A Primer of Just In Time (JIT) compilations

With a bias towards applications to scientific computing

Dr. Kolen Cheung, Research Software Engineer

Research Software & Analytics Group, University of Exeter

August 5th, 2026

Introduction

Why this talk

TODO — one line framing the problem

TODO: the observation

TODO

TODO: what this talk covers

TODO

TODO — scope note, e.g. what is deliberately left out.

Ahead-of-time vs just-in-time

Just-in-time

  • TODO
  • TODO
  • TODO

Ahead-of-time

  • TODO
  • TODO
  • TODO

TODO — the panels read as good/bad by default; recolour or reword if the comparison is meant to be neutral.

How a JIT gets to machine code

Trace

TODO

Specialise

TODO

Optimise

TODO

Emit

TODO

TODO

JIT in scientific computing

The landscape

TODO: option A

TODO

TODO: option B

TODO

What it costs, what it buys

TODO

TODO — metric one

TODO

TODO — metric two

TODO

TODO — metric three

TODO — cite the benchmark these numbers come from.

When JIT is the wrong tool

TODO

TODO

TODO

TODO

Wrapping up

Takeaways

  • TODO
  • TODO
  • TODO

Further reading: TODO

Discussion

Discussion

Questions, comments, or anything you have hit in your own work?