Presentation: Automatically Retrofitting JIT Compilers

Laurence Tratt, a renowned researcher in language engineering at King’s College London, discusses yk, an open-source meta-tracing JIT compiler framework. The presentation focuses on automatically retrofitting JIT compilers to speed up C-based language interpreters like Lua and MicroPython with minimal code changes. Tratt shares a simple demonstration of a Mandelbrot program running in the standard Lua VM, which takes 3.2 seconds to execute.

When the yk technology is applied to the same Lua VM, the program completes in just 0.8 seconds, showcasing a 4x performance improvement. While the example used is cherry-picked for its effectiveness, Tratt emphasizes that improvements are observed across a wider set of Lua benchmarks, achieving a 2x average speedup. The talk also explores the application of yk to MicroPython, resulting in a roughly linear performance boost on the Fannkuch benchmark, improving execution time from 15 seconds to around 7-8 seconds.

The session aims to motivate the development of more efficient language implementations, detailing the technical challenges and progress made so far, while acknowledging the collaborative effort with Edd Barrett, Lukas Diekmann, and Pavel Durov. The work is funded by Shopify and the Royal Academy of Engineering, with Tratt highlighting the relatively small codebase changes introduced by yk, adding only 400 lines to the normal Lua VM.