Comprehensive C++ Hashmap Benchmarks (2022)

2 days ago

The benchmark evaluates 29 different C++ hashmap implementations with various allocators and hashes, totaling 174 combinations and 1914 benchmark evaluations.
Key benchmarks include copying maps, inserting and erasing elements, random insert and access, iteration, and find operations for both integers and strings.
Performance varies significantly across implementations, with some excelling in specific areas like search speed, memory usage, or insertion/erasure efficiency.
The author's own implementation, ankerl::unordered_dense::map, is highlighted as an excellent all-rounder with fast search and iteration speeds.
Google's absl::flat_hash_map and absl::node_hash_map are noted for their performance, especially in large maps, but are sensitive to hash quality.
Boost's unordered_map shows significant improvements in lookup speed over std::unordered_map but is slower in insert, erase, and copy operations.
Specialized allocators like PoolAllocator can improve memory usage and performance for node-based containers.
Hash quality significantly impacts performance, with std::hash and boost::hash performing poorly for integral types due to lack of avalanching.
The benchmark infrastructure uses a controlled environment with isolated cores, disabled frequency scaling, and specific compiler flags to ensure consistency.
The results are presented in a detailed table, allowing users to sort and filter based on specific performance metrics.

Hasty Briefsbeta