some new features

.venv/lib/python3.12/site-packages/scipy/sparse/linalg/_dsolve/__init__.py

@@ -0,0 +1,71 @@
+"""
+Linear Solvers
+==============
+
+The default solver is SuperLU (included in the scipy distribution),
+which can solve real or complex linear systems in both single and
+double precisions.  It is automatically replaced by UMFPACK, if
+available.  Note that UMFPACK works in double precision only, so
+switch it off by::
+
+    >>> from scipy.sparse.linalg import spsolve, use_solver
+    >>> use_solver(useUmfpack=False)
+
+to solve in the single precision. See also use_solver documentation.
+
+Example session::
+
+    >>> from scipy.sparse import csc_matrix, spdiags
+    >>> from numpy import array
+    >>>
+    >>> print("Inverting a sparse linear system:")
+    >>> print("The sparse matrix (constructed from diagonals):")
+    >>> a = spdiags([[1, 2, 3, 4, 5], [6, 5, 8, 9, 10]], [0, 1], 5, 5)
+    >>> b = array([1, 2, 3, 4, 5])
+    >>> print("Solve: single precision complex:")
+    >>> use_solver( useUmfpack = False )
+    >>> a = a.astype('F')
+    >>> x = spsolve(a, b)
+    >>> print(x)
+    >>> print("Error: ", a@x-b)
+    >>>
+    >>> print("Solve: double precision complex:")
+    >>> use_solver( useUmfpack = True )
+    >>> a = a.astype('D')
+    >>> x = spsolve(a, b)
+    >>> print(x)
+    >>> print("Error: ", a@x-b)
+    >>>
+    >>> print("Solve: double precision:")
+    >>> a = a.astype('d')
+    >>> x = spsolve(a, b)
+    >>> print(x)
+    >>> print("Error: ", a@x-b)
+    >>>
+    >>> print("Solve: single precision:")
+    >>> use_solver( useUmfpack = False )
+    >>> a = a.astype('f')
+    >>> x = spsolve(a, b.astype('f'))
+    >>> print(x)
+    >>> print("Error: ", a@x-b)
+
+"""
+
+#import umfpack
+#__doc__ = '\n\n'.join( (__doc__,  umfpack.__doc__) )
+#del umfpack
+
+from .linsolve import *
+from ._superlu import SuperLU
+from . import _add_newdocs
+from . import linsolve
+
+__all__ = [
+    'MatrixRankWarning', 'SuperLU', 'factorized',
+    'spilu', 'splu', 'spsolve',
+    'spsolve_triangular', 'use_solver'
+]
+
+from scipy._lib._testutils import PytestTester
+test = PytestTester(__name__)
+del PytestTester

.venv/lib/python3.12/site-packages/scipy/sparse/linalg/_dsolve/_add_newdocs.py

@@ -0,0 +1,153 @@
+from numpy.lib import add_newdoc
+
+add_newdoc('scipy.sparse.linalg._dsolve._superlu', 'SuperLU',
+    """
+    LU factorization of a sparse matrix.
+
+    Factorization is represented as::
+
+        Pr @ A @ Pc = L @ U
+
+    To construct these `SuperLU` objects, call the `splu` and `spilu`
+    functions.
+
+    Attributes
+    ----------
+    shape
+    nnz
+    perm_c
+    perm_r
+    L
+    U
+
+    Methods
+    -------
+    solve
+
+    Notes
+    -----
+
+    .. versionadded:: 0.14.0
+
+    Examples
+    --------
+    The LU decomposition can be used to solve matrix equations. Consider:
+
+    >>> import numpy as np
+    >>> from scipy.sparse import csc_matrix
+    >>> from scipy.sparse.linalg import splu
+    >>> A = csc_matrix([[1,2,0,4], [1,0,0,1], [1,0,2,1], [2,2,1,0.]])
+
+    This can be solved for a given right-hand side:
+
+    >>> lu = splu(A)
+    >>> b = np.array([1, 2, 3, 4])
+    >>> x = lu.solve(b)
+    >>> A.dot(x)
+    array([ 1.,  2.,  3.,  4.])
+
+    The ``lu`` object also contains an explicit representation of the
+    decomposition. The permutations are represented as mappings of
+    indices:
+
+    >>> lu.perm_r
+    array([2, 1, 3, 0], dtype=int32)  # may vary
+    >>> lu.perm_c
+    array([0, 1, 3, 2], dtype=int32)  # may vary
+
+    The L and U factors are sparse matrices in CSC format:
+
+    >>> lu.L.toarray()
+    array([[ 1. ,  0. ,  0. ,  0. ],  # may vary
+           [ 0.5,  1. ,  0. ,  0. ],
+           [ 0.5, -1. ,  1. ,  0. ],
+           [ 0.5,  1. ,  0. ,  1. ]])
+    >>> lu.U.toarray()
+    array([[ 2. ,  2. ,  0. ,  1. ],  # may vary
+           [ 0. , -1. ,  1. , -0.5],
+           [ 0. ,  0. ,  5. , -1. ],
+           [ 0. ,  0. ,  0. ,  2. ]])
+
+    The permutation matrices can be constructed:
+
+    >>> Pr = csc_matrix((np.ones(4), (lu.perm_r, np.arange(4))))
+    >>> Pc = csc_matrix((np.ones(4), (np.arange(4), lu.perm_c)))
+
+    We can reassemble the original matrix:
+
+    >>> (Pr.T @ (lu.L @ lu.U) @ Pc.T).toarray()
+    array([[ 1.,  2.,  0.,  4.],
+           [ 1.,  0.,  0.,  1.],
+           [ 1.,  0.,  2.,  1.],
+           [ 2.,  2.,  1.,  0.]])
+    """)
+
+add_newdoc('scipy.sparse.linalg._dsolve._superlu', 'SuperLU', ('solve',
+    """
+    solve(rhs[, trans])
+
+    Solves linear system of equations with one or several right-hand sides.
+
+    Parameters
+    ----------
+    rhs : ndarray, shape (n,) or (n, k)
+        Right hand side(s) of equation
+    trans : {'N', 'T', 'H'}, optional
+        Type of system to solve::
+
+            'N':   A   @ x == rhs  (default)
+            'T':   A^T @ x == rhs
+            'H':   A^H @ x == rhs
+
+        i.e., normal, transposed, and hermitian conjugate.
+
+    Returns
+    -------
+    x : ndarray, shape ``rhs.shape``
+        Solution vector(s)
+    """))
+
+add_newdoc('scipy.sparse.linalg._dsolve._superlu', 'SuperLU', ('L',
+    """
+    Lower triangular factor with unit diagonal as a
+    `scipy.sparse.csc_matrix`.
+
+    .. versionadded:: 0.14.0
+    """))
+
+add_newdoc('scipy.sparse.linalg._dsolve._superlu', 'SuperLU', ('U',
+    """
+    Upper triangular factor as a `scipy.sparse.csc_matrix`.
+
+    .. versionadded:: 0.14.0
+    """))
+
+add_newdoc('scipy.sparse.linalg._dsolve._superlu', 'SuperLU', ('shape',
+    """
+    Shape of the original matrix as a tuple of ints.
+    """))
+
+add_newdoc('scipy.sparse.linalg._dsolve._superlu', 'SuperLU', ('nnz',
+    """
+    Number of nonzero elements in the matrix.
+    """))
+
+add_newdoc('scipy.sparse.linalg._dsolve._superlu', 'SuperLU', ('perm_c',
+    """
+    Permutation Pc represented as an array of indices.
+
+    The column permutation matrix can be reconstructed via:
+
+    >>> Pc = np.zeros((n, n))
+    >>> Pc[np.arange(n), perm_c] = 1
+    """))
+
+add_newdoc('scipy.sparse.linalg._dsolve._superlu', 'SuperLU', ('perm_r',
+    """
+    Permutation Pr represented as an array of indices.
+
+    The row permutation matrix can be reconstructed via:
+
+    >>> Pr = np.zeros((n, n))
+    >>> Pr[perm_r, np.arange(n)] = 1
+    """))

.venv/lib/python3.12/site-packages/scipy/sparse/linalg/_dsolve/linsolve.py

@@ -0,0 +1,742 @@
+from warnings import warn, catch_warnings, simplefilter
+
+import numpy as np
+from numpy import asarray
+from scipy.sparse import (issparse,
+                          SparseEfficiencyWarning, csc_matrix, eye, diags)
+from scipy.sparse._sputils import is_pydata_spmatrix, convert_pydata_sparse_to_scipy
+from scipy.linalg import LinAlgError
+import copy
+
+from . import _superlu
+
+noScikit = False
+try:
+    import scikits.umfpack as umfpack
+except ImportError:
+    noScikit = True
+
+useUmfpack = not noScikit
+
+__all__ = ['use_solver', 'spsolve', 'splu', 'spilu', 'factorized',
+           'MatrixRankWarning', 'spsolve_triangular']
+
+
+class MatrixRankWarning(UserWarning):
+    pass
+
+
+def use_solver(**kwargs):
+    """
+    Select default sparse direct solver to be used.
+
+    Parameters
+    ----------
+    useUmfpack : bool, optional
+        Use UMFPACK [1]_, [2]_, [3]_, [4]_. over SuperLU. Has effect only
+        if ``scikits.umfpack`` is installed. Default: True
+    assumeSortedIndices : bool, optional
+        Allow UMFPACK to skip the step of sorting indices for a CSR/CSC matrix.
+        Has effect only if useUmfpack is True and ``scikits.umfpack`` is
+        installed. Default: False
+
+    Notes
+    -----
+    The default sparse solver is UMFPACK when available
+    (``scikits.umfpack`` is installed). This can be changed by passing
+    useUmfpack = False, which then causes the always present SuperLU
+    based solver to be used.
+
+    UMFPACK requires a CSR/CSC matrix to have sorted column/row indices. If
+    sure that the matrix fulfills this, pass ``assumeSortedIndices=True``
+    to gain some speed.
+
+    References
+    ----------
+    .. [1] T. A. Davis, Algorithm 832:  UMFPACK - an unsymmetric-pattern
+           multifrontal method with a column pre-ordering strategy, ACM
+           Trans. on Mathematical Software, 30(2), 2004, pp. 196--199.
+           https://dl.acm.org/doi/abs/10.1145/992200.992206
+
+    .. [2] T. A. Davis, A column pre-ordering strategy for the
+           unsymmetric-pattern multifrontal method, ACM Trans.
+           on Mathematical Software, 30(2), 2004, pp. 165--195.
+           https://dl.acm.org/doi/abs/10.1145/992200.992205
+
+    .. [3] T. A. Davis and I. S. Duff, A combined unifrontal/multifrontal
+           method for unsymmetric sparse matrices, ACM Trans. on
+           Mathematical Software, 25(1), 1999, pp. 1--19.
+           https://doi.org/10.1145/305658.287640
+
+    .. [4] T. A. Davis and I. S. Duff, An unsymmetric-pattern multifrontal
+           method for sparse LU factorization, SIAM J. Matrix Analysis and
+           Computations, 18(1), 1997, pp. 140--158.
+           https://doi.org/10.1137/S0895479894246905T.
+
+    Examples
+    --------
+    >>> import numpy as np
+    >>> from scipy.sparse.linalg import use_solver, spsolve
+    >>> from scipy.sparse import csc_matrix
+    >>> R = np.random.randn(5, 5)
+    >>> A = csc_matrix(R)
+    >>> b = np.random.randn(5)
+    >>> use_solver(useUmfpack=False) # enforce superLU over UMFPACK
+    >>> x = spsolve(A, b)
+    >>> np.allclose(A.dot(x), b)
+    True
+    >>> use_solver(useUmfpack=True) # reset umfPack usage to default
+    """
+    if 'useUmfpack' in kwargs:
+        globals()['useUmfpack'] = kwargs['useUmfpack']
+    if useUmfpack and 'assumeSortedIndices' in kwargs:
+        umfpack.configure(assumeSortedIndices=kwargs['assumeSortedIndices'])
+
+def _get_umf_family(A):
+    """Get umfpack family string given the sparse matrix dtype."""
+    _families = {
+        (np.float64, np.int32): 'di',
+        (np.complex128, np.int32): 'zi',
+        (np.float64, np.int64): 'dl',
+        (np.complex128, np.int64): 'zl'
+    }
+
+    # A.dtype.name can only be "float64" or
+    # "complex128" in control flow
+    f_type = getattr(np, A.dtype.name)
+    # control flow may allow for more index
+    # types to get through here
+    i_type = getattr(np, A.indices.dtype.name)
+
+    try:
+        family = _families[(f_type, i_type)]
+
+    except KeyError as e:
+        msg = ('only float64 or complex128 matrices with int32 or int64 '
+               f'indices are supported! (got: matrix: {f_type}, indices: {i_type})')
+        raise ValueError(msg) from e
+
+    # See gh-8278. Considered converting only if
+    # A.shape[0]*A.shape[1] > np.iinfo(np.int32).max,
+    # but that didn't always fix the issue.
+    family = family[0] + "l"
+    A_new = copy.copy(A)
+    A_new.indptr = np.asarray(A.indptr, dtype=np.int64)
+    A_new.indices = np.asarray(A.indices, dtype=np.int64)
+
+    return family, A_new
+
+def _safe_downcast_indices(A):
+    # check for safe downcasting
+    max_value = np.iinfo(np.intc).max
+
+    if A.indptr[-1] > max_value:  # indptr[-1] is max b/c indptr always sorted
+        raise ValueError("indptr values too large for SuperLU")
+
+    if max(*A.shape) > max_value:  # only check large enough arrays
+        if np.any(A.indices > max_value):
+            raise ValueError("indices values too large for SuperLU")
+
+    indices = A.indices.astype(np.intc, copy=False)
+    indptr = A.indptr.astype(np.intc, copy=False)
+    return indices, indptr
+
+def spsolve(A, b, permc_spec=None, use_umfpack=True):
+    """Solve the sparse linear system Ax=b, where b may be a vector or a matrix.
+
+    Parameters
+    ----------
+    A : ndarray or sparse matrix
+        The square matrix A will be converted into CSC or CSR form
+    b : ndarray or sparse matrix
+        The matrix or vector representing the right hand side of the equation.
+        If a vector, b.shape must be (n,) or (n, 1).
+    permc_spec : str, optional
+        How to permute the columns of the matrix for sparsity preservation.
+        (default: 'COLAMD')
+
+        - ``NATURAL``: natural ordering.
+        - ``MMD_ATA``: minimum degree ordering on the structure of A^T A.
+        - ``MMD_AT_PLUS_A``: minimum degree ordering on the structure of A^T+A.
+        - ``COLAMD``: approximate minimum degree column ordering [1]_, [2]_.
+
+    use_umfpack : bool, optional
+        if True (default) then use UMFPACK for the solution [3]_, [4]_, [5]_,
+        [6]_ . This is only referenced if b is a vector and
+        ``scikits.umfpack`` is installed.
+
+    Returns
+    -------
+    x : ndarray or sparse matrix
+        the solution of the sparse linear equation.
+        If b is a vector, then x is a vector of size A.shape[1]
+        If b is a matrix, then x is a matrix of size (A.shape[1], b.shape[1])
+
+    Notes
+    -----
+    For solving the matrix expression AX = B, this solver assumes the resulting
+    matrix X is sparse, as is often the case for very sparse inputs.  If the
+    resulting X is dense, the construction of this sparse result will be
+    relatively expensive.  In that case, consider converting A to a dense
+    matrix and using scipy.linalg.solve or its variants.
+
+    References
+    ----------
+    .. [1] T. A. Davis, J. R. Gilbert, S. Larimore, E. Ng, Algorithm 836:
+           COLAMD, an approximate column minimum degree ordering algorithm,
+           ACM Trans. on Mathematical Software, 30(3), 2004, pp. 377--380.
+           :doi:`10.1145/1024074.1024080`
+
+    .. [2] T. A. Davis, J. R. Gilbert, S. Larimore, E. Ng, A column approximate
+           minimum degree ordering algorithm, ACM Trans. on Mathematical
+           Software, 30(3), 2004, pp. 353--376. :doi:`10.1145/1024074.1024079`
+
+    .. [3] T. A. Davis, Algorithm 832:  UMFPACK - an unsymmetric-pattern
+           multifrontal method with a column pre-ordering strategy, ACM
+           Trans. on Mathematical Software, 30(2), 2004, pp. 196--199.
+           https://dl.acm.org/doi/abs/10.1145/992200.992206
+
+    .. [4] T. A. Davis, A column pre-ordering strategy for the
+           unsymmetric-pattern multifrontal method, ACM Trans.
+           on Mathematical Software, 30(2), 2004, pp. 165--195.
+           https://dl.acm.org/doi/abs/10.1145/992200.992205
+
+    .. [5] T. A. Davis and I. S. Duff, A combined unifrontal/multifrontal
+           method for unsymmetric sparse matrices, ACM Trans. on
+           Mathematical Software, 25(1), 1999, pp. 1--19.
+           https://doi.org/10.1145/305658.287640
+
+    .. [6] T. A. Davis and I. S. Duff, An unsymmetric-pattern multifrontal
+           method for sparse LU factorization, SIAM J. Matrix Analysis and
+           Computations, 18(1), 1997, pp. 140--158.
+           https://doi.org/10.1137/S0895479894246905T.
+
+
+    Examples
+    --------
+    >>> import numpy as np
+    >>> from scipy.sparse import csc_matrix
+    >>> from scipy.sparse.linalg import spsolve
+    >>> A = csc_matrix([[3, 2, 0], [1, -1, 0], [0, 5, 1]], dtype=float)
+    >>> B = csc_matrix([[2, 0], [-1, 0], [2, 0]], dtype=float)
+    >>> x = spsolve(A, B)
+    >>> np.allclose(A.dot(x).toarray(), B.toarray())
+    True
+    """
+    is_pydata_sparse = is_pydata_spmatrix(b)
+    pydata_sparse_cls = b.__class__ if is_pydata_sparse else None
+    A = convert_pydata_sparse_to_scipy(A)
+    b = convert_pydata_sparse_to_scipy(b)
+
+    if not (issparse(A) and A.format in ("csc", "csr")):
+        A = csc_matrix(A)
+        warn('spsolve requires A be CSC or CSR matrix format',
+             SparseEfficiencyWarning, stacklevel=2)
+
+    # b is a vector only if b have shape (n,) or (n, 1)
+    b_is_sparse = issparse(b)
+    if not b_is_sparse:
+        b = asarray(b)
+    b_is_vector = ((b.ndim == 1) or (b.ndim == 2 and b.shape[1] == 1))
+
+    # sum duplicates for non-canonical format
+    A.sum_duplicates()
+    A = A._asfptype()  # upcast to a floating point format
+    result_dtype = np.promote_types(A.dtype, b.dtype)
+    if A.dtype != result_dtype:
+        A = A.astype(result_dtype)
+    if b.dtype != result_dtype:
+        b = b.astype(result_dtype)
+
+    # validate input shapes
+    M, N = A.shape
+    if (M != N):
+        raise ValueError(f"matrix must be square (has shape {(M, N)})")
+
+    if M != b.shape[0]:
+        raise ValueError(f"matrix - rhs dimension mismatch ({A.shape} - {b.shape[0]})")
+
+    use_umfpack = use_umfpack and useUmfpack
+
+    if b_is_vector and use_umfpack:
+        if b_is_sparse:
+            b_vec = b.toarray()
+        else:
+            b_vec = b
+        b_vec = asarray(b_vec, dtype=A.dtype).ravel()
+
+        if noScikit:
+            raise RuntimeError('Scikits.umfpack not installed.')
+
+        if A.dtype.char not in 'dD':
+            raise ValueError("convert matrix data to double, please, using"
+                  " .astype(), or set linsolve.useUmfpack = False")
+
+        umf_family, A = _get_umf_family(A)
+        umf = umfpack.UmfpackContext(umf_family)
+        x = umf.linsolve(umfpack.UMFPACK_A, A, b_vec,
+                         autoTranspose=True)
+    else:
+        if b_is_vector and b_is_sparse:
+            b = b.toarray()
+            b_is_sparse = False
+
+        if not b_is_sparse:
+            if A.format == "csc":
+                flag = 1  # CSC format
+            else:
+                flag = 0  # CSR format
+
+            indices = A.indices.astype(np.intc, copy=False)
+            indptr = A.indptr.astype(np.intc, copy=False)
+            options = dict(ColPerm=permc_spec)
+            x, info = _superlu.gssv(N, A.nnz, A.data, indices, indptr,
+                                    b, flag, options=options)
+            if info != 0:
+                warn("Matrix is exactly singular", MatrixRankWarning, stacklevel=2)
+                x.fill(np.nan)
+            if b_is_vector:
+                x = x.ravel()
+        else:
+            # b is sparse
+            Afactsolve = factorized(A)
+
+            if not (b.format == "csc" or is_pydata_spmatrix(b)):
+                warn('spsolve is more efficient when sparse b '
+                     'is in the CSC matrix format',
+                     SparseEfficiencyWarning, stacklevel=2)
+                b = csc_matrix(b)
+
+            # Create a sparse output matrix by repeatedly applying
+            # the sparse factorization to solve columns of b.
+            data_segs = []
+            row_segs = []
+            col_segs = []
+            for j in range(b.shape[1]):
+                # TODO: replace this with
+                # bj = b[:, j].toarray().ravel()
+                # once 1D sparse arrays are supported.
+                # That is a slightly faster code path.
+                bj = b[:, [j]].toarray().ravel()
+                xj = Afactsolve(bj)
+                w = np.flatnonzero(xj)
+                segment_length = w.shape[0]
+                row_segs.append(w)
+                col_segs.append(np.full(segment_length, j, dtype=int))
+                data_segs.append(np.asarray(xj[w], dtype=A.dtype))
+            sparse_data = np.concatenate(data_segs)
+            sparse_row = np.concatenate(row_segs)
+            sparse_col = np.concatenate(col_segs)
+            x = A.__class__((sparse_data, (sparse_row, sparse_col)),
+                           shape=b.shape, dtype=A.dtype)
+
+            if is_pydata_sparse:
+                x = pydata_sparse_cls.from_scipy_sparse(x)
+
+    return x
+
+
+def splu(A, permc_spec=None, diag_pivot_thresh=None,
+         relax=None, panel_size=None, options=dict()):
+    """
+    Compute the LU decomposition of a sparse, square matrix.
+
+    Parameters
+    ----------
+    A : sparse matrix
+        Sparse matrix to factorize. Most efficient when provided in CSC
+        format. Other formats will be converted to CSC before factorization.
+    permc_spec : str, optional
+        How to permute the columns of the matrix for sparsity preservation.
+        (default: 'COLAMD')
+
+        - ``NATURAL``: natural ordering.
+        - ``MMD_ATA``: minimum degree ordering on the structure of A^T A.
+        - ``MMD_AT_PLUS_A``: minimum degree ordering on the structure of A^T+A.
+        - ``COLAMD``: approximate minimum degree column ordering
+
+    diag_pivot_thresh : float, optional
+        Threshold used for a diagonal entry to be an acceptable pivot.
+        See SuperLU user's guide for details [1]_
+    relax : int, optional
+        Expert option for customizing the degree of relaxing supernodes.
+        See SuperLU user's guide for details [1]_
+    panel_size : int, optional
+        Expert option for customizing the panel size.
+        See SuperLU user's guide for details [1]_
+    options : dict, optional
+        Dictionary containing additional expert options to SuperLU.
+        See SuperLU user guide [1]_ (section 2.4 on the 'Options' argument)
+        for more details. For example, you can specify
+        ``options=dict(Equil=False, IterRefine='SINGLE'))``
+        to turn equilibration off and perform a single iterative refinement.
+
+    Returns
+    -------
+    invA : scipy.sparse.linalg.SuperLU
+        Object, which has a ``solve`` method.
+
+    See also
+    --------
+    spilu : incomplete LU decomposition
+
+    Notes
+    -----
+    This function uses the SuperLU library.
+
+    References
+    ----------
+    .. [1] SuperLU https://portal.nersc.gov/project/sparse/superlu/
+
+    Examples
+    --------
+    >>> import numpy as np
+    >>> from scipy.sparse import csc_matrix
+    >>> from scipy.sparse.linalg import splu
+    >>> A = csc_matrix([[1., 0., 0.], [5., 0., 2.], [0., -1., 0.]], dtype=float)
+    >>> B = splu(A)
+    >>> x = np.array([1., 2., 3.], dtype=float)
+    >>> B.solve(x)
+    array([ 1. , -3. , -1.5])
+    >>> A.dot(B.solve(x))
+    array([ 1.,  2.,  3.])
+    >>> B.solve(A.dot(x))
+    array([ 1.,  2.,  3.])
+    """
+
+    if is_pydata_spmatrix(A):
+        def csc_construct_func(*a, cls=type(A)):
+            return cls.from_scipy_sparse(csc_matrix(*a))
+        A = A.to_scipy_sparse().tocsc()
+    else:
+        csc_construct_func = csc_matrix
+
+    if not (issparse(A) and A.format == "csc"):
+        A = csc_matrix(A)
+        warn('splu converted its input to CSC format',
+             SparseEfficiencyWarning, stacklevel=2)
+
+    # sum duplicates for non-canonical format
+    A.sum_duplicates()
+    A = A._asfptype()  # upcast to a floating point format
+
+    M, N = A.shape
+    if (M != N):
+        raise ValueError("can only factor square matrices")  # is this true?
+
+    indices, indptr = _safe_downcast_indices(A)
+
+    _options = dict(DiagPivotThresh=diag_pivot_thresh, ColPerm=permc_spec,
+                    PanelSize=panel_size, Relax=relax)
+    if options is not None:
+        _options.update(options)
+
+    # Ensure that no column permutations are applied
+    if (_options["ColPerm"] == "NATURAL"):
+        _options["SymmetricMode"] = True
+
+    return _superlu.gstrf(N, A.nnz, A.data, indices, indptr,
+                          csc_construct_func=csc_construct_func,
+                          ilu=False, options=_options)
+
+
+def spilu(A, drop_tol=None, fill_factor=None, drop_rule=None, permc_spec=None,
+          diag_pivot_thresh=None, relax=None, panel_size=None, options=None):
+    """
+    Compute an incomplete LU decomposition for a sparse, square matrix.
+
+    The resulting object is an approximation to the inverse of `A`.
+
+    Parameters
+    ----------
+    A : (N, N) array_like
+        Sparse matrix to factorize. Most efficient when provided in CSC format.
+        Other formats will be converted to CSC before factorization.
+    drop_tol : float, optional
+        Drop tolerance (0 <= tol <= 1) for an incomplete LU decomposition.
+        (default: 1e-4)
+    fill_factor : float, optional
+        Specifies the fill ratio upper bound (>= 1.0) for ILU. (default: 10)
+    drop_rule : str, optional
+        Comma-separated string of drop rules to use.
+        Available rules: ``basic``, ``prows``, ``column``, ``area``,
+        ``secondary``, ``dynamic``, ``interp``. (Default: ``basic,area``)
+
+        See SuperLU documentation for details.
+
+    Remaining other options
+        Same as for `splu`
+
+    Returns
+    -------
+    invA_approx : scipy.sparse.linalg.SuperLU
+        Object, which has a ``solve`` method.
+
+    See also
+    --------
+    splu : complete LU decomposition
+
+    Notes
+    -----
+    To improve the better approximation to the inverse, you may need to
+    increase `fill_factor` AND decrease `drop_tol`.
+
+    This function uses the SuperLU library.
+
+    Examples
+    --------
+    >>> import numpy as np
+    >>> from scipy.sparse import csc_matrix
+    >>> from scipy.sparse.linalg import spilu
+    >>> A = csc_matrix([[1., 0., 0.], [5., 0., 2.], [0., -1., 0.]], dtype=float)
+    >>> B = spilu(A)
+    >>> x = np.array([1., 2., 3.], dtype=float)
+    >>> B.solve(x)
+    array([ 1. , -3. , -1.5])
+    >>> A.dot(B.solve(x))
+    array([ 1.,  2.,  3.])
+    >>> B.solve(A.dot(x))
+    array([ 1.,  2.,  3.])
+    """
+
+    if is_pydata_spmatrix(A):
+        def csc_construct_func(*a, cls=type(A)):
+            return cls.from_scipy_sparse(csc_matrix(*a))
+        A = A.to_scipy_sparse().tocsc()
+    else:
+        csc_construct_func = csc_matrix
+
+    if not (issparse(A) and A.format == "csc"):
+        A = csc_matrix(A)
+        warn('spilu converted its input to CSC format',
+             SparseEfficiencyWarning, stacklevel=2)
+
+    # sum duplicates for non-canonical format
+    A.sum_duplicates()
+    A = A._asfptype()  # upcast to a floating point format
+
+    M, N = A.shape
+    if (M != N):
+        raise ValueError("can only factor square matrices")  # is this true?
+
+    indices, indptr = _safe_downcast_indices(A)
+
+    _options = dict(ILU_DropRule=drop_rule, ILU_DropTol=drop_tol,
+                    ILU_FillFactor=fill_factor,
+                    DiagPivotThresh=diag_pivot_thresh, ColPerm=permc_spec,
+                    PanelSize=panel_size, Relax=relax)
+    if options is not None:
+        _options.update(options)
+
+    # Ensure that no column permutations are applied
+    if (_options["ColPerm"] == "NATURAL"):
+        _options["SymmetricMode"] = True
+
+    return _superlu.gstrf(N, A.nnz, A.data, indices, indptr,
+                          csc_construct_func=csc_construct_func,
+                          ilu=True, options=_options)
+
+
+def factorized(A):
+    """
+    Return a function for solving a sparse linear system, with A pre-factorized.
+
+    Parameters
+    ----------
+    A : (N, N) array_like
+        Input. A in CSC format is most efficient. A CSR format matrix will
+        be converted to CSC before factorization.
+
+    Returns
+    -------
+    solve : callable
+        To solve the linear system of equations given in `A`, the `solve`
+        callable should be passed an ndarray of shape (N,).
+
+    Examples
+    --------
+    >>> import numpy as np
+    >>> from scipy.sparse.linalg import factorized
+    >>> from scipy.sparse import csc_matrix
+    >>> A = np.array([[ 3. ,  2. , -1. ],
+    ...               [ 2. , -2. ,  4. ],
+    ...               [-1. ,  0.5, -1. ]])
+    >>> solve = factorized(csc_matrix(A)) # Makes LU decomposition.
+    >>> rhs1 = np.array([1, -2, 0])
+    >>> solve(rhs1) # Uses the LU factors.
+    array([ 1., -2., -2.])
+
+    """
+    if is_pydata_spmatrix(A):
+        A = A.to_scipy_sparse().tocsc()
+
+    if useUmfpack:
+        if noScikit:
+            raise RuntimeError('Scikits.umfpack not installed.')
+
+        if not (issparse(A) and A.format == "csc"):
+            A = csc_matrix(A)
+            warn('splu converted its input to CSC format',
+                 SparseEfficiencyWarning, stacklevel=2)
+
+        A = A._asfptype()  # upcast to a floating point format
+
+        if A.dtype.char not in 'dD':
+            raise ValueError("convert matrix data to double, please, using"
+                  " .astype(), or set linsolve.useUmfpack = False")
+
+        umf_family, A = _get_umf_family(A)
+        umf = umfpack.UmfpackContext(umf_family)
+
+        # Make LU decomposition.
+        umf.numeric(A)
+
+        def solve(b):
+            with np.errstate(divide="ignore", invalid="ignore"):
+                # Ignoring warnings with numpy >= 1.23.0, see gh-16523
+                result = umf.solve(umfpack.UMFPACK_A, A, b, autoTranspose=True)
+
+            return result
+
+        return solve
+    else:
+        return splu(A).solve
+
+
+def spsolve_triangular(A, b, lower=True, overwrite_A=False, overwrite_b=False,
+                       unit_diagonal=False):
+    """
+    Solve the equation ``A x = b`` for `x`, assuming A is a triangular matrix.
+
+    Parameters
+    ----------
+    A : (M, M) sparse matrix
+        A sparse square triangular matrix. Should be in CSR or CSC format.
+    b : (M,) or (M, N) array_like
+        Right-hand side matrix in ``A x = b``
+    lower : bool, optional
+        Whether `A` is a lower or upper triangular matrix.
+        Default is lower triangular matrix.
+    overwrite_A : bool, optional
+        Allow changing `A`.
+        Enabling gives a performance gain. Default is False.
+    overwrite_b : bool, optional
+        Allow overwriting data in `b`.
+        Enabling gives a performance gain. Default is False.
+        If `overwrite_b` is True, it should be ensured that
+        `b` has an appropriate dtype to be able to store the result.
+    unit_diagonal : bool, optional
+        If True, diagonal elements of `a` are assumed to be 1.
+
+        .. versionadded:: 1.4.0
+
+    Returns
+    -------
+    x : (M,) or (M, N) ndarray
+        Solution to the system ``A x = b``. Shape of return matches shape
+        of `b`.
+
+    Raises
+    ------
+    LinAlgError
+        If `A` is singular or not triangular.
+    ValueError
+        If shape of `A` or shape of `b` do not match the requirements.
+
+    Notes
+    -----
+    .. versionadded:: 0.19.0
+
+    Examples
+    --------
+    >>> import numpy as np
+    >>> from scipy.sparse import csc_array
+    >>> from scipy.sparse.linalg import spsolve_triangular
+    >>> A = csc_array([[3, 0, 0], [1, -1, 0], [2, 0, 1]], dtype=float)
+    >>> B = np.array([[2, 0], [-1, 0], [2, 0]], dtype=float)
+    >>> x = spsolve_triangular(A, B)
+    >>> np.allclose(A.dot(x), B)
+    True
+    """
+
+    if is_pydata_spmatrix(A):
+        A = A.to_scipy_sparse().tocsc()
+    
+    trans = "N"
+    if issparse(A) and A.format == "csr":
+        A = A.T
+        trans = "T"
+        lower = not lower
+
+    if not (issparse(A) and A.format == "csc"): 
+        warn('CSC or CSR matrix format is required. Converting to CSC matrix.',
+             SparseEfficiencyWarning, stacklevel=2)
+        A = csc_matrix(A)
+    elif not overwrite_A:
+        A = A.copy()
+
+
+    M, N = A.shape
+    if M != N:
+        raise ValueError(
+            f'A must be a square matrix but its shape is {A.shape}.')
+
+    if unit_diagonal:
+        with catch_warnings():
+            simplefilter('ignore', SparseEfficiencyWarning)
+            A.setdiag(1)
+    else:
+        diag = A.diagonal()
+        if np.any(diag == 0):
+            raise LinAlgError(
+                'A is singular: zero entry on diagonal.')
+        invdiag = 1/diag
+        if trans == "N":
+            A = A @ diags(invdiag)
+        else:
+            A = (A.T @ diags(invdiag)).T
+
+    # sum duplicates for non-canonical format
+    A.sum_duplicates()
+
+    b = np.asanyarray(b)
+
+    if b.ndim not in [1, 2]:
+        raise ValueError(
+            f'b must have 1 or 2 dims but its shape is {b.shape}.')
+    if M != b.shape[0]:
+        raise ValueError(
+            'The size of the dimensions of A must be equal to '
+            'the size of the first dimension of b but the shape of A is '
+            f'{A.shape} and the shape of b is {b.shape}.'
+        )
+
+    result_dtype = np.promote_types(np.promote_types(A.dtype, np.float32), b.dtype)
+    if A.dtype != result_dtype:
+        A = A.astype(result_dtype)
+    if b.dtype != result_dtype:
+        b = b.astype(result_dtype)
+    elif not overwrite_b:
+        b = b.copy()
+
+    if lower:
+        L = A
+        U = csc_matrix((N, N), dtype=result_dtype)
+    else:
+        L = eye(N, dtype=result_dtype, format='csc')
+        U = A
+        U.setdiag(0)
+
+    x, info = _superlu.gstrs(trans,
+                             N, L.nnz, L.data, L.indices, L.indptr,
+                             N, U.nnz, U.data, U.indices, U.indptr,
+                             b)
+    if info:
+        raise LinAlgError('A is singular.')
+
+    if not unit_diagonal:
+        invdiag = invdiag.reshape(-1, *([1] * (len(x.shape) - 1)))
+        x = x * invdiag
+
+    return x
+

.venv/lib/python3.12/site-packages/scipy/sparse/linalg/_dsolve/tests/test_linsolve.py

@@ -0,0 +1,883 @@
+import sys
+import threading
+
+import numpy as np
+from numpy import array, finfo, arange, eye, all, unique, ones, dot
+import numpy.random as random
+from numpy.testing import (
+        assert_array_almost_equal, assert_almost_equal,
+        assert_equal, assert_array_equal, assert_, assert_allclose,
+        assert_warns, suppress_warnings)
+import pytest
+from pytest import raises as assert_raises
+
+import scipy.linalg
+from scipy.linalg import norm, inv
+from scipy.sparse import (spdiags, SparseEfficiencyWarning, csc_matrix,
+        csr_matrix, identity, issparse, dok_matrix, lil_matrix, bsr_matrix)
+from scipy.sparse.linalg import SuperLU
+from scipy.sparse.linalg._dsolve import (spsolve, use_solver, splu, spilu,
+        MatrixRankWarning, _superlu, spsolve_triangular, factorized)
+import scipy.sparse
+
+from scipy._lib._testutils import check_free_memory
+from scipy._lib._util import ComplexWarning
+
+
+sup_sparse_efficiency = suppress_warnings()
+sup_sparse_efficiency.filter(SparseEfficiencyWarning)
+
+# scikits.umfpack is not a SciPy dependency but it is optionally used in
+# dsolve, so check whether it's available
+try:
+    import scikits.umfpack as umfpack
+    has_umfpack = True
+except ImportError:
+    has_umfpack = False
+
+def toarray(a):
+    if issparse(a):
+        return a.toarray()
+    else:
+        return a
+
+
+def setup_bug_8278():
+    N = 2 ** 6
+    h = 1/N
+    Ah1D = scipy.sparse.diags([-1, 2, -1], [-1, 0, 1],
+                              shape=(N-1, N-1))/(h**2)
+    eyeN = scipy.sparse.eye(N - 1)
+    A = (scipy.sparse.kron(eyeN, scipy.sparse.kron(eyeN, Ah1D))
+         + scipy.sparse.kron(eyeN, scipy.sparse.kron(Ah1D, eyeN))
+         + scipy.sparse.kron(Ah1D, scipy.sparse.kron(eyeN, eyeN)))
+    b = np.random.rand((N-1)**3)
+    return A, b
+
+
+class TestFactorized:
+    def setup_method(self):
+        n = 5
+        d = arange(n) + 1
+        self.n = n
+        self.A = spdiags((d, 2*d, d[::-1]), (-3, 0, 5), n, n).tocsc()
+        random.seed(1234)
+
+    def _check_singular(self):
+        A = csc_matrix((5,5), dtype='d')
+        b = ones(5)
+        assert_array_almost_equal(0. * b, factorized(A)(b))
+
+    def _check_non_singular(self):
+        # Make a diagonal dominant, to make sure it is not singular
+        n = 5
+        a = csc_matrix(random.rand(n, n))
+        b = ones(n)
+
+        expected = splu(a).solve(b)
+        assert_array_almost_equal(factorized(a)(b), expected)
+
+    def test_singular_without_umfpack(self):
+        use_solver(useUmfpack=False)
+        with assert_raises(RuntimeError, match="Factor is exactly singular"):
+            self._check_singular()
+
+    @pytest.mark.skipif(not has_umfpack, reason="umfpack not available")
+    def test_singular_with_umfpack(self):
+        use_solver(useUmfpack=True)
+        with suppress_warnings() as sup:
+            sup.filter(RuntimeWarning, "divide by zero encountered in double_scalars")
+            assert_warns(umfpack.UmfpackWarning, self._check_singular)
+
+    def test_non_singular_without_umfpack(self):
+        use_solver(useUmfpack=False)
+        self._check_non_singular()
+
+    @pytest.mark.skipif(not has_umfpack, reason="umfpack not available")
+    def test_non_singular_with_umfpack(self):
+        use_solver(useUmfpack=True)
+        self._check_non_singular()
+
+    def test_cannot_factorize_nonsquare_matrix_without_umfpack(self):
+        use_solver(useUmfpack=False)
+        msg = "can only factor square matrices"
+        with assert_raises(ValueError, match=msg):
+            factorized(self.A[:, :4])
+
+    @pytest.mark.skipif(not has_umfpack, reason="umfpack not available")
+    def test_factorizes_nonsquare_matrix_with_umfpack(self):
+        use_solver(useUmfpack=True)
+        # does not raise
+        factorized(self.A[:,:4])
+
+    def test_call_with_incorrectly_sized_matrix_without_umfpack(self):
+        use_solver(useUmfpack=False)
+        solve = factorized(self.A)
+        b = random.rand(4)
+        B = random.rand(4, 3)
+        BB = random.rand(self.n, 3, 9)
+
+        with assert_raises(ValueError, match="is of incompatible size"):
+            solve(b)
+        with assert_raises(ValueError, match="is of incompatible size"):
+            solve(B)
+        with assert_raises(ValueError,
+                           match="object too deep for desired array"):
+            solve(BB)
+
+    @pytest.mark.skipif(not has_umfpack, reason="umfpack not available")
+    def test_call_with_incorrectly_sized_matrix_with_umfpack(self):
+        use_solver(useUmfpack=True)
+        solve = factorized(self.A)
+        b = random.rand(4)
+        B = random.rand(4, 3)
+        BB = random.rand(self.n, 3, 9)
+
+        # does not raise
+        solve(b)
+        msg = "object too deep for desired array"
+        with assert_raises(ValueError, match=msg):
+            solve(B)
+        with assert_raises(ValueError, match=msg):
+            solve(BB)
+
+    def test_call_with_cast_to_complex_without_umfpack(self):
+        use_solver(useUmfpack=False)
+        solve = factorized(self.A)
+        b = random.rand(4)
+        for t in [np.complex64, np.complex128]:
+            with assert_raises(TypeError, match="Cannot cast array data"):
+                solve(b.astype(t))
+
+    @pytest.mark.skipif(not has_umfpack, reason="umfpack not available")
+    def test_call_with_cast_to_complex_with_umfpack(self):
+        use_solver(useUmfpack=True)
+        solve = factorized(self.A)
+        b = random.rand(4)
+        for t in [np.complex64, np.complex128]:
+            assert_warns(ComplexWarning, solve, b.astype(t))
+
+    @pytest.mark.skipif(not has_umfpack, reason="umfpack not available")
+    def test_assume_sorted_indices_flag(self):
+        # a sparse matrix with unsorted indices
+        unsorted_inds = np.array([2, 0, 1, 0])
+        data = np.array([10, 16, 5, 0.4])
+        indptr = np.array([0, 1, 2, 4])
+        A = csc_matrix((data, unsorted_inds, indptr), (3, 3))
+        b = ones(3)
+
+        # should raise when incorrectly assuming indices are sorted
+        use_solver(useUmfpack=True, assumeSortedIndices=True)
+        with assert_raises(RuntimeError,
+                           match="UMFPACK_ERROR_invalid_matrix"):
+            factorized(A)
+
+        # should sort indices and succeed when not assuming indices are sorted
+        use_solver(useUmfpack=True, assumeSortedIndices=False)
+        expected = splu(A.copy()).solve(b)
+
+        assert_equal(A.has_sorted_indices, 0)
+        assert_array_almost_equal(factorized(A)(b), expected)
+
+    @pytest.mark.slow
+    @pytest.mark.skipif(not has_umfpack, reason="umfpack not available")
+    def test_bug_8278(self):
+        check_free_memory(8000)
+        use_solver(useUmfpack=True)
+        A, b = setup_bug_8278()
+        A = A.tocsc()
+        f = factorized(A)
+        x = f(b)
+        assert_array_almost_equal(A @ x, b)
+
+
+class TestLinsolve:
+    def setup_method(self):
+        use_solver(useUmfpack=False)
+
+    def test_singular(self):
+        A = csc_matrix((5,5), dtype='d')
+        b = array([1, 2, 3, 4, 5],dtype='d')
+        with suppress_warnings() as sup:
+            sup.filter(MatrixRankWarning, "Matrix is exactly singular")
+            x = spsolve(A, b)
+        assert_(not np.isfinite(x).any())
+
+    def test_singular_gh_3312(self):
+        # "Bad" test case that leads SuperLU to call LAPACK with invalid
+        # arguments. Check that it fails moderately gracefully.
+        ij = np.array([(17, 0), (17, 6), (17, 12), (10, 13)], dtype=np.int32)
+        v = np.array([0.284213, 0.94933781, 0.15767017, 0.38797296])
+        A = csc_matrix((v, ij.T), shape=(20, 20))
+        b = np.arange(20)
+
+        try:
+            # should either raise a runtime error or return value
+            # appropriate for singular input (which yields the warning)
+            with suppress_warnings() as sup:
+                sup.filter(MatrixRankWarning, "Matrix is exactly singular")
+                x = spsolve(A, b)
+            assert not np.isfinite(x).any()
+        except RuntimeError:
+            pass
+
+    @pytest.mark.parametrize('format', ['csc', 'csr'])
+    @pytest.mark.parametrize('idx_dtype', [np.int32, np.int64])
+    def test_twodiags(self, format: str, idx_dtype: np.dtype):
+        A = spdiags([[1, 2, 3, 4, 5], [6, 5, 8, 9, 10]], [0, 1], 5, 5,
+                    format=format)
+        b = array([1, 2, 3, 4, 5])
+
+        # condition number of A
+        cond_A = norm(A.toarray(), 2) * norm(inv(A.toarray()), 2)
+
+        for t in ['f','d','F','D']:
+            eps = finfo(t).eps  # floating point epsilon
+            b = b.astype(t)
+            Asp = A.astype(t)
+            Asp.indices = Asp.indices.astype(idx_dtype, copy=False)
+            Asp.indptr = Asp.indptr.astype(idx_dtype, copy=False)
+
+            x = spsolve(Asp, b)
+            assert_(norm(b - Asp@x) < 10 * cond_A * eps)
+
+    def test_bvector_smoketest(self):
+        Adense = array([[0., 1., 1.],
+                        [1., 0., 1.],
+                        [0., 0., 1.]])
+        As = csc_matrix(Adense)
+        random.seed(1234)
+        x = random.randn(3)
+        b = As@x
+        x2 = spsolve(As, b)
+
+        assert_array_almost_equal(x, x2)
+
+    def test_bmatrix_smoketest(self):
+        Adense = array([[0., 1., 1.],
+                        [1., 0., 1.],
+                        [0., 0., 1.]])
+        As = csc_matrix(Adense)
+        random.seed(1234)
+        x = random.randn(3, 4)
+        Bdense = As.dot(x)
+        Bs = csc_matrix(Bdense)
+        x2 = spsolve(As, Bs)
+        assert_array_almost_equal(x, x2.toarray())
+
+    @sup_sparse_efficiency
+    def test_non_square(self):
+        # A is not square.
+        A = ones((3, 4))
+        b = ones((4, 1))
+        assert_raises(ValueError, spsolve, A, b)
+        # A2 and b2 have incompatible shapes.
+        A2 = csc_matrix(eye(3))
+        b2 = array([1.0, 2.0])
+        assert_raises(ValueError, spsolve, A2, b2)
+
+    @sup_sparse_efficiency
+    def test_example_comparison(self):
+        row = array([0,0,1,2,2,2])
+        col = array([0,2,2,0,1,2])
+        data = array([1,2,3,-4,5,6])
+        sM = csr_matrix((data,(row,col)), shape=(3,3), dtype=float)
+        M = sM.toarray()
+
+        row = array([0,0,1,1,0,0])
+        col = array([0,2,1,1,0,0])
+        data = array([1,1,1,1,1,1])
+        sN = csr_matrix((data, (row,col)), shape=(3,3), dtype=float)
+        N = sN.toarray()
+
+        sX = spsolve(sM, sN)
+        X = scipy.linalg.solve(M, N)
+
+        assert_array_almost_equal(X, sX.toarray())
+
+    @sup_sparse_efficiency
+    @pytest.mark.skipif(not has_umfpack, reason="umfpack not available")
+    def test_shape_compatibility(self):
+        use_solver(useUmfpack=True)
+        A = csc_matrix([[1., 0], [0, 2]])
+        bs = [
+            [1, 6],
+            array([1, 6]),
+            [[1], [6]],
+            array([[1], [6]]),
+            csc_matrix([[1], [6]]),
+            csr_matrix([[1], [6]]),
+            dok_matrix([[1], [6]]),
+            bsr_matrix([[1], [6]]),
+            array([[1., 2., 3.], [6., 8., 10.]]),
+            csc_matrix([[1., 2., 3.], [6., 8., 10.]]),
+            csr_matrix([[1., 2., 3.], [6., 8., 10.]]),
+            dok_matrix([[1., 2., 3.], [6., 8., 10.]]),
+            bsr_matrix([[1., 2., 3.], [6., 8., 10.]]),
+            ]
+
+        for b in bs:
+            x = np.linalg.solve(A.toarray(), toarray(b))
+            for spmattype in [csc_matrix, csr_matrix, dok_matrix, lil_matrix]:
+                x1 = spsolve(spmattype(A), b, use_umfpack=True)
+                x2 = spsolve(spmattype(A), b, use_umfpack=False)
+
+                # check solution
+                if x.ndim == 2 and x.shape[1] == 1:
+                    # interprets also these as "vectors"
+                    x = x.ravel()
+
+                assert_array_almost_equal(toarray(x1), x,
+                                          err_msg=repr((b, spmattype, 1)))
+                assert_array_almost_equal(toarray(x2), x,
+                                          err_msg=repr((b, spmattype, 2)))
+
+                # dense vs. sparse output  ("vectors" are always dense)
+                if issparse(b) and x.ndim > 1:
+                    assert_(issparse(x1), repr((b, spmattype, 1)))
+                    assert_(issparse(x2), repr((b, spmattype, 2)))
+                else:
+                    assert_(isinstance(x1, np.ndarray), repr((b, spmattype, 1)))
+                    assert_(isinstance(x2, np.ndarray), repr((b, spmattype, 2)))
+
+                # check output shape
+                if x.ndim == 1:
+                    # "vector"
+                    assert_equal(x1.shape, (A.shape[1],))
+                    assert_equal(x2.shape, (A.shape[1],))
+                else:
+                    # "matrix"
+                    assert_equal(x1.shape, x.shape)
+                    assert_equal(x2.shape, x.shape)
+
+        A = csc_matrix((3, 3))
+        b = csc_matrix((1, 3))
+        assert_raises(ValueError, spsolve, A, b)
+
+    @sup_sparse_efficiency
+    def test_ndarray_support(self):
+        A = array([[1., 2.], [2., 0.]])
+        x = array([[1., 1.], [0.5, -0.5]])
+        b = array([[2., 0.], [2., 2.]])
+
+        assert_array_almost_equal(x, spsolve(A, b))
+
+    def test_gssv_badinput(self):
+        N = 10
+        d = arange(N) + 1.0
+        A = spdiags((d, 2*d, d[::-1]), (-3, 0, 5), N, N)
+
+        for spmatrix in (csc_matrix, csr_matrix):
+            A = spmatrix(A)
+            b = np.arange(N)
+
+            def not_c_contig(x):
+                return x.repeat(2)[::2]
+
+            def not_1dim(x):
+                return x[:,None]
+
+            def bad_type(x):
+                return x.astype(bool)
+
+            def too_short(x):
+                return x[:-1]
+
+            badops = [not_c_contig, not_1dim, bad_type, too_short]
+
+            for badop in badops:
+                msg = f"{spmatrix!r} {badop!r}"
+                # Not C-contiguous
+                assert_raises((ValueError, TypeError), _superlu.gssv,
+                              N, A.nnz, badop(A.data), A.indices, A.indptr,
+                              b, int(spmatrix == csc_matrix), err_msg=msg)
+                assert_raises((ValueError, TypeError), _superlu.gssv,
+                              N, A.nnz, A.data, badop(A.indices), A.indptr,
+                              b, int(spmatrix == csc_matrix), err_msg=msg)
+                assert_raises((ValueError, TypeError), _superlu.gssv,
+                              N, A.nnz, A.data, A.indices, badop(A.indptr),
+                              b, int(spmatrix == csc_matrix), err_msg=msg)
+
+    def test_sparsity_preservation(self):
+        ident = csc_matrix([
+            [1, 0, 0],
+            [0, 1, 0],
+            [0, 0, 1]])
+        b = csc_matrix([
+            [0, 1],
+            [1, 0],
+            [0, 0]])
+        x = spsolve(ident, b)
+        assert_equal(ident.nnz, 3)
+        assert_equal(b.nnz, 2)
+        assert_equal(x.nnz, 2)
+        assert_allclose(x.toarray(), b.toarray(), atol=1e-12, rtol=1e-12)
+
+    def test_dtype_cast(self):
+        A_real = scipy.sparse.csr_matrix([[1, 2, 0],
+                                          [0, 0, 3],
+                                          [4, 0, 5]])
+        A_complex = scipy.sparse.csr_matrix([[1, 2, 0],
+                                             [0, 0, 3],
+                                             [4, 0, 5 + 1j]])
+        b_real = np.array([1,1,1])
+        b_complex = np.array([1,1,1]) + 1j*np.array([1,1,1])
+        x = spsolve(A_real, b_real)
+        assert_(np.issubdtype(x.dtype, np.floating))
+        x = spsolve(A_real, b_complex)
+        assert_(np.issubdtype(x.dtype, np.complexfloating))
+        x = spsolve(A_complex, b_real)
+        assert_(np.issubdtype(x.dtype, np.complexfloating))
+        x = spsolve(A_complex, b_complex)
+        assert_(np.issubdtype(x.dtype, np.complexfloating))
+
+    @pytest.mark.slow
+    @pytest.mark.skipif(not has_umfpack, reason="umfpack not available")
+    def test_bug_8278(self):
+        check_free_memory(8000)
+        use_solver(useUmfpack=True)
+        A, b = setup_bug_8278()
+        x = spsolve(A, b)
+        assert_array_almost_equal(A @ x, b)
+
+
+class TestSplu:
+    def setup_method(self):
+        use_solver(useUmfpack=False)
+        n = 40
+        d = arange(n) + 1
+        self.n = n
+        self.A = spdiags((d, 2*d, d[::-1]), (-3, 0, 5), n, n, format='csc')
+        random.seed(1234)
+
+    def _smoketest(self, spxlu, check, dtype, idx_dtype):
+        if np.issubdtype(dtype, np.complexfloating):
+            A = self.A + 1j*self.A.T
+        else:
+            A = self.A
+
+        A = A.astype(dtype)
+        A.indices = A.indices.astype(idx_dtype, copy=False)
+        A.indptr = A.indptr.astype(idx_dtype, copy=False)
+        lu = spxlu(A)
+
+        rng = random.RandomState(1234)
+
+        # Input shapes
+        for k in [None, 1, 2, self.n, self.n+2]:
+            msg = f"k={k!r}"
+
+            if k is None:
+                b = rng.rand(self.n)
+            else:
+                b = rng.rand(self.n, k)
+
+            if np.issubdtype(dtype, np.complexfloating):
+                b = b + 1j*rng.rand(*b.shape)
+            b = b.astype(dtype)
+
+            x = lu.solve(b)
+            check(A, b, x, msg)
+
+            x = lu.solve(b, 'T')
+            check(A.T, b, x, msg)
+
+            x = lu.solve(b, 'H')
+            check(A.T.conj(), b, x, msg)
+
+    @sup_sparse_efficiency
+    def test_splu_smoketest(self):
+        self._internal_test_splu_smoketest()
+
+    def _internal_test_splu_smoketest(self):
+        # Check that splu works at all
+        def check(A, b, x, msg=""):
+            eps = np.finfo(A.dtype).eps
+            r = A @ x
+            assert_(abs(r - b).max() < 1e3*eps, msg)
+
+        for dtype in [np.float32, np.float64, np.complex64, np.complex128]:
+            for idx_dtype in [np.int32, np.int64]:
+                self._smoketest(splu, check, dtype, idx_dtype)
+
+    @sup_sparse_efficiency
+    def test_spilu_smoketest(self):
+        self._internal_test_spilu_smoketest()
+
+    def _internal_test_spilu_smoketest(self):
+        errors = []
+
+        def check(A, b, x, msg=""):
+            r = A @ x
+            err = abs(r - b).max()
+            assert_(err < 1e-2, msg)
+            if b.dtype in (np.float64, np.complex128):
+                errors.append(err)
+
+        for dtype in [np.float32, np.float64, np.complex64, np.complex128]:
+            for idx_dtype in [np.int32, np.int64]:
+                self._smoketest(spilu, check, dtype, idx_dtype)
+
+        assert_(max(errors) > 1e-5)
+
+    @sup_sparse_efficiency
+    def test_spilu_drop_rule(self):
+        # Test passing in the drop_rule argument to spilu.
+        A = identity(2)
+
+        rules = [
+            b'basic,area'.decode('ascii'),  # unicode
+            b'basic,area',  # ascii
+            [b'basic', b'area'.decode('ascii')]
+        ]
+        for rule in rules:
+            # Argument should be accepted
+            assert_(isinstance(spilu(A, drop_rule=rule), SuperLU))
+
+    def test_splu_nnz0(self):
+        A = csc_matrix((5,5), dtype='d')
+        assert_raises(RuntimeError, splu, A)
+
+    def test_spilu_nnz0(self):
+        A = csc_matrix((5,5), dtype='d')
+        assert_raises(RuntimeError, spilu, A)
+
+    def test_splu_basic(self):
+        # Test basic splu functionality.
+        n = 30
+        rng = random.RandomState(12)
+        a = rng.rand(n, n)
+        a[a < 0.95] = 0
+        # First test with a singular matrix
+        a[:, 0] = 0
+        a_ = csc_matrix(a)
+        # Matrix is exactly singular
+        assert_raises(RuntimeError, splu, a_)
+
+        # Make a diagonal dominant, to make sure it is not singular
+        a += 4*eye(n)
+        a_ = csc_matrix(a)
+        lu = splu(a_)
+        b = ones(n)
+        x = lu.solve(b)
+        assert_almost_equal(dot(a, x), b)
+
+    def test_splu_perm(self):
+        # Test the permutation vectors exposed by splu.
+        n = 30
+        a = random.random((n, n))
+        a[a < 0.95] = 0
+        # Make a diagonal dominant, to make sure it is not singular
+        a += 4*eye(n)
+        a_ = csc_matrix(a)
+        lu = splu(a_)
+        # Check that the permutation indices do belong to [0, n-1].
+        for perm in (lu.perm_r, lu.perm_c):
+            assert_(all(perm > -1))
+            assert_(all(perm < n))
+            assert_equal(len(unique(perm)), len(perm))
+
+        # Now make a symmetric, and test that the two permutation vectors are
+        # the same
+        # Note: a += a.T relies on undefined behavior.
+        a = a + a.T
+        a_ = csc_matrix(a)
+        lu = splu(a_)
+        assert_array_equal(lu.perm_r, lu.perm_c)
+
+    @pytest.mark.parametrize("splu_fun, rtol", [(splu, 1e-7), (spilu, 1e-1)])
+    def test_natural_permc(self, splu_fun, rtol):
+        # Test that the "NATURAL" permc_spec does not permute the matrix
+        np.random.seed(42)
+        n = 500
+        p = 0.01
+        A = scipy.sparse.random(n, n, p)
+        x = np.random.rand(n)
+        # Make A diagonal dominant to make sure it is not singular
+        A += (n+1)*scipy.sparse.identity(n)
+        A_ = csc_matrix(A)
+        b = A_ @ x
+
+        # without permc_spec, permutation is not identity
+        lu = splu_fun(A_)
+        assert_(np.any(lu.perm_c != np.arange(n)))
+
+        # with permc_spec="NATURAL", permutation is identity
+        lu = splu_fun(A_, permc_spec="NATURAL")
+        assert_array_equal(lu.perm_c, np.arange(n))
+
+        # Also, lu decomposition is valid
+        x2 = lu.solve(b)
+        assert_allclose(x, x2, rtol=rtol)
+
+    @pytest.mark.skipif(not hasattr(sys, 'getrefcount'), reason="no sys.getrefcount")
+    def test_lu_refcount(self):
+        # Test that we are keeping track of the reference count with splu.
+        n = 30
+        a = random.random((n, n))
+        a[a < 0.95] = 0
+        # Make a diagonal dominant, to make sure it is not singular
+        a += 4*eye(n)
+        a_ = csc_matrix(a)
+        lu = splu(a_)
+
+        # And now test that we don't have a refcount bug
+        rc = sys.getrefcount(lu)
+        for attr in ('perm_r', 'perm_c'):
+            perm = getattr(lu, attr)
+            assert_equal(sys.getrefcount(lu), rc + 1)
+            del perm
+            assert_equal(sys.getrefcount(lu), rc)
+
+    def test_bad_inputs(self):
+        A = self.A.tocsc()
+
+        assert_raises(ValueError, splu, A[:,:4])
+        assert_raises(ValueError, spilu, A[:,:4])
+
+        for lu in [splu(A), spilu(A)]:
+            b = random.rand(42)
+            B = random.rand(42, 3)
+            BB = random.rand(self.n, 3, 9)
+            assert_raises(ValueError, lu.solve, b)
+            assert_raises(ValueError, lu.solve, B)
+            assert_raises(ValueError, lu.solve, BB)
+            assert_raises(TypeError, lu.solve,
+                          b.astype(np.complex64))
+            assert_raises(TypeError, lu.solve,
+                          b.astype(np.complex128))
+
+    @sup_sparse_efficiency
+    def test_superlu_dlamch_i386_nan(self):
+        # SuperLU 4.3 calls some functions returning floats without
+        # declaring them. On i386@linux call convention, this fails to
+        # clear floating point registers after call. As a result, NaN
+        # can appear in the next floating point operation made.
+        #
+        # Here's a test case that triggered the issue.
+        n = 8
+        d = np.arange(n) + 1
+        A = spdiags((d, 2*d, d[::-1]), (-3, 0, 5), n, n)
+        A = A.astype(np.float32)
+        spilu(A)
+        A = A + 1j*A
+        B = A.toarray()
+        assert_(not np.isnan(B).any())
+
+    @sup_sparse_efficiency
+    def test_lu_attr(self):
+
+        def check(dtype, complex_2=False):
+            A = self.A.astype(dtype)
+
+            if complex_2:
+                A = A + 1j*A.T
+
+            n = A.shape[0]
+            lu = splu(A)
+
+            # Check that the decomposition is as advertised
+
+            Pc = np.zeros((n, n))
+            Pc[np.arange(n), lu.perm_c] = 1
+
+            Pr = np.zeros((n, n))
+            Pr[lu.perm_r, np.arange(n)] = 1
+
+            Ad = A.toarray()
+            lhs = Pr.dot(Ad).dot(Pc)
+            rhs = (lu.L @ lu.U).toarray()
+
+            eps = np.finfo(dtype).eps
+
+            assert_allclose(lhs, rhs, atol=100*eps)
+
+        check(np.float32)
+        check(np.float64)
+        check(np.complex64)
+        check(np.complex128)
+        check(np.complex64, True)
+        check(np.complex128, True)
+
+    @pytest.mark.slow
+    @sup_sparse_efficiency
+    def test_threads_parallel(self):
+        oks = []
+
+        def worker():
+            try:
+                self.test_splu_basic()
+                self._internal_test_splu_smoketest()
+                self._internal_test_spilu_smoketest()
+                oks.append(True)
+            except Exception:
+                pass
+
+        threads = [threading.Thread(target=worker)
+                   for k in range(20)]
+        for t in threads:
+            t.start()
+        for t in threads:
+            t.join()
+
+        assert_equal(len(oks), 20)
+
+    def test_singular_matrix(self):
+        # Test that SuperLU does not print to stdout when a singular matrix is
+        # passed. See gh-20993.
+        A = identity(10, format='csr').tocsr()
+        A[-1, -1] = 0
+        b = np.zeros(10)
+        with pytest.warns(MatrixRankWarning):
+            res = spsolve(A, b)
+            assert np.isnan(res).all()
+
+
+class TestGstrsErrors:
+    def setup_method(self):
+      self.A = array([[1.0,2.0,3.0],[4.0,5.0,6.0],[7.0,8.0,9.0]], dtype=np.float64)
+      self.b = np.array([[1.0],[2.0],[3.0]], dtype=np.float64)
+
+    def test_trans(self):
+        L = scipy.sparse.tril(self.A, format='csc')
+        U = scipy.sparse.triu(self.A, k=1, format='csc')
+        with assert_raises(ValueError, match="trans must be N, T, or H"):
+            _superlu.gstrs('X', L.shape[0], L.nnz, L.data, L.indices, L.indptr,
+                                U.shape[0], U.nnz, U.data, U.indices, U.indptr, self.b)
+
+    def test_shape_LU(self):
+        L = scipy.sparse.tril(self.A[0:2,0:2], format='csc')
+        U = scipy.sparse.triu(self.A, k=1, format='csc')
+        with assert_raises(ValueError, match="L and U must have the same dimension"):
+            _superlu.gstrs('N', L.shape[0], L.nnz, L.data, L.indices, L.indptr,
+                                U.shape[0], U.nnz, U.data, U.indices, U.indptr, self.b)
+
+    def test_shape_b(self):
+        L = scipy.sparse.tril(self.A, format='csc')
+        U = scipy.sparse.triu(self.A, k=1, format='csc')
+        with assert_raises(ValueError, match="right hand side array has invalid shape"):
+            _superlu.gstrs('N', L.shape[0], L.nnz, L.data, L.indices, L.indptr,
+                                U.shape[0], U.nnz, U.data, U.indices, U.indptr, 
+                                self.b[0:2])
+
+    def test_types_differ(self):
+        L = scipy.sparse.tril(self.A.astype(np.float32), format='csc')
+        U = scipy.sparse.triu(self.A, k=1, format='csc')
+        with assert_raises(TypeError, match="nzvals types of L and U differ"):
+            _superlu.gstrs('N', L.shape[0], L.nnz, L.data, L.indices, L.indptr,
+                                U.shape[0], U.nnz, U.data, U.indices, U.indptr, self.b)
+
+    def test_types_unsupported(self):
+        L = scipy.sparse.tril(self.A.astype(np.uint8), format='csc')
+        U = scipy.sparse.triu(self.A.astype(np.uint8), k=1, format='csc')
+        with assert_raises(TypeError, match="nzvals is not of a type supported"):
+            _superlu.gstrs('N', L.shape[0], L.nnz, L.data, L.indices, L.indptr,
+                                U.shape[0], U.nnz, U.data, U.indices, U.indptr, 
+                                self.b.astype(np.uint8))
+
+class TestSpsolveTriangular:
+    def setup_method(self):
+        use_solver(useUmfpack=False)
+
+    @pytest.mark.parametrize("fmt",["csr","csc"])
+    def test_zero_diagonal(self,fmt):
+        n = 5
+        rng = np.random.default_rng(43876432987)
+        A = rng.standard_normal((n, n))
+        b = np.arange(n)
+        A = scipy.sparse.tril(A, k=0, format=fmt)
+
+        x = spsolve_triangular(A, b, unit_diagonal=True, lower=True)
+
+        A.setdiag(1)
+        assert_allclose(A.dot(x), b)
+
+        # Regression test from gh-15199
+        A = np.array([[0, 0, 0], [1, 0, 0], [1, 1, 0]], dtype=np.float64)
+        b = np.array([1., 2., 3.])
+        with suppress_warnings() as sup:
+            sup.filter(SparseEfficiencyWarning, "CSC or CSR matrix format is")
+            spsolve_triangular(A, b, unit_diagonal=True)
+
+    @pytest.mark.parametrize("fmt",["csr","csc"])
+    def test_singular(self,fmt):
+        n = 5
+        if fmt == "csr":
+            A = csr_matrix((n, n))
+        else:
+            A = csc_matrix((n, n))
+        b = np.arange(n)
+        for lower in (True, False):
+            assert_raises(scipy.linalg.LinAlgError,
+                          spsolve_triangular, A, b, lower=lower)
+
+    @sup_sparse_efficiency
+    def test_bad_shape(self):
+        # A is not square.
+        A = np.zeros((3, 4))
+        b = ones((4, 1))
+        assert_raises(ValueError, spsolve_triangular, A, b)
+        # A2 and b2 have incompatible shapes.
+        A2 = csr_matrix(eye(3))
+        b2 = array([1.0, 2.0])
+        assert_raises(ValueError, spsolve_triangular, A2, b2)
+
+    @sup_sparse_efficiency
+    def test_input_types(self):
+        A = array([[1., 0.], [1., 2.]])
+        b = array([[2., 0.], [2., 2.]])
+        for matrix_type in (array, csc_matrix, csr_matrix):
+            x = spsolve_triangular(matrix_type(A), b, lower=True)
+            assert_array_almost_equal(A.dot(x), b)
+
+    @pytest.mark.slow
+    @sup_sparse_efficiency
+    @pytest.mark.parametrize("n", [10, 10**2, 10**3])
+    @pytest.mark.parametrize("m", [1, 10])
+    @pytest.mark.parametrize("lower", [True, False])
+    @pytest.mark.parametrize("format", ["csr", "csc"])
+    @pytest.mark.parametrize("unit_diagonal", [False, True])
+    @pytest.mark.parametrize("choice_of_A", ["real", "complex"])
+    @pytest.mark.parametrize("choice_of_b", ["floats", "ints", "complexints"])
+    def test_random(self, n, m, lower, format, unit_diagonal, choice_of_A, choice_of_b):
+        def random_triangle_matrix(n, lower=True, format="csr", choice_of_A="real"):
+            if choice_of_A == "real":
+                dtype = np.float64
+            elif choice_of_A == "complex":
+                dtype = np.complex128
+            else:
+                raise ValueError("choice_of_A must be 'real' or 'complex'.")
+            rng = np.random.default_rng(789002319)
+            rvs = rng.random
+            A = scipy.sparse.random(n, n, density=0.1, format='lil', dtype=dtype,
+                    random_state=rng, data_rvs=rvs)
+            if lower:
+                A = scipy.sparse.tril(A, format="lil")
+            else:
+                A = scipy.sparse.triu(A, format="lil")
+            for i in range(n):
+                A[i, i] = np.random.rand() + 1
+            if format == "csc":
+                A = A.tocsc(copy=False)
+            else:
+                A = A.tocsr(copy=False)
+            return A
+
+        np.random.seed(1234)
+        A = random_triangle_matrix(n, lower=lower)
+        if choice_of_b == "floats":
+            b = np.random.rand(n, m)
+        elif choice_of_b == "ints":
+            b = np.random.randint(-9, 9, (n, m))
+        elif choice_of_b == "complexints":
+            b = np.random.randint(-9, 9, (n, m)) + np.random.randint(-9, 9, (n, m)) * 1j
+        else:
+            raise ValueError(
+                "choice_of_b must be 'floats', 'ints', or 'complexints'.")
+        x = spsolve_triangular(A, b, lower=lower, unit_diagonal=unit_diagonal)
+        if unit_diagonal:
+            A.setdiag(1)
+        assert_allclose(A.dot(x), b, atol=1.5e-6)
+
+
+