ubuntu22.04 使用 make 编译失败，提示大量 `/usr/bin/ld: ../../../libphi_core.so: undefined reference`

### 问题描述 Issue Description



`commit 7725a4d2d497dabc8e8ff6a387111b0def95a328`



这两天的版本没法通过编译，提示大量的 `/usr/bin/ld: ../../../libphi_core.so: undefined reference`，每次都在 60% 左右就出错了，删除 build 目录里面的东西重新编译，问题同样存在 ～



``` shell

> time cmake .. -DPY_VERSION=3.8 -DWITH_GPU=ON -DWITH_TESTING=OFF -DWITH_CUSTOM_DEVICE=OFF -DWITH_NCCL=OFF



> time make -j6 2>&1

```



``` shell

[ 59%] Linking CXX static library libplace.a

[ 59%] Built target place

[ 59%] Building CXX object paddle/fluid/platform/profiler/CMakeFiles/cpu_utilization.dir/cpu_utilization.cc.o

[ 59%] Building CXX object paddle/fluid/distributed/fleet_executor/CMakeFiles/task_loop_thread_pool.dir/task_loop_thread_pool.cc.o

[ 59%] Building CXX object paddle/fluid/platform/dynload/CMakeFiles/dynload_cuda.dir/nvrtc.cc.o

[ 59%] Linking CXX static library libcpu_utilization.a

[ 59%] Built target cpu_utilization

[ 59%] Building CXX object paddle/fluid/distributed/fleet_executor/CMakeFiles/task_loop_thread_pool.dir/task_loop_thread.cc.o

[ 59%] Building CXX object paddle/fluid/platform/dynload/CMakeFiles/dynload_cuda.dir/cuda_driver.cc.o

[ 59%] Building CXX object paddle/fluid/platform/dynload/CMakeFiles/dynload_cuda.dir/cupti.cc.o

[ 59%] Linking CXX static library libdynload_cuda.a

[ 59%] Built target dynload_cuda

[ 59%] Building CXX object paddle/fluid/distributed/collective/CMakeFiles/processgroup_comm_utils.dir/processgroup_comm_utils.cc.o

[ 59%] Building CXX object paddle/fluid/operators/generator/CMakeFiles/op_compat_infos.dir/__/ops_signature/fused_bn_activation_sig.cc.o

[ 59%] Linking CXX executable jit_kernel_benchmark

/usr/bin/ld: ../../../libphi_core.so: undefined reference to `void phi::PowGradKernel<phi::dtype::float16, phi::GPUContext>(phi::GPUContext const&, phi::DenseTensor const&, phi::DenseTensor const&, paddle::experimental::ScalarBase<phi::DenseTensor> const&, phi::DenseTensor*)'

/usr/bin/ld: ../../../libphi_core.so: undefined reference to `void phi::PNormGradKernel<float, phi::GPUContext>(phi::GPUContext const&, phi::DenseTensor const&, phi::DenseTensor const&, phi::DenseTensor const&, float, int, float, bool, bool, phi::DenseTensor*)'

/usr/bin/ld: ../../../libphi_core.so: undefined reference to `void phi::UniformKernel<phi::dtype::float16, phi::GPUContext>(phi::GPUContext const&, paddle::experimental::IntArrayBase<phi::DenseTensor> const&, phi::DataType, paddle::experimental::ScalarBase<phi::DenseTensor> const&, paddle::experimental::ScalarBase<phi::DenseTensor> const&, int, phi::DenseTensor*)'

/usr/bin/ld: ../../../libphi_core.so: undefined reference to `void phi::AtanGradKernel<double, phi::GPUContext>(phi::GPUContext const&, phi::DenseTensor const&, phi::DenseTensor const&, phi::DenseTensor*)'



...



/usr/bin/ld: ../../../libphi_core.so: undefined reference to `void phi::BatchNormKernel<phi::dtype::float16, phi::GPUContext>(phi::GPUContext const&, phi::DenseTensor const&, phi::DenseTensor const&, phi::DenseTensor const&, paddle::optional<phi::DenseTensor> const&, paddle::optional<phi::DenseTensor> const&, bool, float, float, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, bool, bool, phi::DenseTensor*, phi::DenseTensor*, phi::DenseTensor*, phi::DenseTensor*, phi::DenseTensor*, phi::DenseTensor*)'

/usr/bin/ld: ../../../libphi_core.so: undefined reference to `void phi::ReluGradKernel<double, phi::GPUContext>(phi::GPUContext const&, phi::DenseTensor const&, phi::DenseTensor const&, phi::DenseTensor*)'

/usr/bin/ld: ../../../libphi_core.so: undefined reference to `void phi::TanhGradKernel<phi::dtype::float16, phi::GPUContext>(phi::GPUContext const&, phi::DenseTensor const&, phi::DenseTensor const&, phi::DenseTensor*)'

collect2: error: ld returned 1 exit status

make[2]: *** [paddle/phi/kernels/funcs/jit/CMakeFiles/jit_kernel_benchmark.dir/build.make:115: paddle/phi/kernels/funcs/jit/jit_kernel_benchmark] Error 1

make[1]: *** [CMakeFiles/Makefile2:4973: paddle/phi/kernels/funcs/jit/CMakeFiles/jit_kernel_benchmark.dir/all] Error 2

make[1]: *** Waiting for unfinished jobs....

[ 59%] Building CXX object paddle/fluid/operators/generator/CMakeFiles/op_compat_infos.dir/__/ops_signature/fused_bn_add_activation_sig.cc.o

[ 59%] Linking CXX static library libstream_callback_manager.a

[ 59%] Built target stream_callback_manager

[ 59%] Building CXX object paddle/fluid/operators/generator/CMakeFiles/op_compat_infos.dir/__/ops_signature/fused_conv_sig.cc.o

[ 59%] Linking CXX static library libprocessgroup_comm_utils.a

[ 59%] Built target processgroup_comm_utils



```



尝试只编译 CPU 版本，可以正常编译：



``` shell

> time cmake .. -DPY_VERSION=3.8 -DWITH_GPU=OFF -DWITH_TESTING=OFF -DWITH_CUSTOM_DEVICE=OFF -DWITH_NCCL=OFF



```



### 版本&环境信息 Version & Environment Information



****************************************

Paddle version: 0.0.0

Paddle With CUDA: True



OS: ubuntu 22.04

GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0

Clang version: 14.0.0-1ubuntu1.1

CMake version: version 3.28.1

Libc version: glibc 2.35

Python version: 3.8.17



CUDA version: 11.7.99

Build cuda_11.7.r11.7/compiler.31442593_0

cuDNN version: 8.5.0

Nvidia driver version: 535.54.03

Nvidia driver List: 

GPU 0: NVIDIA P106-100

[]