diff --git a/CMakeLists.txt b/CMakeLists.txt
index 4613155f7700b25b2a8d7c250832722085b332fa..8ce51bb0f4109f596840b4b4cadf8d01525f6dba 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -98,6 +98,8 @@ else()
         add_definitions(-DPADDLE_USE_DSO)
     endif(WITH_DSO)
 
+    set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-g -O3 --use_fast_math")
+
     # Include cuda and cudnn
     include_directories(${CUDNN_INCLUDE_DIR})
     include_directories(${CUDA_TOOLKIT_INCLUDE})
@@ -117,6 +119,9 @@ endif(NOT WITH_TIMER)
 if(WITH_AVX)
     set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${AVX_FLAGS}")
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${AVX_FLAGS}")
+    if(AVX_FOUND)
+        set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xcompiler -mavx")
+    endif(AVX_FOUND)
 else(WITH_AVX)
     set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -msse3")
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -msse3")
diff --git a/cmake/util.cmake b/cmake/util.cmake
index 0fa36f070cc11be543efe9573b93173ec771b9be..74227bf14e37c8560bf5a19aeb716ff8e41cf242 100644
--- a/cmake/util.cmake
+++ b/cmake/util.cmake
@@ -178,13 +178,6 @@ macro(add_simple_unittest TARGET_NAME)
     add_unittest(${TARGET_NAME} ${TARGET_NAME}.cpp)
 endmacro()
 
-macro(add_paddle_culib TARGET_NAME)
-    set(NVCC_FLAG ${CUDA_NVCC_FLAGS})
-    set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS};--use_fast_math)
-    cuda_add_library(${TARGET_NAME} STATIC ${ARGN})
-    set(CUDA_NVCC_FLAGS ${NVCC_FLAG})
-endmacro()
-
 
 # Creates C resources file from files in given resource file
 function(create_resources res_file output)
diff --git a/paddle/gserver/CMakeLists.txt b/paddle/gserver/CMakeLists.txt
index 9ac4d210f6d376639df20800b6782f1f8c03d6aa..a066f80c221ee8ab4383ee6463f7b111984b58ff 100644
--- a/paddle/gserver/CMakeLists.txt
+++ b/paddle/gserver/CMakeLists.txt
@@ -50,7 +50,7 @@ if(NOT WITH_PYTHON)
 endif()
 
 if(WITH_GPU)
-    add_paddle_culib(paddle_gserver ${GSERVER_SOURCES})
+    cuda_add_library(paddle_gserver ${GSERVER_SOURCES})
 else()
     add_library(paddle_gserver STATIC
         ${GSERVER_SOURCES})
diff --git a/paddle/math/CMakeLists.txt b/paddle/math/CMakeLists.txt
index 1bb271ba80e1aa899f967afe27d6f5b8cf71eecd..f5657c4690ca71200346efd4e2c5244c02c92eb1 100644
--- a/paddle/math/CMakeLists.txt
+++ b/paddle/math/CMakeLists.txt
@@ -25,7 +25,7 @@ if(NOT WITH_GPU)
     add_library(paddle_math STATIC
         ${MATH_SOURCES})
 else()
-    add_paddle_culib(paddle_math ${MATH_SOURCES})
+    cuda_add_library(paddle_math ${MATH_SOURCES})
 endif()
 
 
diff --git a/paddle/math/tests/CMakeLists.txt b/paddle/math/tests/CMakeLists.txt
index 07fd1adc1be9c969ca3500cd4e8ce5275b5b5956..ccc3deac7be8708c8dd19b6e3404ba43cb052401 100644
--- a/paddle/math/tests/CMakeLists.txt
+++ b/paddle/math/tests/CMakeLists.txt
@@ -15,7 +15,7 @@ add_simple_unittest(test_perturbation)
 add_simple_unittest(test_CpuGpuVector)
 add_simple_unittest(test_Allocator)
 if(COMPILER_SUPPORT_CXX11)
-  LIST(APPEND CUDA_NVCC_FLAGS -std=c++11 -Xcompiler -fPIC --use_fast_math)
+  LIST(APPEND CUDA_NVCC_FLAGS -std=c++11)
   CUDA_ADD_EXECUTABLE(test_Tensor test_Tensor.cu)
   link_paddle_test(test_Tensor)
 endif()
diff --git a/paddle/math/tests/test_TrainingAlgorithm.cpp b/paddle/math/tests/test_TrainingAlgorithm.cpp
index b465ae6c26264317515dfdab027305fa391c1834..1759d221e51de121f304ba01dcc9ee26c78e1800 100644
--- a/paddle/math/tests/test_TrainingAlgorithm.cpp
+++ b/paddle/math/tests/test_TrainingAlgorithm.cpp
@@ -81,7 +81,7 @@ int VectorCheckErr(const VectorPtr& vector1, const VectorPtr& vector2) {
 
 #else
 
-#include "paddle/common/Stat.h"
+#include "paddle/utils/Stat.h"
 
 #define CHECK_VECTORPTR(vector1, vector2)