LAPACK

LAPACK performance and correctness probed via NumPy eigvalsh (dsyevd) — the LAPACK-heavy half of the NumPy benchmark.

Benchmark source: opensolvers/benchmarks/numpy (bench_blas.py). For DGEMM (A @ B) see the NumPy or BLAS verification sections.

Orange Pi RV2 (8 threads)

Patched RVV OpenBLAS via FlexiBLAS (SciPy-bundle module, easyconfigs#26444).

Kernel	Scalar	Patched RVV	Speedup
EIGH (eigvalsh, N=2048)	10.54 s	6.72 s	1.6×

Stock EESSI default RVV: eigvalsh returns NaN / fails to converge. See difftest.

Banana Pi F3 (cross-board)

Kernel	Scalar	Patched RVV	Speedup
EIGH N=2048	9.59 s	5.94 s	1.6×

Stock unpatched RVV: LinAlgError: Eigenvalues did not converge.

For the full NumPy benchmark (DGEMM + EIGH), see NumPy. For a heavier eigensolver probe, see ELPA.