Skip to main content Module faer_ndarray Copy item path Source FaerArrayView FaerCholeskyFactor FaerColView FaerLblt $LBL^\top$ decomposition FaerLdlt $L D L^\top$ decomposition FaerLlt $L L^\top$ decomposition FaerLinalgError FaerSymmetricFactor FaerCholesky FaerEigh FaerQr FaerSolve SolveCoreFaerSvd array1_to_col_matmut array2_to_matmut default_rrqr_rank_alpha factorize_symmetricwith_fallback Factorize a symmetric system with LLT -> LDLT -> LBLT fallback. fast_ab Compute A * B using faer’s SIMD-optimized GEMM.
For A of shape (n, p) and B of shape (p, q), this computes the (n, q) result.
Uses zero-copy views when possible. fast_ab_into Write faer matmul result A*B directly into a pre-allocated ndarray Array2.
Avoids the intermediate faer::Mat allocation and mat_to_array copy. fast_ata Compute A^T * A using faer’s SIMD-optimized GEMM.
This is MUCH faster than ndarray’s .t().dot() for matrices where n > ~100. fast_ata_into Compute A^T * A into a pre-allocated output buffer.
out must be shaped (p, p) where A is (n, p). fast_atb Compute A^T * B using faer’s SIMD-optimized GEMM.
For A of shape (n, p) and B of shape (n, q), this computes the (p, q) result.
Uses zero-copy views when possible. fast_atb_with_parallelism Compute A^T * B with an explicit faer parallelism policy for callers that
are already running independent products in an outer Rayon task. fast_atv Compute A^T * v using faer’s SIMD-optimized GEMV.
For A of shape (n, p) and v of shape (n,), this computes the (p,) result. fast_atv_into Compute A^T * v into a pre-allocated output buffer.
out must be length p where A is (n, p) and v is length n. fast_av Compute A * v using faer’s SIMD-optimized GEMV.
For A of shape (n, p) and v of shape (p,), this computes the (n,) result. fast_av_into Compute A * v into a pre-allocated output buffer.
out must be length n where A is (n, p) and v is length p. fast_av_view_into Compute A * v into a pre-allocated ArrayViewMut1 slice. Like
fast_av_into&mut Array1,
so callers can write directly into a sub-range of a larger buffer
without intermediate allocation. fast_joint_hessian_2x2 Compute the 2×2 block joint Hessian in a single streaming pass:
[X_a^T diag(w_aa) X_a, X_a^T diag(w_ab) X_b]
[X_b^T diag(w_ab) X_a, X_b^T diag(w_bb) X_b] fast_xt_diag_x Compute A^T * diag(W) * A using streaming chunks to avoid O(n*p) allocation. fast_xt_diag_x_with_parallelism Compute A^T * diag(W) * A with an explicit faer parallelism policy for
callers that parallelize multiple independent Hessian blocks externally. fast_xt_diag_y Compute A^T * diag(W) * B using streaming chunks. ldlt_rook Computes a symmetric-indefinite rook-pivoted LBL^T factorization. rrqr_nullspace_basis Compute an orthonormal basis for null(a^T) using column-pivoted QR on a.