Vectorization and multitasking with a Monte Carlo code for neutron transport problems

Abstract

This paper summarizes two improvements of a Monte Carlo code by resorting to vectorization and multitasking techniques.

After a short presentation of the physical problem to solve and a description of main difficulties to produce an efficient coding, this paper introduces the vectorization principles employed and briefly describes the vectorized algorithm functioning.

Next, measured performances on CRAY 1S. CYBER 205 and CRAY X-MP/48 are compared.

The second part of this paper is devoted to multitasking technique. Starting from the standard multitasking tools available with FORTRAN on CRAY X-MP. a multitasked algorithm and its measured speed-ups are presented.

In conclusion we prove that vector and parallel computers are a great opportunity for such Monte Carlo algorithms.