You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
ljx/doc/ext_profiler.html

366 lines
13 KiB
HTML

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN" "http://www.w3.org/TR/html4/strict.dtd">
<html>
<head>
<title>Profiler</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) 2005-2016, 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">
</head>
<body>
<div id="site">
<a href="http://luajit.org"><span>Lua<span id="logo">JIT</span></span></a>
</div>
<div id="head">
<h1>Profiler</h1>
</div>
<div id="nav">
<ul><li>
<a href="luajit.html">LuaJIT</a>
<ul><li>
<a href="http://luajit.org/download.html">Download <span class="ext">&raquo;</span></a>
</li><li>
<a href="install.html">Installation</a>
</li><li>
<a href="running.html">Running</a>
</li></ul>
</li><li>
<a href="extensions.html">Extensions</a>
<ul><li>
<a href="ext_ffi.html">FFI Library</a>
<ul><li>
<a href="ext_ffi_tutorial.html">FFI Tutorial</a>
</li><li>
<a href="ext_ffi_api.html">ffi.* API</a>
</li><li>
<a href="ext_ffi_semantics.html">FFI Semantics</a>
</li></ul>
</li><li>
<a href="ext_jit.html">jit.* Library</a>
</li><li>
<a href="ext_c_api.html">Lua/C API</a>
</li><li>
<a class="current" href="ext_profiler.html">Profiler</a>
</li></ul>
</li><li>
<a href="status.html">Status</a>
<ul><li>
<a href="changes.html">Changes</a>
</li></ul>
</li><li>
<a href="faq.html">FAQ</a>
</li><li>
<a href="http://luajit.org/performance.html">Performance <span class="ext">&raquo;</span></a>
</li><li>
<a href="http://wiki.luajit.org/">Wiki <span class="ext">&raquo;</span></a>
</li><li>
<a href="http://luajit.org/list.html">Mailing List <span class="ext">&raquo;</span></a>
</li></ul>
</div>
<div id="main">
<p>
LuaJIT has an integrated statistical profiler with very low overhead. It
allows sampling the currently executing stack and other parameters in
regular intervals.
</p>
<p>
The integrated profiler can be accessed from three levels:
</p>
<ul>
<li>The <a href="#hl_profiler">bundled high-level profiler</a>, invoked by the
<a href="#j_p"><tt>-jp</tt></a> command line option.</li>
<li>A <a href="#ll_lua_api">low-level Lua API</a> to control the profiler.</li>
<li>A <a href="#ll_c_api">low-level C API</a> to control the profiler.</li>
</ul>
<h2 id="hl_profiler">High-Level Profiler</h2>
<p>
The bundled high-level profiler offers basic profiling functionality. It
generates simple textual summaries or source code annotations. It can be
accessed with the <a href="#j_p"><tt>-jp</tt></a> command line option
or from Lua code by loading the underlying <tt>jit.p</tt> module.
</p>
<p>
To cut to the chase &mdash; run this to get a CPU usage profile by
function name:
</p>
<pre class="code">
luajit -jp myapp.lua
</pre>
<p>
It's <em>not</em> a stated goal of the bundled profiler to add every
possible option or to cater for special profiling needs. The low-level
profiler APIs are documented below. They may be used by third-party
authors to implement advanced functionality, e.g. IDE integration or
graphical profilers.
</p>
<p>
Note: Sampling works for both interpreted and JIT-compiled code. The
results for JIT-compiled code may sometimes be surprising. LuaJIT
heavily optimizes and inlines Lua code &mdash; there's no simple
one-to-one correspondence between source code lines and the sampled
machine code.
</p>
<h3 id="j_p"><tt>-jp=[options[,output]]</tt></h3>
<p>
The <tt>-jp</tt> command line option starts the high-level profiler.
When the application run by the command line terminates, the profiler
stops and writes the results to <tt>stdout</tt> or to the specified
<tt>output</tt> file.
</p>
<p>
The <tt>options</tt> argument specifies how the profiling is to be
performed:
</p>
<ul>
<li><tt>f</tt> &mdash; Stack dump: function name, otherwise module:line.
This is the default mode.</li>
<li><tt>F</tt> &mdash; Stack dump: ditto, but dump module:name.</li>
<li><tt>l</tt> &mdash; Stack dump: module:line.</li>
<li><tt>&lt;number&gt;</tt> &mdash; stack dump depth (callee &larr;
caller). Default: 1.</li>
<li><tt>-&lt;number&gt;</tt> &mdash; Inverse stack dump depth (caller
&rarr; callee).</li>
<li><tt>s</tt> &mdash; Split stack dump after first stack level. Implies
depth&nbsp;&ge;&nbsp;2 or depth&nbsp;&le;&nbsp;-2.</li>
<li><tt>p</tt> &mdash; Show full path for module names.</li>
<li><tt>v</tt> &mdash; Show VM states.</li>
<li><tt>z</tt> &mdash; Show <a href="#jit_zone">zones</a>.</li>
<li><tt>r</tt> &mdash; Show raw sample counts. Default: show percentages.</li>
<li><tt>a</tt> &mdash; Annotate excerpts from source code files.</li>
<li><tt>A</tt> &mdash; Annotate complete source code files.</li>
<li><tt>G</tt> &mdash; Produce raw output suitable for graphical tools.</li>
<li><tt>m&lt;number&gt;</tt> &mdash; Minimum sample percentage to be shown.
Default: 3%.</li>
<li><tt>i&lt;number&gt;</tt> &mdash; Sampling interval in milliseconds.
Default: 10ms.<br>
Note: The actual sampling precision is OS-dependent.</li>
</ul>
<p>
The default output for <tt>-jp</tt> is a list of the most CPU consuming
spots in the application. Increasing the stack dump depth with (say)
<tt>-jp=2</tt> may help to point out the main callers or callees of
hotspots. But sample aggregation is still flat per unique stack dump.
</p>
<p>
To get a two-level view (split view) of callers/callees, use
<tt>-jp=s</tt> or <tt>-jp=-s</tt>. The percentages shown for the second
level are relative to the first level.
</p>
<p>
To see how much time is spent in each line relative to a function, use
<tt>-jp=fl</tt>.
</p>
<p>
To see how much time is spent in different VM states or
<a href="#jit_zone">zones</a>, use <tt>-jp=v</tt> or <tt>-jp=z</tt>.
</p>
<p>
Combinations of <tt>v/z</tt> with <tt>f/F/l</tt> produce two-level
views, e.g. <tt>-jp=vf</tt> or <tt>-jp=fv</tt>. This shows the time
spent in a VM state or zone vs. hotspots. This can be used to answer
questions like "Which time consuming functions are only interpreted?" or
"What's the garbage collector overhead for a specific function?".
</p>
<p>
Multiple options can be combined &mdash; but not all combinations make
sense, see above. E.g. <tt>-jp=3si4m1</tt> samples three stack levels
deep in 4ms intervals and shows a split view of the CPU consuming
functions and their callers with a 1% threshold.
</p>
<p>
Source code annotations produced by <tt>-jp=a</tt> or <tt>-jp=A</tt> are
always flat and at the line level. Obviously, the source code files need
to be readable by the profiler script.
</p>
<p>
The high-level profiler can also be started and stopped from Lua code with:
</p>
<pre class="code">
require("jit.p").start(options, output)
...
require("jit.p").stop()
</pre>
<h3 id="jit_zone"><tt>jit.zone</tt> &mdash; Zones</h3>
<p>
Zones can be used to provide information about different parts of an
application to the high-level profiler. E.g. a game could make use of an
<tt>"AI"</tt> zone, a <tt>"PHYS"</tt> zone, etc. Zones are hierarchical,
organized as a stack.
</p>
<p>
The <tt>jit.zone</tt> module needs to be loaded explicitly:
</p>
<pre class="code">
local zone = require("jit.zone")
</pre>
<ul>
<li><tt>zone("name")</tt> pushes a named zone to the zone stack.</li>
<li><tt>zone()</tt> pops the current zone from the zone stack and
returns its name.</li>
<li><tt>zone:get()</tt> returns the current zone name or <tt>nil</tt>.</li>
<li><tt>zone:flush()</tt> flushes the zone stack.</li>
</ul>
<p>
To show the time spent in each zone use <tt>-jp=z</tt>. To show the time
spent relative to hotspots use e.g. <tt>-jp=zf</tt> or <tt>-jp=fz</tt>.
</p>
<h2 id="ll_lua_api">Low-level Lua API</h2>
<p>
The <tt>jit.profile</tt> module gives access to the low-level API of the
profiler from Lua code. This module needs to be loaded explicitly:
<pre class="code">
local profile = require("jit.profile")
</pre>
<p>
This module can be used to implement your own higher-level profiler.
A typical profiling run starts the profiler, captures stack dumps in
the profiler callback, adds them to a hash table to aggregate the number
of samples, stops the profiler and then analyzes all of the captured
stack dumps. Other parameters can be sampled in the profiler callback,
too. But it's important not to spend too much time in the callback,
since this may skew the statistics.
</p>
<h3 id="profile_start"><tt>profile.start(mode, cb)</tt>
&mdash; Start profiler</h3>
<p>
This function starts the profiler. The <tt>mode</tt> argument is a
string holding options:
</p>
<ul>
<li><tt>f</tt> &mdash; Profile with precision down to the function level.</li>
<li><tt>l</tt> &mdash; Profile with precision down to the line level.</li>
<li><tt>i&lt;number&gt;</tt> &mdash; Sampling interval in milliseconds (default
10ms).</br>
Note: The actual sampling precision is OS-dependent.
</li>
</ul>
<p>
The <tt>cb</tt> argument is a callback function which is called with
three arguments: <tt>(thread, samples, vmstate)</tt>. The callback is
called on a separate coroutine, the <tt>thread</tt> argument is the
state that holds the stack to sample for profiling. Note: do
<em>not</em> modify the stack of that state or call functions on it.
</p>
<p>
<tt>samples</tt> gives the number of accumulated samples since the last
callback (usually 1).
</p>
<p>
<tt>vmstate</tt> holds the VM state at the time the profiling timer
triggered. This may or may not correspond to the state of the VM when
the profiling callback is called. The state is either <tt>'N'</tt>
native (compiled) code, <tt>'I'</tt> interpreted code, <tt>'C'</tt>
C&nbsp;code, <tt>'G'</tt> the garbage collector, or <tt>'J'</tt> the JIT
compiler.
</p>
<h3 id="profile_stop"><tt>profile.stop()</tt>
&mdash; Stop profiler</h3>
<p>
This function stops the profiler.
</p>
<h3 id="profile_dump"><tt>dump = profile.dumpstack([thread,] fmt, depth)</tt>
&mdash; Dump stack </h3>
<p>
This function allows taking stack dumps in an efficient manner. It
returns a string with a stack dump for the <tt>thread</tt> (coroutine),
formatted according to the <tt>fmt</tt> argument:
</p>
<ul>
<li><tt>p</tt> &mdash; Preserve the full path for module names. Otherwise
only the file name is used.</li>
<li><tt>f</tt> &mdash; Dump the function name if it can be derived. Otherwise
use module:line.</li>
<li><tt>F</tt> &mdash; Ditto, but dump module:name.</li>
<li><tt>l</tt> &mdash; Dump module:line.</li>
<li><tt>Z</tt> &mdash; Zap the following characters for the last dumped
frame.</li>
<li>All other characters are added verbatim to the output string.</li>
</ul>
<p>
The <tt>depth</tt> argument gives the number of frames to dump, starting
at the topmost frame of the thread. A negative number dumps the frames in
inverse order.
</p>
<p>
The first example prints a list of the current module names and line
numbers of up to 10 frames in separate lines. The second example prints
semicolon-separated function names for all frames (up to 100) in inverse
order:
</p>
<pre class="code">
print(profile.dumpstack(thread, "l\n", 10))
print(profile.dumpstack(thread, "lZ;", -100))
</pre>
<h2 id="ll_c_api">Low-level C API</h2>
<p>
The profiler can be controlled directly from C&nbsp;code, e.g. for
use by IDEs. The declarations are in <tt>"luajit.h"</tt> (see
<a href="ext_c_api.html">Lua/C API</a> extensions).
</p>
<h3 id="luaJIT_profile_start"><tt>luaJIT_profile_start(L, mode, cb, data)</tt>
&mdash; Start profiler</h3>
<p>
This function starts the profiler. <a href="#profile_start">See
above</a> for a description of the <tt>mode</tt> argument.
</p>
<p>
The <tt>cb</tt> argument is a callback function with the following
declaration:
</p>
<pre class="code">
typedef void (*luaJIT_profile_callback)(void *data, lua_State *L,
int samples, int vmstate);
</pre>
<p>
<tt>data</tt> is available for use by the callback. <tt>L</tt> is the
state that holds the stack to sample for profiling. Note: do
<em>not</em> modify this stack or call functions on this stack &mdash;
use a separate coroutine for this purpose. <a href="#profile_start">See
above</a> for a description of <tt>samples</tt> and <tt>vmstate</tt>.
</p>
<h3 id="luaJIT_profile_stop"><tt>luaJIT_profile_stop(L)</tt>
&mdash; Stop profiler</h3>
<p>
This function stops the profiler.
</p>
<h3 id="luaJIT_profile_dumpstack"><tt>p = luaJIT_profile_dumpstack(L, fmt, depth, len)</tt>
&mdash; Dump stack </h3>
<p>
This function allows taking stack dumps in an efficient manner.
<a href="#profile_dump">See above</a> for a description of <tt>fmt</tt>
and <tt>depth</tt>.
</p>
<p>
This function returns a <tt>const&nbsp;char&nbsp;*</tt> pointing to a
private string buffer of the profiler. The <tt>int&nbsp;*len</tt>
argument returns the length of the output string. The buffer is
overwritten on the next call and deallocated when the profiler stops.
You either need to consume the content immediately or copy it for later
use.
</p>
<br class="flush">
</div>
<div id="foot">
<hr class="hide">
Copyright &copy; 2005-2016 Mike Pall
<span class="noprint">
&middot;
<a href="contact.html">Contact</a>
</span>
</div>
</body>
</html>