crc32fast

crc32fast is a Rust crate providing fast, SIMD-accelerated CRC32 (IEEE) checksum computation. It ships a portable baseline implementation alongside optimized paths using sse/pclmulqdq on x86 and crc32 instructions on aarch64, with automatic runtime CPU feature detection choosing the fastest available implementation via the Hasher::new constructor.

With over half a billion downloads, it is one of the most widely depended-upon crates in the Rust ecosystem, used transitively by compression, archive, and networking crates that need CRC32 checksums (such as flate2 and zip). It supports no_std environments for embedded use when the optional std feature is disabled.

What You Get

• A simple hash(&[u8]) -> u32 convenience function for one-shot checksum computation
• An incremental Hasher type for streaming large inputs through update/finalize
• Runtime CPU feature detection that transparently picks the fastest available implementation (baseline, pclmulqdq on x86, crc32 on aarch64)
• no_std support via an optional std feature, for embedded and kernel-level use
• Fuzz-tested (via cargo-fuzz/libFuzzer) and address-sanitized unsafe SIMD code paths

Common Use Cases

• Computing CRC32 checksums for file formats that require them (ZIP, gzip, PNG) inside archive/compression crates
• Verifying data integrity for network protocols or storage formats that embed a CRC32 trailer
• High-throughput checksum computation in performance-sensitive Rust services processing large byte streams
• Embedded/no_std firmware that needs a dependency-light CRC32 implementation

Under The Hood

Architecture - The crate is organized around a public Hasher and a free hash function in lib.rs, which dispatch to one of several backend implementations: baseline.rs provides a portable, table-driven implementation (table.rs holds the precomputed CRC lookup tables) used whenever no SIMD path is available or std is disabled, while src/specialized/ contains architecture-specific pclmulqdq (x86/x86_64) and crc32 (aarch64) intrinsic implementations selected via runtime CPU feature detection performed once in Hasher::new. combine.rs implements CRC combination math so two checksums computed over adjacent byte ranges can be merged without re-hashing, enabling parallel/chunked checksum computation. Tech Stack - Pure Rust, edition 2021, with a single runtime dependency (cfg-if) for conditional compilation across the baseline/specialized backends; build.rs detects target capabilities at compile time. Dev-dependencies (bencher, quickcheck, rand) support benchmarking and property-based testing. Code Quality - Correctness of the unsafe SIMD code is validated through property-based tests (quickcheck) plus fuzz testing via cargo-fuzz/libFuzzer across debug and release profiles (see the fuzz/ directory), and every commit is run under an address sanitizer in CI — a notably rigorous safety bar for a crate built around raw SIMD intrinsics. API Design - The API is deliberately minimal: a one-line hash() for simple cases and a Hasher with update/finalize for streaming, mirroring the shape of Rust’s standard std::hash::Hasher trait so it feels idiomatic to Rust users, with feature flags (std, nightly) clearly documented for constrained targets.