By Wenhua Guan, March 14th. 2015.
This program is for computing matrix vector multiplication of 
the Rotne-Prager matrix tensor. It calls four Laplace FMM kernel 
and can reach different accuracy. In the direct evaluation, the
self to self interaction is ignored. If self-to-self interaction needs
to be computed, just add it in the subroutine rpy3supeval().

In this code, four times of FMM whole procedure have been called. 
Actually, initilization and  graph just need to be established once. 
Computation and post can be updated each time. 

To do list: 
- How to parallelize the computation of 4 FMM Callings.