fast-slice-utils

Highly optimized slice utilities using SIMD instructions when available

Overview

fast-slice-utils provides optimized implementations of a couple of slice operations using platform-specific SIMD instructions (AVX2, NEON, etc.).

The library is no_std compatible.

Functions

• find_prefix_overlap: Returns the number of matching bytes at the start of two slices
• starts_with: A faster replacement for the standard library's slice::starts_with

Platform Support

• x86_64 with AVX2: Uses 256-bit AVX2 instructions
• ARM with NEON: Uses 128-bit NEON instructions
• Portable SIMD: Uses portable SIMD. (requires nightly feature)

Features

• nightly: Enables optimizations that require a nightly compiler (e.g., AVX-512, Portable SIMD).
• miri_safe: Enables compatibility with Miri, at the expense of performance.

Safety (miri issue)

Some implementations may over-read the provided slices, which causes Miri to report errors. The code ensures that it never reads across a page boundary. Enabling the miri_safe feature ensures compatibility with Miri. If a fix can be found (that doesn't hurt performance too badly) we will gladly integrate it.

Contributing

Any contribution is welcome including bug fixes, optimizations, additional functions, and support for additional platforms.