diff --git a/src/gpuarrays.jl b/src/gpuarrays.jl
index af87b5912..94e2e73d6 100644
--- a/src/gpuarrays.jl
+++ b/src/gpuarrays.jl
@@ -1,48 +1,4 @@
-struct ROCArrayBackend <: AbstractGPUBackend end
-
-struct ROCKernelContext <: AbstractKernelContext end
-
-@inline function GPUArrays.gpu_call(
-    ::ROCArrayBackend, f, args, threads::Int, blocks::Int; name::Maybe{String},
-)
-    @roc gridsize=blocks groupsize=threads name=name f(ROCKernelContext(), args...)
-end
-
-# indexing
-
-for (f, froc) in (
-    (:blockidx, :blockIdx),
-    (:blockdim, :blockDim),
-    (:threadidx, :threadIdx),
-    (:griddim, :gridGroupDim)
-)
-    @eval @inline GPUArrays.$f(::ROCKernelContext) = AMDGPU.$froc().x
-end
-
-# math
-
-@inline GPUArrays.cos(::ROCKernelContext, x) = cos(x)
-@inline GPUArrays.sin(::ROCKernelContext, x) = sin(x)
-@inline GPUArrays.sqrt(::ROCKernelContext, x) = sqrt(x)
-@inline GPUArrays.log(::ROCKernelContext, x) = log(x)
-
-# memory
-
-@inline function GPUArrays.LocalMemory(::ROCKernelContext, ::Type{T}, ::Val{dims}, ::Val{id}) where {T,dims,id}
-    ptr = AMDGPU.Device.alloc_special(Val{id}(), T, Val{AMDGPU.AS.Local}(), Val{prod(dims)}())
-    ROCDeviceArray(dims, ptr)
-end
-
-# synchronization
-
-@inline function GPUArrays.synchronize_threads(::ROCKernelContext)
-    sync_workgroup()
-    return
-end
-
-GPUArrays.device(x::ROCArray) = x.buf[].device
-
-GPUArrays.backend(::Type{<:ROCArray}) = ROCArrayBackend()
+import KernelAbstractions
 
 function GPUArrays.derive(
     ::Type{T}, x::ROCArray, dims::Dims{N}, offset::Int,