warp-lang 1.7.0__py3-none-manylinux_2_34_aarch64.whl

warp/tests/test_fem.py

@@ -0,0 +1,2003 @@
+# SPDX-FileCopyrightText: Copyright (c) 2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import math
+import unittest
+from typing import Any
+
+import numpy as np
+
+import warp as wp
+import warp.fem as fem
+from warp.fem import Coords, D, Domain, Field, Sample, curl, div, grad, integrand, normal
+from warp.fem.cache import dynamic_kernel
+from warp.fem.geometry.closest_point import project_on_tet_at_origin, project_on_tri_at_origin
+from warp.fem.linalg import inverse_qr, spherical_part, symmetric_eigenvalues_qr, symmetric_part
+from warp.fem.space import shape
+from warp.fem.types import make_free_sample
+from warp.fem.utils import (
+    grid_to_hexes,
+    grid_to_quads,
+    grid_to_tets,
+    grid_to_tris,
+)
+from warp.sparse import bsr_zeros
+from warp.tests.unittest_utils import *
+
+vec6f = wp.vec(length=6, dtype=float)
+mat66f = wp.mat(shape=(6, 6), dtype=float)
+
+
+@integrand
+def linear_form(s: Sample, u: Field):
+    return u(s)
+
+
+@integrand
+def scaled_linear_form(s: Sample, u: Field, scale: wp.array(dtype=float)):
+    return u(s) * scale[0]
+
+
+@wp.kernel
+def atomic_sum(v: wp.array(dtype=float), sum: wp.array(dtype=float)):
+    i = wp.tid()
+    wp.atomic_add(sum, 0, v[i])
+
+
+def test_integrate_gradient(test, device):
+    with wp.ScopedDevice(device):
+        # Grid geometry
+        geo = fem.Grid2D(res=wp.vec2i(5))
+
+        # Domain and function spaces
+        domain = fem.Cells(geometry=geo)
+        quadrature = fem.RegularQuadrature(domain=domain, order=3)
+
+        scalar_space = fem.make_polynomial_space(geo, degree=3)
+
+        u = scalar_space.make_field()
+        u.dof_values = wp.zeros_like(u.dof_values, requires_grad=True)
+
+        result = wp.empty(dtype=wp.float32, shape=(1), requires_grad=True)
+        tape = wp.Tape()
+
+        # forward pass
+        with tape:
+            fem.integrate(linear_form, quadrature=quadrature, fields={"u": u}, output=result)
+        tape.backward(result)
+
+        test_field = fem.make_test(space=scalar_space, domain=domain)
+
+        u_adj = wp.empty_like(u.dof_values, requires_grad=True)
+        scale = wp.ones(1, requires_grad=True)
+        loss = wp.zeros(1, requires_grad=True)
+
+        tape2 = wp.Tape()
+        with tape2:
+            fem.integrate(
+                scaled_linear_form,
+                quadrature=quadrature,
+                fields={"u": test_field},
+                values={"scale": scale},
+                assembly="generic",
+                output=u_adj,
+            )
+            wp.launch(atomic_sum, dim=u_adj.shape, inputs=[u_adj, loss])
+
+        # gradient of scalar integral w.r.t dofs should be equal to linear form vector
+        assert_np_equal(u_adj.numpy(), u.dof_values.grad.numpy(), tol=1.0e-8)
+        test.assertAlmostEqual(loss.numpy()[0], 1.0, places=4)
+
+        # Check gradient of linear form vec w.r.t value params
+        tape.zero()
+        tape2.backward(loss=loss)
+
+        test.assertAlmostEqual(loss.numpy()[0], scale.grad.numpy()[0], places=4)
+        tape2.zero()
+        test.assertEqual(scale.grad.numpy()[0], 0.0)
+
+        # Same, with dispatched assembly
+        tape2.reset()
+        loss.zero_()
+        with tape2:
+            fem.integrate(
+                scaled_linear_form,
+                quadrature=quadrature,
+                fields={"u": test_field},
+                values={"scale": scale},
+                assembly="dispatch",
+                output=u_adj,
+            )
+            wp.launch(atomic_sum, dim=u_adj.shape, inputs=[u_adj, loss])
+        tape2.backward(loss=loss)
+        test.assertAlmostEqual(loss.numpy()[0], scale.grad.numpy()[0], places=4)
+
+
+@fem.integrand
+def bilinear_field(s: fem.Sample, domain: fem.Domain):
+    x = domain(s)
+    return x[0] * x[1]
+
+
+@fem.integrand
+def grad_field(s: fem.Sample, p: fem.Field):
+    return fem.grad(p, s)
+
+
+def test_interpolate_gradient(test, device):
+    with wp.ScopedDevice(device):
+        # Quad mesh with single element
+        # so we can test gradient with respect to vertex positions
+        positions = wp.array([[0.0, 0.0], [0.0, 2.0], [2.0, 0.0], [2.0, 2.0]], dtype=wp.vec2, requires_grad=True)
+        quads = wp.array([[0, 2, 3, 1]], dtype=int)
+        geo = fem.Quadmesh2D(quads, positions)
+
+        # Quadratic scalar space
+        scalar_space = fem.make_polynomial_space(geo, degree=2)
+
+        # Point-based vector space
+        # So we can test gradient with respect to interpolation point position
+        point_coords = wp.array([[[0.5, 0.5, 0.0]]], dtype=fem.Coords, requires_grad=True)
+        point_quadrature = fem.ExplicitQuadrature(
+            domain=fem.Cells(geo), points=point_coords, weights=wp.array([[1.0]], dtype=float)
+        )
+        interpolation_nodes = fem.PointBasisSpace(point_quadrature)
+        vector_space = fem.make_collocated_function_space(interpolation_nodes, dtype=wp.vec2)
+
+        # Initialize scalar field with known function
+        scalar_field = scalar_space.make_field()
+        scalar_field.dof_values.requires_grad = True
+        fem.interpolate(bilinear_field, dest=scalar_field)
+
+        # Interpolate gradient at center point
+        vector_field = vector_space.make_field()
+        vector_field.dof_values.requires_grad = True
+        vector_field_restriction = fem.make_restriction(vector_field)
+        tape = wp.Tape()
+        with tape:
+            fem.interpolate(
+                grad_field,
+                dest=vector_field_restriction,
+                fields={"p": scalar_field},
+                kernel_options={"enable_backward": True},
+            )
+
+        assert_np_equal(vector_field.dof_values.numpy(), np.array([[1.0, 1.0]]))
+
+        vector_field.dof_values.grad.assign([1.0, 0.0])
+        tape.backward()
+
+        assert_np_equal(scalar_field.dof_values.grad.numpy(), np.array([0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -0.5, 0.0, 0.5]))
+        assert_np_equal(
+            geo.positions.grad.numpy(),
+            np.array(
+                [
+                    [0.25, 0.25],
+                    [0.25, 0.25],
+                    [-0.25, -0.25],
+                    [-0.25, -0.25],
+                ]
+            ),
+        )
+        assert_np_equal(point_coords.grad.numpy(), np.array([[[0.0, 2.0, 0.0]]]))
+
+        tape.zero()
+        scalar_field.dof_values.grad.zero_()
+        geo.positions.grad.zero_()
+        point_coords.grad.zero_()
+
+        vector_field.dof_values.grad.assign([0.0, 1.0])
+        tape.backward()
+
+        assert_np_equal(scalar_field.dof_values.grad.numpy(), np.array([0.0, 0.0, 0.0, 0.0, 0.0, -0.5, 0.0, 0.5, 0.0]))
+        assert_np_equal(
+            geo.positions.grad.numpy(),
+            np.array(
+                [
+                    [0.25, 0.25],
+                    [-0.25, -0.25],
+                    [0.25, 0.25],
+                    [-0.25, -0.25],
+                ]
+            ),
+        )
+        assert_np_equal(point_coords.grad.numpy(), np.array([[[2.0, 0.0, 0.0]]]))
+
+        # Compare against jacobian
+        scalar_trial = fem.make_trial(scalar_space)
+        jacobian = bsr_zeros(
+            rows_of_blocks=point_quadrature.total_point_count(),
+            cols_of_blocks=scalar_space.node_count(),
+            block_type=wp.mat(shape=(2, 1), dtype=float),
+        )
+        fem.interpolate(
+            grad_field,
+            dest=jacobian,
+            quadrature=point_quadrature,
+            fields={"p": scalar_trial},
+            kernel_options={"enable_backward": False},
+        )
+        assert jacobian.nnz_sync() == 4  # one non-zero per edge center
+        assert_np_equal((jacobian @ scalar_field.dof_values.grad).numpy(), [[0.0, 0.5]])
+
+
+@integrand
+def vector_divergence_form(s: Sample, u: Field, q: Field):
+    return div(u, s) * q(s)
+
+
+@integrand
+def vector_grad_form(s: Sample, u: Field, q: Field):
+    return wp.dot(u(s), grad(q, s))
+
+
+@integrand
+def vector_boundary_form(domain: Domain, s: Sample, u: Field, q: Field):
+    return wp.dot(u(s) * q(s), normal(domain, s))
+
+
+def test_vector_divergence_theorem(test, device):
+    rng = np.random.default_rng(123)
+
+    with wp.ScopedDevice(device):
+        # Grid geometry
+        geo = fem.Grid2D(res=wp.vec2i(5))
+
+        # Domain and function spaces
+        interior = fem.Cells(geometry=geo)
+        boundary = fem.BoundarySides(geometry=geo)
+
+        vector_space = fem.make_polynomial_space(geo, degree=2, dtype=wp.vec2)
+        scalar_space = fem.make_polynomial_space(geo, degree=1, dtype=float)
+
+        u = vector_space.make_field()
+        u.dof_values = rng.random(size=(u.dof_values.shape[0], 2))
+
+        # Divergence theorem
+        constant_one = scalar_space.make_field()
+        constant_one.dof_values.fill_(1.0)
+
+        interior_quadrature = fem.RegularQuadrature(domain=interior, order=vector_space.degree)
+        boundary_quadrature = fem.RegularQuadrature(domain=boundary, order=vector_space.degree)
+        div_int = fem.integrate(
+            vector_divergence_form,
+            quadrature=interior_quadrature,
+            fields={"u": u, "q": constant_one},
+            kernel_options={"enable_backward": False},
+        )
+        boundary_int = fem.integrate(
+            vector_boundary_form,
+            quadrature=boundary_quadrature,
+            fields={"u": u.trace(), "q": constant_one.trace()},
+            kernel_options={"enable_backward": False},
+        )
+
+        test.assertAlmostEqual(div_int, boundary_int, places=5)
+
+        # Integration by parts
+        q = scalar_space.make_field()
+        q.dof_values = rng.random(size=q.dof_values.shape[0])
+
+        interior_quadrature = fem.RegularQuadrature(domain=interior, order=vector_space.degree + scalar_space.degree)
+        boundary_quadrature = fem.RegularQuadrature(domain=boundary, order=vector_space.degree + scalar_space.degree)
+        div_int = fem.integrate(
+            vector_divergence_form,
+            quadrature=interior_quadrature,
+            fields={"u": u, "q": q},
+            kernel_options={"enable_backward": False},
+        )
+        grad_int = fem.integrate(
+            vector_grad_form,
+            quadrature=interior_quadrature,
+            fields={"u": u, "q": q},
+            kernel_options={"enable_backward": False},
+        )
+        boundary_int = fem.integrate(
+            vector_boundary_form,
+            quadrature=boundary_quadrature,
+            fields={"u": u.trace(), "q": q.trace()},
+            kernel_options={"enable_backward": False},
+        )
+
+        test.assertAlmostEqual(div_int + grad_int, boundary_int, places=5)
+
+
+@integrand
+def tensor_divergence_form(s: Sample, tau: Field, v: Field):
+    return wp.dot(div(tau, s), v(s))
+
+
+@integrand
+def tensor_grad_form(s: Sample, tau: Field, v: Field):
+    return wp.ddot(wp.transpose(tau(s)), grad(v, s))
+
+
+@integrand
+def tensor_boundary_form(domain: Domain, s: Sample, tau: Field, v: Field):
+    return wp.dot(tau(s) * v(s), normal(domain, s))
+
+
+def test_tensor_divergence_theorem(test, device):
+    rng = np.random.default_rng(123)
+
+    with wp.ScopedDevice(device):
+        # Grid geometry
+        geo = fem.Grid2D(res=wp.vec2i(5))
+
+        # Domain and function spaces
+        interior = fem.Cells(geometry=geo)
+        boundary = fem.BoundarySides(geometry=geo)
+
+        tensor_space = fem.make_polynomial_space(geo, degree=2, dtype=wp.mat22)
+        vector_space = fem.make_polynomial_space(geo, degree=1, dtype=wp.vec2)
+
+        tau = tensor_space.make_field()
+        tau.dof_values = rng.random(size=(tau.dof_values.shape[0], 2, 2))
+
+        # Divergence theorem
+        constant_vec = vector_space.make_field()
+        constant_vec.dof_values.fill_(wp.vec2(0.5, 2.0))
+
+        interior_quadrature = fem.RegularQuadrature(domain=interior, order=tensor_space.degree)
+        boundary_quadrature = fem.RegularQuadrature(domain=boundary, order=tensor_space.degree)
+        div_int = fem.integrate(
+            tensor_divergence_form,
+            quadrature=interior_quadrature,
+            fields={"tau": tau, "v": constant_vec},
+            kernel_options={"enable_backward": False},
+        )
+        boundary_int = fem.integrate(
+            tensor_boundary_form,
+            quadrature=boundary_quadrature,
+            fields={"tau": tau.trace(), "v": constant_vec.trace()},
+            kernel_options={"enable_backward": False},
+        )
+
+        test.assertAlmostEqual(div_int, boundary_int, places=5)
+
+        # Integration by parts
+        v = vector_space.make_field()
+        v.dof_values = rng.random(size=(v.dof_values.shape[0], 2))
+
+        interior_quadrature = fem.RegularQuadrature(domain=interior, order=tensor_space.degree + vector_space.degree)
+        boundary_quadrature = fem.RegularQuadrature(domain=boundary, order=tensor_space.degree + vector_space.degree)
+        div_int = fem.integrate(
+            tensor_divergence_form,
+            quadrature=interior_quadrature,
+            fields={"tau": tau, "v": v},
+            kernel_options={"enable_backward": False},
+        )
+        grad_int = fem.integrate(
+            tensor_grad_form,
+            quadrature=interior_quadrature,
+            fields={"tau": tau, "v": v},
+            kernel_options={"enable_backward": False},
+        )
+        boundary_int = fem.integrate(
+            tensor_boundary_form,
+            quadrature=boundary_quadrature,
+            fields={"tau": tau.trace(), "v": v.trace()},
+            kernel_options={"enable_backward": False},
+        )
+
+        test.assertAlmostEqual(div_int + grad_int, boundary_int, places=5)
+
+
+@integrand
+def grad_decomposition(s: Sample, u: Field, v: Field):
+    return wp.length_sq(grad(u, s) * v(s) - D(u, s) * v(s) - wp.cross(curl(u, s), v(s)))
+
+
+def test_grad_decomposition(test, device):
+    rng = np.random.default_rng(123)
+
+    with wp.ScopedDevice(device):
+        # Grid geometry
+        geo = fem.Grid3D(res=wp.vec3i(5))
+
+        # Domain and function spaces
+        domain = fem.Cells(geometry=geo)
+        quadrature = fem.RegularQuadrature(domain=domain, order=4)
+
+        vector_space = fem.make_polynomial_space(geo, degree=2, dtype=wp.vec3)
+        u = vector_space.make_field()
+
+        u.dof_values = rng.random(size=(u.dof_values.shape[0], 3))
+
+        err = fem.integrate(grad_decomposition, quadrature=quadrature, fields={"u": u, "v": u})
+        test.assertLess(err, 1.0e-8)
+
+
+def _gen_trimesh(Nx, Ny):
+    x = np.linspace(0.0, 1.0, Nx + 1)
+    y = np.linspace(0.0, 1.0, Ny + 1)
+
+    positions = np.transpose(np.meshgrid(x, y, indexing="ij"), axes=(1, 2, 0)).reshape(-1, 2)
+
+    vidx = grid_to_tris(Nx, Ny)
+
+    return wp.array(positions, dtype=wp.vec2), wp.array(vidx, dtype=int)
+
+
+def _gen_quadmesh(N):
+    x = np.linspace(0.0, 1.0, N + 1)
+    y = np.linspace(0.0, 1.0, N + 1)
+
+    positions = np.transpose(np.meshgrid(x, y, indexing="ij"), axes=(1, 2, 0)).reshape(-1, 2)
+
+    vidx = grid_to_quads(N, N)
+
+    return wp.array(positions, dtype=wp.vec2), wp.array(vidx, dtype=int)
+
+
+def _gen_tetmesh(Nx, Ny, Nz):
+    x = np.linspace(0.0, 1.0, Nx + 1)
+    y = np.linspace(0.0, 1.0, Ny + 1)
+    z = np.linspace(0.0, 1.0, Nz + 1)
+
+    positions = np.transpose(np.meshgrid(x, y, z, indexing="ij"), axes=(1, 2, 3, 0)).reshape(-1, 3)
+
+    vidx = grid_to_tets(Nx, Ny, Nz)
+
+    return wp.array(positions, dtype=wp.vec3), wp.array(vidx, dtype=int)
+
+
+def _gen_hexmesh(N):
+    x = np.linspace(0.0, 1.0, N + 1)
+    y = np.linspace(0.0, 1.0, N + 1)
+    z = np.linspace(0.0, 1.0, N + 1)
+
+    positions = np.transpose(np.meshgrid(x, y, z, indexing="ij"), axes=(1, 2, 3, 0)).reshape(-1, 3)
+
+    vidx = grid_to_hexes(N, N, N)
+
+    return wp.array(positions, dtype=wp.vec3), wp.array(vidx, dtype=int)
+
+
+@fem.integrand(kernel_options={"enable_backward": False})
+def _test_geo_cells(
+    s: fem.Sample,
+    domain: fem.Domain,
+    cell_measures: wp.array(dtype=float),
+):
+    wp.atomic_add(cell_measures, s.element_index, fem.measure(domain, s) * s.qp_weight)
+
+
+@fem.integrand(kernel_options={"enable_backward": False, "max_unroll": 1})
+def _test_geo_sides(
+    s: fem.Sample,
+    domain: fem.Domain,
+    ref_measure: float,
+    side_measures: wp.array(dtype=float),
+):
+    side_index = s.element_index
+    coords = s.element_coords
+
+    cells = fem.cells(domain)
+
+    inner_s = fem.to_inner_cell(domain, s)
+    outer_s = fem.to_outer_cell(domain, s)
+
+    pos_side = domain(s)
+    pos_inner = cells(inner_s)
+    pos_outer = cells(outer_s)
+
+    for k in range(type(pos_side).length):
+        wp.expect_near(pos_side[k], pos_inner[k], 0.0001)
+        wp.expect_near(pos_side[k], pos_outer[k], 0.0001)
+
+    inner_side_s = fem.to_cell_side(domain, inner_s, side_index)
+    outer_side_s = fem.to_cell_side(domain, outer_s, side_index)
+
+    wp.expect_near(coords, inner_side_s.element_coords, 0.0001)
+    wp.expect_near(coords, outer_side_s.element_coords, 0.0001)
+
+    area = fem.measure(domain, s)
+    wp.atomic_add(side_measures, side_index, area * s.qp_weight)
+
+    F = fem.deformation_gradient(domain, s)
+    F_det = fem.Geometry._element_measure(F)
+    wp.expect_near(F_det * ref_measure, area)
+
+
+@fem.integrand(kernel_options={"enable_backward": False, "max_unroll": 1})
+def _test_side_normals(
+    s: fem.Sample,
+    domain: fem.Domain,
+):
+    # test consistency of side normal, measure, and deformation gradient
+    F = fem.deformation_gradient(domain, s)
+
+    nor = fem.normal(domain, s)
+    F_cross = fem.Geometry._element_normal(F)
+
+    for k in range(type(nor).length):
+        wp.expect_near(F_cross[k], nor[k], 0.0001)
+
+
+def _launch_test_geometry_kernel(geo: fem.Geometry, device):
+    cell_measures = wp.zeros(dtype=float, device=device, shape=geo.cell_count())
+    cell_quadrature = fem.RegularQuadrature(fem.Cells(geo), order=2)
+
+    side_measures = wp.zeros(dtype=float, device=device, shape=geo.side_count())
+    side_quadrature = fem.RegularQuadrature(fem.Sides(geo), order=2)
+
+    with wp.ScopedDevice(device):
+        fem.interpolate(
+            _test_geo_cells,
+            quadrature=cell_quadrature,
+            values={"cell_measures": cell_measures},
+        )
+        fem.interpolate(
+            _test_geo_sides,
+            quadrature=side_quadrature,
+            values={"side_measures": side_measures, "ref_measure": geo.reference_side().measure()},
+        )
+
+        if geo.side_normal is not None:
+            fem.interpolate(
+                _test_side_normals,
+                quadrature=side_quadrature,
+            )
+
+    return side_measures, cell_measures
+
+
+def test_grid_2d(test, device):
+    N = 3
+
+    geo = fem.Grid2D(res=wp.vec2i(N))
+
+    test.assertEqual(geo.cell_count(), N**2)
+    test.assertEqual(geo.vertex_count(), (N + 1) ** 2)
+    test.assertEqual(geo.side_count(), 2 * (N + 1) * N)
+    test.assertEqual(geo.boundary_side_count(), 4 * N)
+
+    side_measures, cell_measures = _launch_test_geometry_kernel(geo, device)
+
+    assert_np_equal(side_measures.numpy(), np.full(side_measures.shape, 1.0 / (N)), tol=1.0e-4)
+    assert_np_equal(cell_measures.numpy(), np.full(cell_measures.shape, 1.0 / (N**2)), tol=1.0e-4)
+
+
+def test_triangle_mesh(test, device):
+    N = 3
+
+    with wp.ScopedDevice(device):
+        positions, tri_vidx = _gen_trimesh(N, N)
+
+    geo = fem.Trimesh2D(tri_vertex_indices=tri_vidx, positions=positions)
+
+    test.assertEqual(geo.cell_count(), 2 * (N) ** 2)
+    test.assertEqual(geo.vertex_count(), (N + 1) ** 2)
+    test.assertEqual(geo.side_count(), 2 * (N + 1) * N + (N**2))
+    test.assertEqual(geo.boundary_side_count(), 4 * N)
+
+    side_measures, cell_measures = _launch_test_geometry_kernel(geo, device)
+
+    assert_np_equal(cell_measures.numpy(), np.full(cell_measures.shape, 0.5 / (N**2)), tol=1.0e-4)
+    test.assertAlmostEqual(np.sum(side_measures.numpy()), 2 * (N + 1) + N * math.sqrt(2.0), places=4)
+
+    # 3d
+
+    positions = positions.numpy()
+    positions = np.hstack((positions, np.ones((positions.shape[0], 1))))
+    positions = wp.array(positions, device=device, dtype=wp.vec3)
+
+    geo = fem.Trimesh3D(tri_vertex_indices=tri_vidx, positions=positions)
+
+    test.assertEqual(geo.cell_count(), 2 * (N) ** 2)
+    test.assertEqual(geo.vertex_count(), (N + 1) ** 2)
+    test.assertEqual(geo.side_count(), 2 * (N + 1) * N + (N**2))
+    test.assertEqual(geo.boundary_side_count(), 4 * N)
+
+    side_measures, cell_measures = _launch_test_geometry_kernel(geo, device)
+
+    assert_np_equal(cell_measures.numpy(), np.full(cell_measures.shape, 0.5 / (N**2)), tol=1.0e-4)
+    test.assertAlmostEqual(np.sum(side_measures.numpy()), 2 * (N + 1) + N * math.sqrt(2.0), places=4)
+
+
+def test_quad_mesh(test, device):
+    N = 3
+
+    with wp.ScopedDevice(device):
+        positions, quad_vidx = _gen_quadmesh(N)
+
+    geo = fem.Quadmesh2D(quad_vertex_indices=quad_vidx, positions=positions)
+
+    test.assertEqual(geo.cell_count(), N**2)
+    test.assertEqual(geo.vertex_count(), (N + 1) ** 2)
+    test.assertEqual(geo.side_count(), 2 * (N + 1) * N)
+    test.assertEqual(geo.boundary_side_count(), 4 * N)
+
+    side_measures, cell_measures = _launch_test_geometry_kernel(geo, device)
+
+    assert_np_equal(side_measures.numpy(), np.full(side_measures.shape, 1.0 / (N)), tol=1.0e-4)
+    assert_np_equal(cell_measures.numpy(), np.full(cell_measures.shape, 1.0 / (N**2)), tol=1.0e-4)
+
+    # 3d
+
+    positions = positions.numpy()
+    positions = np.hstack((positions, np.ones((positions.shape[0], 1))))
+    positions = wp.array(positions, device=device, dtype=wp.vec3)
+
+    geo = fem.Quadmesh3D(quad_vertex_indices=quad_vidx, positions=positions)
+
+    test.assertEqual(geo.cell_count(), N**2)
+    test.assertEqual(geo.vertex_count(), (N + 1) ** 2)
+    test.assertEqual(geo.side_count(), 2 * (N + 1) * N)
+    test.assertEqual(geo.boundary_side_count(), 4 * N)
+
+    side_measures, cell_measures = _launch_test_geometry_kernel(geo, device)
+
+    assert_np_equal(side_measures.numpy(), np.full(side_measures.shape, 1.0 / (N)), tol=1.0e-4)
+    assert_np_equal(cell_measures.numpy(), np.full(cell_measures.shape, 1.0 / (N**2)), tol=1.0e-4)
+
+
+def test_grid_3d(test, device):
+    N = 3
+
+    geo = fem.Grid3D(res=wp.vec3i(N))
+
+    test.assertEqual(geo.cell_count(), (N) ** 3)
+    test.assertEqual(geo.vertex_count(), (N + 1) ** 3)
+    test.assertEqual(geo.side_count(), 3 * (N + 1) * N**2)
+    test.assertEqual(geo.boundary_side_count(), 6 * N * N)
+    test.assertEqual(geo.edge_count(), 3 * N * (N + 1) ** 2)
+
+    side_measures, cell_measures = _launch_test_geometry_kernel(geo, device)
+
+    assert_np_equal(side_measures.numpy(), np.full(side_measures.shape, 1.0 / (N**2)), tol=1.0e-4)
+    assert_np_equal(cell_measures.numpy(), np.full(cell_measures.shape, 1.0 / (N**3)), tol=1.0e-4)
+
+
+def test_tet_mesh(test, device):
+    N = 3
+
+    with wp.ScopedDevice(device):
+        positions, tet_vidx = _gen_tetmesh(N, N, N)
+
+    geo = fem.Tetmesh(tet_vertex_indices=tet_vidx, positions=positions)
+
+    test.assertEqual(geo.cell_count(), 5 * (N) ** 3)
+    test.assertEqual(geo.vertex_count(), (N + 1) ** 3)
+    test.assertEqual(geo.side_count(), 6 * (N + 1) * N**2 + (N**3) * 4)
+    test.assertEqual(geo.boundary_side_count(), 12 * N * N)
+    test.assertEqual(geo.edge_count(), 3 * N * (N + 1) * (2 * N + 1))
+
+    side_measures, cell_measures = _launch_test_geometry_kernel(geo, device)
+
+    test.assertAlmostEqual(np.sum(cell_measures.numpy()), 1.0, places=4)
+    test.assertAlmostEqual(np.sum(side_measures.numpy()), 0.5 * 6 * (N + 1) + N * 2 * math.sqrt(3.0), places=4)
+
+
+def test_hex_mesh(test, device):
+    N = 3
+
+    with wp.ScopedDevice(device):
+        positions, tet_vidx = _gen_hexmesh(N)
+
+    geo = fem.Hexmesh(hex_vertex_indices=tet_vidx, positions=positions)
+
+    test.assertEqual(geo.cell_count(), (N) ** 3)
+    test.assertEqual(geo.vertex_count(), (N + 1) ** 3)
+    test.assertEqual(geo.side_count(), 3 * (N + 1) * N**2)
+    test.assertEqual(geo.boundary_side_count(), 6 * N * N)
+    test.assertEqual(geo.edge_count(), 3 * N * (N + 1) ** 2)
+
+    side_measures, cell_measures = _launch_test_geometry_kernel(geo, device)
+
+    assert_np_equal(side_measures.numpy(), np.full(side_measures.shape, 1.0 / (N**2)), tol=1.0e-4)
+    assert_np_equal(cell_measures.numpy(), np.full(cell_measures.shape, 1.0 / (N**3)), tol=1.0e-4)
+
+
+def test_nanogrid(test, device):
+    N = 8
+
+    points = wp.array([[0.5, 0.5, 0.5]], dtype=float, device=device)
+    volume = wp.Volume.allocate_by_tiles(
+        tile_points=points, voxel_size=1.0 / N, translation=(0.0, 0.0, 0.0), bg_value=None, device=device
+    )
+
+    geo = fem.Nanogrid(volume)
+
+    test.assertEqual(geo.cell_count(), (N) ** 3)
+    test.assertEqual(geo.vertex_count(), (N + 1) ** 3)
+    test.assertEqual(geo.side_count(), 3 * (N + 1) * N**2)
+    test.assertEqual(geo.boundary_side_count(), 6 * N * N)
+    test.assertEqual(geo.edge_count(), 3 * N * (N + 1) ** 2)
+
+    side_measures, cell_measures = _launch_test_geometry_kernel(geo, device)
+
+    assert_np_equal(side_measures.numpy(), np.full(side_measures.shape, 1.0 / (N**2)), tol=1.0e-4)
+    assert_np_equal(cell_measures.numpy(), np.full(cell_measures.shape, 1.0 / (N**3)), tol=1.0e-4)
+
+
+@wp.func
+def _refinement_field(x: wp.vec3):
+    return 4.0 * (wp.length(x) - 0.5)
+
+
+def test_adaptive_nanogrid(test, device):
+    # 3 res-1 voxels, 8 res-0 voxels
+
+    res0 = wp.array(
+        [
+            [2, 2, 0],
+            [2, 3, 0],
+            [3, 2, 0],
+            [3, 3, 0],
+            [2, 2, 1],
+            [2, 3, 1],
+            [3, 2, 1],
+            [3, 3, 1],
+        ],
+        dtype=int,
+        device=device,
+    )
+    res1 = wp.array(
+        [
+            [0, 0, 0],
+            [0, 1, 0],
+            [1, 0, 0],
+            [1, 1, 0],
+        ],
+        dtype=int,
+        device=device,
+    )
+
+    grid0 = wp.Volume.allocate_by_voxels(res0, 0.5, device=device)
+    grid1 = wp.Volume.allocate_by_voxels(res1, 1.0, device=device)
+    geo = fem.adaptive_nanogrid_from_hierarchy([grid0, grid1])
+
+    test.assertEqual(geo.cell_count(), 3 + 8)
+    test.assertEqual(geo.vertex_count(), 2 * 9 + 27 - 8)
+    test.assertEqual(geo.side_count(), 2 * 4 + 6 * 2 + (3 * (2 + 1) * 2**2 - 6))
+    test.assertEqual(geo.boundary_side_count(), 2 * 4 + 4 * 2 + (4 * 4 - 4))
+    # test.assertEqual(geo.edge_count(), 6 * 4 + 9 + (3 * 2 * (2 + 1) ** 2 - 12))
+    test.assertEqual(geo.stacked_face_count(), geo.side_count() + 2)
+    test.assertEqual(geo.stacked_edge_count(), 6 * 4 + 9 + (3 * 2 * (2 + 1) ** 2 - 12) + 7)
+
+    side_measures, cell_measures = _launch_test_geometry_kernel(geo, device)
+
+    test.assertAlmostEqual(np.sum(cell_measures.numpy()), 4.0, places=4)
+    test.assertAlmostEqual(np.sum(side_measures.numpy()), 20 + 3.0, places=4)
+
+    # Test with non-graded geometry
+    ref_field = fem.ImplicitField(fem.Cells(geo), func=_refinement_field)
+    non_graded_geo = fem.adaptive_nanogrid_from_field(grid1, level_count=3, refinement_field=ref_field)
+    _launch_test_geometry_kernel(geo, device)
+
+    # Test automatic grading
+    graded_geo = fem.adaptive_nanogrid_from_field(grid1, level_count=3, refinement_field=ref_field, grading="face")
+    test.assertEqual(non_graded_geo.cell_count() + 7, graded_geo.cell_count())
+
+
+@integrand
+def _rigid_deformation_field(s: Sample, domain: Domain, translation: wp.vec3, rotation: wp.vec3, scale: float):
+    q = wp.quat_from_axis_angle(wp.normalize(rotation), wp.length(rotation))
+    return translation + scale * wp.quat_rotate(q, domain(s)) - domain(s)
+
+
+def test_deformed_geometry(test, device):
+    N = 3
+
+    with wp.ScopedDevice(device):
+        positions, tet_vidx = _gen_tetmesh(N, N, N)
+
+        geo = fem.Tetmesh(tet_vertex_indices=tet_vidx, positions=positions)
+
+        translation = [1.0, 2.0, 3.0]
+        rotation = [0.0, math.pi / 4.0, 0.0]
+        scale = 2.0
+
+        vector_space = fem.make_polynomial_space(geo, dtype=wp.vec3, degree=2)
+        pos_field = vector_space.make_field()
+        fem.interpolate(
+            _rigid_deformation_field,
+            dest=pos_field,
+            values={"translation": translation, "rotation": rotation, "scale": scale},
+        )
+
+        deformed_geo = pos_field.make_deformed_geometry()
+
+        # rigidly-deformed geometry
+
+        test.assertEqual(geo.cell_count(), 5 * (N) ** 3)
+        test.assertEqual(geo.vertex_count(), (N + 1) ** 3)
+        test.assertEqual(geo.side_count(), 6 * (N + 1) * N**2 + (N**3) * 4)
+        test.assertEqual(geo.boundary_side_count(), 12 * N * N)
+
+        side_measures, cell_measures = _launch_test_geometry_kernel(deformed_geo, wp.get_device())
+
+        test.assertAlmostEqual(
+            np.sum(cell_measures.numpy()), scale**3, places=4, msg=f"cell_measures = {cell_measures.numpy()}"
+        )
+        test.assertAlmostEqual(
+            np.sum(side_measures.numpy()), scale**2 * (0.5 * 6 * (N + 1) + N * 2 * math.sqrt(3.0)), places=4
+        )
+
+        @wp.kernel
+        def _test_deformed_geometry_normal(
+            geo_index_arg: geo.SideIndexArg, geo_arg: geo.SideArg, def_arg: deformed_geo.SideArg, rotation: wp.vec3
+        ):
+            i = wp.tid()
+            side_index = deformed_geo.boundary_side_index(geo_index_arg, i)
+
+            s = make_free_sample(side_index, Coords(0.5, 0.5, 0.0))
+            geo_n = geo.side_normal(geo_arg, s)
+            def_n = deformed_geo.side_normal(def_arg, s)
+
+            q = wp.quat_from_axis_angle(wp.normalize(rotation), wp.length(rotation))
+            wp.expect_near(wp.quat_rotate(q, geo_n), def_n, 0.001)
+
+        wp.launch(
+            _test_deformed_geometry_normal,
+            dim=geo.boundary_side_count(),
+            inputs=[
+                geo.side_index_arg_value(wp.get_device()),
+                geo.side_arg_value(wp.get_device()),
+                deformed_geo.side_arg_value(wp.get_device()),
+                rotation,
+            ],
+        )
+
+    wp.synchronize()
+
+
+@wp.kernel
+def _test_closest_point_on_tri_kernel(
+    e0: wp.vec2,
+    e1: wp.vec2,
+    points: wp.array(dtype=wp.vec2),
+    sq_dist: wp.array(dtype=float),
+    coords: wp.array(dtype=Coords),
+):
+    i = wp.tid()
+    d2, c = project_on_tri_at_origin(points[i], e0, e1)
+    sq_dist[i] = d2
+    coords[i] = c
+
+
+@wp.kernel
+def _test_closest_point_on_tet_kernel(
+    e0: wp.vec3,
+    e1: wp.vec3,
+    e2: wp.vec3,
+    points: wp.array(dtype=wp.vec3),
+    sq_dist: wp.array(dtype=float),
+    coords: wp.array(dtype=Coords),
+):
+    i = wp.tid()
+    d2, c = project_on_tet_at_origin(points[i], e0, e1, e2)
+    sq_dist[i] = d2
+    coords[i] = c
+
+
+def test_closest_point_queries(test, device):
+    # Test some simple lookup queries
+    e0 = wp.vec2(2.0, 0.0)
+    e1 = wp.vec2(0.0, 2.0)
+
+    points = wp.array(
+        (
+            [-1.0, -1.0],
+            [0.5, 0.5],
+            [1.0, 1.0],
+            [2.0, 2.0],
+        ),
+        dtype=wp.vec2,
+        device=device,
+    )
+    expected_sq_dist = np.array([2.0, 0.0, 0.0, 2.0])
+    expected_coords = np.array([[1.0, 0.0, 0.0], [0.5, 0.25, 0.25], [0.0, 0.5, 0.5], [0.0, 0.5, 0.5]])
+
+    sq_dist = wp.empty(shape=points.shape, dtype=float, device=device)
+    coords = wp.empty(shape=points.shape, dtype=Coords, device=device)
+    wp.launch(
+        _test_closest_point_on_tri_kernel, dim=points.shape, device=device, inputs=[e0, e1, points, sq_dist, coords]
+    )
+
+    assert_np_equal(coords.numpy(), expected_coords)
+    assert_np_equal(sq_dist.numpy(), expected_sq_dist)
+
+    # Tet
+
+    e0 = wp.vec3(3.0, 0.0, 0.0)
+    e1 = wp.vec3(0.0, 3.0, 0.0)
+    e2 = wp.vec3(0.0, 0.0, 3.0)
+
+    points = wp.array(
+        (
+            [-1.0, -1.0, -1.0],
+            [0.5, 0.5, 0.5],
+            [1.0, 1.0, 1.0],
+            [2.0, 2.0, 2.0],
+        ),
+        dtype=wp.vec3,
+        device=device,
+    )
+    expected_sq_dist = np.array([3.0, 0.0, 0.0, 3.0])
+    expected_coords = np.array(
+        [
+            [0.0, 0.0, 0.0],
+            [1.0 / 6.0, 1.0 / 6.0, 1.0 / 6.0],
+            [1.0 / 3.0, 1.0 / 3.0, 1.0 / 3.0],
+            [1.0 / 3.0, 1.0 / 3.0, 1.0 / 3.0],
+        ]
+    )
+
+    sq_dist = wp.empty(shape=points.shape, dtype=float, device=device)
+    coords = wp.empty(shape=points.shape, dtype=Coords, device=device)
+    wp.launch(
+        _test_closest_point_on_tet_kernel, dim=points.shape, device=device, inputs=[e0, e1, e2, points, sq_dist, coords]
+    )
+
+    assert_np_equal(coords.numpy(), expected_coords, tol=1.0e-4)
+    assert_np_equal(sq_dist.numpy(), expected_sq_dist, tol=1.0e-4)
+
+
+def test_regular_quadrature(test, device):
+    from warp.fem.geometry.element import LinearEdge, Polynomial, Triangle
+
+    for family in Polynomial:
+        # test integrating monomials
+        for degree in range(8):
+            coords, weights = LinearEdge().instantiate_quadrature(degree, family=family)
+            res = sum(w * pow(c[0], degree) for w, c in zip(weights, coords))
+            ref = 1.0 / (degree + 1)
+
+            test.assertAlmostEqual(ref, res, places=4)
+
+        # test integrating y^k1 (1 - x)^k2 on triangle using transformation to square
+        for x_degree in range(4):
+            for y_degree in range(4):
+                coords, weights = Triangle().instantiate_quadrature(x_degree + y_degree, family=family)
+                res = 0.5 * sum(w * pow(1.0 - c[1], x_degree) * pow(c[2], y_degree) for w, c in zip(weights, coords))
+
+                ref = 1.0 / ((x_degree + y_degree + 2) * (y_degree + 1))
+                # print(x_degree, y_degree, family, len(coords), res, ref)
+                test.assertAlmostEqual(ref, res, places=4)
+
+    # test integrating y^k1 (1 - x)^k2 on triangle using direct formulas
+    for x_degree in range(5):
+        for y_degree in range(5):
+            coords, weights = Triangle().instantiate_quadrature(x_degree + y_degree, family=None)
+            res = 0.5 * sum(w * pow(1.0 - c[1], x_degree) * pow(c[2], y_degree) for w, c in zip(weights, coords))
+
+            ref = 1.0 / ((x_degree + y_degree + 2) * (y_degree + 1))
+            test.assertAlmostEqual(ref, res, places=4)
+
+
+def test_dof_mapper(test, device):
+    matrix_types = [wp.mat22, wp.mat33]
+
+    # Symmetric mapper
+    for mapping in fem.SymmetricTensorMapper.Mapping:
+        for dtype in matrix_types:
+            mapper = fem.SymmetricTensorMapper(dtype, mapping=mapping)
+            dof_dtype = mapper.dof_dtype
+
+            for k in range(dof_dtype._length_):
+                elem = np.array(dof_dtype(0.0))
+                elem[k] = 1.0
+                dof_vec = dof_dtype(elem)
+
+                mat = mapper.dof_to_value(dof_vec)
+                dof_round_trip = mapper.value_to_dof(mat)
+
+                # Check that value_to_dof(dof_to_value) is idempotent
+                assert_np_equal(np.array(dof_round_trip), np.array(dof_vec))
+
+                # Check that value is unitary for Frobenius norm 0.5 * |tau:tau|
+                frob_norm2 = 0.5 * wp.ddot(mat, mat)
+                test.assertAlmostEqual(frob_norm2, 1.0, places=6)
+
+    # Skew-symmetric mapper
+    for dtype in matrix_types:
+        mapper = fem.SkewSymmetricTensorMapper(dtype)
+        dof_dtype = mapper.dof_dtype
+
+        if hasattr(dof_dtype, "_length_"):
+            for k in range(dof_dtype._length_):
+                elem = np.array(dof_dtype(0.0))
+                elem[k] = 1.0
+                dof_vec = dof_dtype(elem)
+
+                mat = mapper.dof_to_value(dof_vec)
+                dof_round_trip = mapper.value_to_dof(mat)
+
+                # Check that value_to_dof(dof_to_value) is idempotent
+                assert_np_equal(np.array(dof_round_trip), np.array(dof_vec))
+
+                # Check that value is unitary for Frobenius norm 0.5 * |tau:tau|
+                frob_norm2 = 0.5 * wp.ddot(mat, mat)
+                test.assertAlmostEqual(frob_norm2, 1.0, places=6)
+        else:
+            dof_val = 1.0
+
+            mat = mapper.dof_to_value(dof_val)
+            dof_round_trip = mapper.value_to_dof(mat)
+
+            test.assertAlmostEqual(dof_round_trip, dof_val)
+
+            # Check that value is unitary for Frobenius norm 0.5 * |tau:tau|
+            frob_norm2 = 0.5 * wp.ddot(mat, mat)
+            test.assertAlmostEqual(frob_norm2, 1.0, places=6)
+
+
+@wp.func
+def _expect_near(a: Any, b: Any, tol: float):
+    wp.expect_near(a, b, tol)
+
+
+@wp.func
+def _expect_near(a: wp.vec2, b: wp.vec2, tol: float):
+    for k in range(2):
+        wp.expect_near(a[k], b[k], tol)
+
+
+def test_shape_function_weight(test, shape: shape.ShapeFunction, coord_sampler, CENTER_COORDS):
+    NODE_COUNT = shape.NODES_PER_ELEMENT
+    weight_fn = shape.make_element_inner_weight()
+    node_coords_fn = shape.make_node_coords_in_element()
+
+    # Weight at node should be 1
+    @dynamic_kernel(suffix=shape.name, kernel_options={"enable_backward": False})
+    def node_unity_test():
+        n = wp.tid()
+        node_w = weight_fn(node_coords_fn(n), n)
+        wp.expect_near(node_w, 1.0, 1e-5)
+
+    wp.launch(node_unity_test, dim=NODE_COUNT, inputs=[])
+
+    # Sum of node quadrature weights should be one (order 0)
+    # Sum of weighted quadrature coords should be element center (order 1)
+    node_quadrature_weight_fn = shape.make_node_quadrature_weight()
+
+    @dynamic_kernel(suffix=shape.name, kernel_options={"enable_backward": False})
+    def node_quadrature_unity_test():
+        sum_node_qp = float(0.0)
+        sum_node_qp_coords = Coords(0.0)
+
+        for n in range(NODE_COUNT):
+            w = node_quadrature_weight_fn(n)
+            sum_node_qp += w
+            sum_node_qp_coords += w * node_coords_fn(n)
+
+        wp.expect_near(sum_node_qp, 1.0, 0.0001)
+        wp.expect_near(sum_node_qp_coords, CENTER_COORDS, 0.0001)
+
+    wp.launch(node_quadrature_unity_test, dim=1, inputs=[])
+
+    @dynamic_kernel(suffix=shape.name, kernel_options={"enable_backward": False})
+    def partition_of_unity_test():
+        rng_state = wp.rand_init(4321, wp.tid())
+        coords = coord_sampler(rng_state)
+
+        # sum of node weights anywhere should be 1.0
+        w_sum = type(weight_fn(coords, 0))(0.0)
+        for n in range(NODE_COUNT):
+            w_sum += weight_fn(coords, n)
+
+        _expect_near(wp.abs(w_sum), type(w_sum)(1.0), 0.0001)
+
+    n_samples = 100
+    wp.launch(partition_of_unity_test, dim=n_samples, inputs=[])
+
+
+def test_shape_function_trace(test, shape: shape.ShapeFunction, CENTER_COORDS):
+    NODE_COUNT = shape.NODES_PER_ELEMENT
+    node_coords_fn = shape.make_node_coords_in_element()
+
+    # Sum of node quadrature weights should be one (order 0)
+    # Sum of weighted quadrature coords should be element center (order 1)
+    trace_node_quadrature_weight_fn = shape.make_trace_node_quadrature_weight()
+
+    @dynamic_kernel(suffix=shape.name, kernel_options={"enable_backward": False})
+    def trace_node_quadrature_unity_test():
+        sum_node_qp = float(0.0)
+        sum_node_qp_coords = Coords(0.0)
+
+        for n in range(NODE_COUNT):
+            coords = node_coords_fn(n)
+
+            if wp.abs(coords[0]) < 1.0e-6:
+                w = trace_node_quadrature_weight_fn(n)
+                sum_node_qp += w
+                sum_node_qp_coords += w * node_coords_fn(n)
+
+        wp.expect_near(sum_node_qp, 1.0, 0.0001)
+        wp.expect_near(sum_node_qp_coords, CENTER_COORDS, 0.0001)
+
+    wp.launch(trace_node_quadrature_unity_test, dim=1, inputs=[])
+
+
+def test_shape_function_gradient(
+    test,
+    shape: shape.ShapeFunction,
+    coord_sampler,
+    coord_delta_sampler,
+    pure_curl: bool = False,
+    pure_spherical: bool = False,
+):
+    weight_fn = shape.make_element_inner_weight()
+    weight_gradient_fn = shape.make_element_inner_weight_gradient()
+
+    @wp.func
+    def scalar_delta(avg_grad: Any, param_delta: Any):
+        return wp.dot(avg_grad, param_delta)
+
+    @wp.func
+    def vector_delta(avg_grad: Any, param_delta: Any):
+        return avg_grad * param_delta
+
+    grad_delta_fn = scalar_delta if shape.value == shape.Value.Scalar else vector_delta
+
+    @dynamic_kernel(suffix=shape.name, kernel_options={"enable_backward": False})
+    def finite_difference_test():
+        i, n = wp.tid()
+        rng_state = wp.rand_init(1234, i)
+
+        coords = coord_sampler(rng_state)
+
+        epsilon = 0.003
+        param_delta, coords_delta = coord_delta_sampler(epsilon, rng_state)
+
+        w_p = weight_fn(coords + coords_delta, n)
+        w_m = weight_fn(coords - coords_delta, n)
+
+        gp = weight_gradient_fn(coords + coords_delta, n)
+        gm = weight_gradient_fn(coords - coords_delta, n)
+
+        # 2nd-order finite-difference test
+        # See Schroeder 2019, Practical course on computing derivatives in code
+        delta_ref = w_p - w_m
+        delta_est = grad_delta_fn(gp + gm, param_delta)
+        _expect_near(delta_ref, delta_est, 0.0001)
+
+        if wp.static(pure_curl):
+            wp.expect_near(wp.ddot(symmetric_part(gp), symmetric_part(gp)), gp.dtype(0.0))
+
+        if wp.static(pure_spherical):
+            deviatoric_part = gp - spherical_part(gp)
+            wp.expect_near(wp.ddot(deviatoric_part, deviatoric_part), gp.dtype(0.0))
+
+    n_samples = 100
+    wp.launch(finite_difference_test, dim=(n_samples, shape.NODES_PER_ELEMENT), inputs=[])
+
+
+def test_square_shape_functions(test, device):
+    SQUARE_CENTER_COORDS = wp.constant(Coords(0.5, 0.5, 0.0))
+    SQUARE_SIDE_CENTER_COORDS = wp.constant(Coords(0.0, 0.5, 0.0))
+
+    @wp.func
+    def square_coord_sampler(state: wp.uint32):
+        return Coords(wp.randf(state), wp.randf(state), 0.0)
+
+    @wp.func
+    def square_coord_delta_sampler(epsilon: float, state: wp.uint32):
+        param_delta = wp.normalize(wp.vec2(wp.randf(state), wp.randf(state))) * epsilon
+        return param_delta, Coords(param_delta[0], param_delta[1], 0.0)
+
+    Q_1 = shape.SquareBipolynomialShapeFunctions(degree=1, family=fem.Polynomial.LOBATTO_GAUSS_LEGENDRE)
+    Q_2 = shape.SquareBipolynomialShapeFunctions(degree=2, family=fem.Polynomial.LOBATTO_GAUSS_LEGENDRE)
+    Q_3 = shape.SquareBipolynomialShapeFunctions(degree=3, family=fem.Polynomial.LOBATTO_GAUSS_LEGENDRE)
+
+    test_shape_function_weight(test, Q_1, square_coord_sampler, SQUARE_CENTER_COORDS)
+    test_shape_function_weight(test, Q_2, square_coord_sampler, SQUARE_CENTER_COORDS)
+    test_shape_function_weight(test, Q_3, square_coord_sampler, SQUARE_CENTER_COORDS)
+    test_shape_function_trace(test, Q_1, SQUARE_SIDE_CENTER_COORDS)
+    test_shape_function_trace(test, Q_2, SQUARE_SIDE_CENTER_COORDS)
+    test_shape_function_trace(test, Q_3, SQUARE_SIDE_CENTER_COORDS)
+    test_shape_function_gradient(test, Q_1, square_coord_sampler, square_coord_delta_sampler)
+    test_shape_function_gradient(test, Q_2, square_coord_sampler, square_coord_delta_sampler)
+    test_shape_function_gradient(test, Q_3, square_coord_sampler, square_coord_delta_sampler)
+
+    Q_1 = shape.SquareBipolynomialShapeFunctions(degree=1, family=fem.Polynomial.GAUSS_LEGENDRE)
+    Q_2 = shape.SquareBipolynomialShapeFunctions(degree=2, family=fem.Polynomial.GAUSS_LEGENDRE)
+    Q_3 = shape.SquareBipolynomialShapeFunctions(degree=3, family=fem.Polynomial.GAUSS_LEGENDRE)
+
+    test_shape_function_weight(test, Q_1, square_coord_sampler, SQUARE_CENTER_COORDS)
+    test_shape_function_weight(test, Q_2, square_coord_sampler, SQUARE_CENTER_COORDS)
+    test_shape_function_weight(test, Q_3, square_coord_sampler, SQUARE_CENTER_COORDS)
+    test_shape_function_gradient(test, Q_1, square_coord_sampler, square_coord_delta_sampler)
+    test_shape_function_gradient(test, Q_2, square_coord_sampler, square_coord_delta_sampler)
+    test_shape_function_gradient(test, Q_3, square_coord_sampler, square_coord_delta_sampler)
+
+    S_2 = shape.SquareSerendipityShapeFunctions(degree=2, family=fem.Polynomial.LOBATTO_GAUSS_LEGENDRE)
+    S_3 = shape.SquareSerendipityShapeFunctions(degree=3, family=fem.Polynomial.LOBATTO_GAUSS_LEGENDRE)
+
+    test_shape_function_weight(test, S_2, square_coord_sampler, SQUARE_CENTER_COORDS)
+    test_shape_function_weight(test, S_3, square_coord_sampler, SQUARE_CENTER_COORDS)
+    test_shape_function_trace(test, S_2, SQUARE_SIDE_CENTER_COORDS)
+    test_shape_function_trace(test, S_3, SQUARE_SIDE_CENTER_COORDS)
+    test_shape_function_gradient(test, S_2, square_coord_sampler, square_coord_delta_sampler)
+    test_shape_function_gradient(test, S_3, square_coord_sampler, square_coord_delta_sampler)
+
+    P_c1 = shape.SquareNonConformingPolynomialShapeFunctions(degree=1)
+    P_c2 = shape.SquareNonConformingPolynomialShapeFunctions(degree=2)
+    P_c3 = shape.SquareNonConformingPolynomialShapeFunctions(degree=3)
+
+    test_shape_function_weight(test, P_c1, square_coord_sampler, SQUARE_CENTER_COORDS)
+    test_shape_function_weight(test, P_c2, square_coord_sampler, SQUARE_CENTER_COORDS)
+    test_shape_function_weight(test, P_c3, square_coord_sampler, SQUARE_CENTER_COORDS)
+    test_shape_function_gradient(test, P_c1, square_coord_sampler, square_coord_delta_sampler)
+    test_shape_function_gradient(test, P_c2, square_coord_sampler, square_coord_delta_sampler)
+    test_shape_function_gradient(test, P_c3, square_coord_sampler, square_coord_delta_sampler)
+
+    N1_1 = shape.SquareNedelecFirstKindShapeFunctions(degree=1)
+    test_shape_function_gradient(test, N1_1, square_coord_sampler, square_coord_delta_sampler)
+    RT_1 = shape.SquareRaviartThomasShapeFunctions(degree=1)
+    test_shape_function_gradient(test, RT_1, square_coord_sampler, square_coord_delta_sampler)
+
+    wp.synchronize()
+
+
+def test_cube_shape_functions(test, device):
+    CUBE_CENTER_COORDS = wp.constant(Coords(0.5, 0.5, 0.5))
+    CUBE_SIDE_CENTER_COORDS = wp.constant(Coords(0.0, 0.5, 0.5))
+
+    @wp.func
+    def cube_coord_sampler(state: wp.uint32):
+        return Coords(wp.randf(state), wp.randf(state), wp.randf(state))
+
+    @wp.func
+    def cube_coord_delta_sampler(epsilon: float, state: wp.uint32):
+        param_delta = wp.normalize(wp.vec3(wp.randf(state), wp.randf(state), wp.randf(state))) * epsilon
+        return param_delta, param_delta
+
+    Q_1 = shape.CubeTripolynomialShapeFunctions(degree=1, family=fem.Polynomial.LOBATTO_GAUSS_LEGENDRE)
+    Q_2 = shape.CubeTripolynomialShapeFunctions(degree=2, family=fem.Polynomial.LOBATTO_GAUSS_LEGENDRE)
+    Q_3 = shape.CubeTripolynomialShapeFunctions(degree=3, family=fem.Polynomial.LOBATTO_GAUSS_LEGENDRE)
+
+    test_shape_function_weight(test, Q_1, cube_coord_sampler, CUBE_CENTER_COORDS)
+    test_shape_function_weight(test, Q_2, cube_coord_sampler, CUBE_CENTER_COORDS)
+    test_shape_function_weight(test, Q_3, cube_coord_sampler, CUBE_CENTER_COORDS)
+    test_shape_function_trace(test, Q_1, CUBE_SIDE_CENTER_COORDS)
+    test_shape_function_trace(test, Q_2, CUBE_SIDE_CENTER_COORDS)
+    test_shape_function_trace(test, Q_3, CUBE_SIDE_CENTER_COORDS)
+    test_shape_function_gradient(test, Q_1, cube_coord_sampler, cube_coord_delta_sampler)
+    test_shape_function_gradient(test, Q_2, cube_coord_sampler, cube_coord_delta_sampler)
+    test_shape_function_gradient(test, Q_3, cube_coord_sampler, cube_coord_delta_sampler)
+
+    Q_1 = shape.CubeTripolynomialShapeFunctions(degree=1, family=fem.Polynomial.GAUSS_LEGENDRE)
+    Q_2 = shape.CubeTripolynomialShapeFunctions(degree=2, family=fem.Polynomial.GAUSS_LEGENDRE)
+    Q_3 = shape.CubeTripolynomialShapeFunctions(degree=3, family=fem.Polynomial.GAUSS_LEGENDRE)
+
+    test_shape_function_weight(test, Q_1, cube_coord_sampler, CUBE_CENTER_COORDS)
+    test_shape_function_weight(test, Q_2, cube_coord_sampler, CUBE_CENTER_COORDS)
+    test_shape_function_weight(test, Q_3, cube_coord_sampler, CUBE_CENTER_COORDS)
+    test_shape_function_gradient(test, Q_1, cube_coord_sampler, cube_coord_delta_sampler)
+    test_shape_function_gradient(test, Q_2, cube_coord_sampler, cube_coord_delta_sampler)
+    test_shape_function_gradient(test, Q_3, cube_coord_sampler, cube_coord_delta_sampler)
+
+    S_2 = shape.CubeSerendipityShapeFunctions(degree=2, family=fem.Polynomial.LOBATTO_GAUSS_LEGENDRE)
+    S_3 = shape.CubeSerendipityShapeFunctions(degree=3, family=fem.Polynomial.LOBATTO_GAUSS_LEGENDRE)
+
+    test_shape_function_weight(test, S_2, cube_coord_sampler, CUBE_CENTER_COORDS)
+    test_shape_function_weight(test, S_3, cube_coord_sampler, CUBE_CENTER_COORDS)
+    test_shape_function_trace(test, S_2, CUBE_SIDE_CENTER_COORDS)
+    test_shape_function_trace(test, S_3, CUBE_SIDE_CENTER_COORDS)
+    test_shape_function_gradient(test, S_2, cube_coord_sampler, cube_coord_delta_sampler)
+    test_shape_function_gradient(test, S_3, cube_coord_sampler, cube_coord_delta_sampler)
+
+    P_c1 = shape.CubeNonConformingPolynomialShapeFunctions(degree=1)
+    P_c2 = shape.CubeNonConformingPolynomialShapeFunctions(degree=2)
+    P_c3 = shape.CubeNonConformingPolynomialShapeFunctions(degree=3)
+
+    test_shape_function_weight(test, P_c1, cube_coord_sampler, CUBE_CENTER_COORDS)
+    test_shape_function_weight(test, P_c2, cube_coord_sampler, CUBE_CENTER_COORDS)
+    test_shape_function_weight(test, P_c3, cube_coord_sampler, CUBE_CENTER_COORDS)
+    test_shape_function_gradient(test, P_c1, cube_coord_sampler, cube_coord_delta_sampler)
+    test_shape_function_gradient(test, P_c2, cube_coord_sampler, cube_coord_delta_sampler)
+    test_shape_function_gradient(test, P_c3, cube_coord_sampler, cube_coord_delta_sampler)
+
+    N1_1 = shape.CubeNedelecFirstKindShapeFunctions(degree=1)
+    test_shape_function_gradient(test, N1_1, cube_coord_sampler, cube_coord_delta_sampler)
+    RT_1 = shape.CubeRaviartThomasShapeFunctions(degree=1)
+    test_shape_function_gradient(test, RT_1, cube_coord_sampler, cube_coord_delta_sampler)
+
+    wp.synchronize()
+
+
+def test_tri_shape_functions(test, device):
+    TRI_CENTER_COORDS = wp.constant(Coords(1 / 3.0, 1 / 3.0, 1 / 3.0))
+    TRI_SIDE_CENTER_COORDS = wp.constant(Coords(0.0, 0.5, 0.5))
+
+    @wp.func
+    def tri_coord_sampler(state: wp.uint32):
+        a = wp.randf(state)
+        b = wp.randf(state)
+        return Coords(1.0 - a - b, a, b)
+
+    @wp.func
+    def tri_coord_delta_sampler(epsilon: float, state: wp.uint32):
+        param_delta = wp.normalize(wp.vec2(wp.randf(state), wp.randf(state))) * epsilon
+        a = param_delta[0]
+        b = param_delta[1]
+        return param_delta, Coords(-a - b, a, b)
+
+    P_1 = shape.TrianglePolynomialShapeFunctions(degree=1)
+    P_2 = shape.TrianglePolynomialShapeFunctions(degree=2)
+    P_3 = shape.TrianglePolynomialShapeFunctions(degree=3)
+
+    test_shape_function_weight(test, P_1, tri_coord_sampler, TRI_CENTER_COORDS)
+    test_shape_function_weight(test, P_2, tri_coord_sampler, TRI_CENTER_COORDS)
+    test_shape_function_weight(test, P_3, tri_coord_sampler, TRI_CENTER_COORDS)
+    test_shape_function_trace(test, P_1, TRI_SIDE_CENTER_COORDS)
+    test_shape_function_trace(test, P_2, TRI_SIDE_CENTER_COORDS)
+    test_shape_function_trace(test, P_3, TRI_SIDE_CENTER_COORDS)
+    test_shape_function_gradient(test, P_1, tri_coord_sampler, tri_coord_delta_sampler)
+    test_shape_function_gradient(test, P_2, tri_coord_sampler, tri_coord_delta_sampler)
+    test_shape_function_gradient(test, P_3, tri_coord_sampler, tri_coord_delta_sampler)
+
+    P_1d = shape.TriangleNonConformingPolynomialShapeFunctions(degree=1)
+    P_2d = shape.TriangleNonConformingPolynomialShapeFunctions(degree=2)
+    P_3d = shape.TriangleNonConformingPolynomialShapeFunctions(degree=3)
+
+    test_shape_function_weight(test, P_1d, tri_coord_sampler, TRI_CENTER_COORDS)
+    test_shape_function_weight(test, P_2d, tri_coord_sampler, TRI_CENTER_COORDS)
+    test_shape_function_weight(test, P_3d, tri_coord_sampler, TRI_CENTER_COORDS)
+    test_shape_function_gradient(test, P_1d, tri_coord_sampler, tri_coord_delta_sampler)
+    test_shape_function_gradient(test, P_2d, tri_coord_sampler, tri_coord_delta_sampler)
+    test_shape_function_gradient(test, P_3d, tri_coord_sampler, tri_coord_delta_sampler)
+
+    N1_1 = shape.TriangleNedelecFirstKindShapeFunctions(degree=1)
+    test_shape_function_gradient(test, N1_1, tri_coord_sampler, tri_coord_delta_sampler, pure_curl=True)
+
+    RT_1 = shape.TriangleNedelecFirstKindShapeFunctions(degree=1)
+    test_shape_function_gradient(test, RT_1, tri_coord_sampler, tri_coord_delta_sampler, pure_spherical=True)
+
+    wp.synchronize()
+
+
+def test_tet_shape_functions(test, device):
+    TET_CENTER_COORDS = wp.constant(Coords(1 / 4.0, 1 / 4.0, 1 / 4.0))
+    TET_SIDE_CENTER_COORDS = wp.constant(Coords(0.0, 1.0 / 3.0, 1.0 / 3.0))
+
+    @wp.func
+    def tet_coord_sampler(state: wp.uint32):
+        return Coords(wp.randf(state), wp.randf(state), wp.randf(state))
+
+    @wp.func
+    def tet_coord_delta_sampler(epsilon: float, state: wp.uint32):
+        param_delta = wp.normalize(wp.vec3(wp.randf(state), wp.randf(state), wp.randf(state))) * epsilon
+        return param_delta, param_delta
+
+    P_1 = shape.TetrahedronPolynomialShapeFunctions(degree=1)
+    P_2 = shape.TetrahedronPolynomialShapeFunctions(degree=2)
+    P_3 = shape.TetrahedronPolynomialShapeFunctions(degree=3)
+
+    test_shape_function_weight(test, P_1, tet_coord_sampler, TET_CENTER_COORDS)
+    test_shape_function_weight(test, P_2, tet_coord_sampler, TET_CENTER_COORDS)
+    test_shape_function_weight(test, P_3, tet_coord_sampler, TET_CENTER_COORDS)
+    test_shape_function_trace(test, P_1, TET_SIDE_CENTER_COORDS)
+    test_shape_function_trace(test, P_2, TET_SIDE_CENTER_COORDS)
+    test_shape_function_trace(test, P_3, TET_SIDE_CENTER_COORDS)
+    test_shape_function_gradient(test, P_1, tet_coord_sampler, tet_coord_delta_sampler)
+    test_shape_function_gradient(test, P_2, tet_coord_sampler, tet_coord_delta_sampler)
+    test_shape_function_gradient(test, P_3, tet_coord_sampler, tet_coord_delta_sampler)
+
+    P_1d = shape.TetrahedronNonConformingPolynomialShapeFunctions(degree=1)
+    P_2d = shape.TetrahedronNonConformingPolynomialShapeFunctions(degree=2)
+    P_3d = shape.TetrahedronNonConformingPolynomialShapeFunctions(degree=3)
+
+    test_shape_function_weight(test, P_1d, tet_coord_sampler, TET_CENTER_COORDS)
+    test_shape_function_weight(test, P_2d, tet_coord_sampler, TET_CENTER_COORDS)
+    test_shape_function_weight(test, P_3d, tet_coord_sampler, TET_CENTER_COORDS)
+    test_shape_function_gradient(test, P_1d, tet_coord_sampler, tet_coord_delta_sampler)
+    test_shape_function_gradient(test, P_2d, tet_coord_sampler, tet_coord_delta_sampler)
+    test_shape_function_gradient(test, P_3d, tet_coord_sampler, tet_coord_delta_sampler)
+
+    N1_1 = shape.TetrahedronNedelecFirstKindShapeFunctions(degree=1)
+    test_shape_function_gradient(test, N1_1, tet_coord_sampler, tet_coord_delta_sampler, pure_curl=True)
+
+    RT_1 = shape.TetrahedronRaviartThomasShapeFunctions(degree=1)
+    test_shape_function_gradient(test, RT_1, tet_coord_sampler, tet_coord_delta_sampler, pure_spherical=True)
+
+    wp.synchronize()
+
+
+def test_point_basis(test, device):
+    geo = fem.Grid2D(res=wp.vec2i(2))
+
+    domain = fem.Cells(geo)
+
+    quadrature = fem.RegularQuadrature(domain, order=2, family=fem.Polynomial.GAUSS_LEGENDRE)
+    point_basis = fem.PointBasisSpace(quadrature)
+
+    point_space = fem.make_collocated_function_space(point_basis)
+    point_test = fem.make_test(point_space, domain=domain)
+
+    # Sample at particle positions
+    ones = fem.integrate(linear_form, fields={"u": point_test}, nodal=True)
+    test.assertAlmostEqual(np.sum(ones.numpy()), 1.0, places=5)
+
+    # Sampling outside of particle positions
+    other_quadrature = fem.RegularQuadrature(domain, order=2, family=fem.Polynomial.LOBATTO_GAUSS_LEGENDRE)
+    zeros = fem.integrate(linear_form, quadrature=other_quadrature, fields={"u": point_test})
+
+    test.assertAlmostEqual(np.sum(zeros.numpy()), 0.0, places=5)
+
+    # test point basis with variable points per cell
+    points = wp.array([[0.25, 0.33], [0.33, 0.25], [0.8, 0.8]], dtype=wp.vec2)
+    pic = fem.PicQuadrature(domain, positions=points)
+
+    test.assertEqual(pic.active_cell_count(), 2)
+    test.assertEqual(pic.total_point_count(), 3)
+    test.assertEqual(pic.max_points_per_element(), 2)
+
+    point_basis = fem.PointBasisSpace(pic)
+    point_space = fem.make_collocated_function_space(point_basis)
+    point_test = fem.make_test(point_space, domain=domain)
+    test.assertEqual(point_test.space_restriction.node_count(), 3)
+
+    ones = fem.integrate(linear_form, fields={"u": point_test}, quadrature=pic)
+    test.assertAlmostEqual(np.sum(ones.numpy()), pic.active_cell_count() / geo.cell_count(), places=5)
+
+    zeros = fem.integrate(linear_form, quadrature=other_quadrature, fields={"u": point_test})
+    test.assertAlmostEqual(np.sum(zeros.numpy()), 0.0, places=5)
+
+    linear_vec = fem.make_polynomial_space(geo, dtype=wp.vec2)
+    linear_test = fem.make_test(linear_vec)
+    point_trial = fem.make_trial(point_space)
+
+    mat = fem.integrate(vector_divergence_form, fields={"u": linear_test, "q": point_trial}, quadrature=pic)
+    test.assertEqual(mat.nrow, 9)
+    test.assertEqual(mat.ncol, 3)
+    test.assertEqual(mat.nnz_sync(), 12)
+
+
+@fem.integrand
+def _bicubic(s: Sample, domain: Domain):
+    x = domain(s)
+    return wp.pow(x[0], 3.0) * wp.pow(x[1], 3.0)
+
+
+@fem.integrand
+def _piecewise_constant(s: Sample):
+    return float(s.element_index)
+
+
+def test_particle_quadratures(test, device):
+    geo = fem.Grid2D(res=wp.vec2i(2))
+
+    domain = fem.Cells(geo)
+
+    # Explicit quadrature
+    points, weights = domain.reference_element().instantiate_quadrature(order=4, family=fem.Polynomial.GAUSS_LEGENDRE)
+    points_per_cell = len(points)
+
+    points = points * domain.element_count()
+    weights = weights * domain.element_count()
+
+    points = wp.array(points, shape=(domain.element_count(), points_per_cell), dtype=Coords, device=device)
+    weights = wp.array(weights, shape=(domain.element_count(), points_per_cell), dtype=float, device=device)
+
+    explicit_quadrature = fem.ExplicitQuadrature(domain, points, weights)
+
+    test.assertEqual(explicit_quadrature.max_points_per_element(), points_per_cell)
+    test.assertEqual(explicit_quadrature.total_point_count(), points_per_cell * geo.cell_count())
+
+    # test integration accuracy
+    val = fem.integrate(_bicubic, quadrature=explicit_quadrature)
+    test.assertAlmostEqual(val, 1.0 / 16, places=5)
+
+    # test indexing validity
+    arr = wp.empty(explicit_quadrature.total_point_count(), dtype=float)
+    fem.interpolate(_piecewise_constant, dest=arr, quadrature=explicit_quadrature)
+    assert_np_equal(arr.numpy(), np.arange(geo.cell_count()).repeat(points_per_cell))
+
+    # PIC quadrature
+    element_indices = wp.array([3, 3, 2], dtype=int, device=device)
+    element_coords = wp.array(
+        [
+            [0.25, 0.5, 0.0],
+            [0.5, 0.25, 0.0],
+            [0.5, 0.5, 0.0],
+        ],
+        dtype=Coords,
+        device=device,
+    )
+
+    pic_quadrature = fem.PicQuadrature(domain, positions=(element_indices, element_coords))
+
+    test.assertEqual(pic_quadrature.max_points_per_element(), 2)
+    test.assertEqual(pic_quadrature.total_point_count(), 3)
+    test.assertEqual(pic_quadrature.active_cell_count(), 2)
+
+    # Test integration accuracy
+    val = fem.integrate(_piecewise_constant, quadrature=pic_quadrature)
+    test.assertAlmostEqual(val, 1.25, places=5)
+
+    # Test differentiability of PicQuadrature w.r.t positions and measures
+    points = wp.array([[0.25, 0.33], [0.33, 0.25], [0.8, 0.8]], dtype=wp.vec2, device=device, requires_grad=True)
+    measures = wp.ones(3, dtype=float, device=device, requires_grad=True)
+
+    tape = wp.Tape()
+    with tape:
+        pic = fem.PicQuadrature(domain, positions=points, measures=measures, requires_grad=True)
+
+    pic.arg_value(device).particle_coords.grad.fill_(1.0)
+    pic.arg_value(device).particle_fraction.grad.fill_(1.0)
+    tape.backward()
+
+    assert_np_equal(points.grad.numpy(), np.full((3, 2), 2.0))  # == 1.0 / cell_size
+    assert_np_equal(measures.grad.numpy(), np.full(3, 4.0))  # == 1.0 / cell_area
+
+
+@fem.integrand
+def _value_at_node(s: fem.Sample, f: fem.Field, values: wp.array(dtype=float)):
+    node_index = fem.operator.node_partition_index(f, s.qp_index)
+    return values[node_index]
+
+
+def test_nodal_quadrature(test, device):
+    geo = fem.Grid2D(res=wp.vec2i(2))
+
+    domain = fem.Cells(geo)
+
+    space = fem.make_polynomial_space(geo, degree=2, discontinuous=True, family=fem.Polynomial.GAUSS_LEGENDRE)
+    nodal_quadrature = fem.NodalQuadrature(domain, space)
+
+    test.assertEqual(nodal_quadrature.max_points_per_element(), 9)
+    test.assertEqual(nodal_quadrature.total_point_count(), 9 * geo.cell_count())
+
+    val = fem.integrate(_bicubic, quadrature=nodal_quadrature)
+    test.assertAlmostEqual(val, 1.0 / 16, places=5)
+
+    # test accessing data associated to a given node
+
+    piecewise_constant_space = fem.make_polynomial_space(geo, degree=0)
+    geo_partition = fem.LinearGeometryPartition(geo, 3, 4)
+    space_partition = fem.make_space_partition(piecewise_constant_space, geo_partition)
+    field = fem.make_discrete_field(piecewise_constant_space, space_partition=space_partition)
+
+    partition_domain = fem.Cells(geo_partition)
+    partition_nodal_quadrature = fem.NodalQuadrature(partition_domain, piecewise_constant_space)
+
+    partition_node_values = wp.array([5.0], dtype=float)
+    val = fem.integrate(
+        _value_at_node,
+        quadrature=partition_nodal_quadrature,
+        fields={"f": field},
+        values={"values": partition_node_values},
+    )
+    test.assertAlmostEqual(val, 5.0 / geo.cell_count(), places=5)
+
+
+@wp.func
+def aniso_bicubic_fn(x: wp.vec2, scale: wp.vec2):
+    return wp.pow(x[0] * scale[0], 3.0) * wp.pow(x[1] * scale[1], 3.0)
+
+
+@wp.func
+def aniso_bicubic_grad(x: wp.vec2, scale: wp.vec2):
+    return wp.vec2(
+        3.0 * scale[0] * wp.pow(x[0] * scale[0], 2.0) * wp.pow(x[1] * scale[1], 3.0),
+        3.0 * scale[1] * wp.pow(x[0] * scale[0], 3.0) * wp.pow(x[1] * scale[1], 2.0),
+    )
+
+
+def test_implicit_fields(test, device):
+    geo = fem.Grid2D(res=wp.vec2i(2))
+    domain = fem.Cells(geo)
+    boundary = fem.BoundarySides(geo)
+
+    space = fem.make_polynomial_space(geo)
+    vec_space = fem.make_polynomial_space(geo, dtype=wp.vec2)
+    discrete_field = fem.make_discrete_field(space)
+    discrete_vec_field = fem.make_discrete_field(vec_space)
+
+    # Uniform
+
+    uniform = fem.UniformField(domain, 5.0)
+    fem.interpolate(uniform, dest=discrete_field)
+    assert_np_equal(discrete_field.dof_values.numpy(), np.full(9, 5.0))
+
+    fem.interpolate(grad_field, fields={"p": uniform}, dest=discrete_vec_field)
+    assert_np_equal(discrete_vec_field.dof_values.numpy(), np.zeros((9, 2)))
+
+    uniform.value = 2.0
+    fem.interpolate(uniform.trace(), dest=fem.make_restriction(discrete_field, domain=boundary))
+    assert_np_equal(discrete_field.dof_values.numpy(), np.array([2.0] * 4 + [5.0] + [2.0] * 4))
+
+    # Implicit
+
+    implicit = fem.ImplicitField(
+        domain, func=aniso_bicubic_fn, values={"scale": wp.vec2(2.0, 4.0)}, grad_func=aniso_bicubic_grad
+    )
+    fem.interpolate(implicit, dest=discrete_field)
+    assert_np_equal(
+        discrete_field.dof_values.numpy(),
+        np.array([0.0, 0.0, 0.0, 0.0, 2.0**3, 4.0**3, 0.0, 2.0**3 * 2.0**3, 4.0**3 * 2.0**3]),
+    )
+
+    fem.interpolate(grad_field, fields={"p": implicit}, dest=discrete_vec_field)
+    assert_np_equal(discrete_vec_field.dof_values.numpy()[0], np.zeros(2))
+    assert_np_equal(discrete_vec_field.dof_values.numpy()[-1], np.full(2, (2.0**9.0 * 3.0)))
+
+    implicit.values.scale = wp.vec2(-2.0, -2.0)
+    fem.interpolate(implicit.trace(), dest=fem.make_restriction(discrete_field, domain=boundary))
+    assert_np_equal(
+        discrete_field.dof_values.numpy(),
+        np.array([0.0, 0.0, 0.0, 0.0, 2.0**3, 2.0**3, 0.0, 2.0**3, 4.0**3]),
+    )
+
+    # Nonconforming
+
+    geo2 = fem.Grid2D(res=wp.vec2i(1), bounds_lo=wp.vec2(0.25, 0.25), bounds_hi=wp.vec2(2.0, 2.0))
+    domain2 = fem.Cells(geo2)
+    boundary2 = fem.BoundarySides(geo2)
+    space2 = fem.make_polynomial_space(geo2)
+    vec_space2 = fem.make_polynomial_space(geo2, dtype=wp.vec2)
+    discrete_field2 = fem.make_discrete_field(space2)
+    discrete_vec_field2 = fem.make_discrete_field(vec_space2)
+
+    nonconforming = fem.NonconformingField(domain2, discrete_field, background=5.0)
+    fem.interpolate(
+        nonconforming,
+        dest=discrete_field2,
+    )
+    assert_np_equal(discrete_field2.dof_values.numpy(), np.array([2.0] + [5.0] * 3))
+
+    fem.interpolate(grad_field, fields={"p": nonconforming}, dest=discrete_vec_field2)
+    assert_np_equal(discrete_vec_field2.dof_values.numpy()[0], np.full(2, 8.0))
+    assert_np_equal(discrete_vec_field2.dof_values.numpy()[-1], np.zeros(2))
+
+    discrete_field2.dof_values.zero_()
+    fem.interpolate(
+        nonconforming.trace(),
+        dest=fem.make_restriction(discrete_field2, domain=boundary2),
+    )
+    assert_np_equal(discrete_field2.dof_values.numpy(), np.array([2.0] + [5.0] * 3))
+
+
+@fem.integrand
+def _expect_pure_curl(s: fem.Sample, field: fem.Field):
+    sym_grad = fem.D(field, s)
+    wp.expect_near(wp.ddot(sym_grad, sym_grad), 0.0)
+    return 0.0
+
+
+@fem.integrand
+def _expect_pure_spherical(s: fem.Sample, field: fem.Field):
+    grad = fem.grad(field, s)
+    deviatoric_part = grad - spherical_part(grad)
+    wp.expect_near(wp.ddot(deviatoric_part, deviatoric_part), 0.0)
+    return 0.0
+
+
+@fem.integrand
+def _expect_normal_continuity(s: fem.Sample, domain: fem.Domain, field: fem.Field):
+    nor = fem.normal(domain, s)
+    wp.expect_near(wp.dot(fem.inner(field, s), nor), wp.dot(fem.outer(field, s), nor), 0.0001)
+    return 0.0
+
+
+@fem.integrand
+def _expect_tangential_continuity(s: fem.Sample, domain: fem.Domain, field: fem.Field):
+    nor = fem.normal(domain, s)
+    in_s = fem.inner(field, s)
+    out_s = fem.outer(field, s)
+    in_t = in_s - wp.dot(in_s, nor) * nor
+    out_t = out_s - wp.dot(out_s, nor) * nor
+
+    _expect_near(in_t, out_t, 0.0001)
+    return 0.0
+
+
+def test_vector_spaces(test, device):
+    # Test covariant / contravariant mappings
+
+    with wp.ScopedDevice(device):
+        positions, hex_vidx = _gen_quadmesh(3)
+
+        geo = fem.Quadmesh2D(quad_vertex_indices=hex_vidx, positions=positions)
+
+        curl_space = fem.make_polynomial_space(geo, element_basis=fem.ElementBasis.NEDELEC_FIRST_KIND)
+        curl_test = fem.make_test(curl_space)
+
+        curl_field = curl_space.make_field()
+        curl_field.dof_values = wp.array(np.linspace(0.0, 1.0, curl_space.node_count()), dtype=float)
+
+        fem.interpolate(
+            _expect_tangential_continuity,
+            quadrature=fem.RegularQuadrature(fem.Sides(geo), order=2),
+            fields={"field": curl_field.trace()},
+        )
+
+        div_space = fem.make_polynomial_space(geo, element_basis=fem.ElementBasis.RAVIART_THOMAS)
+        div_test = fem.make_test(div_space)
+
+        div_field = div_space.make_field()
+        div_field.dof_values = wp.array(np.linspace(0.0, 1.0, div_space.node_count()), dtype=float)
+
+        fem.interpolate(
+            _expect_normal_continuity,
+            quadrature=fem.RegularQuadrature(fem.Sides(geo), order=2),
+            fields={"field": div_field.trace()},
+        )
+
+    with wp.ScopedDevice(device):
+        positions, hex_vidx = _gen_hexmesh(3)
+
+        geo = fem.Hexmesh(hex_vertex_indices=hex_vidx, positions=positions)
+
+        curl_space = fem.make_polynomial_space(geo, element_basis=fem.ElementBasis.NEDELEC_FIRST_KIND)
+        curl_test = fem.make_test(curl_space)
+
+        curl_field = curl_space.make_field()
+        curl_field.dof_values = wp.array(np.linspace(0.0, 1.0, curl_space.node_count()), dtype=float)
+
+        fem.interpolate(
+            _expect_tangential_continuity,
+            quadrature=fem.RegularQuadrature(fem.Sides(geo), order=2),
+            fields={"field": curl_field.trace()},
+        )
+
+        div_space = fem.make_polynomial_space(geo, element_basis=fem.ElementBasis.RAVIART_THOMAS)
+        div_test = fem.make_test(div_space)
+
+        div_field = div_space.make_field()
+        div_field.dof_values = wp.array(np.linspace(0.0, 1.0, div_space.node_count()), dtype=float)
+
+        fem.interpolate(
+            _expect_normal_continuity,
+            quadrature=fem.RegularQuadrature(fem.Sides(geo), order=2),
+            fields={"field": div_field.trace()},
+        )
+
+    return
+
+    with wp.ScopedDevice(device):
+        positions, tri_vidx = _gen_trimesh(3, 5)
+
+        geo = fem.Trimesh2D(tri_vertex_indices=tri_vidx, positions=positions)
+
+        curl_space = fem.make_polynomial_space(geo, element_basis=fem.ElementBasis.NEDELEC_FIRST_KIND)
+        curl_test = fem.make_test(curl_space)
+
+        fem.integrate(_expect_pure_curl, fields={"field": curl_test}, assembly="generic")
+
+        curl_field = curl_space.make_field()
+        curl_field.dof_values.fill_(1.0)
+        fem.interpolate(
+            _expect_pure_curl, quadrature=fem.RegularQuadrature(fem.Cells(geo), order=2), fields={"field": curl_field}
+        )
+
+        fem.interpolate(
+            _expect_tangential_continuity,
+            quadrature=fem.RegularQuadrature(fem.Sides(geo), order=2),
+            fields={"field": curl_field.trace()},
+        )
+
+        div_space = fem.make_polynomial_space(geo, element_basis=fem.ElementBasis.RAVIART_THOMAS)
+        div_test = fem.make_test(div_space)
+
+        fem.integrate(_expect_pure_spherical, fields={"field": div_test}, assembly="generic")
+
+        div_field = div_space.make_field()
+        div_field.dof_values.fill_(1.0)
+        fem.interpolate(
+            _expect_pure_spherical,
+            quadrature=fem.RegularQuadrature(fem.Cells(geo), order=2),
+            fields={"field": div_field},
+        )
+
+        fem.interpolate(
+            _expect_normal_continuity,
+            quadrature=fem.RegularQuadrature(fem.Sides(geo), order=2),
+            fields={"field": div_field.trace()},
+        )
+
+    with wp.ScopedDevice(device):
+        positions, tet_vidx = _gen_tetmesh(3, 5, 7)
+
+        geo = fem.Tetmesh(tet_vertex_indices=tet_vidx, positions=positions)
+
+        curl_space = fem.make_polynomial_space(geo, element_basis=fem.ElementBasis.NEDELEC_FIRST_KIND)
+        curl_test = fem.make_test(curl_space)
+
+        fem.integrate(_expect_pure_curl, fields={"field": curl_test}, assembly="generic")
+
+        curl_field = curl_space.make_field()
+        curl_field.dof_values.fill_(1.0)
+        fem.interpolate(
+            _expect_pure_curl, quadrature=fem.RegularQuadrature(fem.Cells(geo), order=2), fields={"field": curl_field}
+        )
+
+        fem.interpolate(
+            _expect_tangential_continuity,
+            quadrature=fem.RegularQuadrature(fem.Sides(geo), order=1),
+            fields={"field": curl_field.trace()},
+        )
+
+        div_space = fem.make_polynomial_space(geo, element_basis=fem.ElementBasis.RAVIART_THOMAS)
+        div_test = fem.make_test(div_space)
+
+        fem.integrate(_expect_pure_spherical, fields={"field": div_test}, assembly="generic")
+
+        div_field = div_space.make_field()
+        div_field.dof_values.fill_(1.0)
+        fem.interpolate(
+            _expect_pure_spherical,
+            quadrature=fem.RegularQuadrature(fem.Cells(geo), order=2),
+            fields={"field": div_field},
+        )
+
+        fem.interpolate(
+            _expect_normal_continuity,
+            quadrature=fem.RegularQuadrature(fem.Sides(geo), order=0),
+            fields={"field": div_field.trace()},
+        )
+
+
+@wp.kernel
+def test_qr_eigenvalues():
+    tol = 1.0e-8
+
+    # zero
+    Zero = wp.mat33(0.0)
+    Id = wp.identity(n=3, dtype=float)
+    D3, P3 = symmetric_eigenvalues_qr(Zero, tol * tol)
+    wp.expect_eq(D3, wp.vec3(0.0))
+    wp.expect_eq(P3, Id)
+
+    # Identity
+    D3, P3 = symmetric_eigenvalues_qr(Id, tol * tol)
+    wp.expect_eq(D3, wp.vec3(1.0))
+    wp.expect_eq(wp.transpose(P3) * P3, Id)
+
+    # rank 1
+    v = wp.vec4(0.0, 1.0, 1.0, 0.0)
+    Rank1 = wp.outer(v, v)
+    D4, P4 = symmetric_eigenvalues_qr(Rank1, tol * tol)
+    wp.expect_near(wp.max(D4), wp.length_sq(v), tol)
+    Err4 = wp.transpose(P4) * wp.diag(D4) * P4 - Rank1
+    wp.expect_near(wp.ddot(Err4, Err4), 0.0, tol)
+
+    # rank 2
+    v2 = wp.vec4(0.0, 0.5, -0.5, 0.0)
+    Rank2 = Rank1 + wp.outer(v2, v2)
+    D4, P4 = symmetric_eigenvalues_qr(Rank2, tol * tol)
+    wp.expect_near(wp.max(D4), wp.length_sq(v), tol)
+    wp.expect_near(D4[0] + D4[1] + D4[2] + D4[3], wp.length_sq(v) + wp.length_sq(v2), tol)
+    Err4 = wp.transpose(P4) * wp.diag(D4) * P4 - Rank2
+    wp.expect_near(wp.ddot(Err4, Err4), 0.0, tol)
+
+    # rank 4
+    v3 = wp.vec4(1.0, 2.0, 3.0, 4.0)
+    v4 = wp.vec4(2.0, 1.0, 0.0, -1.0)
+    Rank4 = Rank2 + wp.outer(v3, v3) + wp.outer(v4, v4)
+    D4, P4 = symmetric_eigenvalues_qr(Rank4, tol * tol)
+    Err4 = wp.transpose(P4) * wp.diag(D4) * P4 - Rank4
+    wp.expect_near(wp.ddot(Err4, Err4), 0.0, tol)
+
+    # test robustness to low requested tolerance
+    Rank6 = wp.matrix_from_cols(
+        vec6f(0.00171076, 0.0, 0.0, 0.0, 0.0, 0.0),
+        vec6f(0.0, 0.00169935, 6.14367e-06, -3.52589e-05, 3.02397e-05, -1.53458e-11),
+        vec6f(0.0, 6.14368e-06, 0.00172217, 2.03568e-05, 1.74589e-05, -2.92627e-05),
+        vec6f(0.0, -3.52589e-05, 2.03568e-05, 0.00172178, 2.53422e-05, 3.02397e-05),
+        vec6f(0.0, 3.02397e-05, 1.74589e-05, 2.53422e-05, 0.00171114, 3.52589e-05),
+        vec6f(0.0, 6.42993e-12, -2.92627e-05, 3.02397e-05, 3.52589e-05, 0.00169935),
+    )
+    D6, P6 = symmetric_eigenvalues_qr(Rank6, 0.0)
+    Err6 = wp.transpose(P6) * wp.diag(D6) * P6 - Rank6
+    wp.expect_near(wp.ddot(Err6, Err6), 0.0, 1.0e-13)
+
+
+@wp.kernel
+def test_qr_inverse():
+    rng = wp.rand_init(4356, wp.tid())
+    M = wp.mat33(
+        wp.randf(rng, 0.0, 10.0),
+        wp.randf(rng, 0.0, 10.0),
+        wp.randf(rng, 0.0, 10.0),
+        wp.randf(rng, 0.0, 10.0),
+        wp.randf(rng, 0.0, 10.0),
+        wp.randf(rng, 0.0, 10.0),
+        wp.randf(rng, 0.0, 10.0),
+        wp.randf(rng, 0.0, 10.0),
+        wp.randf(rng, 0.0, 10.0),
+    )
+
+    if wp.determinant(M) != 0.0:
+        tol = 1.0e-8
+        Mi = inverse_qr(M)
+        Id = wp.identity(n=3, dtype=float)
+        Err = M * Mi - Id
+        wp.expect_near(wp.ddot(Err, Err), 0.0, tol)
+        Err = Mi * M - Id
+        wp.expect_near(wp.ddot(Err, Err), 0.0, tol)
+
+
+def test_array_axpy(test, device):
+    N = 10
+    alpha = 0.5
+    beta = 4.0
+
+    x = wp.full(N, 2.0, device=device, dtype=float, requires_grad=True)
+    y = wp.array(np.arange(N), device=device, dtype=wp.float64, requires_grad=True)
+
+    tape = wp.Tape()
+    with tape:
+        fem.utils.array_axpy(x=x, y=y, alpha=alpha, beta=beta)
+
+    assert_np_equal(x.numpy(), np.full(N, 2.0))
+    assert_np_equal(y.numpy(), alpha * x.numpy() + beta * np.arange(N))
+
+    y.grad.fill_(1.0)
+    tape.backward()
+
+    assert_np_equal(x.grad.numpy(), alpha * np.ones(N))
+    assert_np_equal(y.grad.numpy(), beta * np.ones(N))
+
+
+devices = get_test_devices()
+cuda_devices = get_selected_cuda_test_devices()
+
+
+class TestFem(unittest.TestCase):
+    pass
+
+
+add_function_test(TestFem, "test_regular_quadrature", test_regular_quadrature)
+add_function_test(TestFem, "test_closest_point_queries", test_closest_point_queries)
+add_function_test(TestFem, "test_grad_decomposition", test_grad_decomposition, devices=devices)
+add_function_test(TestFem, "test_integrate_gradient", test_integrate_gradient, devices=devices)
+add_function_test(TestFem, "test_interpolate_gradient", test_interpolate_gradient, devices=devices)
+add_function_test(TestFem, "test_vector_divergence_theorem", test_vector_divergence_theorem, devices=devices)
+add_function_test(TestFem, "test_tensor_divergence_theorem", test_tensor_divergence_theorem, devices=devices)
+add_function_test(TestFem, "test_grid_2d", test_grid_2d, devices=devices)
+add_function_test(TestFem, "test_triangle_mesh", test_triangle_mesh, devices=devices)
+add_function_test(TestFem, "test_quad_mesh", test_quad_mesh, devices=devices)
+add_function_test(TestFem, "test_grid_3d", test_grid_3d, devices=devices)
+add_function_test(TestFem, "test_tet_mesh", test_tet_mesh, devices=devices)
+add_function_test(TestFem, "test_hex_mesh", test_hex_mesh, devices=devices)
+add_function_test(TestFem, "test_nanogrid", test_nanogrid, devices=cuda_devices)
+add_function_test(TestFem, "test_adaptive_nanogrid", test_adaptive_nanogrid, devices=cuda_devices)
+add_function_test(TestFem, "test_deformed_geometry", test_deformed_geometry, devices=devices)
+add_function_test(TestFem, "test_vector_spaces", test_vector_spaces, devices=devices)
+add_function_test(TestFem, "test_dof_mapper", test_dof_mapper)
+add_function_test(TestFem, "test_point_basis", test_point_basis)
+add_function_test(TestFem, "test_particle_quadratures", test_particle_quadratures)
+add_function_test(TestFem, "test_nodal_quadrature", test_nodal_quadrature)
+add_function_test(TestFem, "test_implicit_fields", test_implicit_fields)
+
+
+class TestFemUtilities(unittest.TestCase):
+    pass
+
+
+add_kernel_test(TestFemUtilities, test_qr_eigenvalues, dim=1, devices=devices)
+add_kernel_test(TestFemUtilities, test_qr_inverse, dim=100, devices=devices)
+add_function_test(TestFemUtilities, "test_array_axpy", test_array_axpy)
+
+
+class TestFemShapeFunctions(unittest.TestCase):
+    pass
+
+
+add_function_test(TestFemShapeFunctions, "test_square_shape_functions", test_square_shape_functions)
+add_function_test(TestFemShapeFunctions, "test_cube_shape_functions", test_cube_shape_functions)
+add_function_test(TestFemShapeFunctions, "test_tri_shape_functions", test_tri_shape_functions)
+add_function_test(TestFemShapeFunctions, "test_tet_shape_functions", test_tet_shape_functions)
+
+
+if __name__ == "__main__":
+    wp.clear_kernel_cache()
+    unittest.main(verbosity=2, failfast=True)