tiny_sort

Crates.iotiny_sort
lib.rstiny_sort
version1.0.5
sourcesrc
created_at2023-06-11 15:41:04.503685
updated_at2023-12-20 17:08:56.68935
descriptionBinary-size optimized stable and unstable sorts
homepage
repositoryhttps://github.com/Voultapher/tiny-sort-rs
max_upload_size
id887452
size34,171
Lukas Bergdoll (Voultapher)

documentation

https://docs.rs/tiny_sort

README

github crates.io docs.rs

tiny_sort binary-size optimized sort implementations

The tiny_sort crate provides two sort implementations tiny_sort::stable::sort and tiny_sort::unstable::sort. The crate is no_std and both versions can be disabled via features, by setting default-features = false. tiny_sort::stable::sort requires alloc, tiny_sort::unstable::sort doesn't. In addition to fn sort<T: Ord>(v: &mut [T]), both sort implementations also provide fn sort_by<T, F: FnMut(&T, &T) -> Ordering>(v: &mut [T], mut compare: F) to sort with a custom comparison function.

Use these sort implementations if you care about binary-size more than you care about performance. Otherwise use slice::sort and slice::sort_unstable.

A closer look at the implementations.

stable::sort

The stable sort is a branchless mergesort. This means:

  • Guaranteed O(N * log(N)) worst case perf
  • No adaptiveness
  • Branch miss-prediction not affected by outcome of comparison function
  • Allocates N auxiliary memory.

unstable::sort

The unstable sort is a branchless heapsort. This means:

  • Guaranteed O(N * log(N)) worst case perf
  • No adaptiveness
  • Branch miss-prediction not affected by outcome of comparison function

Benchmarks

Setup

Linux 6.3
rustc 1.76.0-nightly (f704f3b93 2023-12-19)
clang version 15.0.7
gcc (GCC) 13.1.1 20230429
AMD Ryzen 9 5900X 12-Core Processor (Zen3 micro-architecture)
CPU boost enabled.

Contestants:

- rust_tinymergesort_stable    | This crate' `stable::sort`
- rust_std_stable              | `slice::sort` https://github.com/rust-lang/rust (1)
- cpp_std_gnu_stable           | libstdc++ `std::sort_stable` (2)
- cpp_std_libcxx_stable        | libc++ `std::sort_stable` (3)

- rust_tinyheapsort_unstable   | This crate' `unstable::sort`
- rust_std_unstable            | `slice::sort_unstable` https://github.com/rust-lang/rust (1)
- cpp_std_gnu_unstable         | libstdc++ `std::sort` (2)
- cpp_std_libcxx_unstable      | libc++ `std::sort` (3)

Footnotes:

  1. Vendored ca. mid 2022.
  2. Built with gcc.
  3. Built with clang.

Binary-size

A minimal program is compiled with --release, lto = "thin" and opt-level = "s" for the Rust code and -Os for the header only C++ code. The C++ code is compiled with gcc. And the resulting binary is stripped with strip.

#[inline(never)]
fn instantiate_sort<T: Ord + std::fmt::Display>(mut v: Vec<T>) {
    tiny_sort::unstable::sort(&mut v);

    // side-effect
    println!("{}", v[v.len() - 1]);
}

fn main() {
    use std::env;
    let len = env::args().len();

    // The vec pattern is hard to predict for the compiler.
    // And the len is unknown by design.
    // Plus instantiate_sort is forced to inline never which means it has to be
    // compiled in a way that could accept all possible layout of v.
    instantiate_sort((0..len as u64).rev().collect());
}

The baseline with the sort uncommented is: 292_864 bytes. The values below are the stripped binary size subtracted from the baseline.

- rust_tinymergesort_stable    | 528 bytes
- rust_std_stable              | 2928 bytes
- cpp_std_gnu_stable           | 5528 bytes
- cpp_std_libxx_stable         | 4368 bytes

- rust_tinyheapsort_unstable   | 320 bytes
- rust_std_unstable            | 3848 bytes
- cpp_std_gnu_unstable         | 2128 bytes
- cpp_std_libcxx_unstable      | 1272 bytes

Run-time

A rough estimate what kind of performance you can get with these sort implementations. If you care about performance use slice::sort and slice::sort_unstable.

stable::sort

hot-u64-10k
cold-u64-scaling-random

unstable::sort

hot-u64-10k
cold-u64-scaling-random

Min required rustc version

The minimum required rustc version is 1.51.

The minimum versions are best-effort and may change with any new major release.

Contributing

Please respect the CODE_OF_CONDUCT.md when contributing.

Versioning

We use SemVer for versioning. For the versions available, see the tags on this repository.

Authors

See also the list of contributors who participated in this project.

License

This project is licensed under the Apache License, Version 2.0 - see the LICENSE.md file for details.

Commit count: 27

cargo fmt