diff --git a/paddle/framework/CMakeLists.txt b/paddle/framework/CMakeLists.txt
index 8bfa41715ffa82c8f6fb321e1a562649d429672f..6788cb34fbaf5941cbb1537c7a83577c623bf76a 100644
--- a/paddle/framework/CMakeLists.txt
+++ b/paddle/framework/CMakeLists.txt
@@ -5,10 +5,18 @@ cc_library(ddim SRCS ddim.cc DEPS eigen3)
 cc_test(ddim_test SRCS ddim_test.cc DEPS ddim)
 nv_test(dim_test SRCS dim_test.cu DEPS ddim)
 
-cc_library(tensor SRCS tensor.cc DEPS ddim place paddle_memory device_context framework_proto)
+if (WITH_GPU)
+  nv_library(tensor SRCS tensor.cc tensor_util.cu DEPS ddim place paddle_memory device_context framework_proto)
+else()
+  cc_library(tensor SRCS tensor.cc tensor_util.cc DEPS ddim place paddle_memory device_context framework_proto)
+endif ()
 
 cc_test(tensor_test SRCS tensor_test.cc DEPS tensor)
-cc_test(tensor_util_test SRCS tensor_util_test.cc DEPS tensor)
+if (WITH_GPU)
+  nv_test(tensor_util_test SRCS tensor_util_test.cc tensor_util_test.cu DEPS tensor)
+else()
+  cc_test(tensor_util_test SRCS tensor_util_test.cc DEPS tensor)
+endif()
 
 cc_test(eigen_test SRCS eigen_test.cc DEPS tensor)
 
diff --git a/paddle/framework/tensor_util.cc b/paddle/framework/tensor_util.cc
new file mode 100644
index 0000000000000000000000000000000000000000..7efc649d0bcda67c663d148e83bcbb6789b0f371
--- /dev/null
+++ b/paddle/framework/tensor_util.cc
@@ -0,0 +1,119 @@
+/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
+
+   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/framework/tensor_util.h"
+
+namespace paddle {
+namespace framework {
+template <typename Predicate, typename DevCtx>
+struct AnyDTypeVisitor {
+  Predicate predicate_;
+  const Tensor& tensor_;
+  const DevCtx& ctx_;
+  Tensor* out_;
+
+  AnyDTypeVisitor(Predicate predicate, const Tensor& tensor, const DevCtx& ctx,
+                  Tensor* out)
+      : predicate_(predicate), tensor_(tensor), ctx_(ctx), out_(out) {}
+
+  template <typename T>
+  void operator()() const {
+    auto t = EigenVector<T>::Flatten(tensor_);
+    auto o = EigenScalar<bool>::From(*out_);
+    // return any of predicate_(t) is true.
+    o.device(*ctx_.eigen_device()) = predicate_(t).any();
+  }
+};
+
+template <typename Predicate, typename DevCtx>
+inline void AnyImpl(Predicate predicate, const framework::Tensor& tensor,
+                    const DevCtx& ctx, framework::Tensor* out) {
+  VisitDataType(ToDataType(tensor.type()), AnyDTypeVisitor<Predicate, DevCtx>(
+                                               predicate, tensor, ctx, out));
+}
+
+template <typename Predicate>
+struct AnyVisitor : public boost::static_visitor<bool> {
+  const framework::Tensor& tensor_;
+  Predicate predicate_;
+
+  AnyVisitor(const framework::Tensor& tensor, Predicate predicate)
+      : tensor_(tensor), predicate_(std::move(predicate)) {}
+
+  template <typename Place>
+  bool operator()(const Place& place) const {
+    framework::Tensor out;
+    out.Resize({1});
+    out.mutable_data<bool>(place);
+    auto* ctx = platform::DeviceContextPool::Instance().GetByPlace(place);
+    AnyImpl(predicate_, tensor_, *ctx, &out);
+    return this->GetResult(out, place);
+  }
+
+  bool GetResult(const framework::Tensor& out,
+                 const platform::CUDAPlace& gpu) const {
+    platform::CPUPlace cpu;
+    framework::Tensor tmp;
+    tmp.Resize({1});
+    tmp.mutable_data<bool>(cpu);
+    auto gpuctx = platform::DeviceContextPool::Instance().Get(gpu);
+    gpuctx->Wait();
+    CopyFrom(out, cpu, *gpuctx, &tmp);
+    gpuctx->Wait();
+    return GetResult(tmp, cpu);
+  }
+
+  bool GetResult(const framework::Tensor& out,
+                 const platform::CPUPlace& cpu) const {
+    return *out.data<bool>();
+  }
+};
+
+template <typename Predicate>
+inline bool Any(const framework::Tensor& tensor, Predicate predicate) {
+  AnyVisitor<Predicate> visitor(tensor, predicate);
+  auto place = tensor.place();
+  return platform::VisitPlace(place, visitor);
+}
+
+struct HasNANPredicate {
+  template <typename T>
+  auto operator()(const T& eigen_vec) const
+      -> decltype(std::declval<T>().isnan()) {
+    // Cast eigen_vector to vector of bool. true if is inf.
+    return eigen_vec.isnan();
+  }
+};
+
+bool HasNAN(const framework::Tensor& tensor) {
+  HasNANPredicate predicate;
+  return Any(tensor, predicate);
+}
+
+struct HasInfPredicate {
+  template <typename T>
+  auto operator()(const T& eigen_vec) const
+      -> decltype(std::declval<T>().isinf()) {
+    // Cast eigen_vector to vector of bool. true if is inf.
+    return eigen_vec.isinf();
+  }
+};
+
+bool HasInf(const framework::Tensor& tensor) {
+  HasInfPredicate predicate;
+  return Any(tensor, predicate);
+}
+
+}  // namespace framework
+}  // namespace paddle
diff --git a/paddle/framework/tensor_util.cu b/paddle/framework/tensor_util.cu
new file mode 120000
index 0000000000000000000000000000000000000000..b00e6e59d93328bf3142597ea4de0dc225501e56
--- /dev/null
+++ b/paddle/framework/tensor_util.cu
@@ -0,0 +1 @@
+./tensor_util.cc
\ No newline at end of file
diff --git a/paddle/framework/tensor_util.h b/paddle/framework/tensor_util.h
index 108006911a5f30691a01c6579a62c8111b653986..a86fab29250fd4df45a3ddd075e1f4c3a332fb42 100644
--- a/paddle/framework/tensor_util.h
+++ b/paddle/framework/tensor_util.h
@@ -14,8 +14,10 @@ limitations under the License. */
 
 #pragma once
 #include "paddle/framework/data_type.h"
+#include "paddle/framework/eigen.h"
 #include "paddle/framework/framework.pb.h"
 #include "paddle/framework/tensor.h"
+#include "paddle/platform/device_context.h"
 
 namespace paddle {
 namespace framework {
@@ -207,6 +209,12 @@ inline void CopyToVector(const Tensor& src, std::vector<T>* dst) {
                src_ptr, size);
 }
 
+// Returns true if a tensor contains NAN, i.e., Not A Number.
+extern bool HasNAN(const framework::Tensor& tensor);
+
+// Returns true if a tensor contains Inf, i.e., Infinity.
+extern bool HasInf(const framework::Tensor& tensor);
+
 inline void SerializeToStream(std::ostream& os, const Tensor& tensor,
                               const platform::DeviceContext& dev_ctx) {
   // TODO(typhoonzero): serialize to ostream
diff --git a/paddle/framework/tensor_util_test.cc b/paddle/framework/tensor_util_test.cc
index 1281e9c2a45858810d505aca6b2a2587e7a7280c..f00ce795488bd4279a6ad9e537f32d152e6131e4 100644
--- a/paddle/framework/tensor_util_test.cc
+++ b/paddle/framework/tensor_util_test.cc
@@ -13,6 +13,7 @@
 
 #include "paddle/framework/tensor_util.h"
 #include <gtest/gtest.h>
+#include <cmath>
 #include <string>
 
 namespace paddle {
@@ -230,6 +231,29 @@ TEST(CopyToVector, Tensor) {
 #endif
 }
 
+TEST(IsNAN, CPU) {
+  using namespace paddle::framework;
+  using namespace paddle::platform;
+  Tensor src;
+  float* buf = src.mutable_data<float>({3}, CPUPlace());
+  buf[0] = 0.0;
+  buf[1] = NAN;
+  buf[2] = 0.0;
+
+  ASSERT_TRUE(HasNAN(src));
+}
+
+TEST(IsInf, CPU) {
+  using namespace paddle::framework;
+  using namespace paddle::platform;
+  Tensor src;
+  double* buf = src.mutable_data<double>({3}, CPUPlace());
+  buf[0] = 1.0;
+  buf[1] = INFINITY;
+  buf[2] = 0.0;
+  ASSERT_TRUE(HasInf(src));
+}
+
 TEST(Tensor, SerializeAndDeserialize) {
   framework::Tensor src_tensor;
   int array[6] = {1, 2, 3, 4, 5, 6};
diff --git a/paddle/framework/tensor_util_test.cu b/paddle/framework/tensor_util_test.cu
new file mode 100644
index 0000000000000000000000000000000000000000..ebd35fdf6c2a1388fec23057070f723c8ef9da9c
--- /dev/null
+++ b/paddle/framework/tensor_util_test.cu
@@ -0,0 +1,57 @@
+/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
+
+   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 "gtest/gtest.h"
+#include "paddle/framework/tensor_util.h"
+#include "paddle/platform/device_context.h"
+#include "paddle/platform/place.h"
+
+namespace paddle {
+namespace framework {
+
+static __global__ void FillNAN(float* buf) {
+  buf[0] = 0.0;
+  buf[1] = 0.1;
+  buf[2] = NAN;
+}
+static __global__ void FillInf(float* buf) {
+  buf[0] = 0.0;
+  buf[1] = INFINITY;
+  buf[2] = 0.5;
+}
+
+TEST(HasNAN, GPU) {
+  Tensor tensor;
+  platform::CUDAPlace gpu(0);
+  auto& pool = platform::DeviceContextPool::Instance();
+  auto* cuda_ctx = pool.GetByPlace(gpu);
+  float* buf = tensor.mutable_data<float>({3}, gpu);
+  FillNAN<<<1, 1, 0, cuda_ctx->stream()>>>(buf);
+  cuda_ctx->Wait();
+  ASSERT_TRUE(HasNAN(tensor));
+}
+
+TEST(HasInf, GPU) {
+  Tensor tensor;
+  platform::CUDAPlace gpu(0);
+  auto& pool = platform::DeviceContextPool::Instance();
+  auto* cuda_ctx = pool.GetByPlace(gpu);
+  float* buf = tensor.mutable_data<float>({3}, gpu);
+  FillInf<<<1, 1, 0, cuda_ctx->stream()>>>(buf);
+  cuda_ctx->Wait();
+  ASSERT_TRUE(HasInf(tensor));
+}
+
+}  // namespace framework
+}  // namespace paddle
diff --git a/paddle/platform/device_context.h b/paddle/platform/device_context.h
index dfef2c16d8f2277d57cbcfe51d108402e518799b..2b366e6383d23e2d31a194edd04412892a8311eb 100644
--- a/paddle/platform/device_context.h
+++ b/paddle/platform/device_context.h
@@ -52,6 +52,14 @@ class CPUDeviceContext : public DeviceContext {
   std::unique_ptr<Eigen::DefaultDevice> eigen_device_;
 };
 
+template <typename Place>
+struct DefaultDeviceContextType;
+
+template <>
+struct DefaultDeviceContextType<platform::CPUPlace> {
+  using TYPE = CPUDeviceContext;
+};
+
 #ifdef PADDLE_WITH_CUDA
 
 class EigenCudaStreamDevice;
@@ -90,6 +98,11 @@ class CUDADeviceContext : public DeviceContext {
   cublasHandle_t cublas_handle_;
 };
 
+template <>
+struct DefaultDeviceContextType<platform::CUDAPlace> {
+  using TYPE = CUDADeviceContext;
+};
+
 class CUDNNDeviceContext : public CUDADeviceContext {
  public:
   explicit CUDNNDeviceContext(CUDAPlace place);
@@ -125,6 +138,13 @@ class DeviceContextPool {
   /*! \brief  Return handle of single device context. */
   const platform::DeviceContext* Get(const platform::Place& place);
 
+  template <typename Place>
+  const typename DefaultDeviceContextType<Place>::TYPE* GetByPlace(
+      const Place& place) {
+    return reinterpret_cast<
+        const typename DefaultDeviceContextType<Place>::TYPE*>(Get(place));
+  }
+
  private:
   static DeviceContextPool* pool;
   constexpr static int LEFT_SHIFT = 8;
diff --git a/paddle/platform/place.h b/paddle/platform/place.h
index d25eaa689f4a4baa951db5c61bbf99288e365ee1..76b5c502cc48431a4e9b13b07505978884576e1d 100644
--- a/paddle/platform/place.h
+++ b/paddle/platform/place.h
@@ -15,7 +15,7 @@ limitations under the License. */
 #pragma once
 
 #include <iostream>
-
+#include "paddle/platform/enforce.h"
 #include "paddle/platform/variant.h"
 
 namespace paddle {
@@ -64,5 +64,31 @@ bool places_are_same_class(const Place &, const Place &);
 
 std::ostream &operator<<(std::ostream &, const Place &);
 
+template <typename Visitor>
+struct PlaceVisitorWrapper
+    : public boost::static_visitor<typename Visitor::result_type> {
+  const Visitor &visitor_;
+  explicit PlaceVisitorWrapper(const Visitor &visitor) : visitor_(visitor) {}
+
+  typename Visitor::result_type operator()(const CPUPlace &cpu) const {
+    return visitor_(cpu);
+  }
+
+  typename Visitor::result_type operator()(const CUDAPlace &cuda) const {
+#ifdef PADDLE_WITH_CUDA
+    return visitor_(cuda);
+#else
+    PADDLE_THROW("Paddle is not compiled with CUDA. Cannot visit cuda device");
+    return typename Visitor::result_type();
+#endif
+  }
+};
+
+template <typename Visitor>
+typename Visitor::result_type VisitPlace(const Place &place,
+                                         const Visitor &visitor) {
+  return boost::apply_visitor(PlaceVisitorWrapper<Visitor>(visitor), place);
+}
+
 }  // namespace platform
 }  // namespace paddle