UMASH is an almost-universal family of hash functions. Each Params struct defines a specific hash function; when the parameters are generated pseudorandomly, the probability that two different inputs of up to s bytes collide (for an independently generated set of parameters) is at most ceil(s / 4096) 2^-55 for the 64-bit hash. The 128-bit fingerprint reduces that probability to less than 2^-70 for inputs of 1 GB or less.

See the reference repo for more details and proofs.

umash-rs: a rust API for UMASH

UMASH is a family of fast hash / fingerprinting functions with collision bounds. This crate builds on x86-64 (with CLMUL) and little-endian aarch64 (with VMULL). The UMASH family of functions is defined independently of hardware specific features, so computes the same values on both architecture.

See https://github.com/backtrace-labs/umash for more details.