Struct SIMDOptimizer 

pub struct SIMDOptimizer { /* private fields */ }

SIMD operations for optimizer kernels

Implementations

impl SIMDOptimizer

pub fn new(config: SIMDConfig) -> Self

Create a new SIMD optimizer with configuration

pub fn detect_capabilities() -> SIMDConfig

Detect available SIMD instruction sets

pub unsafe fn adam_update_avx2( &self, params: &mut [f32], gradients: &[f32], momentum: &mut [f32], velocity: &mut [f32], lr: f32, beta1: f32, beta2: f32, eps: f32, step: i32, ) -> Result<()>

SIMD-optimized Adam update with AVX2

Safety

Caller must ensure the CPU supports AVX2 instructions. This is guaranteed when the function is invoked via the #[target_feature(enable = "avx2")] dispatch path only on hardware where the feature is available.

pub unsafe fn adamw_update_avx2( &self, params: &mut [f32], gradients: &[f32], momentum: &mut [f32], velocity: &mut [f32], lr: f32, beta1: f32, beta2: f32, eps: f32, weight_decay: f32, step: i32, ) -> Result<()>

SIMD-optimized AdamW update with AVX2 (decoupled weight decay)

Safety

Caller must ensure the CPU supports AVX2 instructions. This is guaranteed when the function is invoked via the #[target_feature(enable = "avx2")] dispatch path only on hardware where the feature is available.

pub unsafe fn sgd_momentum_update_avx2( &self, params: &mut [f32], gradients: &[f32], momentum: &mut [f32], lr: f32, momentum_factor: f32, weight_decay: f32, nesterov: bool, ) -> Result<()>

SIMD-optimized SGD with momentum update

Safety

Caller must ensure the CPU supports AVX2 instructions. This is guaranteed when the function is invoked via the #[target_feature(enable = "avx2")] dispatch path only on hardware where the feature is available.

pub unsafe fn clip_gradients_avx2( &self, gradients: &mut [f32], max_norm: f32, ) -> Result<f32>

SIMD-optimized gradient clipping

Safety

Caller must ensure the CPU supports AVX2 instructions. This is guaranteed when the function is invoked via the #[target_feature(enable = "avx2")] dispatch path only on hardware where the feature is available.

pub unsafe fn vector_add_avx2( &self, a: &mut [f32], b: &[f32], scale: f32, ) -> Result<()>

SIMD-optimized vector addition (for gradient accumulation)

Safety

Caller must ensure the CPU supports AVX2 instructions. This is guaranteed when the function is invoked via the #[target_feature(enable = "avx2")] dispatch path only on hardware where the feature is available.

pub unsafe fn dot_product_avx2(&self, a: &[f32], b: &[f32]) -> Result<f32>

SIMD-optimized dot product

Safety

Caller must ensure the CPU supports AVX2 instructions. This is guaranteed when the function is invoked via the #[target_feature(enable = "avx2")] dispatch path only on hardware where the feature is available.

pub fn adam_update_fallback( &self, params: &mut [f32], gradients: &[f32], momentum: &mut [f32], velocity: &mut [f32], lr: f32, beta1: f32, beta2: f32, eps: f32, step: i32, ) -> Result<()>

Fallback implementations for non-x86 architectures

pub fn adam_update( &self, params: &mut [f32], gradients: &[f32], momentum: &mut [f32], velocity: &mut [f32], lr: f32, beta1: f32, beta2: f32, eps: f32, step: i32, ) -> Result<()>

Auto-dispatch to best available implementation

pub fn adamw_update( &self, params: &mut [f32], gradients: &[f32], momentum: &mut [f32], velocity: &mut [f32], lr: f32, beta1: f32, beta2: f32, eps: f32, weight_decay: f32, step: i32, ) -> Result<()>

Auto-dispatch AdamW

pub fn get_performance_info(&self) -> SIMDPerformanceInfo

Get performance statistics

Trait Implementations

impl Default for SIMDOptimizer

fn default() -> Self

Returns the “default value” for a type.