[BUG] CumlArray error while handling empty cupy array

[tfeher]

Describe the bug
CumlArray has a problem while wrapping empty cupy array. Depending on where the array has 0 dim, we can run into the following exception:

Steps/Code to reproduce bug
Used 21.08 nightly to test (rapidsai/rapidsai-core-dev-nightly:21.08-cuda11.2-devel-ubuntu18.04-py3.8 on V100-SXM2-16GB, driver 460.32.03).

import cupy as cp
from cuml.common.array import CumlArray
dtype = cp.float32

# This works
x = cp.empty(shape=(0, 1), dtype=dtype, order='F')
x_cp = cp.asarray(x)

# This also works
x = CumlArray.empty(shape=(1, 0), dtype=dtype, order='F')
x_cp = cp.asarray(x) 

# This does not work
x = CumlArray.empty(shape=(0, 1), dtype=dtype, order='F')
x_cp = cp.asarray(x)

Output:

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
<ipython-input-3-71e2f5f68c60> in <module>
     13 # This does not work
     14 x = CumlArray.empty(shape=(0, 1), dtype=dtype, order='F')
---> 15 x_cp = cp.asarray(x)

/opt/conda/envs/rapids/lib/python3.8/site-packages/cupy/_creation/from_data.py in asarray(a, dtype, order)
     64 
     65     """
---> 66     return _core.array(a, dtype, False, order)
     67 
     68 

cupy/_core/core.pyx in cupy._core.core.array()

cupy/_core/core.pyx in cupy._core.core.array()

cupy/_core/core.pyx in cupy._core.core._convert_object_with_cuda_array_interface()

cupy/cuda/memory.pyx in cupy.cuda.memory.UnownedMemory.__init__()

RuntimeError: UnownedMemory requires explicit device ID for a null pointer.

Expected behavior
Return an empty cupy array in all cases, without error.

Additional context
This problem can be triggered if a model has parameter arrays with zero size (see #4033). (Admittedly it is a rare case.)

[github-actions]

This issue has been labeled inactive-90d due to no recent activity in the past 90 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed.

[github-actions]

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.