Restructuring Compilers for Parallel Architectures
Members of the project team
Research objectives
A restructuring compiler is an important tool for retargeting existing
serial software to parallel architectures.
Existing restructuring compilers often lack flexibility: transformations are
hard coded and there are no possibilities to control their application
sequence, which means that these compilers behave like a black box that
cannot be controlled.
Also, these compilers are directed towards a certain target architecture.
The goal of the present research effort
is to enhance the flexibility of compilers in a number of ways.
A restructuring compiler for full Fortran 77 is being developed that offers
an interactive environment for the application of program transformations.
The transformations themselves can be specified in a transformation
definition language.
This language is highly expressive accommodating all usual transformations.
Transformations that are specific for a certain target architecture can be
formulated making the compiler suitable for a range of architectures.
The language also incorporates features for expressing global
transformations.
Furthermore, mechanisms are being studied for guiding the application of
transformations by means of strategies.
A strategy will select (semi) automatically an optimal sequence.
These strategies in turn can be specified by the user, enabling the use of
different strategies for different architectures.
Collaborations
Through the Esprit BRA project APPARC there exists collaboration with
F. Bodin and W. Jalby (IRISA Rennes), M. Gerndt (KFA Juelich), and
R.H. Perrot (Queen's University of Belfast).
Deliverables
An implementation of a powerful restructuring compiler, together with
an interactive environment for program transformations consisting of
a Transformation Definition Language, and a Transformation Application
Engine. The definition of a range of transformations in the language.
Recent publications
- P.J.H. Brinkhaus, Compiler Analysis of Procedure Calls,
Master Thesis INF/SCR-93-13, Utrecht University.
- P.M.W. Knijnenburg and H.A.G. Wijshoff,
On improving data locality in sparse matrix computations,
technical report 94-15, Leiden University.
- A.J.C. Bik and H.A.G. Wijshoff,
MT1 - A prototype restructuring compiler,
technical report 93-32, Leiden University.
Last modified on August 15, 1997 by Lex Wolters.