Update dpnp.diagonal()

[vlad-perevezentsev]

This PR suggests an updated implementation of dpnp.diagonal() including support for axis1, axis2 and negative offset parameters.

The new implementation shows performance improvements over the old implementation:

In [1]: import dpnp
 
In [2]: import numpy
 
In [3]: na = numpy.random.randint(-10**4, 10**4, size=(4096,4096))
 
In [4]: a = numpy.array(na,dtype='int32')
 
In [5]: a_dp = dpnp.array(a)
 
In [6]: res_dp = dpnp.diagonal(a_dp)
 
In [7]: %timeit -n 10 res_dp = dpnp.diagonal(a_dp)
16.9 ms ± 1.59 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
 
In [8]: %timeit -n 10 res_dp = dpnp.diagonal(a_dp,new=True)
17.8 µs ± 6.47 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)
 
In [9]: a = numpy.array(na,dtype='float32')
 
In [10]: a_dp = dpnp.array(a)
 
In [11]: res_dp = dpnp.diagonal(a_dp)
 
In [12]: %timeit -n 10 res_dp = dpnp.diagonal(a_dp)
18 ms ± 1.21 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
 
In [13]: %timeit -n 10 res_dp = dpnp.diagonal(a_dp,new=True)
17.2 µs ± 4.5 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

The backend implementation (dpnp_diagonal) is not removed due to the fact that dpnp_trace uses it

Additionally updating dpnp.diagonal() will allow to update dpnp.diag() function via reusing dpnp.diagonal()

• Have you provided a meaningful PR description?
• Have you added a test, reproducer or referred to 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?
• If this PR is a work in progress, are you filing the PR as a draft?

[github-actions]

View rendered docs @ https://intelpython.github.io/dpnp/pull//index.html

[antonwolfy]

Thank you @vlad-perevezentsev