“Reconfigurable Architecture for Chip Multiprocessors”
Dr. Matthew A. Watkins, Harvey Mudd College
Prior research in reconfigurable computing involved augmenting a single processor core with reconfigurable logic. Despite significant performance gains for some applications, the area and power costs can easily outweigh the benefits, especially for workloads that do not make good use of the fabric. This prior work also focused almost exclusively on uniprocessor systems and did not address the unique requirements of parallel applications.
This talk proposes a novel reconfigurable architecture for chip multiprocessors (CMPs) running a mix of serial and parallel applications. In our approach, the reconfigurable fabric is shared among multiple threads to amortize the area and power costs and increase fabric utilization. To further reduce the overhead, we propose a heterogeneous CMP with different regions optimized for different tasks, including regions with shared reconfigurable fabrics, and other regions with only conventional cores. Within a reconfigurable region, the architecture dynamically manages the assignment of threads to fabric clusters and the dynamic partitioning of the fabric and includes mechanisms that accelerate parallel applications and enable parallelization of otherwise sequential applications. This talk will discuss the architecture of the shared fabric, the management schemes developed to optimize the performance of the shared fabric and the use of the fabric for multiple forms of fine-grained communication.
Matthew Watkins is a visiting assistant professor of engineering at Harvey Mudd College. His research interests are in the general area of computer architecture and digital design with a special focus on reconfigurable architectures for chip multiprocessors and the design of large multicore systems. He received his PhD and MS degrees in electrical and computer engineering from Cornell University. He was a 2005 National Science Foundation Graduate Research Fellow, and he’s a member of IEEE and the ACM.