📄 luajit_performance.html
字号:
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN" "http://www.w3.org/TR/html4/strict.dtd"><html><head><title>LuaJIT Performance</title><meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1"><meta name="Author" content="Mike Pall"><meta name="Copyright" content="Copyright (C) 2005-2008, Mike Pall"><meta name="Language" content="en"><link rel="stylesheet" type="text/css" href="bluequad.css" media="screen"><link rel="stylesheet" type="text/css" href="bluequad-print.css" media="print"><style type="text/css">table.bench { line-height: 1.2;}tr.benchhead td { font-weight: bold;}td img, li img { vertical-align: middle;}td.barhead, td.bar { font-size: 8pt; font-family: Courier New, Courier, monospace; width: 360px; padding: 0;}td.bar { background: url('img/backbar.png');}td.speedup { text-align: center;}</style></head><body><div id="site"><a href="http://luajit.org/"><span>Lua<span id="logo">JIT</span></span></a></div><div id="head"><h1>LuaJIT Performance</h1></div><div id="nav"><ul><li><a href="index.html">Index</a></li><li><a href="luajit.html">LuaJIT</a><ul><li><a href="luajit_features.html">Features</a></li><li><a href="luajit_install.html">Installation</a></li><li><a href="luajit_run.html">Running</a></li><li><a href="luajit_api.html">API Extensions</a></li><li><a href="luajit_intro.html">Introduction</a></li><li><a class="current" href="luajit_performance.html">Performance</a></li><li><a href="luajit_debug.html">Debugging</a></li><li><a href="luajit_changes.html">Changes</a></li></ul></li><li><a href="coco.html">Coco</a><ul><li><a href="coco_portability.html">Portability</a></li><li><a href="coco_api.html">API Extensions</a></li><li><a href="coco_changes.html">Changes</a></li></ul></li><li><a href="dynasm.html">DynASM</a><ul><li><a href="dynasm_features.html">Features</a></li><li><a href="dynasm_examples.html">Examples</a></li></ul></li><li><a href="http://luajit.org/download.html">Download <span class="ext">»</span></a></li></ul></div><div id="main"><p>Here are some performance measurements, based on a few benchmarks.</p><h2 id="interpretation">Interpreting the Results</h2><p>As is always the case with benchmarks, care must be taken tointerpret the results:</p><p>First, the standard Lua interpreter is already <em>very</em> fast.It's commonly the fastest of it's class (interpreters) in the<a href="http://shootout.alioth.debian.org/"><span class="ext">»</span> Great Computer Language Shootout</a>.Only true machine code compilers get a better overall score.</p><p>Any performance improvements due to LuaJIT can only be incremental.You can't expect a speedup of 50x if the fastest compiled languageis only 5x faster than interpreted Lua in a particular benchmark.LuaJIT can't do miracles.</p><p>Also please note that most of the benchmarks below are <em>not</em>trivial micro-benchmarks, which are often cited with marvelous numbers.Micro-benchmarks do not realistically model the performance gains youcan expect in your own programs.</p><p>It's easy to make up a few one-liners like:<br><tt> local function f(...) end; for i=1,1e7 do f() end</tt><br>This is more than 30x faster with LuaJIT. But you won't findthis in a real-world program.</p><h2 id="methods">Measurement Methods</h2><p>All measurements have been taken on a Pentium III 1.139 GHzrunning Linux 2.6. Both Lua and LuaJIT have been compiled withGCC 3.3.6 with <tt>-O3 -fomit-frame-pointer</tt>.You'll definitely get different results on different machines orwith different C compiler options. <sup>*</sup></p><p>The base for the comparison are the user CPU times as reported by<tt>/usr/bin/time</tt>. The runtime of each benchmark is parametrizedand has been adjusted to minimize the variation between several runs.The ratio between the times for LuaJIT and Lua gives the speedup.Only this number is shown because it's less dependent on a specific system.</p><p>E.g. a speedup of 6.74 means the same benchmark runs almost 7 timesfaster with <tt>luajit -O</tt> than with standard Lua (or with<tt>-j off</tt>). Your mileage may vary.</p><p style="font-size: 80%;"><sup>*</sup> Yes, LuaJIT relies on quite a bit of the Lua core infrastructurelike table and string handling. All of this is written in C andshould be compiled with full optimization turned on, or performancewill suffer.</p><h2 id="lua_luajit" class="pagebreak">Comparing Lua to LuaJIT</h2><p>Here is a comparison using the current benchmark collection of the<a href="http://shootout.alioth.debian.org/"><span class="ext">»</span> Great Computer Language Shootout</a> (as of 3/2006):</p><div class="tablewrap"><table class="bench"><tr class="benchhead"><td>Benchmark</td><td class="speedup">Speedup</td><td class="barhead"><img src="img/spacer.png" width="360" height="12" alt="-----1x----2x----3x----4x----5x----6x----7x----8x"></td></tr><tr class="odd"><td>mandelbrot</td><td class="speedup">6.74</td><td class="bar"><img src="img/bluebar.png" width="303" height="12" alt="========================================"></td></tr><tr class="even"><td>recursive</td><td class="speedup">6.64</td><td class="bar"><img src="img/bluebar.png" width="299" height="12" alt="========================================"></td></tr><tr class="odd"><td>fannkuch</td><td class="speedup">5.37</td><td class="bar"><img src="img/bluebar.png" width="242" height="12" alt="================================"></td></tr><tr class="even"><td>chameneos</td><td class="speedup">5.08</td><td class="bar"><img src="img/bluebar.png" width="229" height="12" alt="=============================="></td></tr><tr class="odd"><td>nsievebits</td><td class="speedup">5.05</td><td class="bar"><img src="img/bluebar.png" width="227" height="12" alt="=============================="></td></tr><tr class="even"><td>pidigits</td><td class="speedup">4.94</td><td class="bar"><img src="img/bluebar.png" width="222" height="12" alt="=============================="></td></tr><tr class="odd"><td>nbody</td><td class="speedup">4.63</td><td class="bar"><img src="img/bluebar.png" width="208" height="12" alt="============================"></td></tr><tr class="even"><td>spectralnorm</td><td class="speedup">4.59</td><td class="bar"><img src="img/bluebar.png" width="207" height="12" alt="============================"></td></tr><tr class="odd"><td>cheapconcr</td><td class="speedup">4.46</td><td class="bar"><img src="img/bluebar.png" width="201" height="12" alt="==========================="></td></tr><tr class="even"><td>partialsums</td><td class="speedup">3.73</td><td class="bar"><img src="img/bluebar.png" width="168" height="12" alt="======================"></td></tr><tr class="odd"><td>fasta</td><td class="speedup">2.68</td><td class="bar"><img src="img/bluebar.png" width="121" height="12" alt="================"></td></tr><tr class="even"><td>cheapconcw</td><td class="speedup">2.52</td><td class="bar"><img src="img/bluebar.png" width="113" height="12" alt="==============="></td></tr><tr class="odd"><td>nsieve</td><td class="speedup">1.95</td><td class="bar"><img src="img/bluebar.png" width="88" height="12" alt="============"></td></tr><tr class="even"><td>revcomp</td><td class="speedup">1.92</td><td class="bar"><img src="img/bluebar.png" width="86" height="12" alt="============"></td></tr><tr class="odd"><td>knucleotide</td><td class="speedup">1.59</td><td class="bar"><img src="img/bluebar.png" width="72" height="12" alt="=========="></td></tr><tr class="even"><td>binarytrees</td><td class="speedup">1.52</td><td class="bar"><img src="img/bluebar.png" width="68" height="12" alt="========="></td></tr><tr class="odd"><td>sumfile</td><td class="speedup">1.27</td><td class="bar"><img src="img/bluebar.png" width="57" height="12" alt="========"></td></tr><tr class="even"><td>regexdna</td><td class="speedup">1.01</td><td class="bar"><img src="img/bluebar.png" width="45" height="12" alt="======"></td></tr></table></div><p>Note that many of these benchmarks have changed over time (both specand code). Benchmark results shown in previous versions of LuaJITare not directly comparable. The next section compares differentversions with the current set of benchmarks.</p><h2 id="luajit_versions" class="pagebreak">Comparing LuaJIT Versions</h2><p>This shows the improvements between the following versions:</p><ul><li>LuaJIT 1.0.x <img src="img/bluebar.png" width="30" height="12" alt="(===)"></li><li>LuaJIT 1.1.x <img src="img/bluebar.png" width="30" height="12" alt="(===##)"><img src="img/magentabar.png" width="20" height="12" alt=""></li></ul><div class="tablewrap"><table class="bench"><tr class="benchhead"><td>Benchmark</td><td class="speedup">Speedup</td><td class="barhead"><img src="img/spacer.png" width="360" height="12" alt="-----1x----2x----3x----4x----5x----6x----7x----8x"></td></tr><tr class="odd"><td>fannkuch</td><td class="speedup">3.96 → 5.37</td><td class="bar"><img src="img/bluebar.png" width="178" height="12" alt="========================"><img src="img/magentabar.png" width="64" height="12" alt="########"></td></tr><tr class="even"><td>chameneos</td><td class="speedup">2.25 → 5.08</td><td class="bar"><img src="img/bluebar.png" width="101" height="12" alt="=============="><img src="img/magentabar.png" width="128" height="12" alt="################"></td></tr><tr class="odd"><td>nsievebits</td><td class="speedup">2.90 → 5.05</td><td class="bar"><img src="img/bluebar.png" width="131" height="12" alt="================="><img src="img/magentabar.png" width="96" height="12" alt="#############"></td></tr><tr class="even"><td>pidigits</td><td class="speedup">3.58 → 4.94</td><td class="bar"><img src="img/bluebar.png" width="161" height="12" alt="====================="><img src="img/magentabar.png" width="61" height="12" alt="#########"></td></tr><tr class="odd"><td>nbody</td><td class="speedup">4.16 → 4.63</td><td class="bar"><img src="img/bluebar.png" width="187" height="12" alt="========================="><img src="img/magentabar.png" width="21" height="12" alt="###"></td></tr><tr class="even"><td>cheapconcr</td><td class="speedup">1.46 → 4.46</td><td class="bar"><img src="img/bluebar.png" width="66" height="12" alt="========="><img src="img/magentabar.png" width="135" height="12" alt="##################"></td></tr><tr class="odd"><td>partialsums</td><td class="speedup">1.71 → 3.73</td><td class="bar"><img src="img/bluebar.png" width="77" height="12" alt="=========="><img src="img/magentabar.png" width="91" height="12" alt="############"></td></tr><tr class="even"><td>fasta</td><td class="speedup">2.37 → 2.68</td><td class="bar"><img src="img/bluebar.png" width="107" height="12" alt="=============="><img src="img/magentabar.png" width="14" height="12" alt="##"></td></tr><tr class="odd"><td>cheapconcw</td><td class="speedup">1.27 → 2.52</td><td class="bar"><img src="img/bluebar.png" width="57" height="12" alt="========"><img src="img/magentabar.png" width="56" height="12" alt="#######"></td></tr><tr class="even"><td>revcomp</td><td class="speedup">1.45 → 1.92</td><td class="bar"><img src="img/bluebar.png" width="65" height="12" alt="========="><img src="img/magentabar.png" width="21" height="12" alt="###"></td></tr><tr class="odd"><td>knucleotide</td><td class="speedup">1.32 → 1.59</td><td class="bar"><img src="img/bluebar.png" width="59" height="12" alt="========"><img src="img/magentabar.png" width="13" height="12" alt="##"></td></tr></table></div><p>All other benchmarks show only minor performance differences.</p><h2 id="summary">Summary</h2><p>These results should give you an idea about what speedupyou can expect depending on the nature of your Lua code:</p><ul><li>LuaJIT is really good at (floating-point) math and loops(mandelbrot, pidigits, spectralnorm, partialsums).</li><li>Function calls (recursive), vararg calls, table lookups (nbody),table iteration and coroutine switching (chameneos, cheapconc)are a lot faster than with plain Lua.</li><li>It's still pretty good for indexed table access (fannkuch, nsieve)and string processing (fasta, revcomp, knucleotide).But there is room for improvement in a future version.</li><li>If your application spends most of the time in C codeyou won't see much of a difference (regexdna, sumfile).Ok, so write more code in pure Lua. :-)</li><li>The real speedup may be shadowed by other dominant factors in a benchmark:<ul><li>Common parts of the Lua core: e.g. memory allocationand GC (binarytrees).</li><li>Language characteristics: e.g. lack of bit operations (nsievebits).</li><li>System characteristics: e.g. CPU cache size and memory speed (nsieve).</li></ul></li></ul><p>The best idea is of course to benchmark your <em>own</em> applications.Please report any interesting results you may find. Thank you!</p><br class="flush"></div><div id="foot"><hr class="hide">Copyright © 2005-2008 Mike Pall<span class="noprint">·<a href="contact.html">Contact</a></span></div></body></html>⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -