NeuralAmpModeler-rs 3.1.0

An opinionated, high-performance Neural Amp Modeler (NAM) client and core implementation in Rust for Linux/PipeWire and CLAP plugins.
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
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

// SPDX-License-Identifier: Apache-2.0
// Copyright (c) 2026 Fábio Henrique de Lima Silva (fhl.bsb@gmail.com) All rights reserved.
#![allow(
    unsafe_op_in_unsafe_fn,
    clippy::missing_safety_doc,
    clippy::too_many_arguments
)]

//! Dot Product 4x kernels — AVX-512 (interleaved).

use core::arch::x86_64::*;

/// Dot product interleaved 4x with 8 ZMM accumulators (2 alternating sets of 4)
/// to break dependency chains in the FMA pipeline.
///
/// Each ZMM processes 4 state values (16 f32 lanes: 4 output channels × 4 replications).
/// The 2 sets alternate every 16 state elements to keep the FMA pipeline saturated.
///
/// # Safety
/// `weights` must have a length greater than or equal to `state.len()`.
/// Both slices must be valid and accessible for reading.
#[target_feature(enable = "avx512f,avx512vl")]
pub unsafe fn dot_product_4x_interleaved_avx512(weights: &[[u16; 4]], state: &[f32]) -> [f32; 4] {
    let len = state.len().min(weights.len());
    debug_assert!(weights.len() >= len);
    let mut i = 0;

    unsafe {
        let perm_idx = _mm512_set_epi32(3, 3, 3, 3, 2, 2, 2, 2, 1, 1, 1, 1, 0, 0, 0, 0);

        let mut sum_a0 = _mm512_setzero_ps();
        let mut sum_a1 = _mm512_setzero_ps();
        let mut sum_a2 = _mm512_setzero_ps();
        let mut sum_a3 = _mm512_setzero_ps();
        let mut sum_b0 = _mm512_setzero_ps();
        let mut sum_b1 = _mm512_setzero_ps();
        let mut sum_b2 = _mm512_setzero_ps();
        let mut sum_b3 = _mm512_setzero_ps();

        while i + 32 <= len {
            _mm_prefetch::<_MM_HINT_T0>(state.as_ptr().wrapping_add(i + 64) as *const i8);
            _mm_prefetch::<_MM_HINT_T0>(weights.as_ptr().wrapping_add(i + 16) as *const i8);
            _mm_prefetch::<_MM_HINT_T0>(weights.as_ptr().wrapping_add(i + 32) as *const i8);
            _mm_prefetch::<_MM_HINT_T0>(weights.as_ptr().wrapping_add(i + 48) as *const i8);

            let s_a = _mm512_permutexvar_ps(
                perm_idx,
                _mm512_castps128_ps512(_mm_loadu_ps(state.as_ptr().add(i))),
            );
            let w_a =
                _mm512_cvtph_ps(_mm256_loadu_si256(weights.as_ptr().add(i) as *const __m256i));
            sum_a0 = _mm512_fmadd_ps(w_a, s_a, sum_a0);

            let s_b = _mm512_permutexvar_ps(
                perm_idx,
                _mm512_castps128_ps512(_mm_loadu_ps(state.as_ptr().add(i + 4))),
            );
            let w_b = _mm512_cvtph_ps(_mm256_loadu_si256(
                weights.as_ptr().add(i + 4) as *const __m256i
            ));
            sum_a1 = _mm512_fmadd_ps(w_b, s_b, sum_a1);

            let s_c = _mm512_permutexvar_ps(
                perm_idx,
                _mm512_castps128_ps512(_mm_loadu_ps(state.as_ptr().add(i + 8))),
            );
            let w_c = _mm512_cvtph_ps(_mm256_loadu_si256(
                weights.as_ptr().add(i + 8) as *const __m256i
            ));
            sum_a2 = _mm512_fmadd_ps(w_c, s_c, sum_a2);

            let s_d = _mm512_permutexvar_ps(
                perm_idx,
                _mm512_castps128_ps512(_mm_loadu_ps(state.as_ptr().add(i + 12))),
            );
            let w_d = _mm512_cvtph_ps(_mm256_loadu_si256(
                weights.as_ptr().add(i + 12) as *const __m256i
            ));
            sum_a3 = _mm512_fmadd_ps(w_d, s_d, sum_a3);

            let s_e = _mm512_permutexvar_ps(
                perm_idx,
                _mm512_castps128_ps512(_mm_loadu_ps(state.as_ptr().add(i + 16))),
            );
            let w_e = _mm512_cvtph_ps(_mm256_loadu_si256(
                weights.as_ptr().add(i + 16) as *const __m256i
            ));
            sum_b0 = _mm512_fmadd_ps(w_e, s_e, sum_b0);

            let s_f = _mm512_permutexvar_ps(
                perm_idx,
                _mm512_castps128_ps512(_mm_loadu_ps(state.as_ptr().add(i + 20))),
            );
            let w_f = _mm512_cvtph_ps(_mm256_loadu_si256(
                weights.as_ptr().add(i + 20) as *const __m256i
            ));
            sum_b1 = _mm512_fmadd_ps(w_f, s_f, sum_b1);

            let s_g = _mm512_permutexvar_ps(
                perm_idx,
                _mm512_castps128_ps512(_mm_loadu_ps(state.as_ptr().add(i + 24))),
            );
            let w_g = _mm512_cvtph_ps(_mm256_loadu_si256(
                weights.as_ptr().add(i + 24) as *const __m256i
            ));
            sum_b2 = _mm512_fmadd_ps(w_g, s_g, sum_b2);

            let s_h = _mm512_permutexvar_ps(
                perm_idx,
                _mm512_castps128_ps512(_mm_loadu_ps(state.as_ptr().add(i + 28))),
            );
            let w_h = _mm512_cvtph_ps(_mm256_loadu_si256(
                weights.as_ptr().add(i + 28) as *const __m256i
            ));
            sum_b3 = _mm512_fmadd_ps(w_h, s_h, sum_b3);

            i += 32;
        }

        while i + 16 <= len {
            let s_a = _mm512_permutexvar_ps(
                perm_idx,
                _mm512_castps128_ps512(_mm_loadu_ps(state.as_ptr().add(i))),
            );
            let w_a =
                _mm512_cvtph_ps(_mm256_loadu_si256(weights.as_ptr().add(i) as *const __m256i));
            sum_a0 = _mm512_fmadd_ps(w_a, s_a, sum_a0);

            let s_b = _mm512_permutexvar_ps(
                perm_idx,
                _mm512_castps128_ps512(_mm_loadu_ps(state.as_ptr().add(i + 4))),
            );
            let w_b = _mm512_cvtph_ps(_mm256_loadu_si256(
                weights.as_ptr().add(i + 4) as *const __m256i
            ));
            sum_a1 = _mm512_fmadd_ps(w_b, s_b, sum_a1);

            let s_c = _mm512_permutexvar_ps(
                perm_idx,
                _mm512_castps128_ps512(_mm_loadu_ps(state.as_ptr().add(i + 8))),
            );
            let w_c = _mm512_cvtph_ps(_mm256_loadu_si256(
                weights.as_ptr().add(i + 8) as *const __m256i
            ));
            sum_a2 = _mm512_fmadd_ps(w_c, s_c, sum_a2);

            let s_d = _mm512_permutexvar_ps(
                perm_idx,
                _mm512_castps128_ps512(_mm_loadu_ps(state.as_ptr().add(i + 12))),
            );
            let w_d = _mm512_cvtph_ps(_mm256_loadu_si256(
                weights.as_ptr().add(i + 12) as *const __m256i
            ));
            sum_a3 = _mm512_fmadd_ps(w_d, s_d, sum_a3);

            i += 16;
        }

        while i + 4 <= len {
            let s = _mm512_permutexvar_ps(
                perm_idx,
                _mm512_castps128_ps512(_mm_loadu_ps(state.as_ptr().add(i))),
            );
            let w = _mm512_cvtph_ps(_mm256_loadu_si256(weights.as_ptr().add(i) as *const __m256i));
            sum_a0 = _mm512_fmadd_ps(w, s, sum_a0);
            i += 4;
        }

        let s_ab = _mm512_add_ps(_mm512_add_ps(sum_a0, sum_a1), _mm512_add_ps(sum_a2, sum_a3));
        let s_all = _mm512_add_ps(
            s_ab,
            _mm512_add_ps(_mm512_add_ps(sum_b0, sum_b1), _mm512_add_ps(sum_b2, sum_b3)),
        );

        let lo = _mm512_extractf32x4_ps(s_all, 0);
        let hi0 = _mm512_extractf32x4_ps(s_all, 1);
        let hi1 = _mm512_extractf32x4_ps(s_all, 2);
        let hi2 = _mm512_extractf32x4_ps(s_all, 3);
        let mut sum128 = _mm_add_ps(_mm_add_ps(lo, hi0), _mm_add_ps(hi1, hi2));

        while i < len {
            let s0 = _mm_load1_ps(state.as_ptr().add(i));
            let w0 = _mm_cvtph_ps(_mm_loadu_si64(
                weights.as_ptr().add(i) as *const u16 as *const u8
            ));
            sum128 = _mm_fmadd_ps(w0, s0, sum128);
            i += 1;
        }

        let mut out = [0.0; 4];
        _mm_storeu_ps(out.as_mut_ptr(), sum128);
        out
    }
}