hashbrown

A direct swiss-table port to Rust that became std::HashMap since Rust 1.36 (2019), delivering 2× speedup over the prior Robin Hood implementation. As of version 0.15+, it uses foldhash as its default hasher, replacing ahash. All standard HashMap usage benefits from foldhash's performance with no code changes. With foldhash in place, hashbrown tracks abseil-flat-hash-map to within ~5% — putting Rust's stdlib map in the same league as Google's reference.

Swiss Table mechanics

hashbrown implements the swiss-table design: each 64-bit hash splits into H1 (group selector) and H2 (7-bit fingerprint). H2 is broadcast across 16 metadata bytes via SIMD and compared simultaneously — checking 16 slots in ~3 CPU instructions. The technique, pioneered by Google's absl::flat_hash_map, now underpins hash maps across C++, Rust, and Go (Go 1.24 adopted Swiss Table as its default runtime map: 31% faster inserts, 21% faster lookups, 70% memory reduction at Datadog).

In the C++ family, boost-unordered-flat-map is the current 2025 all-around best performer (overflow byte avoids Abseil's tombstone rot), and folly-f14 wins on memory footprint. See swiss-table and fastest-hash-map-2025 for the full landscape.

The hash function dominates

The most impactful change to Rust hash performance in recent years was not a table change. Switching from SipHash (the safe std::HashMap default for hash-flooding resistance) to foldhash yields 2–5× speedup in hash-heavy workloads. The compiler rustc itself saw a 6% overall speedup from switching to fxhash. Hashbrown ships with foldhash by default; if you control the hasher in a custom hash structure, rapidhash leads overall throughput at 4.25 ns geometric mean.

Public crate vs std

The hashbrown crate exists for users who want a faster HashMap than std's default DDoS-resistant SipHash configuration, or who need newer hashbrown features ahead of the stdlib re-export. Most code should just use std::collections::HashMap — it is hashbrown. Reach for the crate only when you need a non-default hasher (e.g., foldhash explicitly) or hashbrown-specific raw entry APIs.

When to step outside hashbrown

For concurrent workloads, papaya (lock-free reads), dashmap (sharded RwLock), and scc (extreme write contention) cover Rust's options — but if you can use C++, parlayhash hits 1,130 Mops at 128 threads, an order of magnitude past anything Rust currently fields. For static read-only key sets, the phf crate provides compile-time perfect-hashing.