Improve performance for dpnp.diag() (#1822)

* Update dpnp.diag to improve perfomance

* Use dpnp.zero_like() instead of dpnp.zeros()

* Update doctrings for dpnp.diagflat/eye/identity

* Update cupy test_matrix.py

dpnp/dpnp_iface_arraycreation.py

@@ -866,9 +866,14 @@ def diag(v, /, k=0, *, device=None, usm_type=None, sycl_queue=None):
     v : array_like
         Input data, in any form that can be converted to an array. This
         includes scalars, lists, lists of tuples, tuples, tuples of tuples,
-        tuples of lists, and ndarrays. If `v` is a 2-D array, return a copy of
-        its k-th diagonal. If `v` is a 1-D array, return a 2-D array with `v`
-        on the k-th diagonal.
+        tuples of lists, and ndarrays.
+        If `v` is a 1-D array, return a 2-D array with `v`
+        on the `k`-th diagonal.
+        If `v` is a 2-D array and is an instance of
+        {dpnp.ndarray, usm_ndarray}, then:
+          - If `device`, `usm_type`, and `sycl_queue` are set to their
+            default values, returns a read/write view of its k-th diagonal.
+          - Otherwise, returns a copy of its k-th diagonal.
     k : int, optional
         Diagonal in question. The default is 0. Use k > 0 for diagonals above
         the main diagonal, and k < 0 for diagonals below the main diagonal.
@@ -894,79 +899,62 @@ def diag(v, /, k=0, *, device=None, usm_type=None, sycl_queue=None):
     --------
     :obj:`diagonal` : Return specified diagonals.
     :obj:`diagflat` : Create a 2-D array with the flattened input as a diagonal.
-    :obj:`trace` : Return sum along diagonals.
-    :obj:`triu` : Return upper triangle of an array.
-    :obj:`tril` : Return lower triangle of an array.
+    :obj:`trace` : Return the sum along diagonals of the array.
+    :obj:`triu` : Upper triangle of an array.
+    :obj:`tril` : Lower triangle of an array.
 
     Examples
     --------
     >>> import dpnp as np
-    >>> x0 = np.arange(9).reshape((3, 3))
-    >>> x0
+    >>> x = np.arange(9).reshape((3, 3))
+    >>> x
     array([[0, 1, 2],
            [3, 4, 5],
            [6, 7, 8]])
 
-    >>> np.diag(x0)
+    >>> np.diag(x)
     array([0, 4, 8])
-    >>> np.diag(x0, k=1)
+    >>> np.diag(x, k=1)
     array([1, 5])
-    >>> np.diag(x0, k=-1)
+    >>> np.diag(x, k=-1)
     array([3, 7])
 
-    >>> np.diag(np.diag(x0))
+    >>> np.diag(np.diag(x))
     array([[0, 0, 0],
            [0, 4, 0],
            [0, 0, 8]])
 
     Creating an array on a different device or with a specified usm_type
 
-    >>> x = np.diag(x0) # default case
-    >>> x, x.device, x.usm_type
+    >>> res = np.diag(x) # default case
+    >>> res, res.device, res.usm_type
     (array([0, 4, 8]), Device(level_zero:gpu:0), 'device')
 
-    >>> y = np.diag(x0, device="cpu")
-    >>> y, y.device, y.usm_type
+    >>> res_cpu = np.diag(x, device="cpu")
+    >>> res_cpu, res_cpu.device, res_cpu.usm_type
     (array([0, 4, 8]), Device(opencl:cpu:0), 'device')
 
-    >>> z = np.diag(x0, usm_type="host")
-    >>> z, z.device, z.usm_type
+    >>> res_host = np.diag(x, usm_type="host")
+    >>> res_host, res_host.device, res_host.usm_type
     (array([0, 4, 8]), Device(level_zero:gpu:0), 'host')
 
     """
 
     if not isinstance(k, int):
         raise TypeError(f"An integer is required, but got {type(k)}")
 
-    v = dpnp.asarray(v, device=device, usm_type=usm_type, sycl_queue=sycl_queue)
+    v = dpnp.asanyarray(
+        v, device=device, usm_type=usm_type, sycl_queue=sycl_queue
+    )
 
-    init0 = max(0, -k)
-    init1 = max(0, k)
     if v.ndim == 1:
         size = v.shape[0] + abs(k)
-        m = dpnp.zeros(
-            (size, size),
-            dtype=v.dtype,
-            usm_type=v.usm_type,
-            sycl_queue=v.sycl_queue,
-        )
-        for i in range(v.shape[0]):
-            m[(init0 + i), init1 + i] = v[i]
-        return m
+        ret = dpnp.zeros_like(v, shape=(size, size))
+        ret.diagonal(k)[:] = v
+        return ret
 
     if v.ndim == 2:
-        size = max(
-            0, min(v.shape[0], v.shape[0] + k, v.shape[1], v.shape[1] - k)
-        )
-        m = dpnp.zeros(
-            (size,),
-            dtype=v.dtype,
-            usm_type=v.usm_type,
-            sycl_queue=v.sycl_queue,
-        )
-        for i in range(size):
-            m[i] = v[(init0 + i), init1 + i]
-        return m
+        return v.diagonal(k)
 
     raise ValueError("Input must be a 1-D or 2-D array.")
 
@@ -1008,9 +996,9 @@ def diagflat(v, /, k=0, *, device=None, usm_type=None, sycl_queue=None):
 
     See Also
     --------
-    :obj:`diag` : Return the extracted diagonal or constructed diagonal array.
-    :obj:`diagonal` : Return specified diagonals.
-    :obj:`trace` : Return sum along diagonals.
+    :obj:`dpnp.diag` : Extract a diagonal or construct a diagonal array.
+    :obj:`dpnp.diagonal` : Return specified diagonals.
+    :obj:`dpnp.trace` : Return sum along diagonals.
 
     Examples
     --------
@@ -1324,6 +1312,11 @@ def eye(
     Parameter `like` is supported only with default value ``None``.
     Otherwise, the function raises `NotImplementedError` exception.
 
+    See Also
+    --------
+    :obj:`dpnp.identity` : Return the identity array.
+    :obj:`dpnp.diag` : Extract a diagonal or construct a diagonal array.
+
     Examples
     --------
     >>> import dpnp as np
@@ -2264,7 +2257,7 @@ def identity(
     :obj:`dpnp.eye` : Return a 2-D array with ones on the diagonal and zeros
                       elsewhere.
     :obj:`dpnp.ones` : Return a new array setting values to one.
-    :obj:`dpnp.diag` : Return diagonal 2-D array from an input 1-D array.
+    :obj:`dpnp.diag` : Extract a diagonal or construct a diagonal array.
 
     Examples
     --------

tests/third_party/cupy/creation_tests/test_matrix.py

@@ -27,35 +27,35 @@ def test_diag3(self, xp):
     def test_diag_extraction_from_nested_list(self, xp):
         a = [[1, 2, 3], [4, 5, 6], [7, 8, 9]]
         r = xp.diag(a, 1)
-        self.assertIsInstance(r, xp.ndarray)
+        assert isinstance(r, xp.ndarray)
         return r
 
     @testing.numpy_cupy_array_equal()
     def test_diag_extraction_from_nested_tuple(self, xp):
         a = ((1, 2, 3), (4, 5, 6), (7, 8, 9))
         r = xp.diag(a, -1)
-        self.assertIsInstance(r, xp.ndarray)
+        assert isinstance(r, xp.ndarray)
         return r
 
     @testing.numpy_cupy_array_equal()
     def test_diag_construction(self, xp):
         a = testing.shaped_arange((3,), xp)
         r = xp.diag(a)
-        self.assertIsInstance(r, xp.ndarray)
+        assert isinstance(r, xp.ndarray)
         return r
 
     @testing.numpy_cupy_array_equal()
     def test_diag_construction_from_list(self, xp):
         a = [1, 2, 3]
         r = xp.diag(a)
-        self.assertIsInstance(r, xp.ndarray)
+        assert isinstance(r, xp.ndarray)
         return r
 
     @testing.numpy_cupy_array_equal()
     def test_diag_construction_from_tuple(self, xp):
         a = (1, 2, 3)
         r = xp.diag(a)
-        self.assertIsInstance(r, xp.ndarray)
+        assert isinstance(r, xp.ndarray)
         return r
 
     def test_diag_scaler(self):