neco-sparse

日本語

Lightweight COO and CSR sparse matrix types for assembling sparse operators and iterating rows efficiently in caller-side sparse linear algebra code.

Matrix formats

neco-sparse focuses on two storage formats:

• CooMat<T> for assembly, where entries are pushed as (row, col, value) triplets
• CsrMat<T> for computation, where rows are contiguous and row-wise traversal runs in O(nnz)

The common workflow is to accumulate entries in COO, allow duplicates during assembly, and convert once to CSR for iteration, diagonal extraction, submatrix construction, or caller-side SpMV.

Usage

Build CSR directly

use neco_sparse::CsrMat;

let csr = CsrMat::try_from_csr_data(
    2,
    3,
    vec![0, 2, 3],
    vec![0, 2, 1],
    vec![1.0, 4.0, 3.0],
).unwrap();

assert_eq!(csr.nnz(), 3);
assert_eq!(csr.get(0, 2), Some(&4.0));

Assemble in COO, then convert

use neco_sparse::{CooMat, CsrMat};

let mut coo = CooMat::new(3, 3);
coo.push(0, 0, 2.0);
coo.push(1, 2, 5.0);
coo.push(0, 0, 1.0);

let csr = CsrMat::from(&coo);
assert_eq!(csr.get(0, 0), Some(&3.0));

Iterate rows and build auxiliary matrices

use neco_sparse::CsrMat;

let eye = CsrMat::identity(4);
let zeros = CsrMat::zeros(3, 5);

for row in eye.row_iter() {
    for (&col, &val) in row.col_indices().iter().zip(row.values()) {
        println!("col={col}, val={val}");
    }
}

assert_eq!(zeros.nnz(), 0);

Extract diagonals and submatrices

use neco_sparse::{CooMat, CsrMat};

let mut coo = CooMat::new(3, 3);
coo.push(0, 0, 4.0);
coo.push(0, 2, 1.0);
coo.push(1, 1, 5.0);
coo.push(2, 0, 2.0);
coo.push(2, 2, 6.0);

let csr = CsrMat::from(&coo);
let diag = csr.diagonal().unwrap();
let sub = csr.submatrix(&[2, 0], &[2, 0]).unwrap();

assert_eq!(diag, vec![4.0, 5.0, 6.0]);
assert_eq!(sub.row(0).values(), &[6.0, 2.0]);

API

License

MIT