Please use enviroment variables instead of hardcoded cuda path, and happy new year - Githubissues

mindspore-ai / mindspore

MindSpore is a new open source deep learning training/inference framework that could be used for mobile, edge and cloud scenarios.

https://gitee.com/mindspore/mindspore

Apache License 2.0

4.17k stars 688 forks source link

Please use enviroment variables instead of hardcoded cuda path, and happy new year #116

Open ytgui opened 3 years ago

ytgui commented 3 years ago

Environment

Hardware Environment: GPU

Software Environment:

MindSpore version (source or binary): 1.1
Python version (e.g., Python 3.7.5): 3.7.9
OS platform and distribution (e.g., Linux Ubuntu 16.04): Ubuntu 20.04
GCC/Compiler version (if compiled from source): gcc 7.5.0

Special notes for this issue

For most of linux x64 users, the cuda path is /usr/lib/x86_64-linux-gnu/libcudart.so, and conda package manager install cuda on ${conda}/envs/${env}/lib/libcudart.so. This is your mindspore/ccsrc/CMakeLists.txt#L306 script:
```
elseif(ENABLE_GPU)
set(MINDSPORE_RPATH ${MINDSPORE_RPATH}:/usr/local/cuda/lib64)
```
Temporary solution: use ln -s command to create a softlink, this could cheat cmake and build mindspore successful without changing original CMakeList.txt
I guess your CI system runs on a Linux x86 system, and cross compile for x64

Related Issues

yuyicg commented 3 years ago

Thank you for your feedback, it will be optimized in the next version.