Module vector 

Vector operations and utilities for RAG system

This module provides high-performance vector operations essential for similarity search and embedding manipulation in the RAG system.

Core Operations

• Similarity Metrics: Cosine similarity for measuring vector similarity
• Distance Metrics: L2 (Euclidean) distance for spatial relationships
• Vector Algebra: Dot product, normalization, and comparison operations

Performance Characteristics

All operations in this module are optimized for hot-path performance:

• Functions are marked with inline hints for small vectors
• Efficient iterator usage for auto-vectorization
• Minimal allocations and cache-friendly access patterns
• SIMD acceleration with runtime CPU detection (3-4x speedup)

SIMD Acceleration

The module automatically uses SIMD instructions when available:

• x86_64: AVX2, SSE2, or scalar fallback
• ARM: NEON or scalar fallback
• Runtime detection ensures optimal performance on any CPU

Use the *_auto() functions for automatic SIMD/scalar selection:

use foxstash_core::vector::{cosine_similarity_auto, dot_product_auto};

let a = vec![1.0; 384];
let b = vec![2.0; 384];

// Automatically uses SIMD if available
let similarity = cosine_similarity_auto(&a, &b).unwrap();
let dot = dot_product_auto(&a, &b).unwrap();

Usage

use foxstash_core::vector::ops::{cosine_similarity, normalize};

let mut embedding = vec![1.0, 2.0, 3.0];
normalize(&mut embedding);

let query = vec![0.6, 0.8, 0.0];
let similarity = cosine_similarity(&embedding, &query).unwrap();

Re-exports

pub use ops::approx_equal;

pub use ops::cosine_similarity;

pub use ops::dot_product;

pub use ops::l2_distance;

pub use ops::normalize;

pub use simd::cosine_distance_prenorm;

pub use simd::cosine_similarity_simd;

pub use simd::dot_product_simd;

pub use simd::l2_distance_simd;

pub use simd::norm_simd;

Modules

ops

Vector operations for RAG system

product_quantize

Product Quantization (PQ) for extreme compression

quantize

Vector quantization for memory-efficient storage

simd

SIMD-accelerated vector operations

Functions

cosine_similarity_auto

Automatically selects between SIMD and scalar cosine similarity.

dot_product_auto

Automatically selects between SIMD and scalar dot product.

l2_distance_auto

Automatically selects between SIMD and scalar L2 distance.