[AMP OP&Test]numel op fp/bf 16 support (#52172)

* [AMP OP&Test]numel op fp/bf 16 support

* dtype update

* remove err case

python/paddle/fluid/tests/unittests/test_numel_op.py

@@ -15,10 +15,11 @@
 import unittest
 
 import numpy as np
-from eager_op_test import OpTest
+from eager_op_test import OpTest, convert_float_to_uint16
 
 import paddle
 from paddle import fluid
+from paddle.fluid import core
 
 
 class TestNumelOp(OpTest):
@@ -26,7 +27,7 @@ class TestNumelOp(OpTest):
         self.op_type = "size"
         self.python_api = paddle.numel
         self.init()
-        x = np.random.random((self.shape)).astype("float64")
+        x = np.random.random((self.shape)).astype(self.dtype)
         self.inputs = {
             'Input': x,
         }
@@ -37,16 +38,65 @@ class TestNumelOp(OpTest):
 
     def init(self):
         self.shape = (6, 56, 8, 55)
+        self.dtype = np.float64
 
 
 class TestNumelOp1(TestNumelOp):
     def init(self):
         self.shape = (11, 66)
+        self.dtype = np.float64
 
 
 class TestNumelOp2(TestNumelOp):
     def init(self):
         self.shape = (0,)
+        self.dtype = np.float64
+
+
+class TestNumelOpFP16(TestNumelOp):
+    def init(self):
+        self.dtype = np.float16
+        self.shape = (6, 56, 8, 55)
+
+
+class TestNumelOp1FP16(TestNumelOp):
+    def init(self):
+        self.dtype = np.float16
+        self.shape = (11, 66)
+
+
+class TestNumelOp2FP16(TestNumelOp):
+    def init(self):
+        self.dtype = np.float16
+        self.shape = (0,)
+
+
+@unittest.skipIf(
+    not core.is_compiled_with_cuda()
+    or not core.is_bfloat16_supported(core.CUDAPlace(0)),
+    "core is not compiled with CUDA and do not support bfloat16",
+)
+class TestNumelOpBF16(OpTest):
+    def setUp(self):
+        self.op_type = "size"
+        self.python_api = paddle.numel
+        self.dtype = np.uint16
+        self.init()
+        x = np.random.random((self.shape)).astype(np.float32)
+        self.inputs = {'Input': convert_float_to_uint16(x)}
+        self.outputs = {'Out': np.array([np.size(x)])}
+
+    def test_check_output(self):
+        place = paddle.CUDAPlace(0)
+        self.check_output_with_place(place)
+
+    def init(self):
+        self.shape = (6, 56, 8, 55)
+
+
+class TestNumelOp1BF16(TestNumelOpBF16):
+    def init(self):
+        self.shape = (11, 66)
 
 
 class TestNumelAPI(unittest.TestCase):