Improve built-in ops and dpnp universal functions testing

[ZzEeKkAa]

Add full coverage on ufuncs and built-in operators. Motivated by #178

It appears that we were not aware that these functions are not supported:

• "cbrt"

We also does not support (may be we need proper input data):

• matmul operator (a @ b)
• imatmul operator (a @= b)
• not_ operator (not a) (because numpy does not support it)

Open question:

• Does it mean if the operator works in parfor, than it will work in kernel?

Checklist:

• 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?
• If this PR is a work in progress, are you filing the PR as a draft?

[diptorupd]

@ZzEeKkAa Thank you for the comprehensive evaluation. I am replying to some of your questions and comments in the PR description.

1. It appears that we were not aware that these functions are not supported:
   "cbrt"
   "mod"

Please log separate issues to track the gap.

2. We also does not support (may be we need proper input data):

   matmul operator (a @ b)
   imatmul operator (a @= b)
   not_ operator (not a)

The matmul operators not being supported is expected. I believe the matmul operators are not treated as parfor and Numba lowers them as library calls. Numba-dpex used to have the same functionality at one point.

The not_ not being supported needs investigation. Please open an issue so that we can keep track of it.

3. Does it mean if the operator works in parfor, than it will work in kernel?

That is a correct assumption. However, we should note a subtle difference. There are various types of expressions that can be lowered to parfor nodes: numpy calls on arrays and built-in operations on arrays (both of which are vector operations), numpy calls and built in operations on scalars inside a prange loop (the operation here is a scalar op). In a kernel, only the scalar versions of all these operators and ufuncs can be used.

It is a good idea to add similar unit tests for the kernel decorator as well. We will be explicitly verifying that lowering a scalar operation works.

[ZzEeKkAa]

not is not supported by numpy, I've managed to make mod work and opened issue for cbrt.

Let's test kernel in separate PR.

[diptorupd]

nice changes. Making unit testing more comprehensive is always welcome!