Crate shrew_cpu 

shrew-cpu

CPU backend implementation for Shrew.

This crate implements the Backend trait for CPU execution. Uses gemm for SIMD-accelerated matrix multiplication (AVX2/AVX-512/FMA) and rayon for parallel batched matmul and large elementwise ops.

Modules

ops

Structs

CpuBackend

CPU backend. Implements Backend by running operations on CPU via iterators.

CpuDevice

The CPU device. Since every machine has exactly one CPU (from our perspective), this is a zero-sized type.

Enums

CpuStorage

Storage of tensor data in CPU memory.

Type Aliases

CpuTensor

Convenience type alias for CPU tensors.