Initialization is not thread safe

[maleadt]

julia> using CUDAnative, CuArrays
[ Info: Precompiling CuArrays [3a865a2d-5b23-5a0f-bc46-62713ec82fae]

julia> Threads.nthreads()
4

julia> ENV["JULIA_DEBUG"] = "CUDAnative"
"CUDAnative"

julia> kernel() = nothing
kernel (generic function with 1 method)

julia> Threads.@threads for i in 1:10
           @cuda kernel()
       end
ERROR: ┌ Debug: Initializing CUDA on thread 3
└ @ CUDAnative ~/Julia/pkg/CUDAnative/src/init.jl:35
TaskFailedException:
AssertionError: CUDAdrv.jl is not functional
Stacktrace:
 [1] libcuda at /home/tim/Julia/pkg/CUDAdrv/src/CUDAdrv.jl:79 [inlined]
 [2] (::CUDAdrv.var"#615#cache_fptr!#139")() at /home/tim/Julia/pkg/CUDAapi/src/call.jl:31
 [3] macro expansion at /home/tim/Julia/pkg/CUDAapi/src/call.jl:39 [inlined]
 [4] macro expansion at /home/tim/Julia/pkg/CUDAdrv/src/libcuda.jl:153 [inlined]
 [5] macro expansion at /home/tim/Julia/pkg/CUDAdrv/src/error.jl:115 [inlined]
 [6] cuCtxGetCurrent at /home/tim/Julia/pkg/CUDAapi/src/call.jl:93 [inlined]
 [7] CuCurrentContext() at /home/tim/Julia/pkg/CUDAdrv/src/context.jl:84
 [8] context() at /home/tim/Julia/pkg/CUDAnative/src/init.jl:75
 [9] cufunction_fast(::Function, ::Type{T} where T, ::Int64; name::Nothing, kwargs::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}) at /home/tim/Julia/pkg/CUDAnative/src/execution.jl:379
 [10] cufunction at /home/tim/Julia/pkg/CUDAnative/src/execution.jl:419 [inlined]
 [11] macro expansion at /home/tim/Julia/pkg/CUDAnative/src/execution.jl:157 [inlined]
 [12] macro expansion at ./REPL[5]:2 [inlined]
 [13] (::var"#2#threadsfor_fun#3"{UnitRange{Int64}})(::Bool) at ./threadingconstructs.jl:61
 [14] (::var"#2#threadsfor_fun#3"{UnitRange{Int64}})() at ./threadingconstructs.jl:28
Stacktrace:
 [1] wait(::Task) at ./task.jl:267
 [2] top-level scope at ./threadingconstructs.jl:69

julia> ┌ Debug: Trying to use artifacts...
└ @ CUDAnative ~/Julia/pkg/CUDAnative/src/bindeps.jl:37
┌ Debug: Using CUDA 10.2.89 from an artifact at /home/tim/Julia/depot/artifacts/93956fcdec9ac5ea76289d25066f02c2f4ebe56e
└ @ CUDAnative ~/Julia/pkg/CUDAnative/src/bindeps.jl:93
┌ Debug: Toolchain with LLVM 8.0.1, CUDA driver 10.2.0 and toolkit 10.2.89 supports devices 3.0, 3.2, 3.5, 3.7, 5.0, 5.2, 5.3, 6.0, 6.1, 6.2, 7.0, 7.2 and 7.5; PTX 3.2, 4.0, 4.1, 4.2, 4.3, 5.0, 6.0, 6.1 and 6.3
└ @ CUDAnative ~/Julia/pkg/CUDAnative/src/bindeps.jl:199
┌ Debug: Incompatibility detected between CUDA and LLVM 8.0+; disabling debug info emission
└ @ CUDAnative ~/Julia/pkg/CUDAnative/src/compiler/irgen.jl:145
# hangs

[maleadt]

#218