Section: Research Program

Parallelism

The traditional parallelization approach, which consists in moving from a sequential to a parallel code, must be transformed into a direct design and implementation of high performance parallel software. All levels of parallelism (vector instructions, multi-cores, many-cores, clusters, grid, clouds) need to be exploited in order to extract the maximum computing power from current hardware resources [6] , [8] , [1] . An important specificity of GenScale is to systematically adopt a design approach where all levels of parallelism are potentially considered.