numba-lapack

UNSAFE, zero-overhead Numba intrinsics that expose the full BLAS/LAPACK C-APIs via SciPy’s __pyx_capi__. Call BLAS/LAPACK directly from @njit in nopython mode.

⚠️ Unsafe means unsafe: raw pointer semantics; you are responsible for valid pointers, shapes, and leading dimensions.

Highlights

• Auto-discovers scipy.linalg.cython_blas and cython_lapack symbols at import.
• Generates Numba @intrinsic wrappers with the exact ABI (no Python overhead).
• Accepts arrays, typed pointers, or by-ref scalars for pointer parameters.
• Ships type stubs so IDEs can see function names & arg docs.

Quick start

import numpy as np
from numba import njit
from numba_lapack import dgemm

@njit(cache=True)
def gemm_nn(A, B, C, alpha, beta):
    m, k = A.shape
    _, n = B.shape
    dgemm(np.uint8(ord('N')), np.uint8(ord('N')),
          m, n, k, alpha, A, m, B, k, beta, C, m)