Methods of adapting parallel dataflow computer system to various task types

In this article the architecture of parallel dataflow computer system and its main features are described. Varying those features one can better adapt the system to one or another specific task type. In particular, choosing the set of basic computation distribution functions one can enhance the applicability of the system to tasks with different modes of computational locality. General principles of building special-purpose parallel computer systems based on dataflow model of computation (with dynamically formed context) are given. The results of applying these methods to particular task (FFT) are shown.