My PhD Thesis

Title: Meta-Tracing Just-in-Time Compilation for RPython

Abstract: Many dynamic languages are implemented using traditional interpreters because implementing just-in-time (JIT) compilers for them is too complex. This limits their performance and restricts their applicability. This thesis describes meta-tracing, a technique that makes it possible to apply a single tracing JIT compiler across a wide variety of languages. Meta-tracing is flexible enough to incorporate typical JIT techniques such as run-time type feedback and unboxing. With the help of several case studies, meta-tracing is applied to various language implementations. Performance measurements of using meta-tracing show improvements by up to factor 50 for implementations of production languages. Meta-tracing can thus simplify the implementation of high-performance virtual machines for dynamic languages significantly, making these languages more practical in a wider context.

PDF: http://cfbolz.de/stuff/cfbolz-meta-tracing.pdf

Official archived version: http://d-nb.info/1057957054

Source Code

The following Mercurial repositories contain code that the thesis describes:

The main PyPy repository, containing the RPython translation toolchain, MetaJIT (Chapters 3, 4, 5, 6) and the PyPy Python interpreter (Chapter 10).
The regular expression repository contains the regular expression engine of Chapter 9.
The Pyrolog repository contains the Pyrolog Prolog interpreter described in Chapter 11.
The benchmarks repository contains the code necessary to re-do the benchmarks in Chapters 4, 6, 9, 10, 11.