KernelAbstraction ndrange using UnitRange

General Usage
,
1

I have an array with size(A)=(N,M) and I would like to run a kernel over ndrange=(2:N-1,3:M-1). Is there a way to do this other than defining an OffsetArray wrapper?

2

Use I = CartesianIndices(ndrange) and do for i in I, or A[I] = ..., @view A[I], etcetera? Not sure about CUDA support, though.

3

For future reference, what we ended up doing was adding the starting index of the range within the kernel. So if we want to iterate over R::CartesianIndices

@kernel function kern(A,@Const(I0))
  I = @index(Global,Cartesian)
  I += I0
  A[I] = func(I)
end
kern(backend,64)(A,R[1]-oneunit(R[1]),ndrange=size(R))

Which seems to work on general backends without a performance penalty.