Crate nabled_linalg 

Ndarray-native linear algebra domains for the nabled workspace.

nabled-linalg provides decomposition routines, dense/sparse kernels, vector/tensor primitives, and matrix-function algorithms over ndarray data.

Feature Flags

1. blas: enables BLAS acceleration through ndarray/blas.
2. openblas-system: enables provider-backed LAPACK paths via system OpenBLAS.
3. openblas-static: enables provider-backed LAPACK paths via statically linked OpenBLAS.
4. netlib-system: enables provider-backed LAPACK paths via system Netlib LAPACK.
5. netlib-static: enables provider-backed LAPACK paths via statically linked Netlib LAPACK.
6. magma-system: enables CUDA-backed MAGMA provider paths for supported decomposition domains.
7. accelerator-rayon: enables selected parallel CPU kernels.
8. accelerator-wgpu: enables bounded GPU (f32/f64) kernel paths.

Execution Model

1. Provider: decomposition implementation source (internal or selected LAPACK provider).
2. Backend: operation-kernel execution target (CpuBackend, GpuBackend).
3. Kernel: operation-family contract (for example matmat, sparse matvec, tensor contraction).

Provider selection and backend selection are orthogonal, compile-time concerns. Public APIs remain ndarray-native and backend/provider agnostic.

Example

use ndarray::arr2;
use nabled_linalg::svd;

let a = arr2(&[[1.0_f64, 2.0], [3.0, 4.0]]);
let decomposition = svd::decompose(&a)?;
assert_eq!(decomposition.singular_values.len(), 2);

Modules

accelerator

Compile-time backend contracts and accelerator kernels.

batched

Batched decomposition helpers over stacks of matrices.

cholesky

Cholesky decomposition over ndarray matrices.

eigen

Eigenvalue decompositions over ndarray matrices.

lu

LU decomposition over ndarray matrices.

matrix

Dense matrix pipeline primitives over ndarray arrays.

matrix_functions

Matrix functions over ndarray matrices.

orthogonalization

Orthogonalization routines over ndarray matrices.

polar

Polar decomposition over ndarray matrices.

qr

QR decomposition over ndarray matrices.

schur

Schur decomposition over ndarray matrices.

sparse

Sparse matrix primitives and iterative solves over CSR/CSC/COO structures.

svd

Singular value decomposition over ndarray matrices.

sylvester

Sylvester and Lyapunov solvers over ndarray matrices.

tensor

Tensor and cube primitives over ndarray higher-rank arrays.

triangular

Ndarray-native triangular solve kernels.

vector

Vector-first primitives for embedding-style workloads.