Abstract: We need to make it easier to design custom accelerators, especially for reconfigurable hardware (i.e., FPGAs). The current mainstream options are hardware description languages (HDLs), which are low-level languages that make it feel like you're wiring up a circuit by hand, and high-level synthesis (HLS) tools, which compile legacy software languages like C or C++ to an HDL.
The thesis of this talk is that better alternatives are possible. We find empirically (and unsurprisingly) that the semantic chasm between C++ and hardware circuits comes with myriad correctness and performance pitfalls. We advocate instead for more research on accelerator design languages: programming models that maintain computational semantics while not attempting to shield developers from thinking about hardware. Our lab is designing a programming language, Dahlia, that uses a type system to restrict HLS programs to a subset with predictable semantics and performance. Beyond Dahlia, we are designing an open-source compiler infrastructure to support the construction of new accelerator design languages.
Bio: Adrian Sampson is an assistant professor in the computer science department at Cornell. He works on programming languages, computer architecture, and the abstractions that separate them. He previously worked on approximate computing, the idea that we should allow machines to expose errors to some kinds of applications as a trade-off for computational efficiency. He sees approximate computing as an instance of a broader breakdown of airtight distinctions between hardware and software concerns.
Registration and a passcode are required for this event. If you would like to attend this talk and need the Zoom information, please register here to receive Zoom links and passcodes for the Systems Lunch events from Emery Berger.