fashex - fast hexadecimal encoding and decoding

Hexadecimal string encoding and decoding with best-effort SIMD acceleration.

Highlights

• Best-effort SIMD acceleration.
• Small, focused public APIs.

Performance

This crate is comparable in performance to hex-simd, slightly better than const-hex, and far superior to faster-hex.

For benchmark results, please refer to BENCHMARK.md.

Usage

Adds fashex as a dependency in your Cargo.toml:

[dependencies]
fashex = "*" # Please specify the latest version instead of "*"!

For detailed usage instructions, please refer to the documentation.

Features

• alloc (default)

  Enables Rust alloc crate support.

• std (default)

  Enables Rust standard library support.

• nightly

  Enables features that require nightly Rust.

• portable-simd

  Enables portable SIMD acceleration support via unstable core::simd, which may help improve performance on platforms other than x86, x86_64, aarch64 (arm64ec), loongarch32 and loongarch64.

  Notes: this requires nightly Rust.

• experimental-loongarch-simd

  Enables experimental SIMD acceleration via Loongson SIMD eXtention or Loongson Advanced SIMD eXtention.

  Notes: this requires nightly Rust.

• runtime-cpu-detection (default)

  Enables CPU features detection at runtime.

no_std support

This crate is no_std-friendly, just disable the default features:

[dependencies]
fashex = { version = "*", default-features = false, features = [] } # Please specify the latest version instead of "*"!

Some notes:

1. You will probably need to enable "alloc" feature manually.
2. OS-independent no_std-friendly CPU features detection is only supported on x86 and x86_64 platforms. On other platforms, runtime-cpu-detection does nothing when "std" feature is disabled.

Disclaimer

Although all tests are passing and hours of fuzzing show no failures, latent bugs may still exist in the code.

1. The SIMD-accelerated code paths for x86 / x86_64 architecture are well tested and fuzzed.
2. The SIMD-accelerated code paths for aarch64 architecture are tested and fuzzed.
3. The SIMD-accelerated code paths for loongarch64 / loongarch32 architecture are cross-tested, but not extensively fuzzed.

If you find any bugs, please report them.

Acknowledgements

This crate is inspired by the following repositories:

• hex-simd, SPDX-License-Identifier: MIT
• const-hex, SPDX-License-Identifier: MIT OR Apache-2.0
• faster-hex, SPDX-License-Identifier: MIT OR Apache-2.0

License