Struct SimdVectorOps

pub struct SimdVectorOps;

SIMD-optimized vector operations for high-performance similarity calculations

Production optimizations:

• AVX-512 support for modern CPUs (8x performance improvement)
• FMA (Fused Multiply-Add) instructions for better precision and performance
• Cache-aware processing for large vector batches
• Memory prefetching for optimal cache utilization
• Aligned memory access patterns

Implementations

impl SimdVectorOps

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

Compute dot product using SIMD instructions

This provides 2-8x speedup over naive implementations by using AVX-512/AVX2/SSE instructions when available, with automatic CPU feature detection and optimal instruction selection.

Performance optimizations:

• AVX-512: 16 f32 operations per instruction (8x speedup)
• FMA instructions: Fused multiply-add for better precision
• Memory prefetching for cache optimization
• Unrolled loops for reduced overhead

pub fn squared_norm(a: &Array1<f32>) -> f32

Compute squared L2 norm using SIMD instructions

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

Compute cosine similarity using SIMD-optimized operations

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

Add two vectors element-wise using SIMD instructions

pub fn scale_vector(a: &Array1<f32>, scalar: f32) -> Array1<f32>

Scale a vector by a scalar using SIMD instructions