add cuDNN dynamic-link libraries into wheel package (#47552)

* add cudnn into whl package

* add cudnn dso into whl package

* let WITH_CUDNN_DSO be consistent with WITH_GPU

* fix WITH_CUDNN_DSO in paddle_build

CMakeLists.txt

@@ -307,6 +307,8 @@ option(WITH_FLPS "FL PS mode" OFF)
 option(WITH_RPC "Compile with rpc support" ${WITH_DISTRIBUTE})
 option(WITH_CUDNN_FRONTEND
        "Compile with CUDNN Frontend API support (experimental)" OFF)
+option(WITH_CUDNN_DSO "Compile PaddlePaddle with cuDNN dynamic-link libraries"
+       ${WITH_GPU})
 
 if(WITH_RECORD_BUILDTIME)
   set_property(
@@ -390,6 +392,16 @@ if(NOT WITH_GPU AND WITH_NCCL)
       CACHE STRING "Disable NCCL when compiling without GPU" FORCE)
 endif()
 
+if(NOT WITH_GPU AND WITH_CUDNN_DSO)
+  message(
+    WARNING
+      "Can't compile with cuDNN libraries when compiling without GPU. Force WITH_CUDNN_DSO=OFF."
+  )
+  set(WITH_CUDNN_DSO
+      OFF
+      CACHE STRING "Disable cuDNN libraries when compiling without GPU" FORCE)
+endif()
+
 # force WITH_XPU on when WITH_XPU_KP
 if(WITH_XPU_KP AND NOT WITH_XPU)
   message(

paddle/scripts/paddle_build.sh

@@ -197,6 +197,8 @@ function cmake_base() {
         INFERENCE_DEMO_INSTALL_DIR=${INFERENCE_DEMO_INSTALL_DIR:-/root/.cache/inference_demo}
     fi
 
+    WITH_CUDNN_DSO=${WITH_GPU:-OFF}
+
     distibuted_flag=${WITH_DISTRIBUTE:-OFF}
     gloo_flag=${distibuted_flag}
 
@@ -214,6 +216,7 @@ function cmake_base() {
         -DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE:-Release}
         ${PYTHON_FLAGS}
         -DWITH_GPU=${WITH_GPU:-OFF}
+        -DWITH_CUDNN_DSO=${WITH_CUDNN_DSO:-OFF}
         -DWITH_TENSORRT=${WITH_TENSORRT:-ON}
         -DWITH_ROCM=${WITH_ROCM:-OFF}
         -DWITH_CINN=${WITH_CINN:-OFF}
@@ -269,6 +272,7 @@ EOF
         -DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE:-Release} \
         ${PYTHON_FLAGS} \
         -DWITH_GPU=${WITH_GPU:-OFF} \
+        -DWITH_CUDNN_DSO=${WITH_CUDNN_DSO:-OFF} \
         -DWITH_TENSORRT=${WITH_TENSORRT:-ON} \
         -DWITH_ROCM=${WITH_ROCM:-OFF} \
         -DWITH_CINN=${WITH_CINN:-OFF} \

python/setup.py.in

@@ -8,6 +8,7 @@ import sys
 import fnmatch
 import errno
 import platform
+import glob
 
 from contextlib import contextmanager
 from setuptools import Command
@@ -471,6 +472,19 @@ shutil.copy('${BLAS_LIB}', libs_path)
 shutil.copy('${LAPACK_LIB}', libs_path)
 shutil.copy('${GFORTRAN_LIB}', libs_path)
 shutil.copy('${GNU_RT_LIB_1}', libs_path)
+if '${WITH_CUDNN_DSO}' == 'ON' and os.path.exists('${CUDNN_LIBRARY}'):
+    package_data['paddle.libs']+=[os.path.basename('${CUDNN_LIBRARY}')]
+    shutil.copy('${CUDNN_LIBRARY}', libs_path)
+    if sys.platform.startswith("linux") and '${CUDNN_MAJOR_VERSION}' == '8':
+        # libcudnn.so includes libcudnn_ops_infer.so, libcudnn_ops_train.so,
+        # libcudnn_cnn_infer.so, libcudnn_cnn_train.so, libcudnn_adv_infer.so,
+        # libcudnn_adv_train.so
+        cudnn_lib_files = glob.glob(os.path.dirname('${CUDNN_LIBRARY}') + '/libcudnn_*so.8')
+        for cudnn_lib in cudnn_lib_files:
+            if os.path.exists(cudnn_lib):
+                package_data['paddle.libs']+=[os.path.basename(cudnn_lib)]
+                shutil.copy(cudnn_lib, libs_path)
+
 if not sys.platform.startswith("linux"):
     package_data['paddle.libs']+=[os.path.basename('${GNU_RT_LIB_2}')]
     shutil.copy('${GNU_RT_LIB_2}', libs_path)