emerge 0.4.10__py3-none-any.whl → 0.5.0__py3-none-any.whl

emerge/_emerge/solve_interfaces/pardiso_interface.py

@@ -0,0 +1,468 @@
+
+# EMerge is an open source Python based FEM EM simulation module.
+# Copyright (C) 2025  Robert Fennis.
+
+# This program is free software; you can redistribute it and/or
+# modify it under the terms of the GNU General Public License
+# as published by the Free Software Foundation; either version 2
+# of the License, or (at your option) any later version.
+
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+
+# You should have received a copy of the GNU General Public License
+# along with this program; if not, see
+# <https://www.gnu.org/licenses/>.
+from __future__ import annotations
+import os
+import sys
+import ctypes
+import re
+import site
+from ctypes.util import find_library
+from enum import Enum
+import numpy as np
+from scipy import sparse
+from pathlib import Path
+from typing import Iterable, Iterator
+import pickle
+from loguru import logger
+
+############################################################
+#                        ERROR CODES                       #
+############################################################
+
+PARDISO_ERROR_CODES = {
+     0: "No error.",
+    -1: "Input inconsistent.",
+    -2: "Not enough memory.",
+    -3: "Reordering problem.",
+    -4: "Zero pivot, numerical fac. or iterative refinement problem.",
+    -5: "Unclassified (internal) error.",
+    -6: "Preordering failed (matrix types 11(real and nonsymmetric), 13(complex and nonsymmetric) only).",
+    -7: "Diagonal Matrix problem.",
+    -8: "32-bit integer overflow problem.",
+   -10: "No license file pardiso.lic found.",
+   -11: "License is expired.",
+   -12: "Wrong username or hostname.",
+  -100: "Reached maximum number of Krylov-subspace iteration in iterative solver.",
+  -101: "No sufficient convergence in Krylov-subspace iteration within 25 iterations.",
+  -102: "Error in Krylov-subspace iteration.",
+  -103: "Bread-Down in Krylov-subspace iteration",
+}
+
+
+############################################################
+#               FINDING THE PARDISO DLL FILES              #
+############################################################
+
+
+#: Environment variable that overrides automatic searching
+ENV_VAR = "PYPARDISO_MKL_RT"
+
+
+def _candidate_dirs() -> Iterable[Path]:
+    """Return directories in which to look for MKL."""
+    # Ordered from most to least likely
+    seen: set[Path] = set()
+
+    for p in (                         # likely “local” env first
+        Path(sys.prefix),
+        Path(getattr(sys, "base_prefix", sys.prefix)),
+        Path(site.USER_BASE),
+        *(Path(x) for x in os.getenv("LD_LIBRARY_PATH", "").split(":") if x),
+    ):
+        if p not in seen:
+            seen.add(p)
+            yield p
+
+def _search_mkl() -> Iterator[Path]:
+    """Yield candidate MKL library paths, shortest first."""
+    pattern = {
+        "win32": r"^mkl_rt.*\.dll$",
+        "darwin": r"^libmkl_rt(\.\d+)*\.dylib$",
+        "linux": r"^libmkl_rt(\.so(\.\d+)*)?$",
+    }.get(sys.platform, r"^libmkl_rt")
+
+    regex = re.compile(pattern, re.IGNORECASE)
+
+    for base in _candidate_dirs():
+        for path in sorted(base.rglob("**/*mkl_rt*"), key=lambda p: len(str(p))):
+            if regex.match(path.name):
+                yield path
+
+def cache_path_result(tag: str, compute_fn, force: bool = False):
+    """
+    Retrieve a cached Path object or compute it and store it.
+    
+    Parameters
+    ----------
+    tag : str
+        Cache key.
+    compute_fn : callable
+        Callable that returns a Path.
+    force : bool
+        If True, bypass and overwrite the cache.
+    """
+    cache_dir = Path(__file__).parent / "__pycache__"
+    cache_dir.mkdir(exist_ok=True)
+    cache_file = cache_dir / f"{tag}.pkl"
+
+    if not force and cache_file.exists():
+        with open(cache_file, "rb") as f:
+            filename = pickle.load(f)
+            print(f"Using cached MKL file: {filename}")
+            return filename
+
+    result = compute_fn()
+    with open(cache_file, "wb") as f:
+        pickle.dump(result, f)
+    return result
+
+def search_mkl() -> str:
+    """Searches for the file path of the PARDISO MKL executable
+
+    Returns:
+        str: The filepath
+    """
+    logger.debug('Searching for MKL executable...')
+    for candidate in _search_mkl():
+        try:
+            ctypes.CDLL(candidate)
+        except OSError:
+            continue  # try the next one
+    logger.debug(f'Executable found: {candidate}')
+    return candidate
+
+def load_mkl() -> ctypes.CDLL:
+    """Locate and load **mkl_rt**; raise ImportError on failure."""
+    # 1 explicit override
+    override = os.getenv(ENV_VAR)
+    if override:
+        try:
+            return ctypes.CDLL(override)
+        except OSError as e:
+            raise ImportError(f"{override!r} could not be loaded: {e}") from None
+
+    # 2 system utility (cheap)
+    lib = find_library("mkl_rt")
+    if lib:
+        try:
+            return ctypes.CDLL(lib)
+        except OSError:
+            pass
+    
+    # 2 system utility (cheap)
+    lib = find_library("mkl_rt.1")
+    if lib:
+        try:
+            return ctypes.CDLL(lib)
+        except OSError:
+            pass
+
+    # 3 filesystem walk (expensive, but last resort)
+    try:
+        filename = cache_path_result('mkl_file',search_mkl)
+        return ctypes.CDLL(filename)
+    except OSError:
+        logger.warning('File name {filename} is no longer valid. Re-executing MKL search.')
+        filename = cache_path_result('mkl_file',search_mkl,force=True)
+        return ctypes.CDLL(filename)
+
+    raise ImportError(
+        "Shared library *mkl_rt* not found. "
+        f"Set the environment variable {ENV_VAR} to its full path if it is in a non-standard location."
+    )
+
+    
+
+############################################################
+#                  ALL C-TYPE DEFINITIONS                 #
+############################################################
+
+
+class MKL_Complex16(ctypes.Structure):
+    _fields_ = [("real", ctypes.c_double),
+                ("imag", ctypes.c_double)]
+
+CINT64 = ctypes.c_int64
+CINT32 = ctypes.c_int32
+CFLOAT32 = ctypes.c_float
+CFLOAT64 = ctypes.c_double
+
+CPX16_P = ctypes.POINTER(MKL_Complex16)
+CINT64_P = ctypes.POINTER(CINT64)
+CINT32_P = ctypes.POINTER(CINT32)
+CNONE_P = ctypes.POINTER(None)
+CFLOAT32_P = ctypes.POINTER(CFLOAT32)
+CFLOAT64_P = ctypes.POINTER(CFLOAT64)
+VOID_SIZE = ctypes.sizeof(ctypes.c_void_p)
+
+if VOID_SIZE == 8:
+    PT_A = CINT64
+    PT_B = np.int64
+elif VOID_SIZE == 4:
+    PT_A = CINT32
+    PT_B = np.int32
+
+def c_int(value: int):
+    return ctypes.byref(ctypes.c_int32(value))
+
+PARDISO_ARG_TYPES = (ctypes.POINTER(PT_A),CINT32_P,CINT32_P,
+    CINT32_P,CINT32_P,CINT32_P,CNONE_P,CINT32_P,CINT32_P,
+    CINT32_P,CINT32_P,CINT32_P,CINT32_P,CNONE_P,CNONE_P,CINT32_P,
+)
+
+
+############################################################
+#                  PARDISO CONFIGURATIONS                 #
+############################################################
+
+
+class PARDISOMType(Enum):
+    REAL_SYM_STRUCT = 1
+    REAL_SYM_POSDEF = 2
+    REAL_SYM_INDEF = -2
+    COMP_SYM_STRUCT = 3
+    COMP_HERM_POSDEF = 4
+    COMP_HERM_INDEF = -4
+    COMP_SYM = 6
+    REAL_NONSYM = 11
+    COMP_NONSYM = 13
+
+class PARDISOPhase(Enum):
+    SYMBOLIC_FACTOR = 11
+    NUMERIC_FACTOR = 12
+    NUMERIC_SOLVE = 33
+    
+
+
+############################################################
+#                   GENERIC MATRIX CLASS                  #
+############################################################
+
+
+class SolveMatrix:
+    def __init__(self, A: sparse.csr_matrix):
+        A = A.tocsr()
+        if not A.has_sorted_indices:
+            A.sort_indices()
+
+        if (A.getnnz(axis=1) == 0).any() or (A.getnnz(axis=0) == 0).any():
+            raise ValueError('Matrix A is singular, because it contains empty rows or columns')
+
+        if A.dtype in (np.float16, np.float32, np.float64):
+            A = A.astype(np.float64)
+        elif np.iscomplexobj(A):
+            A = A.astype(np.complex128)
+        self.mat: sparse.csr_matrix = A
+
+        self.factorized_symb: bool = False
+        self.factorized_num: bool = False
+
+    
+    def is_same(self, matrix: SolveMatrix):
+        return id(self.mat) is id(matrix.mat)
+
+    def same_structure(self, matrix: SolveMatrix):
+        return np.all(matrix.mat.indices == self.mat.indices) & np.all(matrix.mat.indptr == self.mat.indptr)
+
+    @property
+    def zerovec(self):
+        return np.zeros_like((self.mat.shape[0], 1), dtype=self.mat.dtype)
+
+
+############################################################
+#                   THE PARDISO INTERFACE                  #
+############################################################
+
+class PardisoInterface:
+
+    def __init__(self):
+
+        self.libmkl: ctypes.CDLL = load_mkl()
+        self._pardiso_interface = self.libmkl.pardiso
+
+        self._pardiso_interface.restype = None
+        self._pardiso_interface.argtypes = PARDISO_ARG_TYPES
+
+        self.PT = np.zeros(64, dtype=PT_B)
+        self.IPARM = np.zeros(64, dtype=np.int32)
+        self.PERM = np.zeros(0, dtype=np.int32)
+        self.MATRIX_TYPE: PARDISOMType = None
+        self.complex: bool = False
+        self.mat_structure: np.ndarray = None
+        self.message_level: int = 0
+
+        self.matrix: SolveMatrix = None
+        self.factored_matrix: SolveMatrix = None
+
+        self.configure_solver()
+
+    def _configure(self, A: sparse.csr_matrix) -> None:
+        """Configures the solver for the appropriate data type (float/complex)
+
+        Args:
+            A (sparse.csr_matrix): The sparse matrix to solve
+        """
+        if np.iscomplexobj(A):
+            self.MATRIX_TYPE = PARDISOMType.COMP_SYM_STRUCT
+            self.complex = True
+        else:
+            self.MATRIX_TYPE = PARDISOMType.REAL_SYM_STRUCT
+            self.complex = False
+
+    def _prepare_B(self, b: np.ndarray) -> np.ndarray:
+        """Fixes the forcing-vector for the solution process
+
+        Args:
+            b (np.ndarray): The forcing vector in Ax=b
+
+        Returns:
+            np.ndarray: The prepared forcing-vector
+        """
+        if sparse.issparse(b):
+            b = b.todense()
+        if np.iscomplexobj(b):
+            b = b.astype(np.complex128)
+        else:
+            b = b.astype(np.float64)
+        return b
+    
+    def symbolic(self, A: sparse.csr_matrix) -> int:
+        """Calls the Symbollic solve routinge
+
+        Returns:
+            int: The error code
+        """
+        print("SYMBOLIC FACTORIZATION")
+        self._configure(A)
+        zerovec = np.zeros_like((A.shape[0], 1), dtype=A.dtype)
+        _, error = self._call_solver(A, zerovec, phase=PARDISOPhase.SYMBOLIC_FACTOR)
+        return error
+
+    def numeric(self, A: sparse.csr_matrix) -> int:
+        """Calls the Numeric solve routine
+
+        Returns:
+            int: The error code
+        """
+        print("NUMERIC FACTORIZATION")
+        self._configure(A)
+        zerovec = np.zeros_like((A.shape[0], 1), dtype=A.dtype)
+        _, error = self._call_solver(A, zerovec, phase=PARDISOPhase.NUMERIC_FACTOR)
+        return error
+        
+    def solve(self, A: sparse.csr_matrix, b: np.ndarray) -> tuple[np.ndarray, int]:
+        """ Solves the linear problem Ax=b with PARDISO
+
+        Args:
+            A (sparse.csr_matrix): The A-matrix
+            b (np.ndarray): The b-vector
+
+        Returns:
+            tuple[np.ndarray, int]: The solution vector x, and error code
+        """
+        self._configure(A)
+        b = self._prepare_B(b)
+        x, error = self._call_solver(A, b, phase=PARDISOPhase.NUMERIC_SOLVE)
+        return x, error
+
+    def configure_solver(self, 
+        perm_algo: int = 3,
+        nthreads: int = None,
+        user_perm: int = 0,
+        n_refine_steps: int = 0,
+        pivot_pert: int = 13,
+        weighted_matching: int = 2):
+        """Configures the solver
+
+        Args:
+            perm_algo (int, optional): The permutation algorithm. Defaults to 3.
+            nthreads (int, optional): The number of threads (Must be greater than OMP_NUM_THREADS). Defaults to None.
+            user_perm (int, optional): 1, if a user permuation is provided (not supported yet). Defaults to 0.
+            n_refine_steps (int, optional): Number of refinement steps. Defaults to 0.
+            pivot_pert (int, optional): _description_. Defaults to 13.
+            weighted_matching (int, optional): weighted matching mode. Defaults to 2.
+        """
+        if nthreads is None:
+            nthreads = int(os.environ.get('OMP_NUM_THREADS'))
+
+        self.IPARM[1] = perm_algo
+        self.IPARM[2] = nthreads
+        self.IPARM[4] = user_perm
+        self.IPARM[7] = n_refine_steps
+        self.IPARM[9] = pivot_pert
+        self.IPARM[12] = weighted_matching
+
+    def _call_solver(self, A: sparse.csr_matrix, b: np.ndarray, phase: PARDISOPhase) -> tuple[np.ndarray, int]:
+        """Calls the PARDISO solver on linear problem Ax=b
+
+        Args:
+            A (sparse.csr_matrix): The A-matrix
+            b (np.ndarray): The b-vector
+            phase (PARDISOPhase): The solution phase
+
+        Returns:
+            tuple[np.ndarray, int]: The solution vector x and error code.
+        """
+
+        # Declare the empty vector
+        x = np.zeros_like(b)
+        error = ctypes.c_int32(0)
+        
+        # Up the pointers as PARDISO uses [1,...] indexing
+        A_index_pointers = A.indptr + 1
+        A_indices = A.indices + 1
+
+        # Define the appropriate data type (complex vs real)
+        if self.complex:
+            VALUE_P = A.data.ctypes.data_as(CPX16_P)
+            RHS_P = b.ctypes.data_as(CPX16_P)
+            X_P = x.ctypes.data_as(CPX16_P)
+        else:
+            VALUE_P = A.data.ctypes.data_as(CFLOAT64_P)
+            RHS_P = b.ctypes.data_as(CFLOAT64_P)
+            X_P = x.ctypes.data_as(CFLOAT64_P)
+
+        # Calls the pardiso function
+        self._pardiso_interface(
+            self.PT.ctypes.data_as(ctypes.POINTER(PT_A)),
+            c_int(1),
+            c_int(1),
+            c_int(self.MATRIX_TYPE.value),
+            c_int(phase.value),
+            c_int(A.shape[0]),
+            VALUE_P,
+            A_index_pointers.ctypes.data_as(CINT32_P),
+            A_indices.ctypes.data_as(CINT32_P),
+            self.PERM.ctypes.data_as(CINT32_P),
+            c_int(1),
+            self.IPARM.ctypes.data_as(CINT32_P),
+            c_int(self.message_level),
+            RHS_P,
+            X_P,
+            ctypes.byref(error))
+        
+        # Returns the solution vector plus error code
+        return np.ascontiguousarray(x), error.value
+
+    def get_error(self, error: int) -> str:
+        """Returns the PARDISO error description string given an error number
+
+        Args:
+            error (int): The error number
+
+        Returns:
+            str: A description string
+        """
+        return PARDISO_ERROR_CODES[error]
+
+    def clear_memory(self):
+        """Clear the memory of this solver plus the PARDISO process.
+        """
+        self.factorized_A = None
+        self.matrix = None
+        self.symbolic(sparse.csr_matrix((0,0), dtype=np.complex128), np.zeros(0, dtype=np.complex128))