Convert diagnostics output for MultiLayerQG from Arrays to Tuples?

[navidcy]

For example, the energies() diagnostic returns KE as nlayer x 1 array, and PE as (nlayer-1) x 1 arrays.

GeophysicalFlows.jl/src/multilayerqg.jl

Lines 809 to 815 in f4c68a5

	for j = 1:nlayers
	CUDA.@allowscalar KE[j] = 1 / (2 * grid.Lx * grid.Ly) * parsevalsum(abs²∇𝐮h[:, :, j], grid) * params.H[j] / sum(params.H)
	end
	for j = 1:nlayers-1
	CUDA.@allowscalar PE[j] = 1 / (2 * grid.Lx * grid.Ly) * params.f₀^2 / params.g′[j] * parsevalsum(abs2.(vars.ψh[:, :, j+1] .- vars.ψh[:, :, j]), grid)
	end

Should we convert these to Tuples? Would that avoid scalar operations, e.g., in

GeophysicalFlows.jl/src/multilayerqg.jl

Line 810 in f4c68a5

CUDA.@allowscalar KE[j] = 1 / (2 * grid.Lx * grid.Ly) * parsevalsum(abs²∇𝐮h[:, :, j], grid) * params.H[j] / sum(params.H)

?

[glwagner]

That might work. Using views might be another option.