DPNP vs Numpy strides mismatch

[abagusetty]

This one bite us dearly on PVCs, when porting an application written in cupy. Also found a similar issue #1649

Looks like dpnp considers elements vs bytes (from numpy). Hoping to get some traction on this one

import dpnp as dp
import numpy as np

shape = (4, 24, 4096)
a_dp, a_np = dp.empty(shape, order='C'), np.empty(shape, order='C')

print("dpnp strides:", a_dp.strides)
print("numpy strides:", a_np.strides)
print("Match:", a_dp.strides == a_np.strides)

[abagusetty]

@antonwolfy gentle ping: dpnp strides are in elements vs NumPy/CuPy bytes
This is breaking an app porting cupy to dpnp port on Aurora/PVC. Any chance to prioritize a compat fix

[antonwolfy]

@abagusetty, sure, I'll look on that.

In general, that implementation was intentional from the beginning and come from dpctl implementation.
But we can change that to improve usability and porting.