hcat of vectors yields "incompatible LLVM IR"

[gustafsson]

Similar to #55 but this issue refers to hcat.

julia> using CuArrays;

julia> x = cu(rand(2));

julia> y = cu(rand(2));

julia> hcat(x,y)
WARNING: Encountered incompatible LLVM IR for kernel(Int64, CUDAnative.CuDeviceArray{Float32,2,CUDAnative.AS.Global}, Tuple{CUDAnative.CuDeviceArray{Float32,1,CUDAnative.AS.Global},CUDAnative.CuDeviceArray{Float32,1,CUDAnative.AS.Global}}) at capability 3.7.0: CUDAnative.InvalidIRError("calls the Julia runtime", ("jl_bounds_error_unboxed_int",   call void @jl_bounds_error_unboxed_int(i8* %50, i8** inttoptr (i64 139624390903792 to i8**), i64 %.), !dbg !49))
ERROR: LLVM IR generated for kernel(Int64, CUDAnative.CuDeviceArray{Float32,2,CUDAnative.AS.Global}, Tuple{CUDAnative.CuDeviceArray{Float32,1,CUDAnative.AS.Global},CUDAnative.CuDeviceArray{Float32,1,CUDAnative.AS.Global}}) at capability 3.7.0 is not compatible
Stacktrace:
 [1] #compile_function#58(::Bool, ::Function, ::Any, ::Any, ::VersionNumber) at ~/.julia/v0.6/CUDAnative/src/jit.jl:434
 [2] cufunction(::CUDAdrv.CuDevice, ::Any, ::Any) at ~/.julia/v0.6/CUDAnative/src/jit.jl:488
 [3] macro expansion at ~/.julia/v0.6/CUDAnative/src/execution.jl:107 [inlined]
 [4] _cuda(::Tuple{Int64,Int64}, ::Int64, ::CUDAdrv.CuStream, ::CuArrays.#kernel#12, ::Int64, ::CUDAnative.CuDeviceArray{Float32,2,CUDAnative.AS.Global}, ::Tuple{CUDAnative.CuDeviceArray{Float32,1,CUDAnative.AS.Global},CUDAnative.CuDeviceArray{Float32,1,CUDAnative.AS.Global}}) at ~/.julia/v0.6/CUDAnative/src/execution.jl:80
 [5] _cat(::Int64, ::CuArray{Float32,2}, ::CuArray{Float32,1}, ::Vararg{CuArray{Float32,1},N} where N) at ~/.julia/v0.6/CuArrays/src/utils.jl:96
 [6] cat_t(::Int64, ::Type{T} where T, ::CuArray{Float32,1}, ::CuArray{Float32,1}, ::Vararg{CuArray{Float32,1},N} where N) at ~/.julia/v0.6/CuArrays/src/utils.jl:104
 [7] hcat(::CuArray{Float32,1}, ::CuArray{Float32,1}) at ~/.julia/v0.6/CuArrays/src/utils.jl:108