JIT, A Maze of Twisty Little Traces
You are in a maze of twisty little traces, all alike. Last year at EUC I gave an in-depth tour of how automated source-to-source transformations were used to generate a proof-of-concept JIT compiler from the VM sources. Since then work has progressed to turn the proof-of-concept into a prototype.
Turning a proof-of-concept into a prototype is a process with many degrees of freedom, not unlike an old-school adventure game where you start out with minimal knowledge of your surroundings and the path to your goal. Compared to a proof-of-concept, a prototype not only has to behave correctly, it must also perform well but the way to get there is not apparent without a lot of experimentation.
This talk will provide a brief introduction to the just-in-time compiler followed by a walk-through of last year’s development. The walk-through will cover the changes made to the initial proof-of-concept to increase performance and also show how the current version of the just-in-time compiler compares to the unmodified system and HiPE.
Talk objectives: Update the Erlang community on the progress of the just-in-time compiler for Erlang.
Target audience: Developers interested in source-to-source transformations, virtual-machine implementation and optimization.
Frej Drejhammar The Brains behind the Erlang JIT
Swedish Institute of Computer Science
Frej Drejhammar received his M.Sc. degree in Computer Science from the Swedish Royal Institute of Technology (KTH) in 2001 followed by a Ph.Lic. from the same institution in 2005. He is currently a senior researcher at the Swedish Institute of Computer Science (SICS) in Kista, Sweden. His research interests are programming languages, protocols and tools for reliable and distributed computer systems.