When adding CUDNN_jll v8.2.4 to an empty environment on Julia 1.9.4, I get:
Failed to precompile CUDNN_jll [62b44479-cb7b-5706-934f-f13b2eb2e645] to "/opt/julia_depot/compiled/v1.9/CUDNN_jll/jl_EkyUPT".
ERROR: LoadError: ArgumentError: Package CUDNN_jll does not have CUDA_loader_jll in its dependencies:
- You may have a partially installed environment. Try `Pkg.instantiate()`
to ensure all packages in the environment are installed.
- Or, if you have CUDNN_jll checked out for development and have
added CUDA_loader_jll as a dependency but haven't updated your primary
environment's manifest file, try `Pkg.resolve()`.
- Otherwise you may need to report an issue with CUDNN_jll
I can add (and precompile) CUDA_loader_jll
and CUDA_Runtime_jll
without issue.
Any tips?
Context: I'm trying to set-up a build env. for the Torch.jl libtorch C API wrapper: https://buildkite.com/julialang/torch-dot-jl/builds/169#018c6623-42be-4771-b995-6b9398f96bde/112-256
From Tim Besard:
CUDA_loader_jll was an old experiment. you'll probably have to rebuild that version of CUDNN_jll based on the modern stack of JLLs (i.e. CUDA_Runtime_jll)
Jesper Stemann Andersen has marked this topic as resolved.
Last updated: Nov 22 2024 at 04:41 UTC