dpctl.tensor.sum function raises an error: SyclAsynchronousError

[npolina4]

Device: opencl:cpu
OS: Linux
Occurrence frequency: periodically

import dpctl.tensor as dpt
x = dpt.arange(10, dtype=dpt.float32)
axes = [None, 0, (0,)]
for axis in axe:
    y = dpt.sum(a, axis=axis, dtype=dpt.complex128)

dpctl/_sycl_queue.pyx:126: in dpctl._sycl_queue.default_async_error_handler
    ???
E   dpctl._sycl_queue.SyclAsynchronousError: 1

The above exception was the direct cause of the following exception:
y = dpt.sum(
/usr/share/miniconda3/envs/test/lib/python3.9/site-packages/dpctl/tensor/_reduction.py:193: in sum
    return _reduction_over_axis(
/usr/share/miniconda3/envs/test/lib/python3.9/site-packages/dpctl/tensor/_reduction.py:139: in _reduction_over_axis
    dpctl.SyclEvent.wait_for(host_tasks_list)
E   SystemError: <built-in function wait_for> returned a result with an error set

[oleksandr-pavlyk]

See https://github.com/IntelPython/dpnp/actions/runs/6475840967/job/17584595766?pr=1585

[oleksandr-pavlyk]

Corrected example:

import dpctl.tensor as dpt

x = dpt.arange(10, dtype=dpt.float32, device='cpu')

for rep in range(16356):
    print(rep)
    axes = [None, 0, (0,)]
    for axis in axes:
        y = dpt.sum(x, axis=axis, dtype=dpt.complex128)

I could not reproduce the crash though.

[ndgrigorian]

I also could not reproduce the crash.

Similar crashes, however, do seem to be appearing in the CI, in various tests, and always where 'wait_for' is being used.
See:
https://github.com/IntelPython/dpctl/actions/runs/6458849200/job/17534736280
https://github.com/IntelPython/dpctl/actions/runs/6461852353/job/17543529221
https://github.com/IntelPython/dpctl/actions/runs/6461789466/job/17543428471
(last is a Windows case)

[oleksandr-pavlyk]

Use of DPCTLEvent_WaitAndThrow has been replaced with use of DPCTLEvent_Wait.

The SyclAsynchronousError error exception class has been removed.