Per Gustafsson
I am a member of the High
Performance Erlang (HiPE)
group at the department of Information Technology at Uppsala
University. We work to improve the performance and usability of the concurrent
functional language Erlang.
In the group we work on several different projects.
- We maintain and develop a native code compiler for Erlang.
This compiler is included in the Erlang/OTP distribution of Erlang
- We develop an SMP aware version of the Erlang runtime system.
This will be available in a Beta version in Erlang/OTP R11.
- We investigate different garbage collection techniques for
functional concurrent languages
- We develop a tool called Dialyzer
which finds potential bugs in Erlang programs by analyzing the type
information present in Erlang programs
My research
I work on improving support for manipulating bit streams in Erlang.
Since Erlang's most important application domain is telecommunications
programming many programs deal with communication based on
bitstreams. To simplify these tasks a pattern matching syntax was
developed.
My first project was to show how our native code compiler could
implement this efficiently without producing to much code through
selective translation. This is described in [5]. This paper did not
deal with any high level aspects, but it only showed how a particular
set of virtual machine instructions could be translated into a register
transfer language.
To make this low level pattern matching efficient it is important to
perform high level optimizations as well. In [3] and more extensively
in [1] we describe how pattern matching compilation could be
adjusted to be used to compile pattern matching on bit streams. We
devised one tree based method to get maximal performance as well as a
guarded sequential automaton based method to minimize code size.
Unfortunatly the syntax for pattern matching in Erlang suffer from some
issues, in particular the binary datatype was based on a byte stream.
This creates a lot of extra work for the users of this construct when
they deal with bit streams forcing them to manually calculate how to
pad their bit streams in order to store it in a byte stream. In [2] we
present a proposal on how to extend the syntax to avoid this issue as
well as some other problems. In addition we introduce the notion of
binary comprehensions similar to list comprehensions.
In [6] we describe how a true bit stream data structure can be
implemented in Erlang and how that makes many applications a lot
simpler to write. In addition to this we compare the result of this
approach with other languages both in terms of performance and in
terms of conciseness. The benchmarks that we use in this paper
is available from this
webpage.
In [7] we describe a formal semantics for pattern matching on bit
streams in the context of a small functional language. This paper also
include most of the work from [6] as this is an earlier longer version of that paper
Publications
[1] Efficient
manipulation of binary data using pattern matching
with Konstantinos Sagonas
in the Journal of
Functional Programming 16(1), pages 35–74, January 2006.
bibtex
[2]
Bit-Level Binaries and Generalized Comprehensions in Erlang
with Konstantinos Sagonas
in the Erlang Workshop
(ERLANG 05).
bibtex
[3] Adaptive
Pattern Matching on Binary Data
with Konstantinos Sagonas
in the European Symposium on
Programming (ESOP 2004).
bibtex
[4] All You
Wanted to Know About the HIPE Compiler (and might have been afraid to
ask)
with K. Sagonas, M. Pettersson, R. Carlsson and T. Lindahl
in the Erlang Workshop
(ERLANG 03).
bibtex
[5] Native
Code Compilation of Erlang's Bit Syntax
with Konstantinos Sagonas
in the Erlang Workshop
(ERLANG 02).
bibtex
[6]
Applications, Implementation and Performance Evaluation of Bit Stream Programming in Erlang
with Konstantinos Sagonas
accepted for PADL'07
Unpublished
[7]
Functional Manipulation of Bit Streams
with Konstantinos Sagonas
Teaching