Predictable Accelerator Design

with Time-Sensitive Affine types


While field-programmable gate arrays (FPGAs) provide an opportunity to co-design applications with hardware accelerators, they remain difficult to program. High-level synthesis (HLS) tools promise to raise the level of abstraction by compiling C or C++ to accelerator designs. We find that the black-box heuristics in HLS tools can be unpredictable: changing parameters in the program that should improve performance can counterintuitively yield slower and larger FPGA implementations.
    This paper proposes a type system that restricts HLS to programs that can predictably compile to hardware accelerators. We implement the type system in Dahlia, a programming language that compiles to HLS C++, and evaluate how its type system can reduce the size of HLS parameter spaces while accepting Pareto-optimal designs.

Programming Languages Design and Implementation