aboutsummaryrefslogtreecommitdiff
path: root/doc/ext_profiler.html
diff options
context:
space:
mode:
Diffstat (limited to 'doc/ext_profiler.html')
-rw-r--r--doc/ext_profiler.html359
1 files changed, 359 insertions, 0 deletions
diff --git a/doc/ext_profiler.html b/doc/ext_profiler.html
new file mode 100644
index 00000000..300a9482
--- /dev/null
+++ b/doc/ext_profiler.html
@@ -0,0 +1,359 @@
1<!DOCTYPE html>
2<html>
3<head>
4<title>Profiler</title>
5<meta charset="utf-8">
6<meta name="Copyright" content="Copyright (C) 2005-2026">
7<meta name="Language" content="en">
8<link rel="stylesheet" type="text/css" href="bluequad.css" media="screen">
9<link rel="stylesheet" type="text/css" href="bluequad-print.css" media="print">
10</head>
11<body>
12<div id="site">
13<a href="https://luajit.org"><span>Lua<span id="logo">JIT</span></span></a>
14</div>
15<div id="head">
16<h1>Profiler</h1>
17</div>
18<div id="nav">
19<ul><li>
20<a href="luajit.html">LuaJIT</a>
21<ul><li>
22<a href="https://luajit.org/download.html">Download <span class="ext">&raquo;</span></a>
23</li><li>
24<a href="install.html">Installation</a>
25</li><li>
26<a href="running.html">Running</a>
27</li></ul>
28</li><li>
29<a href="extensions.html">Extensions</a>
30<ul><li>
31<a href="ext_ffi.html">FFI Library</a>
32<ul><li>
33<a href="ext_ffi_tutorial.html">FFI Tutorial</a>
34</li><li>
35<a href="ext_ffi_api.html">ffi.* API</a>
36</li><li>
37<a href="ext_ffi_semantics.html">FFI Semantics</a>
38</li></ul>
39</li><li>
40<a href="ext_buffer.html">String Buffers</a>
41</li><li>
42<a href="ext_jit.html">jit.* Library</a>
43</li><li>
44<a href="ext_c_api.html">Lua/C API</a>
45</li><li>
46<a class="current" href="ext_profiler.html">Profiler</a>
47</li></ul>
48</li><li>
49<a href="https://luajit.org/status.html">Status <span class="ext">&raquo;</span></a>
50</li><li>
51<a href="https://luajit.org/faq.html">FAQ <span class="ext">&raquo;</span></a>
52</li><li>
53<a href="https://luajit.org/list.html">Mailing List <span class="ext">&raquo;</span></a>
54</li></ul>
55</div>
56<div id="main">
57<p>
58LuaJIT has an integrated statistical profiler with very low overhead. It
59allows sampling the currently executing stack and other parameters in
60regular intervals.
61</p>
62<p>
63The integrated profiler can be accessed from three levels:
64</p>
65<ul>
66<li>The <a href="#hl_profiler">bundled high-level profiler</a>, invoked by the
67<a href="#j_p"><tt>-jp</tt></a> command line option.</li>
68<li>A <a href="#ll_lua_api">low-level Lua API</a> to control the profiler.</li>
69<li>A <a href="#ll_c_api">low-level C API</a> to control the profiler.</li>
70</ul>
71
72<h2 id="hl_profiler">High-Level Profiler</h2>
73<p>
74The bundled high-level profiler offers basic profiling functionality. It
75generates simple textual summaries or source code annotations. It can be
76accessed with the <a href="#j_p"><tt>-jp</tt></a> command line option
77or from Lua code by loading the underlying <tt>jit.p</tt> module.
78</p>
79<p>
80To cut to the chase &mdash; run this to get a CPU usage profile by
81function name:
82</p>
83<pre class="code">
84luajit -jp myapp.lua
85</pre>
86<p>
87It's <em>not</em> a stated goal of the bundled profiler to add every
88possible option or to cater for special profiling needs. The low-level
89profiler APIs are documented below. They may be used by third-party
90authors to implement advanced functionality, e.g. IDE integration or
91graphical profilers.
92</p>
93<p>
94Note: Sampling works for both interpreted and JIT-compiled code. The
95results for JIT-compiled code may sometimes be surprising. LuaJIT
96heavily optimizes and inlines Lua code &mdash; there's no simple
97one-to-one correspondence between source code lines and the sampled
98machine code.
99</p>
100
101<h3 id="j_p"><tt>-jp=[options[,output]]</tt></h3>
102<p>
103The <tt>-jp</tt> command line option starts the high-level profiler.
104When the application run by the command line terminates, the profiler
105stops and writes the results to <tt>stdout</tt> or to the specified
106<tt>output</tt> file.
107</p>
108<p>
109The <tt>options</tt> argument specifies how the profiling is to be
110performed:
111</p>
112<ul>
113<li><tt>f</tt> &mdash; Stack dump: function name, otherwise module:line.
114This is the default mode.</li>
115<li><tt>F</tt> &mdash; Stack dump: ditto, but dump module:name.</li>
116<li><tt>l</tt> &mdash; Stack dump: module:line.</li>
117<li><tt>&lt;number&gt;</tt> &mdash; stack dump depth (callee &larr;
118caller). Default: 1.</li>
119<li><tt>-&lt;number&gt;</tt> &mdash; Inverse stack dump depth (caller
120&rarr; callee).</li>
121<li><tt>s</tt> &mdash; Split stack dump after first stack level. Implies
122depth&nbsp;&ge;&nbsp;2 or depth&nbsp;&le;&nbsp;-2.</li>
123<li><tt>p</tt> &mdash; Show full path for module names.</li>
124<li><tt>v</tt> &mdash; Show VM states.</li>
125<li><tt>z</tt> &mdash; Show <a href="#jit_zone">zones</a>.</li>
126<li><tt>r</tt> &mdash; Show raw sample counts. Default: show percentages.</li>
127<li><tt>a</tt> &mdash; Annotate excerpts from source code files.</li>
128<li><tt>A</tt> &mdash; Annotate complete source code files.</li>
129<li><tt>G</tt> &mdash; Produce raw output suitable for graphical tools.</li>
130<li><tt>m&lt;number&gt;</tt> &mdash; Minimum sample percentage to be shown.
131Default: 3%.</li>
132<li><tt>i&lt;number&gt;</tt> &mdash; Sampling interval in milliseconds.
133Default: 10ms.<br>
134Note: The actual sampling precision is OS-dependent.</li>
135</ul>
136<p>
137The default output for <tt>-jp</tt> is a list of the most CPU consuming
138spots in the application. Increasing the stack dump depth with (say)
139<tt>-jp=2</tt> may help to point out the main callers or callees of
140hotspots. But sample aggregation is still flat per unique stack dump.
141</p>
142<p>
143To get a two-level view (split view) of callers/callees, use
144<tt>-jp=s</tt> or <tt>-jp=-s</tt>. The percentages shown for the second
145level are relative to the first level.
146</p>
147<p>
148To see how much time is spent in each line relative to a function, use
149<tt>-jp=fl</tt>.
150</p>
151<p>
152To see how much time is spent in different VM states or
153<a href="#jit_zone">zones</a>, use <tt>-jp=v</tt> or <tt>-jp=z</tt>.
154</p>
155<p>
156Combinations of <tt>v/z</tt> with <tt>f/F/l</tt> produce two-level
157views, e.g. <tt>-jp=vf</tt> or <tt>-jp=fv</tt>. This shows the time
158spent in a VM state or zone vs. hotspots. This can be used to answer
159questions like "Which time-consuming functions are only interpreted?" or
160"What's the garbage collector overhead for a specific function?".
161</p>
162<p>
163Multiple options can be combined &mdash; but not all combinations make
164sense, see above. E.g. <tt>-jp=3si4m1</tt> samples three stack levels
165deep in 4ms intervals and shows a split view of the CPU consuming
166functions and their callers with a 1% threshold.
167</p>
168<p>
169Source code annotations produced by <tt>-jp=a</tt> or <tt>-jp=A</tt> are
170always flat and at the line level. Obviously, the source code files need
171to be readable by the profiler script.
172</p>
173<p>
174The high-level profiler can also be started and stopped from Lua code with:
175</p>
176<pre class="code">
177require("jit.p").start(options, output)
178...
179require("jit.p").stop()
180</pre>
181
182<h3 id="jit_zone"><tt>jit.zone</tt> &mdash; Zones</h3>
183<p>
184Zones can be used to provide information about different parts of an
185application to the high-level profiler. E.g. a game could make use of an
186<tt>"AI"</tt> zone, a <tt>"PHYS"</tt> zone, etc. Zones are hierarchical,
187organized as a stack.
188</p>
189<p>
190The <tt>jit.zone</tt> module needs to be loaded explicitly:
191</p>
192<pre class="code">
193local zone = require("jit.zone")
194</pre>
195<ul>
196<li><tt>zone("name")</tt> pushes a named zone to the zone stack.</li>
197<li><tt>zone()</tt> pops the current zone from the zone stack and
198returns its name.</li>
199<li><tt>zone:get()</tt> returns the current zone name or <tt>nil</tt>.</li>
200<li><tt>zone:flush()</tt> flushes the zone stack.</li>
201</ul>
202<p>
203To show the time spent in each zone use <tt>-jp=z</tt>. To show the time
204spent relative to hotspots use e.g. <tt>-jp=zf</tt> or <tt>-jp=fz</tt>.
205</p>
206
207<h2 id="ll_lua_api">Low-level Lua API</h2>
208<p>
209The <tt>jit.profile</tt> module gives access to the low-level API of the
210profiler from Lua code. This module needs to be loaded explicitly:
211<pre class="code">
212local profile = require("jit.profile")
213</pre>
214<p>
215This module can be used to implement your own higher-level profiler.
216A typical profiling run starts the profiler, captures stack dumps in
217the profiler callback, adds them to a hash table to aggregate the number
218of samples, stops the profiler and then analyzes all captured
219stack dumps. Other parameters can be sampled in the profiler callback,
220too. But it's important not to spend too much time in the callback,
221since this may skew the statistics.
222</p>
223
224<h3 id="profile_start"><tt>profile.start(mode, cb)</tt>
225&mdash; Start profiler</h3>
226<p>
227This function starts the profiler. The <tt>mode</tt> argument is a
228string holding options:
229</p>
230<ul>
231<li><tt>f</tt> &mdash; Profile with precision down to the function level.</li>
232<li><tt>l</tt> &mdash; Profile with precision down to the line level.</li>
233<li><tt>i&lt;number&gt;</tt> &mdash; Sampling interval in milliseconds (default
23410ms).</br>
235Note: The actual sampling precision is OS-dependent.
236</li>
237</ul>
238<p>
239The <tt>cb</tt> argument is a callback function which is called with
240three arguments: <tt>(thread, samples, vmstate)</tt>. The callback is
241called on a separate coroutine, the <tt>thread</tt> argument is the
242state that holds the stack to sample for profiling. Note: do
243<em>not</em> modify the stack of that state or call functions on it.
244</p>
245<p>
246<tt>samples</tt> gives the number of accumulated samples since the last
247callback (usually 1).
248</p>
249<p>
250<tt>vmstate</tt> holds the VM state at the time the profiling timer
251triggered. This may or may not correspond to the state of the VM when
252the profiling callback is called. The state is either <tt>'N'</tt>
253native (compiled) code, <tt>'I'</tt> interpreted code, <tt>'C'</tt>
254C&nbsp;code, <tt>'G'</tt> the garbage collector, or <tt>'J'</tt> the JIT
255compiler.
256</p>
257
258<h3 id="profile_stop"><tt>profile.stop()</tt>
259&mdash; Stop profiler</h3>
260<p>
261This function stops the profiler.
262</p>
263
264<h3 id="profile_dump"><tt>dump = profile.dumpstack([thread,] fmt, depth)</tt>
265&mdash; Dump stack </h3>
266<p>
267This function allows taking stack dumps in an efficient manner. It
268returns a string with a stack dump for the <tt>thread</tt> (coroutine),
269formatted according to the <tt>fmt</tt> argument:
270</p>
271<ul>
272<li><tt>p</tt> &mdash; Preserve the full path for module names. Otherwise,
273only the file name is used.</li>
274<li><tt>f</tt> &mdash; Dump the function name if it can be derived. Otherwise,
275use module:line.</li>
276<li><tt>F</tt> &mdash; Ditto, but dump module:name.</li>
277<li><tt>l</tt> &mdash; Dump module:line.</li>
278<li><tt>Z</tt> &mdash; Zap the following characters for the last dumped
279frame.</li>
280<li>All other characters are added verbatim to the output string.</li>
281</ul>
282<p>
283The <tt>depth</tt> argument gives the number of frames to dump, starting
284at the topmost frame of the thread. A negative number dumps the frames in
285inverse order.
286</p>
287<p>
288The first example prints a list of the current module names and line
289numbers of up to 10 frames in separate lines. The second example prints
290semicolon-separated function names for all frames (up to 100) in inverse
291order:
292</p>
293<pre class="code">
294print(profile.dumpstack(thread, "l\n", 10))
295print(profile.dumpstack(thread, "lZ;", -100))
296</pre>
297
298<h2 id="ll_c_api">Low-level C API</h2>
299<p>
300The profiler can be controlled directly from C&nbsp;code, e.g. for
301use by IDEs. The declarations are in <tt>"luajit.h"</tt> (see
302<a href="ext_c_api.html">Lua/C API</a> extensions).
303</p>
304
305<h3 id="luaJIT_profile_start"><tt>luaJIT_profile_start(L, mode, cb, data)</tt>
306&mdash; Start profiler</h3>
307<p>
308This function starts the profiler. <a href="#profile_start">See
309above</a> for a description of the <tt>mode</tt> argument.
310</p>
311<p>
312The <tt>cb</tt> argument is a callback function with the following
313declaration:
314</p>
315<pre class="code">
316typedef void (*luaJIT_profile_callback)(void *data, lua_State *L,
317 int samples, int vmstate);
318</pre>
319<p>
320<tt>data</tt> is available for use by the callback. <tt>L</tt> is the
321state that holds the stack to sample for profiling. Note: do
322<em>not</em> modify this stack or call functions on this stack &mdash;
323use a separate coroutine for this purpose. <a href="#profile_start">See
324above</a> for a description of <tt>samples</tt> and <tt>vmstate</tt>.
325</p>
326
327<h3 id="luaJIT_profile_stop"><tt>luaJIT_profile_stop(L)</tt>
328&mdash; Stop profiler</h3>
329<p>
330This function stops the profiler.
331</p>
332
333<h3 id="luaJIT_profile_dumpstack"><tt>p = luaJIT_profile_dumpstack(L, fmt, depth, len)</tt>
334&mdash; Dump stack </h3>
335<p>
336This function allows taking stack dumps in an efficient manner.
337<a href="#profile_dump">See above</a> for a description of <tt>fmt</tt>
338and <tt>depth</tt>.
339</p>
340<p>
341This function returns a <tt>const&nbsp;char&nbsp;*</tt> pointing to a
342private string buffer of the profiler. The <tt>int&nbsp;*len</tt>
343argument returns the length of the output string. The buffer is
344overwritten on the next call and deallocated when the profiler stops.
345You either need to consume the content immediately or copy it for later
346use.
347</p>
348<br class="flush">
349</div>
350<div id="foot">
351<hr class="hide">
352Copyright &copy; 2005-2026
353<span class="noprint">
354&middot;
355<a href="contact.html">Contact</a>
356</span>
357</div>
358</body>
359</html>
generated by cgit v1.2.3-55-g6feb (git 2.39.1) at 2026-06-04 05:45:11 +0000