cmake 编译过程中提示缺少 libphi.a
wsdxyz opened this issue · 3 comments
描述：jetson nano 下使用make从源码编译，cmake 构建成功后，使用make -j4编译，提示 缺少 libphi.a文件；
问题： 原因，如何解决？
日志如下：，环境 如下，cmakelists.txt 见附件
[ 11%] Building CUDA object paddle/phi/CMakeFiles/phi.dir/kernels/sparse/gpu/softmax_kernel.cu.o
[ 11%] Building CXX object paddle/phi/CMakeFiles/phi.dir/kernels/cpu/activation_grad_kernel.cc.o
[ 11%] Building CXX object paddle/phi/CMakeFiles/phi.dir/kernels/cpu/warprnnt_kernel.cc.o
[ 11%] Linking CXX static library libphi.a
[ 42%] Built target phi
[ 42%] Built target string_helper
[ 42%] Built target pretty_log
make[2]: *** No rule to make target 'paddle/phi/libphi.a', needed by 'paddle/phi/kernels/funcs/jit/jit_ker nel_benchmark'. Stop.
CMakeFiles/Makefile2:5471: recipe for target 'paddle/phi/kernels/funcs/jit/CMakeFiles/jit_kernel_benchmark .dir/all' failed
make[1]: *** [paddle/phi/kernels/funcs/jit/CMakeFiles/jit_kernel_benchmark.dir/all] Error 2
make[1]: *** Waiting for unfinished jobs....
[ 42%] Built target pybind_util
make[2]: *** No rule to make target 'paddle/phi/libphi.a', needed by 'paddle/phi/tools/print_phi_kernels'. Stop.
CMakeFiles/Makefile2:5510: recipe for target 'paddle/phi/tools/CMakeFiles/print_phi_kernels.dir/all' faile d
make[1]: *** [paddle/phi/tools/CMakeFiles/print_phi_kernels.dir/all] Error 2
[ 42%] Built target enforce
Makefile:148: recipe for target 'all' failed
make: *** [all] Error 2
更多细节：
硬件环境：
cuda 10.2.89
cudnn 8.2.1
jetpack 4.6.4
tensorRT 8.x （32G 的是8.2.1，单独安装见3.1）
sudo find / -name "libnvinfer.so" /usr/lib/aarch64-linux-gnu/libnvinfer.so
ubuntu 18.4 LTS
GPU架构 maxwell （Jetson Nano | NVIDIA Developer）
python 3.9.11
paddlepaddle 2.5.0
paddleDetection 2.6.1
gcc 7.5 --> 转 8.2
gfortran :
cmake 3.10 ->3.16 /usr/local/ --> 3.19.8 (必须，见4.1）
nvidia-docker Docker version 20.10.21, build 20.10.21-0ubuntu1~18.04.3
nccl= 无
CMakeLists.txt 见附件
CMakeLists.txt
你好，
可以先在 https://www.paddlepaddle.org.cn/inference/v2.6/guides/install/download_lib.html 这里获取jetson对应的预测库；
此外，可以参考 https://github.com/PaddlePaddle/Paddle-Inference-Demo/tree/master/c%2B%2B/gpu/resnet50 实现编译和部署
你好，请切换Paddle分支到develop之后再编译试试呢。
谢谢你们的告知，我先尝试下另一个jetson nano环境下编译paddlepaddle，之后再反馈，谢谢