VECTOR PARALLEL COMPUTATION OF A PARTICLE-IN-CELL (PIC) CODE

A vector parallel Particle-in-Cell (PIC) code is presented. The code w as designed to improve vector as well as parallel performance on a dis tributed memory system where each node consists of a shared memory sym metric vector multiprocessor. The code features sub-domains defined in shared memory and includes a retry algorithm for vectorizing the part icle loop in the deposition interpolation procedure. The particle loop contains data structure accesses that can have data dependencies. Ben chmarks were performed on an NEC SX-4/16A to evaluate methods to addre ss dependent data accesses on a shared memory system. One method, the retry algorithm, used in the deposition interpolation procedure, achie ves one of the highest performance levels for list vectors on the syst em, and a parallel efficiency of 93% using 16 processors. This provide s a 2.7 times performance improvement compared to the work-vector algo rithm that is commonly used.