fix batch_norm_xpu bug & remove xpusimulator dependence (#28430)

*test=kunlun

cmake/external/xpu.cmake

@@ -4,30 +4,26 @@ endif()
 
 INCLUDE(ExternalProject)
 SET(XPU_PROJECT                 "extern_xpu")
-SET(XPU_URL    "https://baidu-kunlun-public.su.bcebos.com/paddle_depence/xpu.tar.gz" CACHE STRING "" FORCE)
+SET(XPU_URL    "https://baidu-kunlun-public.su.bcebos.com/paddle_depence/xpu_2020_09_22_api_2020_11_05.tar.gz" CACHE STRING "" FORCE)
 SET(XPU_SOURCE_DIR              "${THIRD_PARTY_PATH}/xpu")
 SET(XPU_DOWNLOAD_DIR            "${XPU_SOURCE_DIR}/src/${XPU_PROJECT}")
 SET(XPU_INSTALL_DIR             "${THIRD_PARTY_PATH}/install/xpu")
-SET(XPU_API_INC_DIR             "${THIRD_PARTY_PATH}/install/xpu/api/include")
-SET(XPU_RUNTIME_INC_DIR         "${THIRD_PARTY_PATH}/install/xpu/runtime/include")
+SET(XPU_API_INC_DIR             "${THIRD_PARTY_PATH}/install/xpu/include")
 SET(XPU_LIB_DIR                 "${THIRD_PARTY_PATH}/install/xpu/lib")
 
 SET(XPU_API_LIB_NAME            "libxpuapi.so")
 SET(XPU_RT_LIB_NAME             "libxpurt.so")
-SET(XPU_SIM_LIB_NAME            "libxpusim.so")
 SET(XPU_API_LIB                 "${XPU_LIB_DIR}/${XPU_API_LIB_NAME}")
 SET(XPU_RT_LIB                  "${XPU_LIB_DIR}/${XPU_RT_LIB_NAME}")
-SET(XPU_SIM_LIB                 "${XPU_LIB_DIR}/${XPU_SIM_LIB_NAME}")
 
 SET(CMAKE_INSTALL_RPATH "${CMAKE_INSTALL_RPATH}" "${XPU_INSTALL_DIR}/lib")
 
 INCLUDE_DIRECTORIES(${XPU_API_INC_DIR})
-INCLUDE_DIRECTORIES(${XPU_RUNTIME_INC_DIR})
 
 FILE(WRITE ${XPU_DOWNLOAD_DIR}/CMakeLists.txt
   "PROJECT(XPU)\n"
   "cmake_minimum_required(VERSION 3.0)\n"
-  "install(DIRECTORY xpu/api xpu/runtime xpu/lib \n"
+  "install(DIRECTORY xpu/include xpu/lib \n"
   "        DESTINATION ${XPU_INSTALL_DIR})\n")
 
 ExternalProject_Add(
@@ -50,5 +46,5 @@ set_property(TARGET shared_xpuapi PROPERTY IMPORTED_LOCATION "${XPU_API_LIB}")
 # for cc_library(xxx SRCS xxx.c DEPS xpulib)
 generate_dummy_static_lib(LIB_NAME "xpulib" GENERATOR "xpu.cmake")
 
-TARGET_LINK_LIBRARIES(xpulib ${XPU_API_LIB} ${XPU_RT_LIB} ${XPU_SIM_LIB})
+TARGET_LINK_LIBRARIES(xpulib ${XPU_API_LIB} ${XPU_RT_LIB})
 ADD_DEPENDENCIES(xpulib ${XPU_PROJECT})

paddle/fluid/operators/batch_norm_op_xpu.cc

@@ -15,6 +15,7 @@ limitations under the License. */
 #ifdef PADDLE_WITH_XPU
 
 #include "paddle/fluid/operators/batch_norm_op.h"
+#include "xpu/refactor/nn.h"
 
 namespace paddle {
 namespace operators {
@@ -72,10 +73,10 @@ class BatchNormXPUKernel : public framework::OpKernel<T> {
       auto* variance_out_data = variance_out->data<T>();
       auto* saved_mean_data = saved_mean->data<T>();
       auto* saved_variance_data = saved_variance->data<T>();
-      int r = xpu::batch_norm_train_forward(
-          dev_ctx.x_context(), epsilon, momentum, N, C, H, W, x_data, y_data,
-          scale_data, bias_data, mean_out_data, variance_out_data,
-          saved_mean_data, saved_variance_data);
+      int r = xpu::batch_norm<T>(dev_ctx.x_context(), x_data, y_data, N, C, H,
+                                 W, epsilon, momentum, scale_data, bias_data,
+                                 saved_mean_data, saved_variance_data,
+                                 mean_out_data, variance_out_data, true);
       PADDLE_ENFORCE_EQ(
           r, XPU_SUCCESS,
           platform::errors::External("XPU API(batch_norm_train_forward) return "

python/setup.py.in

@@ -319,10 +319,8 @@ if '${WITH_XPU}' == 'ON':
                 raise Exception("patch ${XPU_API_LIB} failed, command: %s" % command)
     shutil.copy('${XPU_API_LIB}', libs_path)
     shutil.copy('${XPU_RT_LIB}', libs_path)
-    shutil.copy('${XPU_SIM_LIB}', libs_path)
     package_data['paddle.libs']+=['${XPU_API_LIB_NAME}',
-                                  '${XPU_RT_LIB_NAME}',
-                                  '${XPU_SIM_LIB_NAME}']
+                                  '${XPU_RT_LIB_NAME}']
 
 # copy libfuild_framework.so to libs
 if os.name != 'nt' and sys.platform != 'darwin':