Home My Page Projects Code Snippets Project Openings SML/NJ
Summary Activity Forums Tracker Lists Tasks Docs Surveys News SCM Files

SCM Repository

[smlnj] View of /pages/trunk/NEWS/110.23-README.html
ViewVC logotype

View of /pages/trunk/NEWS/110.23-README.html

Parent Directory Parent Directory | Revision Log Revision Log


Revision 953 - (download) (as text) (annotate)
Thu Oct 11 09:52:12 2001 UTC (18 years, 7 months ago) by macqueen
File size: 4253 byte(s)
Initial revision
<!DOCTYPE HTML PUBLIC "-//IETF//DTD HTML//EN">
<html>
  <head>
    <title>SML/NJ Version 110.23 NEWS</title>
  </head>

  <body bgcolor="white">
   <center><h1>Standard ML of New Jersey<BR>
        Version 110.23, October 9, 1999</h1>
   </center>

   <center>
      <tt> http://cm.bell-labs.com/cm/cs/what/smlnj/index.html </tt>
   </center>

<blockquote>
<center>
<h2> Warning </h2>
</center>
 <blockquote>
 <blockquote>
    <em>
  	This version is intended for compiler hackers.  
	We are in the midst of substantial structural changes, 
	and this is a snapshot. 
    </em>
 </blockquote> 
 </blockquote>
      
<hr>

<center><h2>Summary:</h2></center>
  This version has some major MLRISC changes (from Allen Leung) and
  bug fixes for CM.   
  The primary MLRISC changes include a sparse algorithm for
  interference  graph construction, dead copy elimination, propagation 
  of branch frequency information, and makes use of a new annotation
  mechanism throughout the system.

<hr>

<center><h2> CM </h2></center>

  Fixed a bug related to consistent handling of pathnames that point
  to the same file but are not identical. (This situation can happen 
  because of path anchors or symbolic links).  
  <p>
  Fixed a  potential exponential blowup in CM.make pointed out
  by Simon Helsen.

<hr>

<center><h2> MLRISC</h2></center>
  <h4> Register Allocator </h4>
     This release uses a brand new graph coloring register allocator.  
     The new register allocator is slightly faster (20% in 
     the register allocator for RISC machines; no change for x86), 
     generates less spill code (about 10% on RISC machines) 
     than the previous one, and has eliminated a few bugs.  It has 
     also incorporated many algorithmic improvements (but the framework 
     is still based on iterated coalescing). 
     Some of the improvements are:
     <p> <ul>
     
     <li> Sparse interference graph construction -- the new algorithm
       uses a sparse method for constructing the interference graph.  
       This is based on the algorithm described in Modern Compiler 
       Implementation, by A.Appel. Since our input is mostly single
       assignment (CPS), this phase runs in effectively linear-time.
     <p><li>
       Dead copy elimination -- an additional phase for removing 
       dead copies is added.  This allows us to remove about 59k of 
       code on the x86 platform. 
     <p><li>
       In addition, these new features are added but not currently used for
       SML/NJ:

	<p><ul>
        <li> coalescing and spilling decisions can now be prioritized by execution
          frequencies

        <li> coalescing and coloring of spill locations 
        </ul>
     </ul>
     <p>
     Please see the 
     <A HREF="http://cm.bell-labs.com/cm/cs/what/smlnj/compiler-notes/index.html#NEW-RA"> paper </A> 
     for more details.

<hr>
<center><h2>Annotations</h2></center>
       The interface of annotations has been changed in this version.
       It is now an abstract type.  See the files 
       <tt>MLRISC/library/annotations.{sig,sml}</tt> and
       <tt>MLRISC/instructions/basicAnnotations.sig</tt> for details. 
       <p>
       A few
      <A HREF="http://cm.bell-labs.com/cm/cs/what/smlnj/compiler-notes/index.html#ANNOTATIONS"> notes </A>
       describing this change are available. 

<h4>Block names</h4>

       Block names are now implemented using basic block annotations instead
       of a special data constructor.

<h4>MLTREE</h4>
       Various fixes dealing with integer to integer conversions have been 
       added.  

<h4>CM Libraries</h4>
       All the target specific files have been turned into libraries. 
       MLRISC itself has been broken up into several libraries a core 
       defined by MLRISC.cm and various extensions.
      <p>
       The new CM library mechanism makes it particularly easy to 
       autoload new optimization phases at the top level. See
       compiler/MLRISC-OPTIMIZER.cm
<hr>

    <font size=-2>
    <address><a href="mailto:george@research.bell-labs.com">
		Lal George</a></address>
<!-- Created: Thu Aug  6 00:13:09 EDT 1998 -->
<!-- hhmts start -->
Last modified: Wed Oct  6 21:37:30 EDT 1999
<!-- hhmts end -->
    </font>
</blockquote>
</body>
</html>

root@smlnj-gforge.cs.uchicago.edu
ViewVC Help
Powered by ViewVC 1.0.0