[BUG] CuPy/Cumlarray UnownedMemory error on ARM CI jobs

[dantegd]

Describe the bug

FAILED test_adapters.py::test_check_array - RuntimeError: UnownedMemory requires explicit device ID for a null pointer.
[2147](https://github.com/rapidsai/cuml/actions/runs/3734738186/jobs/6337919174#step:6:2148)
FAILED test_svm.py::test_svm_no_support_vectors - RuntimeError: UnownedMemory requires explicit device ID for a null pointer.

Steps/Code to reproduce bug
Tests will be xfailed to merge GHA PR, but can be seen here: https://github.com/rapidsai/cuml/actions/runs/3734738186/jobs/6337919174

Full error:

 _______________________________ test_check_array _______________________________
 [gw0] linux -- Python 3.9.15 /opt/conda/envs/test/bin/python3.9
 
     def test_check_array():
         # accept_sparse
         arr = coo_matrix((3, 4), dtype=cp.float64)
         check_array(arr, accept_sparse=True)
         with pytest.raises(ValueError):
             check_array(arr, accept_sparse=False)
     
         # dtype
         arr = cp.array([[1, 2]], dtype=cp.int64)
         check_array(arr, dtype=cp.int64, copy=False)
     
         arr = cp.array([[1, 2]], dtype=cp.float32)
         new_arr = check_array(arr, dtype=cp.int64)
         assert new_arr.dtype == cp.int64
     
         # order
         arr = cp.array([[1, 2]], dtype=cp.int64, order='F')
         new_arr = check_array(arr, order='F')
         assert new_arr.flags.f_contiguous
         new_arr = check_array(arr, order='C')
         assert new_arr.flags.c_contiguous
     
         # force_all_finite
         arr = cp.array([[1, cp.inf]])
         check_array(arr, force_all_finite=False)
         with pytest.raises(ValueError):
             check_array(arr, force_all_finite=True)
     
         # ensure_2d
         arr = cp.array([1, 2], dtype=cp.float32)
         check_array(arr, ensure_2d=False)
         with pytest.raises(ValueError):
             check_array(arr, ensure_2d=True)
     
         # ensure_2d
         arr = cp.array([[1, 2, 3], [4, 5, 6]], dtype=cp.float32)
         check_array(arr, ensure_2d=True)
     
         # ensure_min_samples
         arr = cp.array([[1, 2]], dtype=cp.float32)
         check_array(arr, ensure_min_samples=1)
         with pytest.raises(ValueError):
             check_array(arr, ensure_min_samples=2)
     
         # ensure_min_features
         arr = cp.array([[]], dtype=cp.float32)
 >       check_array(arr, ensure_min_features=0)
 
 python/cuml/tests/test_adapters.py:124: 
 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
 /opt/conda/envs/test/lib/python3.9/site-packages/cuml/thirdparty_adapters/adapters.py:288: in check_array
     X, n_rows, n_cols, dtype = input_to_cupy_array(array,
 /opt/conda/envs/test/lib/python3.9/contextlib.py:79: in inner
     return func(*args, **kwds)
 /opt/conda/envs/test/lib/python3.9/site-packages/cuml/internals/input_utils.py:434: in input_to_cupy_array
     return out_data._replace(array=out_data.array.to_output("cupy"))
 /opt/conda/envs/test/lib/python3.9/site-packages/cuml/internals/memory_utils.py:85: in cupy_rmm_wrapper
     return func(*args, **kwargs)
 /opt/conda/envs/test/lib/python3.9/contextlib.py:79: in inner
     return func(*args, **kwds)
 /opt/conda/envs/test/lib/python3.9/site-packages/cuml/internals/array.py:630: in to_output
     return output_mem_type.xpy.asarray(
 /opt/conda/envs/test/lib/python3.9/site-packages/cupy/_creation/from_data.py:76: in asarray
     return _core.array(a, dtype, False, order)
 cupy/_core/core.pyx:2249: in cupy._core.core.array
     ???
 cupy/_core/core.pyx:2261: in cupy._core.core.array
     ???
 cupy/_core/core.pyx:2301: in cupy._core.core._array_from_cuda_array_interface
     ???
 cupy/_core/core.pyx:2632: in cupy._core.core._convert_object_with_cuda_array_interface
     ???
 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
 
 >   ???
 E   RuntimeError: UnownedMemory requires explicit device ID for a null pointer.
 
 cupy/cuda/memory.pyx:190: RuntimeError
 _________________________ test_svm_no_support_vectors __________________________
 [gw1] linux -- Python 3.9.15 /opt/conda/envs/test/bin/python3.9
 
     def test_svm_no_support_vectors():
         n_rows = 10
         n_cols = 3
         X = cp.random.uniform(size=(n_rows, n_cols), dtype=cp.float64)
         y = cp.ones((n_rows, 1))
         model = cuml.svm.SVR(kernel="linear", C=10)
         model.fit(X, y)
         pred = model.predict(X)
     
         assert svm_array_equal(pred, y, 0)
     
         assert model.n_support_ == 0
         assert abs(model.intercept_ - 1) <= 1e-6
         assert svm_array_equal(model.coef_, cp.zeros((1, n_cols)))
 >       assert model.dual_coef_.shape == (1, 0)
 
 python/cuml/tests/test_svm.py:552: 
 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
 /opt/conda/envs/test/lib/python3.9/site-packages/cuml/common/array_descriptor.py:134: in __get__
     return self._to_output(instance, output_type)
 /opt/conda/envs/test/lib/python3.9/site-packages/cuml/common/array_descriptor.py:99: in _to_output
     output = cuml_arr.to_output(output_type=to_output_type,
 /opt/conda/envs/test/lib/python3.9/site-packages/cuml/internals/memory_utils.py:85: in cupy_rmm_wrapper
     return func(*args, **kwargs)
 /opt/conda/envs/test/lib/python3.9/contextlib.py:79: in inner
     return func(*args, **kwds)
 /opt/conda/envs/test/lib/python3.9/site-packages/cuml/internals/array.py:630: in to_output
     return output_mem_type.xpy.asarray(
 /opt/conda/envs/test/lib/python3.9/site-packages/cupy/_creation/from_data.py:76: in asarray
     return _core.array(a, dtype, False, order)
 cupy/_core/core.pyx:2249: in cupy._core.core.array
     ???
 cupy/_core/core.pyx:2261: in cupy._core.core.array
     ???
 cupy/_core/core.pyx:2301: in cupy._core.core._array_from_cuda_array_interface
     ???
 cupy/_core/core.pyx:2632: in cupy._core.core._convert_object_with_cuda_array_interface
     ???
 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
 
 >   ???
 E   RuntimeError: UnownedMemory requires explicit device ID for a null pointer.
 

Environment details (please complete the following information):

• Environment location: [Bare-metal, Docker, Cloud(specify cloud provider)]
• Linux Distro/Architecture: [Ubuntu 16.04 amd64]
• GPU Model/Driver: [V100 and driver 396.44]
• CUDA: [9.2]
• Method of cuDF & cuML install: [conda, Docker, or from source]
  • If method of install is [conda], run conda list and include results here
  • If method of install is [Docker], provide docker pull & docker run commands used
  • If method of install is [from source], provide versions of cmake & gcc/g++ and commit hash of build

Additional context
Related to #4095