emhash

A C++ hash map family that wins raw insert/erase throughput and tolerates extreme load factors up to 0.999. In Martin Ankerl's 2022 benchmark suite (29 hash maps × 6 hash functions × 11 benchmarks), emhash7 clocked 3.6 ns/op for integer find-hit versus 15 ns for std::unordered_map — a 4.2× advantage and the fastest result on insert-and-erase workloads.

What makes it fast

Three-way linear probing with x86 bit-scan instructions, no tombstones. Where Abseil degrades under heavy insert/erase cycling because of tombstone accumulation, emhash sidesteps the problem with active reorganization. Combined with linear probe groups it stays cache-friendly even when the table is nearly full.

The 0.999 load factor is the headline: most open-addressed hash maps cap at 0.875 (boost-unordered-flat-map, abseil-flat-hash-map) because collision costs dominate above that. emhash's probing strategy keeps unsuccessful lookups bounded even at 99.9% occupancy, which makes it the tightest-packing option for memory-pressured workloads.

The catch

Less community adoption and testing than boost-unordered-flat-map or abseil-flat-hash-map. The author has shipped multiple variants (emhash5, emhash6, emhash7, emhash8) with subtly different tradeoffs, and there is no official documentation explaining when to pick which. Production deployments are rarer; the API is less polished.

When to use it

• Insert/erase throughput is the dominant cost.
• Memory pressure forces extreme load factors.
• You can afford to benchmark variants and pick.

For general-purpose C++ work, boost-unordered-flat-map is the safer default. emhash is a specialist's tool that delivers when you have the workload to justify the integration cost. See fastest-hash-map-2025 for context.