diff --git a/paddle/framework/CMakeLists.txt b/paddle/framework/CMakeLists.txt
index 0d1617424ecffdcdaaccba6cbd761b2563f6b073..f69a3cfbf8b195cceb15fc2ce81d8bb9d28e5830 100644
--- a/paddle/framework/CMakeLists.txt
+++ b/paddle/framework/CMakeLists.txt
@@ -15,7 +15,7 @@ nv_test(lod_tensor_gpu_test SRCS lod_tensor_test.cu DEPS lod_tensor)
 
 cc_test(variable_test SRCS variable_test.cc)
 
-cc_library(scope SRCS scope.cc)
+cc_library(scope SRCS scope.cc DEPS glog)
 cc_test(scope_test SRCS scope_test.cc DEPS scope)
 
 
diff --git a/paddle/framework/scope.cc b/paddle/framework/scope.cc
index 19e25fba05f2f1c959da32c950320d3a44d5109d..14cc530448379eb6d4bf0435f607494aa01ef5b5 100644
--- a/paddle/framework/scope.cc
+++ b/paddle/framework/scope.cc
@@ -16,6 +16,7 @@ limitations under the License. */
 
 #include <memory>  // for unique_ptr
 #include <mutex>   // for call_once
+#include "glog/logging.h"
 #include "paddle/string/printf.h"
 
 namespace paddle {
@@ -23,7 +24,10 @@ namespace framework {
 
 Scope::~Scope() {
   DropKids();
-  for (auto& kv : vars_) delete kv.second;
+  for (auto& kv : vars_) {
+    VLOG(3) << "Destroy variable " << kv.first;
+    delete kv.second;
+  }
 }
 
 Scope& Scope::NewScope() const {
@@ -38,6 +42,7 @@ Variable* Scope::Var(const std::string& name) {
   }
   Variable* v = new Variable();
   vars_[name] = v;
+  VLOG(3) << "Create variable " << name << " on scope";
   v->name_ = &(vars_.find(name)->first);
   return v;
 }
diff --git a/paddle/memory/CMakeLists.txt b/paddle/memory/CMakeLists.txt
index 9cc4233e43267472d405c3e4e617f0782e1430ea..aed5275dbf9be707cc6e19e729133ba8eab58195 100644
--- a/paddle/memory/CMakeLists.txt
+++ b/paddle/memory/CMakeLists.txt
@@ -1,6 +1,6 @@
 add_subdirectory(detail)
 
-cc_library(memory SRCS memory.cc)
+cc_library(memory SRCS memory.cc DEPS place)
 cc_library(memcpy SRCS memcpy.cc)
 
 cc_library(paddle_memory
diff --git a/paddle/memory/detail/meta_cache.cc b/paddle/memory/detail/meta_cache.cc
index 30ff80e7bac0b595fe60aeab0a3c59f4e23eae2d..f0721c3b94b74eed3a02e4bc744c24b97ac170a9 100644
--- a/paddle/memory/detail/meta_cache.cc
+++ b/paddle/memory/detail/meta_cache.cc
@@ -13,6 +13,7 @@
    limitations under the License. */
 
 #include "paddle/memory/detail/meta_cache.h"
+#include "glog/logging.h"
 #include "paddle/memory/detail/memory_block.h"
 #include "paddle/platform/assert.h"
 
@@ -28,7 +29,9 @@ Metadata MetadataCache::load(const MemoryBlock* block) {
     PADDLE_ASSERT(existing_metadata->second.check_guards());
     return existing_metadata->second;
   } else {
-    PADDLE_ASSERT(reinterpret_cast<const Metadata*>(block)->check_guards());
+    auto* meta = reinterpret_cast<const Metadata*>(block);
+    VLOG(3) << "Load MetaData type=" << meta->type;
+    PADDLE_ASSERT(meta->check_guards());
     return *reinterpret_cast<const Metadata*>(block);
   }
 }
diff --git a/paddle/memory/memory.cc b/paddle/memory/memory.cc
index 8e561528f0e7e6ff524fc51b4776efc4e5bd28cd..0b648642f90a09db7452cce97eb04cedfcf55f4f 100644
--- a/paddle/memory/memory.cc
+++ b/paddle/memory/memory.cc
@@ -39,11 +39,15 @@ BuddyAllocator* GetCPUBuddyAllocator() {
 
 template <>
 void* Alloc<platform::CPUPlace>(platform::CPUPlace place, size_t size) {
-  return GetCPUBuddyAllocator()->Alloc(size);
+  VLOG(3) << "Allocate " << size << " bytes on " << platform::Place(place);
+  void* p = GetCPUBuddyAllocator()->Alloc(size);
+  VLOG(3) << "  pointer=" << p;
+  return p;
 }
 
 template <>
 void Free<platform::CPUPlace>(platform::CPUPlace place, void* p) {
+  VLOG(3) << "Free pointer=" << p << " on " << platform::Place(place);
   GetCPUBuddyAllocator()->Free(p);
 }
 
diff --git a/paddle/operators/sequence_pool_op.h b/paddle/operators/sequence_pool_op.h
index ead30e8e90b25165664b690491895ae68c8fc0ab..07bf61df45bf51c8648180ffc9eb97306865fab6 100644
--- a/paddle/operators/sequence_pool_op.h
+++ b/paddle/operators/sequence_pool_op.h
@@ -144,11 +144,11 @@ class SequencePoolGradKernel : public framework::OpKernel<T> {
           Eigen::Map<const Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic>>
               in_t_map(in_t.data<T>(), h, w);
           int row_id;
-          Eigen::array<int, 2> extents = {1, 1};
+          Eigen::array<int, 2> extents{{1, 1}};
           for (int col_id = 0; col_id < w; col_id++) {
             in_t_map.col(col_id).maxCoeff(&row_id);
-            Eigen::array<int, 2> in_offsets = {row_id, col_id};
-            Eigen::array<int, 2> out_offsets = {0, col_id};
+            Eigen::array<int, 2> in_offsets{{row_id, col_id}};
+            Eigen::array<int, 2> out_offsets{{0, col_id}};
             in_g_e.slice(in_offsets, extents).device(place) =
                 out_g_e.slice(out_offsets, extents);
           }
diff --git a/paddle/operators/sign_op.cc b/paddle/operators/sign_op.cc
new file mode 100644
index 0000000000000000000000000000000000000000..1b2f879d6d305e4e77be41683d8249904337a6f8
--- /dev/null
+++ b/paddle/operators/sign_op.cc
@@ -0,0 +1,70 @@
+/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+   http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License. */
+
+#include "paddle/operators/sign_op.h"
+
+namespace paddle {
+namespace operators {
+
+class SignOp : public framework::OperatorWithKernel {
+ public:
+  using framework::OperatorWithKernel::OperatorWithKernel;
+
+  void InferShape(framework::InferShapeContext *ctx) const override {
+    PADDLE_ENFORCE(ctx->HasInput("X"),
+                   "Input(X) of SignOp should not be null.");
+    PADDLE_ENFORCE(ctx->HasOutput("Out"),
+                   "Output(Out) of SignOp should not be null.");
+    ctx->SetOutputDim("Out", ctx->GetInputDim("X"));
+    ctx->ShareLoD("X", /*->*/ "Out");
+  }
+};
+
+template <typename AttrType>
+class SignOpMaker : public framework::OpProtoAndCheckerMaker {
+ public:
+  SignOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
+      : OpProtoAndCheckerMaker(proto, op_checker) {
+    AddInput("X", "(Tensor) Input tensor of sign operator.");
+    AddOutput("Out", "(Tensor) Output tensor of sign operator.");
+    AddComment(R"DOC(Sign operator
+
+The equation is: Out = X.sign()
+)DOC");
+  }
+};
+
+class SignGradMaker : public framework::SingleGradOpDescMaker {
+ public:
+  using framework::SingleGradOpDescMaker::SingleGradOpDescMaker;
+
+  std::unique_ptr<framework::OpDescBind> Apply() const override {
+    auto *grad_op = new framework::OpDescBind();
+    grad_op->SetType("scale");
+    grad_op->SetInput("X", OutputGrad("Out"));
+    grad_op->SetOutput("Out", InputGrad("X"));
+    grad_op->SetAttr("scale", 0.0f);
+    return std::unique_ptr<framework::OpDescBind>(grad_op);
+  }
+};
+
+}  // namespace operators
+}  // namespace paddle
+
+namespace ops = paddle::operators;
+
+REGISTER_OPERATOR(sign, ops::SignOp, ops::SignOpMaker<float>,
+                  ops::SignGradMaker);
+REGISTER_OP_CPU_KERNEL(sign,
+                       ops::SignKernel<paddle::platform::CPUPlace, float>);
diff --git a/paddle/operators/sign_op.cu b/paddle/operators/sign_op.cu
new file mode 100644
index 0000000000000000000000000000000000000000..4d0638cb97d84bf650fb23e4d2a201adc51a4b68
--- /dev/null
+++ b/paddle/operators/sign_op.cu
@@ -0,0 +1,18 @@
+/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+   http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License. */
+
+#include "paddle/operators/sign_op.h"
+
+REGISTER_OP_GPU_KERNEL(
+    sign, paddle::operators::SignKernel<paddle::platform::GPUPlace, float>);
diff --git a/paddle/operators/sign_op.h b/paddle/operators/sign_op.h
new file mode 100644
index 0000000000000000000000000000000000000000..ab5cd4bac019d602c63ea51629fb85fa7e206841
--- /dev/null
+++ b/paddle/operators/sign_op.h
@@ -0,0 +1,38 @@
+/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+   http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License. */
+
+#pragma once
+
+#include "paddle/framework/eigen.h"
+#include "paddle/framework/op_registry.h"
+
+namespace paddle {
+namespace operators {
+template <typename Place, typename T>
+class SignKernel : public framework::OpKernel<T> {
+ public:
+  virtual void Compute(const framework::ExecutionContext& context) const {
+    auto* out = context.Output<framework::Tensor>("Out");
+    auto* in = context.Input<framework::Tensor>("X");
+    out->mutable_data<T>(in->place());
+
+    auto eigen_out = framework::EigenVector<T>::Flatten(*out);
+    auto eigen_in = framework::EigenVector<T>::Flatten(*in);
+    auto& place = context.GetEigenDevice<Place>();
+    eigen_out.device(place) = eigen_in.sign();
+  }
+};
+
+}  // namespace operators
+}  // namespace paddle
diff --git a/python/paddle/v2/framework/tests/test_sign_op.py b/python/paddle/v2/framework/tests/test_sign_op.py
new file mode 100644
index 0000000000000000000000000000000000000000..c6b59bcfd8ba71e54d4c3a2b7a3dac1f2a346265
--- /dev/null
+++ b/python/paddle/v2/framework/tests/test_sign_op.py
@@ -0,0 +1,22 @@
+import unittest
+import numpy as np
+from op_test import OpTest
+
+
+class TestSignOp(OpTest):
+    def setUp(self):
+        self.op_type = "sign"
+        self.inputs = {
+            'X': np.random.uniform(-10, 10, (10, 10)).astype("float32")
+        }
+        self.outputs = {'Out': np.sign(self.inputs['X'])}
+
+    def test_check_output(self):
+        self.check_output()
+
+    def test_check_grad(self):
+        self.check_grad(['X'], 'Out')
+
+
+if __name__ == "__main__":
+    unittest.main()