diff --git a/paddle/framework/CMakeLists.txt b/paddle/framework/CMakeLists.txt
index 2af10a996c1ff2829d7b262809a0a78095ffeef9..46dce7d1d273540d94359cbe28c0679824a041ea 100644
--- a/paddle/framework/CMakeLists.txt
+++ b/paddle/framework/CMakeLists.txt
@@ -12,7 +12,11 @@ else()
 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
index 293c65a065402adb00571e0651ed1091384789ba..7efc649d0bcda67c663d148e83bcbb6789b0f371 100644
--- a/paddle/framework/tensor_util.cc
+++ b/paddle/framework/tensor_util.cc
@@ -31,6 +31,7 @@ struct AnyDTypeVisitor {
   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();
   }
 };
@@ -66,9 +67,10 @@ struct AnyVisitor : public boost::static_visitor<bool> {
     framework::Tensor tmp;
     tmp.Resize({1});
     tmp.mutable_data<bool>(cpu);
-    platform::DeviceContextPool::Instance().Get(gpu)->Wait();
-    CopyFrom(out, cpu, &tmp);
-    platform::DeviceContextPool::Instance().Get(gpu)->Wait();
+    auto gpuctx = platform::DeviceContextPool::Instance().Get(gpu);
+    gpuctx->Wait();
+    CopyFrom(out, cpu, *gpuctx, &tmp);
+    gpuctx->Wait();
     return GetResult(tmp, cpu);
   }
 
@@ -89,6 +91,7 @@ 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();
   }
 };
@@ -102,6 +105,7 @@ 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();
   }
 };
diff --git a/paddle/framework/tensor_util.h b/paddle/framework/tensor_util.h
index e71d8e5672a3323ce789afa2d575517dc596b12c..784170dae3f475b99dab3943c05a6496a0cdcd21 100644
--- a/paddle/framework/tensor_util.h
+++ b/paddle/framework/tensor_util.h
@@ -208,7 +208,10 @@ 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);
 
 }  // namespace framework
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