Crate hamming_bitwise_fast 

Fast bitwise Hamming distance using auto-vectorization with runtime SIMD detection on x86.

Quick Start

use hamming_bitwise_fast::array;

let a: [u8; 128] = [0xFF; 128];  // 1024-bit vectors
let b: [u8; 128] = [0x00; 128];

// Single comparison
let distance = array::distance(&a, &b);  // 1024

// One source vs many targets
let targets = vec![a, b];
let mut distances = vec![0u32; 2];
array::batch(&a, &targets, &mut distances);

Choosing an API

Fixed-size arrays vs slices

If the vector size is known at compile time (e.g., 1024-bit embeddings are [u8; 128]), use the array module for the best performance.

Use slice when sizes vary at runtime or are not known until program execution.

Single vs batch

Use array::batch or slice::batch when comparing one source against many targets. Batch is the fastest approach for one-to-many comparisons.

Platform Behavior

Platform	Configuration	Behavior
x86/x86_64	Default	Runtime CPU detection via multiversion (AVX-512/AVX2/SSE4.2)
x86/x86_64	default-features = false	Baseline SSE2 only (slow)
ARM	Default	NEON is baseline; already optimized

On x86, the default build automatically detects and uses the best available SIMD instructions at runtime:

cargo add hamming-bitwise-fast

For best single-call performance on x86, enable LTO so the compiler can auto-vectorize across the crate boundary:

[profile.release]
lto = true

For maximum performance, also compile with -C target-cpu=native (eliminates runtime dispatch overhead, at the cost of portability).

On ARM (including Apple Silicon), the default build is already fast.

Feature Flags

• multiversion_x86 (enabled by default): Enables runtime CPU detection for optimal SIMD on x86 via the multiversion crate. Disable with default-features = false if you need zero dependencies or are targeting a known CPU with -C target-cpu=native.

Modules

array

Fixed-size array APIs for bitwise Hamming distance.

slice

Variable-length slice APIs for bitwise Hamming distance.

Functions

hamming_bitwise_fast

Convenience alias for slice::distance that matches the crate name.