ka_amd_offsetarray.jl

using KernelAbstractions
using OffsetArrays
using AMDGPU
using ROCKernels
using Test

import KernelAbstractions: Event
using ROCKernels: ROCEvent

function Event(::ROCDevice)
    queue = AMDGPU.get_default_queue()
    event = AMDGPU.barrier_and!(queue, AMDGPU.active_kernels(queue))
    MultiEvent(Tuple(ROCEvent(s) for s in event.signals))
end

@kernel function copy_kernel!(A, B)
    I = @index(Global)
    @inbounds A[I] = B[I]
end

function mycopy!(A, B)
    @assert size(A) == size(B)
    copy_kernel!(ROCDevice(), 256)(A, B, ndrange=length(A))
end

A = OffsetArray(zeros(Float32, 1024) |> ROCArray, -3)
B = OffsetArray(ones(Float32, 1024) |> ROCArray, -3)

event = mycopy!(A, B)
wait(event)

@test A == B