Jetxon Xavier NX paddle installation g++ segmentation fault on paddle/phi/kernels/funcs/CMakeFiles/math_function.dir/math_function.cc.o

[lmdnyo]

问题描述 Issue Description

I am trying to build paddlepaddle on the Jetxon Xavier NX 16 GB using given instructions. The necessary packages and per-requisites are installed. But it is getting stuck in make TARGET=ARMV8 -j$(nproc). Here is the errror.

[ 6%] Building CXX object paddle/phi/kernels/funcs/CMakeFiles/math_function.dir/math_function.cc.o
g++: internal compiler error: Segmentation fault signal terminated program cc1plus
Please submit a full bug report,
with preprocessed source if appropriate.
See file:///usr/share/doc/gcc-8/README.Bugs for instructions.
paddle/phi/kernels/funcs/CMakeFiles/math_function.dir/build.make:75: recipe for target 'paddle/phi/kernels/funcs/CMakeFiles/math_function.dir/math_function.cc.o' failed
make[2]: *** [paddle/phi/kernels/funcs/CMakeFiles/math_function.dir/math_function.cc.o] Error 4
CMakeFiles/Makefile2:57196: recipe for target 'paddle/phi/kernels/funcs/CMakeFiles/math_function.dir/all' failed
make[1]: *** [paddle/phi/kernels/funcs/CMakeFiles/math_function.dir/all] Error 2
Makefile:135: recipe for target 'all' failed
make: *** [all] Error 2

I am using following configuration.
Ubuntu 18.04
uname -a = Linux tegra-ubuntu 4.9.253-tegra #4 SMP PREEMPT Wed Mar 9 19:40:11 CST 2022 aarch64 aarch64 aarch64 GNU/Linux
Jetpack = 4.6
Cuda = 10.2
Cudnn = 8
gcc = gcc-7 (tried gcc-8 as well)
paddle version = 2.3.0

I also tried to install 2.1.3 and 2.0.2 but same error. Tried to install ready packages of NVIDIA JETSON NANO paddlepaddle-gpu-2.0.0 wheel from QEngineering, tried gpu-1.6 and 1.83 wheels but no luck. Got errors in fluid.install_check.run_check().

After that I even tried with make TARGET=ARMV8 and -j1 with following cmake configuration.

cmake .. -DWITH_MKL=OFF -DWITH_MKLDNN=OFF -DWITH_AVX=OFF -DWITH_GPU=ON -DON_INFER=ON -DWITH_TESTING=OFF -DCMAKE_BUILD_TYPE=Release -DCUDA_ARCH_NAME=All -DWITH_ARM=ON -DWITH_NV_JETSON=ON -DWITH_PYTHON=ON -DPY_VERSION=3.6 -DPYTHON_INCLUDE_DIR:PATH=/usr/include/python3.6 -DPYTHON_EXECUTABLE:FILEPATH=/usr/bin/python3.6 -DPYTHON_LIBRARY:FILEPATH=/usr/lib/aarch64-linux-gnu/libpython3.6m.so -DWITH_XBYAK=OFF -DWITH_RCCL=OFF -DWITH_ROCM=OFF -DCMAKE_CUDA_COMPILER=/usr/local/cuda-10.2/bin/nvcc

But still the same error every time. Please help since I am not able to compile the same on Xavier NX.

版本&环境信息 Version & Environment Information

---

Paddle version: 2.0.0
Paddle With CUDA: True

OS: Ubuntu 18.04
Python version: 3.6.9

CUDA version: 10.2.300
Build cuda_10.2_r440.TC440_70.29663091_0
cuDNN version: None.None.None
Nvidia driver version: None

---

[paddle-bot-old]

您好，我们已经收到了您的问题，会安排技术人员尽快解答您的问题，请耐心等待。请您再次检查是否提供了清晰的问题描述、复现代码、环境&版本、报错信息等。同时，您也可以通过查看官网API文档、常见问题、历史Issue、AI社区来寻求解答。祝您生活愉快～

Hi! We've received your issue and please be patient to get responded. We will arrange technicians to answer your questions as soon as possible. Please make sure that you have posted enough message to demo your request. You may also check out the API，FAQ，Github Issue and AI community to get the answer.Have a nice day!

[lmdnyo]

I was wondering why cudnn was not shown by the above script. So I checked and found it is already there. Here are the details

nvidia@tegra-ubuntu:~$ find /usr -name "cudnn*.h"
/usr/include/aarch64-linux-gnu/cudnn_v8.h
/usr/include/aarch64-linux-gnu/cudnn_cnn_train_v8.h
/usr/include/aarch64-linux-gnu/cudnn_adv_infer_v8.h
/usr/include/aarch64-linux-gnu/cudnn_adv_train_v8.h
/usr/include/aarch64-linux-gnu/cudnn_cnn_infer_v8.h
/usr/include/aarch64-linux-gnu/cudnn_ops_train_v8.h
/usr/include/aarch64-linux-gnu/cudnn_backend_v8.h
/usr/include/aarch64-linux-gnu/cudnn_ops_infer_v8.h
/usr/include/aarch64-linux-gnu/cudnn_version_v8.h
/usr/include/cudnn.h
/usr/include/cudnn_cnn_infer.h
/usr/include/cudnn_adv_infer.h
/usr/include/cudnn_ops_train.h
/usr/include/cudnn_version.h
/usr/include/cudnn_cnn_train.h
/usr/include/cudnn_backend.h
/usr/include/cudnn_adv_train.h
/usr/include/cudnn_ops_infer.h

[ZhangHandi]

Hi, try to build in a single-threaded environment instead of multi-threaded