dpt.sum over axis for transposed input array

[antonwolfy]

Due to merging gh-1364 the below example has been broken:

import numpy, dpctl, dpctl.tensor as dpt
from math import prod

sh = (1, 2, 3)
a = dpt.arange(prod(sh), dtype='i4')
a = dpt.reshape(a, sh)
aT = dpt.permute_dims(a, (2, 1, 0))

aT.shape
# Out: (3, 2, 1)

dpt.sum(aT, axis=2)
# Out:
# usm_ndarray([[0, 1],
# [2, 3],
# [4, 5]])

b = numpy.arange(prod(sh), dtype='i4').reshape(sh)
bT = b.T

bT.shape
# Out: (3, 2, 1)

bT.sum(axis=2)
# Out:
# array([[0, 3],
# [1, 4],
# [2, 5]])

[antonwolfy]

Similar issue with another shape:

sh = (3, 3, 3)
a = dpt.arange(prod(sh), dtype='i4')
a = dpt.reshape(a, sh)
aT = dpt.permute_dims(a, (2, 1, 0))

aT.strides
# Out: (1, 3, 9)

dpt.sum(aT, axis=2)
# Out:
# usm_ndarray([[27, 30, 33],
# [36, 39, 42],
# [45, 48, 51]])

b = numpy.arange(prod(sh), dtype='i4').reshape(sh)
bT = b.T

bT.strides
# Out: (4, 12, 36)

bT.sum(axis=2)
# Out:
# array([[27, 36, 45],
# [30, 39, 48],
# [33, 42, 51]])