Merge pull request #1192 from zhangjun/cherrypick-0.6.0

Cherrypick 1179 1182 to  0.6.0

cmake/paddlepaddle.cmake

@@ -136,8 +136,10 @@ LINK_DIRECTORIES(${PADDLE_INSTALL_DIR}/third_party/install/mklml/lib)
 SET(CMAKE_INSTALL_RPATH "${CMAKE_INSTALL_RPATH}" "${PADDLE_INSTALL_DIR}/third_party/install/mkldnn/lib")
 LINK_DIRECTORIES(${PADDLE_INSTALL_DIR}/third_party/install/mkldnn/lib)
 
-ADD_LIBRARY(openblas STATIC IMPORTED GLOBAL)
-SET_PROPERTY(TARGET openblas PROPERTY IMPORTED_LOCATION ${PADDLE_INSTALL_DIR}/third_party/install/openblas/lib/libopenblas.a)
+if (NOT WITH_MKLML)
+    ADD_LIBRARY(openblas STATIC IMPORTED GLOBAL)
+    SET_PROPERTY(TARGET openblas PROPERTY IMPORTED_LOCATION ${PADDLE_INSTALL_DIR}/third_party/install/openblas/lib/libopenblas.a)
+endif()
 
 ADD_LIBRARY(paddle_inference STATIC IMPORTED GLOBAL)
 SET_PROPERTY(TARGET paddle_inference PROPERTY IMPORTED_LOCATION ${PADDLE_INSTALL_DIR}/lib/libpaddle_inference.a)

doc/BAIDU_KUNLUN_XPU_SERVING.md

@@ -22,7 +22,7 @@ cmake -DPYTHON_INCLUDE_DIR=/usr/include/python3.7m/ \
     -DSERVER=ON ..
 make -j10
 ```
-You can run `make install` to produce the target in `./output` directory. Add `-DCMAKE_INSTALL_PREFIX=./output` to specify the output path to CMake command shown above。
+You can run `make install` to produce the target in `./output` directory. Add `-DCMAKE_INSTALL_PREFIX=./output` to specify the output path to CMake command shown above. Please specify `-DWITH_MKL=ON` on Intel CPU platform with AVX2 support.  
 * Compile the Serving Client
 ```
 mkdir -p client-build-arm && cd client-build-arm

python/examples/xpu/bert/README.md

@@ -26,7 +26,7 @@ this script will download Chinese Dictionary File vocab.txt and Chinese Sample D
 ### Start Service
 
 ```
-python3 bert_web_service.py serving_server 7703
+python3 -m paddle_serving_server.serve --model serving_server --port 7703 --use_lite --use_xpu --ir_optim
 ```
 
 ### Client Prediction

python/examples/xpu/bert/bert_client.py

@@ -31,7 +31,7 @@ client.connect(endpoint_list)
 for line in sys.stdin:
     feed_dict = reader.process(line)
     for key in feed_dict.keys():
-        feed_dict[key] = np.array(feed_dict[key]).reshape((128, 1))
+        feed_dict[key] = np.array(feed_dict[key]).reshape((1, 128))
     #print(feed_dict)
-    result = client.predict(feed=feed_dict, fetch=fetch, batch=False)
+    result = client.predict(feed=feed_dict, fetch=fetch, batch=True)
 print(result)

python/examples/xpu/ernie/ernie_client.py

@@ -23,7 +23,7 @@ args = benchmark_args()
 
 reader = ChineseErnieReader({"max_seq_len": 128})
 fetch = ["save_infer_model/scale_0"]
-endpoint_list = ['127.0.0.1:12000']
+endpoint_list = ['127.0.0.1:7704']
 client = Client()
 client.load_client_config(args.model)
 client.connect(endpoint_list)