Invited Paper: Acceleration of Computationally- Intensive Kernels in the Reconfigurable Era

Abstract—One of the major topics that attracts constantly the interest of research community is the acceleration of computationally-intensive applications. Towards this direction, different technologies are competing each other and are all characterized by a common tendency; they evolve continuously towards improving their products so as to serve better their customers and attract new ones. Each company stresses the stronger advantages of its technology to convince people from academia and industry, and either individuals to use it. However, none of the existing technologies monopolizes all domains yet. This is because each one has its own benefits and drawbacks and cannot serve well all application domains. Another reason is that different people prefer to acquire skills in a specific technology and avoid switching between different technologies, even though this could benefit the application under development. Hence, from the one side a technology itself imposes performance margins in terms of factors such as speed, power, energy, and cost, while on the other side stands the willingness of the user to adopt a technology. This adoption can be affected heavily by the flexibility and the “friendliness ” (i.e. easy-to-use) of the technology. Present paper discusses problems in using reconfigurable technology and suggests some research directions. We delve into the details of two systems equipped with specific reconfigurable platforms and explore the capabilities offered to the user for accessing this technology. We point out some problems in adopt-ing reconfigurable technology and we study some performance characteristics. Aim of the paper is to signify issues that could gain the attention of the research community, and trigger the exploration of solutions that will foster the adoption of reconfigurable technology by a wider range of people. I