docs/guides/cmsis-dsp

CMSIS-DSP Guide

Home | Installation | Getting Started | CMSIS-DSP

---

Nimphea includes a comprehensive wrapper for ARM's CMSIS-DSP library, providing hardware-accelerated math functions optimized for the Cortex-M7 processor.

Features

• Fast Math: Optimized sine, cosine, and square root functions.
• Vector Operations: Element-wise addition, multiplication, and scaling.
• Filtering: FIR and Biquad (IIR) filters with zero heap allocation.
• Transforms: Real and Complex FFT (up to 4096 points).
• Matrix: High-performance matrix arithmetic.

Basic Usage

All CMSIS functions are designed to be real-time safe (no heap allocations).

Fast Math

import nimphea/cmsis/dsp_fastmath

let s = fastSin(PI / 2.0)
let root = fastSqrt(16.0)

Vector Operations

import nimphea/cmsis/dsp_basic

var v1 = [1.0'f32, 2.0, 3.0, 4.0]
var v2 = [10.0'f32, 20.0, 30.0, 40.0]
var res: array[4, float32]

add(res, v1, v2) # res = [11, 22, 33, 44]

Advanced DSP

FIR Filtering

Filters use compile-time sizing for maximum efficiency.

import nimphea/cmsis/dsp_filtering

const coeffs: array[3, float32] = [0.5, 0.5, 0.5]
var filter: FirFilter[3, 48] # 3 taps, max block size 48
filter.init(addr coeffs[0])

proc audioCallback(input, output: AudioBuffer, size: int) {.cdecl.} =
  filter.process(input[0], output[0])

Fast Fourier Transform (FFT)

import nimphea/cmsis/dsp_transforms

var fft: FftInstance[256] # 256-point complex FFT
fft.init()

var data: array[512, float32] # Interleaved [re, im, re, im, ...]
fft.forward(data) # In-place FFT

Performance Tips

1. Use static arrays: Avoid seq in your audio path.
2. Pre-initialize: Initialize filters and FFT instances in your main() or setup code, not inside the audio callback.
3. Use the right type: Prefer float32 (f32) for most audio tasks as it is hardware-accelerated by the FPU.