GitHub speeds case‑folding to >45 GiB/s per core, slashing code‑search latency
They built a branch‑free, byte‑space arithmetic loop in the Rust crate casefold, processing every byte of source code at over 45 GiB/s on a single core. This lets GitHub’s Blackbird index and query‑match 180 million repositories faster, because sweeping the whole buffer—rather than stopping at the first non‑ASCII byte—wins on the ASCII‑heavy workload.