Trueno Sparse: Sparse matrix formats and operations

Provides CSR, COO, and BSR sparse matrix formats with SIMD-accelerated SpMV (sparse matrix-vector multiply) and SpMM (sparse matrix-dense matrix multiply).

Design

• Provable contracts: Format invariants validated at construction time
• Backward error bounded: Numerical accuracy follows LAProof bounds
• SIMD dispatch: Scalar → AVX2 → AVX-512 runtime selection
• GPU ready: Formats are GPU-transfer-friendly (contiguous arrays)

Quick Start

use trueno_sparse::{CsrMatrix, CooMatrix, SparseOps};

// Build from COO (triplets)
let coo = CooMatrix::new(3, 3, vec![0, 1, 2], vec![0, 1, 2], vec![1.0_f32, 2.0, 3.0]).unwrap();
let csr = CsrMatrix::from_coo(&coo);

// SpMV: y = A * x
let x = vec![1.0_f32, 1.0, 1.0];
let mut y = vec![0.0_f32; 3];
csr.spmv(1.0, &x, 0.0, &mut y);
assert!((y[0] - 1.0).abs() < 1e-6);
assert!((y[1] - 2.0).abs() < 1e-6);
assert!((y[2] - 3.0).abs() < 1e-6);

References

• Merrill & Garland, "Merge-Based Parallel SpMV", PPoPP 2016
• LAProof (Princeton): formal backward error bounds for CSR SpMV