Simple tricks for faster Rust programs

Rust is pretty fast. Let’s get that out of the way. But sometimes, pretty fast is not fast enough.

Fortunately, it’s also pretty easy to slightly improve the performance of your Rust binaries with minimal code changes. I’m gonna go over some of these tricks that I’ve picked up from many sources across the web (I’ll post a small list of very good blogs run by smart Rustaceans who cover interesting Rust related things).

Turn on full LTO#

Rust by default runs a “thin” LTO pass across each individual codegen unit. This can be optimized with a very simple addition to your Cargo.toml

codegen-units = 1
lto = "fat"

This makes the following changes to the release profile:

  • Forces rustc to build the entire crate as a single unit, which lets LLVM make smarter decisions about optimization thanks to all the code being together.
  • Switches LTO to the fat variant. In fat mode, LTO will perform optimization across the entire dependency graph as opposed to the default option of doing it just to the local crate.

Use a different memory allocator#

Some time ago, Rust switched from using jemalloc on all platforms to the OS-native allocator. This caused serious performance regressions in many programs like fd. To switch back to jemalloc, check out this PR for the changes required.

Note that this alone is not guaranteed to be helpful, and a lot of programs see little to no benefit, so please run your own benchmarks with hyperfine to confirm whether or not it helped you.


Rustaceans love their cows, and it’s one of the most underrated APIs in the Rust standard library. It’s claim to fame is relatively simple - it’s a smart copy-on-write pointer. Or well, a smart clone-on-write pointer, as copy means something different in Rust as opposed to other languages.

Given a data wrapped in a std::borrow::Cow, you can avoid cloning the data if you only want immutable read access, which saves memory and improves runtime performance as well. Over a large codebase, these savings pile up to create a noticeable enough difference. Here’s an example from the Rust standard library that explains this well.

use std::borrow::Cow;

fn abs_all(input: &mut Cow<[i32]>) {
    for i in 0..input.len() {
        let v = input[i];
        if v < 0 {
            // Clones into a vector if not already owned.
            input.to_mut()[i] = -v;

// No clone occurs because `input` doesn't need to be mutated.
let slice = [0, 1, 2];
let mut input = Cow::from(&slice[..]);
abs_all(&mut input);

// Clone occurs because `input` needs to be mutated.
let slice = [-1, 0, 1];
let mut input = Cow::from(&slice[..]);
abs_all(&mut input);

// No clone occurs because `input` is already owned.
let mut input = Cow::from(vec![-1, 0, 1]);
abs_all(&mut input);


  • Pascal Hertleif’s blog - He’s a very popular and active Rust developer and writes amazing, insightful articles.
  • Amos Wenger’s blog - Amos' articles often go over important topics like API design through a comparison angle between Rust and another language to highlight differences and benefits to each approach.
  • Stjepan Glavina’s blog - He’s done a lot of interesting perf-related work including optimising sorting in the stdlib and building async libraries. His writeups for the library work are very intriguing and go into great detail about the process.

You can sponsor me through GitHub to support this blog, my work on Password Store and all my other open source projects.