Use linspace() function from dpctl.tensor.

[npolina4]

Use linspace() function from dpctl.tensor module instead of DPNP backend implementation.

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

[oleksandr-pavlyk]

My experiment that ran into an error:

In [24]: x0, x1 = dpnp.ones((2,2)), dpnp.full((2,2,), 3)

In [25]: x0.device, x1.device
Out[25]: (Device(level_zero:gpu:0), Device(level_zero:gpu:0))

In [26]: x0.dtype, x1.dtype
Out[26]: (dtype('float32'), dtype('int64'))

In [27]: dpnp.linspace(x0, x1, num=5)   # raises ValueError: Device Intel(R) Graphics [0x9a49] does not provide native support for double-precision floating point type.

A variation dpnp.linspace(x0, dpnp.asarray(x1, 'f4'), num=5) works as expected.

I should also note that dpnp.linspace(x0[0,0], dpnp.asarray(x1, 'f4')[0,0], num=5) fails while attempting to fall back to NumPy.

When the array support works, it is rather slow. It would be nice to understand why.

In [54]: x2 = dpnp.asarray(x1, dtype=x0.dtype)

In [55]: %time dpnp.linspace(0.0, 3.0, num=1_000_000)
CPU times: user 9.2 ms, sys: 0 ns, total: 9.2 ms
Wall time: 9.95 ms
Out[55]: <dpnp.dpnp_array.dpnp_array at 0x7f4d75cd37f0>

In [56]: %time dpnp.linspace(x0, x2, num=1_000_000)
CPU times: user 133 ms, sys: 7.55 ms, total: 140 ms
Wall time: 187 ms
Out[56]: <dpnp.dpnp_array.dpnp_array at 0x7f4d6693f340>

[oleksandr-pavlyk]

x = dpnp.linspace(0, 3 + 2j, num=1000) throws TypeError: can't convert complex to float.

x = dpt.linspace(0, 3 + 2j, num=1000) works as expected.

[antonwolfy]

In case start and stop are both dpnp_array or usm_ndarray, we might change usm_type type here for resulting _start and _stop and so zero copy will not be applied (i.e. extra memory copy might take place).

[antonwolfy]

Thank you! LGMT!

[oleksandr-pavlyk]

Thank you @npolina4 !

[antonwolfy]

The PR is going to resolve issue #1249 .