Repository navigation
Fix #423: Julia backend no longer installs all of CUDA toolkit - #424
Merged
Merged
Conversation
|
This branch was successfully deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.



Fixes the issue described in #423 by having the Julia CUDA backend install and loading CUDACore and NVML instead of CUDA by default. If CUDA.jl is installed, it reuses that. When the backend has to install CUDACore and finds a CUDA toolkit already on the system that works with the driver, it tells CUDA.jl to use that toolkit instead of downloading one. "Works" here means CUDA 12 or newer, a major version no higher than the driver supports, and the parts CUDA.jl needs all present. It looks in the same places CUDA.jl does (
CUDA_HOME/CUDA_PATH/CUDA_ROOT, ptxas on PATH,/usr/local/cuda*,/opt/cuda*) and prints a warning saying which toolkit it chose. The setting is written into the project before installation, because CUDA.jl decides what to download at install time.If so desired, a new environment variable can be used,
KERNEL_TUNER_JULIA_LOCAL_CUDA_TOOLKIT, to control this: auto (default), yes or no.This PR has been tested on CUDA & Metal with both
noxand the tests of KernelTuner.jl as well as the upcoming SC notebook.