Module tensor 

Zero-copy, const-generic N-dimensional strided tensor view.

Design

TensorView<'a, T, const N: usize> is a rank-N view over a borrowed slice. Shape and strides are [usize; N] arrays resolved at compile time — the const generic N is erased after monomorphization, leaving no runtime overhead vs. a hand-written 2-D or 3-D struct.

Layout Markers

Two zero-sized layout markers tag contiguous storage assumptions:

• RowMajor — row-major (C-order) storage; strides[i] = ∏_{j>i} shape[j].
• ColMajor — column-major (Fortran-order) storage.

Zero-Copy Contract

• new(data, shape) — zero allocation; computes row-major strides from shape.
• with_strides(data, shape, strides) — zero allocation; caller supplies strides.
• row_view(i) — returns a TensorView<'_, T, {N-1}> sharing the same slice.
• reshape(new_shape) — returns a new view if the layout is contiguous; no copy.
• All get / iter_rows operations are also zero-copy.

Module Structure

Leaf module	Contents
layout	RowMajor, ColMajor ZSTs; sealed Layout trait
error	TensorError enum
view	TensorView core + rank_ops/simd_bridge leaves
cow	TensorCow enum + all impl blocks
helpers (priv)	row_major_strides, compute_offset

Re-exports

pub use cow::TensorCow;

pub use error::TensorError;

pub use layout::ColMajor;

pub use layout::Layout;

pub use layout::RowMajor;

pub use view::TensorView;

Modules

cow

Clone-on-Write tensor container backed by AlignedVec.

error

Error type for tensor construction and indexing.

layout

Layout marker ZSTs and the sealed Layout trait.

view

Zero-copy N-dimensional strided tensor view.