facet_to_subgrid_2d_method_2

ska_sdp_exec_swiftly.fourier_transform_dask.facet_to_subgrid_2d_method_2(facet, distr_fft_class, base_arrays, use_dask=False)[source]
Approach 2: First, do prepare_facet on the horizontal axis

(axis=0), then for loop for the horizontal subgrid direction, and do extract_subgrid within same loop do prepare_facet in the vertical case (axis=1), then go into the fila subgrid loop in the vertical dir and do extract_subgrid for that

However, remember that prepare_facet increases the amount of data involved, which in turn means that we need to shuffle more data through subsequent computations. Therefore it is actually more efficient to first do the subgrid-specific reduction, and then continue with the (constant) facet preparation along the other axis. We can tackle both axes in whatever order we like, it doesn’t make a difference for the result.

Parameters:

  • facet – 2D numpy array of facets

  • distr_fft_class – StreamingDistributedFFT class object

  • base_arrays – BaseArrays class object

  • use_dask – use dask.delayed or not

Returns:

approximate subgrid array (subgrids derived from facets)