Update dpnp.linalg.svd() to run on CUDA

[vlad-perevezentsev]

This PR suggests updating dpnp.linagl.svd() implementation to support running on CUDA devices.
Since cuSolver gesvd only supports m>=n (Remark 1) the previous implementation crashed with Segmentation fault (core dumped)

$ ONEAPI_DEVICE_SELECTOR=cuda:gpu pytest -v dpnp/tests/test_linalg.py::TestSvd::test_svd

Python error: Segmentation fault

This suggests adding checks for m>=n otherwise transpose the input array.

Passing the transposed array to oneapi::mkl::lapack::gesvd increases the performance of dpnp.linalg.svd() due to the reducing a matrix with m >= n to bidiagonal form (inside lapack::gesvd) is more efficient

# 2D array 

$ a_shape = (1024,2048)
$ na = generate_random_numpy_array(a_shape, dtype='f4', seed_value=81)

$ a_dp = dpnp.array(na, device='gpu')

# GPU 
$ %timeit res_dp = dpnp.linalg.svd(a_dp);q.wait()
# 1.07 s ± 11 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

$ %timeit res_dp = dpnp.linalg.svd(a_dp,new=True);q.wait()
# 881 ms ± 2.52 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

$ a_dp = dpnp.array(na, device='cpu')

# CPU 
$ %timeit res_dp = dpnp.linalg.svd(a_dp);q.wait()
# 1.1 s ± 3.63 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

$ %timeit res_dp = dpnp.linalg.svd(a_dp,new=True);q.wait()
# 897 ms ± 3.89 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

# 3D array

$ a_shape = (16,256,512)
$ na = generate_random_numpy_array(a_shape, dtype='f4', seed_value=81)


# GPU
$ %timeit res_dp = dpnp.linalg.svd(a_dp);q.wait()
# 979 ms ± 9.84 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

$ %timeit res_dp = dpnp.linalg.svd(a_dp,new=True);q.wait()
# 895 ms ± 6.79 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

# CPU

$ %timeit res_dp = dpnp.linalg.svd(a_dp);q.wait()
# 753 ms ± 39.7 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

$ %timeit res_dp = dpnp.linalg.svd(a_dp,new=True);q.wait()
# 679 ms ± 26.8 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

• 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/index.html

[antonwolfy]

Thank you @vlad-perevezentsev, LGTM