accelerate-rs

Safe Rust bindings for Apple's Accelerate framework on macOS using a Swift bridge over the C APIs.

The GitHub repository is accelerate-rs; the published crates.io package is apple-accelerate.

Install

cargo add apple-accelerate

Enable the raw-ffi feature if you also want the underlying C declarations:

cargo add apple-accelerate --features raw-ffi

Quick start

use apple_accelerate::{
    add_f32, integrate, solve_linear_system_f32, sdot, QuadratureOptions,
};

fn main() -> Result<(), Box<dyn std::error::Error>> {
    let added = add_f32(&[1.0, 2.0, 3.0], &[4.0, 5.0, 6.0])?;
    assert_eq!(added, vec![5.0, 7.0, 9.0]);

    let dot = sdot(&[1.0, 2.0, 3.0], &[4.0, 5.0, 6.0])?;
    assert_eq!(dot, 32.0);

    // LAPACK matrices are column-major.
    let solution = solve_linear_system_f32(&[3.0, 1.0, 1.0, 2.0], 2, &[9.0, 8.0])?;
    assert!((solution[0] - 2.0).abs() < 1.0e-5);
    assert!((solution[1] - 3.0).abs() < 1.0e-5);

    let integral = integrate(|x| x * x, 0.0, 1.0, QuadratureOptions::default())?;
    assert!((integral.integral - (1.0 / 3.0)).abs() < 1.0e-9);
    Ok(())
}

v0.2 surface

• vDSP: FFT setup, biquad setup, vector arithmetic, reductions, and window generation
• vForce: element-wise transcendental and root functions over f32 slices
• BLAS: sdot, row-major sgemv, and row-major sgemm
• LAPACK: LU factorization and linear solves for column-major single-precision matrices
• BNNS: safe ReLU/sigmoid vector activations plus the existing thin unsafe filter owner
• Sparse: sparse-vector dot products and sparse-to-dense accumulation
• vImage: ARGB8888 rotate / box-convolve / scale and Planar8 contrast stretch
• simd: SIMD4 add / dot / length / normalize helpers
• Quadrature: one-dimensional adaptive numerical integration with Rust closures
• raw-ffi feature: re-exports the underlying C declarations for all wrapped areas

Smoke examples

for ex in examples/*.rs; do
  cargo run --example "$(basename "$ex" .rs)"
done

The numbered examples cover every logical area of the crate.