Coverage: 83.41% (+0.09%) from 83.322% when pulling d1d67b9 on boolean-reductions into 1320d39 on master.

Array API standard conformance tests for dpctl=0.14.3dev1=py310h76be34b_50 ran successfully.
Passed: 157
Failed: 843
Skipped: 116

View rendered docs @ https://intelpython.github.io/dpctl/pulls/1204/index.html

Array API standard conformance tests for dpctl=0.14.3dev1=py310h76be34b_50 ran successfully.
Passed: 157
Failed: 843
Skipped: 116

Array API standard conformance tests for dpctl=0.14.3dev1=py310h76be34b_52 ran successfully.
Passed: 157
Failed: 843
Skipped: 116

Array API standard conformance tests for dpctl=0.14.3dev1=py310h76be34b_55 ran successfully.
Passed: 157
Failed: 843
Skipped: 116

This PR requires docstrings

Array API standard conformance tests for dpctl=0.14.3dev1=py310h76be34b_56 ran successfully.
Passed: 157
Failed: 843
Skipped: 116

Array API standard conformance tests for dpctl=0.14.3dev1=py310h76be34b_57 ran successfully.
Passed: 157
Failed: 843
Skipped: 116

Array API standard conformance tests for dpctl=0.14.3dev1=py310h76be34b_62 ran successfully.
Passed: 157
Failed: 843
Skipped: 116

I think this PR is ready. The last change helped as witnessed by performance gap between axis=None and axis=(0,1,2,) below:

In [1]: import dpctl.tensor as dpt

In [2]: dpt.asarray([float('inf'),]*30 + [float('nan'),] * 12 + [-float('inf'), ] * 18, device="cpu").shape
Out[2]: (60,)

In [3]: x = dpt.reshape(dpt.asarray([float('inf'),]*30 + [float('nan'),] * 12 + [-float('inf'), ] * 18, device="cpu"), (5,4,3))

In [4]: x.flags
Out[4]:
  C_CONTIGUOUS : True
  F_CONTIGUOUS : False
  WRITABLE : True

In [5]: dpt.all(x)
Out[5]: usm_ndarray(True)

In [6]: %timeit dpt.all(x)
235 µs ± 24.3 µs per loop (mean ± std. dev. of 7 runs, 10,000 loops each)

In [7]: %timeit dpt.all(x, axis=(0,1,2))
246 µs ± 27.5 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

In [8]: %timeit dpt.all(x, axis=None)
254 µs ± 10.8 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

In [9]: %timeit dpt.all(x, axis=(0,1,2))
248 µs ± 31.5 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

In [10]: %timeit dpt.all(x, axis=None)
205 µs ± 11.8 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

In [11]: %timeit dpt.all(x, axis=(0,1,2))
251 µs ± 17 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

Deleted rendered PR docs from intelpython.github.com/dpctl, latest should be updated shortly. 🤞

Array API standard conformance tests for dpctl=0.14.3dev1=py310h76be34b_62 ran successfully.
Passed: 157
Failed: 843
Skipped: 116