Module kernel 

Runtime-dispatched SIMD kernel abstraction.

SimdKernel is a trait that abstracts batch floating-point arithmetic. Concrete implementations cover the scalar fallback, AVX2, and NEON paths.

global_kernel selects the best available kernel once per process via OnceLock — subsequent calls are a single pointer load.

Structs

Avx2Kernel

AVX2/FMA-accelerated kernel.

ScalarKernel

Pure-scalar fallback kernel. Works on every platform.

Traits

SimdKernel

Batch arithmetic operations, abstracted over SIMD width and ISA.

Functions

global_kernel

Returns the best available SimdKernel for this process.