search

uwsbel / autonomy-research-testbed

https://projects.sbel.org/autonomy-toolkit
MIT License
15 stars 5 forks source link

CUDA_CUDA_LIBRARY Error Cmake #150

Open bretwitt opened 4 months ago

bretwitt commented 4 months ago

Host system has NVIDIA Container Toolkit 15.2, CUDA 11.4 on Ubuntu 20.04 (if it matters)

I'm building with atk dev -ba -s chrono --optionals gpus vnc to build a fresh Chrono image (the dockerhub one is out of date for the current Chrono feature/ros branch) I'm also building for humble

I'm running into the following error at step 13

image

Would appreciate some guidance!

bretwitt commented 4 months ago

As an update, error is reproduced with CUDA 12.2 and NVIDIA CUDA toolkit 12.2 and nvidia_container_toolkit

bretwitt commented 4 months ago

This was fixed by appending -DCMAKE_LIBRARY_PATH=/usr/local/cuda/lib64/stubs to the CMake build command. The demo_ART_cone.py needed some small fixes to accomodate a (new?) PyChrono API. I'd be happy to make it a PR/fork if needed

AaronYoung5 commented 3 months ago

@bretwitt Ah sorry, I must have notifcations disabled! Thanks for the update. Odd that it worked okay before and not anymore. Looks like we can just set -DCMAKE_LIBRARY_PATH=${LIBRARY_PATH} since nvidia includes the cuda library path in the LIBRARY_PATH env variable (I wonder if there was a change in chrono which caused this to break, probably okay to do this for now tho). I will make a PR with this change, as well as updating the chrono branch to main since feature/ros has been merged in the latest release.

As for the PyChrono demo fixes, feel free to make a PR with those! I meant to update those, but didn't get to it. Thanks for debugging that!!