Extend dpnp.pad to support pad_width as a dictionary

CHANGELOG.md

@@ -39,6 +39,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 * Added missing includes to files in ufunc and VM pybind11 extensions [#2571](https://github.com/IntelPython/dpnp/pull/2571)
 * Refactored backend implementation of `dpnp.linalg.solve` to use oneMKL LAPACK `gesv` directly [#2558](https://github.com/IntelPython/dpnp/pull/2558)
 * Improved performance of `dpnp.isclose` function by implementing a dedicated kernel for scalar `rtol` and `atol` arguments [#2540](https://github.com/IntelPython/dpnp/pull/2540)
+* Extended `dpnp.pad` to support `pad_width` keyword as a dictionary [#2535](https://github.com/IntelPython/dpnp/pull/2535)
 
 ### Deprecated
 

dpnp/dpnp_iface_manipulation.py

@@ -2508,14 +2508,17 @@ def pad(array, pad_width, mode="constant", **kwargs):
     ----------
     array : {dpnp.ndarray, usm_ndarray}
         The array of rank ``N`` to pad.
-    pad_width : {sequence, array_like, int}
+    pad_width : {sequence, array_like, int, dict}
         Number of values padded to the edges of each axis.
         ``((before_1, after_1), ... (before_N, after_N))`` unique pad widths
         for each axis.
         ``(before, after)`` or ``((before, after),)`` yields same before
         and after pad for each axis.
         ``(pad,)`` or ``int`` is a shortcut for ``before = after = pad`` width
         for all axes.
+        If a dictionary, each key is an axis and its corresponding value is an
+        integer or a pair of integers describing the padding ``(before, after)``
+        or ``pad`` width for that axis.
     mode : {str, function}, optional
         One of the following string values or a user supplied function.
 
@@ -2698,6 +2701,26 @@ def pad(array, pad_width, mode="constant", **kwargs):
            [100, 100, 100, 100, 100, 100, 100],
            [100, 100, 100, 100, 100, 100, 100]])
 
+    >>> a = np.arange(1, 7).reshape(2, 3)
+    >>> np.pad(a, {1: (1, 2)})
+    array([[0, 1, 2, 3, 0, 0],
+           [0, 4, 5, 6, 0, 0]])
+    >>> np.pad(a, {-1: 2})
+    array([[0, 0, 1, 2, 3, 0, 0],
+           [0, 0, 4, 5, 6, 0, 0]])
+    >>> np.pad(a, {0: (3, 0)})
+    array([[0, 0, 0],
+           [0, 0, 0],
+           [0, 0, 0],
+           [1, 2, 3],
+           [4, 5, 6]])
+    >>> np.pad(a, {0: (3, 0), 1: 2})
+    array([[0, 0, 0, 0, 0, 0, 0],
+           [0, 0, 0, 0, 0, 0, 0],
+           [0, 0, 0, 0, 0, 0, 0],
+           [0, 0, 1, 2, 3, 0, 0],
+           [0, 0, 4, 5, 6, 0, 0]])
+
     """
 
     dpnp.check_supported_arrays_type(array)

dpnp/dpnp_utils/dpnp_utils_pad.py

@@ -273,6 +273,47 @@ def _get_stats(padded, axis, width_pair, length_pair, stat_func):
     return left_stat, right_stat
 
 
+def _pad_normalize_dict_width(pad_width, ndim):
+    """
+    Normalize pad width passed as a dictionary.
+
+    Parameters
+    ----------
+    pad_width : dict
+        Padding specification. The keys must be integer axis indices, and
+        the values must be either:
+          - a single int (same padding before and after),
+          - a tuple of two ints (before, after).
+    ndim : int
+        Number of dimensions in the input array.
+
+    Returns
+    -------
+    seq : list
+        A (ndim, 2) list of padding widths for each axis.
+
+    Raises
+    ------
+    TypeError
+        If the padding format for any axis is invalid.
+
+    """
+
+    seq = [(0, 0)] * ndim
+    for axis, width in pad_width.items():
+        if isinstance(width, int):
+            seq[axis] = (width, width)
+        elif (
+            isinstance(width, tuple)
+            and len(width) == 2
+            and all(isinstance(w, int) for w in width)
+        ):
+            seq[axis] = width
+        else:
+            raise TypeError(f"Invalid pad width for axis {axis}: {width}")
+    return seq
+
+
 def _pad_simple(array, pad_width, fill_value=None):
     """
     Copied from numpy/lib/_arraypad_impl.py
@@ -616,21 +657,25 @@ def _view_roi(array, original_area_slice, axis):
 def dpnp_pad(array, pad_width, mode="constant", **kwargs):
     """Pad an array."""
 
+    nd = array.ndim
+
     if isinstance(pad_width, int):
         if pad_width < 0:
             raise ValueError("index can't contain negative values")
-        pad_width = ((pad_width, pad_width),) * array.ndim
+        pad_width = ((pad_width, pad_width),) * nd
     else:
         if dpnp.is_supported_array_type(pad_width):
             pad_width = dpnp.asnumpy(pad_width)
         else:
+            if isinstance(pad_width, dict):
+                pad_width = _pad_normalize_dict_width(pad_width, nd)
             pad_width = numpy.asarray(pad_width)
 
         if not pad_width.dtype.kind == "i":
             raise TypeError("`pad_width` must be of integral type.")
 
-        # Broadcast to shape (array.ndim, 2)
-        pad_width = _as_pairs(pad_width, array.ndim, as_index=True)
+        # Broadcast to shape (nd, 2)
+        pad_width = _as_pairs(pad_width, nd, as_index=True)
 
     if callable(mode):
         function = mode
@@ -683,7 +728,7 @@ def dpnp_pad(array, pad_width, mode="constant", **kwargs):
         if (
             dpnp.isscalar(values)
             and values == 0
-            and (array.ndim == 1 or array.size < 3e7)
+            and (nd == 1 or array.size < 3e7)
         ):
             # faster path for 1d arrays or small n-dimensional arrays
             return _pad_simple(array, pad_width, 0)[0]

dpnp/tests/test_arraypad.py

@@ -539,3 +539,21 @@ def test_as_pairs_exceptions(self):
             dpnp_as_pairs([[1, 2], [3, 4]], 3)
         with pytest.raises(ValueError, match="could not be broadcast"):
             dpnp_as_pairs(dpnp.ones((2, 3)), 3)
+
+    @testing.with_requires("numpy>=2.4")
+    @pytest.mark.parametrize(
+        "sh, pad_width",
+        [
+            ((3, 4, 5), {-2: (1, 3)}),
+            ((3, 4, 5), {0: (5, 2)}),
+            ((3, 4, 5), {0: (5, 2), -1: (3, 4)}),
+            ((3, 4, 5), {1: 5}),
+        ],
+    )
+    def test_dict_pad_width(self, sh, pad_width):
+        a = numpy.zeros(sh)
+        ia = dpnp.array(a)
+
+        result = dpnp.pad(ia, pad_width)
+        expected = numpy.pad(a, pad_width)
+        assert_equal(result, expected)