AXPY

A macro-based alternative to expression templates for efficient n-ary linear combinations of slice-like objects, i.e. objects that implement .iter() and .iter_mut(). Compiled with optimizations, resulting source code elides bound checks and will be auto-vectorized by LLVM.

Examples

#[macro_use] extern crate axpy;
fn test(a: f64, x: &[f64], y: &[f64], z: &mut [f64]) {
    // some random expression
    axpy![z = a * x + z - 2.*y];

    // this becomes:
    // for (z, (x, y)) in z.iter_mut().zip(x.iter().zip(y.iter())) {
    //     *z = a * *x + 1. * *z - 2. * *y;
    // }
}

Virtually any "reasonable" linear combination of any number of vectors (up to the compiler macro recursion limit) is permitted, along with other assignment statements, e.g. += or -= in addition to =. The assigned variable may freely appear anywhere in the expression, permitting in-place modifications without auxiliary variables. Refer to the source code for more information -- as far as macro code goes, it is fairly well commented. The only restriction is that in the expression, if a scalar and vector entry are multiplied, the scalar must occur on the left.

License

Licensed under

• Apache License 2.0, or
• MIT License, or
• BSD 2-Clause License,

at your option.

Contribution

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be tri-licensed as above, without any additional terms or conditions.

Acknowledgments

• static-cond was how I learned to do token equality matching, which was used in this code to permit the assigned variable appearing throughout the expression.

Versions

• 0.3.0 -- restored ability to use integer types by rethinking macro patterns
• 0.2.0 -- simplification of macro by relying on further (verified) optimizations (e.g. 1*x and x-0 are no-ops)
• 0.1.0 -- initial implementation of macro