Supporting order arg to asnumpy

[abagusetty]

Fixed dpnp.asnumpy and dpnp.ndarray.asnumpy ignoring the order arg
Fixes: #2568

• Have you provided a meaningful PR description?
• Have you added a test, reproducer or referred to an issue with a reproducer?
• Have you tested your changes locally for CPU and GPU devices?
• Have you made sure that new changes do not introduce compiler warnings?
• Have you checked performance impact of proposed changes?
• Have you added documentation for your changes, if necessary?
• Have you added your changes to the changelog?

[intel-python-devops]

Can one of the admins verify this patch?

[coveralls]

coverage: 78.071% (-0.002%) from 78.073% — abagusetty:fix-asnumpy-order-2884 into IntelPython:master

[antonwolfy]

missing test for order="A" and order=None

[antonwolfy]

Messing test with zero-sized ndarray

[abagusetty]

@antonwolfy Thanks for taking time. Addressed the above suggestions

[ndgrigorian]

@abagusetty
while this works in some cases, in other edge cases, this will not work, especially in terms of keeping "K" order:

In [1]: import numpy as np

In [2]: import dpnp as dnp

In [3]: x = np.arange(10)

In [4]: x_np = np.arange(10)

In [5]: x = dnp.arange(10)

In [6]: x
Out[6]: array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])

In [7]: x_np
Out[7]: array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])

In [8]: x = x[::-1]

In [9]: x_np = x_np[::-1]

In [10]: dnp.asnumpy(x)
Out[10]: array([9, 8, 7, 6, 5, 4, 3, 2, 1, 0])

In [11]: _.strides
Out[11]: (8,)

In [12]: np.asarray(x_np)
Out[12]: array([9, 8, 7, 6, 5, 4, 3, 2, 1, 0])

In [13]: _.strides
Out[13]: (-8,)

the issue is that there are no kernels for copying to strided host memory, only contiguous. So in such a case as this, we get a C-contiguous copy back. With cases like F-contiguous memory, order="K" will preserve it, because we can copy to F-contiguous host memory, as it is contiguous.

A proper solution requires new kernels. At minimum, the limitation should be documented.

[ndgrigorian]

part of the reason is that this will preserve the strides of the source result array, which are not necessarily correct per NumPy's order="K" semantics themselves

[ndgrigorian]

@abagusetty
it seems order="K" by default wasn't changed for dnp.asnumpy, so that observation was incorrect!

In [10]: dpt.asnumpy(x_dpt)
Out[10]: array([9, 8, 7, 6, 5, 4, 3, 2, 1, 0])

In [11]: _.strides
Out[11]: (-8,)

In [12]: dnp.asnumpy(x_dnp, order="K")
Out[12]: array([9, 8, 7, 6, 5, 4, 3, 2, 1, 0])

In [13]: _.strides
Out[13]: (-8,)

I will look for other edge cases, but if this works in this case, I don't see a likely issue otherwise