TURNUS: An open-source design space exploration framework for dynamic stream programs

Although the research on the design of heterogeneous concurrent systems has a long and rich history, a unified design methodology and tool support have not emerged so far. Therefore, the creation of such systems remains a difficult, time-consuming and error-prone process. The absence of principled support for system evaluation and optimization at high level of abstraction makes the quality of the resulting implementation strongly dependent on the experience or individual preferences of the designer. In this work we are presenting TURNUS, a unified dataflow design space exploration framework for heterogeneous parallel systems. This open source framework represents a decade of research on high-level modelling and simulation methods and tools for system level performance estimation and optimization. Last year we presented heuristic algorithms that were focused on the results of exploration in terms of algorithmic optimization, rapid performance estimation, application throughput, buffer size dimensioning and power optimization. This year we are presenting the novelties that have been introduced in TURNUS such as clock gating, pipelining optimization, kernel splitting algorithms, advanced partitioning algorithms and scheduling optimization based on model predictive control techniques.

I. Introduction

The recent resurgence of interest in languages and tools for stream processing is fuelled by the ubiquity of parallel computing platforms as they are used in nearly every application area, including embedded and portable devices. It is also due to the increasing demand for highly complex and efficient streaming algorithms, for instance in video coding, image and signal processing, network packet processing or cryptography. In each of these or other fields designers have to face the problem of implementing a complex algorithm operating on one or more streams of data on a platform with multiply connected processing elements. In this demo we are discussing some of the concepts and tools required to optimize stream programs, using an at-size case study to validate them in practice. Finally, the TURNUS framework [1], showing how a designer can efficiently optimize and prototype one's design, is presented.