Module reorder

Expand description

Sparse matrix reordering algorithms.

Reordering reduces bandwidth (RCM) or fill-in (AMD) of sparse matrices, improving the performance of direct and iterative solvers.

Reverse Cuthill-McKee (RCM): BFS from a peripheral node, reversed. Reduces matrix bandwidth, which improves cache locality for SpMV and reduces fill-in for banded preconditioners.
Approximate Minimum Degree (AMD): Greedy elimination choosing the node with minimum degree. Reduces fill-in for Cholesky/LU factorization.

let perm = rcm_ordering(matrix)?;
let reordered = permute_csr(matrix, &perm)?;

Functions§

amd_ordering: Compute the Approximate Minimum Degree (AMD) ordering of a sparse matrix.
amd_ordering_host: Host-side AMD ordering computation.
bandwidth: Compute the bandwidth of a matrix given its CSR structure.
inverse_permutation: Compute the inverse permutation.
permute_csr: Apply a permutation to a CSR matrix: P * A * P^T.
rcm_ordering: Compute the Reverse Cuthill-McKee (RCM) ordering of a sparse matrix.
rcm_ordering_host: Host-side RCM ordering computation.