Cannot attach to the transpose of a np.ndarray

[manopapad]

Reported by @elliottslaughter

When executing this file:

import numpy as np
import cupynumeric as cn

x = np.ones((4,5))
y = x.transpose()
cn.asarray(y)

we get the following error:

Traceback (most recent call last):
  File "/home/mpapadakis/b/cupynumeric/a.py", line 6, in <module>
    cn.asarray(y)
  File "runtime.pyx", line 1188, in legate.core._lib.runtime.runtime.track_provenance.decorator.wrapper
  File "runtime.pyx", line 1189, in legate.core._lib.runtime.runtime.track_provenance.decorator.wrapper
  File "/home/mpapadakis/b/cupynumeric/cupynumeric/_utils/coverage.py", line 105, in wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/home/mpapadakis/b/cupynumeric/cupynumeric/_module/creation_data.py", line 141, in asarray
    thunk = runtime.get_numpy_thunk(a, share=True, dtype=dtype)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/mpapadakis/b/cupynumeric/cupynumeric/runtime.py", line 314, in get_numpy_thunk
    return self.find_or_create_array_thunk(obj, transfer)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/mpapadakis/b/cupynumeric/cupynumeric/runtime.py", line 416, in find_or_create_array_thunk
    raise NotImplementedError(
NotImplementedError: cuPyNumeric does not currently know how to attach to array views that are not affine transforms of their parent array.

Essentially cuPyNumeric is saying "I can't reverse-engineer how y is derived from x using an affine transformation", but a transpose is an affine transformation, so this is likely a bug in compute_parent_child_mapping.

At a high level, perhaps this check isn't necessary anymore. What cuPyNumeric is trying to avoid is this pattern:

x = np.ones(...)
a = cn.asarray(x[0:10])
b = cn.asarray(x[5:15])

where two different Stores are created by attaching to different pieces of the same ndarray. Instead of attaching to the sub-arrays, cuPyNumeric first attaches to the full x, then creates subregions for a and b.

@lightsighter do you remember why you needed to guard against this? @magnatelee do you expect any issue with current Legate, if you have two Stores created in this fashion?