Modeling and Verification Methods for Application Design in Heterogeneous Architectures

Abstract

A methodology of application design for massive parallelism systems on the example of GPGPU systems focused on the algorithmic design stage is proposed. Two stages of design are considered: creation of a formal specification and its research and verification. For the first stage, the use of mathematical apparatus of the system of algorithmic algebras/modified system of algorithmic algebras and transition systems is proposed. For the second stage, the use of network and automatic models is analyzed, and the advantages of each model are given. In particular, the computational model in NVIDIA CUDA using Petri nets, as well as the linear-temporal logic formulas and automatic model, was studied.