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
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266

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

/// F32-native 8-lane interleaved dot product (scalar reference for SIMD validation).
///
/// Uses `mul_add` (FMA3 fused multiply-add) to match the instruction set of the
/// AVX2/FMA kernel (`dot_product_8x_f32_avx2`). Both paths produce **mathematically
/// equivalent** results with FMA‑level precision.
///
/// # Safety
/// Caller must ensure `weights.len() >= state.len()`.
#[inline]
pub unsafe fn dot_product_8x_f32_scalar(weights: &[[f32; 8]], state: &[f32]) -> [f32; 8] {
    let len = core::cmp::min(weights.len(), state.len());
    let mut r = [0.0f32; 8];
    for i in 0..len {
        let w = weights.get_unchecked(i);
        let s = state.get_unchecked(i);
        r[0] = (*w.get_unchecked(0)).mul_add(*s, r[0]);
        r[1] = (*w.get_unchecked(1)).mul_add(*s, r[1]);
        r[2] = (*w.get_unchecked(2)).mul_add(*s, r[2]);
        r[3] = (*w.get_unchecked(3)).mul_add(*s, r[3]);
        r[4] = (*w.get_unchecked(4)).mul_add(*s, r[4]);
        r[5] = (*w.get_unchecked(5)).mul_add(*s, r[5]);
        r[6] = (*w.get_unchecked(6)).mul_add(*s, r[6]);
        r[7] = (*w.get_unchecked(7)).mul_add(*s, r[7]);
    }
    r
}

/// F32-native 16-lane interleaved dot product (scalar oracle for SIMD validation).
///
/// Uses `mul_add` (FMA3 fused multiply-add) to match the instruction set of the
/// AVX‑512 kernel (`dot_product_16x_f32_avx512`) and the AVX2/FMA kernel
/// (`dot_product_16x_f32_avx2`). This function serves exclusively as the
/// correctness reference in test and benchmark suites. It MUST NOT be called
/// from any production code path (trait implementations, model inference loops).
///
/// # Safety
/// Caller must ensure `weights.len() >= state.len()`.
#[inline]
pub unsafe fn dot_product_16x_f32_scalar(weights: &[[f32; 16]], state: &[f32]) -> [f32; 16] {
    let len = core::cmp::min(weights.len(), state.len());
    let mut r = [0.0f32; 16];
    for i in 0..len {
        let w = weights.get_unchecked(i);
        let s = state.get_unchecked(i);
        r[0] = (*w.get_unchecked(0)).mul_add(*s, r[0]);
        r[1] = (*w.get_unchecked(1)).mul_add(*s, r[1]);
        r[2] = (*w.get_unchecked(2)).mul_add(*s, r[2]);
        r[3] = (*w.get_unchecked(3)).mul_add(*s, r[3]);
        r[4] = (*w.get_unchecked(4)).mul_add(*s, r[4]);
        r[5] = (*w.get_unchecked(5)).mul_add(*s, r[5]);
        r[6] = (*w.get_unchecked(6)).mul_add(*s, r[6]);
        r[7] = (*w.get_unchecked(7)).mul_add(*s, r[7]);
        r[8] = (*w.get_unchecked(8)).mul_add(*s, r[8]);
        r[9] = (*w.get_unchecked(9)).mul_add(*s, r[9]);
        r[10] = (*w.get_unchecked(10)).mul_add(*s, r[10]);
        r[11] = (*w.get_unchecked(11)).mul_add(*s, r[11]);
        r[12] = (*w.get_unchecked(12)).mul_add(*s, r[12]);
        r[13] = (*w.get_unchecked(13)).mul_add(*s, r[13]);
        r[14] = (*w.get_unchecked(14)).mul_add(*s, r[14]);
        r[15] = (*w.get_unchecked(15)).mul_add(*s, r[15]);
    }
    r
}

/// F32-native 8-lane interleaved dual-frame dot product (scalar reference).
///
/// Processes two state vectors against the same weight slice using `mul_add`
/// (FMA3) to match the rounding of the AVX2/FMA kernel.
///
/// # Safety
/// Caller must ensure `weights.len() >= state_f0.len()` and
/// `weights.len() >= state_f1.len()`.
#[inline]
pub unsafe fn dot_product_8x_f32_dual_scalar(
    weights: &[[f32; 8]],
    state_f0: &[f32],
    state_f1: &[f32],
) -> ([f32; 8], [f32; 8]) {
    let len = core::cmp::min(
        weights.len(),
        core::cmp::min(state_f0.len(), state_f1.len()),
    );
    let mut r0 = [0.0f32; 8];
    let mut r1 = [0.0f32; 8];
    for i in 0..len {
        let w = weights.get_unchecked(i);
        let s0 = state_f0.get_unchecked(i);
        let s1 = state_f1.get_unchecked(i);
        r0[0] = (*w.get_unchecked(0)).mul_add(*s0, r0[0]);
        r0[1] = (*w.get_unchecked(1)).mul_add(*s0, r0[1]);
        r0[2] = (*w.get_unchecked(2)).mul_add(*s0, r0[2]);
        r0[3] = (*w.get_unchecked(3)).mul_add(*s0, r0[3]);
        r0[4] = (*w.get_unchecked(4)).mul_add(*s0, r0[4]);
        r0[5] = (*w.get_unchecked(5)).mul_add(*s0, r0[5]);
        r0[6] = (*w.get_unchecked(6)).mul_add(*s0, r0[6]);
        r0[7] = (*w.get_unchecked(7)).mul_add(*s0, r0[7]);
        r1[0] = (*w.get_unchecked(0)).mul_add(*s1, r1[0]);
        r1[1] = (*w.get_unchecked(1)).mul_add(*s1, r1[1]);
        r1[2] = (*w.get_unchecked(2)).mul_add(*s1, r1[2]);
        r1[3] = (*w.get_unchecked(3)).mul_add(*s1, r1[3]);
        r1[4] = (*w.get_unchecked(4)).mul_add(*s1, r1[4]);
        r1[5] = (*w.get_unchecked(5)).mul_add(*s1, r1[5]);
        r1[6] = (*w.get_unchecked(6)).mul_add(*s1, r1[6]);
        r1[7] = (*w.get_unchecked(7)).mul_add(*s1, r1[7]);
    }
    (r0, r1)
}

/// F32-native 16-lane interleaved dual-frame dot product (scalar reference).
///
/// Processes two state vectors against the same weight slice using `mul_add`
/// (FMA3) to match the rounding of the AVX2/FMA and AVX-512 kernels.
/// This function serves exclusively as the correctness reference in test and
/// benchmark suites. It MUST NOT be called from any production code path.
///
/// # Safety
/// Caller must ensure `weights.len() >= state_f0.len()` and
/// `weights.len() >= state_f1.len()`.
#[inline]
pub unsafe fn dot_product_16x_f32_dual_scalar(
    weights: &[[f32; 16]],
    state_f0: &[f32],
    state_f1: &[f32],
) -> ([f32; 16], [f32; 16]) {
    let len = core::cmp::min(
        weights.len(),
        core::cmp::min(state_f0.len(), state_f1.len()),
    );
    let mut r0 = [0.0f32; 16];
    let mut r1 = [0.0f32; 16];
    for i in 0..len {
        let w = weights.get_unchecked(i);
        let s0 = state_f0.get_unchecked(i);
        let s1 = state_f1.get_unchecked(i);
        r0[0] = (*w.get_unchecked(0)).mul_add(*s0, r0[0]);
        r0[1] = (*w.get_unchecked(1)).mul_add(*s0, r0[1]);
        r0[2] = (*w.get_unchecked(2)).mul_add(*s0, r0[2]);
        r0[3] = (*w.get_unchecked(3)).mul_add(*s0, r0[3]);
        r0[4] = (*w.get_unchecked(4)).mul_add(*s0, r0[4]);
        r0[5] = (*w.get_unchecked(5)).mul_add(*s0, r0[5]);
        r0[6] = (*w.get_unchecked(6)).mul_add(*s0, r0[6]);
        r0[7] = (*w.get_unchecked(7)).mul_add(*s0, r0[7]);
        r0[8] = (*w.get_unchecked(8)).mul_add(*s0, r0[8]);
        r0[9] = (*w.get_unchecked(9)).mul_add(*s0, r0[9]);
        r0[10] = (*w.get_unchecked(10)).mul_add(*s0, r0[10]);
        r0[11] = (*w.get_unchecked(11)).mul_add(*s0, r0[11]);
        r0[12] = (*w.get_unchecked(12)).mul_add(*s0, r0[12]);
        r0[13] = (*w.get_unchecked(13)).mul_add(*s0, r0[13]);
        r0[14] = (*w.get_unchecked(14)).mul_add(*s0, r0[14]);
        r0[15] = (*w.get_unchecked(15)).mul_add(*s0, r0[15]);
        r1[0] = (*w.get_unchecked(0)).mul_add(*s1, r1[0]);
        r1[1] = (*w.get_unchecked(1)).mul_add(*s1, r1[1]);
        r1[2] = (*w.get_unchecked(2)).mul_add(*s1, r1[2]);
        r1[3] = (*w.get_unchecked(3)).mul_add(*s1, r1[3]);
        r1[4] = (*w.get_unchecked(4)).mul_add(*s1, r1[4]);
        r1[5] = (*w.get_unchecked(5)).mul_add(*s1, r1[5]);
        r1[6] = (*w.get_unchecked(6)).mul_add(*s1, r1[6]);
        r1[7] = (*w.get_unchecked(7)).mul_add(*s1, r1[7]);
        r1[8] = (*w.get_unchecked(8)).mul_add(*s1, r1[8]);
        r1[9] = (*w.get_unchecked(9)).mul_add(*s1, r1[9]);
        r1[10] = (*w.get_unchecked(10)).mul_add(*s1, r1[10]);
        r1[11] = (*w.get_unchecked(11)).mul_add(*s1, r1[11]);
        r1[12] = (*w.get_unchecked(12)).mul_add(*s1, r1[12]);
        r1[13] = (*w.get_unchecked(13)).mul_add(*s1, r1[13]);
        r1[14] = (*w.get_unchecked(14)).mul_add(*s1, r1[14]);
        r1[15] = (*w.get_unchecked(15)).mul_add(*s1, r1[15]);
    }
    (r0, r1)
}

/// F32-native 8-lane interleaved dot product with init accumulation
/// (scalar reference for SIMD validation).
///
/// Computes `dot_product_8x_f32(weights, state)` then adds `init` element-wise.
///
/// # Safety
/// Caller must ensure `weights.len() >= state.len()`.
#[inline]
pub unsafe fn dot_product_8x_f32_accumulate_scalar(
    weights: &[[f32; 8]],
    state: &[f32],
    init: &[f32; 8],
) -> [f32; 8] {
    let r = dot_product_8x_f32_scalar(weights, state);
    let mut out = [0.0f32; 8];
    for i in 0..8 {
        out[i] = init[i] + r[i];
    }
    out
}

/// F32-native 8-lane interleaved dual-frame dot product with init accumulation
/// (scalar reference for SIMD validation).
///
/// # Safety
/// Caller must ensure `weights.len() >= state_f0.len()` and
/// `weights.len() >= state_f1.len()`.
#[inline]
pub unsafe fn dot_product_8x_f32_dual_accumulate_scalar(
    weights: &[[f32; 8]],
    state_f0: &[f32],
    state_f1: &[f32],
    init_f0: &[f32; 8],
    init_f1: &[f32; 8],
) -> ([f32; 8], [f32; 8]) {
    let (r0, r1) = dot_product_8x_f32_dual_scalar(weights, state_f0, state_f1);
    let mut out_f0 = [0.0f32; 8];
    let mut out_f1 = [0.0f32; 8];
    for i in 0..8 {
        out_f0[i] = init_f0[i] + r0[i];
        out_f1[i] = init_f1[i] + r1[i];
    }
    (out_f0, out_f1)
}

/// F32-native 16-lane interleaved dot product with init accumulation
/// (scalar oracle for SIMD validation).
///
/// This function serves exclusively as the correctness reference in test and
/// benchmark suites. It MUST NOT be called from any production code path.
///
/// # Safety
/// Caller must ensure `weights.len() >= state.len()`.
#[inline]
pub unsafe fn dot_product_16x_f32_accumulate_scalar(
    weights: &[[f32; 16]],
    state: &[f32],
    init: &[f32; 16],
) -> [f32; 16] {
    let r = dot_product_16x_f32_scalar(weights, state);
    let mut out = [0.0f32; 16];
    for i in 0..16 {
        out[i] = init[i] + r[i];
    }
    out
}

/// F32-native 16-lane interleaved dual-frame dot product with init accumulation
/// (scalar oracle for SIMD validation).
///
/// This function serves exclusively as the correctness reference in test and
/// benchmark suites. It MUST NOT be called from any production code path.
///
/// # Safety
/// Caller must ensure `weights.len() >= state_f0.len()` and
/// `weights.len() >= state_f1.len()`.
#[inline]
pub unsafe fn dot_product_16x_f32_dual_accumulate_scalar(
    weights: &[[f32; 16]],
    state_f0: &[f32],
    state_f1: &[f32],
    init_f0: &[f32; 16],
    init_f1: &[f32; 16],
) -> ([f32; 16], [f32; 16]) {
    let (r0, r1) = dot_product_16x_f32_dual_scalar(weights, state_f0, state_f1);
    let mut out_f0 = [0.0f32; 16];
    let mut out_f1 = [0.0f32; 16];
    for i in 0..16 {
        out_f0[i] = init_f0[i] + r0[i];
        out_f1[i] = init_f1[i] + r1[i];
    }
    (out_f0, out_f1)
}