[BUG] Build Failure on HPC Cluster when enabling CUDA support

[jinhzhou]

[BUG] Build Failure on HPC Cluster when enabling CUDA support

[jinhzhou]

Hi Lorenzo, thank you for replying!

I'm also inclined to believe it's a local issue. I shall give another CUDA-enabled software a try and update accordingly!

[lorenzo-rovigatti]

The cicc: command not found seems to be related to an issue with the local CUDA installation. Is there a reason why you had to install CUDA with miniconda? If nvcc is provided by the HPC environment, you shouldn't have to install it yourself.

[jinhzhou]

Hi! Thank you for responding.

Here's the output error I got after calling cmake without cuda installed in the anaconda environment. I doubled checked the installed packages in the HPC environment and saw that cuda was indeed already installed (specifically cuda 9.1, 9.0, 8.0, 10.1, and 10.0 are installed in the HPC environment).

cmake_output_without_cuda_in_venv.txt

[jinhzhou]

Another thing to note: the HPC environment does not have cudatoolkit installed.

Edit One:
I think I was confused when I originally made the post. After running nvcc --version without the anaconda environment active, I got a -bash: nvcc: command not found output in the terminal (when cuda was not installed in the env). When cuda is installed in the env and is not active, I am able to run nvcc --version and get the version output in the terminal

Edit Two:
Testing on another computer resulted in the same cicc: not found error as in the original post.
Steps:

• Create a new anaconda venv by running: conda create --name venv
• conda activate venv
• git clone https://github.com/lorenzo-rovigatti/oxDNA.git
• cd oxDNA/
• mkdir build
• cd build
• conda install nvidia::cuda-toolkit
• cmake -DCUDA=ON ..
• make -j4

[lorenzo-rovigatti]

To me this looks like a local issue that has nothing to do with oxDNA per se. Have you tried compiling another CUDA-enabled software just to see how it goes?