187 lines
7.9 KiB
HTML
Executable File

<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN"
"http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">
<html xmlns="http://www.w3.org/1999/xhtml" xml:lang="en" lang="en">
<head>
<title>LuaProfiler - Time profiler for the Lua programming language</title>
<link rel="stylesheet" href="http://www.keplerproject.org/doc.css" type="text/css"/>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8"/>
</head>
<body>
<div id="container">
<div id="product">
<div id="product_logo"><a href="http://www.keplerproject.org">
<img alt="LuaProfiler logo" src="luaprofiler.png"/>
</a></div>
<div id="product_name"><big><b>LuaProfiler</b></big></div>
<div id="product_description">Time profiler for the Lua programming language</div>
</div> <!-- id="product" -->
<div id="main">
<div id="navigation">
<h1>LuaProfiler</h1>
<ul>
<li><a href="index.html">Home</a>
<ul>
<li> <a href="index.html#over">Overview</a></li>
<li> <a href="index.html#status">Status</a></li>
<li> <a href="index.html#download">Download</a></li>
<li> <a href="index.html#history">History</a></li>
<li> <a href="index.html#credits">Credits</a></li>
<li> <a href="index.html#contact">Contact</a></li>
</ul>
</li>
<li><strong>Manual</strong>
<ul>
<li> <a href="manual.html#introduction">Introduction</a></li>
<li> <a href="manual.html#install">Installation</a></li>
<li> <a href="manual.html#analyzer">Using LuaProfiler</a></li>
</ul>
</li>
<li><a href="http://luaforge.net/projects/luaprofiler/">Project</a>
<ul>
<li><a href="http://luaforge.net/tracker/?group_id=25">Bug Tracker</a></li>
<li><a href="http://luaforge.net/scm/?group_id=25">CVS</a></li>
</ul>
</li>
<li><a href="license.html">License</a></li>
</ul>
</div>
<div id="content">
<h2><a name="introduction"></a>Introduction</h2>
<cite>More computing sins are committed in the name of efficiency (without necessarily
achieving it) than for any other single reason - including blind stupidity. - W.A. Wulf</cite>
<p> LuaProfiler can profile programs written in Lua 5.1, Lua 5.0, Lua 4.0, Lua 3.2 and LuaNG
(used by older versions of CGILua).</p>
<p>There are several reasons to not optimize your program prematurely. Some
of them include unnecessary loss of time, maintainability and robustness.</p>
<p>A profiler tells you what parts of your program are slow, so you can
focus on them and forget about the rest. The slow parts are called bottlenecks.
A bottleneck is a part of your program that is much slower than
the rest of the program.</p>
<p>If you are programming in Lua, as a rule of thumb once you found a bottleneck
you have two options: fine tune your code or, arguably the better option,
redesigning your code.</p>
<p>To fine tune your Lua code, all you need to know is how slow the functions
are. Then you can rewrite some of them, possibly in C. This option not necessarily
improves significantly your performance.</p>
<p>To redesign your Lua code, you need more than that: you'll need to know how slow
the functions are and which context makes them slow. A general solution aims to avoid
unnecessary calls and loops.</p>
<p>When you do that, you reduce the complexity of your code, which can give you much
more significant speed improvements in some situations. If you still need to speed
things up and redesigning is not a viable option anymore, go ahead and fine tune the
slower functions.</p>
<p>To fine tune, you need only timing numbers, but to redesign you need a better view
of your program. This profiler helps you in these two processes.</p>
<p>For flexibility reasons, LuaProfiler is divided in 2 parts: the Lua code profiler,
written in C to maximize the precision of timing, and the analyzer that generates
information about the program execution, written in Lua to allow greater variance
in the analysis process.</p>
<p>The code profiler produces as result a table of function calls and times
that is usable by the analyzer itself or by a spreadsheet program, so you can
analyze the results numerically to help you to fine tune your program.</p>
<h2><a name="install"></a>Installation</h2>
<p>LuaProfiler source is distributed as a group of C files and some makefile templates.
For older versions of Lua (3.2 and 4.0), LuaProfiler must be embedded in your application
(linked with it). You must call the function <code>init_profiler()</code> to start profiling from
C code.</p>
<p>LuaProfiler follows the
<a href="http://www.inf.puc-rio.br/~roberto/pil2/chapter15.pdf">package model</a>
for Lua 5.1, therefore it should be "installed" in your <code>package.path</code>.</p>
<p>Windows users can use the pre-compiled version of LuaProfiler
(<code>profiler.dll</code>) available at
<a href="http://luaforge.net/projects/luaprofiler/files">LuaForge</a>.</p>
<h2><a name="analyzer"></a>Using LuaProfiler</h2>
<p>To profile your program for versions of Lua prior to 5.0, include <code>luaProfiler.h</code>
into your main module, include a call to <code>init_profiler(L)</code> after opening the Lua library
and recompile it using the appropriate LuaProfiler lib. If you are profiling Lua 3.2 programs,
consider <em>L</em> to be <code>NULL</code>.</p>
<p>For Lua 5.1 and 5.0, just <code>require"profiler"</code> in your script and this will define a
<em>profiler</em> module with two functions:</p>
<dl>
<dt><strong>start([filename])</strong></dt>
<dd>Starts the profiler using the optional <em>filename</em> as the log file.</dd>
<dt><strong>stop()</strong></dt>
<dd>Stops the profiler.</dd>
</dl>
<p>You can restart the profiler after a <code>stop</code> with another call to <code>start</code>.</p>
<p>All you need now is to run your program and get the output generated by the
profiler. The default log file is written to the working directory of the program,
in a file like <code>lprof_<em>randomid</em>.out</code> where <em>randomid</em> is
a random number. The log format uses a line for every function call in your program which may
result in huge files, so be careful with disk space if your program runs for a long time, or use
localized profiling wrapping the target section with a <code>start()</code>/<code>stop()</code>
call sequence.</p>
<p>LuaProfiler includes a script, <code>summary.lua</code>, that analyzes the output of the profiler
and generates a table with the functions used by your program and their running
times, ordered according to how much of the total execution time of your program the function
used. This script needs to be interpreted by Lua.</p>
<p>You can also pass the <code>-v</code> option to <code>summary.lua</code>, just before the name of the
log file, to generate a more comprehensive summary, containing the number of times each function
was called, their average running times, and the percentage of the total running time of your
script each function took.</p>
<p>Assuming the Lua interpreter is in your path, and both summary.lua and your output
file are in the same directory, switch to the directory where the output file is and type:</p>
<pre class="example">
lua summary.lua lprof_123456.out
</pre>
<p>For the verbose output, type:</p>
<pre class="example">
lua summary.lua -v lprof_123456.out
</pre>
<p>You can also import the output file into a spreadsheet and analyze it there. Each line
of the output file is a function call. The first line is a header that explains what each
column contains. If you are using Excel for example, you can easily create a pivot table that
shows the sum of times for the functions and then sort the cells so the bottlenecks are at the top.</p>
</div> <!-- id="content" -->
</div> <!-- id="main" -->
<div id="about">
<p><a href="http://validator.w3.org/check?uri=referer"><img src="http://www.w3.org/Icons/valid-xhtml10" alt="Valid XHTML 1.0!" height="31" width="88" /></a></p>
<p><small>
$Id: manual.html,v 1.13 2007/08/22 18:20:40 carregal Exp $
</small></p>
</div> <!-- id="about" -->
</div> <!-- id="container" -->
</body>
</html>