Module kernels 

Generic CPU compute kernels, written over the element traits and plain slices + Layout. The impl FloatOps/IntOps/BoolOps for Cpu blocks dispatch the storage enums to these.

Modules

element

Per-element numeric traits. Generic CPU kernels are written over these, then the storage enums dispatch to the concrete f32/f64/i32/… instantiation.

elementwise

Generic elementwise kernels (binary, binary-scalar, unary, comparison, in-place accumulate). Ported from luma-core’s arith.rs, generalized over the CpuNum element trait and returning plain Vecs.

indexing

Indexing kernels: index_select, gather, index_add, scatter_add. Ported from luma-core storage.rs. All require contiguous inputs.

iter

Shared strided-iteration helpers used by the generic CPU kernels, plus the DimArray view for iterating a single (possibly strided) dimension.

matmul

Batched matmul. Ported from luma-core matmul.rs. f32/f64 go through the gemm crate; other element types use a naive triple loop.

nn

Fused nn kernels: softmax and rms_norm. Ported from luma-core nn.rs.

reduce

Generic reduction kernels. Ported from luma-core reduce.rs.

shape

Shape-movement kernels that materialize data: concatenation.