Deinterleave data

General Usage Performance
6

I have done some additional trials using these suggestions, again on the Raspberry PI.

The new code is:

function deinterleavesj(data::Vector{T}, num_channels::Integer, num_samples_per_channel::Integer) where T <: AbstractFloat
   if length(data) == num_channels * num_samples_per_channel
       datamatrix = transpose(reshape(a, num_channels, num_samples_per_channel));
   else
       error("Dimension mismatch")
   end
   return datamatrix
end

function deinterleaverg(data::Vector{T}, num_channels::Integer, num_samples_per_channel::Integer) where T <: AbstractFloat
    if length(data) == num_channels * num_samples_per_channel
        datamatrix = transpose(reshape(view(a, :), num_channels, num_samples_per_channel));
    else
        error("Dimension mismatch")
    end
    return datamatrix
 end

giving the following performance benchmarks:

julia> @btime deinterleavesj(a, 4, length(a) ÷ 4 );
  729.622 ns (6 allocations: 96 bytes)

julia> @btime deinterleaverg(a, 4, length(a) ÷ 4 );
  1.252 μs (6 allocations: 120 bytes)

This gives close to a 100x speed improvement with minimal use of memory.

Thanks for the pointers!