[Phi] Support scale dygraph final state (#41321)

* support scale final state

* fix inplace error

* pass arg directly

* pass arg directly for inplace api

* fix type

paddle/fluid/eager/auto_code_generator/final_state_generator/python_c_gen.py

@@ -23,7 +23,7 @@ from codegen_utils import ParseYamlForward, GetInplacedFunctionName
 ###########################
 ## Global Configurations ##
 ###########################
-skipped_forward_api_names = set(["scale"])
+skipped_forward_api_names = set([])
 
 
 def SkipAPIGeneration(forward_api_name):

python/paddle/fluid/layers/nn.py

@@ -11779,6 +11779,9 @@ def scale(x, scale=1.0, bias=0.0, bias_after_scale=True, act=None, name=None):
 
     """
 
+    if in_dygraph_mode():
+        out = _C_ops.final_state_scale(x, scale, float(bias), bias_after_scale)
+        return dygraph_utils._append_activation_in_dygraph(out)
     if _non_static_mode():
         _scale = scale.numpy().item(0) if isinstance(scale, Variable) else scale
         out = _C_ops.scale(x, 'scale',

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

@@ -27,6 +27,7 @@ from paddle.static import Program, program_guard
 class TestScaleOp(OpTest):
     def setUp(self):
         self.op_type = "scale"
+        self.python_api = paddle.scale
         self.dtype = np.float64
         self.init_dtype_type()
         self.inputs = {'X': np.random.random((10, 10)).astype(self.dtype)}
@@ -39,15 +40,16 @@ class TestScaleOp(OpTest):
         pass
 
     def test_check_output(self):
-        self.check_output()
+        self.check_output(check_eager=True)
 
     def test_check_grad(self):
-        self.check_grad(['X'], 'Out')
+        self.check_grad(['X'], 'Out', check_eager=True)
 
 
 class TestScaleOpScaleVariable(OpTest):
     def setUp(self):
         self.op_type = "scale"
+        self.python_api = paddle.scale
         self.dtype = np.float64
         self.init_dtype_type()
         self.scale = -2.3
@@ -62,10 +64,10 @@ class TestScaleOpScaleVariable(OpTest):
         pass
 
     def test_check_output(self):
-        self.check_output()
+        self.check_output(check_eager=True)
 
     def test_check_grad(self):
-        self.check_grad(['X'], 'Out')
+        self.check_grad(['X'], 'Out', check_eager=True)
 
 
 class TestScaleOpSelectedRows(unittest.TestCase):
@@ -144,18 +146,19 @@ class TestScaleFp16Op(TestScaleOp):
     def test_check_output(self):
         place = core.CUDAPlace(0)
         if core.is_float16_supported(place):
-            self.check_output_with_place(place, atol=0.002)
+            self.check_output_with_place(place, atol=0.002, check_eager=True)
 
     def test_check_grad(self):
         place = core.CUDAPlace(0)
         if core.is_float16_supported(place):
             self.check_grad_with_place(
-                place, ["X"], "Out", max_relative_error=0.05)
+                place, ["X"], "Out", max_relative_error=0.05, check_eager=True)
 
 
 class TestScaleBF16Op(OpTest):
     def setUp(self):
         self.op_type = "scale"
+        self.python_api = paddle.scale
         self.dtype = np.uint16
         self.attrs = {'scale': -2.3}
         x = np.random.random((10, 10)).astype(np.float32)
@@ -164,10 +167,10 @@ class TestScaleBF16Op(OpTest):
         self.outputs = {'Out': convert_float_to_uint16(out)}
 
     def test_check_output(self):
-        self.check_output()
+        self.check_output(check_eager=True)
 
     def test_check_grad(self):
-        self.check_grad(['X'], 'Out', numeric_grad_delta=0.8)
+        self.check_grad(['X'], 'Out', numeric_grad_delta=0.8, check_eager=True)
 
 
 @unittest.skipIf(not core.is_compiled_with_cuda(),

python/paddle/tensor/math.py

@@ -98,10 +98,13 @@ def scale_(x, scale=1.0, bias=0.0, bias_after_scale=True, act=None, name=None):
     Inplace version of ``scale`` API, the output Tensor will be inplaced with input ``x``.
     Please refer to :ref:`api_tensor_scale`.
     """
-    _scale = scale.numpy().item(0) if isinstance(scale, Variable) else scale
-    return _C_ops.scale_(x, 'scale',
-                            float(_scale), 'bias',
-                            float(bias), 'bias_after_scale', bias_after_scale)
+    if in_dygraph_mode():
+        return _C_ops.final_state_scale_(x, scale, float(bias), bias_after_scale)
+    if _in_legacy_dygraph():
+        _scale = scale.numpy().item(0) if isinstance(scale, Variable) else scale
+        return _C_ops.scale_(x, 'scale',
+                                float(_scale), 'bias',
+                                float(bias), 'bias_after_scale', bias_after_scale)
 
 
 def pow(x, y, name=None):

python/paddle/utils/code_gen/api.yaml

@@ -1345,6 +1345,7 @@
   kernel :
     func : scale, scale_sr
   inplace : (x -> out)
+  backward : scale_grad
 
 - api : scatter
   args : (Tensor x, Tensor index, Tensor updates, bool overwrite)

python/paddle/utils/code_gen/backward.yaml

@@ -947,7 +947,7 @@
 
 - backward_api : scale_grad
   forward : scale (Tensor x, Scalar scale, float bias, bool bias_after_scale) -> Tensor(out)
-  args : (Tensor out_grad, Scalar scale, float bias=0.0, bool bias_after_scale=true)
+  args : (Tensor out_grad, Scalar scale=1.0, float bias=0.0, bool bias_after_scale=true)
   output : Tensor(x_grad)
   invoke : scale(out_grad, scale, bias, bias_after_scale)