feectools 0.1.11__tar.gz → 0.3.0__tar.gz

{feectools-0.1.11/feectools.egg-info → feectools-0.3.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: feectools
-Version: 0.1.11
+Version: 0.3.0
 Summary: Slimmed-down fork of Psydac (https://github.com/pyccel/psydac) with less functionality and fewer dependencies.
 Author-email: Psydac development team <psydac@googlegroups.com>
 Maintainer-email: Stefan Possanner <stefan.possanner@ipp.mpg.de>, Max Lindqvist <max.lindqvist@ipp.mpg.de>, Yaman Güçlü <yaman.guclu@gmail.com>, Martin Campos Pinto <martin.campos-pinto@ipp.mpg.de>, Ahmed Ratnani <ratnaniahmed@gmail.com>

{feectools-0.1.11 → feectools-0.3.0}/feectools/api/essential_bc.py

@@ -72,6 +72,9 @@ def apply_essential_bc_stencil(a, *, axis, ext, order, identity=False):
     if isinstance(a, StencilVector):
         V = a.space
         n = V.ndim
+        # Boundary entries may be ghost entries of neighbouring processes, which
+        # all call this function: their ghost regions are no longer up to date.
+        a.ghost_regions_in_sync = False
     elif isinstance(a, StencilMatrix):
         V = a.codomain
         n = V.ndim * 2

{feectools-0.1.11 → feectools-0.3.0}/feectools/ddm/cart.py

@@ -1,5 +1,6 @@
 # coding: utf-8
 
+import copy
 import os
 import numpy as np
 import cunumpy as xp
@@ -409,42 +410,100 @@ class DomainDecomposition:
     def refine(self, ncells, global_element_starts, global_element_ends):
         """ Create the new Cartesian decomposition of the refined domain.
 
+        The process topology (and its communicators) is shared with ``self``.
+
         Parameters
         ----------
         ncells : list or tuple of int
             Number of cells of refined space.
 
-        global_starts: list of list of int
-            The starts of the coefficients for every process along each direction.
+        global_element_starts : list of list of int
+            The element starts for every process along each direction.
 
-        global_ends: list of list of int
-            The ends of the coefficients for every process along each direction.
+        global_element_ends : list of list of int
+            The element ends for every process along each direction.
 
         Returns
         -------
-        domain : CartDecomposition
-            Cartesian decomposition of the refined domain.
+        domain : DomainDecomposition
+            Domain decomposition of the refined domain.
         """
 
         # Check input arguments
         assert len( ncells ) == len( self.ncells )
         assert all(nc>=snc for nc, snc in zip(ncells, self.ncells))
 
-        domain         = DomainDecomposition(self.ncells, self.periods, comm=self.comm,
-                                            global_comm=self.global_comm, num_threads=self.num_threads,
-                                            size=self.size)
-        domain._ncells = tuple ( ncells )
+        return self._with_element_partition(ncells, global_element_starts, global_element_ends)
+
+    def coarsen(self, factors):
+        """ Create the Cartesian decomposition of a coarsened domain, aligned with ``self``.
+
+        Along axis ``i`` every ``factors[i]`` consecutive cells are merged into one coarse cell.
+        Each process owns the coarse cells covering exactly its fine cells, so the process
+        topology (and its communicators) is shared with ``self``. This requires that the
+        element starts and ends+1 of every process are divisible by ``factors[i]``.
+
+        Parameters
+        ----------
+        factors : list or tuple of int
+            Coarsening factor (>= 1) along each direction.
+
+        Returns
+        -------
+        domain : DomainDecomposition
+            Domain decomposition of the coarse domain.
+        """
+
+        assert len( factors ) == self.ndim
+        assert all( isinstance(f, (int, np.integer)) and f >= 1 for f in factors )
+
+        ncells = []
+        global_element_starts = []
+        global_element_ends   = []
+        for axis, f in enumerate(factors):
+            gs = xp.asarray(self._global_element_starts[axis])
+            ge = xp.asarray(self._global_element_ends  [axis])
+            if self._ncells[axis] % f != 0 or xp.any(gs % f != 0) or xp.any((ge + 1) % f != 0):
+                raise ValueError(
+                    f"Cannot coarsen axis {axis} by a factor {f}: ncells={self._ncells[axis]}, "
+                    f"element starts={gs.tolist()}, ends={ge.tolist()} are not all aligned."
+                )
+            ncells.append(self._ncells[axis] // f)
+            global_element_starts.append(xp.array(gs // f))
+            global_element_ends  .append(xp.array((ge + 1) // f - 1))
+
+        return self._with_element_partition(ncells, global_element_starts, global_element_ends)
+
+    def _with_element_partition(self, ncells, global_element_starts, global_element_ends):
+        """ Return a copy of ``self`` with the same process topology but a new element partition.
+
+        Communicators are shared (not duplicated), hence this method is not collective.
+        """
+
+        assert len( ncells ) == self.ndim
+        for axis in range(self.ndim):
+            gs = xp.asarray(global_element_starts[axis])
+            ge = xp.asarray(global_element_ends  [axis])
+            assert len(gs) == len(ge) == self._nprocs[axis], \
+                f"Axis {axis}: need one block per process ({self._nprocs[axis]}), got {len(gs)}."
+            assert gs[0] == 0 and ge[-1] == ncells[axis] - 1, \
+                f"Axis {axis}: blocks must cover [0, {ncells[axis] - 1}]."
+            assert xp.all(ge >= gs), f"Axis {axis}: empty blocks are not allowed."
+            assert xp.all(gs[1:] == ge[:-1] + 1), f"Axis {axis}: blocks must be contiguous."
+
+        domain = copy.copy(self)
+        domain._ncells = tuple( int(n) for n in ncells )
 
         # Store arrays with all the starts and ends along each direction for every process
-        domain._global_element_starts = tuple(global_element_starts)
-        domain._global_element_ends   = tuple(global_element_ends)
+        domain._global_element_starts = list(global_element_starts)
+        domain._global_element_ends   = list(global_element_ends)
         if self.is_comm_null:return domain
 
         # Start/end values of global indices (without ghost regions)
         domain._starts = tuple( domain._global_element_starts[axis][c] for axis,c in zip(range(self._ndims), self._coords) )
         domain._ends   = tuple( domain._global_element_ends  [axis][c] for axis,c in zip(range(self._ndims), self._coords) )
 
-        domain._local_ncells = tuple(e-s+1 for s,e in zip(self._starts, self._ends))
+        domain._local_ncells = tuple(e-s+1 for s,e in zip(domain._starts, domain._ends))
         return domain
 
 #==================================================================================

feectools-0.3.0/feectools/ddm/tests/test_coarsen.py

@@ -0,0 +1,67 @@
+import numpy as np
+import pytest
+
+from feectools.ddm.cart import DomainDecomposition
+from feectools.ddm.mpi import mpi as MPI
+
+
+def _comm(parallel):
+    return MPI.COMM_WORLD if parallel else None
+
+
+def _as_list(arrs):
+    return [np.asarray(a).tolist() for a in arrs]
+
+
+@pytest.mark.parametrize("parallel", [False, pytest.param(True, marks=pytest.mark.mpi)])
+@pytest.mark.parametrize("periods", [(True, True, True), (False, True, False)])
+def test_coarsen(parallel, periods):
+    comm = _comm(parallel)
+    fine = DomainDecomposition([32, 16, 8], periods, comm=comm)
+    coarse = fine.coarsen([2, 4, 1])
+
+    assert coarse.ncells == (16, 4, 8)
+    assert coarse.periods == fine.periods
+    assert tuple(coarse.nprocs) == tuple(fine.nprocs)
+    assert coarse.comm_cart is fine.comm_cart
+    assert tuple(coarse.coords) == tuple(fine.coords)
+
+    # every process owns exactly the coarse cells covering its fine cells
+    for axis, f in enumerate([2, 4, 1]):
+        assert coarse.starts[axis] * f == fine.starts[axis]
+        assert (coarse.ends[axis] + 1) * f == fine.ends[axis] + 1
+        assert coarse.local_ncells[axis] * f == fine.local_ncells[axis]
+
+    # the original object is untouched
+    assert fine.ncells == (32, 16, 8)
+
+    # refining back gives the original partition
+    back = coarse.refine(
+        fine.ncells,
+        [np.asarray(s) * f for s, f in zip(coarse.global_element_starts, [2, 4, 1])],
+        [(np.asarray(e) + 1) * f - 1 for e, f in zip(coarse.global_element_ends, [2, 4, 1])],
+    )
+    assert back.ncells == fine.ncells
+    assert _as_list(back.global_element_starts) == _as_list(fine.global_element_starts)
+    assert _as_list(back.global_element_ends) == _as_list(fine.global_element_ends)
+    assert back.starts == fine.starts
+    assert back.ends == fine.ends
+    assert back.local_ncells == fine.local_ncells
+
+
+@pytest.mark.parametrize("parallel", [False, pytest.param(True, marks=pytest.mark.mpi)])
+def test_coarsen_repeated(parallel):
+    comm = _comm(parallel)
+    dd = DomainDecomposition([64, 32, 1], (True, True, True), comm=comm, mpi_dims_mask=[True, True, False])
+    for _ in range(3):
+        dd = dd.coarsen([2, 2, 1])
+    assert dd.ncells == (8, 4, 1)
+    assert dd.local_ncells[2] == 1
+
+
+def test_coarsen_misaligned():
+    dd = DomainDecomposition([6, 4, 4], (True, True, True))
+    with pytest.raises(ValueError):
+        dd.coarsen([4, 1, 1])
+    with pytest.raises(AssertionError):
+        dd.coarsen([0, 1, 1])

{feectools-0.1.11 → feectools-0.3.0}/feectools/linalg/basic.py

@@ -7,6 +7,7 @@ provides the fundamental classes for linear algebra operations.
 
 """
 
+import itertools
 from abc import ABC, abstractmethod
 from types import LambdaType 
 from inspect import signature
@@ -280,13 +281,143 @@ class LinearOperator(ABC):
         upon convertion to matrix.
         """
 
-    @abstractmethod
-    def tosparse(self):
-        """ Convert to a sparse matrix in any of the formats supported by scipy.sparse."""
+    def toarray(self, out=None, is_sparse=False, format='csr'):
+        """
+        Assemble the global matrix of the linear operator column by column.
 
-    @abstractmethod
-    def toarray(self):
-        """ Convert to Numpy 2D array. """
+        Column j is computed as ``self.dot(e_j)``, where e_j is the j-th
+        canonical basis vector of the domain, in the global numbering of
+        ``Vector.toarray()`` (C-ordered within each StencilVector, blocks
+        concatenated). Since only ``dot`` is used, this works for any linear
+        operator, including matrix-free ones, whose domain is a
+        StencilVectorSpace or a (possibly nested) BlockVectorSpace thereof.
+
+        The cost is one call to ``dot`` per global degree of freedom of the
+        domain, hence this default is meant for testing and small problems.
+        Subclasses with an explicit matrix representation should override it.
+
+        In parallel, all ranks call ``dot`` collectively, each rank fills the
+        rows it owns, and every rank receives the full matrix.
+
+        Parameters
+        ----------
+        out : numpy.ndarray, optional
+            Dense array of shape ``self.shape`` into which the result is
+            written in place. Must be None if ``is_sparse`` is True.
+
+        is_sparse : bool
+            If True, return a scipy.sparse matrix, otherwise a dense
+            numpy.ndarray.
+
+        format : str
+            Sparse format, only used if ``is_sparse`` is True: one of 'csr'
+            (default), 'csc', 'bsr', 'lil', 'dok', 'coo' or 'dia'.
+
+        Returns
+        -------
+        numpy.ndarray or scipy.sparse matrix
+            The global matrix of shape ``self.shape``, identical on all ranks.
+        """
+        from feectools.linalg.block import BlockVectorSpace
+        from feectools.linalg.stencil import StencilVectorSpace
+
+        # Flatten the (possibly nested) block structure of the domain
+        def leaves(w):
+            if isinstance(w.space, StencilVectorSpace):
+                return [w]
+            elif isinstance(w.space, BlockVectorSpace):
+                return [lw for b in w.blocks for lw in leaves(b)]
+            else:
+                raise TypeError(f'{type(self).__name__}.toarray() requires a domain made of '
+                                f'StencilVectorSpaces, not {type(w.space).__name__}.')
+
+        e_j = self.domain.zeros()
+        Ae_j = self.codomain.zeros()
+        e_j_leaves = leaves(e_j)
+        offsets = xp.cumsum([0] + [lw.space.dimension for lw in e_j_leaves[:-1]])
+
+        if is_sparse:
+            assert out is None, 'out must be None if is_sparse is True.'
+            assert format in ('csr', 'csc', 'bsr', 'lil', 'dok', 'coo', 'dia'), \
+                f'Unknown sparse format {format!r}.'
+            rows, cols, data = [], [], []
+        elif out is None:
+            out = xp.zeros(self.shape, dtype=self.dtype)
+        else:
+            assert isinstance(out, xp.ndarray)
+            assert out.shape == self.shape, f'out has shape {out.shape}, expected {self.shape}.'
+
+        # Index ranges owned by each rank, for every leaf
+        bounds = [(lw.starts, lw.ends) for lw in e_j_leaves]
+        if e_j_leaves[0].space.parallel:
+            comm = e_j_leaves[0].space.cart.comm
+            rank = comm.Get_rank()
+            all_bounds = comm.allgather(bounds)
+        else:
+            comm = None
+            rank = 0
+            all_bounds = [bounds]
+
+        # All ranks loop over all columns, since dot() is collective;
+        # only the owner of index i sets the entry of e_j to one.
+        for owner, owner_bounds in enumerate(all_bounds):
+            for lw, offset, (starts, ends) in zip(e_j_leaves, offsets, owner_bounds):
+                for i in itertools.product(*(range(s, e + 1) for s, e in zip(starts, ends))):
+                    if rank == owner:
+                        lw[i] = 1
+                    lw.update_ghost_regions()
+                    self.dot(e_j, out=Ae_j)
+                    if rank == owner:
+                        lw[i] = 0
+
+                    # Global vector with nonzeros only in the rows owned by this rank
+                    col_j = Ae_j.toarray()
+                    j = offset + xp.ravel_multi_index(i, lw.space.npts)
+                    if is_sparse:
+                        nz = xp.flatnonzero(col_j)
+                        rows.append(nz)
+                        cols.append(xp.full(nz.size, j))
+                        data.append(col_j[nz])
+                    else:
+                        out[:, j] = col_j
+                # Clear the ghost regions of the last entry set to one
+                lw.update_ghost_regions()
+
+        if not is_sparse:
+            if comm is not None:
+                from feectools.ddm.mpi import mpi as MPI
+                comm.Allreduce(MPI.IN_PLACE, out, op=MPI.SUM)
+            return out
+
+        rows = xp.concatenate(rows) if rows else xp.zeros(0, dtype=int)
+        cols = xp.concatenate(cols) if cols else xp.zeros(0, dtype=int)
+        data = xp.concatenate(data) if data else xp.zeros(0, dtype=self.dtype)
+        if comm is not None:
+            rows = xp.concatenate(comm.allgather(rows))
+            cols = xp.concatenate(comm.allgather(cols))
+            data = xp.concatenate(comm.allgather(data))
+
+        return coo_matrix((data, (rows, cols)), shape=self.shape).asformat(format)
+
+    def tosparse(self, format='csr'):
+        """
+        Assemble the global matrix of the linear operator as a scipy.sparse matrix.
+
+        Default implementation calling the generic ``LinearOperator.toarray``
+        with ``is_sparse=True``; see there for cost and parallel behavior.
+        Subclasses with an explicit matrix representation should override it.
+
+        Parameters
+        ----------
+        format : str
+            One of 'csr' (default), 'csc', 'bsr', 'lil', 'dok', 'coo' or 'dia'.
+
+        Returns
+        -------
+        scipy.sparse matrix
+            The global matrix of shape ``self.shape``, identical on all ranks.
+        """
+        return LinearOperator.toarray(self, is_sparse=True, format=format)
 
     @abstractmethod
     def dot(self, v, out=None):
@@ -977,9 +1108,6 @@ class ComposedLinearOperator(LinearOperator):
     def dtype(self):
         return None
 
-    def toarray(self):
-        raise NotImplementedError('toarray() is not defined for ComposedLinearOperators.')
-
     def tosparse(self):
         mats = [M.tosparse() for M in self._multiplicants]
         M = mats[0]
@@ -1084,12 +1212,6 @@ class PowerLinearOperator(LinearOperator):
         """ Returns the power to which the operator is raised. """
         return self._factorial
 
-    def toarray(self):
-        raise NotImplementedError('toarray() is not defined for PowerLinearOperators.')
-
-    def tosparse(self):
-        raise NotImplementedError('tosparse() is not defined for PowerLinearOperators.')
-
     def transpose(self, conjugate=False):
         return PowerLinearOperator(domain=self.codomain, codomain=self.domain, A=self._operator.transpose(conjugate=conjugate), n=self._factorial)
 
@@ -1207,12 +1329,6 @@ class InverseLinearOperator(LinearOperator):
             elif key == 'verbose':
                 assert isinstance(value, bool), "verbose must be a bool"
 
-    def toarray(self):
-        raise NotImplementedError('toarray() is not defined for InverseLinearOperators.')
-
-    def tosparse(self):
-        raise NotImplementedError('tosparse() is not defined for InverseLinearOperators.')
-
     def get_info(self):
         """ Returns the previous convergence information. """
         return self._info
@@ -1366,12 +1482,6 @@ class MatrixFreeLinearOperator(LinearOperator):
                     
         return out
         
-    def toarray(self):
-        raise NotImplementedError('toarray() is not defined for MatrixFreeLinearOperator.')
-
-    def tosparse(self):
-        raise NotImplementedError('tosparse() is not defined for MatrixFreeLinearOperator.')
-    
     def transpose(self, conjugate=False):
         if self._dot_transpose is None:
             raise NotImplementedError('no transpose dot method was given -- cannot create the transpose operator')

{feectools-0.1.11 → feectools-0.3.0}/feectools/linalg/fft.py

@@ -21,12 +21,6 @@ class DistributedFFTBase(LinearOperator):
         The function at position i is applied to the i-th tensor direction.
         If only a single callable is given, it is used for all directions.
     """
-    def toarray(self):
-        raise NotImplementedError('toarray() is not defined for DistributedFFTBase.')
-
-    def tosparse(self):
-        raise NotImplementedError('tosparse() is not defined for DistributedFFTBase.')
-
     # Possible additions for the future:
     # * split off the LinearSolver class when used with the space ndarray (as used in the KroneckerLinearSolver),
     #   and make it state if it works in-place (or if it needs temporary memory), and what its optimal

{feectools-0.1.11 → feectools-0.3.0}/feectools/linalg/kron.py

@@ -528,12 +528,6 @@ class KroneckerLinearSolver(LinearOperator):
     @property
     def dtype(self):
         return None
-    
-    def toarray(self):
-        raise NotImplementedError('toarray() is not defined for KroneckerLinearSolvers.')
-    
-    def tosparse(self):
-        raise NotImplementedError('tosparse() is not defined for KroneckerLinearSolvers.')
 
     def transpose(self, conjugate=False):
         new_domain = self._codomain

{feectools-0.1.11 → feectools-0.3.0}/feectools/linalg/stencil.py

@@ -332,7 +332,7 @@ class StencilVectorSpace(VectorSpace):
                 import cupy as cp
                 y._interface_data[axis, ext][:] = cp.asarray(y_int_np)
 
-        x._sync = x._sync and y._sync
+        y._sync = x._sync and y._sync
 
     #--------------------------------------
     # Other properties/methods

feectools-0.3.0/feectools/linalg/tests/test_axpy_ghost_sync.py

@@ -0,0 +1,31 @@
+from feectools.ddm.cart import DomainDecomposition, CartDecomposition
+from feectools.linalg.stencil import StencilVectorSpace
+
+
+def _space():
+    dd = DomainDecomposition([8, 8], [True, True])
+    cart = CartDecomposition(dd, [8, 8], [[0], [0]], [[7], [7]], [2, 2], [1, 1])
+    return StencilVectorSpace(cart)
+
+
+def test_axpy_ghost_sync():
+    """y += a*x must mark y out of sync if x is, and must not modify the flag of x."""
+    V = _space()
+    x, y = V.zeros(), V.zeros()
+
+    x.ghost_regions_in_sync = False
+    y.ghost_regions_in_sync = True
+    y.mul_iadd(2.0, x)
+    assert not y.ghost_regions_in_sync
+    assert not x.ghost_regions_in_sync
+
+    x.ghost_regions_in_sync = True
+    y.ghost_regions_in_sync = False
+    y.mul_iadd(2.0, x)
+    assert not y.ghost_regions_in_sync
+    assert x.ghost_regions_in_sync
+
+    x.ghost_regions_in_sync = True
+    y.ghost_regions_in_sync = True
+    y.mul_iadd(2.0, x)
+    assert y.ghost_regions_in_sync

feectools-0.3.0/feectools/linalg/tests/test_toarray.py

@@ -0,0 +1,162 @@
+# -*- coding: UTF-8 -*-
+#
+# Tests for the generic LinearOperator.toarray() and LinearOperator.tosparse(),
+# which assemble the matrix of any linear operator from its dot() method.
+#
+import pytest
+import cunumpy as xp
+
+from feectools.ddm.mpi import mpi as MPI
+from feectools.ddm.cart import DomainDecomposition, CartDecomposition
+from feectools.linalg.basic import LinearOperator, MatrixFreeLinearOperator, IdentityOperator
+from feectools.linalg.stencil import StencilVectorSpace, StencilMatrix
+from feectools.linalg.block import BlockVectorSpace, BlockLinearOperator
+from feectools.linalg.solvers import inverse
+from feectools.linalg.tests.test_block import compute_global_starts_ends
+
+SPARSE_FORMATS = ['csr', 'csc', 'bsr', 'lil', 'dok', 'coo', 'dia']
+
+#===============================================================================
+# HELPERS
+#===============================================================================
+def get_space(npts, pads, periods, comm=None):
+    D = DomainDecomposition(npts, periods=periods, comm=comm)
+    global_starts, global_ends = compute_global_starts_ends(D, npts)
+    C = CartDecomposition(D, npts, global_starts, global_ends, pads=pads, shifts=[1] * len(npts))
+    return StencilVectorSpace(C)
+
+def get_random_matrix(V, W, seed):
+    """ Random StencilMatrix, identical on all ranks in the global numbering. """
+    rng = xp.random.default_rng(seed)
+    M = StencilMatrix(V, W)
+    # Draw the full (global) band and keep the local part, so the matrix
+    # does not depend on the domain decomposition.
+    shape = tuple(W.npts) + M._data.shape[W.ndim:]
+    band = rng.random(shape) - 0.5
+    idx = tuple(slice(s, e + 1) for s, e in zip(W.starts, W.ends))
+    M[idx] = band[idx]
+    M.remove_spurious_entries()
+    return M
+
+def matrix_free(A):
+    """ Hide the explicit matrix of A, so that the generic toarray() is used. """
+    return MatrixFreeLinearOperator(A.domain, A.codomain, lambda v, out=None: A.dot(v, out=out))
+
+def reference(A):
+    """ Global dense matrix of A, from its own (row-local in parallel) tosparse(). """
+    local = A.tosparse().toarray()
+    comm = A.domain.cart.comm if isinstance(A.domain, StencilVectorSpace) else A.domain.spaces[0].cart.comm
+    if A.domain.parallel:
+        glob = xp.zeros_like(local)
+        comm.Allreduce(local, glob, op=MPI.SUM)
+        return glob
+    return local
+
+def check_all(O, ref):
+    """ Check dense, in-place and all sparse outputs of the generic toarray()/tosparse(). """
+    assert xp.allclose(O.toarray(), ref)
+
+    out = xp.full(O.shape, 7.0)
+    res = O.toarray(out=out)
+    assert res is out
+    assert xp.allclose(out, ref)
+
+    for fmt in SPARSE_FORMATS:
+        S = O.toarray(is_sparse=True, format=fmt)
+        assert S.format == fmt
+        assert S.shape == O.shape
+        assert xp.allclose(S.toarray(), ref)
+
+    S = O.tosparse()
+    assert S.format == 'csr'
+    assert xp.allclose(S.toarray(), ref)
+    assert O.tosparse('csc').format == 'csc'
+
+def stencil_case(periods, comm):
+    V = get_space([6, 5], [2, 1], periods, comm=comm)
+    return get_random_matrix(V, V, seed=0)
+
+def block_case(periods, comm):
+    """ 2x2 block operator, and a nested block operator [[B, 0], [0, A]]. """
+    V = get_space([6, 5], [2, 1], periods, comm=comm)
+    A = get_random_matrix(V, V, seed=0)
+    A01 = get_random_matrix(V, V, seed=1)
+    VV = BlockVectorSpace(V, V)
+    B = BlockLinearOperator(VV, VV, blocks=[[A, A01], [None, A]])
+    Vn = BlockVectorSpace(VV, V)
+    N = BlockLinearOperator(Vn, Vn, blocks=[[B, None], [None, A]])
+    return A, B, N
+
+#===============================================================================
+# SERIAL TESTS
+#===============================================================================
+@pytest.mark.parametrize('periods', [[False, False], [True, False], [True, True]])
+def test_toarray_stencil(periods):
+    A = stencil_case(periods, comm=None)
+    check_all(matrix_free(A), reference(A))
+
+@pytest.mark.parametrize('periods', [[False, False], [True, True]])
+def test_toarray_block(periods):
+    A, B, N = block_case(periods, comm=None)
+    refA, refB = reference(A), reference(B)
+    check_all(matrix_free(B), refB)
+
+    Z = xp.zeros((refB.shape[0], refA.shape[1]))
+    refN = xp.block([[refB, Z], [Z.T, refA]])
+    check_all(matrix_free(N), refN)
+
+def test_toarray_composite_operators():
+    """ Operators that used to raise NotImplementedError in toarray()/tosparse(). """
+    A = stencil_case([True, False], comm=None)
+    ref = reference(A)
+    V = A.domain
+
+    assert xp.allclose((A @ A).toarray(), ref @ ref)
+    assert xp.allclose((A ** 2).toarray(), ref @ ref)
+    assert xp.allclose((A ** 2).tosparse().toarray(), ref @ ref)
+
+    M = A.T @ A + IdentityOperator(V)
+    refM = ref.T @ ref + xp.eye(ref.shape[0])
+    Minv = inverse(M, 'cg', tol=1e-13, maxiter=1000)
+    assert xp.allclose(Minv.toarray(), xp.linalg.inv(refM), atol=1e-8)
+
+def test_toarray_invalid_input():
+    O = matrix_free(stencil_case([False, False], comm=None))
+    with pytest.raises(AssertionError):
+        O.toarray(out=xp.zeros((O.shape[0], O.shape[1] + 1)))
+    with pytest.raises(AssertionError):
+        O.toarray(out=xp.zeros(O.shape), is_sparse=True)
+    with pytest.raises(AssertionError):
+        O.toarray(is_sparse=True, format='xyz')
+
+#===============================================================================
+# PARALLEL TESTS
+#===============================================================================
+@pytest.mark.parametrize('periods', [[False, False], [True, False], [True, True]])
+@pytest.mark.parallel
+def test_toarray_stencil_parallel(periods):
+    A = stencil_case(periods, comm=MPI.COMM_WORLD)
+    check_all(matrix_free(A), reference(A))
+
+@pytest.mark.parametrize('periods', [[False, False], [True, True]])
+@pytest.mark.parallel
+def test_toarray_block_parallel(periods):
+    A, B, N = block_case(periods, comm=MPI.COMM_WORLD)
+    refA, refB = reference(A), reference(B)
+    check_all(matrix_free(B), refB)
+
+    Z = xp.zeros((refB.shape[0], refA.shape[1]))
+    refN = xp.block([[refB, Z], [Z.T, refA]])
+    check_all(matrix_free(N), refN)
+
+@pytest.mark.parallel
+def test_toarray_composite_operators_parallel():
+    A = stencil_case([True, False], comm=MPI.COMM_WORLD)
+    ref = reference(A)
+    assert xp.allclose((A @ A).toarray(), ref @ ref)
+    assert xp.allclose((A ** 2).tosparse().toarray(), ref @ ref)
+
+#===============================================================================
+if __name__ == '__main__':
+    import sys
+    pytest.main(sys.argv)

{feectools-0.1.11 → feectools-0.3.0/feectools.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: feectools
-Version: 0.1.11
+Version: 0.3.0
 Summary: Slimmed-down fork of Psydac (https://github.com/pyccel/psydac) with less functionality and fewer dependencies.
 Author-email: Psydac development team <psydac@googlegroups.com>
 Maintainer-email: Stefan Possanner <stefan.possanner@ipp.mpg.de>, Max Lindqvist <max.lindqvist@ipp.mpg.de>, Yaman Güçlü <yaman.guclu@gmail.com>, Martin Campos Pinto <martin.campos-pinto@ipp.mpg.de>, Ahmed Ratnani <ratnaniahmed@gmail.com>

{feectools-0.1.11 → feectools-0.3.0}/feectools.egg-info/SOURCES.txt

@@ -41,6 +41,7 @@ feectools/ddm/tests/__init__.py
 feectools/ddm/tests/test_cart_1d.py
 feectools/ddm/tests/test_cart_2d.py
 feectools/ddm/tests/test_cart_3d.py
+feectools/ddm/tests/test_coarsen.py
 feectools/ddm/tests/test_multicart_2d.py
 feectools/ddm/tests/test_partition.py
 feectools/feec/__init__.py
@@ -91,6 +92,7 @@ feectools/linalg/kernels/stencil2IJV_kernels.py
 feectools/linalg/kernels/stencil2coo_kernels.py
 feectools/linalg/kernels/transpose_kernels.py
 feectools/linalg/tests/__init__.py
+feectools/linalg/tests/test_axpy_ghost_sync.py
 feectools/linalg/tests/test_block.py
 feectools/linalg/tests/test_fft.py
 feectools/linalg/tests/test_kron_stencil_matrix.py
@@ -100,6 +102,7 @@ feectools/linalg/tests/test_solvers.py
 feectools/linalg/tests/test_stencil_interface_matrix.py
 feectools/linalg/tests/test_stencil_vector.py
 feectools/linalg/tests/test_stencil_vector_space.py
+feectools/linalg/tests/test_toarray.py
 feectools/linalg/tests/utilities.py
 feectools/utilities/__init__.py
 feectools/utilities/quadratures.py

{feectools-0.1.11 → feectools-0.3.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name            = "feectools"
-version         = "0.1.11"
+version         = "0.3.0"
 description     = "Slimmed-down fork of Psydac (https://github.com/pyccel/psydac) with less functionality and fewer dependencies."
 readme          = "README.md"
 requires-python = ">= 3.10"