Date: Tue, 26 Nov 1996 00:18:23 GMT Server: NCSA/1.5.1 Last-modified: Fri, 23 Aug 1996 17:07:13 GMT Content-type: text/html Content-length: 4529

The Massively Scalar Compiler Group

Rice University
Houston, Texas, USA The Massively Scalar Compiler Group at Rice University

Purpose

The Massively Scalar Compiler Project is designing and building a laboratory compiler to explore new compiler-based code-improvement techniques for microprocessor-based systems. Our focus is on uniprocessor systems.

Mission

The goals of the Massively Scalar Compiler Project are We view industrial compiler groups as our primary customers, with academic researchers as our second priority.

Our primary focus is on problems that arise for uniprocessor, microprocessor-based systems. We try to ensure that our techniques work equally well in multiprocessor systems.

This work is supported by both ARPA and the State of Texas.

Problems that we attack

In general, we investigate problems that arise in compiling code for uniprocessor microprocessor-based systems. Our particular areas of interest are code improvement techniques (often misnamed "optimization") and code generation issues. A common misconception is that these problems were all solved in 1980.

Methodology

As a group, we have a strong belief in the development of new algorithms coupled with experimental evaluation. We view research and development as two halves of a whole effort. Each of our research activities is directly coupled to a development activity. It is our intent to build a reference implementation for each new technique that proves its worth; these are written in nuweb, a literate programming tool. The implementation is intended to serve as a reference guide for reimplementation in a commercial setting.

For the truly interested, a longer discussion is available.

Personnel


The group scaling a massive object


The Massively Scalar Compiler Project is led by four researchers
  1. Keith D. Cooper
  2. Ken Kennedy
  3. Taylor Simpson
  4. Linda Torczon
The prime reason that these people are productive is because they have a supporting cast of staff and graduate students.

Software

One of the prime work products of the project is software. As a phase of the compiler achieves a semblance of stability and maturity, we place it into the pool of released software. The code is available (in nuweb form) via anonymous ftp.

The compiler uses a low-level intermediate language named ILOC. We have gathered together in one place the complete ILOC archives, including unpublished papers and newsletters. We hope that someone learns from our experience!

Accomplishments

This project produces several kinds of results. The first, of course, is students. Preston Briggs of Tera Computers and Cliff Click of Hewlett-Packard are both recent graduates. We publish papers on our work in several different venues. We write software and make it available via the web. For more information, we have a brief summary of results available online.

More information

Our most recent annual report can be found in the ARPA archives. Overhead transparencies and poster materials from the Spring 96 ARPA PI meeting are also archived online.
This page has been accessed times since we started counting in September 1995.