Gearhash gets a 2× speedup on ARM64 with a NEON backend

Speeding up gearhash on ARM64

Gearhash gets a 2× speedup on ARM64 with a NEON backend

The gearhash crate's new NEON backend makes it roughly twice as fast on ARM64 at typical chunk sizes, selected automatically on aarch64. The author details the journey from a disappointing 0.92× direct port to 1.81×, overcoming the serial dependency chain and compiler reassociation. Benchmarks show speedups up to 2.17× for larger chunks, and the crate now powers Hugging Face's Xet client with 10k–20k daily downloads.

You cannot stop the compiler reassociating a sum, but you can (try to) stop it seeing one.

More from this day

2026-09-18