optimized_weights_computation_reg

[mumairsadiq]

I suggest certain changes to weight computations for p2p phase (specifically init_reg_body function inside fmm.cpp).

Currently, We iterate over all the bodies to see if each body influences some cells. This is an exhaustive way of doing that.

I now pick a target cell and just see the bodies in its neighbouring cells if they have some influence on the target cell. This avoids looking into all leaf cells, which slows down computation particularly as the number of particles grows.

Also, I consider only those bodies for regularization which doesn't have weight one in their own cell.

I have tried a few experiments with this update on my workstation:

./test_reg -n 33876 -P 6 -m 16 -a fd -r 0.01

init_reg_body time is reduced from 0.47098 to 0.00800 seconds

./test/test_reg -n 120000 -P 6 -m 16 -a fd -r 0.01

init_reg_body time is reduced from 12.26815 to 0.05770 seconds

[jooooow]

@mumairsadiq
Aside from the suggestion above, I think the PeriodicInteractionMapP2P part is a very good optimization and suitable for parallel computing.