Interoperability

CUDA.jl

Views can be converted into CuArrays using Base.unsafe_wrap:

unsafe_wrap(CuArray, v::Kokkos.View)

And CuArrays can be converted into views with Kokkos.view_wrap:

julia> A = CuArray{Int64}(undef, 4, 4);

julia> CUDA.@allowscalar for i in eachindex(A)
           A[i] = i
       end

julia> A
4×4 CuArray{Int64, 2, CUDA.Mem.DeviceBuffer}:
 1  5   9  13
 2  6  10  14
 3  7  11  15
 4  8  12  16

julia> A_v = Kokkos.view_wrap(A)
4×4 Kokkos.Views.View{Int64, 2, Kokkos.LayoutLeft, Kokkos.CudaSpace}: <inaccessible view>

SubArrays of CuArrays (or, more precisely, any CUDA.StridedSubCuArray), can also be converted into views with a LayoutStride:

julia> sub_A = @view A[2:3, 2:3];

julia> sub_A isa CUDA.StridedSubCuArray
true

julia> size(sub_A), strides(sub_A)
((2, 2), (1, 4))

julia> sub_A_v = Kokkos.view_wrap(sub_A)
2×2 Kokkos.Views.View{Int64, 2, Kokkos.LayoutStride, Kokkos.CudaSpace}: <inaccessible view>

julia> size(sub_A_v), strides(sub_A_v)
((2, 2), (1, 4))

Unlike Kokkos.View, it is possible to perform arithmetic operations on a CuArray from the host, as well as indexing device memory (if permitted by CUDA.allowscalar(true) or CUDA.@allowscalar).

AMDGPU.jl

Views can be converted into ROCArrays using Base.unsafe_wrap:

unsafe_wrap(ROCArray, v::Kokkos.View)

And ROCArrays can be converted into views with Kokkos.view_wrap, in the same manner as for CuArrays.