def _AffineExpansionStorage(args: (
tuple_of(Form),
tuple_of(Matrix.Type()),
tuple_of(Vector.Type()),
tuple_of((Form, Matrix.Type())),
tuple_of((Form, Vector.Type()))
)):
return AffineExpansionStorage_Form(args)
def _init_rhs(self, rhs: Vector.Type(),
bcs: (list_of(DirichletBC), ProductOutputDirichletBC,
dict_of(str, list_of(DirichletBC)),
dict_of(str, ProductOutputDirichletBC))):
# Create a copy of rhs, in order not to change
# the original references when applying bcs
self.rhs = rhs.copy()
def _evaluate(expression: (Matrix.Type(), Vector.Type(), Function.Type(),
TensorsList, FunctionsList,
ParametrizedTensorFactory,
ParametrizedExpressionFactory),
at: (ReducedMesh, ReducedVertices, None) = None,
**kwargs):
return evaluate_base(expression, at, **kwargs)
def _product(thetas: ThetaType, operators: tuple_of(Vector.Type())):
output = tensor_copy(operators[0])
output.zero()
for (theta, operator) in zip(thetas, operators):
theta = float(theta)
output.add_local(theta * operator.get_local())
output.apply("add")
return ProductOutput(output)
def __init__(self, args):
AffineExpansionStorage_Base.__init__(self, args)
content = list()
for arg in args:
if isinstance(arg, Form):
content.append(assemble(arg, keep_diagonal=True))
elif isinstance(arg, (Matrix.Type(), Vector.Type())):
content.append(arg)
else:
raise RuntimeError("Invalid argument to AffineExpansionStorage")
self._content = tuple(content)
def _abs(vector: Vector.Type()):
# Note: PETSc offers VecAbs and VecMax, but for symmetry with the matrix case we do the same by hand
vec = to_petsc4py(vector)
row_start, row_end = vec.getOwnershipRange()
i_max = None
value_max = None
for i in range(row_start, row_end):
val = vec.getValue(i)
if value_max is None or fabs(val) > fabs(value_max):
i_max = i
value_max = val
assert i_max is not None
assert value_max is not None
#
mpi_comm = vec.comm.tompi4py()
(global_value_max, global_i_max) = parallel_max(value_max, (i_max, ), fabs, mpi_comm)
return AbsOutput(global_value_max, global_i_max)
def _export(
solution: (
Function.Type(),
Matrix.Type(),
Vector.Type()
),
directory: (
Folders.Folder,
str
),
filename: str,
suffix: (
int,
None
) = None,
component: (
int,
str,
None
) = None
):
export_base(solution, directory, filename, suffix, component)
def AdditionalIsFunction(arg):
return isinstance(arg, Operator)
def ConvertAdditionalFunctionTypes(arg):
assert isinstance(arg, Operator)
return function_from_ufl_operators(arg)
backend = ModuleWrapper(BasisFunctionsMatrix, evaluate, Function,
FunctionsList, Matrix, NonAffineExpansionStorage,
ParametrizedTensorFactory, TensorsList, Vector)
wrapping = ModuleWrapper(function_to_vector, matrix_mul_vector,
vector_mul_vector,
vectorized_matrix_inner_vectorized_matrix)
online_backend = ModuleWrapper(OnlineMatrix=OnlineMatrix,
OnlineVector=OnlineVector)
online_wrapping = ModuleWrapper()
transpose_base = basic_transpose(backend, wrapping, online_backend,
online_wrapping, AdditionalIsFunction,
ConvertAdditionalFunctionTypes)
@backend_for("dolfin",
inputs=((BasisFunctionsMatrix, Function.Type(), FunctionsList,
Matrix.Type(), Operator, ParametrizedTensorFactory,
TensorsList, Vector.Type()), ))
def transpose(arg):
return transpose_base(arg)
# You should have received a copy of the GNU Lesser General Public License
# along with RBniCS. If not, see <http://www.gnu.org/licenses/>.
#
from math import fabs
from ufl.core.operator import Operator
from rbnics.backends.dolfin.matrix import Matrix
from rbnics.backends.dolfin.vector import Vector
from rbnics.backends.dolfin.function import Function
from rbnics.backends.dolfin.wrapping import function_from_ufl_operators, get_global_dof_coordinates, get_global_dof_component, to_petsc4py
from rbnics.utils.decorators import backend_for, overload
from rbnics.utils.mpi import parallel_max
# abs function to compute maximum absolute value of an expression, matrix or vector (for EIM). To be used in combination with max
# even though here we actually carry out both the max and the abs!
@backend_for("dolfin", inputs=((Matrix.Type(), Vector.Type(), Function.Type(), Operator), ))
def abs(expression):
return _abs(expression)
@overload
def _abs(matrix: Matrix.Type()):
# Note: PETSc offers a method MatGetRowMaxAbs, but it is not wrapped in petsc4py. We do the same by hand
mat = to_petsc4py(matrix)
row_start, row_end = mat.getOwnershipRange()
i_max, j_max = None, None
value_max = None
for i in range(row_start, row_end):
cols, vals = mat.getRow(i)
for (c, v) in zip(cols, vals):
if value_max is None or fabs(v) > fabs(value_max):
i_max = i
from rbnics.backends.dolfin.vector import Vector
from rbnics.backends.dolfin.wrapping import (build_dof_map_reader_mapping,
form_argument_space,
function_extend_or_restrict,
function_load, get_function_space,
get_function_subspace,
to_petsc4py)
from rbnics.backends.dolfin.wrapping.tensor_load import basic_tensor_load
from rbnics.utils.decorators import backend_for, ModuleWrapper
from rbnics.utils.io import Folders
backend = ModuleWrapper(Function, Matrix, Vector)
wrapping_for_wrapping = ModuleWrapper(build_dof_map_reader_mapping,
form_argument_space, to_petsc4py)
tensor_load = basic_tensor_load(backend, wrapping_for_wrapping)
wrapping = ModuleWrapper(function_extend_or_restrict,
function_load,
get_function_space,
get_function_subspace,
tensor_load=tensor_load)
import_base = basic_import_(backend, wrapping)
# Import a solution from file
@backend_for("dolfin",
inputs=((Function.Type(), Matrix.Type(), Vector.Type()),
(Folders.Folder, str), str, (int, None), (int, str, None))
)
def import_(solution, directory, filename, suffix=None, component=None):
import_base(solution, directory, filename, suffix, component)
# Copyright (C) 2015-2021 by the RBniCS authors
#
# This file is part of RBniCS.
#
# SPDX-License-Identifier: LGPL-3.0-or-later
from rbnics.backends.basic import copy as basic_copy
from rbnics.backends.dolfin.function import Function
from rbnics.backends.dolfin.matrix import Matrix
from rbnics.backends.dolfin.vector import Vector
from rbnics.backends.dolfin.wrapping import function_copy, tensor_copy
from rbnics.utils.decorators import backend_for, list_of, ModuleWrapper
backend = ModuleWrapper(Function, Matrix, Vector)
wrapping = ModuleWrapper(function_copy, tensor_copy)
copy_base = basic_copy(backend, wrapping)
@backend_for("dolfin",
inputs=((Function.Type(), list_of(Function.Type()), Matrix.Type(),
Vector.Type()), ))
def copy(arg):
return copy_base(arg)
def _assign(object_to: Vector.Type(), object_from: Vector.Type()):
if object_from is not object_to:
to_petsc4py(object_from).copy(to_petsc4py(object_to))
# Copyright (C) 2015-2021 by the RBniCS authors
#
# This file is part of RBniCS.
#
# SPDX-License-Identifier: LGPL-3.0-or-later
from ufl.core.operator import Operator
from dolfin import assign as dolfin_assign
from rbnics.backends.dolfin.function import Function
from rbnics.backends.dolfin.matrix import Matrix
from rbnics.backends.dolfin.vector import Vector
from rbnics.backends.dolfin.wrapping import function_from_ufl_operators, to_petsc4py
from rbnics.utils.decorators import backend_for, list_of, overload
@backend_for("dolfin", inputs=((Function.Type(), list_of(Function.Type()), Matrix.Type(), Vector.Type()),
(Function.Type(), list_of(Function.Type()), Matrix.Type(), Operator, Vector.Type())))
def assign(object_to, object_from):
_assign(object_to, object_from)
@overload
def _assign(object_to: Function.Type(), object_from: Function.Type()):
if object_from is not object_to:
dolfin_assign(object_to, object_from)
@overload
def _assign(object_to: Function.Type(), object_from: Operator):
dolfin_assign(object_to, function_from_ufl_operators(object_from))
# You should have received a copy of the GNU Lesser General Public License
# along with RBniCS. If not, see <http://www.gnu.org/licenses/>.
#
from ufl.core.operator import Operator
from dolfin import assign as dolfin_assign
from rbnics.backends.dolfin.function import Function
from rbnics.backends.dolfin.matrix import Matrix
from rbnics.backends.dolfin.vector import Vector
from rbnics.backends.dolfin.wrapping import function_from_ufl_operators, to_petsc4py
from rbnics.utils.decorators import backend_for, list_of, overload
@backend_for("dolfin",
inputs=((Function.Type(), list_of(Function.Type()), Matrix.Type(),
Vector.Type()),
(Function.Type(), list_of(Function.Type()), Matrix.Type(),
Operator, Vector.Type())))
def assign(object_to, object_from):
_assign(object_to, object_from)
@overload
def _assign(object_to: Function.Type(), object_from: Function.Type()):
if object_from is not object_to:
dolfin_assign(object_to, object_from)
@overload
def _assign(object_to: Function.Type(), object_from: Operator):
dolfin_assign(object_to, function_from_ufl_operators(object_from))
from ufl import Form
from dolfin import assemble, DirichletBC, PETScLUSolver
from rbnics.backends.abstract import LinearSolver as AbstractLinearSolver, LinearProblemWrapper
from rbnics.backends.dolfin.evaluate import evaluate
from rbnics.backends.dolfin.function import Function
from rbnics.backends.dolfin.matrix import Matrix
from rbnics.backends.dolfin.parametrized_tensor_factory import ParametrizedTensorFactory
from rbnics.backends.dolfin.vector import Vector
from rbnics.backends.dolfin.wrapping.dirichlet_bc import ProductOutputDirichletBC
from rbnics.utils.decorators import BackendFor, dict_of, list_of, overload
@BackendFor("dolfin",
inputs=((Form, Matrix.Type(), ParametrizedTensorFactory,
LinearProblemWrapper), Function.Type(),
(Form, ParametrizedTensorFactory, Vector.Type(),
None), (list_of(DirichletBC), ProductOutputDirichletBC,
dict_of(str, list_of(DirichletBC)),
dict_of(str, ProductOutputDirichletBC), None)))
class LinearSolver(AbstractLinearSolver):
@overload((Form, Matrix.Type(), ParametrizedTensorFactory),
Function.Type(),
(Form, ParametrizedTensorFactory, Vector.Type()),
(list_of(DirichletBC), ProductOutputDirichletBC,
dict_of(str, list_of(DirichletBC)),
dict_of(str, ProductOutputDirichletBC), None))
def __init__(self, lhs, solution, rhs, bcs=None):
self.solution = solution
self._init_lhs(lhs, bcs)
self._init_rhs(rhs, bcs)
self._apply_bcs(bcs)
evaluate_sparse_vector_at_dofs,
expression_on_reduced_mesh=expression_on_reduced_mesh,
expression_on_truth_mesh=expression_on_truth_mesh,
form_on_reduced_function_space=form_on_reduced_function_space,
form_on_truth_function_space=form_on_truth_function_space)
online_backend = ModuleWrapper(OnlineFunction=OnlineFunction,
OnlineMatrix=OnlineMatrix,
OnlineVector=OnlineVector)
online_wrapping = ModuleWrapper()
evaluate_base = basic_evaluate(backend, wrapping, online_backend,
online_wrapping)
# Evaluate a parametrized expression, possibly at a specific location
@backend_for("dolfin",
inputs=((Matrix.Type(), Vector.Type(), Function.Type(), Operator,
TensorsList, FunctionsList, ParametrizedTensorFactory,
ParametrizedExpressionFactory), (ReducedMesh,
ReducedVertices, None)))
def evaluate(expression, at=None, **kwargs):
return _evaluate(expression, at, **kwargs)
@overload
def _evaluate(expression: (Matrix.Type(), Vector.Type(), TensorsList),
at: ReducedMesh, **kwargs):
assert len(kwargs) == 0
return evaluate_base(expression, at)
@overload
from ufl.core.operator import Operator
from rbnics.backends.basic import transpose as basic_transpose
from rbnics.backends.dolfin.basis_functions_matrix import BasisFunctionsMatrix
from rbnics.backends.dolfin.evaluate import evaluate
from rbnics.backends.dolfin.function import Function
from rbnics.backends.dolfin.functions_list import FunctionsList
from rbnics.backends.dolfin.matrix import Matrix
from rbnics.backends.dolfin.non_affine_expansion_storage import NonAffineExpansionStorage
from rbnics.backends.dolfin.parametrized_tensor_factory import ParametrizedTensorFactory
from rbnics.backends.dolfin.tensors_list import TensorsList
from rbnics.backends.dolfin.vector import Vector
from rbnics.backends.dolfin.wrapping import function_from_ufl_operators, function_to_vector, matrix_mul_vector, vector_mul_vector, vectorized_matrix_inner_vectorized_matrix
from rbnics.backends.online import OnlineMatrix, OnlineVector
from rbnics.utils.decorators import backend_for, ModuleWrapper
def AdditionalIsFunction(arg):
return isinstance(arg, Operator)
def ConvertAdditionalFunctionTypes(arg):
assert isinstance(arg, Operator)
return function_from_ufl_operators(arg)
backend = ModuleWrapper(BasisFunctionsMatrix, evaluate, Function, FunctionsList, Matrix, NonAffineExpansionStorage, ParametrizedTensorFactory, TensorsList, Vector)
wrapping = ModuleWrapper(function_to_vector, matrix_mul_vector, vector_mul_vector, vectorized_matrix_inner_vectorized_matrix)
online_backend = ModuleWrapper(OnlineMatrix=OnlineMatrix, OnlineVector=OnlineVector)
online_wrapping = ModuleWrapper()
transpose_base = basic_transpose(backend, wrapping, online_backend, online_wrapping, AdditionalIsFunction, ConvertAdditionalFunctionTypes)
@backend_for("dolfin", inputs=((BasisFunctionsMatrix, Function.Type(), FunctionsList, Matrix.Type(), Operator, ParametrizedTensorFactory, TensorsList, Vector.Type()), ))
def transpose(arg):
return transpose_base(arg)
class LinearSolver(AbstractLinearSolver):
@overload((Form, Matrix.Type(), ParametrizedTensorFactory),
Function.Type(), (Form, ParametrizedTensorFactory, Vector.Type()),
(list_of(DirichletBC), ProductOutputDirichletBC, dict_of(str, list_of(DirichletBC)),
dict_of(str, ProductOutputDirichletBC), None))
def __init__(self, lhs, solution, rhs, bcs=None):
self.solution = solution
self._init_lhs(lhs, bcs)
self._init_rhs(rhs, bcs)
self._apply_bcs(bcs)
self._linear_solver = "default"
self.monitor = None
@overload(LinearProblemWrapper, Function.Type())
def __init__(self, problem_wrapper, solution):
self.__init__(problem_wrapper.matrix_eval(), solution, problem_wrapper.vector_eval(), problem_wrapper.bc_eval())
self.monitor = problem_wrapper.monitor
@overload(Form, (list_of(DirichletBC), ProductOutputDirichletBC, dict_of(str, list_of(DirichletBC)),
dict_of(str, ProductOutputDirichletBC), None))
def _init_lhs(self, lhs, bcs):
self.lhs = assemble(lhs, keep_diagonal=True)
@overload(ParametrizedTensorFactory, (list_of(DirichletBC), ProductOutputDirichletBC,
dict_of(str, list_of(DirichletBC)), dict_of(str, ProductOutputDirichletBC),
None))
def _init_lhs(self, lhs, bcs):
self.lhs = evaluate(lhs)
@overload(Matrix.Type(), None)
def _init_lhs(self, lhs, bcs):
self.lhs = lhs
@overload(Matrix.Type(), (list_of(DirichletBC), ProductOutputDirichletBC, dict_of(str, list_of(DirichletBC)),
dict_of(str, ProductOutputDirichletBC)))
def _init_lhs(self, lhs, bcs):
# Create a copy of lhs, in order not to change
# the original references when applying bcs
self.lhs = lhs.copy()
@overload(Form, (list_of(DirichletBC), ProductOutputDirichletBC, dict_of(str, list_of(DirichletBC)),
dict_of(str, ProductOutputDirichletBC), None))
def _init_rhs(self, rhs, bcs):
self.rhs = assemble(rhs)
@overload(ParametrizedTensorFactory, (list_of(DirichletBC), ProductOutputDirichletBC,
dict_of(str, list_of(DirichletBC)), dict_of(str, ProductOutputDirichletBC),
None))
def _init_rhs(self, rhs, bcs):
self.rhs = evaluate(rhs)
@overload(Vector.Type(), None)
def _init_rhs(self, rhs, bcs):
self.rhs = rhs
@overload(Vector.Type(), (list_of(DirichletBC), ProductOutputDirichletBC, dict_of(str, list_of(DirichletBC)),
dict_of(str, ProductOutputDirichletBC)))
def _init_rhs(self, rhs, bcs):
# Create a copy of rhs, in order not to change
# the original references when applying bcs
self.rhs = rhs.copy()
@overload(None)
def _apply_bcs(self, bcs):
pass
@overload((list_of(DirichletBC), ProductOutputDirichletBC))
def _apply_bcs(self, bcs):
for bc in bcs:
bc.apply(self.lhs, self.rhs)
@overload((dict_of(str, list_of(DirichletBC)), dict_of(str, ProductOutputDirichletBC)))
def _apply_bcs(self, bcs):
for key in bcs:
for bc in bcs[key]:
bc.apply(self.lhs, self.rhs)
def set_parameters(self, parameters):
assert len(parameters) in (0, 1)
if len(parameters) == 1:
assert "linear_solver" in parameters
self._linear_solver = parameters.get("linear_solver", "default")
def solve(self):
solver = PETScLUSolver(self._linear_solver)
solver.solve(self.lhs, self.solution.vector(), self.rhs)
if self.monitor is not None:
self.monitor(self.solution)
# SPDX-License-Identifier: LGPL-3.0-or-later
from ufl import Form
from dolfin import assemble, DirichletBC, PETScLUSolver
from rbnics.backends.abstract import LinearSolver as AbstractLinearSolver, LinearProblemWrapper
from rbnics.backends.dolfin.evaluate import evaluate
from rbnics.backends.dolfin.function import Function
from rbnics.backends.dolfin.matrix import Matrix
from rbnics.backends.dolfin.parametrized_tensor_factory import ParametrizedTensorFactory
from rbnics.backends.dolfin.vector import Vector
from rbnics.backends.dolfin.wrapping.dirichlet_bc import ProductOutputDirichletBC
from rbnics.utils.decorators import BackendFor, dict_of, list_of, overload
@BackendFor("dolfin", inputs=((Form, Matrix.Type(), ParametrizedTensorFactory, LinearProblemWrapper),
Function.Type(), (Form, ParametrizedTensorFactory, Vector.Type(), None),
(list_of(DirichletBC), ProductOutputDirichletBC, dict_of(str, list_of(DirichletBC)),
dict_of(str, ProductOutputDirichletBC), None)))
class LinearSolver(AbstractLinearSolver):
@overload((Form, Matrix.Type(), ParametrizedTensorFactory),
Function.Type(), (Form, ParametrizedTensorFactory, Vector.Type()),
(list_of(DirichletBC), ProductOutputDirichletBC, dict_of(str, list_of(DirichletBC)),
dict_of(str, ProductOutputDirichletBC), None))
def __init__(self, lhs, solution, rhs, bcs=None):
self.solution = solution
self._init_lhs(lhs, bcs)
self._init_rhs(rhs, bcs)
self._apply_bcs(bcs)
self._linear_solver = "default"
self.monitor = None
def _init_rhs(self, rhs: Vector.Type(), bcs: None):
self.rhs = rhs
def _assign(object_to: (Matrix.Type(), Vector.Type()),
object_from: (Matrix.Type(), Vector.Type())):
if object_from is not object_to:
to_petsc4py(object_from).copy(
to_petsc4py(object_to),
to_petsc4py(object_to).Structure.SAME_NONZERO_PATTERN)
# GNU Lesser General Public License for more details.
#
# You should have received a copy of the GNU Lesser General Public License
# along with RBniCS. If not, see <http://www.gnu.org/licenses/>.
#
from ufl import Form
from dolfin import assemble, DirichletBC
from rbnics.backends.abstract import AffineExpansionStorage as AbstractAffineExpansionStorage
from rbnics.backends.dolfin.matrix import Matrix
from rbnics.backends.dolfin.vector import Vector
from rbnics.backends.dolfin.function import Function
from rbnics.utils.decorators import backend_for, list_of, overload, tuple_of
# Generic backend
@backend_for("dolfin", inputs=((tuple_of(list_of(DirichletBC)), tuple_of(Form), tuple_of(Function.Type()), tuple_of(Matrix.Type()), tuple_of(Vector.Type()), tuple_of((Form, Matrix.Type())), tuple_of((Form, Vector.Type()))), ))
def AffineExpansionStorage(args):
return _AffineExpansionStorage(args)
# Base implementation
class AffineExpansionStorage_Base(AbstractAffineExpansionStorage):
def __init__(self, args):
self._content = None
def __getitem__(self, key):
return self._content[key]
def __iter__(self):
return self._content.__iter__()
def __len__(self):
def _evaluate(expression: (Matrix.Type(), Vector.Type(), TensorsList),
at: ReducedMesh, **kwargs):
assert len(kwargs) == 0
return evaluate_base(expression, at)
function_extend_or_restrict, function_from_ufl_operators,
function_save, get_function_subspace, to_petsc4py)
from rbnics.backends.dolfin.wrapping.tensor_save import basic_tensor_save
from rbnics.utils.decorators import backend_for, ModuleWrapper, overload
from rbnics.utils.io import Folders
backend = ModuleWrapper(Function, Matrix, Vector)
wrapping_for_wrapping = ModuleWrapper(build_dof_map_writer_mapping, form_argument_space, to_petsc4py,
form_name=form_name)
tensor_save = basic_tensor_save(backend, wrapping_for_wrapping)
wrapping = ModuleWrapper(function_extend_or_restrict, function_save, get_function_subspace, tensor_save=tensor_save)
export_base = basic_export(backend, wrapping)
# Export a solution to file
@backend_for("dolfin", inputs=((Function.Type(), Matrix.Type(), Operator, Vector.Type()), (Folders.Folder, str),
str, (int, None), (int, str, None)))
def export(solution, directory, filename, suffix=None, component=None):
_export(solution, directory, filename, suffix, component)
@overload
def _export(
solution: (
Function.Type(),
Matrix.Type(),
Vector.Type()
),
directory: (
Folders.Folder,
str
