Struct SimdOptimizer 

pub struct SimdOptimizer;

SIMD-optimized tensor operations using stable Rust features

This is the main entry point for SIMD operations, providing a unified interface to all optimized functions while maintaining backward compatibility.

Implementations

impl SimdOptimizer

pub fn add_f32_unchecked(a: &[f32], b: &[f32], result: &mut [f32])

Fast inline element-wise addition without bounds checking (for hot paths)

pub fn add_f32_optimized(a: &[f32], b: &[f32], result: &mut [f32]) -> Result<()>

Element-wise addition with SciRS2-Core SIMD auto-vectorization (safe version)

pub fn add_f32_auto(a: &[f32], b: &[f32], result: &mut [f32]) -> Result<()>

Auto-selection element-wise addition (chooses optimal implementation based on size)

pub fn mul_f32_unchecked(a: &[f32], b: &[f32], result: &mut [f32])

Fast inline element-wise multiplication without bounds checking (for hot paths)

pub fn mul_f32_optimized(a: &[f32], b: &[f32], result: &mut [f32]) -> Result<()>

Element-wise multiplication with optimization hints for f32 tensors (safe version)

pub fn sub_f32_optimized(a: &[f32], b: &[f32], result: &mut [f32]) -> Result<()>

Element-wise subtraction with optimization hints for f32 tensors (safe version)

pub fn relu_f32_optimized(input: &[f32], output: &mut [f32]) -> Result<()>

Optimized ReLU activation function

pub fn sigmoid_f32_optimized(input: &[f32], output: &mut [f32]) -> Result<()>

Optimized sigmoid activation with fast approximation

pub fn matmul_f32_blocked( a: &[f32], b: &[f32], c: &mut [f32], m: usize, n: usize, k: usize, block_size: usize, ) -> Result<()>

Advanced cache-friendly matrix multiplication using blocking

pub fn dot_product_f32_optimized(a: &[f32], b: &[f32]) -> Result<f32>

Optimized dot product for f32 vectors using Kahan summation

pub fn sum_f32_unchecked(input: &[f32]) -> f32

Ultra-fast vectorized horizontal sum reduction

pub fn max_f32_unchecked(input: &[f32]) -> f32

Vectorized maximum element finding with SIMD optimization

pub fn detect_capabilities() -> SimdCapabilities

Platform-specific capabilities detection (stable approach)

pub fn reduce_sum_f32_optimized(input: &[f32]) -> Result<f32>

Advanced vectorized reduction operations with numerical stability

pub fn reduce_min_max_f32_optimized(input: &[f32]) -> Result<(f32, f32)>

Vectorized min/max operations with efficient branching

pub fn normalize_f32_optimized( input: &[f32], output: &mut [f32], eps: f32, ) -> Result<()>

Advanced normalization with streaming computation for memory efficiency