Canno precompile CUDAExt with fixed CUDA runtime version

[Moelf]

]add XGBoost, CUDA

using CUDA
CUDA.set_runtime_version!(v"11.8"; local_toolkit=true)

then try to recompile XGBoost gives:

ERROR: LoadError: InitError: UndefVarError: `libxgboost` not defined
Stacktrace:
  [1] XGBRegisterLogCallback
    @ ~/.julia/packages/XGBoost/nqMqQ/src/Lib.jl:56 [inlined]
  [2] __init__()
    @ XGBoost ~/.julia/packages/XGBoost/nqMqQ/src/XGBoost.jl:40
  [3] run_module_init(mod::Module, i::Int64)
    @ Base ./loading.jl:1134
  [4] register_restored_modules(sv::Core.SimpleVector, pkg::Base.PkgId, path::String)
    @ Base ./loading.jl:1122
  [5] _include_from_serialized(pkg::Base.PkgId, path::String, ocachepath::String, depmods::Vector{Any})
    @ Base ./loading.jl:1067
  [6] _require_search_from_serialized(pkg::Base.PkgId, sourcepath::String, build_id::UInt128)

[tylerjthomas9]

I also get this error when local_toolkit=true. I do not get it when local_toolkit=false. I am not sure what is causing it.

[Moelf]

bump?

[tylerjthomas9]

I haven't had a chance to investigate this further. I've been battling carpal tunnel for a few months, and my open-source time/ability has been impacted the most.

[trivialfis]

Have you tried using CUDA after setting the local toolkit and restarting Julia? Is there any error?