[XPU] bump XCCL to 3.0.1.1 (#69752)

dynamicheart · web-flow · commit dc1f0eb365bc · 2024-11-28T15:17:49.000+08:00
* [XPU] bump XCCL to 3.0.1.1

* fix

* fix

* fix
diff --git a/cmake/external/xpu.cmake b/cmake/external/xpu.cmake
@@ -22,6 +22,7 @@ set(XPU_API_LIB_NAME "libxpuapi.so")
 set(XPU_RT_LIB_NAME "libxpurt.so")
 set(XPU_CUDA_LIB_NAME "libxpucuda.so")
 set(XPU_CUDA_RT_LIB_NAME "libcudart.so")
+set(XPU_ML_LIB_NAME "libxpuml.so")
 set(XPU_XFT_LIB_NAME "libxft.so")
 set(XPU_XPTI_LIB_NAME "libxpti.so")
 set(XPU_XBLAS_LIB_NAME "libxpu_blas.so")
@@ -31,7 +32,7 @@ set(XPU_XPUDNN_LIB_NAME "libxpu_dnn.so")
 if(NOT DEFINED XPU_XHPC_BASE_DATE)
   set(XPU_XHPC_BASE_DATE "dev/20241127")
 endif()
-set(XPU_XCCL_BASE_VERSION "3.0.0.5") # For XRE5
+set(XPU_XCCL_BASE_VERSION "3.0.1.1") # For XRE5
 if(NOT DEFINED XPU_XFT_BASE_VERSION)
   set(XPU_XFT_BASE_VERSION "20230602")
 endif()
@@ -146,6 +147,7 @@ set(XPU_XBLAS_LIB "${XPU_LIB_DIR}/${XPU_XBLAS_LIB_NAME}")
 set(XPU_RT_LIB "${XPU_LIB_DIR}/${XPU_RT_LIB_NAME}")
 set(XPU_CUDA_LIB "${XPU_LIB_DIR}/${XPU_CUDA_LIB_NAME}")
 set(XPU_CUDA_RT_LIB "${XPU_LIB_DIR}/${XPU_CUDA_RT_LIB_NAME}")
+set(XPU_ML_LIB "${XPU_LIB_DIR}/${XPU_ML_LIB_NAME}")
 set(XPU_XFA_LIB "${XPU_LIB_DIR}/${XPU_XFA_LIB_NAME}")
 set(XPU_XPUDNN_LIB "${XPU_LIB_DIR}/${XPU_XPUDNN_LIB_NAME}")
 
@@ -190,6 +192,7 @@ if(WITH_XPU_XRE5)
     BUILD_BYPRODUCTS ${XPU_XFA_LIB}
     BUILD_BYPRODUCTS ${XPU_RT_LIB}
     BUILD_BYPRODUCTS ${XPU_CUDA_RT_LIB}
+    BUILD_BYPRODUCTS ${XPU_ML_LIB}
     BUILD_BYPRODUCTS ${XPU_BKCL_LIB})
 else()
   ExternalProject_Add(
@@ -221,12 +224,6 @@ set_property(TARGET shared_xpuapi PROPERTY IMPORTED_LOCATION "${XPU_API_LIB}")
 # for cc_library(xxx SRCS xxx.c DEPS xpulib)
 generate_dummy_static_lib(LIB_NAME "xpulib" GENERATOR "xpu.cmake")
 
-if(WITH_XPU_XRE5)
-  target_link_libraries(xpulib ${XPU_API_LIB} ${XPU_RT_LIB} ${XPU_CUDA_RT_LIB})
-else()
-  target_link_libraries(xpulib ${XPU_API_LIB} ${XPU_RT_LIB})
-endif()
-
 if(WITH_XPU_XFT)
   message(STATUS "Compile with XPU XFT!")
   add_definitions(-DPADDLE_WITH_XPU_XFT)
@@ -272,13 +269,20 @@ if(WITH_XPU_XRE5)
     xpulib
     ${XPU_RT_LIB}
     ${XPU_CUDA_RT_LIB}
-    ${XPU_BKCL_LIB}
     ${XPU_XBLAS_LIB}
     ${XPU_API_LIB}
     ${XPU_XFA_LIB}
     ${XPU_XPUDNN_LIB})
 else()
-  target_link_libraries(xpulib ${XPU_RT_LIB} ${XPU_BKCL_LIB} ${XPU_API_LIB})
+  target_link_libraries(xpulib ${XPU_RT_LIB} ${XPU_API_LIB})
+endif()
+
+if(WITH_XPU_BKCL)
+  if(WITH_XPU_XRE5)
+    target_link_libraries(xpulib ${XPU_ML_LIB} ${XPU_BKCL_LIB})
+  else()
+    target_link_libraries(xpulib ${XPU_BKCL_LIB})
+  endif()
 endif()
 
 add_dependencies(xpulib ${XPU_PROJECT})
diff --git a/python/env_dict.py.in b/python/env_dict.py.in
@@ -65,6 +65,8 @@ env_dict={
     'WITH_XPU_XRE5':'@WITH_XPU_XRE5@',
     'XPU_CUDA_RT_LIB':'@XPU_CUDA_RT_LIB@',
     'XPU_CUDA_RT_LIB_NAME':'@XPU_CUDA_RT_LIB_NAME@',
+    'XPU_ML_LIB':'@XPU_ML_LIB@',
+    'XPU_ML_LIB_NAME':'@XPU_ML_LIB_NAME@',
     'WITH_XPU_BKCL':'@WITH_XPU_BKCL@',
     'XPU_BKCL_LIB':'@XPU_BKCL_LIB@',
     'XPU_BKCL_LIB_NAME':'@XPU_BKCL_LIB_NAME@',
diff --git a/python/setup.py.in b/python/setup.py.in
@@ -1027,6 +1027,10 @@ if '${WITH_XPU}' == 'ON':
         for xpu_cuda_rt_lib_file in xpu_cuda_rt_lib_list:
             shutil.copy(xpu_cuda_rt_lib_file, libs_path)
             package_data['paddle.libs'] += [os.path.basename(xpu_cuda_rt_lib_file)]
+        xpu_ml_lib_list = glob.glob('${XPU_ML_LIB}*')
+        for xpu_ml_lib_file in xpu_ml_lib_list:
+            shutil.copy(xpu_ml_lib_file, libs_path)
+            package_data['paddle.libs'] += [os.path.basename(xpu_ml_lib_file)]
         shutil.copy('${XPU_XBLAS_LIB}', libs_path)
         package_data['paddle.libs'] += ['${XPU_XBLAS_LIB_NAME}']
         shutil.copy('${XPU_XFA_LIB}', libs_path)
diff --git a/setup.py b/setup.py
@@ -1482,6 +1482,12 @@ def get_package_data_and_package_dir():
                 package_data['paddle.libs'] += [
                     os.path.basename(xpu_cuda_rt_lib_file)
                 ]
+            xpu_ml_lib_list = glob.glob(env_dict.get("XPU_ML_LIB") + '*')
+            for xpu_ml_lib_file in xpu_ml_lib_list:
+                shutil.copy(xpu_ml_lib_file, libs_path)
+                package_data['paddle.libs'] += [
+                    os.path.basename(xpu_ml_lib_file)
+                ]
             shutil.copy(env_dict.get("XPU_XBLAS_LIB"), libs_path)
             package_data['paddle.libs'] += [env_dict.get("XPU_XBLAS_LIB_NAME")]
             shutil.copy(env_dict.get("XPU_XFA_LIB"), libs_path)
diff --git a/tools/xpu/pack_paddle_dependence.sh b/tools/xpu/pack_paddle_dependence.sh
@@ -64,6 +64,7 @@ function xre_prepare() {
   check_files ${XRE_DIR_NAME}/include/xpu/runtime.h ${XRE_DIR_NAME}/so/libxpurt.so
   if [ "$WITH_XPU_XRE5" -eq 1 ]; then
     check_files ${XRE_DIR_NAME}/so/libcudart.so
+    check_files ${XRE_DIR_NAME}/so/libxpuml.so
   fi
   cp -r ${XRE_DIR_NAME}/include/xpu/* xpu/include/xpu/
   cp -r ${XRE_DIR_NAME}/so/* xpu/lib/
@@ -94,6 +95,8 @@ function xccl_prepare() {
   check_files ${XCCL_DIR_NAME}/include/bkcl.h ${XCCL_DIR_NAME}/so/libbkcl.so
   cp -r ${XCCL_DIR_NAME}/include/* xpu/include/xpu/
   cp -r ${XCCL_DIR_NAME}/so/* xpu/lib/
+  # FIXME(yangjianbang): 待bkcl增加RPATH后, 删除以下代码
+  patchelf --set-rpath '$ORIGIN/' xpu/lib/libbkcl.so
 }
 
 function local_prepare() {
