kylematsuda / cornedbeef

Writing a hashmap in Rust

cornedbeef

Writing a hashmap in Rust for learning and fun.

Please see the accompanying blog posts:

• Part 1: Intro and two naive maps
• Part 2: Swiss Table metadata and MaybeUninit
• Part 3a: SIMD probing
• Part 3b: Exception safety
• Part 3c: Resizing with SIMD

The design was inspired by Google's Swiss Tables map and Rust's std::collections::HashMap aka hashbrown (based on Swiss Tables).

This repo contains 6 iterations on a hashmap, building from a naive design toward (a simplified) Swiss Tables:

• first::Map: separate chaining using std::collections::LinkedList
• second::Map: open addressing (quadratic probing)
• third::Map: open addressing with Swiss tables metadata
• fourth::Map: same as third but using std::mem::MaybeUninit as an optimization
• fifth::Map: same as fourth but adding SIMD probing
• sixth::Map (unfinished): same as fifth but putting the metadata and backing storage in the same allocation (with a lot of unsafe)

Speed comparison with std

These are done with the benchmarks in /benches.

Reported times are for my laptop (Intel i7-10750H, 32 GB RAM, Fedora 37).

Benchmark name	std runtime (ms)	fifth::Map runtime (ms)	Ratio
insert_grow_seq 1	3.85	4.58	1.19
insert_grow_seq 8	4.81	5.79	1.20
insert_grow_random 1	3.95	4.75	1.20
insert_grow_random 8	4.87	5.87	1.21
insert_reserved 1	2.03	2.26	1.11
insert_reserved 8	2.47	2.74	1.11
lookup 1	2.14	2.61	1.22
lookup 8	3.11	4.3	1.38
lookup string 1	4.32	4.61	1.07
lookup string 8	7.07	7.17	1.01
lookup miss 1	1.84	1.99	1.08
lookup miss 8	2.02	2.26	1.12
remove 1	3.07	3.49	1.14
remove 8	5.00	5.60	1.12