Closed vatai closed 5 years ago
This could be done by analyzing comm_table[]
An idea: considering the npart
x npart
x nlevel * n
size comm_table
(from x to x vvidx), this could be reduced (by summation for example) to an npart
x npart
matrix (from x to) which would be optimal if the largest elements are in the diagonals.
See reduce_comm.{c,h}
For now the idea is to go trough all the permutations P:[0,..,npart-1] -> [0,..,npart-1], such that sum of overlapping indices is maximal, where overlapping indices are {i : P(part[i]) = old_part[i]}.