address comments

Author: Vahid Tavanashad

CHANGELOG.md

@@ -16,7 +16,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ### Changed
 
 * Allowed input array of `uint64` dtype in `dpnp.bincount` [#2361](https://github.com/IntelPython/dpnp/pull/2361)
-* The vector norms `ord={1, 2, inf}` and the matrix norms `ord={1, 2, inf, "fro", "nuc"}` now consistently return zero for empty arrays, which are arrays with at least one axis of size zero. This change affects `dpnp.linalg.norm`, `dpnp.linalg.vector_norm`, and `dpnp.linalg.matrix_norm`. Previously, dpnp would either raise errors or return zero depending on the parameters provided [#2371](https://github.com/IntelPython/dpnp/pull/2371)
+* The vector norms `ord={None, 1, 2, inf}` and the matrix norms `ord={None, 1, 2, inf, "fro", "nuc"}` now consistently return zero for empty arrays, which are arrays with at least one axis of size zero. This change affects `dpnp.linalg.norm`, `dpnp.linalg.vector_norm`, and `dpnp.linalg.matrix_norm`. Previously, dpnp would either raise errors or return zero depending on the parameters provided [#2371](https://github.com/IntelPython/dpnp/pull/2371)
 
 ### Fixed
 

dpnp/linalg/dpnp_utils_linalg.py

@@ -1185,6 +1185,9 @@ def _norm_int_axis(x, ord, axis, keepdims):
     """
 
     if ord == dpnp.inf:
+        if x.shape[axis] == 0:
+            x = dpnp.moveaxis(x, axis, -1)
+            return dpnp.zeros_like(x, shape=x.shape[:-1])
         return dpnp.abs(x).max(axis=axis, keepdims=keepdims)
     if ord == -dpnp.inf:
         return dpnp.abs(x).min(axis=axis, keepdims=keepdims)
@@ -1220,6 +1223,10 @@ def _norm_tuple_axis(x, ord, row_axis, col_axis, keepdims):
     """
 
     axis = (row_axis, col_axis)
+    flag = x.shape[row_axis] == 0 or x.shape[col_axis] == 0
+    if flag and ord in [1, 2, dpnp.inf]:
+        x = dpnp.moveaxis(x, axis, (-2, -1))
+        return dpnp.zeros_like(x, shape=x.shape[:-2])
     if row_axis == col_axis:
         raise ValueError("Duplicate axes given.")
     if ord == 2:
@@ -2401,17 +2408,10 @@ def dpnp_norm(x, ord=None, axis=None, keepdims=False):
         axis = (axis,)
 
     if len(axis) == 1:
-        if x.shape[axis[0]] == 0 and ord in [1, 2, dpnp.inf]:
-            x = dpnp.moveaxis(x, axis, -1)
-            return dpnp.zeros_like(x, shape=x.shape[:-1])
         axis = normalize_axis_index(axis[0], ndim)
         return _norm_int_axis(x, ord, axis, keepdims)
 
     if len(axis) == 2:
-        flag = x.shape[axis[0]] == 0 or x.shape[axis[1]] == 0
-        if flag and ord in ["fro", "nuc", 1, 2, dpnp.inf]:
-            x = dpnp.moveaxis(x, axis, (-2, -1))
-            return dpnp.zeros_like(x, shape=x.shape[:-2])
         row_axis, col_axis = axis
         row_axis = normalize_axis_index(row_axis, ndim)
         col_axis = normalize_axis_index(col_axis, ndim)

dpnp/tests/test_linalg.py

@@ -2097,13 +2097,13 @@ def test_empty(self, shape, ord, axis, keepdims):
             assert_raises(ValueError, dpnp.linalg.norm, ia, **kwarg)
             assert_raises(ValueError, numpy.linalg.norm, a, **kwarg)
         elif axis is None and a.ndim != 1 and a.shape[-1] == 0:
-            # TODO: when similar changes in numpy are available,
-            # instead of assert_equal with zero, we should compare with numpy
             if ord in [-2, -1, 0, 3]:
                 # reduction cannot be performed over zero-size axes
                 assert_raises(ValueError, dpnp.linalg.norm, ia, **kwarg)
                 assert_raises(ValueError, numpy.linalg.norm, a, **kwarg)
             else:
+                # TODO: when similar changes in numpy are available, instead
+                # of assert_equal with zero, we should compare with numpy
                 # ord in [None, 1, 2]
                 assert_equal(dpnp.linalg.norm(ia, **kwarg), 0)
         else:
@@ -2295,14 +2295,15 @@ def test_matrix_norm(self, ord, keepdims):
 
     @pytest.mark.parametrize("dtype", [dpnp.float32, dpnp.int32])
     @pytest.mark.parametrize(
-        "shape_axis", [[(2, 0), None], [(2, 0, 3), (0, 1)]]
+        "shape_axis", [[(2, 0), None], [(2, 0), (0, 1)], [(0, 2), (0, 1)]]
     )
     def test_matrix_norm_empty(self, dtype, shape_axis):
         shape, axis = shape_axis[0], shape_axis[1]
         x = dpnp.zeros(shape, dtype=dtype)
 
         # TODO: when similar changes in numpy are available,
         # instead of assert_equal with zero, we should compare with numpy
+        assert_equal(dpnp.linalg.norm(x, axis=axis), 0)
         assert_equal(dpnp.linalg.norm(x, axis=axis, ord="fro"), 0)
         assert_equal(dpnp.linalg.norm(x, axis=axis, ord="nuc"), 0)
         assert_equal(dpnp.linalg.norm(x, axis=axis, ord=2), 0)
@@ -2315,6 +2316,7 @@ def test_vector_norm_empty(self, dtype, axis):
         x = dpnp.zeros(0, dtype=dtype)
         # TODO: when similar changes in numpy are available,
         # instead of assert_equal with zero, we should compare with numpy
+        assert_equal(dpnp.linalg.vector_norm(x, axis=axis), 0)
         assert_equal(dpnp.linalg.vector_norm(x, axis=axis, ord=1), 0)
         assert_equal(dpnp.linalg.vector_norm(x, axis=axis, ord=2), 0)
         assert_equal(dpnp.linalg.vector_norm(x, axis=axis, ord=dpnp.inf), 0)