AMDGPU does not load any more without error when installation not functional

[omlins]

It has been found a very useful property of GPU packages over the years to load without error when the installation not functional. Now there is thrown the following error when no installation is present:

Error: ROCm discovery failed!
│ Discovered ROCm path: .
│ Use `ROCM_PATH` env variable to specify ROCm directory.
│ 
│   exception =
│    could not load symbol "hipRuntimeGetVersion":
│    /opt/hostedtoolcache/julia/1.11.0/x64/bin/julia: undefined symbol: hipRuntimeGetVersion
│    Stacktrace:
│      [1] _hip_runtime_version()
│        @ AMDGPU.ROCmDiscovery ~/.julia/packages/AMDGPU/jLWP2/src/discovery/discovery.jl:44
│      [2] __init__()
│        @ AMDGPU.ROCmDiscovery ~/.julia/packages/AMDGPU/jLWP2/src/discovery/discovery.jl:95
│      [3] run_module_init(mod::Module, i::Int64)
│        @ Base ./loading.jl:1336
│      [4] register_restored_modules(sv::Core.SimpleVector, pkg::Base.PkgId, path::String)
│        @ Base ./loading.jl:1324
│      [5] _include_from_serialized(pkg::Base.PkgId, path::String, ocachepath::String, depmods::Vector{Any}, ignore_native::Nothing; register::Bool)
│        @ Base ./loading.jl:1213
│      [6] _include_from_serialized (repeats 2 times)
│        @ ./loading.jl:1169 [inlined]
│      [7] _require_search_from_serialized(pkg::Base.PkgId, sourcepath::String, build_id::UInt128, stalecheck::Bool; reasons::Dict{String, Int64}, DEPOT_PATH::Vector{String})
│        @ Base ./loading.jl:1975
│      [8] _require(pkg::Base.PkgId, env::String)
│        @ Base ./loading.jl:2435
│      [9] __require_prelocked(uuidkey::Base.PkgId, env::String)
│        @ Base ./loading.jl:2300
│     [10] #invoke_in_world#3
│        @ ./essentials.jl:1088 [inlined]
│     [11] invoke_in_world
│        @ ./essentials.jl:1085 [inlined]
│     [12] _require_prelocked(uuidkey::Base.PkgId, env::String)
│        @ Base ./loading.jl:2287
│     [13] macro expansion
│        @ ./loading.jl:2226 [inlined]
│     [14] macro expansion
│        @ ./lock.jl:273 [inlined]
│     [15] __require(into::Module, mod::Symbol)
│        @ Base ./loading.jl:2183
│     [16] #invoke_in_world#3
│        @ ./essentials.jl:1088 [inlined]
│     [17] invoke_in_world
│        @ ./essentials.jl:1085 [inlined]
│     [18] require(into::Module, mod::Symbol)
│        @ Base ./loading.jl:2176
│     [19] top-level scope
│        @ ~/work/ParallelStencil.jl/ParallelStencil.jl/test/ParallelKernel/test_allocators.jl:16
│     [20] include(mod::Module, _path::String)
│        @ Base ./Base.jl:557
│     [21] exec_options(opts::Base.JLOptions)
│        @ Base ./client.jl:323
│     [22] _start()
│        @ Base ./client.jl:531
└ @ AMDGPU.ROCmDiscovery ~/.julia/packages/AMDGPU/jLWP2/src/discovery/discovery.jl:113

It can be observed for example here in the ParallelStencil CI: https://github.com/omlins/ParallelStencil.jl/actions/runs/11251525928/job/31282759336?pr=169#step:6:578

In occurrence this breaks all our CI...

[luraess]

@pxl-th is this a feature or a bug from the latest discovery refactor?

[pxl-th]

That's definitely a bug... I'll open a PR with fixes

[omlins]

That's definitely a bug... I'll open a PR with fixes

Thanks @pxl-th !