Module kernel 

Defines runtime-executable compute kernels.

A kernel is a pure compute primitive that reads one or more input Tensors and writes results into a pre-allocated output Tensor. Kernels do not allocate output storage and are expected to validate basic argument invariants (arity, shape, rank) before any compute.

To allocate Tensors, see the crate::graph::Graph API.

Structs

AddKernel

Elementwise addition kernel.

MatMulKernel

Naive 2-D matrix multiplication kernel.

ReluKernel

Rectified Linear Unit (ReLU) activation kernel.

Enums

KernelError

Kernel-level errors raised during argument validation or execution.

Traits

Kernel

A runtime-executable compute primitive.