"""Unit tests for the Vector interface""" # Copyright (C) 2011-2014 Garth N. Wells # # This file is part of DOLFIN. # # DOLFIN is free software: you can redistribute it and/or modify # it under the terms of the GNU Lesser General Public License as published by # the Free Software Foundation, either version 3 of the License, or # (at your option) any later version. # # DOLFIN 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 Lesser General Public License for more details. # # You should have received a copy of the GNU Lesser General Public License # along with DOLFIN. If not, see . # # Modified by Anders Logg 2011 import pytest import numpy from copy import copy from ctypes import c_long from dolfin import * from dolfin_utils.test import * # TODO: Use the fixture setup from matrix in a shared conftest.py when # we move tests to one flat folder. # Lists of backends supporting or not supporting GenericVector::data() # access data_backends = [] no_data_backends = ["PETSc", "Tpetra"] # Add serial only backends if MPI.size(MPI.comm_world) == 1: # TODO: What about "Dense" and "Sparse"? The sub_backend wasn't # used in the old test. data_backends += ["Eigen"] # Remove backends we haven't built with data_backends = list(filter(has_linear_algebra_backend, data_backends)) no_data_backends = list(filter(has_linear_algebra_backend, no_data_backends)) any_backends = data_backends + no_data_backends # Fixtures setting up and resetting the global linear algebra backend # for a list of backends data_backend = set_parameters_fixture("linear_algebra_backend", data_backends) no_data_backend = set_parameters_fixture("linear_algebra_backend", no_data_backends) any_backend = set_parameters_fixture("linear_algebra_backend", any_backends) class TestVectorForAnyBackend: def test_create_empty_vector(self, any_backend): v0 = Vector() info(v0) info(v0, True) assert v0.size() == 0 def test_create_vector(self, any_backend): n = 301 v1 = Vector(MPI.comm_world, n) assert v1.size() == n def test_copy_vector(self, any_backend): n = 301 v0 = Vector(MPI.comm_world, n) v1 = Vector(v0) assert v0.size() == n del v0 assert v1.size() == n def test_assign_and_copy_vector(self, any_backend): n = 301 v0 = Vector(MPI.comm_world, n) v0[:] = 1.0 assert v0.sum() == n v1 = Vector(v0) del v0 assert v1.sum() == n def test_zero(self, any_backend): v0 = Vector(MPI.comm_world, 301) v0.zero() assert v0.sum() == 0.0 def test_apply(self, any_backend): v0 = Vector(MPI.comm_world, 301) v0.apply("insert") v0.apply("add") def test_str(self, any_backend): v0 = Vector(MPI.comm_world, 13) tmp = v0.str(False) tmp = v0.str(True) def test_init_range(self, any_backend): n = 301 local_range = MPI.local_range(MPI.comm_world, n) v0 = Vector(MPI.comm_world) v0.init(local_range) assert v0.local_range() == local_range def test_size(self, any_backend): n = 301 v0 = Vector(MPI.comm_world, 301) assert v0.size() == n def test_local_size(self, any_backend): n = 301 local_range = MPI.local_range(MPI.comm_world, n) v0 = Vector(MPI.comm_world) v0.init(local_range) assert v0.local_size() == local_range[1] - local_range[0] def test_owns_index(self, any_backend): m, n = 301, 25 v0 = Vector(MPI.comm_world, m) local_range = v0.local_range() in_range = local_range[0] <= n < local_range[1] assert v0.owns_index(n) == in_range #def test_set(self, any_backend): #def test_add(self, any_backend): def test_get_local(self, any_backend): from numpy import empty n = 301 v0 = Vector(MPI.comm_world, n) data = v0.get_local() def test_set_local(self, any_backend): from numpy import zeros n = 301 v0 = Vector(MPI.comm_world, n) data = zeros((v0.local_size()), dtype='d') v0.set_local(data) data = zeros((v0.local_size()*2), dtype='d') def test_add_local(self, any_backend): from numpy import zeros n = 301 v0 = Vector(MPI.comm_world, n) data = zeros((v0.local_size()), dtype='d') v0.add_local(data) data = zeros((v0.local_size()*2), dtype='d') def test_gather(self, any_backend): # Gather not implemented in Eigen if any_backend == "Eigen" or any_backend == "Tpetra": return # Create distributed vector of local size 1 x = DefaultFactory().create_vector(MPI.comm_world) r = MPI.rank(x.mpi_comm()) x.init((r, r+1)) # Create local vector y = DefaultFactory().create_vector(MPI.comm_self) # Do the actual test across all rank permutations for target_rank in range(MPI.size(x.mpi_comm())): # Set nonzero value on single rank if r == target_rank: x[0] = 42.0 # Set using local index else: x[0] = 0.0 # Set using local index assert numpy.isclose(x.sum(), 42.0) # Gather (using global index) and check the result x.gather(y, numpy.array([target_rank], dtype=la_index_dtype())) assert numpy.isclose(y[0], 42.0) # NumPy array version out = x.gather(numpy.array([target_rank], dtype=la_index_dtype())) assert out.shape == (1,) and numpy.isclose(out[0], 42.0) # Test gather on zero out = x.gather_on_zero() if r == 0: expected = numpy.array([42.0 if i == target_rank else 0.0 for i in range(x.size())]) else: expected = numpy.array([]) assert out.shape == expected.shape and numpy.allclose(out, expected) # Test also the corner case of empty indices on one process if r == target_rank: out = x.gather(numpy.array([], dtype=la_index_dtype())) expected = numpy.array([]) else: out = x.gather(numpy.array([target_rank], dtype=la_index_dtype())) expected = numpy.array([42.0]) assert out.shape == expected.shape and numpy.allclose(out, expected) # Check that distributed gather vector is not accepted if MPI.size(MPI.comm_world) > 1: z = DefaultFactory().create_vector(MPI.comm_world) with pytest.raises(RuntimeError): x.gather(z, numpy.array([0], dtype=la_index_dtype())) # Check that gather vector of wrong size is not accepted z = DefaultFactory().create_vector(MPI.comm_self) z.init(3) with pytest.raises(RuntimeError): x.gather(z, numpy.array([0], dtype=la_index_dtype())) def test_axpy(self, any_backend): n = 301 v0 = Vector(MPI.comm_world, n) v0[:] = 1.0 v1 = Vector(v0) v0.axpy(2.0, v1) assert v0.sum() == 2*n + n def test_abs(self, any_backend): n = 301 v0 = Vector(MPI.comm_world, n) v0[:] = -1.0 v0.abs() assert v0.sum() == n def test_inner(self, any_backend): n = 301 v0 = Vector(MPI.comm_world, n) v0[:] = 2.0 v1 = Vector(MPI.comm_world, n) v1[:] = 3.0 assert v0.inner(v1) == 6*n def test_norm(self, any_backend): n = 301 v0 = Vector(MPI.comm_world, n) v0[:] = -2.0 assert v0.norm("l1") == 2.0*n assert v0.norm("l2") == sqrt(4.0*n) assert v0.norm("linf") == 2.0 def test_min(self, any_backend): v0 = Vector(MPI.comm_world, 301) v0[:] = 2.0 assert v0.min() == 2.0 def test_max(self, any_backend): v0 = Vector(MPI.comm_world,301) v0[:] = -2.0 assert v0.max() == -2.0 def test_sum(self, any_backend): n = 301 v0 = Vector(MPI.comm_world, n) v0[:] = -2.0 assert v0.sum() == -2.0*n def test_sum_entries(self, any_backend): from numpy import zeros n = 301 v0 = Vector(MPI.comm_world, n) v0[:] = -2.0 entries = zeros(5, dtype='uintp') assert v0.sum(entries) == -2.0 entries[0] = 2 entries[1] = 1 entries[2] = 236 entries[3] = 123 entries[4] = 97 assert v0.sum(entries) == -2.0*5 def test_scalar_mult(self, any_backend): n = 301 v0 = Vector(MPI.comm_world, n) v0[:] = -1.0 v0 *= 2.0 assert v0.sum() == -2.0*n def test_vector_element_mult(self, any_backend): n = 301 v0 = Vector(MPI.comm_world, n) v1 = Vector(MPI.comm_world, n) v0[:] = -2.0 v1[:] = 3.0 v0 *= v1 assert v0.sum() == -6.0*n def test_scalar_divide(self, any_backend): n = 301 v0 = Vector(MPI.comm_world, n) v0[:] = -1.0 v0 /= -2.0 assert v0.sum() == 0.5*n def test_vector_add(self, any_backend): n = 301 v0 = Vector(MPI.comm_world, n) v1 = Vector(MPI.comm_world, n) v0[:] = -1.0 v1[:] = 2.0 v0 += v1 assert v0.sum() == n def test_scalar_add(self, any_backend): n = 301 v0 = Vector(MPI.comm_world, n) v1 = Vector(MPI.comm_world, n) v0[:] = -1.0 v0 += 2.0 assert v0.sum() == n v0 -= 2.0 assert v0.sum() == -n v0 = v0 + 3.0 assert v0.sum() == 2*n v0 = v0 - 1.0 assert v0.sum() == n def test_vector_subtract(self, any_backend): n = 301 v0 = Vector(MPI.comm_world, n) v1 = Vector(MPI.comm_world, n) v0[:] = -1.0 v1[:] = 2.0 v0 -= v1 assert v0.sum() == -3.0*n def test_vector_assignment(self, any_backend): m, n = 301, 345 v0 = Vector(MPI.comm_world, m) v1 = Vector(MPI.comm_world, n) v0[:] = -1.0 v1[:] = 2.0 v0 = v1 assert v0.sum() == 2.0*n def test_vector_assignment_length(self, any_backend): # Test that assigning vectors of different lengths fails m, n = 301, 345 v0 = Vector(MPI.comm_world, m) v1 = Vector(MPI.comm_world, n) def wrong_assignment(v0, v1): v0[:] = v1 with pytest.raises(RuntimeError): wrong_assignment(v0, v1) def test_vector_assignment_length(self, any_backend): # Test that assigning with diffrent parallel layouts fails if MPI.size(MPI.comm_world) > 1: m = 301 local_range0 = MPI.local_range(MPI.comm_world, m) print("local range", local_range0[0], local_range0[1]) # Shift parallel partitiong but preserve global size if MPI.rank(MPI.comm_world) == 0: local_range1 = (local_range0[0], local_range0[1] + 1) elif MPI.rank(MPI.comm_world) == MPI.size(MPI.comm_world) - 1: local_range1 = (local_range0[0] + 1, local_range0[1]) else: local_range1 = (local_range0[0] + 1, local_range0[1] + 1) v0 = Vector(MPI.comm_world) v0.init(local_range0) v1 = Vector(MPI.comm_world) v1.init(local_range1) assert v0.size() == v1.size() def wrong_assignment(v0, v1): v0[:] = v1 with pytest.raises(RuntimeError): wrong_assignment(v0, v1) # Test the access of the raw data through pointers # This is only available for Eigen backend def test_vector_data(self, data_backend): # Test for ordinary Vector v = Vector(MPI.comm_world, 301) v = as_backend_type(v) rw_array = v.array_view() assert rw_array.flags.owndata == False with pytest.raises(Exception): rw_array.resize([10]) # Check that the array is a writable view rw_array[0] = 42 ro_array = v.get_local() assert ro_array[0] == 42 # Test for as_backend_type Vector v = as_backend_type(v) rw_array2 = v.array_view() assert (rw_array2 == ro_array).all() # xfail on TypeError xfail_type = pytest.mark.xfail(strict=True, raises=TypeError) xfail_type_py3 = pytest.mark.xfail(strict=True, raises=TypeError) # some systems (32 bit arches) do not have numpy.float128 try: test_float128 = numpy.float128(42.0) except AttributeError: test_float128 = numpy.longdouble(42.0) @pytest.mark.parametrize("operand", [int(42), 42.0, numpy.sin(1.0), float(42.0), numpy.float64(42.0), int(42.0), c_long(42).value, numpy.float16(42.0), numpy.float16(42.0), numpy.float32(42.0), test_float128, numpy.longdouble(42.0), numpy.int8(42.0), numpy.int16(42.0), numpy.int32(42.0), numpy.intc(42.0), numpy.longdouble(42.0), numpy.intp(42.0), numpy.int64(42.0), numpy.int_(42.0), numpy.longlong(42.0), ]) def test_vector_type_priority_with_numpy(self, any_backend, operand): """Test that DOLFIN return types are prefered over NumPy types for binary operations on NumPy objects """ def _test_binary_ops(v, operand): assert isinstance(v + operand, cpp.la.GenericVector) assert isinstance(v - operand, cpp.la.GenericVector) assert isinstance(v*operand, cpp.la.GenericVector) assert isinstance(v/operand, cpp.la.GenericVector) assert isinstance(operand + v, cpp.la.GenericVector) assert isinstance(operand - v, cpp.la.GenericVector) assert isinstance(operand*v, cpp.la.GenericVector) assert isinstance(v+v, cpp.la.GenericVector) assert isinstance(v-v, cpp.la.GenericVector) assert isinstance(v*v, cpp.la.GenericVector) v += v.copy(); assert isinstance(v, cpp.la.GenericVector) v -= v.copy(); assert isinstance(v, cpp.la.GenericVector) v *= v.copy(); assert isinstance(v, cpp.la.GenericVector) v += operand; assert isinstance(v, cpp.la.GenericVector) v -= operand; assert isinstance(v, cpp.la.GenericVector) v *= operand; assert isinstance(v, cpp.la.GenericVector) v /= operand; assert isinstance(v, cpp.la.GenericVector) op = copy(operand); op += v; assert isinstance(op, cpp.la.GenericVector) op = copy(operand); op -= v; assert isinstance(op, cpp.la.GenericVector) op = copy(operand); op *= v; assert isinstance(op, cpp.la.GenericVector) # Test with vector wrapper v = Vector(MPI.comm_world, 8) _test_binary_ops(v, operand) # Test with vector casted to backend type v = as_backend_type(v) _test_binary_ops(v, operand)