diff --git a/paddle/fluid/inference/api/demo_ci/CMakeLists.txt b/paddle/fluid/inference/api/demo_ci/CMakeLists.txt
index d656cc1d4f41fdf7a8b391f320fe24e8f2ebfc08..f6c4bed79f2be3a66e15af85acc465d72ecdb17f 100644
--- a/paddle/fluid/inference/api/demo_ci/CMakeLists.txt
+++ b/paddle/fluid/inference/api/demo_ci/CMakeLists.txt
@@ -4,6 +4,7 @@ option(WITH_MKL        "Compile demo with MKL/OpenBlas support, default use MKL.
 option(WITH_GPU        "Compile demo with GPU/CPU, default use CPU."                    OFF)
 option(WITH_STATIC_LIB "Compile demo with static/shared library, default use static."   ON)
 option(USE_TENSORRT "Compile demo with TensorRT."   OFF)
+
 if(NOT WITH_STATIC_LIB)
   add_definitions("-DPADDLE_WITH_SHARED_LIB")
 endif()
@@ -44,6 +45,7 @@ link_directories("${PADDLE_LIB}/paddle/lib")
 
 if (WIN32)
   add_definitions("/DGOOGLE_GLOG_DLL_DECL=")
+  option(MSVC_STATIC_CRT "use static C Runtime library by default" ON)
   if (MSVC_STATIC_CRT)
     set(CMAKE_C_FLAGS_DEBUG   "${CMAKE_C_FLAGS_DEBUG} /bigobj /MTd")
     set(CMAKE_C_FLAGS_RELEASE  "${CMAKE_C_FLAGS_RELEASE} /bigobj /MT")
@@ -96,10 +98,8 @@ if(WITH_MKL)
   set(MATH_LIB_PATH "${PADDLE_LIB_THIRD_PARTY_PATH}mklml")
   include_directories("${MATH_LIB_PATH}/include")
   if(WIN32)
-    set(MATH_DLL ${MATH_LIB_PATH}/lib/mklml${CMAKE_SHARED_LIBRARY_SUFFIX}
-            ${MATH_LIB_PATH}/lib/libiomp5md${CMAKE_SHARED_LIBRARY_SUFFIX}
-            ${MATH_LIB_PATH}/lib/msvcr120${CMAKE_SHARED_LIBRARY_SUFFIX}
-            )
+    set(MATH_LIB ${MATH_LIB_PATH}/lib/mklml${CMAKE_STATIC_LIBRARY_SUFFIX}
+                 ${MATH_LIB_PATH}/lib/libiomp5md${CMAKE_STATIC_LIBRARY_SUFFIX})
   else()
     set(MATH_LIB ${MATH_LIB_PATH}/lib/libmklml_intel${CMAKE_SHARED_LIBRARY_SUFFIX}
                  ${MATH_LIB_PATH}/lib/libiomp5${CMAKE_SHARED_LIBRARY_SUFFIX})
@@ -114,20 +114,23 @@ if(WITH_MKL)
     endif(WIN32)
   endif()
 else()
+  set(OPENBLAS_LIB_PATH "${PADDLE_LIB_THIRD_PARTY_PATH}openblas")
+  include_directories("${OPENBLAS_LIB_PATH}/include/openblas")
   if(WIN32)
-    set(MATH_DLL ${PADDLE_LIB_THIRD_PARTY_PATH}openblas/lib/openblas${CMAKE_SHARED_LIBRARY_SUFFIX})
-    # Note: fix the openblas static library not work on windows, and change the static library to import library.
-    set(MATH_LIB ${PADDLE_LIB_THIRD_PARTY_PATH}openblas/lib/openblas${CMAKE_STATIC_LIBRARY_SUFFIX})
+    set(MATH_LIB ${OPENBLAS_LIB_PATH}/lib/openblas${CMAKE_STATIC_LIBRARY_SUFFIX})
   else()
-    set(MATH_LIB ${PADDLE_LIB_THIRD_PARTY_PATH}openblas/lib/libopenblas${CMAKE_STATIC_LIBRARY_SUFFIX})
+    set(MATH_LIB ${OPENBLAS_LIB_PATH}/lib/libopenblas${CMAKE_STATIC_LIBRARY_SUFFIX})
   endif()
 endif()
 
-# Note: libpaddle_inference_api.so/a must put before libpaddle_fluid.so/a
 if(WITH_STATIC_LIB)
   set(DEPS ${PADDLE_LIB}/paddle/lib/libpaddle_fluid${CMAKE_STATIC_LIBRARY_SUFFIX})
 else()
-  set(DEPS ${PADDLE_LIB}/paddle/lib/libpaddle_fluid${CMAKE_SHARED_LIBRARY_SUFFIX})
+  if(WIN32)
+    set(DEPS ${PADDLE_LIB}/paddle/lib/paddle_fluid${CMAKE_STATIC_LIBRARY_SUFFIX})
+  else()
+    set(DEPS ${PADDLE_LIB}/paddle/lib/libpaddle_fluid${CMAKE_SHARED_LIBRARY_SUFFIX})
+  endif()
 endif()
 
 if (NOT WIN32)
@@ -151,7 +154,7 @@ if(WITH_GPU)
     endif()
     set(DEPS ${DEPS} ${CUDA_LIB}/libcudart${CMAKE_SHARED_LIBRARY_SUFFIX})
   else()
-    if (USE_TENSORRT)
+    if(USE_TENSORRT)
       set(DEPS ${DEPS} ${TENSORRT_LIB_DIR}/nvinfer${CMAKE_STATIC_LIBRARY_SUFFIX})
       set(DEPS ${DEPS} ${TENSORRT_LIB_DIR}/nvinfer_plugin${CMAKE_STATIC_LIBRARY_SUFFIX})
     endif()
@@ -170,16 +173,23 @@ if(WIN32)
               ${CMAKE_BINARY_DIR}/${CMAKE_BUILD_TYPE}
             COMMAND ${CMAKE_COMMAND} -E copy ${TENSORRT_LIB_DIR}/nvinfer_plugin${CMAKE_SHARED_LIBRARY_SUFFIX}
               ${CMAKE_BINARY_DIR}/${CMAKE_BUILD_TYPE}
-            )
+    )
   endif()
   if(WITH_MKL)
     add_custom_command(TARGET ${DEMO_NAME} POST_BUILD
-          COMMAND ${CMAKE_COMMAND} -E copy ${MKLDNN_PATH}/lib/mkldnn.dll ${CMAKE_BINARY_DIR}/${CMAKE_BUILD_TYPE}
-          COMMAND ${CMAKE_COMMAND} -E copy ${MATH_DLL} ${CMAKE_BINARY_DIR}/${CMAKE_BUILD_TYPE}
-          )
-    else()
+          COMMAND ${CMAKE_COMMAND} -E copy ${MATH_LIB_PATH}/lib/mklml.dll ${CMAKE_BINARY_DIR}/Release
+          COMMAND ${CMAKE_COMMAND} -E copy ${MATH_LIB_PATH}/lib/libiomp5md.dll ${CMAKE_BINARY_DIR}/Release
+          COMMAND ${CMAKE_COMMAND} -E copy ${MATH_LIB_PATH}/lib/msvcr120.dll ${CMAKE_BINARY_DIR}/Release
+          COMMAND ${CMAKE_COMMAND} -E copy ${MKLDNN_PATH}/lib/mkldnn.dll  ${CMAKE_BINARY_DIR}/Release
+    )
+  else()
     add_custom_command(TARGET ${DEMO_NAME} POST_BUILD
-            COMMAND ${CMAKE_COMMAND} -E copy ${MATH_DLL} ${CMAKE_BINARY_DIR}/${CMAKE_BUILD_TYPE}
-            )
-    endif()
+          COMMAND ${CMAKE_COMMAND} -E copy ${OPENBLAS_LIB_PATH}/lib/openblas.dll ${CMAKE_BINARY_DIR}/Release
+    )
+  endif()
+  if(NOT WITH_STATIC_LIB)
+    add_custom_command(TARGET ${DEMO_NAME} POST_BUILD
+          COMMAND ${CMAKE_COMMAND} -E copy ${PADDLE_LIB}/paddle/lib/paddle_fluid${CMAKE_SHARED_LIBRARY_SUFFIX} ${CMAKE_BINARY_DIR}/Release
+    )
+  endif()
 endif()
diff --git a/paddle/fluid/inference/api/demo_ci/run.sh b/paddle/fluid/inference/api/demo_ci/run.sh
index 054f9de3d7e51097c9a8597d2e337dbc71c4ef7b..bfa273d4468dbb8e43995bfaadfa6dea932fd7c4 100755
--- a/paddle/fluid/inference/api/demo_ci/run.sh
+++ b/paddle/fluid/inference/api/demo_ci/run.sh
@@ -6,6 +6,7 @@ TEST_GPU_CPU=$3 # test both GPU/CPU mode or only CPU mode
 DATA_DIR=$4 # dataset
 TENSORRT_INCLUDE_DIR=$5 # TensorRT header file dir, default to /usr/local/TensorRT/include
 TENSORRT_LIB_DIR=$6 # TensorRT lib file dir, default to /usr/local/TensorRT/lib
+
 inference_install_dir=${PADDLE_ROOT}/build/fluid_inference_install_dir
 
 cd `dirname $0`
@@ -35,7 +36,7 @@ function download() {
   mkdir -p $dir_name
   cd $dir_name
   if [[ -e "${PREFIX}${dir_name}.tar.gz" ]]; then
-    echo "${PREFIX}{dir_name}.tar.gz has been downloaded."
+    echo "${PREFIX}${dir_name}.tar.gz has been downloaded."
   else
       wget -q ${URL_ROOT}$dir_name.tar.gz
       tar xzf *.tar.gz
@@ -49,78 +50,123 @@ for vis_demo_name in $vis_demo_list; do
   download $vis_demo_name
 done
 
+# download word2vec data
+mkdir -p word2vec
+cd word2vec
+if [[ -e "word2vec.inference.model.tar.gz" ]]; then
+  echo "word2vec.inference.model.tar.gz has been downloaded."
+else
+    wget -q http://paddle-inference-dist.bj.bcebos.com/word2vec.inference.model.tar.gz
+    tar xzf *.tar.gz
+fi
+
 # compile and test the demo
 cd $current_dir
 mkdir -p build
 cd build
+rm -rf *
 
-for WITH_STATIC_LIB in ON OFF; do
-# TODO(Superjomn) reopen this
-# something wrong with the TensorArray reset.
-:<<D
-  # -----simple_on_word2vec-----
-  rm -rf *
-  cmake .. -DPADDLE_LIB=${inference_install_dir} \
+if [ $(echo `uname` | grep "Win") != "" ]; then
+  # -----simple_on_word2vec on windows-----
+  cmake .. -G "Visual Studio 14 2015" -A x64 -DPADDLE_LIB=${inference_install_dir} \
     -DWITH_MKL=$TURN_ON_MKL \
     -DDEMO_NAME=simple_on_word2vec \
     -DWITH_GPU=$TEST_GPU_CPU \
-    -DWITH_STATIC_LIB=$WITH_STATIC_LIB
-  make -j
-  word2vec_model=$DATA_DIR'/word2vec/word2vec.inference.model'
-  if [ -d $word2vec_model ]; then
-    for use_gpu in $use_gpu_list; do
-      ./simple_on_word2vec \
-        --dirname=$word2vec_model \
-        --use_gpu=$use_gpu
-      if [ $? -ne 0 ]; then
-        echo "simple_on_word2vec demo runs fail."
-        exit 1
-      fi
-    done
+    -DWITH_STATIC_LIB=ON
+  msbuild  /maxcpucount /property:Configuration=Release cpp_inference_demo.sln
+  Release/simple_on_word2vec.exe \
+      --dirname=$DATA_DIR/word2vec/word2vec.inference.model \
+      --use_gpu=False
+  if [ $? -ne 0 ]; then
+    echo "simple_on_word2vec demo runs fail."
+    exit 1
   fi
-D
-  # ---------vis_demo---------
+
+  # -----vis_demo on windows-----
   rm -rf *
-  cmake .. -DPADDLE_LIB=${inference_install_dir} \
+  cmake .. -G "Visual Studio 14 2015" -A x64 -DPADDLE_LIB=${inference_install_dir} \
     -DWITH_MKL=$TURN_ON_MKL \
     -DDEMO_NAME=vis_demo \
     -DWITH_GPU=$TEST_GPU_CPU \
-    -DWITH_STATIC_LIB=$WITH_STATIC_LIB
-  make -j
-  for use_gpu in $use_gpu_list; do
-    for vis_demo_name in $vis_demo_list; do
-      ./vis_demo \
-        --modeldir=$DATA_DIR/$vis_demo_name/model \
-        --data=$DATA_DIR/$vis_demo_name/data.txt \
-        --refer=$DATA_DIR/$vis_demo_name/result.txt \
-        --use_gpu=$use_gpu
-      if [ $? -ne 0 ]; then
-        echo "vis demo $vis_demo_name runs fail."
-        exit 1
-      fi
-    done
+    -DWITH_STATIC_LIB=ON
+  msbuild  /maxcpucount /property:Configuration=Release cpp_inference_demo.sln
+  for vis_demo_name in $vis_demo_list; do
+    Release/vis_demo.exe \
+      --modeldir=$DATA_DIR/$vis_demo_name/model \
+      --data=$DATA_DIR/$vis_demo_name/data.txt \
+      --refer=$DATA_DIR/$vis_demo_name/result.txt \
+      --use_gpu=False
+    if [ $? -ne 0 ]; then
+      echo "vis demo $vis_demo_name runs fail."
+      exit 1
+    fi
   done
+else
+  for WITH_STATIC_LIB in ON OFF; do
+    # -----simple_on_word2vec on linux/mac-----
+    rm -rf *
+    cmake .. -DPADDLE_LIB=${inference_install_dir} \
+      -DWITH_MKL=$TURN_ON_MKL \
+      -DDEMO_NAME=simple_on_word2vec \
+      -DWITH_GPU=$TEST_GPU_CPU \
+      -DWITH_STATIC_LIB=$WITH_STATIC_LIB
+    make -j$(nproc)
+    word2vec_model=$DATA_DIR'/word2vec/word2vec.inference.model'
+    if [ -d $word2vec_model ]; then
+      for use_gpu in $use_gpu_list; do
+        ./simple_on_word2vec \
+          --dirname=$DATA_DIR/word2vec/word2vec.inference.model \
+          --use_gpu=$use_gpu
+        if [ $? -ne 0 ]; then
+          echo "simple_on_word2vec demo runs fail."
+          exit 1
+        fi
+      done
+    fi
 
-  # --------tensorrt mobilenet------
-  if [ $USE_TENSORRT == ON -a $TEST_GPU_CPU == ON ]; then
+    # ---------vis_demo on linux/mac---------
     rm -rf *
     cmake .. -DPADDLE_LIB=${inference_install_dir} \
       -DWITH_MKL=$TURN_ON_MKL \
-      -DDEMO_NAME=trt_mobilenet_demo \
+      -DDEMO_NAME=vis_demo \
       -DWITH_GPU=$TEST_GPU_CPU \
-      -DWITH_STATIC_LIB=$WITH_STATIC_LIB \
-      -DUSE_TENSORRT=$USE_TENSORRT \
-      -DTENSORRT_INCLUDE_DIR=$TENSORRT_INCLUDE_DIR \
-      -DTENSORRT_LIB_DIR=$TENSORRT_LIB_DIR
-    make -j
-    ./trt_mobilenet_demo \
-      --modeldir=$DATA_DIR/mobilenet/model \
-      --data=$DATA_DIR/mobilenet/data.txt \
-      --refer=$DATA_DIR/mobilenet/result.txt 
-    if [ $? -ne 0 ]; then
-      echo "trt demo trt_mobilenet_demo runs fail."
-      exit 1
+      -DWITH_STATIC_LIB=$WITH_STATIC_LIB
+    make -j$(nproc)
+    for use_gpu in $use_gpu_list; do
+      for vis_demo_name in $vis_demo_list; do
+        ./vis_demo \
+          --modeldir=$DATA_DIR/$vis_demo_name/model \
+          --data=$DATA_DIR/$vis_demo_name/data.txt \
+          --refer=$DATA_DIR/$vis_demo_name/result.txt \
+          --use_gpu=$use_gpu
+        if [ $? -ne 0 ]; then
+          echo "vis demo $vis_demo_name runs fail."
+          exit 1
+        fi
+      done
+    done
+
+    # --------tensorrt mobilenet on linux/mac------
+    if [ $USE_TENSORRT == ON -a $TEST_GPU_CPU == ON ]; then
+      rm -rf *
+      cmake .. -DPADDLE_LIB=${inference_install_dir} \
+        -DWITH_MKL=$TURN_ON_MKL \
+        -DDEMO_NAME=trt_mobilenet_demo \
+        -DWITH_GPU=$TEST_GPU_CPU \
+        -DWITH_STATIC_LIB=$WITH_STATIC_LIB \
+        -DUSE_TENSORRT=$USE_TENSORRT \
+        -DTENSORRT_INCLUDE_DIR=$TENSORRT_INCLUDE_DIR \
+        -DTENSORRT_LIB_DIR=$TENSORRT_LIB_DIR
+      make -j$(nproc)
+      ./trt_mobilenet_demo \
+        --modeldir=$DATA_DIR/mobilenet/model \
+        --data=$DATA_DIR/mobilenet/data.txt \
+        --refer=$DATA_DIR/mobilenet/result.txt 
+      if [ $? -ne 0 ]; then
+        echo "trt demo trt_mobilenet_demo runs fail."
+        exit 1
+      fi
     fi
-  fi
-done
+  done
+fi
 set +x