Struct BatchSimdOps 

pub struct BatchSimdOps;

Batch SIMD operations for processing multiple vectors at once

Production optimizations:

• Cache-aware processing for large batches
• Memory-aligned operations for optimal SIMD performance
• Parallel processing with work-stealing for large datasets
• Branch-prediction friendly algorithms
• Memory bandwidth optimization

Implementations

impl BatchSimdOps

pub fn pairwise_cosine_similarities(vectors: &[Array1<f32>]) -> Vec<Vec<f32>>

Compute pairwise cosine similarities between all vectors in a batch

pub fn find_k_most_similar( query: &Array1<f32>, vectors: &[Array1<f32>], k: usize, ) -> Vec<(usize, f32)>

Find the k most similar vectors to a query vector

pub fn compute_centroid(vectors: &[Array1<f32>]) -> Array1<f32>

Compute centroid of a batch of vectors

pub fn fast_cosine_similarity_with_norms( a: &Array1<f32>, b: &Array1<f32>, norm_a: f32, norm_b: f32, ) -> f32

Fast cosine similarity with pre-computed norms (production optimization)

pub fn cache_optimized_batch_similarities( query: &Array1<f32>, vectors: &[Array1<f32>], batch_size: usize, ) -> Vec<f32>

Cache-optimized batch similarity calculation for large datasets

pub fn aligned_dot_product(a: &Array1<f32>, b: &Array1<f32>) -> f32

Memory-aligned vector operations for optimal SIMD performance

pub fn batch_normalize(vectors: &mut [Array1<f32>])

High-performance batch normalization

pub fn matrix_vector_multiply( matrix: &[Array1<f32>], vector: &Array1<f32>, ) -> Array1<f32>

SIMD-optimized matrix-vector multiplication for batch operations