Running into Could NOT find CUDA (missing: CUDA_INCLUDE_DIRS) (found version "12.6") Error with CUDA 12.6
#56
Labels
bug
Something isn't working
Comments
|
Please try again with all packages coming from just conda-forge. If you still see an issue with that, please report that as a new issue with the information requested Mixing channels in this way is unsupported |
This was referenced Dec 15, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Solution to issue cannot be found in the documentation.
Issue
We are trying to build FBGEMM against the CUDA 12.6 (pytorch/FBGEMM#3503), and are running into the following error:
For some context:
All of our builds and tests are performed inside a Conda environment, and all of the tool chains and dependencies that we use (gcc, clang, torch, CUDA) are installed from scratch using
conda install, for maximum environment isolation and build reproducibility. We have been successfully building FBGEMM against CUDA 11.7, 11.8, 12.1, and 12.4 for a while now, and have been able to rigorously document the full build process here.We usually install the full CUDA metapackage, i.e.
We did notice that for CUDA 12.6, when we look for
cuda.h, we see:whereas with CUDA 12.4, we have:
and with CUDA 11.8, we have:
In other words, the CUDA headers do not appear to be installed onto
$CONDA_PREFIX/include. It seems to me at least that conda install process for CUDA 12.6 is incorrect, or has been updated in a way that CMake is unable to find the path.My questions are:
$CONDA_PREFIX/includeto$CONDA_PREFIX/targets/x86_64-linux/include) expected?CUDA_INCLUDE_DIRSand/orCUDA_TOOLKIT_ROOT_DIR? I've read from here and here that we should manually setCUDA_TOOLKIT_ROOT_DIR, but those apply to non-Conda installations of CUDA, where everything is in one directory.Installed packages
Environment info
(build_env) bash-5.2# conda info active environment : build_env active env location : /root/miniconda/envs/build_env shell level : 2 user config file : /root/.condarc populated config files : /root/miniconda/.condarc conda version : 24.11.1 conda-build version : not installed python version : 3.12.8.final.0 solver : libmamba (default) virtual packages : __archspec=1=skylake __conda=24.11.1=0 __glibc=2.34=0 __linux=5.15.49=0 __unix=0=0 base environment : /root/miniconda (writable) conda av data dir : /root/miniconda/etc/conda conda av metadata url : None channel URLs : https://repo.anaconda.com/pkgs/main/linux-64 https://repo.anaconda.com/pkgs/main/noarch https://repo.anaconda.com/pkgs/r/linux-64 https://repo.anaconda.com/pkgs/r/noarch package cache : /root/miniconda/pkgs /root/.conda/pkgs envs directories : /root/miniconda/envs /root/.conda/envs platform : linux-64 user-agent : conda/24.11.1 requests/2.32.3 CPython/3.12.8 Linux/5.15.49-linuxkit amzn/2023.6.20241121 glibc/2.34 solver/libmamba conda-libmamba-solver/24.9.0 libmambapy/1.5.11 aau/0.4.4 c/y9hCWxGMVafoJzXeH0wkog s/kQYdsacincQ_QGd77ljlgQ e/vkuBvTVNMhBRt74NJNj-rA UID:GID : 0:0 netrc file : None offline mode : False (build_env) bash-5.2# nvcc --version nvcc warning : incompatible redefinition for option 'compiler-bindir', the last value of this option was used nvcc: NVIDIA (R) Cuda compiler driver Copyright (c) 2005-2024 NVIDIA Corporation Built on Tue_Oct_29_23:50:19_PDT_2024 Cuda compilation tools, release 12.6, V12.6.85 Build cuda_12.6.r12.6/compiler.35059454_0 (build_env) bash-5.2#The text was updated successfully, but these errors were encountered: