Rust for C-Programmers

22.7 Introduction to Rayon for Data Parallelism

Parallelizing loops by manually spawning threads can be tedious. Rayon is a popular crate that automates data-parallel operations. You write code using iterators, and Rayon splits the work across a thread pool, using work stealing for load balancing.

22.7.1 Basic Rayon Usage

Add Rayon to your Cargo.toml:

[dependencies]
rayon = "1.7"

Then:

use rayon::prelude::*;

Replace .iter() or .iter_mut() with .par_iter() or .par_iter_mut():

use rayon::prelude::*;

fn main() {
    let numbers: Vec<u64> = (0..1_000_000).collect();
    let sum_of_squares: u64 = numbers
        .par_iter()
        .map(|x| x.pow(2))
        .sum();

    println!("Sum of squares = {}", sum_of_squares);
}

Rayon automatically manages thread creation and scheduling behind the scenes.

22.7.2 Balancing and Performance

Although Rayon simplifies parallelism, for very small datasets or trivial computations, its overhead might outweigh the gains. Always profile to ensure parallelization is beneficial.

22.7.3 The join() Function

Rayon also provides join() to run two closures in parallel:

fn parallel_compute() -> (i32, i32) {
    rayon::join(
        || heavy_task_1(),
        || heavy_task_2(),
    )
}

fn heavy_task_1() -> i32 { 42 }
fn heavy_task_2() -> i32 { 47 }

Internally, Rayon reuses a fixed-size thread pool and balances workloads via work stealing.