Скачать или смотреть An Efficient Compiler for the Gradually Typed Lambda Calculus

An Efficient Compiler for the Gradually Typed Lambda Calculus
by Andre Kuhlenschmidt (speaker), Deyaaeldeen Almahallawi, Jeremy G. Siek, Indiana University, USA

Scheme Workshop 2018
ICFP (International Conference on Functional Programming)
Scheme Workshop 2018
https://icfp18.sigplan.org/track/sche...
Fri 28 Sep 2018
Union Station Hotel
St. Louis, Missouri, USA


Abstract

Gradual typing combines static and dynamic typing in the same program. One would hope that the performance in a gradually typed language would range between that of a dynamically typed language and a statically typed language. Existing implementations of gradually typed languages have not achieved this goal due to overheads associated with runtime casts. Takikawa et al. (2016) report up to 100× slowdowns for partially typed programs. In this paper we present a compiler, named Grift, for evaluating implementation techniques for gradual typing. We take a straightforward but unexplored implementation approach for gradual typing, that is, ahead-of-time compilation to native assembly code using space-efficient coercions.

Our experiments show that this approach achieves performance on par with OCaml on statically typed programs but is slower than other dynamically typed languages on untyped programs. On partially typed code, the speedup with respect to untyped code ranges from 0.42× to 19.90× across the benchmarks. We implement casts using coercions and show that coercions eliminate catastrophic slowdowns without introducing significant overhead in all benchmarks. Across the benchmarks, coercions range from 9% slower (fft) to 13.65× faster (quicksort) than regular casts. We also implement the monotonic references and show that they eliminate all overhead in statically typed code. For partially typed code, they are faster on average than the traditional proxied references, sometimes up to 1.65×.