf1);
    config.apply_blending(&mut buf2);
    assert_eq!(&buf1[..2], &buf2[..2]);
}