Next: Out-of-Core Algorithm Up: Implementation Previous: LU Factor and

RCS Computation

The RCS computation program implemented on the CM-5 is written in Fortran 77 with the CMMD message passing library. The first task is to read in the current coefficients from a storage device. The CMMD global-read function, which is the same as the CMMD global-write described in the matrix fill subsection, is employed to carry out this task. Then, the far field contributed by all currents is computed. Finally, the radar cross section is computed. The far field computation is the major task discribed in this subsection. There are two cases; one is for a single current vector and another is for multiple current vectors.

Let and denote the numbers of the transmitters and receivers, respectively, and and be respectively the numbers of the polarizations of the transmitter and the receivers.

For a single current vector, if is a small number then only one node is active and the rest of the nodes stay inactive. If is considerably larger it is worthwhile to divide the current vector into pieces and assign each node to do some work. The result of each node is summed up in node 0. For current vectors, if , then make the first nodes do the computation and the rest of nodes remain idle. If , then each node is associated with a slab of the current matrix. The result of each node is again summed up in node 0. The divide and conquer algorithm is applied to sum the result of each node in node 0. The algorithm is illustrated in Figure 4.10, where .

A pseudo code of the algorithm implemented is given in Figure 4.11.


xshen@
Sat Dec 3 17:51:03 EST 1994