[xpu] support paddle compile with XFT (#51082)

CMakeLists.txt

@@ -52,6 +52,7 @@ option(WITH_MPI "Compile PaddlePaddle with MPI" OFF)
 option(WITH_TENSORRT "Compile PaddlePaddle with NVIDIA TensorRT" OFF)
 option(WITH_XPU "Compile PaddlePaddle with BAIDU KUNLUN XPU" OFF)
 option(WITH_XPU_KP "Compile PaddlePaddle with BAIDU XPU compiler " OFF)
+option(WITH_XPU_XFT "Compile PaddlePaddle with BAIDU XPU-XFT" OFF)
 option(WITH_MLU "Compile PaddlePaddle with CAMBRICON MLU" OFF)
 option(WITH_WIN_DUMP_DBG "Compile with windows core dump debug mode" OFF)
 option(WITH_ASCEND "Compile PaddlePaddle with ASCEND" OFF)
@@ -74,6 +75,9 @@ endif()
 if(WITH_GPU AND WITH_XPU_KP)
   message(FATAL_ERROR "Error when compile GPU and XPU2 at the same time")
 endif()
+if(WITH_GPU AND WITH_XPU_XFT)
+  message(FATAL_ERROR "Error when compile GPU and XPU-XFT at the same time")
+endif()
 if(WITH_GPU AND WITH_ASCEND)
   message(FATAL_ERROR "Error when compile GPU and ASCEND at the same time")
 endif()
@@ -419,6 +423,15 @@ if(WITH_XPU_KP AND NOT WITH_XPU)
       CACHE STRING "Enable WITH_XPU when compiling with WITH_XPU_KP" FORCE)
 endif()
 
+if(NOT WITH_XPU AND WITH_XPU_XFT)
+  message(
+    WARNING
+      "Enable WITH_XPU when compiling with WITH_XPU_XFT. Force WITH_XPU=ON.")
+  set(WITH_XPU
+      ON
+      CACHE STRING "Enable WITH_XPU when compiling with WITH_XPU_XFT" FORCE)
+endif()
+
 if(NOT WITH_XPU AND WITH_XPU_BKCL)
   message(
     WARNING "Disable BKCL when compiling without XPU. Force WITH_XPU_BKCL=OFF.")

cmake/external/xpu.cmake

@@ -6,9 +6,11 @@ include(ExternalProject)
 set(XPU_PROJECT "extern_xpu")
 set(XPU_API_LIB_NAME "libxpuapi.so")
 set(XPU_RT_LIB_NAME "libxpurt.so")
+set(XPU_XFT_LIB_NAME "libxft.so")
 
 set(XPU_BASE_DATE "20230227")
 set(XPU_XCCL_BASE_VERSION "1.0.10")
+set(XPU_XFT_BASE_VERSION "latest")
 
 if(NOT DEFINED XPU_BASE_URL)
   set(XPU_BASE_URL_WITHOUT_DATE
@@ -22,30 +24,42 @@ set(XPU_XCCL_BASE_URL
     "https://klx-sdk-release-public.su.bcebos.com/xccl/release/${XPU_XCCL_BASE_VERSION}"
 )
 
+if(NOT XPU_XFT_BASE_URL)
+  set(XPU_XFT_BASE_URL
+      "https://klx-sdk-release-public.su.bcebos.com/xft/dev/${XPU_XFT_BASE_VERSION}"
+  )
+endif()
+
 if(WITH_AARCH64)
   set(XPU_XRE_DIR_NAME "xre-kylin_aarch64")
   set(XPU_XDNN_DIR_NAME "xdnn-kylin_aarch64")
   set(XPU_XCCL_DIR_NAME "xccl-kylin_aarch64")
+  set(XPU_XFT_DIR_NAME "") # TODO: xft has no kylin output at now.
 elseif(WITH_SUNWAY)
   set(XPU_XRE_DIR_NAME "xre-deepin_sw6_64")
   set(XPU_XDNN_DIR_NAME "xdnn-deepin_sw6_64")
   set(XPU_XCCL_DIR_NAME "xccl-deepin_sw6_64")
+  set(XPU_XFT_DIR_NAME "") # TODO: xft has no deepin output at now.
 elseif(WITH_BDCENTOS)
   set(XPU_XRE_DIR_NAME "xre-bdcentos_x86_64")
   set(XPU_XDNN_DIR_NAME "xdnn-bdcentos_x86_64")
   set(XPU_XCCL_DIR_NAME "xccl-bdcentos_x86_64")
+  set(XPU_XFT_DIR_NAME "xft_bdcentos6u3_x86_64_gcc82")
 elseif(WITH_UBUNTU)
   set(XPU_XRE_DIR_NAME "xre-ubuntu_x86_64")
   set(XPU_XDNN_DIR_NAME "xdnn-ubuntu_x86_64")
   set(XPU_XCCL_DIR_NAME "xccl-ubuntu_x86_64")
+  set(XPU_XFT_DIR_NAME "xft_ubuntu1604_x86_64")
 elseif(WITH_CENTOS)
   set(XPU_XRE_DIR_NAME "xre-centos7_x86_64")
   set(XPU_XDNN_DIR_NAME "xdnn-centos7_x86_64")
   set(XPU_XCCL_DIR_NAME "xccl-bdcentos_x86_64")
+  set(XPU_XFT_DIR_NAME "xft_bdcentos6u3_x86_64_gcc82")
 else()
   set(XPU_XRE_DIR_NAME "xre-ubuntu_x86_64")
   set(XPU_XDNN_DIR_NAME "xdnn-ubuntu_x86_64")
   set(XPU_XCCL_DIR_NAME "xccl-ubuntu_x86_64")
+  set(XPU_XFT_DIR_NAME "xft_ubuntu1604_x86_64")
 endif()
 
 set(XPU_XRE_URL
@@ -57,12 +71,16 @@ set(XPU_XDNN_URL
 set(XPU_XCCL_URL
     "${XPU_XCCL_BASE_URL}/${XPU_XCCL_DIR_NAME}.tar.gz"
     CACHE STRING "" FORCE)
+set(XPU_XFT_URL "${XPU_XFT_BASE_URL}/${XPU_XFT_DIR_NAME}.tar.gz")
 set(XPU_PACK_DEPENCE_URL
     "https://baidu-kunlun-public.su.bcebos.com/paddle_depence/pack_paddle_depence.sh"
     CACHE STRING "" FORCE)
 set(XPU_CHECK_DEPENCE_URL
     "https://baidu-kunlun-public.su.bcebos.com/paddle_depence/check_xpu_dependence.sh"
     CACHE STRING "" FORCE)
+set(XPU_XFT_GET_DEPENCE_URL
+    "https://baidu-kunlun-public.su.bcebos.com/paddle_depence/get_xft_dependence.sh"
+    CACHE STRING "" FORCE)
 
 set(SNAPPY_PREFIX_DIR "${THIRD_PARTY_PATH}/xpu")
 set(XPU_DOWNLOAD_DIR "${SNAPPY_PREFIX_DIR}/src/${XPU_PROJECT}")
@@ -90,7 +108,9 @@ ExternalProject_Add(
     wget ${XPU_CHECK_DEPENCE_URL} && bash check_xpu_dependence.sh
     ${XPU_BASE_URL} ${XPU_XCCL_BASE_URL} && wget ${XPU_PACK_DEPENCE_URL} && bash
     pack_paddle_depence.sh ${XPU_XRE_URL} ${XPU_XRE_DIR_NAME} ${XPU_XDNN_URL}
-    ${XPU_XDNN_DIR_NAME} ${XPU_XCCL_URL} ${XPU_XCCL_DIR_NAME}
+    ${XPU_XDNN_DIR_NAME} ${XPU_XCCL_URL} ${XPU_XCCL_DIR_NAME} && wget
+    ${XPU_XFT_GET_DEPENCE_URL} && bash get_xft_dependence.sh ${XPU_XFT_URL}
+    ${XPU_XFT_DIR_NAME}
   DOWNLOAD_NO_PROGRESS 1
   UPDATE_COMMAND ""
   CMAKE_ARGS -DCMAKE_INSTALL_PREFIX=${XPU_INSTALL_ROOT}
@@ -116,7 +136,22 @@ if(WITH_XPU_BKCL)
   set(XPU_BKCL_LIB "${XPU_LIB_DIR}/${XPU_BKCL_LIB_NAME}")
   set(XPU_BKCL_INC_DIR "${THIRD_PARTY_PATH}/install/xpu/include")
   include_directories(${XPU_BKCL_INC_DIR})
+endif()
+
+if(WITH_XPU_XFT)
+  message(STATUS "Compile with XPU XFT!")
+  add_definitions(-DPADDLE_WITH_XPU_XFT)
+
+  set(XPU_XFT_LIB "${XPU_LIB_DIR}/${XPU_XFT_LIB_NAME}")
+endif()
+
+if(WITH_XPU_BKCL AND WITH_XPU_XFT)
+  target_link_libraries(xpulib ${XPU_API_LIB} ${XPU_RT_LIB} ${XPU_BKCL_LIB}
+                        ${XPU_XFT_LIB})
+elseif(WITH_XPU_BKCL)
   target_link_libraries(xpulib ${XPU_API_LIB} ${XPU_RT_LIB} ${XPU_BKCL_LIB})
+elseif(WITH_XPU_XFT)
+  target_link_libraries(xpulib ${XPU_API_LIB} ${XPU_RT_LIB} ${XPU_XFT_LIB})
 else()
   target_link_libraries(xpulib ${XPU_API_LIB} ${XPU_RT_LIB})
 endif()

python/env_dict.py.in

@@ -56,6 +56,9 @@ env_dict={
     'WITH_XPU_BKCL':'@WITH_XPU_BKCL@',
     'XPU_BKCL_LIB':'@XPU_BKCL_LIB@',
     'XPU_BKCL_LIB_NAME':'@XPU_BKCL_LIB_NAME@',
+    'WITH_XPU_XFT':'@WITH_XPU_XFT@',
+    'XPU_XFT_LIB':'@XPU_XFT_LIB@',
+    'XPU_XFT_LIB_NAME':'@XPU_XFT_LIB_NAME@',
     'THIRD_PARTY_PATH':'@THIRD_PARTY_PATH@',
     'SETUP_LOG_FILE':'@SETUP_LOG_FILE@',
     'WITH_STRIP':'@WITH_STRIP@',

python/setup.py.in

@@ -646,6 +646,10 @@ if '${WITH_XPU_BKCL}' == 'ON':
     shutil.copy('${XPU_BKCL_LIB}', libs_path)
     package_data['paddle.libs']+=['${XPU_BKCL_LIB_NAME}']
 
+if '${WITH_XPU_XFT}' == 'ON':
+    shutil.copy('${XPU_XFT_LIB}', libs_path)
+    package_data['paddle.libs']+=['${XPU_XFT_LIB_NAME}']
+
 # remove unused paddle/libs/__init__.py
 if os.path.isfile(libs_path+'/__init__.py'):
     os.remove(libs_path+'/__init__.py')

setup.py

@@ -1052,6 +1052,9 @@ def get_package_data_and_package_dir():
         shutil.copy(env_dict.get("XPU_BKCL_LIB"), libs_path)
         package_data['paddle.libs'] += [env_dict.get("XPU_BKCL_LIB_NAME")]
 
+    if env_dict.get("WITH_XPU_XFT") == 'ON':
+        shutil.copy(env_dict.get("XPU_XFT_LIB"), libs_path)
+        package_data['paddle.libs'] += [env_dict.get("XPU_XFT_LIB_NAME")]
     # remove unused paddle/libs/__init__.py
     if os.path.isfile(libs_path + '/__init__.py'):
         os.remove(libs_path + '/__init__.py')