http:^^softlib.rice.edu^mscp^mscp.html

来自「This data set contains WWW-pages collect」· HTML 代码 · 共 128 行

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

<H1>The Massively Scalar Compiler Group</H1>
<!WA0><a href=http://riceinfo.rice.edu><i>Rice University</i></a><br><i>Houston, Texas, USA</i>

  <TITLE>The Massively Scalar Compiler Group at Rice University</TITLE>

  <H2>Purpose</H1>

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.<p>

<H2>Mission</H2>

The goals of the Massively Scalar Compiler Project are
<UL>

<LI>	to develop new techniques in code optimization and code
	generation that represent real improvements over existing
	methods,<p>

<LI>	to transfer this knowledge to industrial compiler groups 
	in a way that enables them to quickly evaluate and deploy
	new methods, and <p>

<LI>	to implement and distribute a laboratory compiler that can
	serve as a testbed for evaluating new techniques in code
	optimization and code generation.<p>
</UL>
We view industrial compiler groups as our primary customers, 
with academic researchers as our second priority.<p>

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.<p>

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

<H2>Problems that we attack</H2>
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 <!WA1><a href="http://softlib.rice.edu/MSCP/problems.html">all solved in 1980.</a><p>

<h2>Methodology</h2>
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 
<!WA2><a href="http://softlib.rice.edu/MSCP/nuweb.html"><i>nuweb</i></a>,
a literate programming tool.
The implementation is intended to serve as a reference guide for 
reimplementation in a commercial setting.<p>

For the truly interested, a longer <!WA3><a href="http://softlib.rice.edu/MSCP/methodology.html">discussion</a>
is available.<p>

<h2> Personnel</h2>

<hr> 
<!WA4><img align=middle src=http://softlib.rice.edu/MSCP/group.gif> <p>
<em>The group scaling a massive object </em>
<hr>

The Massively Scalar Compiler Project is led by four researchers
<ol>
<li><!WA5><a href="http://softlib.rice.edu/MSCP/cooper.html">Keith D. Cooper</a>
<li><!WA6><a href="http://www.cs.rice.edu/~ken/index.html">Ken Kennedy</a>
<li><!WA7><a href="http://www.cs.rice.edu/~lts">Taylor Simpson</a>
<li><!WA8><a href="http://softlib.rice.edu/MSCP/torczon.html">Linda Torczon</a>
</ol>
The prime reason that these people are productive is because they have
a supporting cast of <!WA9><a href=http://softlib.rice.edu/MSCP/prog_list.html>staff</a> 
and <!WA10><a href=http://softlib.rice.edu/MSCP/grad_stud_list.html>graduate students</a>.<p>

<h2>Software</h2>
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 <!WA11><a href="http://softlib.rice.edu/MSCP/software.html">code</a> is available (in <i>nuweb</i> form) 
via anonymous ftp. 
</p>
The compiler uses a low-level intermediate language named ILOC.  
We have gathered together in one place the 
<!WA12><a href="http://softlib.rice.edu/MSCP/iloc.html">complete ILOC archives</a>,
including unpublished papers and newsletters. 
We hope that someone learns from our experience! 


<h2>Accomplishments</h2>
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 <!WA13><a href="http://softlib.rice.edu/MSCP/publications.html">papers</a> on our work in 
several different venues.
We write software and make it 
<!WA14><a href="http://softlib.rice.edu/MSCP/software.html">available</a> via the web.
For more information, we have a brief 
<!WA15><a href="http://softlib.rice.edu/MSCP/results.ps">summary of results</a> available online.


<h2>More information</h2>
Our most recent annual report can be found <!WA16><a href="http://www.ito.darpa.mil/Summaries95/D295--Rice.html">in the 
ARPA archives</a>.

<!WA17><a href=http://softlib.rice.edu/MSCP/talk.ps>Overhead transparencies</a> and 
<!WA18><a href=http://softlib.rice.edu/MSCP/poster.ps>poster materials</a> from the Spring 96 ARPA PI
meeting are also archived online.

<hr>
This page has been accessed 
<!WA19><img src="http://riceinfo.rice.edu/cgi-bin/counter.xbm?http://www.cs.rice.edu/MSCP/index.html">
times since we started counting in September 1995.


</body>
</html>