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

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

//! Optimized FastTanh (Padé rational approximation) activation kernels.

use crate::activation_simd_avx2;
use crate::activation_simd_avx512;
use core::arch::x86_64::*;

/// Applies FastTanh (Padé rational approximation) to a slice using AVX2+FMA.
///
/// # Safety
/// Requires AVX2 and FMA support.
#[target_feature(enable = "avx2,fma")]
#[expect(
    clippy::excessive_precision,
    reason = "High-precision constants required for bit-exact numerical validation against reference"
)]
pub unsafe fn fast_tanh_slice_avx2(data: &mut [f32]) {
    let ca = _mm256_set1_ps(2.45550750702956_f32);
    let cb = _mm256_set1_ps(0.893229853513558_f32);
    let cc = _mm256_set1_ps(0.821226666969744_f32);
    let cd = _mm256_set1_ps(2.44506634652299_f32);
    let ce = _mm256_set1_ps(0.814642734961073_f32);
    let sign_mask = _mm256_set1_ps(-0.0_f32);
    let mut i = 0;
    let len = data.len();
    unsafe {
        activation_simd_avx2!(
            i,
            len,
            {
                let x1 = _mm256_loadu_ps(data.as_ptr().add(i));
                let x2 = _mm256_loadu_ps(data.as_ptr().add(i + 8));
                let ax1 = _mm256_andnot_ps(sign_mask, x1);
                let ax2 = _mm256_andnot_ps(sign_mask, x2);
                let x21 = _mm256_mul_ps(x1, x1);
                let x22 = _mm256_mul_ps(x2, x2);
                let num_inner1 = _mm256_fmadd_ps(cc, ax1, cb);
                let num_inner2 = _mm256_fmadd_ps(cc, ax2, cb);
                let num_poly1 = _mm256_fmadd_ps(num_inner1, x21, _mm256_fmadd_ps(ca, ax1, ca));
                let num_poly2 = _mm256_fmadd_ps(num_inner2, x22, _mm256_fmadd_ps(ca, ax2, ca));
                let num1 = _mm256_mul_ps(x1, num_poly1);
                let num2 = _mm256_mul_ps(x2, num_poly2);
                let xe1 = _mm256_mul_ps(ce, _mm256_mul_ps(x1, ax1));
                let xe2 = _mm256_mul_ps(ce, _mm256_mul_ps(x2, ax2));
                let xterm1 = _mm256_add_ps(x1, xe1);
                let xterm2 = _mm256_add_ps(x2, xe2);
                let abs_xterm1 = _mm256_andnot_ps(sign_mask, xterm1);
                let abs_xterm2 = _mm256_andnot_ps(sign_mask, xterm2);
                let den_inner1 = _mm256_add_ps(cd, x21);
                let den_inner2 = _mm256_add_ps(cd, x22);
                let den1 = _mm256_fmadd_ps(den_inner1, abs_xterm1, cd);
                let den2 = _mm256_fmadd_ps(den_inner2, abs_xterm2, cd);
                let y1 = _mm256_div_ps(num1, den1);
                let y2 = _mm256_div_ps(num2, den2);
                _mm256_storeu_ps(data.as_mut_ptr().add(i), y1);
                _mm256_storeu_ps(data.as_mut_ptr().add(i + 8), y2);
            },
            {
                let x = _mm256_loadu_ps(data.as_ptr().add(i));
                let ax = _mm256_andnot_ps(sign_mask, x);
                let x2 = _mm256_mul_ps(x, x);
                let num_inner = _mm256_fmadd_ps(cc, ax, cb);
                let num_poly = _mm256_fmadd_ps(num_inner, x2, _mm256_fmadd_ps(ca, ax, ca));
                let num = _mm256_mul_ps(x, num_poly);
                let xe = _mm256_mul_ps(ce, _mm256_mul_ps(x, ax));
                let xterm = _mm256_add_ps(x, xe);
                let abs_xterm = _mm256_andnot_ps(sign_mask, xterm);
                let den_inner = _mm256_add_ps(cd, x2);
                let den = _mm256_fmadd_ps(den_inner, abs_xterm, cd);
                let y = _mm256_div_ps(num, den);
                _mm256_storeu_ps(data.as_mut_ptr().add(i), y);
            }
        );
    }
    for x in data.iter_mut().skip(i) {
        *x = fast_tanh(*x);
    }
}

/// Applies FastTanh (Padé rational approximation) to a slice using AVX-512+FMA.
///
/// # Safety
/// Requires AVX-512F, AVX-512VL, AVX-512DQ, and FMA support.
#[target_feature(enable = "avx512f,avx512vl,avx512dq,fma")]
#[expect(
    clippy::excessive_precision,
    reason = "High-precision constants required for bit-exact numerical validation against reference"
)]
pub unsafe fn fast_tanh_slice_avx512(data: &mut [f32]) {
    let ca = _mm512_set1_ps(2.45550750702956_f32);
    let cb = _mm512_set1_ps(0.893229853513558_f32);
    let cc = _mm512_set1_ps(0.821226666969744_f32);
    let cd = _mm512_set1_ps(2.44506634652299_f32);
    let ce = _mm512_set1_ps(0.814642734961073_f32);
    let sign_mask = _mm512_set1_ps(-0.0_f32);
    let mut i = 0;
    let len = data.len();
    unsafe {
        activation_simd_avx512!(i, len, {
            let x = _mm512_loadu_ps(data.as_ptr().add(i));
            let ax = _mm512_andnot_ps(sign_mask, x);
            let x2 = _mm512_mul_ps(x, x);
            let num_inner = _mm512_fmadd_ps(cc, ax, cb);
            let num_poly = _mm512_fmadd_ps(num_inner, x2, _mm512_fmadd_ps(ca, ax, ca));
            let num = _mm512_mul_ps(x, num_poly);
            let xe = _mm512_mul_ps(ce, _mm512_mul_ps(x, ax));
            let xterm = _mm512_add_ps(x, xe);
            let abs_xterm = _mm512_andnot_ps(sign_mask, xterm);
            let den_inner = _mm512_add_ps(cd, x2);
            let den = _mm512_fmadd_ps(den_inner, abs_xterm, cd);
            let y = _mm512_div_ps(num, den);
            _mm512_storeu_ps(data.as_mut_ptr().add(i), y);
        });
    }
    for x in data.iter_mut().skip(i) {
        *x = fast_tanh(*x);
    }
}

/// Fast rational Padé approximation for the tanh function.
#[inline(always)]
#[expect(
    clippy::excessive_precision,
    reason = "High-precision constants required for bit-exact numerical validation against reference"
)]
pub fn fast_tanh(x: f32) -> f32 {
    let ax = x.abs();
    let x2 = x * x;

    (x * (2.45550750702956
        + 2.45550750702956 * ax
        + (0.893229853513558 + 0.821226666969744 * ax) * x2))
        / (2.44506634652299 + (2.44506634652299 + x2) * (x + 0.814642734961073 * x * ax).abs())
}