6 posts
OxiBLAS 0.2.2 is a production-readiness hardening release: a 16-domain multi-agent audit plus adversarial re-verification closes numerical correctness bugs across BLAS/LAPACK/sparse, hardens PackedRef/BandedRef construction to unsafe fn, fixes an integer-overflow bug in the Matrix Market parser, turns 57 of 64 previously-`ignore`d doctests into real executed tests, and adds a libFuzzer harness for both untrusted-input parsers. No breaking API changes outside two now-`unsafe` constructors. Pure Rust BLAS/LAPACK for SciRS2.
Production-grade BLAS and LAPACK entirely in Rust. Up to 172% of OpenBLAS performance on Apple M3, full sparse solvers, f128 precision, RuntimeAutoTuner, no_std support. The sovereign mathematical foundation for SciRS2 and the entire COOLJAPAN scientific computing stack.
OxiBLAS 0.2.0 is a major step up: cache-oblivious recursive and parallel factorizations, batched BLAS, runtime auto-tuning, multifrontal sparse solvers, mixed-precision refinement, NUMA-aware allocation, and no_std support — with the Fortran FFI retired in favor of a fully pure Rust workspace.
OxiBLAS 0.1.2 brings complex-aware LAPACK to ndarray: SVD and QR for complex matrices, Cholesky and eigendecomposition for Hermitian matrices, plus relaxed trait bounds so solve() works with Complex32/Complex64. Pure Rust BLAS/LAPACK, no C or Fortran.
OxiBLAS 0.1.1 fixes a critical sign bug in the symmetric eigenvalue QR algorithm so eigenvectors accumulate correctly for matrices that need multiple QR sweeps. The pure Rust BLAS/LAPACK foundation now powering the just-launched SciRS2 core.
The first public release of OxiBLAS — a pure Rust implementation of BLAS and LAPACK. Full Level 1/2/3 BLAS, LU/Cholesky/QR/SVD/EVD, 9 sparse formats, f16/f128 precision, and DGEMM already matching OpenBLAS on large matrices. No C, no Fortran, no MKL.