diff --git a/paddle/fluid/API.spec b/paddle/fluid/API.spec
index c4a65bb13e25c550b3cdd08a0f1c8d11bee5abe8..308ab13c638fec67d29e51ac4c9d4f59c55dde3f 100644
--- a/paddle/fluid/API.spec
+++ b/paddle/fluid/API.spec
@@ -178,14 +178,14 @@ paddle.fluid.layers.unstack ArgSpec(args=['x', 'axis', 'num'], varargs=None, key
 paddle.fluid.layers.sequence_enumerate ArgSpec(args=['input', 'win_size', 'pad_value', 'name'], varargs=None, keywords=None, defaults=(0, None))
 paddle.fluid.layers.expand ArgSpec(args=['x', 'expand_times', 'name'], varargs=None, keywords=None, defaults=(None,))
 paddle.fluid.layers.sequence_concat ArgSpec(args=['input', 'name'], varargs=None, keywords=None, defaults=(None,))
-paddle.fluid.layers.scale ArgSpec(args=['x', 'scale', 'bias', 'bias_after_scale', 'act', 'name'], varargs=None, keywords=None, defaults=(1.0, 0.0, True, None, None))
-paddle.fluid.layers.elementwise_add ArgSpec(args=['x', 'y', 'axis', 'use_mkldnn', 'act', 'name'], varargs=None, keywords=None, defaults=(-1, False, None, None))
-paddle.fluid.layers.elementwise_div ArgSpec(args=['x', 'y', 'axis', 'use_mkldnn', 'act', 'name'], varargs=None, keywords=None, defaults=(-1, False, None, None))
-paddle.fluid.layers.elementwise_sub ArgSpec(args=['x', 'y', 'axis', 'use_mkldnn', 'act', 'name'], varargs=None, keywords=None, defaults=(-1, False, None, None))
-paddle.fluid.layers.elementwise_mul ArgSpec(args=['x', 'y', 'axis', 'use_mkldnn', 'act', 'name'], varargs=None, keywords=None, defaults=(-1, False, None, None))
-paddle.fluid.layers.elementwise_max ArgSpec(args=['x', 'y', 'axis', 'use_mkldnn', 'act', 'name'], varargs=None, keywords=None, defaults=(-1, False, None, None))
-paddle.fluid.layers.elementwise_min ArgSpec(args=['x', 'y', 'axis', 'use_mkldnn', 'act', 'name'], varargs=None, keywords=None, defaults=(-1, False, None, None))
-paddle.fluid.layers.elementwise_pow ArgSpec(args=['x', 'y', 'axis', 'use_mkldnn', 'act', 'name'], varargs=None, keywords=None, defaults=(-1, False, None, None))
+paddle.fluid.layers.scale ArgSpec(args=['x', 'scale', 'bias', 'bias_after_scale', 'out', 'act', 'name'], varargs=None, keywords=None, defaults=(1.0, 0.0, True, None, None, None))
+paddle.fluid.layers.elementwise_add ArgSpec(args=['x', 'y', 'out', 'axis', 'use_mkldnn', 'act', 'name'], varargs=None, keywords=None, defaults=(None, -1, False, None, None))
+paddle.fluid.layers.elementwise_div ArgSpec(args=['x', 'y', 'out', 'axis', 'use_mkldnn', 'act', 'name'], varargs=None, keywords=None, defaults=(None, -1, False, None, None))
+paddle.fluid.layers.elementwise_sub ArgSpec(args=['x', 'y', 'out', 'axis', 'use_mkldnn', 'act', 'name'], varargs=None, keywords=None, defaults=(None, -1, False, None, None))
+paddle.fluid.layers.elementwise_mul ArgSpec(args=['x', 'y', 'out', 'axis', 'use_mkldnn', 'act', 'name'], varargs=None, keywords=None, defaults=(None, -1, False, None, None))
+paddle.fluid.layers.elementwise_max ArgSpec(args=['x', 'y', 'out', 'axis', 'use_mkldnn', 'act', 'name'], varargs=None, keywords=None, defaults=(None, -1, False, None, None))
+paddle.fluid.layers.elementwise_min ArgSpec(args=['x', 'y', 'out', 'axis', 'use_mkldnn', 'act', 'name'], varargs=None, keywords=None, defaults=(None, -1, False, None, None))
+paddle.fluid.layers.elementwise_pow ArgSpec(args=['x', 'y', 'out', 'axis', 'use_mkldnn', 'act', 'name'], varargs=None, keywords=None, defaults=(None, -1, False, None, None))
 paddle.fluid.layers.data ArgSpec(args=['name', 'shape', 'append_batch_size', 'dtype', 'lod_level', 'type', 'stop_gradient'], varargs=None, keywords=None, defaults=(True, 'float32', 0, VarType.LOD_TENSOR, True))
 paddle.fluid.layers.open_files ArgSpec(args=['filenames', 'shapes', 'lod_levels', 'dtypes', 'thread_num', 'buffer_size', 'pass_num', 'is_test'], varargs=None, keywords=None, defaults=(None, None, 1, None))
 paddle.fluid.layers.read_file ArgSpec(args=['reader'], varargs=None, keywords=None, defaults=None)
diff --git a/python/paddle/fluid/layers/nn.py b/python/paddle/fluid/layers/nn.py
index 6e0f3de4141aff3efb98b6d1162823a13f83a7bb..2cb61a9cd25c744710ab7ac9ea591902740f78da 100644
--- a/python/paddle/fluid/layers/nn.py
+++ b/python/paddle/fluid/layers/nn.py
@@ -6471,12 +6471,14 @@ def _elementwise_op(helper):
     assert y is not None, 'y cannot be None in {}'.format(op_type)
     axis = helper.kwargs.get('axis', -1)
     use_mkldnn = helper.kwargs.get('use_mkldnn', False)
-    name = helper.kwargs.get('name', None)
-    if name is None:
-        out = helper.create_tmp_variable(dtype=x.dtype)
-    else:
-        out = helper.create_variable(
-            name=name, dtype=x.dtype, persistable=False)
+    out = helper.kwargs.get('out', None)
+    if out is None:
+        name = helper.kwargs.get('name', None)
+        if name is None:
+            out = helper.create_tmp_variable(dtype=x.dtype)
+        else:
+            out = helper.create_variable(
+                name=name, dtype=x.dtype, persistable=False)
 
     helper.append_op(
         type=op_type,
@@ -6489,7 +6491,13 @@ def _elementwise_op(helper):
 
 
 @templatedoc()
-def scale(x, scale=1.0, bias=0.0, bias_after_scale=True, act=None, name=None):
+def scale(x,
+          scale=1.0,
+          bias=0.0,
+          bias_after_scale=True,
+          out=None,
+          act=None,
+          name=None):
     """
     ${comment}
 
@@ -6498,6 +6506,7 @@ def scale(x, scale=1.0, bias=0.0, bias_after_scale=True, act=None, name=None):
         scale(${scale_type}): ${scale_comment}
         bias(${bias_type}): ${bias_comment}
         bias_after_scale(${bias_after_scale_type}): ${bias_after_scale_comment}
+        out(Tensor): Output tensor.
         act(basestring|None): Activation applied to the output.
         name(basestring|None): Name of the output. 
 
@@ -6506,11 +6515,12 @@ def scale(x, scale=1.0, bias=0.0, bias_after_scale=True, act=None, name=None):
     """
 
     helper = LayerHelper('scale', **locals())
-    if name is None:
-        out = helper.create_tmp_variable(dtype=x.dtype)
-    else:
-        out = helper.create_variable(
-            name=name, dtype=x.dtype, persistable=False)
+    if out is None:
+        if name is None:
+            out = helper.create_tmp_variable(dtype=x.dtype)
+        else:
+            out = helper.create_variable(
+                name=name, dtype=x.dtype, persistable=False)
 
     helper.append_op(
         type='scale',
@@ -6524,31 +6534,73 @@ def scale(x, scale=1.0, bias=0.0, bias_after_scale=True, act=None, name=None):
     return helper.append_activation(out)
 
 
-def elementwise_add(x, y, axis=-1, use_mkldnn=False, act=None, name=None):
+def elementwise_add(x,
+                    y,
+                    out=None,
+                    axis=-1,
+                    use_mkldnn=False,
+                    act=None,
+                    name=None):
     return _elementwise_op(LayerHelper('elementwise_add', **locals()))
 
 
-def elementwise_div(x, y, axis=-1, use_mkldnn=False, act=None, name=None):
+def elementwise_div(x,
+                    y,
+                    out=None,
+                    axis=-1,
+                    use_mkldnn=False,
+                    act=None,
+                    name=None):
     return _elementwise_op(LayerHelper('elementwise_div', **locals()))
 
 
-def elementwise_sub(x, y, axis=-1, use_mkldnn=False, act=None, name=None):
+def elementwise_sub(x,
+                    y,
+                    out=None,
+                    axis=-1,
+                    use_mkldnn=False,
+                    act=None,
+                    name=None):
     return _elementwise_op(LayerHelper('elementwise_sub', **locals()))
 
 
-def elementwise_mul(x, y, axis=-1, use_mkldnn=False, act=None, name=None):
+def elementwise_mul(x,
+                    y,
+                    out=None,
+                    axis=-1,
+                    use_mkldnn=False,
+                    act=None,
+                    name=None):
     return _elementwise_op(LayerHelper('elementwise_mul', **locals()))
 
 
-def elementwise_max(x, y, axis=-1, use_mkldnn=False, act=None, name=None):
+def elementwise_max(x,
+                    y,
+                    out=None,
+                    axis=-1,
+                    use_mkldnn=False,
+                    act=None,
+                    name=None):
     return _elementwise_op(LayerHelper('elementwise_max', **locals()))
 
 
-def elementwise_min(x, y, axis=-1, use_mkldnn=False, act=None, name=None):
+def elementwise_min(x,
+                    y,
+                    out=None,
+                    axis=-1,
+                    use_mkldnn=False,
+                    act=None,
+                    name=None):
     return _elementwise_op(LayerHelper('elementwise_min', **locals()))
 
 
-def elementwise_pow(x, y, axis=-1, use_mkldnn=False, act=None, name=None):
+def elementwise_pow(x,
+                    y,
+                    out=None,
+                    axis=-1,
+                    use_mkldnn=False,
+                    act=None,
+                    name=None):
     return _elementwise_op(LayerHelper('elementwise_pow', **locals()))
 
 
@@ -6560,6 +6612,7 @@ for func in [
     func.__doc__ = _generate_doc_string_(
         op_proto,
         additional_args_lines=[
+            "out (Tensor): The output tensor of elementwise op.",
             "act (basestring|None): Activation applied to the output.",
             "name (basestring|None): Name of the output."
         ])