From 95f2743d85f86dfab4342d4f7268905dd6e791c8 Mon Sep 17 00:00:00 2001
From: fandawei <fandawei2@huawei.com>
Date: Wed, 8 Feb 2023 22:26:07 +0800
Subject: [PATCH] modify the logic of float_power for complex

---
 .../mindspore/ops/function/math_func.py       | 26 ++++++++-----------
 tests/st/ops/test_ops_float_power.py          | 19 --------------
 tests/st/tensor/test_float_power.py           | 19 --------------
 3 files changed, 11 insertions(+), 53 deletions(-)

diff --git a/mindspore/python/mindspore/ops/function/math_func.py b/mindspore/python/mindspore/ops/function/math_func.py
index 7a46dc7d579..305aefbdeee 100644
--- a/mindspore/python/mindspore/ops/function/math_func.py
+++ b/mindspore/python/mindspore/ops/function/math_func.py
@@ -954,11 +954,8 @@ def divide(x, other, *, rounding_mode=None):
 def float_power(x, exponent):
     """
     Computes `x` to the power of the exponent.
-    For the real number type, use mindspore.float64 to calculate.
-    For the complex type, use the same type of calculation as the input data.
-
-    .. Note::
-        On GPU, complex dtypes are not supported.
+    For the real number type, cast x and expoent to mindspore.float64 to calculate.
+    Currently, complex type calculation is not supported.
 
     Args:
         x (Union[Tensor, Number]): The first input is a tensor or a number.
@@ -983,22 +980,21 @@ def float_power(x, exponent):
         >>> print(output)
         [2.25 0.   4.  ]
     """
-    if not (isinstance(x, (Tensor, Tensor_)) or isinstance(exponent, (Tensor, Tensor_))):
+    if not (isinstance(x, Tensor) or isinstance(exponent, Tensor)):
         raise TypeError("At least one of the types of inputs must be tensor, " + \
                         f"but the type of 'x' got is {type(x)}, " + \
                         f"and the type of 'exponent' is {type(exponent)}.")
-    if not isinstance(x, (Tensor, Tensor_, numbers.Number)):
+    if not isinstance(x, (Tensor, numbers.Number)):
         raise TypeError(f"The type of 'x' must be Tensor or Number, but got {type(x)}.")
-    if not isinstance(exponent, (Tensor, Tensor_, numbers.Number)):
+    if not isinstance(exponent, (Tensor, numbers.Number)):
         raise TypeError(f"The type of 'exponent' must be Tensor or Number, but got {type(exponent)}.")
 
-    if isinstance(x, (Tensor, Tensor_)) and is_complex(x) and isinstance(exponent, numbers.Number):
-        exponent = cast_(exponent, x.dtype)
-    elif isinstance(exponent, (Tensor, Tensor_)) and is_complex(exponent) and isinstance(x, numbers.Number):
-        x = cast_(x, exponent.dtype)
-    # If both x and exponent are complex Tensor, no processing is required.
-    elif not (isinstance(x, (Tensor, Tensor_)) and is_complex(x) and \
-              isinstance(exponent, (Tensor, Tensor_)) and is_complex(exponent)):
+    if (isinstance(x, Tensor) and is_complex(x)) or \
+            (isinstance(exponent, Tensor) and is_complex(exponent)) or \
+            isinstance(x, complex) or isinstance(exponent, complex):
+        x = cast_(x, mstype.complex128)
+        exponent = cast_(exponent, mstype.complex128)
+    else:
         x = cast_(x, mstype.float64)
         exponent = cast_(exponent, mstype.float64)
     return pow(x, exponent)
diff --git a/tests/st/ops/test_ops_float_power.py b/tests/st/ops/test_ops_float_power.py
index f2dbe6ef7b3..78588d6dbda 100644
--- a/tests/st/ops/test_ops_float_power.py
+++ b/tests/st/ops/test_ops_float_power.py
@@ -46,22 +46,3 @@ def test_float_power_real(mode):
     except_case = np.array([9.0000, 4.0000, 1.0000, 1.0000, 4.0000, 9.0000], dtype=np.float32)
     assert output_case.asnumpy().dtype == np.float64
     assert np.allclose(output_case.asnumpy(), except_case)
-
-
-@pytest.mark.level0
-@pytest.mark.platform_x86_cpu
-@pytest.mark.platform_arm_cpu
-@pytest.mark.env_onecard
-@pytest.mark.parametrize('mode', [ms.GRAPH_MODE, ms.PYNATIVE_MODE])
-def test_float_power_complex(mode):
-    """
-    Feature: ops.float_power
-    Description: Verify the result of float_power
-    Expectation: success
-    """
-    ms.set_context(mode=mode)
-    net = Net()
-    input_case = ms.Tensor(np.array([complex(2, 3), complex(3, 4)]), ms.complex64)
-    output_case = net(input_case, 2)
-    except_case = np.array([complex(-5, 12), complex(-7, 24)], dtype=np.complex64)
-    assert np.allclose(output_case.asnumpy(), except_case)
diff --git a/tests/st/tensor/test_float_power.py b/tests/st/tensor/test_float_power.py
index 4db89b374ca..fccbb3bdfd6 100644
--- a/tests/st/tensor/test_float_power.py
+++ b/tests/st/tensor/test_float_power.py
@@ -44,22 +44,3 @@ def test_float_power_real(mode):
     except_case = np.array([9.0000, 4.0000, 1.0000, 1.0000, 4.0000, 9.0000], dtype=np.float32)
     assert output_case.asnumpy().dtype == np.float64
     assert np.allclose(output_case.asnumpy(), except_case)
-
-
-@pytest.mark.level0
-@pytest.mark.platform_x86_cpu
-@pytest.mark.platform_arm_cpu
-@pytest.mark.env_onecard
-@pytest.mark.parametrize('mode', [ms.GRAPH_MODE, ms.PYNATIVE_MODE])
-def test_float_power_complex(mode):
-    """
-    Feature: tensor.float_power
-    Description: Verify the result of float_power
-    Expectation: success
-    """
-    ms.set_context(mode=mode)
-    net = Net()
-    input_case = ms.Tensor(np.array([complex(2, 3), complex(3, 4)]), ms.complex64)
-    output_case = net(input_case, 2)
-    except_case = np.array([complex(-5, 12), complex(-7, 24)], dtype=np.complex64)
-    assert np.allclose(output_case.asnumpy(), except_case)