The Fastest Way to Match Characters on ARM Processors?
The blog post demonstrates that ARM’s new SVE2 `match` instruction can classify JSON structural and whitespace characters faster than traditional NEON code. By loading a 16‑byte lookup set into a vector, `match` produces a predicate mask in a single step, eliminating multiple equality comparisons. Benchmarks on an AWS Graviton 4 (Neoverse V2) show 16 GB/s throughput and a 25 % reduction in instruction count versus the NEON implementation. The author argues that SVE2 may now be the fastest way to match characters on ARM processors.
Converting Floats to Strings Quickly
Converting binary floating‑point numbers to decimal strings is a core step in JSON, CSV, and logging pipelines. Recent research benchmarks modern algorithms—Dragonbox, Schubfach, and Ryū—showing they are roughly ten times faster than the original Dragon4 from 1990. The study finds...