The implementation of assign_ngb_cells is very sub-optimal and requires allocating a totncells^2*8 bytes array. For small rmax, totncells = NLATMAX * NLATMAX * NLATMAX = 10^6. Therefore, the code wants to allocate a 10^12 * 8 ~ 7.5 GiB array. Not only is this wasteful, performance is severely compromised. The assign_ngb_cells takes ~3seconds while the actual pair-counting takes only ~0.5s.
This issue came up while trying to compare to the range_search routine in @mlpack.
The implementation of
assign_ngb_cellsis very sub-optimal and requires allocating atotncells^2*8bytes array. For small rmax,totncells = NLATMAX * NLATMAX * NLATMAX = 10^6. Therefore, the code wants to allocate a10^12 * 8 ~ 7.5 GiBarray. Not only is this wasteful, performance is severely compromised. Theassign_ngb_cellstakes ~3seconds while the actual pair-counting takes only ~0.5s.This issue came up while trying to compare to the
range_searchroutine in @mlpack.