With slowing technology scaling, specialized accelerators are increasingly attractive, but naive specialization limits accelerators to narrow domains. This is problematic practically because algorithms are constantly evolving, and intellectually as innovations are often siloed into their respective domains.
We believe this problem can be solved jointly by, 1. making accelerators more general, and 2. automating their design. I'll first overview a "general purpose accelerator ISA" that can abstract the typical behaviors of domain-specific accelerators. Our evaluation shows accelerators with these can achieve order-of-magnitude improvements over GPUs, without sacrificing programmability. However, many of their features are expensive and not useful for every workload. Therefore, our second direction is automated codesign: We developed a framework, DSAGEN, that enables users to search for the best programmable architecture given a set of input C/C++ kernels, using principles of modular hardware and compilation. Our overall vision is that hardware and ISA design can be nearly completely automated.
Bio: Tony Nowatzki is an assistant professor of computer science at the University of California, Los Angeles, where he leads the PolyArch Research Group. He joined UCLA in 2017 after completing his PhD at the University of Wisconsin -- Madison. Dr. Nowatzki's research interests include computer architecture, microarchitecture, hardware specialization, and compiler codesign. His work has been recognized with an NSF CAREER grant, three IEEE Micro Top Picks awards, CACM Research Highlights, an IEEE Best of CAL award, and a PLDI Distinguished Paper Award.