Fix for dpnp buffer-arg ndarray ignoring offset

[abagusetty]

Fixes: #3067

• 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?

[intel-python-devops]

Can one of the admins verify this patch?

[coveralls]

coverage: 78.48% (+0.003%) from 78.477% — abagusetty:fix-dpnp-bufferarg-offset into IntelPython:master

[antonwolfy]

The dtype-defaulting could be hoisted out of the branch so it runs once for any buffer with a .dtype, instead of being duplicated between the inner if dtype is None and the elif. That also makes the ordering explicit: the offset math needs dtype for new_itemsize.

if isinstance(buffer, dpnp_array):
    buffer = buffer.get_array()

if dtype is None and hasattr(buffer, "dtype"):
    dtype = buffer.dtype

if isinstance(buffer, dpt.usm_ndarray):
    byte_offset = buffer._element_offset * buffer.itemsize
    new_itemsize = dpnp.dtype(dtype).itemsize
    add_offset, rem = divmod(byte_offset, new_itemsize)
    if rem != 0:
        raise ValueError(...)
    offset += add_offset

[abagusetty]

Done!

[antonwolfy]

Two small gaps worth filling while we're here (non-blocking):

1. No test exercises a non-zero incoming offset= arg together with a dtype-mismatched buffer — the offset=1 case uses a same-itemsize dtype, so the offset += add_offset addition is never checked with a non-trivial add_offset. A case like dpnp.ndarray((size,), dtype=new_dt, buffer=sl, offset=1) would cover it.

2. The misalignment ValueError is only asserted for a dpnp_array buffer (test_misaligned_offset_buffer_ctor_error). Since the fix routes bare usm_ndarray buffers through the same branch, a matching case with buffer=base[3:].get_array() would guard that path too.

[abagusetty]

Addressed both aspects

[antonwolfy]

This test hard-fails on devices without native fp64/complex128 support.

[abagusetty]

guarded it