[antonwolfy]

For inputs where a has 2 or more dimensions and b has more than 2 dimensions, dpnp.dot and dpnp.ndarray.dot place the leading dimensions of b in the middle of the result, unlike dpnp.matmul. NumPy 2.6 deprecates this case, and this PR mirrors that deprecation.

a = dpnp.ones((2, 3))
b = dpnp.ones((4, 3, 5))

dpnp.dot(a, b).shape                        # (2, 4, 5), now emits DeprecationWarning
dpnp.tensordot(a, b, axes=(-1, -2)).shape   # (2, 4, 5), same result
dpnp.matmul(a, b).shape                     # (4, 2, 5), batched matrix product

The result is unchanged, but the call now emits:

dpnp.dot received arrays with a.ndim >= 2 and b.ndim > 2. This places the leading dimensions of b in the middle of the result, which is deprecated and will eventually raise an error. Use dpnp.tensordot(a, b, axes=(-1, -2)) instead.

Only that case warns. 0-D, 1-D and 2-D inputs, a 1-D a with an N-D b, and an N-D a with a 1-D or 2-D b behave as before. As in NumPy, the warning comes after shape validation, so a shape mismatch raises ValueError without a warning.

The implementation moves into a private _dot, which both dpnp.dot and dpnp.ndarray.dot call directly. This way the warning points to the user's line for the method as well; otherwise it would be attributed to dpnp_array.py, and Python's default warning filters would hide it.

The dpnp.dot docstring gets a "Warnings" section, and the 4-D example is removed.

• 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?
• Have you checked performance impact of proposed changes?
• Have you added documentation for your changes, if necessary?
• Have you added your changes to the changelog?

[github-actions]

View rendered docs @ https://intelpython.github.io/dpnp/pull/3124/index.html

[coveralls]

coverage: 82.643% (+0.02%) from 82.627% — deprecate-dot-nd-interleave into master