NeuralAmpModeler-rs 0.1.0

High-performance Neural Amp Modeler DSP core: WaveNet/LSTM/ConvNet inference, SIMD math (x86-64-v3), .nam/.namb loader, cabinet IR, resampling and noise gate.
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

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

//! SIMD register, vector, and basic operation abstractions.

use super::super::InstructionSet;

/// Abstraction trait for SIMD register types (e.g., `__m256`, `__m512`).
pub trait SimdRegister: Copy {
    /// ISA classification associated with this register type.
    const ISA: InstructionSet;
}

/// Vector abstraction trait for SIMD math containers.
pub trait SimdVector {
    /// Element scalar type.
    type Element;
    /// Register representation type.
    type Register: SimdRegister;
}

/// Floating-point SIMD vector trait.
pub trait SimdFloat: SimdVector {
    /// Indicates whether this vector precision represents BF16.
    const IS_BF16: bool = false;
}

/// Basic SIMD operation abstractions.
pub trait SimdOps: SimdFloat {
    /// Store register contents as packed BF16 values to memory.
    ///
    /// # Safety
    /// `ptr` must point to allocated memory with sufficient capacity for the register.
    unsafe fn store_bf16(ptr: *mut u16, v: Self::Register);
}