Reuse dpctl.tensor.isnan(), dpctl.tensor.isinf(), and dpctl.tensor.isfinite() functions.

dpnp/dpnp_algo/dpnp_algo_logic.pxi

@@ -40,9 +40,6 @@ __all__ += [
     "dpnp_allclose",
     "dpnp_any",
     "dpnp_isclose",
-    "dpnp_isfinite",
-    "dpnp_isinf",
-    "dpnp_isnan",
 ]
 
 
@@ -174,48 +171,3 @@ cpdef utils.dpnp_descriptor dpnp_isclose(utils.dpnp_descriptor input1,
         result.get_pyobj()[i] = numpy.isclose(input1.get_pyobj()[i], input2.get_pyobj()[i], rtol, atol, equal_nan)
 
     return result
-
-
-cpdef utils.dpnp_descriptor dpnp_isfinite(utils.dpnp_descriptor input1):
-    input1_obj = input1.get_array()
-    cdef utils.dpnp_descriptor result = utils_py.create_output_descriptor_py(input1.shape,
-                                                                             dpnp.bool,
-                                                                             None,
-                                                                             device=input1_obj.sycl_device,
-                                                                             usm_type=input1_obj.usm_type,
-                                                                             sycl_queue=input1_obj.sycl_queue)
-
-    for i in range(result.size):
-        result.get_pyobj()[i] = numpy.isfinite(input1.get_pyobj()[i])
-
-    return result
-
-
-cpdef utils.dpnp_descriptor dpnp_isinf(utils.dpnp_descriptor input1):
-    input1_obj = input1.get_array()
-    cdef utils.dpnp_descriptor result = utils_py.create_output_descriptor_py(input1.shape,
-                                                                             dpnp.bool,
-                                                                             None,
-                                                                             device=input1_obj.sycl_device,
-                                                                             usm_type=input1_obj.usm_type,
-                                                                             sycl_queue=input1_obj.sycl_queue)
-
-    for i in range(result.size):
-        result.get_pyobj()[i] = numpy.isinf(input1.get_pyobj()[i])
-
-    return result
-
-
-cpdef utils.dpnp_descriptor dpnp_isnan(utils.dpnp_descriptor input1):
-    input1_obj = input1.get_array()
-    cdef utils.dpnp_descriptor result = utils_py.create_output_descriptor_py(input1.shape,
-                                                                             dpnp.bool,
-                                                                             None,
-                                                                             device=input1_obj.sycl_device,
-                                                                             usm_type=input1_obj.usm_type,
-                                                                             sycl_queue=input1_obj.sycl_queue)
-
-    for i in range(result.size):
-        result.get_pyobj()[i] = numpy.isnan(input1.get_pyobj()[i])
-
-    return result

dpnp/dpnp_algo/dpnp_elementwise_common.py

@@ -49,6 +49,9 @@
     "dpnp_floor_divide",
     "dpnp_greater",
     "dpnp_greater_equal",
+    "dpnp_isfinite",
+    "dpnp_isinf",
+    "dpnp_isnan",
     "dpnp_less",
     "dpnp_less_equal",
     "dpnp_log",
@@ -466,6 +469,112 @@ def dpnp_greater_equal(x1, x2, out=None, order="K"):
     return dpnp_array._create_from_usm_ndarray(res_usm)
 
 
+_isfinite_docstring = """
+isfinite(x, out=None, order="K")
+
+Checks if each element of input array is a finite number.
+
+Args:
+    x (dpnp.ndarray):
+        Input array, expected to have numeric data type.
+    out ({None, dpnp.ndarray}, optional):
+        Output array to populate.
+        Array have the correct shape and the expected data type.
+    order ("C","F","A","K", optional):
+        Memory layout of the newly output array, if parameter `out` is `None`.
+        Default: "K".
+Returns:
+    dpnp.ndarray:
+        An array which is True where `x` is not positive infinity,
+        negative infinity, or NaN, False otherwise.
+        The data type of the returned array is `bool`.
+"""
+
+
+def dpnp_isfinite(x, out=None, order="K"):
+    """Invokes isfinite() from dpctl.tensor implementation for isfinite() function."""
+
+    # dpctl.tensor only works with usm_ndarray
+    x1_usm = dpnp.get_usm_ndarray(x)
+    out_usm = None if out is None else dpnp.get_usm_ndarray(out)
+
+    func = UnaryElementwiseFunc(
+        "isfinite", ti._isfinite_result_type, ti._isfinite, _isfinite_docstring
+    )
+    res_usm = func(x1_usm, out=out_usm, order=order)
+    return dpnp_array._create_from_usm_ndarray(res_usm)
+
+
+_isinf_docstring = """
+isinf(x, out=None, order="K")
+
+Checks if each element of input array is an infinity.
+
+Args:
+    x (dpnp.ndarray):
+        Input array, expected to have numeric data type.
+    out ({None, dpnp.ndarray}, optional):
+        Output array to populate.
+        Array have the correct shape and the expected data type.
+    order ("C","F","A","K", optional):
+        Memory layout of the newly output array, if parameter `out` is `None`.
+        Default: "K".
+Returns:
+    dpnp.ndarray:
+        An array which is True where `x` is positive or negative infinity,
+        False otherwise. The data type of the returned array is `bool`.
+"""
+
+
+def dpnp_isinf(x, out=None, order="K"):
+    """Invokes isinf() from dpctl.tensor implementation for isinf() function."""
+
+    # dpctl.tensor only works with usm_ndarray
+    x1_usm = dpnp.get_usm_ndarray(x)
+    out_usm = None if out is None else dpnp.get_usm_ndarray(out)
+
+    func = UnaryElementwiseFunc(
+        "isinf", ti._isinf_result_type, ti._isinf, _isinf_docstring
+    )
+    res_usm = func(x1_usm, out=out_usm, order=order)
+    return dpnp_array._create_from_usm_ndarray(res_usm)
+
+
+_isnan_docstring = """
+isnan(x, out=None, order="K")
+
+Checks if each element of an input array is a NaN.
+
+Args:
+    x (dpnp.ndarray):
+        Input array, expected to have numeric data type.
+    out ({None, dpnp.ndarray}, optional):
+        Output array to populate.
+        Array have the correct shape and the expected data type.
+    order ("C","F","A","K", optional):
+        Memory layout of the newly output array, if parameter `out` is `None`.
+        Default: "K".
+Returns:
+    dpnp.ndarray:
+        An array which is True where x is NaN, False otherwise.
+        The data type of the returned array is `bool`.
+"""
+
+
+def dpnp_isnan(x, out=None, order="K"):
+    """Invokes isnan() from dpctl.tensor implementation for isnan() function."""
+
+    # dpctl.tensor only works with usm_ndarray
+    x1_usm = dpnp.get_usm_ndarray(x)
+    out_usm = None if out is None else dpnp.get_usm_ndarray(out)
+
+    func = UnaryElementwiseFunc(
+        "isnan", ti._isnan_result_type, ti._isnan, _isnan_docstring
+    )
+    res_usm = func(x1_usm, out=out_usm, order=order)
+    return dpnp_array._create_from_usm_ndarray(res_usm)
+
+
 _less_docstring_ = """
 less(x1, x2, out=None, order="K")
 

dpnp/dpnp_iface_logic.py

@@ -51,6 +51,9 @@
     dpnp_equal,
     dpnp_greater,
     dpnp_greater_equal,
+    dpnp_isfinite,
+    dpnp_isinf,
+    dpnp_isnan,
     dpnp_less,
     dpnp_less_equal,
     dpnp_logical_and,
@@ -492,19 +495,20 @@ def isclose(x1, x2, rtol=1e-05, atol=1e-08, equal_nan=False):
     )
 
 
-def isfinite(x1, out=None, **kwargs):
+def isfinite(
+    x, /, out=None, *, where=True, order="K", dtype=None, subok=True, **kwargs
+):
     """
     Test element-wise for finiteness (not infinity or not Not a Number).
 
     For full documentation refer to :obj:`numpy.isfinite`.
 
     Limitations
     -----------
-    Input array is supported as :obj:`dpnp.ndarray`.
+    Parameters `x` is only supported as either :class:`dpnp.ndarray` or :class:`dpctl.tensor.usm_ndarray`.
+    Parameters `where`, `dtype` and `subok` are supported with their default values.
     Otherwise the function will be executed sequentially on CPU.
     Input array data types are limited by supported DPNP :ref:`Data types`.
-    Parameter ``out`` is supported only with default value ``None``.
-    Parameter ``where`` is supported only with default value ``True``.
 
     See Also
     --------
@@ -525,35 +529,38 @@ def isfinite(x1, out=None, **kwargs):
     --------
     >>> import dpnp as np
     >>> x = np.array([-np.inf, 0., np.inf])
-    >>> out = np.isfinite(x)
-    >>> [i for i in out]
-    [False, True, False]
+    >>> np.isfinite(x)
+    array([False,  True, False])
 
     """
 
-    # x1_desc = dpnp.get_dpnp_descriptor(x1)
-    # if x1_desc and kwargs:
-    #     if out is not None:
-    #         pass
-    #     else:
-    #         return dpnp_isfinite(x1_desc).get_pyobj()
-
-    return call_origin(numpy.isfinite, x1, out, **kwargs)
+    return check_nd_call_func(
+        numpy.isfinite,
+        dpnp_isfinite,
+        x,
+        out=out,
+        where=where,
+        order=order,
+        dtype=dtype,
+        subok=subok,
+        **kwargs,
+    )
 
 
-def isinf(x1, out=None, **kwargs):
+def isinf(
+    x, /, out=None, *, where=True, order="K", dtype=None, subok=True, **kwargs
+):
     """
     Test element-wise for positive or negative infinity.
 
     For full documentation refer to :obj:`numpy.isinf`.
 
     Limitations
     -----------
-    Input array is supported as :obj:`dpnp.ndarray`.
+    Parameters `x` is only supported as either :class:`dpnp.ndarray` or :class:`dpctl.tensor.usm_ndarray`.
+    Parameters `where`, `dtype` and `subok` are supported with their default values.
     Otherwise the function will be executed sequentially on CPU.
     Input array data types are limited by supported DPNP :ref:`Data types`.
-    Parameter ``out`` is supported only with default value ``None``.
-    Parameter ``where`` is supported only with default value ``True``.
 
     See Also
     --------
@@ -569,35 +576,38 @@ def isinf(x1, out=None, **kwargs):
     --------
     >>> import dpnp as np
     >>> x = np.array([-np.inf, 0., np.inf])
-    >>> out = np.isinf(x)
-    >>> [i for i in out]
-    [True, False, True]
+    >>> np.isinf(x)
+    array([ True, False,  True])
 
     """
 
-    # x1_desc = dpnp.get_dpnp_descriptor(x1)
-    # if x1_desc and kwargs:
-    #     if out is not None:
-    #         pass
-    #     else:
-    #         return dpnp_isinf(x1_desc).get_pyobj()
-
-    return call_origin(numpy.isinf, x1, out, **kwargs)
+    return check_nd_call_func(
+        numpy.isinf,
+        dpnp_isinf,
+        x,
+        out=out,
+        where=where,
+        order=order,
+        dtype=dtype,
+        subok=subok,
+        **kwargs,
+    )
 
 
-def isnan(x1, out=None, **kwargs):
+def isnan(
+    x, /, out=None, *, where=True, order="K", dtype=None, subok=True, **kwargs
+):
     """
     Test element-wise for NaN and return result as a boolean array.
 
     For full documentation refer to :obj:`numpy.isnan`.
 
     Limitations
     -----------
-    Input array is supported as :obj:`dpnp.ndarray`.
+    Parameters `x` is only supported as either :class:`dpnp.ndarray` or :class:`dpctl.tensor.usm_ndarray`.
+    Parameters `where`, `dtype` and `subok` are supported with their default values.
     Otherwise the function will be executed sequentially on CPU.
     Input array data types are limited by supported DPNP :ref:`Data types`.
-    Parameter ``out`` is supported only with default value ``None``.
-    Parameter ``where`` is supported only with default value ``True``.
 
     See Also
     --------
@@ -614,20 +624,22 @@ def isnan(x1, out=None, **kwargs):
     --------
     >>> import dpnp as np
     >>> x = np.array([np.inf, 0., np.nan])
-    >>> out = np.isnan(x)
-    >>> [i for i in out]
-    [False, False, True]
+    >>> np.isnan(x)
+    array([False, False,  True])
 
     """
 
-    # x1_desc = dpnp.get_dpnp_descriptor(x1)
-    # if x1_desc and kwargs:
-    #     if out is not None:
-    #         pass
-    #     else:
-    #         return dpnp_isnan(x1_desc).get_pyobj()
-
-    return call_origin(numpy.isnan, x1, out, **kwargs)
+    return check_nd_call_func(
+        numpy.isnan,
+        dpnp_isnan,
+        x,
+        out=out,
+        where=where,
+        order=order,
+        dtype=dtype,
+        subok=subok,
+        **kwargs,
+    )
 
 
 def less(

tests/test_logic.py

@@ -4,7 +4,11 @@
 
 import dpnp
 
-from .helper import get_all_dtypes, has_support_aspect64
+from .helper import (
+    get_all_dtypes,
+    get_float_complex_dtypes,
+    has_support_aspect64,
+)
 
 
 @pytest.mark.parametrize("type", get_all_dtypes())
@@ -297,3 +301,25 @@ def test_comparison_no_broadcast_with_shapes(op, sh1, sh2):
     with pytest.raises(ValueError):
         getattr(dpnp, op)(x1, x2)
         getattr(numpy, op)(x1.asnumpy(), x2.asnumpy())
+
+
+@pytest.mark.parametrize(
+    "op", ["isfinite", "isinf", "isnan"], ids=["isfinite", "isinf", "isnan"]
+)
+@pytest.mark.parametrize(
+    "data",
+    [
+        [dpnp.inf, -1, 0, 1, dpnp.nan],
+        [[dpnp.inf, dpnp.nan], [dpnp.nan, 0], [1, dpnp.inf]],
+    ],
+    ids=[
+        "[dpnp.inf, -1, 0, 1, dpnp.nan]",
+        "[[dpnp.inf, dpnp.nan], [dpnp.nan, 0], [1, dpnp.inf]]",
+    ],
+)
+@pytest.mark.parametrize("dtype", get_float_complex_dtypes())
+def test_finite(op, data, dtype):
+    x = dpnp.asarray(data, dtype=dtype)
+    np_res = getattr(dpnp, op)(x)
+    dpnp_res = getattr(numpy, op)(x.asnumpy())
+    assert_equal(dpnp_res, np_res)