• Simdjson Shows More Speed-Ups Possible For SIMD In JSON Parsing: Another 30% Boost Simdjson is the open-source project for high performance JSON parsing by leveraging SIMD instructions for “parsing gigabytes of JSON per second.” Notably it showed years ago the huge performance advantage to using AVX-512 in JSON parsing for surprisingly big benefits. • Simdjson has continued advancing since then with various optimizations over the years and today is out with simdjson 4.3 that brings yet more SIMD optimizations. • Simdjson 4.3 released today and now handles runtime dispaching on LoongArch around the LSX and LASX instructions, Microsoft’s Visual C++ Group improving the build throughput by 30% through build optimizations, and various other enhancements. • Catching my eye with the new simdjson 4.3 release is this pull for SIMD string escaping and batch integer formatting optimizations. • For string-heavy JSON parsing it can yield some massive benefits. • This pull that is part of simdjson 4.3 yields a 30% improvement for the Twitter benchmark and a 6% improvement for the CITM Catalog benchmark that is more numeric heavy than string heavy.
Article Summaries:
- Simdjson, the open‑source JSON parsing library that uses SIMD instructions, has released version 4.3 today. The update adds runtime dispatch for LoongArch LSX and LASX, improves Microsoft Visual C++ build throughput by 30 %, and introduces new SIMD optimizations for string escaping and batch integer formatting. Benchmarks show a 30 % speed‑up on the Twitter dataset and a 6 % improvement on the CITM Catalog, which is more numeric‑heavy. The changes target ARM64 NEON and SSE2 instruction sets, aiming to boost performance for string‑heavy JSON workloads. The new release and benchmarks are available on GitHub.
Sources: