| 1 : |
blume |
780 |
<!DOCTYPE HTML PUBLIC "-//IETF//DTD HTML//EN">
|
| 2 : |
|
|
<html>
|
| 3 : |
|
|
<head>
|
| 4 : |
|
|
<title>SML/NJ Version 110.31 NEWS</title>
|
| 5 : |
|
|
</head>
|
| 6 : |
|
|
|
| 7 : |
|
|
<body bgcolor="white">
|
| 8 : |
|
|
<center><h1>Standard ML of New Jersey<BR>
|
| 9 : |
|
|
Version 110.31, December 8, 2000</h1>
|
| 10 : |
|
|
</center>
|
| 11 : |
|
|
|
| 12 : |
|
|
<center>
|
| 13 : |
|
|
<a href=http://cm.bell-labs.com/cm/cs/what/smlnj/index.html>
|
| 14 : |
|
|
Home page</a>
|
| 15 : |
|
|
</center>
|
| 16 : |
|
|
|
| 17 : |
|
|
<blockquote>
|
| 18 : |
|
|
<center>
|
| 19 : |
|
|
<h2> Warning </h2>
|
| 20 : |
|
|
</center>
|
| 21 : |
|
|
<blockquote>
|
| 22 : |
|
|
<blockquote>
|
| 23 : |
|
|
<em>
|
| 24 : |
|
|
This version is intended for compiler hackers. The
|
| 25 : |
|
|
version ought to be stable, however we have not run
|
| 26 : |
|
|
our full regression testing.
|
| 27 : |
|
|
</em>
|
| 28 : |
|
|
</blockquote>
|
| 29 : |
|
|
</blockquote>
|
| 30 : |
|
|
|
| 31 : |
|
|
|
| 32 : |
|
|
<hr>
|
| 33 : |
|
|
|
| 34 : |
|
|
<center><h2>Summary:</h2></center>
|
| 35 : |
|
|
<UL>
|
| 36 : |
|
|
<LI> Socket related bug fixes.
|
| 37 : |
|
|
<p>
|
| 38 : |
|
|
<LI> Improvements to CM autoloading.
|
| 39 : |
|
|
<p>
|
| 40 : |
|
|
<LI> General cleanup in the use of CM libraries in the compiler.
|
| 41 : |
|
|
<p>
|
| 42 : |
|
|
<LI> A new x86 fp compilation strategy.
|
| 43 : |
|
|
<p>
|
| 44 : |
|
|
<LI> Removal of regmaps from MLRISC.
|
| 45 : |
|
|
<p>
|
| 46 : |
|
|
</UL>
|
| 47 : |
|
|
<hr>
|
| 48 : |
|
|
|
| 49 : |
|
|
<center><h3>Bug fixes </h3></center>
|
| 50 : |
|
|
<ul>
|
| 51 : |
|
|
<li> 1514. sockets c-library broken
|
| 52 : |
|
|
<p>
|
| 53 : |
|
|
<li> 1582. SysErr exception connecting to socket
|
| 54 : |
|
|
<p>
|
| 55 : |
|
|
<li> 1585. getpeername in sockets
|
| 56 : |
|
|
<p>
|
| 57 : |
|
|
</ul>
|
| 58 : |
|
|
|
| 59 : |
|
|
<hr>
|
| 60 : |
|
|
<center><h3>CM </h3></center>
|
| 61 : |
|
|
Drastically improved link traversal code, resulting is faster load
|
| 62 : |
|
|
times for CM and CMB.
|
| 63 : |
|
|
<p>
|
| 64 : |
|
|
Changed CM tool-plugin mechanism. See new manual
|
| 65 : |
|
|
<p>
|
| 66 : |
|
|
Made pickle-lib.cm and eliminated use of comp-lib.cm
|
| 67 : |
|
|
<p>
|
| 68 : |
|
|
|
| 69 : |
|
|
<hr>
|
| 70 : |
|
|
|
| 71 : |
|
|
<center><h3>SML/NJ Library </h3></center>
|
| 72 : |
|
|
Fixed "where" clause to GraphSCCFn.
|
| 73 : |
|
|
<hr>
|
| 74 : |
|
|
<center><h3>MLRISC </h3></center>
|
| 75 : |
|
|
<h4>Intel x86 floating point </h4>
|
| 76 : |
|
|
As of 110.31, there is an alternative floating point code generator
|
| 77 : |
|
|
and register allocator for the x86. Since this is still experimental,
|
| 78 : |
|
|
by default this is turned off. To turn this on, do:
|
| 79 : |
|
|
|
| 80 : |
|
|
<pre>
|
| 81 : |
|
|
CM.autoload "$smlnj/compiler.cm";
|
| 82 : |
|
|
Compiler.Control.MLRISC.getFlag "x86-fast-fp" := true;
|
| 83 : |
|
|
</pre>
|
| 84 : |
|
|
|
| 85 : |
|
|
The new floating point code generator treats the x86 fp stack as
|
| 86 : |
|
|
7 registers, plus one temporary, and directly allocates floating point
|
| 87 : |
|
|
values into these registers.
|
| 88 : |
|
|
Currently, fp parameter passing is still
|
| 89 : |
|
|
done through memory, so the new code generator only benefits floating
|
| 90 : |
|
|
point heavy loops. However, code compiled under the old and new
|
| 91 : |
|
|
code generator can coexist. The algorithm is described in
|
| 92 : |
|
|
<a href=http://cm.bell-labs.com/cm/cs/what/smlnj/compiler-notes/x86-fp.ps>
|
| 93 : |
|
|
this paper</a>.
|
| 94 : |
|
|
<p>
|
| 95 : |
|
|
We compared Version 110.30 compiling the PCLubIN entry in the
|
| 96 : |
|
|
ICFP'00 programming context.
|
| 97 : |
|
|
|
| 98 : |
|
|
<pre>
|
| 99 : |
|
|
110.30 new fp Speedup
|
| 100 : |
|
|
(ICFP00, PCLubIN)
|
| 101 : |
|
|
chess.gml 22.16 20.98 5.63%
|
| 102 : |
|
|
cone-fractal.gml 5.70 5.45 4.51%
|
| 103 : |
|
|
cylinder.gml 1.61 1.58 2.28%
|
| 104 : |
|
|
dice.gml 7.33 6.88 6.57%
|
| 105 : |
|
|
ellipsoid.gml 1.35 1.30 4.16%
|
| 106 : |
|
|
fov.gml 2.63 2.51 4.70%
|
| 107 : |
|
|
Fractal.gml 42.08 41.03 2.56%
|
| 108 : |
|
|
golf.gml 3.09 2.95 4.75%
|
| 109 : |
|
|
holes.gml 3.72 3.50 6.40%
|
| 110 : |
|
|
house.gml 1.41 1.33 5.71%
|
| 111 : |
|
|
intercyl.gml 3.02 2.78 8.41%
|
| 112 : |
|
|
large.gml 8.01 7.81 2.64%
|
| 113 : |
|
|
pipe.gml 6.35 5.78 10.01%
|
| 114 : |
|
|
snowgoon.gml 4.70 4.31 8.95%
|
| 115 : |
|
|
spheres.gml 1.26 1.17 6.98%
|
| 116 : |
|
|
spotlight.gml 0.71 0.68 4.69%
|
| 117 : |
|
|
</pre>
|
| 118 : |
|
|
|
| 119 : |
|
|
By inlining Array2 in the same benchmark we get the following results:
|
| 120 : |
|
|
|
| 121 : |
|
|
<pre>
|
| 122 : |
|
|
110.30 new fp Speedup
|
| 123 : |
|
|
chess.gml 21.85s 21.46s 1.83%
|
| 124 : |
|
|
cone-fractal.gml 5.82s 5.47s 6.28%
|
| 125 : |
|
|
cylinder.gml 1.57s 1.61s -2.85%
|
| 126 : |
|
|
dice.gml 7.57s 6.85s 10.50%
|
| 127 : |
|
|
ellipsoid.gml 1.33s 1.25s 6.74%
|
| 128 : |
|
|
fov.gml 2.75s 2.57s 7.01%
|
| 129 : |
|
|
fractal.gml 22.64s 21.52s 5.20%
|
| 130 : |
|
|
golf.gml 3.04s 2.92s 4.25%
|
| 131 : |
|
|
holes.gml 3.66s 3.48s 5.11%
|
| 132 : |
|
|
house.gml 1.39s 1.29s 7.74%
|
| 133 : |
|
|
intercyl.gml 3.00s 2.78s 7.91%
|
| 134 : |
|
|
large.gml 7.91s 7.82s 1.13%
|
| 135 : |
|
|
pipe.gml 6.44s 5.65s 13.98%
|
| 136 : |
|
|
snowgoon.gml 4.75s 4.29s 10.53%
|
| 137 : |
|
|
spheres.gml 1.22s 1.12s 8.36%
|
| 138 : |
|
|
spotlight.gml 0.71s 0.68s 5.62%
|
| 139 : |
|
|
</pre>
|
| 140 : |
|
|
|
| 141 : |
|
|
Results from other benchmarks:
|
| 142 : |
|
|
<pre>
|
| 143 : |
|
|
barnes-hut 1.714 1.696 1.0%
|
| 144 : |
|
|
fft 0.954 0.906 5.2%
|
| 145 : |
|
|
mandelbrot 19.91 14.99 32.8%
|
| 146 : |
|
|
matrix-multiply(a) 47.77 45.81 4.3%
|
| 147 : |
|
|
matrix-multiply(b) 17.04 15.42 10.5%
|
| 148 : |
|
|
simple 3.02 2.69 12.3%
|
| 149 : |
|
|
tsp 1.75 1.656 5.6%
|
| 150 : |
|
|
</pre>
|
| 151 : |
|
|
NOTE: Matrix multiply(b) has all bounds checking removed.
|
| 152 : |
|
|
|
| 153 : |
|
|
Each test is run 10 times and I take the average.
|
| 154 : |
|
|
|
| 155 : |
|
|
Overall, the numbers do not improve as much as I was hoping, except for
|
| 156 : |
|
|
mandelbrot. The following benchmarks compare smlnj with mlton and C:
|
| 157 : |
|
|
|
| 158 : |
|
|
<pre>
|
| 159 : |
|
|
mandelbrot fft barnes-hut
|
| 160 : |
|
|
sml/nj 110.30 19.91 0.96 1.71
|
| 161 : |
|
|
sml/nj new fp 14.99 0.90 1.71
|
| 162 : |
|
|
gcc -O 14.83
|
| 163 : |
|
|
gcc -O6 14.01 0.68
|
| 164 : |
|
|
mlton -O6 17.46 1.04 1.62
|
| 165 : |
|
|
(version 200000906)
|
| 166 : |
|
|
</pre>
|
| 167 : |
|
|
|
| 168 : |
|
|
<h4>Internal <tt> regmap</tt> Changes </h4>
|
| 169 : |
|
|
<ol>
|
| 170 : |
|
|
<p><li> Changed interface to <tt>CELLS</tt> and the type of <tt>cell</tt>, <tt>cellkind</tt>, <tt>cellset</tt> etc.
|
| 171 : |
|
|
|
| 172 : |
|
|
<p><li> No more regmaps!! The attributes of cells, including its current color,
|
| 173 : |
|
|
are accessible from <tt>CELLS</tt> interface. Cells can now take arbitrary
|
| 174 : |
|
|
annotations. [They will also have a width attribute in the next
|
| 175 : |
|
|
go around.]
|
| 176 : |
|
|
|
| 177 : |
|
|
<p><li> Interface of <tt>STREAM</tt> etc have changed (again, no more regmap).
|
| 178 : |
|
|
|
| 179 : |
|
|
<p><li> Some MLTREE constructors, like <tt>IF</tt>, <tt>BCC</tt>, <tt>JMP</tt>, <tt>CALL</tt>, etc have
|
| 180 : |
|
|
been simplified. <tt>CVTI2I</tt> has been renamed into <tt>SX</tt> and <tt>ZX</tt> respectively,
|
| 181 : |
|
|
following the lambda rtl convention.
|
| 182 : |
|
|
|
| 183 : |
|
|
<p><li> The old RA interface was getting too complicated. There are now
|
| 184 : |
|
|
two functors, <tt>RISC_RA</tt> (in ra/risc_ra.sml) and <tt>X86RA</tt> (in x86/ra/x86RA.sml)
|
| 185 : |
|
|
which abstract out from all the ugly business. The first is for
|
| 186 : |
|
|
RISC machines, and the second is for x86. Please let us know if you
|
| 187 : |
|
|
use these functors.
|
| 188 : |
|
|
|
| 189 : |
|
|
<p><li> The cell change broke the peephole phases, because they used to
|
| 190 : |
|
|
pattern match on specific cell number. I (Allen) hacked up a simple
|
| 191 : |
|
|
tool to translate fake SML with where clauses into real ML. This makes
|
| 192 : |
|
|
it much easier to write the rules. Seems to work. (See Tools/WhereGen)
|
| 193 : |
|
|
</ol>
|
| 194 : |
|
|
|
| 195 : |
|
|
<hr>
|
| 196 : |
|
|
|
| 197 : |
|
|
<font size=-2>
|
| 198 : |
|
|
<address><a href="mailto:george@research.bell-labs.com">
|
| 199 : |
|
|
Lal George</a></address>
|
| 200 : |
|
|
<!-- Created: Thu Aug 6 00:13:09 EDT 1998 -->
|
| 201 : |
|
|
<!-- hhmts start -->
|
| 202 : |
|
|
Last modified: Fri Dec 8 15:15:08 EST 2000
|
| 203 : |
|
|
<!-- hhmts end -->
|
| 204 : |
|
|
</font>
|
| 205 : |
|
|
</blockquote>
|
| 206 : |
|
|
</body>
|
| 207 : |
|
|
</html>
|
Parent Directory
| 
Revision Log
