dpnp.erf function does not support compute follows data

[diptorupd]

The dpnp black_scholes implementation throws an exception and fails to execute:

Traceback (most recent call last):
  File "/dpbench/dpbench/infrastructure/benchmark.py", line 379, in _exec
    warmup(self.impl_fn, inputs)
  File "/dpbench/dpbench/infrastructure/timeout_decorator.py", line 65, in inner
    result = fn(*args, **kwargs)
  File "/dpbench/dpbench/infrastructure/benchmark.py", line 375, in warmup
    self.fmwrk.execute(impl_fn, inputs)
  File "/dpbench/dpbench/infrastructure/framework.py", line 123, in execute
    return impl_fn(**input_args)
  File "/dpbench/dpbench/benchmarks/black_scholes/black_scholes_dpnp.py", line 28, in black_scholes
    d1 = 0.5 + 0.5 * np.erf(w1)
  File "/.conda/envs/dpbench-devel/lib/python3.9/site-packages/dpnp/dpnp_iface_libmath.py", line 80, in erf
    x1_desc = dpnp.get_dpnp_descriptor(in_array1, copy_when_strides=False)
  File "/.conda/envs/dpbench-devel/lib/python3.9/site-packages/dpnp/dpnp_iface.py", line 234, in get_dpnp_descriptor
    ext_obj = array(arr_obj, sycl_queue=default_queue)
  File "/.conda/envs/dpbench-devel/lib/python3.9/site-packages/dpnp/dpnp_iface_arraycreation.py", line 217, in array
    return dpnp_container.asarray(x1,
  File "/.conda/envs/dpbench-devel/lib/python3.9/site-packages/dpnp/dpnp_container.py", line 81, in asarray
    array_obj = dpt.asarray(x1_obj,
  File "/.conda/envs/dpbench-devel/lib/python3.9/site-packages/dpctl/tensor/_ctors.py", line 328, in asarray
    return _asarray_from_usm_ndarray(
  File "/.conda/envs/dpbench-devel/lib/python3.9/site-packages/dpctl/tensor/_ctors.py", line 188, in _asarray_from_usm_ndarray
    res[(slice(None, None, None),) * res.ndim] = usm_ndary
  File "dpctl/tensor/_usmarray.pyx", line 921, in dpctl.tensor._usmarray.usm_ndarray.__setitem__
  File "/.conda/envs/dpbench-devel/lib/python3.9/site-packages/dpctl/tensor/_copy_utils.py", line 199, in _copy_from_usm_ndarray_to_usm_ndarray
    _copy_same_shape(dst, src)
  File "/.conda/envs/dpbench-devel/lib/python3.9/site-packages/dpctl/tensor/_copy_utils.py", line 169, in _copy_same_shape
    hev, ev = ti._copy_usm_ndarray_into_usm_ndarray(
ValueError: Execution queue context is not the same as allocation contexts

On investigating, I found that the dpnp implementation of erf does not yet support the compute follows data programming mode. Here is a reproducer:

>>> import dpnp
>>> 
>>> 
>>> 
>>> a = dpnp.ones(100)
>>> b = dpnp.erf(a)
>>> a.sycl_queue
<dpctl.SyclQueue at 0x7f537d3fbc40>
>>> b.sycl_queue
<dpctl.SyclQueue at 0x7f53b21f0f00>
>>> a.sycl_queue == b.sycl_queue
False

It is expected that the output of erf should be on the same queue as the input array. But, that is not the case.

[antonwolfy]

@diptorupd, I've created PR #1194 to fix the issue in dpnp.

[antonwolfy]

The issue has been fixed and available in dppy/label/dev channel.