Incorrect types following CUDA.jl refactoring

[clintonTE]

Related to #1194 and JuliaGPU/CUDA.jl#14

using CUDA, Flux
function mwecuda()
  A = rand(100,10)

  @info "A: $(typeof(A))"
  @info "gpu(A): $(typeof(gpu(A)))"
  @info "cu(A): $(typeof(CUDA.cu(A)))"
  @assert typeof(CUDA.cu(A)) <: CUDA.CuArray
  @assert typeof(gpu(A)) <: CUDA.CuArray
end
mwecuda()

Output:

[ Info: A: Array{Float64,2}
[ Info: gpu(A): CuArrays.CuArray{Float32,2,Nothing}
[ Info: cu(A): CuArray{Float32,2,Nothing}
ERROR: LoadError: AssertionError: typeof(gpu(A)) <: CUDA.CuArray

[cossio]

Can you check what version of Flux you have? I think the latest Flux is incompatible with CUDA so maybe you are running an older version.

Also let's wait til #1204 gets merged.

[clintonTE]

Master.

Note I also get a MethodError the first time I run the above on a clean session, but it goes away on subsequent passes. Here is a stack trace for the MethodError in case it is helpful:

┌ Error: Exception while generating log record in module Main at REPL[4]:6
│   exception =
│    MethodError: no method matching context!(::CUDAdrv.CuContext)
│    Closest candidates are:
│      context!(::CuContext) at C:\Users\Clinton\.julia\packages\CUDA\INQwl\src\state.jl:124
│      context!(::Function, ::CuContext) at C:\Users\Clinton\.julia\packages\CUDA\INQwl\src\state.jl:144
│    Stacktrace:
│     [1] switched_tasks(::Int64, ::Task) at C:\Users\Clinton\.julia\packages\CUDA\INQwl\src\state.jl:73
│     [2] prepare_cuda_call() at C:\Users\Clinton\.julia\packages\CUDA\INQwl\src\state.jl:29
│     [3] initialize_api() at C:\Users\Clinton\.julia\packages\CUDA\INQwl\lib\cuda\error.jl:98
│     [4] macro expansion at C:\Users\Clinton\.julia\packages\CUDA\INQwl\lib\cuda\libcuda.jl:348 [inlined]
│     [5] macro expansion at C:\Users\Clinton\.julia\packages\CUDA\INQwl\lib\cuda\error.jl:108 [inlined]
│     [6] cuMemAlloc_v2(::Base.RefValue{CuPtr{Nothing}}, ::Int64) at C:\Users\Clinton\.julia\packages\CUDA\INQwl\lib\utils\call.jl:93
│     [7] alloc at C:\Users\Clinton\.julia\packages\CUDA\INQwl\lib\cuda\memory.jl:73 [inlined]
│     [8] macro expansion at C:\Users\Clinton\.julia\packages\TimerOutputs\NvIUx\src\TimerOutput.jl:245 [inlined]
│     [9] macro expansion at C:\Users\Clinton\.julia\packages\CUDA\INQwl\src\memory.jl:114 [inlined]
│     [10] macro expansion at .\util.jl:234 [inlined]
│     [11] actual_alloc(::Int64) at C:\Users\Clinton\.julia\packages\CUDA\INQwl\src\memory.jl:113
│     [12] actual_alloc at C:\Users\Clinton\.julia\packages\CUDA\INQwl\src\memory\binned.jl:53 [inlined]
│     [13] macro expansion at C:\Users\Clinton\.julia\packages\CUDA\INQwl\src\memory\binned.jl:248 [inlined]
│     [14] macro expansion at C:\Users\Clinton\.julia\packages\TimerOutputs\NvIUx\src\TimerOutput.jl:245 [inlined]
│     [15] pool_alloc(::Int64, ::Int64) at C:\Users\Clinton\.julia\packages\CUDA\INQwl\src\memory\binned.jl:247
│     [16] alloc(::Int64) at C:\Users\Clinton\.julia\packages\CUDA\INQwl\src\memory\binned.jl:389
│     [17] macro expansion at C:\Users\Clinton\.julia\packages\TimerOutputs\NvIUx\src\TimerOutput.jl:245 [inlined]
│     [18] macro expansion at C:\Users\Clinton\.julia\packages\CUDA\INQwl\src\memory.jl:225 [inlined]
│     [19] macro expansion at .\util.jl:234 [inlined]
│     [20] alloc at C:\Users\Clinton\.julia\packages\CUDA\INQwl\src\memory.jl:224 [inlined]
│     [21] CuArray{Float32,2,P} where P(::UndefInitializer, ::Tuple{Int64,Int64}) at C:\Users\Clinton\.julia\packages\CUDA\INQwl\src\array.jl:109
│     [22] CuArray at C:\Users\Clinton\.julia\packages\CUDA\INQwl\src\array.jl:117 [inlined]
│     [23] similar at .\abstractarray.jl:671 [inlined]
│     [24] convert(::Type{CuArray{Float32,N,P} where P where N}, ::Array{Float64,2}) at C:\Users\Clinton\.julia\packages\GPUArrays\Ij4YR\src\host\construction.jl:91
│     [25] adapt_storage at C:\Users\Clinton\.julia\packages\CUDA\INQwl\src\array.jl:255 [inlined]
│     [26] adapt_structure at C:\Users\Clinton\.julia\packages\Adapt\3Zk8x\src\Adapt.jl:83 [inlined]
│     [27] adapt at C:\Users\Clinton\.julia\packages\Adapt\3Zk8x\src\Adapt.jl:81 [inlined]
│     [28] cu at C:\Users\Clinton\.julia\packages\CUDA\INQwl\src\array.jl:334 [inlined]
│     [29] macro expansion at .\logging.jl:322 [inlined]
│     [30] mwecuda() at .\REPL[4]:6
│     [31] top-level scope at REPL[5]:1
│     [32] eval(::Module, ::Any) at .\boot.jl:331
│     [33] eval_user_input(::Any, ::REPL.REPLBackend) at D:\buildbot\worker\package_win64\build\usr\share\julia\stdlib\v1.4\REPL\src\REPL.jl:86
│     [34] macro expansion at D:\buildbot\worker\package_win64\build\usr\share\julia\stdlib\v1.4\REPL\src\REPL.jl:118 [inlined]
│     [35] (::REPL.var"#26#27"{REPL.REPLBackend})() at .\task.jl:358

[ToucheSir]

The MWE works and I don't think this has been an issue for a while.