Generating C: Heterogeneous metaprogramming system description

Heterogeneous metaprogramming systems use a higher-level host language to generate code in a lower-level object language. Their appeal is taking advantage of the module system, higher-order functions, data types, type system and verification tools of the host language to quicker produce high-performant lower-level code with some correctness guarantees.We present two heterogeneous metaprogramming systems whose host language is OCaml and object language is C. The first relies on offshoring: treating a subset of (MetaOCaml-generated) OCaml as a different notation for (a subset of) C. The second embeds C in OCaml in tagless-final style. The systems have been used in several projects, including the generation of C supersets OpenCL and OpenMP. Generating C with some correctness guarantees is far less trivial than it may appear, with pitfalls abound: e.g., local variables may only be introduced in statement context; mutable variables are not first-class. Maintenance has challenges of its own, e.g., extensibility. On many examples, we expound the pitfalls we have come across in our experience, and describe the design and implementation to address them.(c) 2023 Elsevier B.V. All rights reserved.

Automated summary

Heterogeneous metaprogramming systems leverage higher-level host languages to generate lower-level object language code, enabling faster production of high-performant code with correctness guarantees. This paper presents two systems with OCaml as the host language and C as the object language, discussing their implementation and applications.