Single-precision singular value decomposition via divide-and-conquer.

Computes $A = U \, \mathrm{diag}(S) \, V^T$. When full_matrices is false the thin SVD is returned: $U \in \mathbb{R}^{m \times k}$, $V^T \in \mathbb{R}^{k \times n}$, $k = \min(m, n)$. When true the full matrices ($U \in \mathbb{R}^{m \times m}$, $V^T \in \mathbb{R}^{n \times n}$) are returned.

Math

$A = U \, \Sigma \, V^T, \quad \Sigma = \mathrm{diag}(S)$

References

LAPACK sgesdd_ (divide-and-conquer driver).