Module distance 

Distance metrics for vector similarity computation

Table of Contents

• Distance: Enum of supported distance metrics
• compute_distance: Core distance computation function
• SIMD optimizations: AVX2/SSE vectorized implementations

Performance

SIMD implementations provide 4-8x speedup on modern CPUs. Auto-vectorization is enabled for scalar fallbacks.

Enums

Distance

Supported distance metrics for similarity computation

Functions

cosine_distance_fast

Fast cosine distance

cosine_distance_normalized

Cosine distance for pre-normalized vectors (fast path) Assumes both vectors have unit length, skips norm calculation

dot_product

Raw dot product (not negated)

dot_product_fast

Fast dot product - uses AVX2 when available, falls back to unrolled scalar

euclidean_distance_squared

Squared Euclidean distance (avoids sqrt for faster comparison)

euclidean_squared_fast

Fast squared Euclidean distance