following NEP-50 for dpnp.einsum (#2120)

Author: vtavana

dpnp/dpnp_iface_statistics.py

@@ -370,7 +370,7 @@ def correlate(x1, x2, mode="valid"):
     -----------
     Input arrays are supported as :obj:`dpnp.ndarray`.
     Size and shape of input arrays are supported to be equal.
-    Parameter `mode` is supported only with default value ``"valid``.
+    Parameter `mode` is supported only with default value ``"valid"``.
     Otherwise the function will be executed sequentially on CPU.
     Input array data types are limited by supported DPNP :ref:`Data types`.
 

dpnp/dpnp_utils/dpnp_utils_einsum.py

@@ -33,9 +33,8 @@
 from dpctl.utils import ExecutionPlacementError
 
 import dpnp
-from dpnp.dpnp_utils import get_usm_allocations
-
-from ..dpnp_array import dpnp_array
+from dpnp.dpnp_array import dpnp_array
+from dpnp.dpnp_utils import get_usm_allocations, map_dtype_to_device
 
 _einsum_symbols = "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ"
 
@@ -1027,17 +1026,16 @@ def dpnp_einsum(
                 "Input and output allocation queues are not compatible"
             )
 
-    result_dtype = dpnp.result_type(*arrays) if dtype is None else dtype
     for id, a in enumerate(operands):
         if dpnp.isscalar(a):
+            scalar_dtype = map_dtype_to_device(type(a), exec_q.sycl_device)
             operands[id] = dpnp.array(
-                a, dtype=result_dtype, usm_type=res_usm_type, sycl_queue=exec_q
+                a, dtype=scalar_dtype, usm_type=res_usm_type, sycl_queue=exec_q
             )
+            arrays.append(operands[id])
     result_dtype = dpnp.result_type(*arrays) if dtype is None else dtype
-    if order in ["a", "A"]:
-        order = (
-            "F" if not any(arr.flags.c_contiguous for arr in arrays) else "C"
-        )
+    if order in "aA":
+        order = "F" if all(arr.flags.fnc for arr in arrays) else "C"
 
     input_subscripts = [
         _parse_ellipsis_subscript(sub, idx, ndim=arr.ndim)

tests/test_linalg.py

@@ -1139,14 +1139,12 @@ def check_einsum_sums(self, dtype, do_opt=False):
         result = inp.einsum(*args, dtype="?", casting="unsafe", optimize=do_opt)
         assert_dtype_allclose(result, expected)
 
-        # with an scalar, NumPy < 2.0.0 uses the other input arrays to
-        # determine the output type while for NumPy > 2.0.0 the scalar
-        # with default machine dtype is used to determine the output
-        # data type
+        # NumPy >= 2.0 follows NEP-50 to determine the output dtype when one of
+        # the inputs is a scalar while NumPy < 2.0 does not
         if numpy.lib.NumpyVersion(numpy.__version__) < "2.0.0":
-            check_type = True
-        else:
             check_type = False
+        else:
+            check_type = True
         a = numpy.arange(9, dtype=dtype)
         a_dp = inp.array(a)
         expected = numpy.einsum(",i->", 3, a)
@@ -1712,7 +1710,7 @@ def test_broadcasting_dot_cases(self):
 
     def test_output_order(self):
         # Ensure output order is respected for optimize cases, the below
-        # conraction should yield a reshaped tensor view
+        # contraction should yield a reshaped tensor view
         a = inp.ones((2, 3, 5), order="F")
         b = inp.ones((4, 3), order="F")
 

tests/third_party/cupy/linalg_tests/test_einsum.py

@@ -475,13 +475,12 @@ def test_einsum_binary(self, xp, dtype_a, dtype_b):
 
 
 class TestEinSumBinaryOperationWithScalar:
-    # with an scalar, NumPy < 2.0.0 uses the other input arrays to determine
-    # the output type while for NumPy > 2.0.0 the scalar with default machine
-    # dtype is used to determine the output type
+    # NumPy >= 2.0 follows NEP-50 to determine the output dtype when one of
+    # the inputs is a scalar while NumPy < 2.0 does not
     if numpy.lib.NumpyVersion(numpy.__version__) < "2.0.0":
-        type_check = has_support_aspect64()
-    else:
         type_check = False
+    else:
+        type_check = has_support_aspect64()
 
     @testing.for_all_dtypes()
     @testing.numpy_cupy_allclose(contiguous_check=False, type_check=type_check)