Composite hardswish

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

@@ -2150,6 +2150,7 @@ def setUp(self):
         self.op_type = 'hard_swish'
         self.init_dtype()
         self.init_shape()
+        self.prim_op_type = "comp"
         self.python_api = paddle.nn.functional.hardswish
 
         np.random.seed(1024)
@@ -2165,18 +2166,42 @@ def setUp(self):
         self.inputs = {'X': x}
         self.attrs = {'threshold': threshold, 'scale': scale, 'offset': offset}
         self.outputs = {'Out': out}
+        self.enable_cinn = False
 
     def init_shape(self):
         self.shape = [10, 12]
 
     def test_check_grad(self):
-        self.check_grad(['X'], 'Out', check_eager=True)
+        self.check_grad(['X'], 'Out', check_eager=True, check_prim=True)
 
     def test_check_output(self):
-        self.check_output(check_eager=True)
+        self.check_output(check_eager=True, check_prim=True)
 
 
 class TestHardSwish_ZeroDim(TestHardSwish):
+    def setUp(self):
+        super().setUp()
+        self.enable_cinn = False
+
+    def init_shape(self):
+        self.shape = []
+
+
+class TestHardSwishFP16(TestHardSwish):
+    def setUp(self):
+        super().setUp()
+        self.only_prim = True
+        self.enable_cinn = False
+
+    def init_dtype(self):
+        self.dtype = np.float16
+
+
+class TestHardSwish_ZeroDim_FP16(TestHardSwishFP16):
+    def setUp(self):
+        super().setUp()
+        self.enable_cinn = False
+
     def init_shape(self):
         self.shape = []
 

python/paddle/incubate/autograd/composite_rules.py

@@ -261,6 +261,31 @@ def bernoulli(shape, dtype, p, seed=0):
     )
 
 
+@REGISTER_COMPOSITE('hard_swish')
+def hard_swish_composite(x):
+    """define composite rule of op hard_swish.
+    offset=3, threshold=6, scale=6
+    out = minimum(
+        maxmum(x + offset, 0), threshold
+    ) * x / scale
+    """
+    offset = 3.0
+    threshold = 6.0
+    scale = 6.0
+    res = (
+        minimum(
+            maximum(
+                x + full(x.shape, offset, dtype=x.dtype),
+                full(x.shape, 0.0, dtype=x.dtype),
+            ),
+            full(x.shape, threshold, dtype=x.dtype),
+        )
+        * x
+        / full(x.shape, scale, dtype=x.dtype)
+    )
+    return res
+
+
 @REGISTER_COMPOSITE('silu')
 def silu_composite(x):
     """

python/paddle/incubate/autograd/primitives.py

@@ -60,6 +60,8 @@
 from paddle.tensor import zeros  # noqa: F401
 from paddle.tensor.creation import assign  # noqa: F401
 from paddle.tensor.manipulation import cast  # noqa: F401
+from paddle.tensor.math import maximum  # noqa: F401
+from paddle.tensor.math import minimum  # noqa: F401
 
 """
 math_op = [
@@ -87,6 +89,8 @@
     'logit',
     'max',
     'min',
+    'minimum',
+    'maximum'
 ]
 
 trigonometric_op = [