SCM Repository
| [smlnj] / sml / trunk / READMES / 110.28-README.html |
View of /sml/trunk/READMES/110.28-README.html
Parent Directory
| 
Revision Log
Revision 637 -
(download)
(as text)
(annotate)
Tue May 2 16:24:56 2000 UTC (22 years, 2 months ago) by dbm
File size: 12110 byte(s)
Tue May 2 16:24:56 2000 UTC (22 years, 2 months ago) by dbm
File size: 12110 byte(s)
created 110.28 README files
<!DOCTYPE HTML PUBLIC "-//IETF//DTD HTML//EN">
<html>
<head>
<title>SML/NJ Version 110.28 NEWS</title>
</head>
<body bgcolor="white">
<center><h1>Standard ML of New Jersey<BR>
Version 110.28, May 1, 2000</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>
<UL>
<LI> This version has some tuning of phase ordering for FLINT, which
mostly solves the excessive space problem during compilation. With
this tuning, 110.28 is slightly better than 110.25 in performance,
but not quite as good as 110.25+latest MLRISC. See the benchmarks
below. Futher work on the register allocator in MLRISC should
improve space performance further.
<p>
<LI>CM:
Installation scripts have been tweaked. A new noweb tool has been
added. ".fun" suffix has been added to those that imply sml compilation.
New version numbering scheme added. Alternative syntax for anchored
paths added. Parallel make made smarter. Support for "unsharing".
Simple "makedepend" functionality added. Further details below.
<p>
See src/system/README and the latest version of the CM manual at
<blockquote>
<//http://www.kurims.kyoto-u.ac.jp/~blume/SMLNJ-DEV/manual/index.html>
<//http://www.kurims.kyoto-u.ac.jp/~blume/SMLNJ-DEV/manual.ps>
</blockquote>
for further information about visible changes.
<p>
<li> A few MLRISC changes, most not SML/NJ related.
<p>
<li> ML-Yacc: updated manual and examples to be SML '97, SML/NJ 110+ compliant.
Examples now compiled with CM.
<li> Reported bug fixes:
<pre>
1498. Specialized real arrays not pretty-printed correctly
1510. Signature matching bug makes "casts" possible
1562. CM complains about unrecognized file extension ".fun"
1563. redundant pathconfig contents with multiple builds
</pre>
<p>
<hr>
<h2> Change Details </h2>
<center><h3> FLINT </h3></center>
Tuned ordering of phases to improve time/space performance. See
src/compiler/FLINT/opt/flintopt.txt for description of some of the
issues.
<p>
<hr>
<center><h3> MLRISC </h3></center>
This update synchronizes my repository with Yale's. Most of these
changes, however, do not affect SML/NJ at all (the RA is an exception).
<p>
<ol>
<li>Register Allocator
<ol type="a">
<li> An improvement in the interference graph construction:
Given a copy
<pre>
s <- t
</pre>
no interference edge between s and t is added for this definition of s.
<p>
<li> I've added two new spill heuristic modules that Fermin and I developed
(in the new library RA.cm). These are unused in SML/NJ but maybe
useful for others (Moby?)
<p>
<li> Added a flag "ra-dump-size" to print out the size of the flowgraph
and the interference graph.
</ol>
<li> X86
<ol type="a">
<li> Various fixes in the backend provided by Fermin [C--] and Lal.
</ol>
<p>
<li> Alpha
<ol type="a">
<li> Added the BSR instruction and code generation that goes with it [C--]
<p>
<li> Other fixes too numerous to recount provided by Fermin [C--]
<p>
<li>PSEUDOARITH was missing in AlphaRewrite. This causes an endless
loop in C--.
</ol>
<p>
<li> Regmaps
<ol type="a">
<li> The regmaps are not initialized with the identity physical bindings
at creation time. This is unneeded.
</ol>
<p>
<li> MLRISC Optimizations
<ol type="a">
<li> The DJ-Graph module can now compute the iterated dominance frontiers
intersects with liveness incrementally in linear time! Woohoo!
This is now used in my new SSA construction algorithm.
<p>
<li> THe branch reorganization module is now smarter about linear chains of
basic blocks.
</ol>
</ol>
<p>
<center><h3> CM </h3></center>
<ul>
<li>Several manual updates
<blockquote>
I always try to keep the manual in sync with CM's latest features.
</blockquote>
<p>
<li>New noweb tool
<blockquote>
The existing system is entirely unaffected by this, but some CM users
have asked for renewed noweb support. Everything is documented in the
CM manual.
<p>
New (plugin) libraries:
<pre>
noweb-tool.cm
nw-ext.cm
</pre>
</blockquote>
<p>
<li>CM version numbering added
<blockquote>
This is an implementation of Lal's proposal for adding version
numbers and version checking to .cm files. Lal said that his
proposal was just that -- a proposal. For the time being I went
ahead and implemented it so that people can comment on it.
Everything is completely backward-compatible (except for the stable
library format, i.e., new bootfiles!).
<p>
As usual, see the CM manual for details.
</blockquote>
<p>
<li> Alternative syntax for anchored paths
<blockquote>
Dave has recently voiced the same concerns that I had when I did
this, so there should be some support. My take is that eventually
I will let support for the current syntax (where anchors are
"implicit") fade out in favor of the new, explicit syntax.
In order to be backward-compatible, both old and new syntax are
currently supported.
<p>
Again, see the CM manual for details.
</blockquote>
<p>
<li> Parallel make is trying to be slightly smarter
<blockquote>
When the master process finds a "bottleneck", i.e., when there is
only one compilation unit that can be compiled and everybody else
is waiting on it, then it will simply compile it directly instead
of clumsily telling one of the slaves to do it.
</blockquote>
<p>
<li> Support for "unsharing" added
<blockquote>
This is necessary in order to be able to have two different
versions of the same library running at the same time (e.g., for
trying out a new MLRISC while still having the old MLRISC linked
into the current compiler, etc.) See the CM manual.
</blockquote>
<p>
<li> makedepend
<blockquote>
Simple "makedepend" functionality added for generating Makefile
dependency information. (This is rather crude at the moment.
Expect some changes here in the future.)
</blockquote>
<p>
<li> New sml class prefix
<blockquote>
".fun" added as a recognized suffix for ML files. Also documented
explicitly in the manual that the fallback behavior (unknown suffix
-> ML file) is not an official feature!
</blockquote>
<p>
<li>Pickler
<blockquote>
Small changes to the pickler for stable libraries.
<blockquote>
<p>
<li>Internal cleanup
<blockquote>
Several internal changes to CM (for cleanup/improvement).
</blockquote>
<p>
<li>install.sh changes
<blockquote>
I changed config/install.sh to remove duplicate entries from the
lib/pathconfig file at the end. Moreover, the final version of
lib/pathconfig is sorted alphabetically. The same (sorting) is done
in src/system/installml.
<p>
The config/install.sh script now consistently uses relative
pathnames in lib/pathconfig whenever the anchor is in the lib
directory. (So far this was true for the libraries that come
pre-compiled and bundled as part of the bootfiles but not for
libraries that are compiled by the script itself.)
<p>
Changed install.sh script to handle archive files without version number
and to use "boot.<arch>-<os>" instead of "sml.boot.<arch>-<os>" for the
name of the boot file archive.
</blockquote>
</ul>
<p>
<hr>
<center><h3> BENCHMARKS </h3></center>
Allen Leung, 28 April 2000:
<p>
I've rerun the benchmarks to see if anything has slowed down in
MLRISC recently, but found nothing. I compared 110.25, 110.25+latest
MLRISC, and 110.27+ (which also has the latest MLRISC). 110.25 uses
cpsopt, and 110.27+ uses flintopt. [110.27+ is repository state
at the end of April, which is essentially 110.28 - dbm].
<p>
The results are:
<pre>
110.25 versus 110.25+latest MLRISC
----------------------------------
Name Compilation Runtime Speedup
barnesHut 5.560 5.368 3.57% 3.690 3.302 11.76%
boyer 8.678 7.672 13.12% 0.313 0.312 0.53%
count-graphs 2.048 1.915 6.96% 28.577 27.128 5.34%
fft 1.325 1.233 7.43% 1.060 0.982 7.98%
knuthBendix 5.337 4.680 14.03% 0.962 0.918 4.72%
lexgen 10.310 9.950 3.62% 0.917 0.902 1.66%
life 1.230 1.127 9.17% 0.128 0.100 28.33%
logic 3.533 3.258 8.44% 5.750 5.413 6.22%
mandelbrot 0.145 0.148 -2.25% 0.700 0.685 2.19%
mlyacc 34.557 33.342 3.64% 0.553 0.533 3.75%
nucleic 6.675 6.507 2.59% 0.173 0.167 4.00%
ratio-regions 6.358 6.218 2.25% 119.753 120.772 -0.84%
ray 2.280 2.247 1.48% 3.563 3.517 1.33%
simple 9.798 9.650 1.54% 2.987 3.083 -3.14%
tsp 1.763 1.528 15.38% 8.657 7.718 12.16%
vliw 33.938 32.570 4.20% 1.982 2.000 -0.92%
Average speedup: 5.32%
Average compile time speedup: 5.95%
</pre>
As you can see, MLRISC on x86 has improved slightly since the
110.25. Now comparing 110.25 with 110.27+:
<pre>
110.25 versus 110.27+
---------------------
Name Compilation Runtime Speedup
barnesHut 5.560 5.568 -0.15% 3.690 3.347 10.26%
boyer 8.678 10.058 -13.72% 0.313 0.322 -2.59%
count-graphs 2.048 2.125 -3.61% 28.577 31.533 -9.38%
fft 1.325 1.297 2.19% 1.060 1.052 0.79%
knuthBendix 5.337 3.737 42.82% 0.962 1.175 -18.16%
lexgen 10.310 9.985 3.25% 0.917 0.992 -7.56%
life 1.230 1.105 11.31% 0.128 0.125 2.67%
logic 3.533 3.302 7.02% 5.750 5.357 7.34%
mandelbrot 0.145 0.162 -10.31% 0.700 0.753 -7.08%
mlyacc 34.557 37.292 -7.33% 0.553 0.493 12.16%
nucleic 6.675 7.717 -13.50% 0.173 0.165 5.05%
ratio-regions 6.358 3.915 62.41% 119.753 126.153 -5.07%
ray 2.280 2.142 6.46% 3.563 2.827 26.06%
simple 9.798 10.067 -2.67% 2.987 2.812 6.22%
tsp 1.763 1.605 9.87% 8.657 8.870 -2.41%
vliw 33.938 46.907 -27.65% 1.982 1.802 9.99%
Average speedup: 1.77%
Average compile time speedup: 4.15%
110.25+latest MLRISC versus 110.27+
-----------------------------------
Name Compilation Runtime Speedup
barnesHut 5.368 5.568 -3.59% 3.302 3.347 -1.34%
boyer 7.672 10.058 -23.73% 0.312 0.322 -3.11%
count-graphs 1.915 2.125 -9.88% 27.128 31.533 -13.97%
fft 1.233 1.297 -4.88% 0.982 1.052 -6.66%
knuthBendix 4.680 3.737 25.25% 0.918 1.175 -21.84%
lexgen 9.950 9.985 -0.35% 0.902 0.992 -9.08%
life 1.127 1.105 1.96% 0.100 0.125 -20.00%
logic 3.258 3.302 -1.31% 5.413 5.357 1.06%
mandelbrot 0.148 0.162 -8.25% 0.685 0.753 -9.07%
mlyacc 33.342 37.292 -10.59% 0.533 0.493 8.11%
nucleic 6.507 7.717 -15.68% 0.167 0.165 1.01%
ratio-regions 6.218 3.915 58.83% 120.772 126.153 -4.27%
ray 2.247 2.142 4.90% 3.517 2.827 24.41%
simple 9.650 10.067 -4.14% 3.083 2.812 9.66%
tsp 1.528 1.605 -4.78% 7.718 8.870 -12.98%
vliw 32.570 46.907 -30.56% 2.000 1.802 11.01%
Average speedup: -2.94%
Average compile time speedup: -1.68%
<pre>
Overall, I'd say 110.27+ is pretty competitive with 110.25. There are
some big improvements (ray, vliw, mlyacc) which can only be attributed to FLINT
changes. But something has slowed down (tsp, knuth-bendix, life, mandelbrot,
count-graphs). [We know the reasons with knuth-bendix.]
Also, compilation time in 110.27+ is generally slower (which is offset
by huge improvements in knuth-bendix and ratio-regions).
<p>
<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: Tue May 2 12:23:30 EDT 2000
<!-- hhmts end -->
</font>
</blockquote>
</body>
</html>
| root@smlnj-gforge.cs.uchicago.edu | ViewVC Help |
| Powered by ViewVC 1.0.0 |
