Monte Carlo calculations on vector supercomputers using GMVP

A multigroup general purpose Monte Carlo code GMVP has been developed. The vectorization algorithm is based on a stack-driven zone selection method. GMVP can treat repeated rectangular and hexagonal lattices together with combinatorial geometry which is quite useful in order to achieve a high gain by vectorization. The performance of the code was evaluated by solving various types of problems. In addition, a continuous energy code is under development and its performance is compared with conventional codes. The code was installed on four other different supercomputers to investigate the portability and computer dependence of code performance. (author)