This project addresses the problem of design space exploration for high-performance embedded computing platforms. Such systems are increasingly heterogeneous, and include powerful multi-core microprocessors (CPUs), many-core Graphics Processing Units (GPUs), and Digital Signal Processors (DSPs). Despite the availability of these powerful platforms, many barriers exist for application developers and platform designers to reap the full benefits of these devices. Some of these challenges include finding effective mappings of applications to a wide range of heterogeneous compute platforms and defining/exploring the most appropriate memory interfaces.
The goal of the project is development of a Heterogeneous Design Environment (HDE), that will allow one to define algorithms at a high level, and automate the search of the design space to select hardware and software implementations to meet desired objectives. Towards this co-design goal, the research includes investigation of methods of expressing algorithms at a conceptual, functional level; exploration of static methods of analyzing parallelism of applications; and extending simulation frameworks to support and accurately model power and performance on CPUs, GPUs, and DSPs at different levels of abstraction. The target application domain for proof of this concept is computer vision, one of the fastest growing market sectors within embedded high-performance computing, in which applications tend to be data-driven, compute-hungry and power-constrained.
The Heterogeneous Design Environment is expected to simplify development of power-efficient embedded computing applications across a set of heterogeneous devices. By providing a rich set of design exploration tools in the context of industry-standard programming frameworks, such as Multi2Sim and OpenCL, the project aims to accelerate the pace at which new computer vision applications are developed. Integrated education and outreach activities include new development of new courseware, delivery of tutorials, involvement of undergraduates, participation in a summer program for middle-school students, and use of the HDE in undergraduate and graduate courses.
