Merge pull request #2339 from IntelPython/reuse-nan-to-num-nan-fns

Reuse `dpnp.nan_to_num` in `dpnp.nansum` and `dpnp.nanprod`

Author: ndgrigorian

dpnp/dpnp_iface_nanfunctions.py

@@ -60,6 +60,36 @@
 ]
 
 
+def _replace_nan_no_mask(a, val):
+    """
+    Replace NaNs in array `a` with `val`.
+
+    If `a` is of inexact type, make a copy of `a`, replace NaNs with
+    the `val` value, and return the copy. If `a` is not of inexact type,
+    do nothing and return `a`.
+
+    Parameters
+    ----------
+    a : {dpnp.ndarray, usm_ndarray}
+        Input array.
+    val : float
+        NaN values are set to `val` before doing the operation.
+
+    Returns
+    -------
+    out : dpnp.ndarray
+        If `a` is of inexact type, return a copy of `a` with the NaNs
+        replaced by the fill value, otherwise return `a`.
+
+    """
+
+    dpnp.check_supported_arrays_type(a)
+    if dpnp.issubdtype(a.dtype, dpnp.inexact):
+        return dpnp.nan_to_num(a, nan=val, posinf=dpnp.inf, neginf=-dpnp.inf)
+
+    return a
+
+
 def _replace_nan(a, val):
     """
     Replace NaNs in array `a` with `val`.
@@ -107,6 +137,18 @@ def nanargmax(a, axis=None, out=None, *, keepdims=False):
 
     For full documentation refer to :obj:`numpy.nanargmax`.
 
+    Warning
+    -------
+    This function synchronizes in order to test for all-NaN slices in the array.
+    This may harm performance in some applications. To avoid synchronization,
+    the user is recommended to filter NaNs themselves and use `dpnp.argmax`
+    on the filtered array.
+
+    Warning
+    -------
+    The results cannot be trusted if a slice contains only NaNs
+    and -Infs.
+
     Parameters
     ----------
     a : {dpnp.ndarray, usm_ndarray}
@@ -136,8 +178,6 @@ def nanargmax(a, axis=None, out=None, *, keepdims=False):
         values ignoring NaNs. The returned array must have the default array
         index data type.
         For all-NaN slices ``ValueError`` is raised.
-        Warning: the results cannot be trusted if a slice contains only NaNs
-        and -Infs.
 
     Limitations
     -----------
@@ -181,6 +221,18 @@ def nanargmin(a, axis=None, out=None, *, keepdims=False):
 
     For full documentation refer to :obj:`numpy.nanargmin`.
 
+    Warning
+    -------
+    This function synchronizes in order to test for all-NaN slices in the array.
+    This may harm performance in some applications. To avoid synchronization,
+    the user is recommended to filter NaNs themselves and use `dpnp.argmax`
+    on the filtered array.
+
+    Warning
+    -------
+    The results cannot be trusted if a slice contains only NaNs
+    and -Infs.
+
     Parameters
     ----------
     a : {dpnp.ndarray, usm_ndarray}
@@ -210,8 +262,6 @@ def nanargmin(a, axis=None, out=None, *, keepdims=False):
         values ignoring NaNs. The returned array must have the default array
         index data type.
         For all-NaN slices ``ValueError`` is raised.
-        Warning: the results cannot be trusted if a slice contains only NaNs
-        and Infs.
 
     Limitations
     -----------
@@ -315,7 +365,7 @@ def nancumprod(a, axis=None, dtype=None, out=None):
 
     """
 
-    a, _ = _replace_nan(a, 1)
+    a = _replace_nan_no_mask(a, 1.0)
     return dpnp.cumprod(a, axis=axis, dtype=dtype, out=out)
 
 
@@ -385,7 +435,7 @@ def nancumsum(a, axis=None, dtype=None, out=None):
 
     """
 
-    a, _ = _replace_nan(a, 0)
+    a = _replace_nan_no_mask(a, 0.0)
     return dpnp.cumsum(a, axis=axis, dtype=dtype, out=out)
 
 
@@ -884,7 +934,7 @@ def nanprod(
 
     """
 
-    a, _ = _replace_nan(a, 1)
+    a = _replace_nan_no_mask(a, 1.0)
     return dpnp.prod(
         a,
         axis=axis,
@@ -988,7 +1038,7 @@ def nansum(
 
     """
 
-    a, _ = _replace_nan(a, 0)
+    a = _replace_nan_no_mask(a, 0.0)
     return dpnp.sum(
         a,
         axis=axis,