Overview - SUIF 2.0 - List of Optimizations - Schedule - People - Sponsors

The SUIF compiler infrastructure is a part of the national compiler infrastructure project funded by Darpa and NSF. It is a collaboration between researchers in four universities (Harvard University, Massachusetts Institute of Technology, Rice University, Stanford University, UC Santa Barbara), and one industrial partner, Portland Group Inc. The infrastructure is based on the SUIF (Stanford University Intermediate Format) parallelizing compiler developed at Stanford University.

The primary objective in the SUIF compiler design is to develop an extensible system that supports a wide range of current research topics including parallelization, object-oriented programming languages, scalar optimizations and machine-specific optimizations. We strive to develop an architecture that is modular, easy to extend and maintain, and supportive of software reuse. The compiler will be tested with standard benchmark programs. As it is not possible for us to create the ultimate system, our hope is to make the system extensible and attractive enough so that the community can band together to create an exciting system for research. Comments and contributions are welcome!

The sources of the SUIF system will be freely available for research and commercial use. The only exception is that the sources of the Digital Fortran front end need to be licensed from Digital.

Overview of the SUIF Compiler Infrastructure

Intermediate Format SUIF 2.0

A revised SUIF program representation (SUIF 2.0) will be released with this new system. The most notable feature in SUIF 2.0 is that the program representation will be user-extensible. This feature is motivated by the extension of the compiler to better support other programming languages and machine-level optimizations. Converters between SUIF 1.x and 2.0 representations will be provided to support backward compatability. More information will be forthcoming in the next few months.

List of Optimizations

Interprocedural Parallelization

Scalar and array data dependence analysis

Scalar and array privatization

Reduction recognition on scalar and array variables

Pointer analysis

Program transformations for parallelism and locality

Affine loop transformations

unimodular transform

loop distribution

loop fusion

loop reindexing

loop scaling

statement reordering

Blocking

Data transformations

Object-oriented programming languages

Capturing object definitions

Virtual function call elimination

Garbage collection

Machine-independent scalar analysis and optimizations

Static single assignment

Dead code elimination

Partial redundancy elimination

Conditional constant propagation

Global value numbering

Strength reduction

Reassociation of expressions

Machine-dependent optimizations

Instruction scheduling

Profiling and instrumentation tools

Schedule

Year 1: SUIF infrastructure base; Sequential Gcc and Fortran compilers
Prototype of object-oriented SUIF
Definition of machine-level SUIF
Year 2: Parallelization and loop transformations
x86 back end
Fortran, Java, C++ front end connections
Partial completion: scalar optimization, register allocation, instruction scheduling
Year 3: Complete system release; Interprocedural analysis

The SUIF Compiler Team

Institution	Principal Investigators	Subjects
Stanford University	Monica Lam	Base SUIF system Interprocedural Parallelization Pointer Analysis
UC Santa Barbara MIT	Urs Holzle Martin Rinard	Object-oriented programming languages
Rice University	Keith Cooper Linda Torczon	Scalar optimizations
Harvard University	Mike Smith	Machine-dependent optimizations
Portland Group Inc.	Vince Schuster	EDG front end connection Validation, release, support

Be sure to visit our home page to learn more about the SUIF compiler system!

If you have any questions or comments, please contact the webmaster.

Last updated on January 12, 1998.