#cache-friendly #byte #sparse-set #container

sparse_set_container

A container based on sparse set. Stable keys, O(1) lookup, cache-friendly iterations, and no hashing.

3 stable releases

1.1.1 Jun 22, 2024
1.0.1 Jun 8, 2024
1.0.0 May 25, 2024

#293 in Data structures

MIT license

73KB
1K SLoC

Sparse Set Container

A container based on a sparse set.

It is useful if you want a container with performance close to Vec but you also want to safely store the indexes to the elements (so that they are not invalidated on removals).
E.g. you have a list of elements in UI that the user can add and remove, but you want to refer to the elements of that list from somewhere else.

crates.io Documentation

Download Status

Usage

Add this to your Cargo.toml:

[dependencies]
sparse_set_container = "1.1"

Description

An array-like container based on sparse set implementation that allows O(1) access to elements without hashing and allows cache-friendly iterations.

Operation SparseSet Vec
push O(1) O(1)
lookup O(1) O(1)
size/len O(1) O(1)
remove O(n) O(n)
swap_remove O(1) O(1)

For iterating over the elements SparseSet exposes an iterator over an internal Vec with values, which is as efficient as iterating over a Vec directly.

Differences to Vec:

  • Instead of using indexes, when adding an element, it returns a lightweight key structure that can be used to access the element later
    • The key is not invalidated when elements are removed from the container
    • If the pointed-at element was removed, the key will not be pointing to any other elements, even if new elements are inserted
  • There is a slight overhead in insertion/lookup/removal operations compared to Vec
  • Consumes more memory:
    • for each value 4*sizeof(usize) bytes on top of the size of the element itself
      • (e.g. 32 bytes per element on 64-bit systems)
    • per each 2^(sizeof(usize)*8) removals the memory consumption will also grow by 2*sizeof(usize)
      • (e.g. 16 bytes per 18446744073709551616 elements removed on 64-bit systems)
  • Many Vec operations are not supported (create an issue on github if you want to request one)

Examples

extern crate sparse_set_container;
use sparse_set_container::SparseSet;

fn main() {
    let mut elements = SparseSet::new();
    elements.push("1");
    let key2 = elements.push("2");
    elements.push("3");

    elements.remove(key2);
    elements.push("4");

    if !elements.contains(key2) {
        println!("Value 2 is not in the container");
    }

    // Prints 1 3 4
    for v in elements.values() {
        print!("{} ", v);
    }

    // Prints 1 3 4
    for k in elements.keys() {
        print!("{} ", elements.get(k).unwrap());
    }
}

Benchmarks

The values captured to illustrate the difference between this SparseSet container implementation, Vec, and standard HashMap:

Benchmark SparseSet<String> Vec<String> HashMap<i32, String>
Create empty 0 ns ±0 0 ns ±0 1 ns ±0
Create with capacity 17 ns ±0 16 ns ±0 32 ns ±1
Push 100 elements 3,254 ns ±14 2,553 ns ±23 5,493 ns ±85
With capacity push 100 3,286 ns ±30 3,156 ns ±106 4,388 ns ±21
Lookup 100 elements 88 ns ±2 39 ns ±14 464 ns ±3
Iterate over 100 elements 30 ns ±0 30 ns ±0 41 ns ±1
Clone with 100 elements 2,184 ns ±23 2,109 ns ±4 1,490 ns ±32
Clone 100 and remove 10 3,055 ns ±107 2,364 ns ±97 1,692 ns ±145
Clone 100 and swap_remove 10 2,475 ns ±119 2,193 ns ±67 N/A

To run the benchmark on your machine, execute cargo run --example bench --release

Or to build this table you can run python tools/collect_benchmark_table.py and then find the results in bench_table.md

License

Licensed under the MIT license: http://opensource.org/licenses/MIT

No runtime deps