warp-lang 1.0.0b5__py3-none-manylinux2014_x86_64.whl → 1.0.0b6__py3-none-manylinux2014_x86_64.whl

warp/tests/test_vec_scalar_ops.py

@@ -0,0 +1,2099 @@
+# Copyright (c) 2022 NVIDIA CORPORATION.  All rights reserved.
+# NVIDIA CORPORATION and its licensors retain all intellectual property
+# and proprietary rights in and to this software, related documentation
+# and any modifications thereto.  Any use, reproduction, disclosure or
+# distribution of this software and related documentation without an express
+# license agreement from NVIDIA CORPORATION is strictly prohibited.
+
+import unittest
+
+import numpy as np
+
+import warp as wp
+from warp.tests.unittest_utils import *
+
+wp.init()
+
+np_signed_int_types = [
+    np.int8,
+    np.int16,
+    np.int32,
+    np.int64,
+    np.byte,
+]
+
+np_unsigned_int_types = [
+    np.uint8,
+    np.uint16,
+    np.uint32,
+    np.uint64,
+    np.ubyte,
+]
+
+np_int_types = np_signed_int_types + np_unsigned_int_types
+
+np_float_types = [np.float16, np.float32, np.float64]
+
+np_scalar_types = np_int_types + np_float_types
+
+
+def randvals(rng, shape, dtype):
+    if dtype in np_float_types:
+        return rng.standard_normal(size=shape).astype(dtype)
+    elif dtype in [np.int8, np.uint8, np.byte, np.ubyte]:
+        return rng.integers(1, high=3, size=shape, dtype=dtype)
+    return rng.integers(1, high=5, size=shape, dtype=dtype)
+
+
+kernel_cache = dict()
+
+
+def getkernel(func, suffix=""):
+    key = func.__name__ + "_" + suffix
+    if key not in kernel_cache:
+        kernel_cache[key] = wp.Kernel(func=func, key=key)
+    return kernel_cache[key]
+
+
+def get_select_kernel(dtype):
+    def output_select_kernel_fn(
+        input: wp.array(dtype=dtype),
+        index: int,
+        out: wp.array(dtype=dtype),
+    ):
+        out[0] = input[index]
+
+    return getkernel(output_select_kernel_fn, suffix=dtype.__name__)
+
+
+def get_select_kernel2(dtype):
+    def output_select_kernel2_fn(
+        input: wp.array(dtype=dtype, ndim=2),
+        index0: int,
+        index1: int,
+        out: wp.array(dtype=dtype),
+    ):
+        out[0] = input[index0, index1]
+
+    return getkernel(output_select_kernel2_fn, suffix=dtype.__name__)
+
+
+def test_arrays(test, device, dtype):
+    rng = np.random.default_rng(123)
+
+    wptype = wp.types.np_dtype_to_warp_type[np.dtype(dtype)]
+    vec2 = wp.types.vector(length=2, dtype=wptype)
+    vec3 = wp.types.vector(length=3, dtype=wptype)
+    vec4 = wp.types.vector(length=4, dtype=wptype)
+    vec5 = wp.types.vector(length=5, dtype=wptype)
+
+    v2_np = randvals(rng, (10, 2), dtype)
+    v3_np = randvals(rng, (10, 3), dtype)
+    v4_np = randvals(rng, (10, 4), dtype)
+    v5_np = randvals(rng, (10, 5), dtype)
+
+    v2 = wp.array(v2_np, dtype=vec2, requires_grad=True, device=device)
+    v3 = wp.array(v3_np, dtype=vec3, requires_grad=True, device=device)
+    v4 = wp.array(v4_np, dtype=vec4, requires_grad=True, device=device)
+    v5 = wp.array(v5_np, dtype=vec5, requires_grad=True, device=device)
+
+    assert_np_equal(v2.numpy(), v2_np, tol=1.0e-6)
+    assert_np_equal(v3.numpy(), v3_np, tol=1.0e-6)
+    assert_np_equal(v4.numpy(), v4_np, tol=1.0e-6)
+    assert_np_equal(v5.numpy(), v5_np, tol=1.0e-6)
+
+    vec2 = wp.types.vector(length=2, dtype=wptype)
+    vec3 = wp.types.vector(length=3, dtype=wptype)
+    vec4 = wp.types.vector(length=4, dtype=wptype)
+
+    v2 = wp.array(v2_np, dtype=vec2, requires_grad=True, device=device)
+    v3 = wp.array(v3_np, dtype=vec3, requires_grad=True, device=device)
+    v4 = wp.array(v4_np, dtype=vec4, requires_grad=True, device=device)
+
+    assert_np_equal(v2.numpy(), v2_np, tol=1.0e-6)
+    assert_np_equal(v3.numpy(), v3_np, tol=1.0e-6)
+    assert_np_equal(v4.numpy(), v4_np, tol=1.0e-6)
+
+
+def test_components(test, device, dtype):
+    # test accessing vector components from Python - this is especially important
+    # for float16, which requires special handling internally
+
+    wptype = wp.types.np_dtype_to_warp_type[np.dtype(dtype)]
+    vec3 = wp.types.vector(length=3, dtype=wptype)
+
+    v = vec3(1, 2, 3)
+
+    # test __getitem__ for individual components
+    test.assertEqual(v[0], 1)
+    test.assertEqual(v[1], 2)
+    test.assertEqual(v[2], 3)
+
+    # test __getitem__ for slices
+    s = v[:]
+    test.assertEqual(s[0], 1)
+    test.assertEqual(s[1], 2)
+    test.assertEqual(s[2], 3)
+
+    s = v[1:]
+    test.assertEqual(s[0], 2)
+    test.assertEqual(s[1], 3)
+
+    s = v[:2]
+    test.assertEqual(s[0], 1)
+    test.assertEqual(s[1], 2)
+
+    s = v[::2]
+    test.assertEqual(s[0], 1)
+    test.assertEqual(s[1], 3)
+
+    # test __setitem__ for individual components
+    v[0] = 4
+    v[1] = 5
+    v[2] = 6
+    test.assertEqual(v[0], 4)
+    test.assertEqual(v[1], 5)
+    test.assertEqual(v[2], 6)
+
+    # test __setitem__ for slices
+    v[:] = [7, 8, 9]
+    test.assertEqual(v[0], 7)
+    test.assertEqual(v[1], 8)
+    test.assertEqual(v[2], 9)
+
+    v[1:] = [10, 11]
+    test.assertEqual(v[0], 7)
+    test.assertEqual(v[1], 10)
+    test.assertEqual(v[2], 11)
+
+    v[:2] = [12, 13]
+    test.assertEqual(v[0], 12)
+    test.assertEqual(v[1], 13)
+    test.assertEqual(v[2], 11)
+
+    v[::2] = [14, 15]
+    test.assertEqual(v[0], 14)
+    test.assertEqual(v[1], 13)
+    test.assertEqual(v[2], 15)
+
+
+def test_py_arithmetic_ops(test, device, dtype):
+    wptype = wp.types.np_dtype_to_warp_type[np.dtype(dtype)]
+
+    def make_vec(*args):
+        if wptype in wp.types.int_types:
+            # Cast to the correct integer type to simulate wrapping.
+            return tuple(wptype._type_(x).value for x in args)
+
+        return args
+
+    vec_cls = wp.vec(3, wptype)
+
+    v = vec_cls(1, -2, 3)
+    test.assertSequenceEqual(+v, make_vec(1, -2, 3))
+    test.assertSequenceEqual(-v, make_vec(-1, 2, -3))
+    test.assertSequenceEqual(v + vec_cls(5, 5, 5), make_vec(6, 3, 8))
+    test.assertSequenceEqual(v - vec_cls(5, 5, 5), make_vec(-4, -7, -2))
+
+    v = vec_cls(2, 4, 6)
+    test.assertSequenceEqual(v * wptype(2), make_vec(4, 8, 12))
+    test.assertSequenceEqual(wptype(2) * v, make_vec(4, 8, 12))
+    test.assertSequenceEqual(v / wptype(2), make_vec(1, 2, 3))
+    test.assertSequenceEqual(wptype(24) / v, make_vec(12, 6, 4))
+
+
+def test_constructors(test, device, dtype, register_kernels=False):
+    rng = np.random.default_rng(123)
+
+    tol = {
+        np.float16: 5.0e-3,
+        np.float32: 1.0e-6,
+        np.float64: 1.0e-8,
+    }.get(dtype, 0)
+
+    wptype = wp.types.np_dtype_to_warp_type[np.dtype(dtype)]
+    vec2 = wp.types.vector(length=2, dtype=wptype)
+    vec3 = wp.types.vector(length=3, dtype=wptype)
+    vec4 = wp.types.vector(length=4, dtype=wptype)
+    vec5 = wp.types.vector(length=5, dtype=wptype)
+
+    def check_scalar_constructor(
+        input: wp.array(dtype=wptype),
+        v2: wp.array(dtype=vec2),
+        v3: wp.array(dtype=vec3),
+        v4: wp.array(dtype=vec4),
+        v5: wp.array(dtype=vec5),
+        v20: wp.array(dtype=wptype),
+        v21: wp.array(dtype=wptype),
+        v30: wp.array(dtype=wptype),
+        v31: wp.array(dtype=wptype),
+        v32: wp.array(dtype=wptype),
+        v40: wp.array(dtype=wptype),
+        v41: wp.array(dtype=wptype),
+        v42: wp.array(dtype=wptype),
+        v43: wp.array(dtype=wptype),
+        v50: wp.array(dtype=wptype),
+        v51: wp.array(dtype=wptype),
+        v52: wp.array(dtype=wptype),
+        v53: wp.array(dtype=wptype),
+        v54: wp.array(dtype=wptype),
+    ):
+        v2result = vec2(input[0])
+        v3result = vec3(input[0])
+        v4result = vec4(input[0])
+        v5result = vec5(input[0])
+
+        v2[0] = v2result
+        v3[0] = v3result
+        v4[0] = v4result
+        v5[0] = v5result
+
+        # multiply outputs by 2 so we've got something to backpropagate
+        v20[0] = wptype(2) * v2result[0]
+        v21[0] = wptype(2) * v2result[1]
+
+        v30[0] = wptype(2) * v3result[0]
+        v31[0] = wptype(2) * v3result[1]
+        v32[0] = wptype(2) * v3result[2]
+
+        v40[0] = wptype(2) * v4result[0]
+        v41[0] = wptype(2) * v4result[1]
+        v42[0] = wptype(2) * v4result[2]
+        v43[0] = wptype(2) * v4result[3]
+
+        v50[0] = wptype(2) * v5result[0]
+        v51[0] = wptype(2) * v5result[1]
+        v52[0] = wptype(2) * v5result[2]
+        v53[0] = wptype(2) * v5result[3]
+        v54[0] = wptype(2) * v5result[4]
+
+    def check_vector_constructors(
+        input: wp.array(dtype=wptype),
+        v2: wp.array(dtype=vec2),
+        v3: wp.array(dtype=vec3),
+        v4: wp.array(dtype=vec4),
+        v5: wp.array(dtype=vec5),
+        v20: wp.array(dtype=wptype),
+        v21: wp.array(dtype=wptype),
+        v30: wp.array(dtype=wptype),
+        v31: wp.array(dtype=wptype),
+        v32: wp.array(dtype=wptype),
+        v40: wp.array(dtype=wptype),
+        v41: wp.array(dtype=wptype),
+        v42: wp.array(dtype=wptype),
+        v43: wp.array(dtype=wptype),
+        v50: wp.array(dtype=wptype),
+        v51: wp.array(dtype=wptype),
+        v52: wp.array(dtype=wptype),
+        v53: wp.array(dtype=wptype),
+        v54: wp.array(dtype=wptype),
+    ):
+        v2result = vec2(input[0], input[1])
+        v3result = vec3(input[2], input[3], input[4])
+        v4result = vec4(input[5], input[6], input[7], input[8])
+        v5result = vec5(input[9], input[10], input[11], input[12], input[13])
+
+        v2[0] = v2result
+        v3[0] = v3result
+        v4[0] = v4result
+        v5[0] = v5result
+
+        # multiply the output by 2 so we've got something to backpropagate:
+        v20[0] = wptype(2) * v2result[0]
+        v21[0] = wptype(2) * v2result[1]
+
+        v30[0] = wptype(2) * v3result[0]
+        v31[0] = wptype(2) * v3result[1]
+        v32[0] = wptype(2) * v3result[2]
+
+        v40[0] = wptype(2) * v4result[0]
+        v41[0] = wptype(2) * v4result[1]
+        v42[0] = wptype(2) * v4result[2]
+        v43[0] = wptype(2) * v4result[3]
+
+        v50[0] = wptype(2) * v5result[0]
+        v51[0] = wptype(2) * v5result[1]
+        v52[0] = wptype(2) * v5result[2]
+        v53[0] = wptype(2) * v5result[3]
+        v54[0] = wptype(2) * v5result[4]
+
+    vec_kernel = getkernel(check_vector_constructors, suffix=dtype.__name__)
+    kernel = getkernel(check_scalar_constructor, suffix=dtype.__name__)
+
+    if register_kernels:
+        return
+
+    input = wp.array(randvals(rng, [1], dtype), requires_grad=True, device=device)
+    v2 = wp.zeros(1, dtype=vec2, device=device)
+    v3 = wp.zeros(1, dtype=vec3, device=device)
+    v4 = wp.zeros(1, dtype=vec4, device=device)
+    v5 = wp.zeros(1, dtype=vec5, device=device)
+    v20 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v21 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v30 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v31 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v32 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v40 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v41 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v42 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v43 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v50 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v51 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v52 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v53 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v54 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+
+    tape = wp.Tape()
+    with tape:
+        wp.launch(
+            kernel,
+            dim=1,
+            inputs=[input],
+            outputs=[v2, v3, v4, v5, v20, v21, v30, v31, v32, v40, v41, v42, v43, v50, v51, v52, v53, v54],
+            device=device,
+        )
+
+    if dtype in np_float_types:
+        for l in [v20, v21]:
+            tape.backward(loss=l)
+            test.assertEqual(tape.gradients[input].numpy()[0], 2.0)
+            tape.zero()
+
+        for l in [v30, v31, v32]:
+            tape.backward(loss=l)
+            test.assertEqual(tape.gradients[input].numpy()[0], 2.0)
+            tape.zero()
+
+        for l in [v40, v41, v42, v43]:
+            tape.backward(loss=l)
+            test.assertEqual(tape.gradients[input].numpy()[0], 2.0)
+            tape.zero()
+
+        for l in [v50, v51, v52, v53, v54]:
+            tape.backward(loss=l)
+            test.assertEqual(tape.gradients[input].numpy()[0], 2.0)
+            tape.zero()
+
+    val = input.numpy()[0]
+    assert_np_equal(v2.numpy()[0], np.array([val, val]), tol=1.0e-6)
+    assert_np_equal(v3.numpy()[0], np.array([val, val, val]), tol=1.0e-6)
+    assert_np_equal(v4.numpy()[0], np.array([val, val, val, val]), tol=1.0e-6)
+    assert_np_equal(v5.numpy()[0], np.array([val, val, val, val, val]), tol=1.0e-6)
+
+    assert_np_equal(v20.numpy()[0], 2 * val, tol=1.0e-6)
+    assert_np_equal(v21.numpy()[0], 2 * val, tol=1.0e-6)
+    assert_np_equal(v30.numpy()[0], 2 * val, tol=1.0e-6)
+    assert_np_equal(v31.numpy()[0], 2 * val, tol=1.0e-6)
+    assert_np_equal(v32.numpy()[0], 2 * val, tol=1.0e-6)
+    assert_np_equal(v40.numpy()[0], 2 * val, tol=1.0e-6)
+    assert_np_equal(v41.numpy()[0], 2 * val, tol=1.0e-6)
+    assert_np_equal(v42.numpy()[0], 2 * val, tol=1.0e-6)
+    assert_np_equal(v43.numpy()[0], 2 * val, tol=1.0e-6)
+    assert_np_equal(v50.numpy()[0], 2 * val, tol=1.0e-6)
+    assert_np_equal(v51.numpy()[0], 2 * val, tol=1.0e-6)
+    assert_np_equal(v52.numpy()[0], 2 * val, tol=1.0e-6)
+    assert_np_equal(v53.numpy()[0], 2 * val, tol=1.0e-6)
+    assert_np_equal(v54.numpy()[0], 2 * val, tol=1.0e-6)
+
+    input = wp.array(randvals(rng, [14], dtype), requires_grad=True, device=device)
+    tape = wp.Tape()
+    with tape:
+        wp.launch(
+            vec_kernel,
+            dim=1,
+            inputs=[input],
+            outputs=[v2, v3, v4, v5, v20, v21, v30, v31, v32, v40, v41, v42, v43, v50, v51, v52, v53, v54],
+            device=device,
+        )
+
+    if dtype in np_float_types:
+        for i, l in enumerate([v20, v21, v30, v31, v32, v40, v41, v42, v43, v50, v51, v52, v53, v54]):
+            tape.backward(loss=l)
+            grad = tape.gradients[input].numpy()
+            expected_grad = np.zeros_like(grad)
+            expected_grad[i] = 2
+            assert_np_equal(grad, expected_grad, tol=tol)
+            tape.zero()
+
+    assert_np_equal(v2.numpy()[0, 0], input.numpy()[0], tol=tol)
+    assert_np_equal(v2.numpy()[0, 1], input.numpy()[1], tol=tol)
+    assert_np_equal(v3.numpy()[0, 0], input.numpy()[2], tol=tol)
+    assert_np_equal(v3.numpy()[0, 1], input.numpy()[3], tol=tol)
+    assert_np_equal(v3.numpy()[0, 2], input.numpy()[4], tol=tol)
+    assert_np_equal(v4.numpy()[0, 0], input.numpy()[5], tol=tol)
+    assert_np_equal(v4.numpy()[0, 1], input.numpy()[6], tol=tol)
+    assert_np_equal(v4.numpy()[0, 2], input.numpy()[7], tol=tol)
+    assert_np_equal(v4.numpy()[0, 3], input.numpy()[8], tol=tol)
+    assert_np_equal(v5.numpy()[0, 0], input.numpy()[9], tol=tol)
+    assert_np_equal(v5.numpy()[0, 1], input.numpy()[10], tol=tol)
+    assert_np_equal(v5.numpy()[0, 2], input.numpy()[11], tol=tol)
+    assert_np_equal(v5.numpy()[0, 3], input.numpy()[12], tol=tol)
+    assert_np_equal(v5.numpy()[0, 4], input.numpy()[13], tol=tol)
+
+    assert_np_equal(v20.numpy()[0], 2 * input.numpy()[0], tol=tol)
+    assert_np_equal(v21.numpy()[0], 2 * input.numpy()[1], tol=tol)
+    assert_np_equal(v30.numpy()[0], 2 * input.numpy()[2], tol=tol)
+    assert_np_equal(v31.numpy()[0], 2 * input.numpy()[3], tol=tol)
+    assert_np_equal(v32.numpy()[0], 2 * input.numpy()[4], tol=tol)
+    assert_np_equal(v40.numpy()[0], 2 * input.numpy()[5], tol=tol)
+    assert_np_equal(v41.numpy()[0], 2 * input.numpy()[6], tol=tol)
+    assert_np_equal(v42.numpy()[0], 2 * input.numpy()[7], tol=tol)
+    assert_np_equal(v43.numpy()[0], 2 * input.numpy()[8], tol=tol)
+    assert_np_equal(v50.numpy()[0], 2 * input.numpy()[9], tol=tol)
+    assert_np_equal(v51.numpy()[0], 2 * input.numpy()[10], tol=tol)
+    assert_np_equal(v52.numpy()[0], 2 * input.numpy()[11], tol=tol)
+    assert_np_equal(v53.numpy()[0], 2 * input.numpy()[12], tol=tol)
+    assert_np_equal(v54.numpy()[0], 2 * input.numpy()[13], tol=tol)
+
+
+def test_anon_type_instance(test, device, dtype, register_kernels=False):
+    rng = np.random.default_rng(123)
+
+    tol = {
+        np.float16: 5.0e-3,
+        np.float32: 1.0e-6,
+        np.float64: 1.0e-8,
+    }.get(dtype, 0)
+
+    wptype = wp.types.np_dtype_to_warp_type[np.dtype(dtype)]
+
+    def check_scalar_init(
+        input: wp.array(dtype=wptype),
+        output: wp.array(dtype=wptype),
+    ):
+        v2result = wp.vector(input[0], length=2)
+        v3result = wp.vector(input[1], length=3)
+        v4result = wp.vector(input[2], length=4)
+        v5result = wp.vector(input[3], length=5)
+
+        idx = 0
+        for i in range(2):
+            output[idx] = wptype(2) * v2result[i]
+            idx = idx + 1
+        for i in range(3):
+            output[idx] = wptype(2) * v3result[i]
+            idx = idx + 1
+        for i in range(4):
+            output[idx] = wptype(2) * v4result[i]
+            idx = idx + 1
+        for i in range(5):
+            output[idx] = wptype(2) * v5result[i]
+            idx = idx + 1
+
+    def check_component_init(
+        input: wp.array(dtype=wptype),
+        output: wp.array(dtype=wptype),
+    ):
+        v2result = wp.vector(input[0], input[1])
+        v3result = wp.vector(input[2], input[3], input[4])
+        v4result = wp.vector(input[5], input[6], input[7], input[8])
+        v5result = wp.vector(input[9], input[10], input[11], input[12], input[13])
+
+        idx = 0
+        for i in range(2):
+            output[idx] = wptype(2) * v2result[i]
+            idx = idx + 1
+        for i in range(3):
+            output[idx] = wptype(2) * v3result[i]
+            idx = idx + 1
+        for i in range(4):
+            output[idx] = wptype(2) * v4result[i]
+            idx = idx + 1
+        for i in range(5):
+            output[idx] = wptype(2) * v5result[i]
+            idx = idx + 1
+
+    scalar_kernel = getkernel(check_scalar_init, suffix=dtype.__name__)
+    component_kernel = getkernel(check_component_init, suffix=dtype.__name__)
+    output_select_kernel = get_select_kernel(wptype)
+
+    if register_kernels:
+        return
+
+    input = wp.array(randvals(rng, [4], dtype), requires_grad=True, device=device)
+    output = wp.zeros(2 + 3 + 4 + 5, dtype=wptype, requires_grad=True, device=device)
+
+    wp.launch(scalar_kernel, dim=1, inputs=[input], outputs=[output], device=device)
+
+    assert_np_equal(output.numpy()[:2], 2 * np.array([input.numpy()[0]] * 2), tol=1.0e-6)
+    assert_np_equal(output.numpy()[2:5], 2 * np.array([input.numpy()[1]] * 3), tol=1.0e-6)
+    assert_np_equal(output.numpy()[5:9], 2 * np.array([input.numpy()[2]] * 4), tol=1.0e-6)
+    assert_np_equal(output.numpy()[9:], 2 * np.array([input.numpy()[3]] * 5), tol=1.0e-6)
+
+    if dtype in np_float_types:
+        out = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+        for i in range(len(output)):
+            tape = wp.Tape()
+            with tape:
+                wp.launch(scalar_kernel, dim=1, inputs=[input], outputs=[output], device=device)
+                wp.launch(output_select_kernel, dim=1, inputs=[output, i], outputs=[out], device=device)
+
+            tape.backward(loss=out)
+            expected = np.zeros_like(input.numpy())
+            if i < 2:
+                expected[0] = 2
+            elif i < 5:
+                expected[1] = 2
+            elif i < 9:
+                expected[2] = 2
+            else:
+                expected[3] = 2
+
+            assert_np_equal(tape.gradients[input].numpy(), expected, tol=tol)
+
+            tape.reset()
+            tape.zero()
+
+    input = wp.array(randvals(rng, [2 + 3 + 4 + 5], dtype), requires_grad=True, device=device)
+    output = wp.zeros(2 + 3 + 4 + 5, dtype=wptype, requires_grad=True, device=device)
+
+    wp.launch(component_kernel, dim=1, inputs=[input], outputs=[output], device=device)
+
+    assert_np_equal(output.numpy(), 2 * input.numpy(), tol=1.0e-6)
+
+    if dtype in np_float_types:
+        out = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+        for i in range(len(output)):
+            tape = wp.Tape()
+            with tape:
+                wp.launch(component_kernel, dim=1, inputs=[input], outputs=[output], device=device)
+                wp.launch(output_select_kernel, dim=1, inputs=[output, i], outputs=[out], device=device)
+
+            tape.backward(loss=out)
+            expected = np.zeros_like(input.numpy())
+            expected[i] = 2
+
+            assert_np_equal(tape.gradients[input].numpy(), expected, tol=tol)
+
+            tape.reset()
+            tape.zero()
+
+
+def test_indexing(test, device, dtype, register_kernels=False):
+    rng = np.random.default_rng(123)
+
+    tol = {
+        np.float16: 5.0e-3,
+        np.float32: 1.0e-6,
+        np.float64: 1.0e-8,
+    }.get(dtype, 0)
+
+    wptype = wp.types.np_dtype_to_warp_type[np.dtype(dtype)]
+    vec2 = wp.types.vector(length=2, dtype=wptype)
+    vec3 = wp.types.vector(length=3, dtype=wptype)
+    vec4 = wp.types.vector(length=4, dtype=wptype)
+    vec5 = wp.types.vector(length=5, dtype=wptype)
+
+    def check_indexing(
+        v2: wp.array(dtype=vec2),
+        v3: wp.array(dtype=vec3),
+        v4: wp.array(dtype=vec4),
+        v5: wp.array(dtype=vec5),
+        v20: wp.array(dtype=wptype),
+        v21: wp.array(dtype=wptype),
+        v30: wp.array(dtype=wptype),
+        v31: wp.array(dtype=wptype),
+        v32: wp.array(dtype=wptype),
+        v40: wp.array(dtype=wptype),
+        v41: wp.array(dtype=wptype),
+        v42: wp.array(dtype=wptype),
+        v43: wp.array(dtype=wptype),
+        v50: wp.array(dtype=wptype),
+        v51: wp.array(dtype=wptype),
+        v52: wp.array(dtype=wptype),
+        v53: wp.array(dtype=wptype),
+        v54: wp.array(dtype=wptype),
+    ):
+        # multiply outputs by 2 so we've got something to backpropagate:
+        v20[0] = wptype(2) * v2[0][0]
+        v21[0] = wptype(2) * v2[0][1]
+
+        v30[0] = wptype(2) * v3[0][0]
+        v31[0] = wptype(2) * v3[0][1]
+        v32[0] = wptype(2) * v3[0][2]
+
+        v40[0] = wptype(2) * v4[0][0]
+        v41[0] = wptype(2) * v4[0][1]
+        v42[0] = wptype(2) * v4[0][2]
+        v43[0] = wptype(2) * v4[0][3]
+
+        v50[0] = wptype(2) * v5[0][0]
+        v51[0] = wptype(2) * v5[0][1]
+        v52[0] = wptype(2) * v5[0][2]
+        v53[0] = wptype(2) * v5[0][3]
+        v54[0] = wptype(2) * v5[0][4]
+
+    kernel = getkernel(check_indexing, suffix=dtype.__name__)
+
+    if register_kernels:
+        return
+
+    v2 = wp.array(randvals(rng, (1, 2), dtype), dtype=vec2, requires_grad=True, device=device)
+    v3 = wp.array(randvals(rng, (1, 3), dtype), dtype=vec3, requires_grad=True, device=device)
+    v4 = wp.array(randvals(rng, (1, 4), dtype), dtype=vec4, requires_grad=True, device=device)
+    v5 = wp.array(randvals(rng, (1, 5), dtype), dtype=vec5, requires_grad=True, device=device)
+    v20 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v21 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v30 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v31 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v32 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v40 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v41 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v42 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v43 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v50 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v51 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v52 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v53 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v54 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+
+    tape = wp.Tape()
+    with tape:
+        wp.launch(
+            kernel,
+            dim=1,
+            inputs=[v2, v3, v4, v5],
+            outputs=[v20, v21, v30, v31, v32, v40, v41, v42, v43, v50, v51, v52, v53, v54],
+            device=device,
+        )
+
+    if dtype in np_float_types:
+        for i, l in enumerate([v20, v21, v30, v31, v32, v40, v41, v42, v43, v50, v51, v52, v53, v54]):
+            tape.backward(loss=l)
+            allgrads = np.concatenate([tape.gradients[v].numpy()[0] for v in [v2, v3, v4, v5]])
+            expected_grads = np.zeros_like(allgrads)
+            expected_grads[i] = 2
+            assert_np_equal(allgrads, expected_grads, tol=tol)
+            tape.zero()
+
+    assert_np_equal(v20.numpy()[0], 2.0 * v2.numpy()[0, 0], tol=tol)
+    assert_np_equal(v21.numpy()[0], 2.0 * v2.numpy()[0, 1], tol=tol)
+    assert_np_equal(v30.numpy()[0], 2.0 * v3.numpy()[0, 0], tol=tol)
+    assert_np_equal(v31.numpy()[0], 2.0 * v3.numpy()[0, 1], tol=tol)
+    assert_np_equal(v32.numpy()[0], 2.0 * v3.numpy()[0, 2], tol=tol)
+    assert_np_equal(v40.numpy()[0], 2.0 * v4.numpy()[0, 0], tol=tol)
+    assert_np_equal(v41.numpy()[0], 2.0 * v4.numpy()[0, 1], tol=tol)
+    assert_np_equal(v42.numpy()[0], 2.0 * v4.numpy()[0, 2], tol=tol)
+    assert_np_equal(v43.numpy()[0], 2.0 * v4.numpy()[0, 3], tol=tol)
+    assert_np_equal(v50.numpy()[0], 2.0 * v5.numpy()[0, 0], tol=tol)
+    assert_np_equal(v51.numpy()[0], 2.0 * v5.numpy()[0, 1], tol=tol)
+    assert_np_equal(v52.numpy()[0], 2.0 * v5.numpy()[0, 2], tol=tol)
+    assert_np_equal(v53.numpy()[0], 2.0 * v5.numpy()[0, 3], tol=tol)
+    assert_np_equal(v54.numpy()[0], 2.0 * v5.numpy()[0, 4], tol=tol)
+
+
+def test_equality(test, device, dtype, register_kernels=False):
+    wptype = wp.types.np_dtype_to_warp_type[np.dtype(dtype)]
+    vec2 = wp.types.vector(length=2, dtype=wptype)
+    vec3 = wp.types.vector(length=3, dtype=wptype)
+    vec4 = wp.types.vector(length=4, dtype=wptype)
+    vec5 = wp.types.vector(length=5, dtype=wptype)
+
+    def check_equality(
+        v20: wp.array(dtype=vec2),
+        v21: wp.array(dtype=vec2),
+        v22: wp.array(dtype=vec2),
+        v30: wp.array(dtype=vec3),
+        v31: wp.array(dtype=vec3),
+        v32: wp.array(dtype=vec3),
+        v33: wp.array(dtype=vec3),
+        v40: wp.array(dtype=vec4),
+        v41: wp.array(dtype=vec4),
+        v42: wp.array(dtype=vec4),
+        v43: wp.array(dtype=vec4),
+        v44: wp.array(dtype=vec4),
+        v50: wp.array(dtype=vec5),
+        v51: wp.array(dtype=vec5),
+        v52: wp.array(dtype=vec5),
+        v53: wp.array(dtype=vec5),
+        v54: wp.array(dtype=vec5),
+        v55: wp.array(dtype=vec5),
+    ):
+        wp.expect_eq(v20[0], v20[0])
+        wp.expect_neq(v21[0], v20[0])
+        wp.expect_neq(v22[0], v20[0])
+
+        wp.expect_eq(v30[0], v30[0])
+        wp.expect_neq(v31[0], v30[0])
+        wp.expect_neq(v32[0], v30[0])
+        wp.expect_neq(v33[0], v30[0])
+
+        wp.expect_eq(v40[0], v40[0])
+        wp.expect_neq(v41[0], v40[0])
+        wp.expect_neq(v42[0], v40[0])
+        wp.expect_neq(v43[0], v40[0])
+        wp.expect_neq(v44[0], v40[0])
+
+        wp.expect_eq(v50[0], v50[0])
+        wp.expect_neq(v51[0], v50[0])
+        wp.expect_neq(v52[0], v50[0])
+        wp.expect_neq(v53[0], v50[0])
+        wp.expect_neq(v54[0], v50[0])
+        wp.expect_neq(v55[0], v50[0])
+
+    kernel = getkernel(check_equality, suffix=dtype.__name__)
+
+    if register_kernels:
+        return
+
+    v20 = wp.array([1.0, 2.0], dtype=vec2, requires_grad=True, device=device)
+    v21 = wp.array([1.0, 3.0], dtype=vec2, requires_grad=True, device=device)
+    v22 = wp.array([3.0, 2.0], dtype=vec2, requires_grad=True, device=device)
+
+    v30 = wp.array([1.0, 2.0, 3.0], dtype=vec3, requires_grad=True, device=device)
+    v31 = wp.array([-1.0, 2.0, 3.0], dtype=vec3, requires_grad=True, device=device)
+    v32 = wp.array([1.0, -2.0, 3.0], dtype=vec3, requires_grad=True, device=device)
+    v33 = wp.array([1.0, 2.0, -3.0], dtype=vec3, requires_grad=True, device=device)
+
+    v40 = wp.array([1.0, 2.0, 3.0, 4.0], dtype=vec4, requires_grad=True, device=device)
+    v41 = wp.array([-1.0, 2.0, 3.0, 4.0], dtype=vec4, requires_grad=True, device=device)
+    v42 = wp.array([1.0, -2.0, 3.0, 4.0], dtype=vec4, requires_grad=True, device=device)
+    v43 = wp.array([1.0, 2.0, -3.0, 4.0], dtype=vec4, requires_grad=True, device=device)
+    v44 = wp.array([1.0, 2.0, 3.0, -4.0], dtype=vec4, requires_grad=True, device=device)
+
+    v50 = wp.array([1.0, 2.0, 3.0, 4.0, 5.0], dtype=vec5, requires_grad=True, device=device)
+    v51 = wp.array([-1.0, 2.0, 3.0, 4.0, 5.0], dtype=vec5, requires_grad=True, device=device)
+    v52 = wp.array([1.0, -2.0, 3.0, 4.0, 5.0], dtype=vec5, requires_grad=True, device=device)
+    v53 = wp.array([1.0, 2.0, -3.0, 4.0, 5.0], dtype=vec5, requires_grad=True, device=device)
+    v54 = wp.array([1.0, 2.0, 3.0, -4.0, 5.0], dtype=vec5, requires_grad=True, device=device)
+    v55 = wp.array([1.0, 2.0, 3.0, 4.0, -5.0], dtype=vec5, requires_grad=True, device=device)
+    wp.launch(
+        kernel,
+        dim=1,
+        inputs=[
+            v20,
+            v21,
+            v22,
+            v30,
+            v31,
+            v32,
+            v33,
+            v40,
+            v41,
+            v42,
+            v43,
+            v44,
+            v50,
+            v51,
+            v52,
+            v53,
+            v54,
+            v55,
+        ],
+        outputs=[],
+        device=device,
+    )
+
+
+def test_scalar_multiplication(test, device, dtype, register_kernels=False):
+    rng = np.random.default_rng(123)
+
+    tol = {
+        np.float16: 5.0e-3,
+        np.float32: 1.0e-6,
+        np.float64: 1.0e-8,
+    }.get(dtype, 0)
+
+    wptype = wp.types.np_dtype_to_warp_type[np.dtype(dtype)]
+    vec2 = wp.types.vector(length=2, dtype=wptype)
+    vec3 = wp.types.vector(length=3, dtype=wptype)
+    vec4 = wp.types.vector(length=4, dtype=wptype)
+    vec5 = wp.types.vector(length=5, dtype=wptype)
+
+    def check_mul(
+        s: wp.array(dtype=wptype),
+        v2: wp.array(dtype=vec2),
+        v3: wp.array(dtype=vec3),
+        v4: wp.array(dtype=vec4),
+        v5: wp.array(dtype=vec5),
+        v20: wp.array(dtype=wptype),
+        v21: wp.array(dtype=wptype),
+        v30: wp.array(dtype=wptype),
+        v31: wp.array(dtype=wptype),
+        v32: wp.array(dtype=wptype),
+        v40: wp.array(dtype=wptype),
+        v41: wp.array(dtype=wptype),
+        v42: wp.array(dtype=wptype),
+        v43: wp.array(dtype=wptype),
+        v50: wp.array(dtype=wptype),
+        v51: wp.array(dtype=wptype),
+        v52: wp.array(dtype=wptype),
+        v53: wp.array(dtype=wptype),
+        v54: wp.array(dtype=wptype),
+    ):
+        v2result = s[0] * v2[0]
+        v3result = s[0] * v3[0]
+        v4result = s[0] * v4[0]
+        v5result = s[0] * v5[0]
+
+        # multiply outputs by 2 so we've got something to backpropagate:
+        v20[0] = wptype(2) * v2result[0]
+        v21[0] = wptype(2) * v2result[1]
+
+        v30[0] = wptype(2) * v3result[0]
+        v31[0] = wptype(2) * v3result[1]
+        v32[0] = wptype(2) * v3result[2]
+
+        v40[0] = wptype(2) * v4result[0]
+        v41[0] = wptype(2) * v4result[1]
+        v42[0] = wptype(2) * v4result[2]
+        v43[0] = wptype(2) * v4result[3]
+
+        v50[0] = wptype(2) * v5result[0]
+        v51[0] = wptype(2) * v5result[1]
+        v52[0] = wptype(2) * v5result[2]
+        v53[0] = wptype(2) * v5result[3]
+        v54[0] = wptype(2) * v5result[4]
+
+    kernel = getkernel(check_mul, suffix=dtype.__name__)
+
+    if register_kernels:
+        return
+
+    s = wp.array(randvals(rng, [1], dtype), requires_grad=True, device=device)
+    v2 = wp.array(randvals(rng, (1, 2), dtype), dtype=vec2, requires_grad=True, device=device)
+    v3 = wp.array(randvals(rng, (1, 3), dtype), dtype=vec3, requires_grad=True, device=device)
+    v4 = wp.array(randvals(rng, (1, 4), dtype), dtype=vec4, requires_grad=True, device=device)
+    v5 = wp.array(randvals(rng, (1, 5), dtype), dtype=vec5, requires_grad=True, device=device)
+    v20 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v21 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v30 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v31 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v32 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v40 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v41 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v42 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v43 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v50 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v51 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v52 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v53 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v54 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    tape = wp.Tape()
+    with tape:
+        wp.launch(
+            kernel,
+            dim=1,
+            inputs=[
+                s,
+                v2,
+                v3,
+                v4,
+                v5,
+            ],
+            outputs=[v20, v21, v30, v31, v32, v40, v41, v42, v43, v50, v51, v52, v53, v54],
+            device=device,
+        )
+
+    assert_np_equal(v20.numpy()[0], 2 * s.numpy()[0] * v2.numpy()[0, 0], tol=tol)
+    assert_np_equal(v21.numpy()[0], 2 * s.numpy()[0] * v2.numpy()[0, 1], tol=tol)
+
+    assert_np_equal(v30.numpy()[0], 2 * s.numpy()[0] * v3.numpy()[0, 0], tol=10 * tol)
+    assert_np_equal(v31.numpy()[0], 2 * s.numpy()[0] * v3.numpy()[0, 1], tol=10 * tol)
+    assert_np_equal(v32.numpy()[0], 2 * s.numpy()[0] * v3.numpy()[0, 2], tol=10 * tol)
+
+    assert_np_equal(v40.numpy()[0], 2 * s.numpy()[0] * v4.numpy()[0, 0], tol=10 * tol)
+    assert_np_equal(v41.numpy()[0], 2 * s.numpy()[0] * v4.numpy()[0, 1], tol=10 * tol)
+    assert_np_equal(v42.numpy()[0], 2 * s.numpy()[0] * v4.numpy()[0, 2], tol=10 * tol)
+    assert_np_equal(v43.numpy()[0], 2 * s.numpy()[0] * v4.numpy()[0, 3], tol=10 * tol)
+
+    assert_np_equal(v50.numpy()[0], 2 * s.numpy()[0] * v5.numpy()[0, 0], tol=10 * tol)
+    assert_np_equal(v51.numpy()[0], 2 * s.numpy()[0] * v5.numpy()[0, 1], tol=10 * tol)
+    assert_np_equal(v52.numpy()[0], 2 * s.numpy()[0] * v5.numpy()[0, 2], tol=10 * tol)
+    assert_np_equal(v53.numpy()[0], 2 * s.numpy()[0] * v5.numpy()[0, 3], tol=10 * tol)
+    assert_np_equal(v54.numpy()[0], 2 * s.numpy()[0] * v5.numpy()[0, 4], tol=10 * tol)
+
+    incmps = np.concatenate([v.numpy()[0] for v in [v2, v3, v4, v5]])
+
+    if dtype in np_float_types:
+        for i, l in enumerate([v20, v21, v30, v31, v32, v40, v41, v42, v43]):
+            tape.backward(loss=l)
+            sgrad = tape.gradients[s].numpy()[0]
+            assert_np_equal(sgrad, 2 * incmps[i], tol=10 * tol)
+            allgrads = np.concatenate([tape.gradients[v].numpy()[0] for v in [v2, v3, v4]])
+            expected_grads = np.zeros_like(allgrads)
+            expected_grads[i] = s.numpy()[0] * 2
+            assert_np_equal(allgrads, expected_grads, tol=10 * tol)
+            tape.zero()
+
+
+def test_scalar_multiplication_rightmul(test, device, dtype, register_kernels=False):
+    rng = np.random.default_rng(123)
+
+    tol = {
+        np.float16: 5.0e-3,
+        np.float32: 1.0e-6,
+        np.float64: 1.0e-8,
+    }.get(dtype, 0)
+
+    wptype = wp.types.np_dtype_to_warp_type[np.dtype(dtype)]
+    vec2 = wp.types.vector(length=2, dtype=wptype)
+    vec3 = wp.types.vector(length=3, dtype=wptype)
+    vec4 = wp.types.vector(length=4, dtype=wptype)
+    vec5 = wp.types.vector(length=5, dtype=wptype)
+
+    def check_rightmul(
+        s: wp.array(dtype=wptype),
+        v2: wp.array(dtype=vec2),
+        v3: wp.array(dtype=vec3),
+        v4: wp.array(dtype=vec4),
+        v5: wp.array(dtype=vec5),
+        v20: wp.array(dtype=wptype),
+        v21: wp.array(dtype=wptype),
+        v30: wp.array(dtype=wptype),
+        v31: wp.array(dtype=wptype),
+        v32: wp.array(dtype=wptype),
+        v40: wp.array(dtype=wptype),
+        v41: wp.array(dtype=wptype),
+        v42: wp.array(dtype=wptype),
+        v43: wp.array(dtype=wptype),
+        v50: wp.array(dtype=wptype),
+        v51: wp.array(dtype=wptype),
+        v52: wp.array(dtype=wptype),
+        v53: wp.array(dtype=wptype),
+        v54: wp.array(dtype=wptype),
+    ):
+        v2result = v2[0] * s[0]
+        v3result = v3[0] * s[0]
+        v4result = v4[0] * s[0]
+        v5result = v5[0] * s[0]
+
+        # multiply outputs by 2 so we've got something to backpropagate:
+        v20[0] = wptype(2) * v2result[0]
+        v21[0] = wptype(2) * v2result[1]
+
+        v30[0] = wptype(2) * v3result[0]
+        v31[0] = wptype(2) * v3result[1]
+        v32[0] = wptype(2) * v3result[2]
+
+        v40[0] = wptype(2) * v4result[0]
+        v41[0] = wptype(2) * v4result[1]
+        v42[0] = wptype(2) * v4result[2]
+        v43[0] = wptype(2) * v4result[3]
+
+        v50[0] = wptype(2) * v5result[0]
+        v51[0] = wptype(2) * v5result[1]
+        v52[0] = wptype(2) * v5result[2]
+        v53[0] = wptype(2) * v5result[3]
+        v54[0] = wptype(2) * v5result[4]
+
+    kernel = getkernel(check_rightmul, suffix=dtype.__name__)
+
+    if register_kernels:
+        return
+
+    s = wp.array(randvals(rng, [1], dtype), requires_grad=True, device=device)
+    v2 = wp.array(randvals(rng, (1, 2), dtype), dtype=vec2, requires_grad=True, device=device)
+    v3 = wp.array(randvals(rng, (1, 3), dtype), dtype=vec3, requires_grad=True, device=device)
+    v4 = wp.array(randvals(rng, (1, 4), dtype), dtype=vec4, requires_grad=True, device=device)
+    v5 = wp.array(randvals(rng, (1, 5), dtype), dtype=vec5, requires_grad=True, device=device)
+    v20 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v21 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v30 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v31 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v32 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v40 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v41 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v42 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v43 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v50 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v51 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v52 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v53 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v54 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    tape = wp.Tape()
+    with tape:
+        wp.launch(
+            kernel,
+            dim=1,
+            inputs=[
+                s,
+                v2,
+                v3,
+                v4,
+                v5,
+            ],
+            outputs=[v20, v21, v30, v31, v32, v40, v41, v42, v43, v50, v51, v52, v53, v54],
+            device=device,
+        )
+
+    assert_np_equal(v20.numpy()[0], 2 * s.numpy()[0] * v2.numpy()[0, 0], tol=tol)
+    assert_np_equal(v21.numpy()[0], 2 * s.numpy()[0] * v2.numpy()[0, 1], tol=tol)
+
+    assert_np_equal(v30.numpy()[0], 2 * s.numpy()[0] * v3.numpy()[0, 0], tol=10 * tol)
+    assert_np_equal(v31.numpy()[0], 2 * s.numpy()[0] * v3.numpy()[0, 1], tol=10 * tol)
+    assert_np_equal(v32.numpy()[0], 2 * s.numpy()[0] * v3.numpy()[0, 2], tol=10 * tol)
+
+    assert_np_equal(v40.numpy()[0], 2 * s.numpy()[0] * v4.numpy()[0, 0], tol=10 * tol)
+    assert_np_equal(v41.numpy()[0], 2 * s.numpy()[0] * v4.numpy()[0, 1], tol=10 * tol)
+    assert_np_equal(v42.numpy()[0], 2 * s.numpy()[0] * v4.numpy()[0, 2], tol=10 * tol)
+    assert_np_equal(v43.numpy()[0], 2 * s.numpy()[0] * v4.numpy()[0, 3], tol=10 * tol)
+
+    assert_np_equal(v50.numpy()[0], 2 * s.numpy()[0] * v5.numpy()[0, 0], tol=10 * tol)
+    assert_np_equal(v51.numpy()[0], 2 * s.numpy()[0] * v5.numpy()[0, 1], tol=10 * tol)
+    assert_np_equal(v52.numpy()[0], 2 * s.numpy()[0] * v5.numpy()[0, 2], tol=10 * tol)
+    assert_np_equal(v53.numpy()[0], 2 * s.numpy()[0] * v5.numpy()[0, 3], tol=10 * tol)
+    assert_np_equal(v54.numpy()[0], 2 * s.numpy()[0] * v5.numpy()[0, 4], tol=10 * tol)
+
+    incmps = np.concatenate([v.numpy()[0] for v in [v2, v3, v4, v5]])
+
+    if dtype in np_float_types:
+        for i, l in enumerate([v20, v21, v30, v31, v32, v40, v41, v42, v43]):
+            tape.backward(loss=l)
+            sgrad = tape.gradients[s].numpy()[0]
+            assert_np_equal(sgrad, 2 * incmps[i], tol=10 * tol)
+            allgrads = np.concatenate([tape.gradients[v].numpy()[0] for v in [v2, v3, v4]])
+            expected_grads = np.zeros_like(allgrads)
+            expected_grads[i] = s.numpy()[0] * 2
+            assert_np_equal(allgrads, expected_grads, tol=10 * tol)
+            tape.zero()
+
+
+def test_cw_multiplication(test, device, dtype, register_kernels=False):
+    rng = np.random.default_rng(123)
+
+    tol = {
+        np.float16: 5.0e-3,
+        np.float32: 1.0e-6,
+        np.float64: 1.0e-8,
+    }.get(dtype, 0)
+
+    wptype = wp.types.np_dtype_to_warp_type[np.dtype(dtype)]
+    vec2 = wp.types.vector(length=2, dtype=wptype)
+    vec3 = wp.types.vector(length=3, dtype=wptype)
+    vec4 = wp.types.vector(length=4, dtype=wptype)
+    vec5 = wp.types.vector(length=5, dtype=wptype)
+
+    def check_cw_mul(
+        s2: wp.array(dtype=vec2),
+        s3: wp.array(dtype=vec3),
+        s4: wp.array(dtype=vec4),
+        s5: wp.array(dtype=vec5),
+        v2: wp.array(dtype=vec2),
+        v3: wp.array(dtype=vec3),
+        v4: wp.array(dtype=vec4),
+        v5: wp.array(dtype=vec5),
+        v20: wp.array(dtype=wptype),
+        v21: wp.array(dtype=wptype),
+        v30: wp.array(dtype=wptype),
+        v31: wp.array(dtype=wptype),
+        v32: wp.array(dtype=wptype),
+        v40: wp.array(dtype=wptype),
+        v41: wp.array(dtype=wptype),
+        v42: wp.array(dtype=wptype),
+        v43: wp.array(dtype=wptype),
+        v50: wp.array(dtype=wptype),
+        v51: wp.array(dtype=wptype),
+        v52: wp.array(dtype=wptype),
+        v53: wp.array(dtype=wptype),
+        v54: wp.array(dtype=wptype),
+    ):
+        v2result = wp.cw_mul(s2[0], v2[0])
+        v3result = wp.cw_mul(s3[0], v3[0])
+        v4result = wp.cw_mul(s4[0], v4[0])
+        v5result = wp.cw_mul(s5[0], v5[0])
+
+        v20[0] = wptype(2) * v2result[0]
+        v21[0] = wptype(2) * v2result[1]
+
+        v30[0] = wptype(2) * v3result[0]
+        v31[0] = wptype(2) * v3result[1]
+        v32[0] = wptype(2) * v3result[2]
+
+        v40[0] = wptype(2) * v4result[0]
+        v41[0] = wptype(2) * v4result[1]
+        v42[0] = wptype(2) * v4result[2]
+        v43[0] = wptype(2) * v4result[3]
+
+        v50[0] = wptype(2) * v5result[0]
+        v51[0] = wptype(2) * v5result[1]
+        v52[0] = wptype(2) * v5result[2]
+        v53[0] = wptype(2) * v5result[3]
+        v54[0] = wptype(2) * v5result[4]
+
+    kernel = getkernel(check_cw_mul, suffix=dtype.__name__)
+
+    if register_kernels:
+        return
+
+    s2 = wp.array(randvals(rng, (1, 2), dtype), dtype=vec2, requires_grad=True, device=device)
+    s3 = wp.array(randvals(rng, (1, 3), dtype), dtype=vec3, requires_grad=True, device=device)
+    s4 = wp.array(randvals(rng, (1, 4), dtype), dtype=vec4, requires_grad=True, device=device)
+    s5 = wp.array(randvals(rng, (1, 5), dtype), dtype=vec5, requires_grad=True, device=device)
+    v2 = wp.array(randvals(rng, (1, 2), dtype), dtype=vec2, requires_grad=True, device=device)
+    v3 = wp.array(randvals(rng, (1, 3), dtype), dtype=vec3, requires_grad=True, device=device)
+    v4 = wp.array(randvals(rng, (1, 4), dtype), dtype=vec4, requires_grad=True, device=device)
+    v5 = wp.array(randvals(rng, (1, 5), dtype), dtype=vec5, requires_grad=True, device=device)
+    v20 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v21 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v30 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v31 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v32 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v40 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v41 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v42 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v43 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v50 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v51 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v52 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v53 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v54 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    tape = wp.Tape()
+    with tape:
+        wp.launch(
+            kernel,
+            dim=1,
+            inputs=[
+                s2,
+                s3,
+                s4,
+                s5,
+                v2,
+                v3,
+                v4,
+                v5,
+            ],
+            outputs=[v20, v21, v30, v31, v32, v40, v41, v42, v43, v50, v51, v52, v53, v54],
+            device=device,
+        )
+
+    assert_np_equal(v20.numpy()[0], 2 * s2.numpy()[0, 0] * v2.numpy()[0, 0], tol=10 * tol)
+    assert_np_equal(v21.numpy()[0], 2 * s2.numpy()[0, 1] * v2.numpy()[0, 1], tol=10 * tol)
+
+    assert_np_equal(v30.numpy()[0], 2 * s3.numpy()[0, 0] * v3.numpy()[0, 0], tol=10 * tol)
+    assert_np_equal(v31.numpy()[0], 2 * s3.numpy()[0, 1] * v3.numpy()[0, 1], tol=10 * tol)
+    assert_np_equal(v32.numpy()[0], 2 * s3.numpy()[0, 2] * v3.numpy()[0, 2], tol=10 * tol)
+
+    assert_np_equal(v40.numpy()[0], 2 * s4.numpy()[0, 0] * v4.numpy()[0, 0], tol=10 * tol)
+    assert_np_equal(v41.numpy()[0], 2 * s4.numpy()[0, 1] * v4.numpy()[0, 1], tol=10 * tol)
+    assert_np_equal(v42.numpy()[0], 2 * s4.numpy()[0, 2] * v4.numpy()[0, 2], tol=10 * tol)
+    assert_np_equal(v43.numpy()[0], 2 * s4.numpy()[0, 3] * v4.numpy()[0, 3], tol=10 * tol)
+
+    assert_np_equal(v50.numpy()[0], 2 * s5.numpy()[0, 0] * v5.numpy()[0, 0], tol=10 * tol)
+    assert_np_equal(v51.numpy()[0], 2 * s5.numpy()[0, 1] * v5.numpy()[0, 1], tol=10 * tol)
+    assert_np_equal(v52.numpy()[0], 2 * s5.numpy()[0, 2] * v5.numpy()[0, 2], tol=10 * tol)
+    assert_np_equal(v53.numpy()[0], 2 * s5.numpy()[0, 3] * v5.numpy()[0, 3], tol=10 * tol)
+    assert_np_equal(v54.numpy()[0], 2 * s5.numpy()[0, 4] * v5.numpy()[0, 4], tol=10 * tol)
+
+    incmps = np.concatenate([v.numpy()[0] for v in [v2, v3, v4, v5]])
+    scmps = np.concatenate([v.numpy()[0] for v in [s2, s3, s4, s5]])
+
+    if dtype in np_float_types:
+        for i, l in enumerate([v20, v21, v30, v31, v32, v40, v41, v42, v43, v50, v51, v52, v53, v54]):
+            tape.backward(loss=l)
+            sgrads = np.concatenate([tape.gradients[v].numpy()[0] for v in [s2, s3, s4, s5]])
+            expected_grads = np.zeros_like(sgrads)
+            expected_grads[i] = incmps[i] * 2
+            assert_np_equal(sgrads, expected_grads, tol=10 * tol)
+
+            allgrads = np.concatenate([tape.gradients[v].numpy()[0] for v in [v2, v3, v4, v5]])
+            expected_grads = np.zeros_like(allgrads)
+            expected_grads[i] = scmps[i] * 2
+            assert_np_equal(allgrads, expected_grads, tol=10 * tol)
+
+            tape.zero()
+
+
+def test_scalar_division(test, device, dtype, register_kernels=False):
+    rng = np.random.default_rng(123)
+
+    tol = {
+        np.float16: 5.0e-3,
+        np.float32: 1.0e-6,
+        np.float64: 1.0e-8,
+    }.get(dtype, 0)
+
+    wptype = wp.types.np_dtype_to_warp_type[np.dtype(dtype)]
+    vec2 = wp.types.vector(length=2, dtype=wptype)
+    vec3 = wp.types.vector(length=3, dtype=wptype)
+    vec4 = wp.types.vector(length=4, dtype=wptype)
+    vec5 = wp.types.vector(length=5, dtype=wptype)
+
+    def check_div(
+        s: wp.array(dtype=wptype),
+        v2: wp.array(dtype=vec2),
+        v3: wp.array(dtype=vec3),
+        v4: wp.array(dtype=vec4),
+        v5: wp.array(dtype=vec5),
+        v20: wp.array(dtype=wptype),
+        v21: wp.array(dtype=wptype),
+        v30: wp.array(dtype=wptype),
+        v31: wp.array(dtype=wptype),
+        v32: wp.array(dtype=wptype),
+        v40: wp.array(dtype=wptype),
+        v41: wp.array(dtype=wptype),
+        v42: wp.array(dtype=wptype),
+        v43: wp.array(dtype=wptype),
+        v50: wp.array(dtype=wptype),
+        v51: wp.array(dtype=wptype),
+        v52: wp.array(dtype=wptype),
+        v53: wp.array(dtype=wptype),
+        v54: wp.array(dtype=wptype),
+    ):
+        v2result = v2[0] / s[0]
+        v3result = v3[0] / s[0]
+        v4result = v4[0] / s[0]
+        v5result = v5[0] / s[0]
+
+        v20[0] = wptype(2) * v2result[0]
+        v21[0] = wptype(2) * v2result[1]
+
+        v30[0] = wptype(2) * v3result[0]
+        v31[0] = wptype(2) * v3result[1]
+        v32[0] = wptype(2) * v3result[2]
+
+        v40[0] = wptype(2) * v4result[0]
+        v41[0] = wptype(2) * v4result[1]
+        v42[0] = wptype(2) * v4result[2]
+        v43[0] = wptype(2) * v4result[3]
+
+        v50[0] = wptype(2) * v5result[0]
+        v51[0] = wptype(2) * v5result[1]
+        v52[0] = wptype(2) * v5result[2]
+        v53[0] = wptype(2) * v5result[3]
+        v54[0] = wptype(2) * v5result[4]
+
+    kernel = getkernel(check_div, suffix=dtype.__name__)
+
+    if register_kernels:
+        return
+
+    s = wp.array(randvals(rng, [1], dtype), requires_grad=True, device=device)
+    v2 = wp.array(randvals(rng, (1, 2), dtype), dtype=vec2, requires_grad=True, device=device)
+    v3 = wp.array(randvals(rng, (1, 3), dtype), dtype=vec3, requires_grad=True, device=device)
+    v4 = wp.array(randvals(rng, (1, 4), dtype), dtype=vec4, requires_grad=True, device=device)
+    v5 = wp.array(randvals(rng, (1, 5), dtype), dtype=vec5, requires_grad=True, device=device)
+    v20 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v21 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v30 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v31 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v32 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v40 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v41 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v42 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v43 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v50 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v51 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v52 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v53 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v54 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    tape = wp.Tape()
+    with tape:
+        wp.launch(
+            kernel,
+            dim=1,
+            inputs=[
+                s,
+                v2,
+                v3,
+                v4,
+                v5,
+            ],
+            outputs=[v20, v21, v30, v31, v32, v40, v41, v42, v43, v50, v51, v52, v53, v54],
+            device=device,
+        )
+
+    if dtype in np_int_types:
+        assert_np_equal(v20.numpy()[0], 2 * (v2.numpy()[0, 0] // (s.numpy()[0])), tol=tol)
+        assert_np_equal(v21.numpy()[0], 2 * (v2.numpy()[0, 1] // (s.numpy()[0])), tol=tol)
+
+        assert_np_equal(v30.numpy()[0], 2 * (v3.numpy()[0, 0] // (s.numpy()[0])), tol=10 * tol)
+        assert_np_equal(v31.numpy()[0], 2 * (v3.numpy()[0, 1] // (s.numpy()[0])), tol=10 * tol)
+        assert_np_equal(v32.numpy()[0], 2 * (v3.numpy()[0, 2] // (s.numpy()[0])), tol=10 * tol)
+
+        assert_np_equal(v40.numpy()[0], 2 * (v4.numpy()[0, 0] // (s.numpy()[0])), tol=10 * tol)
+        assert_np_equal(v41.numpy()[0], 2 * (v4.numpy()[0, 1] // (s.numpy()[0])), tol=10 * tol)
+        assert_np_equal(v42.numpy()[0], 2 * (v4.numpy()[0, 2] // (s.numpy()[0])), tol=10 * tol)
+        assert_np_equal(v43.numpy()[0], 2 * (v4.numpy()[0, 3] // (s.numpy()[0])), tol=10 * tol)
+
+        assert_np_equal(v50.numpy()[0], 2 * (v5.numpy()[0, 0] // (s.numpy()[0])), tol=10 * tol)
+        assert_np_equal(v51.numpy()[0], 2 * (v5.numpy()[0, 1] // (s.numpy()[0])), tol=10 * tol)
+        assert_np_equal(v52.numpy()[0], 2 * (v5.numpy()[0, 2] // (s.numpy()[0])), tol=10 * tol)
+        assert_np_equal(v53.numpy()[0], 2 * (v5.numpy()[0, 3] // (s.numpy()[0])), tol=10 * tol)
+        assert_np_equal(v54.numpy()[0], 2 * (v5.numpy()[0, 4] // (s.numpy()[0])), tol=10 * tol)
+
+    else:
+        assert_np_equal(v20.numpy()[0], 2 * v2.numpy()[0, 0] / (s.numpy()[0]), tol=tol)
+        assert_np_equal(v21.numpy()[0], 2 * v2.numpy()[0, 1] / (s.numpy()[0]), tol=tol)
+
+        assert_np_equal(v30.numpy()[0], 2 * v3.numpy()[0, 0] / (s.numpy()[0]), tol=10 * tol)
+        assert_np_equal(v31.numpy()[0], 2 * v3.numpy()[0, 1] / (s.numpy()[0]), tol=10 * tol)
+        assert_np_equal(v32.numpy()[0], 2 * v3.numpy()[0, 2] / (s.numpy()[0]), tol=10 * tol)
+
+        assert_np_equal(v40.numpy()[0], 2 * v4.numpy()[0, 0] / (s.numpy()[0]), tol=10 * tol)
+        assert_np_equal(v41.numpy()[0], 2 * v4.numpy()[0, 1] / (s.numpy()[0]), tol=10 * tol)
+        assert_np_equal(v42.numpy()[0], 2 * v4.numpy()[0, 2] / (s.numpy()[0]), tol=10 * tol)
+        assert_np_equal(v43.numpy()[0], 2 * v4.numpy()[0, 3] / (s.numpy()[0]), tol=10 * tol)
+
+        assert_np_equal(v50.numpy()[0], 2 * v5.numpy()[0, 0] / (s.numpy()[0]), tol=10 * tol)
+        assert_np_equal(v51.numpy()[0], 2 * v5.numpy()[0, 1] / (s.numpy()[0]), tol=10 * tol)
+        assert_np_equal(v52.numpy()[0], 2 * v5.numpy()[0, 2] / (s.numpy()[0]), tol=10 * tol)
+        assert_np_equal(v53.numpy()[0], 2 * v5.numpy()[0, 3] / (s.numpy()[0]), tol=10 * tol)
+        assert_np_equal(v54.numpy()[0], 2 * v5.numpy()[0, 4] / (s.numpy()[0]), tol=10 * tol)
+
+    incmps = np.concatenate([v.numpy()[0] for v in [v2, v3, v4, v5]])
+
+    if dtype in np_float_types:
+        for i, l in enumerate([v20, v21, v30, v31, v32, v40, v41, v42, v43, v50, v51, v52, v53, v54]):
+            tape.backward(loss=l)
+            sgrad = tape.gradients[s].numpy()[0]
+
+            # d/ds v/s = -v/s^2
+            assert_np_equal(sgrad, -2 * incmps[i] / (s.numpy()[0] * s.numpy()[0]), tol=10 * tol)
+
+            allgrads = np.concatenate([tape.gradients[v].numpy()[0] for v in [v2, v3, v4, v5]])
+            expected_grads = np.zeros_like(allgrads)
+            expected_grads[i] = 2 / s.numpy()[0]
+
+            # d/dv v/s = 1/s
+            assert_np_equal(allgrads, expected_grads, tol=tol)
+            tape.zero()
+
+
+def test_cw_division(test, device, dtype, register_kernels=False):
+    rng = np.random.default_rng(123)
+
+    tol = {
+        np.float16: 1.0e-2,
+        np.float32: 1.0e-6,
+        np.float64: 1.0e-8,
+    }.get(dtype, 0)
+
+    wptype = wp.types.np_dtype_to_warp_type[np.dtype(dtype)]
+    vec2 = wp.types.vector(length=2, dtype=wptype)
+    vec3 = wp.types.vector(length=3, dtype=wptype)
+    vec4 = wp.types.vector(length=4, dtype=wptype)
+    vec5 = wp.types.vector(length=5, dtype=wptype)
+
+    def check_cw_div(
+        s2: wp.array(dtype=vec2),
+        s3: wp.array(dtype=vec3),
+        s4: wp.array(dtype=vec4),
+        s5: wp.array(dtype=vec5),
+        v2: wp.array(dtype=vec2),
+        v3: wp.array(dtype=vec3),
+        v4: wp.array(dtype=vec4),
+        v5: wp.array(dtype=vec5),
+        v20: wp.array(dtype=wptype),
+        v21: wp.array(dtype=wptype),
+        v30: wp.array(dtype=wptype),
+        v31: wp.array(dtype=wptype),
+        v32: wp.array(dtype=wptype),
+        v40: wp.array(dtype=wptype),
+        v41: wp.array(dtype=wptype),
+        v42: wp.array(dtype=wptype),
+        v43: wp.array(dtype=wptype),
+        v50: wp.array(dtype=wptype),
+        v51: wp.array(dtype=wptype),
+        v52: wp.array(dtype=wptype),
+        v53: wp.array(dtype=wptype),
+        v54: wp.array(dtype=wptype),
+    ):
+        v2result = wp.cw_div(v2[0], s2[0])
+        v3result = wp.cw_div(v3[0], s3[0])
+        v4result = wp.cw_div(v4[0], s4[0])
+        v5result = wp.cw_div(v5[0], s5[0])
+
+        v20[0] = wptype(2) * v2result[0]
+        v21[0] = wptype(2) * v2result[1]
+
+        v30[0] = wptype(2) * v3result[0]
+        v31[0] = wptype(2) * v3result[1]
+        v32[0] = wptype(2) * v3result[2]
+
+        v40[0] = wptype(2) * v4result[0]
+        v41[0] = wptype(2) * v4result[1]
+        v42[0] = wptype(2) * v4result[2]
+        v43[0] = wptype(2) * v4result[3]
+
+        v50[0] = wptype(2) * v5result[0]
+        v51[0] = wptype(2) * v5result[1]
+        v52[0] = wptype(2) * v5result[2]
+        v53[0] = wptype(2) * v5result[3]
+        v54[0] = wptype(2) * v5result[4]
+
+    kernel = getkernel(check_cw_div, suffix=dtype.__name__)
+
+    if register_kernels:
+        return
+
+    s2 = wp.array(randvals(rng, (1, 2), dtype), dtype=vec2, requires_grad=True, device=device)
+    s3 = wp.array(randvals(rng, (1, 3), dtype), dtype=vec3, requires_grad=True, device=device)
+    s4 = wp.array(randvals(rng, (1, 4), dtype), dtype=vec4, requires_grad=True, device=device)
+    s5 = wp.array(randvals(rng, (1, 5), dtype), dtype=vec5, requires_grad=True, device=device)
+    v2 = wp.array(randvals(rng, (1, 2), dtype), dtype=vec2, requires_grad=True, device=device)
+    v3 = wp.array(randvals(rng, (1, 3), dtype), dtype=vec3, requires_grad=True, device=device)
+    v4 = wp.array(randvals(rng, (1, 4), dtype), dtype=vec4, requires_grad=True, device=device)
+    v5 = wp.array(randvals(rng, (1, 5), dtype), dtype=vec5, requires_grad=True, device=device)
+    v20 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v21 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v30 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v31 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v32 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v40 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v41 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v42 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v43 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v50 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v51 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v52 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v53 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v54 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    tape = wp.Tape()
+    with tape:
+        wp.launch(
+            kernel,
+            dim=1,
+            inputs=[
+                s2,
+                s3,
+                s4,
+                s5,
+                v2,
+                v3,
+                v4,
+                v5,
+            ],
+            outputs=[v20, v21, v30, v31, v32, v40, v41, v42, v43, v50, v51, v52, v53, v54],
+            device=device,
+        )
+
+    if dtype in np_int_types:
+        assert_np_equal(v20.numpy()[0], 2 * (v2.numpy()[0, 0] // s2.numpy()[0, 0]), tol=tol)
+        assert_np_equal(v21.numpy()[0], 2 * (v2.numpy()[0, 1] // s2.numpy()[0, 1]), tol=tol)
+
+        assert_np_equal(v30.numpy()[0], 2 * (v3.numpy()[0, 0] // s3.numpy()[0, 0]), tol=tol)
+        assert_np_equal(v31.numpy()[0], 2 * (v3.numpy()[0, 1] // s3.numpy()[0, 1]), tol=tol)
+        assert_np_equal(v32.numpy()[0], 2 * (v3.numpy()[0, 2] // s3.numpy()[0, 2]), tol=tol)
+
+        assert_np_equal(v40.numpy()[0], 2 * (v4.numpy()[0, 0] // s4.numpy()[0, 0]), tol=tol)
+        assert_np_equal(v41.numpy()[0], 2 * (v4.numpy()[0, 1] // s4.numpy()[0, 1]), tol=tol)
+        assert_np_equal(v42.numpy()[0], 2 * (v4.numpy()[0, 2] // s4.numpy()[0, 2]), tol=tol)
+        assert_np_equal(v43.numpy()[0], 2 * (v4.numpy()[0, 3] // s4.numpy()[0, 3]), tol=tol)
+
+        assert_np_equal(v50.numpy()[0], 2 * (v5.numpy()[0, 0] // s5.numpy()[0, 0]), tol=tol)
+        assert_np_equal(v51.numpy()[0], 2 * (v5.numpy()[0, 1] // s5.numpy()[0, 1]), tol=tol)
+        assert_np_equal(v52.numpy()[0], 2 * (v5.numpy()[0, 2] // s5.numpy()[0, 2]), tol=tol)
+        assert_np_equal(v53.numpy()[0], 2 * (v5.numpy()[0, 3] // s5.numpy()[0, 3]), tol=tol)
+        assert_np_equal(v54.numpy()[0], 2 * (v5.numpy()[0, 4] // s5.numpy()[0, 4]), tol=tol)
+    else:
+        assert_np_equal(v20.numpy()[0], 2 * v2.numpy()[0, 0] / s2.numpy()[0, 0], tol=tol)
+        assert_np_equal(v21.numpy()[0], 2 * v2.numpy()[0, 1] / s2.numpy()[0, 1], tol=tol)
+
+        assert_np_equal(v30.numpy()[0], 2 * v3.numpy()[0, 0] / s3.numpy()[0, 0], tol=tol)
+        assert_np_equal(v31.numpy()[0], 2 * v3.numpy()[0, 1] / s3.numpy()[0, 1], tol=tol)
+        assert_np_equal(v32.numpy()[0], 2 * v3.numpy()[0, 2] / s3.numpy()[0, 2], tol=tol)
+
+        assert_np_equal(v40.numpy()[0], 2 * v4.numpy()[0, 0] / s4.numpy()[0, 0], tol=tol)
+        assert_np_equal(v41.numpy()[0], 2 * v4.numpy()[0, 1] / s4.numpy()[0, 1], tol=tol)
+        assert_np_equal(v42.numpy()[0], 2 * v4.numpy()[0, 2] / s4.numpy()[0, 2], tol=tol)
+        assert_np_equal(v43.numpy()[0], 2 * v4.numpy()[0, 3] / s4.numpy()[0, 3], tol=tol)
+
+        assert_np_equal(v50.numpy()[0], 2 * v5.numpy()[0, 0] / s5.numpy()[0, 0], tol=tol)
+        assert_np_equal(v51.numpy()[0], 2 * v5.numpy()[0, 1] / s5.numpy()[0, 1], tol=tol)
+        assert_np_equal(v52.numpy()[0], 2 * v5.numpy()[0, 2] / s5.numpy()[0, 2], tol=tol)
+        assert_np_equal(v53.numpy()[0], 2 * v5.numpy()[0, 3] / s5.numpy()[0, 3], tol=tol)
+        assert_np_equal(v54.numpy()[0], 2 * v5.numpy()[0, 4] / s5.numpy()[0, 4], tol=tol)
+
+    if dtype in np_float_types:
+        incmps = np.concatenate([v.numpy()[0] for v in [v2, v3, v4, v5]])
+        scmps = np.concatenate([v.numpy()[0] for v in [s2, s3, s4, s5]])
+
+        for i, l in enumerate([v20, v21, v30, v31, v32, v40, v41, v42, v43, v50, v51, v52, v53, v54]):
+            tape.backward(loss=l)
+            sgrads = np.concatenate([tape.gradients[v].numpy()[0] for v in [s2, s3, s4, s5]])
+            expected_grads = np.zeros_like(sgrads)
+
+            # d/ds v/s = -v/s^2
+            expected_grads[i] = -incmps[i] * 2 / (scmps[i] * scmps[i])
+            assert_np_equal(sgrads, expected_grads, tol=20 * tol)
+
+            allgrads = np.concatenate([tape.gradients[v].numpy()[0] for v in [v2, v3, v4, v5]])
+            expected_grads = np.zeros_like(allgrads)
+
+            # d/dv v/s = 1/s
+            expected_grads[i] = 2 / scmps[i]
+            assert_np_equal(allgrads, expected_grads, tol=tol)
+
+            tape.zero()
+
+
+def test_addition(test, device, dtype, register_kernels=False):
+    rng = np.random.default_rng(123)
+
+    tol = {
+        np.float16: 5.0e-3,
+        np.float32: 1.0e-6,
+        np.float64: 1.0e-8,
+    }.get(dtype, 0)
+
+    wptype = wp.types.np_dtype_to_warp_type[np.dtype(dtype)]
+    vec2 = wp.types.vector(length=2, dtype=wptype)
+    vec3 = wp.types.vector(length=3, dtype=wptype)
+    vec4 = wp.types.vector(length=4, dtype=wptype)
+    vec5 = wp.types.vector(length=5, dtype=wptype)
+
+    def check_add(
+        s2: wp.array(dtype=vec2),
+        s3: wp.array(dtype=vec3),
+        s4: wp.array(dtype=vec4),
+        s5: wp.array(dtype=vec5),
+        v2: wp.array(dtype=vec2),
+        v3: wp.array(dtype=vec3),
+        v4: wp.array(dtype=vec4),
+        v5: wp.array(dtype=vec5),
+        v20: wp.array(dtype=wptype),
+        v21: wp.array(dtype=wptype),
+        v30: wp.array(dtype=wptype),
+        v31: wp.array(dtype=wptype),
+        v32: wp.array(dtype=wptype),
+        v40: wp.array(dtype=wptype),
+        v41: wp.array(dtype=wptype),
+        v42: wp.array(dtype=wptype),
+        v43: wp.array(dtype=wptype),
+        v50: wp.array(dtype=wptype),
+        v51: wp.array(dtype=wptype),
+        v52: wp.array(dtype=wptype),
+        v53: wp.array(dtype=wptype),
+        v54: wp.array(dtype=wptype),
+    ):
+        v2result = v2[0] + s2[0]
+        v3result = v3[0] + s3[0]
+        v4result = v4[0] + s4[0]
+        v5result = v5[0] + s5[0]
+
+        v20[0] = wptype(2) * v2result[0]
+        v21[0] = wptype(2) * v2result[1]
+
+        v30[0] = wptype(2) * v3result[0]
+        v31[0] = wptype(2) * v3result[1]
+        v32[0] = wptype(2) * v3result[2]
+
+        v40[0] = wptype(2) * v4result[0]
+        v41[0] = wptype(2) * v4result[1]
+        v42[0] = wptype(2) * v4result[2]
+        v43[0] = wptype(2) * v4result[3]
+
+        v50[0] = wptype(2) * v5result[0]
+        v51[0] = wptype(2) * v5result[1]
+        v52[0] = wptype(2) * v5result[2]
+        v53[0] = wptype(2) * v5result[3]
+        v54[0] = wptype(2) * v5result[4]
+
+    kernel = getkernel(check_add, suffix=dtype.__name__)
+
+    if register_kernels:
+        return
+
+    s2 = wp.array(randvals(rng, (1, 2), dtype), dtype=vec2, requires_grad=True, device=device)
+    s3 = wp.array(randvals(rng, (1, 3), dtype), dtype=vec3, requires_grad=True, device=device)
+    s4 = wp.array(randvals(rng, (1, 4), dtype), dtype=vec4, requires_grad=True, device=device)
+    s5 = wp.array(randvals(rng, (1, 5), dtype), dtype=vec5, requires_grad=True, device=device)
+    v2 = wp.array(randvals(rng, (1, 2), dtype), dtype=vec2, requires_grad=True, device=device)
+    v3 = wp.array(randvals(rng, (1, 3), dtype), dtype=vec3, requires_grad=True, device=device)
+    v4 = wp.array(randvals(rng, (1, 4), dtype), dtype=vec4, requires_grad=True, device=device)
+    v5 = wp.array(randvals(rng, (1, 5), dtype), dtype=vec5, requires_grad=True, device=device)
+    v20 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v21 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v30 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v31 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v32 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v40 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v41 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v42 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v43 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v50 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v51 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v52 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v53 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    v54 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    tape = wp.Tape()
+    with tape:
+        wp.launch(
+            kernel,
+            dim=1,
+            inputs=[
+                s2,
+                s3,
+                s4,
+                s5,
+                v2,
+                v3,
+                v4,
+                v5,
+            ],
+            outputs=[v20, v21, v30, v31, v32, v40, v41, v42, v43, v50, v51, v52, v53, v54],
+            device=device,
+        )
+
+    assert_np_equal(v20.numpy()[0], 2 * (v2.numpy()[0, 0] + s2.numpy()[0, 0]), tol=tol)
+    assert_np_equal(v21.numpy()[0], 2 * (v2.numpy()[0, 1] + s2.numpy()[0, 1]), tol=tol)
+
+    assert_np_equal(v30.numpy()[0], 2 * (v3.numpy()[0, 0] + s3.numpy()[0, 0]), tol=tol)
+    assert_np_equal(v31.numpy()[0], 2 * (v3.numpy()[0, 1] + s3.numpy()[0, 1]), tol=tol)
+    assert_np_equal(v32.numpy()[0], 2 * (v3.numpy()[0, 2] + s3.numpy()[0, 2]), tol=tol)
+
+    assert_np_equal(v40.numpy()[0], 2 * (v4.numpy()[0, 0] + s4.numpy()[0, 0]), tol=tol)
+    assert_np_equal(v41.numpy()[0], 2 * (v4.numpy()[0, 1] + s4.numpy()[0, 1]), tol=tol)
+    assert_np_equal(v42.numpy()[0], 2 * (v4.numpy()[0, 2] + s4.numpy()[0, 2]), tol=tol)
+    assert_np_equal(v43.numpy()[0], 2 * (v4.numpy()[0, 3] + s4.numpy()[0, 3]), tol=tol)
+
+    assert_np_equal(v50.numpy()[0], 2 * (v5.numpy()[0, 0] + s5.numpy()[0, 0]), tol=tol)
+    assert_np_equal(v51.numpy()[0], 2 * (v5.numpy()[0, 1] + s5.numpy()[0, 1]), tol=tol)
+    assert_np_equal(v52.numpy()[0], 2 * (v5.numpy()[0, 2] + s5.numpy()[0, 2]), tol=tol)
+    assert_np_equal(v53.numpy()[0], 2 * (v5.numpy()[0, 3] + s5.numpy()[0, 3]), tol=tol)
+    assert_np_equal(v54.numpy()[0], 2 * (v5.numpy()[0, 4] + s5.numpy()[0, 4]), tol=2 * tol)
+
+    if dtype in np_float_types:
+        for i, l in enumerate([v20, v21, v30, v31, v32, v40, v41, v42, v43, v50, v51, v52, v53, v54]):
+            tape.backward(loss=l)
+            sgrads = np.concatenate([tape.gradients[v].numpy()[0] for v in [s2, s3, s4, s5]])
+            expected_grads = np.zeros_like(sgrads)
+
+            expected_grads[i] = 2
+            assert_np_equal(sgrads, expected_grads, tol=10 * tol)
+
+            allgrads = np.concatenate([tape.gradients[v].numpy()[0] for v in [v2, v3, v4, v5]])
+            assert_np_equal(allgrads, expected_grads, tol=tol)
+
+            tape.zero()
+
+
+def test_dotproduct(test, device, dtype, register_kernels=False):
+    rng = np.random.default_rng(123)
+
+    tol = {
+        np.float16: 1.0e-2,
+        np.float32: 1.0e-6,
+        np.float64: 1.0e-8,
+    }.get(dtype, 0)
+
+    wptype = wp.types.np_dtype_to_warp_type[np.dtype(dtype)]
+    vec2 = wp.types.vector(length=2, dtype=wptype)
+    vec3 = wp.types.vector(length=3, dtype=wptype)
+    vec4 = wp.types.vector(length=4, dtype=wptype)
+    vec5 = wp.types.vector(length=5, dtype=wptype)
+
+    def check_dot(
+        s2: wp.array(dtype=vec2),
+        s3: wp.array(dtype=vec3),
+        s4: wp.array(dtype=vec4),
+        s5: wp.array(dtype=vec5),
+        v2: wp.array(dtype=vec2),
+        v3: wp.array(dtype=vec3),
+        v4: wp.array(dtype=vec4),
+        v5: wp.array(dtype=vec5),
+        dot2: wp.array(dtype=wptype),
+        dot3: wp.array(dtype=wptype),
+        dot4: wp.array(dtype=wptype),
+        dot5: wp.array(dtype=wptype),
+    ):
+        dot2[0] = wptype(2) * wp.dot(v2[0], s2[0])
+        dot3[0] = wptype(2) * wp.dot(v3[0], s3[0])
+        dot4[0] = wptype(2) * wp.dot(v4[0], s4[0])
+        dot5[0] = wptype(2) * wp.dot(v5[0], s5[0])
+
+    kernel = getkernel(check_dot, suffix=dtype.__name__)
+
+    if register_kernels:
+        return
+
+    s2 = wp.array(randvals(rng, (1, 2), dtype), dtype=vec2, requires_grad=True, device=device)
+    s3 = wp.array(randvals(rng, (1, 3), dtype), dtype=vec3, requires_grad=True, device=device)
+    s4 = wp.array(randvals(rng, (1, 4), dtype), dtype=vec4, requires_grad=True, device=device)
+    s5 = wp.array(randvals(rng, (1, 5), dtype), dtype=vec5, requires_grad=True, device=device)
+    v2 = wp.array(randvals(rng, (1, 2), dtype), dtype=vec2, requires_grad=True, device=device)
+    v3 = wp.array(randvals(rng, (1, 3), dtype), dtype=vec3, requires_grad=True, device=device)
+    v4 = wp.array(randvals(rng, (1, 4), dtype), dtype=vec4, requires_grad=True, device=device)
+    v5 = wp.array(randvals(rng, (1, 5), dtype), dtype=vec5, requires_grad=True, device=device)
+    dot2 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    dot3 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    dot4 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    dot5 = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    tape = wp.Tape()
+    with tape:
+        wp.launch(
+            kernel,
+            dim=1,
+            inputs=[
+                s2,
+                s3,
+                s4,
+                s5,
+                v2,
+                v3,
+                v4,
+                v5,
+            ],
+            outputs=[dot2, dot3, dot4, dot5],
+            device=device,
+        )
+
+    assert_np_equal(dot2.numpy()[0], 2.0 * (v2.numpy() * s2.numpy()).sum(), tol=10 * tol)
+    assert_np_equal(dot3.numpy()[0], 2.0 * (v3.numpy() * s3.numpy()).sum(), tol=10 * tol)
+    assert_np_equal(dot4.numpy()[0], 2.0 * (v4.numpy() * s4.numpy()).sum(), tol=10 * tol)
+    assert_np_equal(dot5.numpy()[0], 2.0 * (v5.numpy() * s5.numpy()).sum(), tol=10 * tol)
+
+    if dtype in np_float_types:
+        tape.backward(loss=dot2)
+        sgrads = tape.gradients[s2].numpy()[0]
+        expected_grads = 2.0 * v2.numpy()[0]
+        assert_np_equal(sgrads, expected_grads, tol=10 * tol)
+
+        vgrads = tape.gradients[v2].numpy()[0]
+        expected_grads = 2.0 * s2.numpy()[0]
+        assert_np_equal(vgrads, expected_grads, tol=tol)
+
+        tape.zero()
+
+        tape.backward(loss=dot3)
+        sgrads = tape.gradients[s3].numpy()[0]
+        expected_grads = 2.0 * v3.numpy()[0]
+        assert_np_equal(sgrads, expected_grads, tol=10 * tol)
+
+        vgrads = tape.gradients[v3].numpy()[0]
+        expected_grads = 2.0 * s3.numpy()[0]
+        assert_np_equal(vgrads, expected_grads, tol=tol)
+
+        tape.zero()
+
+        tape.backward(loss=dot4)
+        sgrads = tape.gradients[s4].numpy()[0]
+        expected_grads = 2.0 * v4.numpy()[0]
+        assert_np_equal(sgrads, expected_grads, tol=10 * tol)
+
+        vgrads = tape.gradients[v4].numpy()[0]
+        expected_grads = 2.0 * s4.numpy()[0]
+        assert_np_equal(vgrads, expected_grads, tol=tol)
+
+        tape.zero()
+
+        tape.backward(loss=dot5)
+        sgrads = tape.gradients[s5].numpy()[0]
+        expected_grads = 2.0 * v5.numpy()[0]
+        assert_np_equal(sgrads, expected_grads, tol=10 * tol)
+
+        vgrads = tape.gradients[v5].numpy()[0]
+        expected_grads = 2.0 * s5.numpy()[0]
+        assert_np_equal(vgrads, expected_grads, tol=10 * tol)
+
+        tape.zero()
+
+
+def test_equivalent_types(test, device, dtype, register_kernels=False):
+    wptype = wp.types.np_dtype_to_warp_type[np.dtype(dtype)]
+
+    # vector types
+    vec2 = wp.types.vector(length=2, dtype=wptype)
+    vec3 = wp.types.vector(length=3, dtype=wptype)
+    vec4 = wp.types.vector(length=4, dtype=wptype)
+    vec5 = wp.types.vector(length=5, dtype=wptype)
+
+    # vector types equivalent to the above
+    vec2_equiv = wp.types.vector(length=2, dtype=wptype)
+    vec3_equiv = wp.types.vector(length=3, dtype=wptype)
+    vec4_equiv = wp.types.vector(length=4, dtype=wptype)
+    vec5_equiv = wp.types.vector(length=5, dtype=wptype)
+
+    # declare kernel with original types
+    def check_equivalence(
+        v2: vec2,
+        v3: vec3,
+        v4: vec4,
+        v5: vec5,
+    ):
+        wp.expect_eq(v2, vec2(wptype(1), wptype(2)))
+        wp.expect_eq(v3, vec3(wptype(1), wptype(2), wptype(3)))
+        wp.expect_eq(v4, vec4(wptype(1), wptype(2), wptype(3), wptype(4)))
+        wp.expect_eq(v5, vec5(wptype(1), wptype(2), wptype(3), wptype(4), wptype(5)))
+
+        wp.expect_eq(v2, vec2_equiv(wptype(1), wptype(2)))
+        wp.expect_eq(v3, vec3_equiv(wptype(1), wptype(2), wptype(3)))
+        wp.expect_eq(v4, vec4_equiv(wptype(1), wptype(2), wptype(3), wptype(4)))
+        wp.expect_eq(v5, vec5_equiv(wptype(1), wptype(2), wptype(3), wptype(4), wptype(5)))
+
+    kernel = getkernel(check_equivalence, suffix=dtype.__name__)
+
+    if register_kernels:
+        return
+
+    # call kernel with equivalent types
+    v2 = vec2_equiv(1, 2)
+    v3 = vec3_equiv(1, 2, 3)
+    v4 = vec4_equiv(1, 2, 3, 4)
+    v5 = vec5_equiv(1, 2, 3, 4, 5)
+
+    wp.launch(kernel, dim=1, inputs=[v2, v3, v4, v5], device=device)
+
+
+def test_conversions(test, device, dtype, register_kernels=False):
+    def check_vectors_equal(
+        v0: wp.vec3,
+        v1: wp.vec3,
+        v2: wp.vec3,
+        v3: wp.vec3,
+    ):
+        wp.expect_eq(v1, v0)
+        wp.expect_eq(v2, v0)
+        wp.expect_eq(v3, v0)
+
+    kernel = getkernel(check_vectors_equal, suffix=dtype.__name__)
+
+    if register_kernels:
+        return
+
+    v0 = wp.vec3(1, 2, 3)
+
+    # test explicit conversions - constructing vectors from different containers
+    v1 = wp.vec3((1, 2, 3))
+    v2 = wp.vec3([1, 2, 3])
+    v3 = wp.vec3(np.array([1, 2, 3], dtype=dtype))
+
+    wp.launch(kernel, dim=1, inputs=[v0, v1, v2, v3], device=device)
+
+    # test implicit conversions - passing different containers as vectors to wp.launch()
+    v1 = (1, 2, 3)
+    v2 = [1, 2, 3]
+    v3 = np.array([1, 2, 3], dtype=dtype)
+
+    wp.launch(kernel, dim=1, inputs=[v0, v1, v2, v3], device=device)
+
+
+def test_constants(test, device, dtype, register_kernels=False):
+    wptype = wp.types.np_dtype_to_warp_type[np.dtype(dtype)]
+    vec2 = wp.types.vector(length=2, dtype=wptype)
+    vec3 = wp.types.vector(length=3, dtype=wptype)
+    vec4 = wp.types.vector(length=4, dtype=wptype)
+    vec5 = wp.types.vector(length=5, dtype=wptype)
+
+    cv2 = wp.constant(vec2(1, 2))
+    cv3 = wp.constant(vec3(1, 2, 3))
+    cv4 = wp.constant(vec4(1, 2, 3, 4))
+    cv5 = wp.constant(vec5(1, 2, 3, 4, 5))
+
+    def check_vector_constants():
+        wp.expect_eq(cv2, vec2(wptype(1), wptype(2)))
+        wp.expect_eq(cv3, vec3(wptype(1), wptype(2), wptype(3)))
+        wp.expect_eq(cv4, vec4(wptype(1), wptype(2), wptype(3), wptype(4)))
+        wp.expect_eq(cv5, vec5(wptype(1), wptype(2), wptype(3), wptype(4), wptype(5)))
+
+    kernel = getkernel(check_vector_constants, suffix=dtype.__name__)
+
+    if register_kernels:
+        return
+
+    wp.launch(kernel, dim=1, inputs=[])
+
+
+def test_minmax(test, device, dtype, register_kernels=False):
+    rng = np.random.default_rng(123)
+
+    # \TODO: not quite sure why, but the numbers are off for 16 bit float
+    # on the cpu (but not cuda). This is probably just the sketchy float16
+    # arithmetic I implemented to get all this stuff working, so
+    # hopefully that can be fixed when we do that correctly.
+    tol = {
+        np.float16: 1.0e-2,
+    }.get(dtype, 0)
+
+    wptype = wp.types.np_dtype_to_warp_type[np.dtype(dtype)]
+    vec2 = wp.types.vector(length=2, dtype=wptype)
+    vec3 = wp.types.vector(length=3, dtype=wptype)
+    vec4 = wp.types.vector(length=4, dtype=wptype)
+    vec5 = wp.types.vector(length=5, dtype=wptype)
+
+    # \TODO: Also not quite sure why: this kernel compiles incredibly
+    # slowly though...
+    def check_vec_min_max(
+        a: wp.array(dtype=wptype, ndim=2),
+        b: wp.array(dtype=wptype, ndim=2),
+        mins: wp.array(dtype=wptype, ndim=2),
+        maxs: wp.array(dtype=wptype, ndim=2),
+    ):
+        for i in range(10):
+            # multiplying by 2 so we've got something to backpropagate:
+            a2read = vec2(a[i, 0], a[i, 1])
+            b2read = vec2(b[i, 0], b[i, 1])
+            c2 = wptype(2) * wp.min(a2read, b2read)
+            d2 = wptype(2) * wp.max(a2read, b2read)
+
+            a3read = vec3(a[i, 2], a[i, 3], a[i, 4])
+            b3read = vec3(b[i, 2], b[i, 3], b[i, 4])
+            c3 = wptype(2) * wp.min(a3read, b3read)
+            d3 = wptype(2) * wp.max(a3read, b3read)
+
+            a4read = vec4(a[i, 5], a[i, 6], a[i, 7], a[i, 8])
+            b4read = vec4(b[i, 5], b[i, 6], b[i, 7], b[i, 8])
+            c4 = wptype(2) * wp.min(a4read, b4read)
+            d4 = wptype(2) * wp.max(a4read, b4read)
+
+            a5read = vec5(a[i, 9], a[i, 10], a[i, 11], a[i, 12], a[i, 13])
+            b5read = vec5(b[i, 9], b[i, 10], b[i, 11], b[i, 12], b[i, 13])
+            c5 = wptype(2) * wp.min(a5read, b5read)
+            d5 = wptype(2) * wp.max(a5read, b5read)
+
+            mins[i, 0] = c2[0]
+            mins[i, 1] = c2[1]
+
+            mins[i, 2] = c3[0]
+            mins[i, 3] = c3[1]
+            mins[i, 4] = c3[2]
+
+            mins[i, 5] = c4[0]
+            mins[i, 6] = c4[1]
+            mins[i, 7] = c4[2]
+            mins[i, 8] = c4[3]
+
+            mins[i, 9] = c5[0]
+            mins[i, 10] = c5[1]
+            mins[i, 11] = c5[2]
+            mins[i, 12] = c5[3]
+            mins[i, 13] = c5[4]
+
+            maxs[i, 0] = d2[0]
+            maxs[i, 1] = d2[1]
+
+            maxs[i, 2] = d3[0]
+            maxs[i, 3] = d3[1]
+            maxs[i, 4] = d3[2]
+
+            maxs[i, 5] = d4[0]
+            maxs[i, 6] = d4[1]
+            maxs[i, 7] = d4[2]
+            maxs[i, 8] = d4[3]
+
+            maxs[i, 9] = d5[0]
+            maxs[i, 10] = d5[1]
+            maxs[i, 11] = d5[2]
+            maxs[i, 12] = d5[3]
+            maxs[i, 13] = d5[4]
+
+    kernel = getkernel(check_vec_min_max, suffix=dtype.__name__)
+    output_select_kernel = get_select_kernel2(wptype)
+
+    if register_kernels:
+        return
+
+    a = wp.array(randvals(rng, (10, 14), dtype), dtype=wptype, requires_grad=True, device=device)
+    b = wp.array(randvals(rng, (10, 14), dtype), dtype=wptype, requires_grad=True, device=device)
+
+    mins = wp.zeros((10, 14), dtype=wptype, requires_grad=True, device=device)
+    maxs = wp.zeros((10, 14), dtype=wptype, requires_grad=True, device=device)
+
+    tape = wp.Tape()
+    with tape:
+        wp.launch(kernel, dim=1, inputs=[a, b], outputs=[mins, maxs], device=device)
+
+    assert_np_equal(mins.numpy(), 2 * np.minimum(a.numpy(), b.numpy()), tol=tol)
+    assert_np_equal(maxs.numpy(), 2 * np.maximum(a.numpy(), b.numpy()), tol=tol)
+
+    out = wp.zeros(1, dtype=wptype, requires_grad=True, device=device)
+    if dtype in np_float_types:
+        for i in range(10):
+            for j in range(14):
+                tape = wp.Tape()
+                with tape:
+                    wp.launch(kernel, dim=1, inputs=[a, b], outputs=[mins, maxs], device=device)
+                    wp.launch(output_select_kernel, dim=1, inputs=[mins, i, j], outputs=[out], device=device)
+
+                tape.backward(loss=out)
+                expected = np.zeros_like(a.numpy())
+                expected[i, j] = 2 if (a.numpy()[i, j] < b.numpy()[i, j]) else 0
+                assert_np_equal(tape.gradients[a].numpy(), expected, tol=tol)
+                expected[i, j] = 2 if (b.numpy()[i, j] < a.numpy()[i, j]) else 0
+                assert_np_equal(tape.gradients[b].numpy(), expected, tol=tol)
+                tape.zero()
+
+                tape = wp.Tape()
+                with tape:
+                    wp.launch(kernel, dim=1, inputs=[a, b], outputs=[mins, maxs], device=device)
+                    wp.launch(output_select_kernel, dim=1, inputs=[maxs, i, j], outputs=[out], device=device)
+
+                tape.backward(loss=out)
+                expected = np.zeros_like(a.numpy())
+                expected[i, j] = 2 if (a.numpy()[i, j] > b.numpy()[i, j]) else 0
+                assert_np_equal(tape.gradients[a].numpy(), expected, tol=tol)
+                expected[i, j] = 2 if (b.numpy()[i, j] > a.numpy()[i, j]) else 0
+                assert_np_equal(tape.gradients[b].numpy(), expected, tol=tol)
+                tape.zero()
+
+
+devices = get_test_devices()
+
+
+class TestVecScalarOps(unittest.TestCase):
+    pass
+
+
+for dtype in np_scalar_types:
+    add_function_test(TestVecScalarOps, f"test_arrays_{dtype.__name__}", test_arrays, devices=devices, dtype=dtype)
+    add_function_test(TestVecScalarOps, f"test_components_{dtype.__name__}", test_components, devices=None, dtype=dtype)
+    add_function_test(
+        TestVecScalarOps, f"test_py_arithmetic_ops_{dtype.__name__}", test_py_arithmetic_ops, devices=None, dtype=dtype
+    )
+    add_function_test_register_kernel(
+        TestVecScalarOps, f"test_constructors_{dtype.__name__}", test_constructors, devices=devices, dtype=dtype
+    )
+    add_function_test_register_kernel(
+        TestVecScalarOps,
+        f"test_anon_type_instance_{dtype.__name__}",
+        test_anon_type_instance,
+        devices=devices,
+        dtype=dtype,
+    )
+    add_function_test_register_kernel(
+        TestVecScalarOps, f"test_indexing_{dtype.__name__}", test_indexing, devices=devices, dtype=dtype
+    )
+    add_function_test_register_kernel(
+        TestVecScalarOps, f"test_equality_{dtype.__name__}", test_equality, devices=devices, dtype=dtype
+    )
+    add_function_test_register_kernel(
+        TestVecScalarOps,
+        f"test_scalar_multiplication_{dtype.__name__}",
+        test_scalar_multiplication,
+        devices=devices,
+        dtype=dtype,
+    )
+    add_function_test_register_kernel(
+        TestVecScalarOps,
+        f"test_scalar_multiplication_rightmul_{dtype.__name__}",
+        test_scalar_multiplication_rightmul,
+        devices=devices,
+        dtype=dtype,
+    )
+    add_function_test_register_kernel(
+        TestVecScalarOps,
+        f"test_cw_multiplication_{dtype.__name__}",
+        test_cw_multiplication,
+        devices=devices,
+        dtype=dtype,
+    )
+    add_function_test_register_kernel(
+        TestVecScalarOps, f"test_scalar_division_{dtype.__name__}", test_scalar_division, devices=devices, dtype=dtype
+    )
+    add_function_test_register_kernel(
+        TestVecScalarOps, f"test_cw_division_{dtype.__name__}", test_cw_division, devices=devices, dtype=dtype
+    )
+    add_function_test_register_kernel(
+        TestVecScalarOps, f"test_addition_{dtype.__name__}", test_addition, devices=devices, dtype=dtype
+    )
+    add_function_test_register_kernel(
+        TestVecScalarOps, f"test_dotproduct_{dtype.__name__}", test_dotproduct, devices=devices, dtype=dtype
+    )
+    add_function_test_register_kernel(
+        TestVecScalarOps, f"test_equivalent_types_{dtype.__name__}", test_equivalent_types, devices=devices, dtype=dtype
+    )
+    add_function_test_register_kernel(
+        TestVecScalarOps, f"test_conversions_{dtype.__name__}", test_conversions, devices=devices, dtype=dtype
+    )
+    add_function_test_register_kernel(
+        TestVecScalarOps, f"test_constants_{dtype.__name__}", test_constants, devices=devices, dtype=dtype
+    )
+
+    # the kernels in this test compile incredibly slowly...
+    # add_function_test_register_kernel(TestVecScalarOps, f"test_minmax_{dtype.__name__}", test_minmax, devices=devices, dtype=dtype)
+
+
+if __name__ == "__main__":
+    wp.build.clear_kernel_cache()
+    unittest.main(verbosity=2, failfast=True)