1. Just-In-Time Compilation Keith W. Krajewski 3/4/2011 paper: A Brief History of Just-In-Time (2003) John Aycock

2. Overview What is JIT? Why know about JIT? Who has used JIT (and to what end)? Simulation JIT in modern languages Classifying JIT techniques Tools for JIT compilation

3. What is JIT? Dynamic compilation “JIT” itself a new term to computing, ~1993 Theoretically unnecessary

4. Goals of JIT Compiled programs are faster Interpreted programs are smaller Interpreted programs are more portable Interpreter has access to runtime info

5. Why know about JIT? Because JIT is like the cure for scurvy.

6. Who has used JIT (and how)? Early sightings LC^2 APL mixed/throw-away code Fortran Smalltalk Self Oberon Erlang O'caml Modern stuff

7. Ancient Runes McCarthy, Lisp (1960), punch cards! Univ. of Michigan IBM 7060 (1966) Thompson, regular expressions (1968)

8. LC^2 Language for conversational computing Mitchell, Perlis, & van Zoeren (1968) Mostly forgettable Cached the actions it performed

9. APL Phillip S. Abrams, 1970 Drag-along Beating

10. Mixed & Throw-Away Code Mixed – Dakin & Poole (1973) & Dawson (1973) Throw-away – Basic

11. Fortran Hansen's 1974 optimization Frequency-of-execution counter Threshold levels Ordered set of machine-specific optimizations

12. Smalltalk Lazy compilation to native code No pagination of compiled code

13. Self David Ungar, Randall Smith, 1987 Pure OO, dynamically typed 3 generations Customization (Pitfall: over-customization) Optimized type info for loops (Pitfall: compilation time) Illusion of speed Influenced Sun's later works (Java)

14. Oberon Niklaus Wirth, 1986 Compiled to “slim binary” Supported dynamic loading of modules Continuous re-optimization in background

15. Erlang Ericsson, 1986 HiPe – explicit invocation Mixed code, switch upon method invocation or exception

16. O'caml Specialized the interpreter's instruction set to include “macro opcodes” … without much benefit

17. Simulation 1 st gen – straight 1 by 1 translation 2 nd gen – 1 by 1 translation with cache 3 rd gen – blocks of source 4 th gen Profiled execution Hot path detection Code generation Bailout

18. HP Dynamo Compiled HPA-8000 code into HPA-8000 code...amazingly, this worked. 30% performance increase Important to be able to bail out

19. Java Static explicit compilation to JVM bytecode Early JVM did straight interpretation Motivation for JIT research Now very portable and relatively quick

20. C#/.Net Static compilation to CIL (MSIL), then bytecode Dynamic per-method compilation to native machine

21. Classifying JIT 3 properties Invocation – explicit/implicit Executability – monoexecutable/polyexecutable Concurrency

22. Toolkits for JIT compilation Keppel (1991), Engler & Proebsting (1994), Ramsey & Fernandez (1995), Engler (1996), Fraser & Proebsting (1999) Address three main problems Binary code generation Cache coherence Execution

23. Summary Dynamic compilation Combines best features of both compiled and interpreted programs Lazy compilation, incremental optimization, mixed code, throw-away code, hotspot detection, machine-specific optimization, customization, simulation Don't re-invent the wheel again

24. Works cited Aycock, John. A Brief History Of Just-In-Time. 2003. Petzold, Charles..NET Book Zero. 2006. http://www.charlespetzold.com/dotnet/DotNetBo okZero11.pdf Wikipedia: Just-In-Time Compilation. http://en.wikipedia.org/wiki/JIT_compilation