Vectorization methods development for a new version of the KENO-V.a criticality safety code

The objective of this research project is to develop a vectorized version of the KENO-V.a Criticality Safety Code, benchmark it against the original version of the code, and determine its speed-up factor for various classes of problems. The current generation of supercomputers are equipped with vector processors which allow the same operation to be simultaneously performed on a string of data. Unfortunately, the Monte Carlo Algorithm used in KENO-V.a, which tracks particles individually, cannot utilize these vector processors. A new Monte Carlo Algorithm needed to be developed which would efficiently utilize the vector processors currently used in computers. The algorithm developed for the vectorized version of KENO-V.a is an event-based, stack-driven, all-zone, tagged particle Monte Carlo algorithm. This algorithm divides the particles into one of four stacks; free-flight, inward-crossing, outward-crossing, or collision. A fifth stack. Kill, contains all particles that have either leaked from the system or been terminated by Russian roulette. The stack containing the largest number of particles is the next stack processed. The generation is completed when the four main stacks are empty. All the particles in the longest stack are processed simultaneously. Only the particle number is transferred between stacks, the particle data remain in permanent vector locations and are updated as the particles traverse through the system. This approach minimizes data transfer between stacks and optimizes the vector length thus maximizing the speed-up. For the 25 benchmark problems, speed-up factors ranging from 1.8 to 5.7 relative to the optimized scalar version of KENO-V.a were obtained. In addition, problem geometry, material composition, and the number of histories per generation all have significant effects on the speed-up factor.