1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
|
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>Benchmarking tips — LLVM 13 documentation</title>
<link rel="stylesheet" href="_static/pygments.css" type="text/css" />
<link rel="stylesheet" href="_static/llvm-theme.css" type="text/css" />
<script id="documentation_options" data-url_root="./" src="_static/documentation_options.js"></script>
<script src="_static/jquery.js"></script>
<script src="_static/underscore.js"></script>
<script src="_static/doctools.js"></script>
<link rel="index" title="Index" href="genindex.html" />
<link rel="search" title="Search" href="search.html" />
<link rel="next" title="Using ARM NEON instructions in big endian mode" href="BigEndianNEON.html" />
<link rel="prev" title="DWARF Extensions For Heterogeneous Debugging" href="AMDGPUDwarfExtensionsForHeterogeneousDebugging.html" />
<style type="text/css">
table.right { float: right; margin-left: 20px; }
table.right td { border: 1px solid #ccc; }
</style>
</head><body>
<div class="logo">
<a href="index.html">
<img src="_static/logo.png"
alt="LLVM Logo" width="250" height="88"/></a>
</div>
<div class="related" role="navigation" aria-label="related navigation">
<h3>Navigation</h3>
<ul>
<li class="right" style="margin-right: 10px">
<a href="genindex.html" title="General Index"
accesskey="I">index</a></li>
<li class="right" >
<a href="BigEndianNEON.html" title="Using ARM NEON instructions in big endian mode"
accesskey="N">next</a> |</li>
<li class="right" >
<a href="AMDGPUDwarfExtensionsForHeterogeneousDebugging.html" title="DWARF Extensions For Heterogeneous Debugging"
accesskey="P">previous</a> |</li>
<li><a href="https://llvm.org/">LLVM Home</a> | </li>
<li><a href="index.html">Documentation</a>»</li>
<li class="nav-item nav-item-1"><a href="UserGuides.html" accesskey="U">User Guides</a> »</li>
<li class="nav-item nav-item-this"><a href="">Benchmarking tips</a></li>
</ul>
</div>
<div class="sphinxsidebar" role="navigation" aria-label="main navigation">
<div class="sphinxsidebarwrapper">
<h3>Documentation</h3>
<ul class="want-points">
<li><a href="https://llvm.org/docs/GettingStartedTutorials.html">Getting Started/Tutorials</a></li>
<li><a href="https://llvm.org/docs/UserGuides.html">User Guides</a></li>
<li><a href="https://llvm.org/docs/Reference.html">Reference</a></li>
</ul>
<h3>Getting Involved</h3>
<ul class="want-points">
<li><a href="https://llvm.org/docs/Contributing.html">Contributing to LLVM</a></li>
<li><a href="https://llvm.org/docs/HowToSubmitABug.html">Submitting Bug Reports</a></li>
<li><a href="https://llvm.org/docs/GettingInvolved.html#mailing-lists">Mailing Lists</a></li>
<li><a href="https://llvm.org/docs/GettingInvolved.html#irc">IRC</a></li>
<li><a href="https://llvm.org/docs/GettingInvolved.html#meetups-and-social-events">Meetups and Social Events</a></li>
</ul>
<h3>Additional Links</h3>
<ul class="want-points">
<li><a href="https://llvm.org/docs/FAQ.html">FAQ</a></li>
<li><a href="https://llvm.org/docs/Lexicon.html">Glossary</a></li>
<li><a href="https://llvm.org/pubs">Publications</a></li>
<li><a href="https://github.com/llvm/llvm-project//">Github Repository</a></li>
</ul>
<div role="note" aria-label="source link">
<h3>This Page</h3>
<ul class="this-page-menu">
<li><a href="_sources/Benchmarking.rst.txt"
rel="nofollow">Show Source</a></li>
</ul>
</div>
<div id="searchbox" style="display: none" role="search">
<h3 id="searchlabel">Quick search</h3>
<div class="searchformwrapper">
<form class="search" action="search.html" method="get">
<input type="text" name="q" aria-labelledby="searchlabel" />
<input type="submit" value="Go" />
</form>
</div>
</div>
<script>$('#searchbox').show(0);</script>
</div>
</div>
<div class="document">
<div class="documentwrapper">
<div class="bodywrapper">
<div class="body" role="main">
<div class="section" id="benchmarking-tips">
<h1>Benchmarking tips<a class="headerlink" href="#benchmarking-tips" title="Permalink to this headline">¶</a></h1>
<div class="section" id="introduction">
<h2>Introduction<a class="headerlink" href="#introduction" title="Permalink to this headline">¶</a></h2>
<p>For benchmarking a patch we want to reduce all possible sources of
noise as much as possible. How to do that is very OS dependent.</p>
<p>Note that low noise is required, but not sufficient. It does not
exclude measurement bias. See
<a class="reference external" href="https://www.cis.upenn.edu/~cis501/papers/producing-wrong-data.pdf">https://www.cis.upenn.edu/~cis501/papers/producing-wrong-data.pdf</a> for
example.</p>
</div>
<div class="section" id="general">
<h2>General<a class="headerlink" href="#general" title="Permalink to this headline">¶</a></h2>
<ul>
<li><p>Use a high resolution timer, e.g. perf under linux.</p></li>
<li><p>Run the benchmark multiple times to be able to recognize noise.</p></li>
<li><p>Disable as many processes or services as possible on the target system.</p></li>
<li><p>Disable frequency scaling, turbo boost and address space
randomization (see OS specific section).</p></li>
<li><p>Static link if the OS supports it. That avoids any variation that
might be introduced by loading dynamic libraries. This can be done
by passing <code class="docutils literal notranslate"><span class="pre">-DLLVM_BUILD_STATIC=ON</span></code> to cmake.</p></li>
<li><p>Try to avoid storage. On some systems you can use tmpfs. Putting the
program, inputs and outputs on tmpfs avoids touching a real storage
system, which can have a pretty big variability.</p>
<p>To mount it (on linux and freebsd at least):</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">mount</span> <span class="o">-</span><span class="n">t</span> <span class="n">tmpfs</span> <span class="o">-</span><span class="n">o</span> <span class="n">size</span><span class="o">=<</span><span class="n">XX</span><span class="o">></span><span class="n">g</span> <span class="n">none</span> <span class="n">dir_to_mount</span>
</pre></div>
</div>
</li>
</ul>
</div>
<div class="section" id="linux">
<h2>Linux<a class="headerlink" href="#linux" title="Permalink to this headline">¶</a></h2>
<ul>
<li><p>Disable address space randomization:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">echo</span> <span class="mi">0</span> <span class="o">></span> <span class="o">/</span><span class="n">proc</span><span class="o">/</span><span class="n">sys</span><span class="o">/</span><span class="n">kernel</span><span class="o">/</span><span class="n">randomize_va_space</span>
</pre></div>
</div>
</li>
<li><p>Set scaling_governor to performance:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="o">/</span><span class="n">sys</span><span class="o">/</span><span class="n">devices</span><span class="o">/</span><span class="n">system</span><span class="o">/</span><span class="n">cpu</span><span class="o">/</span><span class="n">cpu</span><span class="o">*/</span><span class="n">cpufreq</span><span class="o">/</span><span class="n">scaling_governor</span>
<span class="n">do</span>
<span class="n">echo</span> <span class="n">performance</span> <span class="o">></span> <span class="o">/</span><span class="n">sys</span><span class="o">/</span><span class="n">devices</span><span class="o">/</span><span class="n">system</span><span class="o">/</span><span class="n">cpu</span><span class="o">/</span><span class="n">cpu</span><span class="o">*/</span><span class="n">cpufreq</span><span class="o">/</span><span class="n">scaling_governor</span>
<span class="n">done</span>
</pre></div>
</div>
</li>
<li><p>Use <a class="reference external" href="https://github.com/lpechacek/cpuset">https://github.com/lpechacek/cpuset</a> to reserve cpus for just the
program you are benchmarking. If using perf, leave at least 2 cores
so that perf runs in one and your program in another:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">cset</span> <span class="n">shield</span> <span class="o">-</span><span class="n">c</span> <span class="n">N1</span><span class="p">,</span><span class="n">N2</span> <span class="o">-</span><span class="n">k</span> <span class="n">on</span>
</pre></div>
</div>
<p>This will move all threads out of N1 and N2. The <code class="docutils literal notranslate"><span class="pre">-k</span> <span class="pre">on</span></code> means
that even kernel threads are moved out.</p>
</li>
<li><p>Disable the SMT pair of the cpus you will use for the benchmark. The
pair of cpu N can be found in
<code class="docutils literal notranslate"><span class="pre">/sys/devices/system/cpu/cpuN/topology/thread_siblings_list</span></code> and
disabled with:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">echo</span> <span class="mi">0</span> <span class="o">></span> <span class="o">/</span><span class="n">sys</span><span class="o">/</span><span class="n">devices</span><span class="o">/</span><span class="n">system</span><span class="o">/</span><span class="n">cpu</span><span class="o">/</span><span class="n">cpuX</span><span class="o">/</span><span class="n">online</span>
</pre></div>
</div>
</li>
<li><p>Run the program with:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">cset</span> <span class="n">shield</span> <span class="o">--</span><span class="n">exec</span> <span class="o">--</span> <span class="n">perf</span> <span class="n">stat</span> <span class="o">-</span><span class="n">r</span> <span class="mi">10</span> <span class="o"><</span><span class="n">cmd</span><span class="o">></span>
</pre></div>
</div>
<p>This will run the command after <code class="docutils literal notranslate"><span class="pre">--</span></code> in the isolated cpus. The
particular perf command runs the <code class="docutils literal notranslate"><span class="pre"><cmd></span></code> 10 times and reports
statistics.</p>
</li>
</ul>
<p>With these in place you can expect perf variations of less than 0.1%.</p>
<div class="section" id="linux-intel">
<h3>Linux Intel<a class="headerlink" href="#linux-intel" title="Permalink to this headline">¶</a></h3>
<ul>
<li><p>Disable turbo mode:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">echo</span> <span class="mi">1</span> <span class="o">></span> <span class="o">/</span><span class="n">sys</span><span class="o">/</span><span class="n">devices</span><span class="o">/</span><span class="n">system</span><span class="o">/</span><span class="n">cpu</span><span class="o">/</span><span class="n">intel_pstate</span><span class="o">/</span><span class="n">no_turbo</span>
</pre></div>
</div>
</li>
</ul>
</div>
</div>
</div>
<div class="clearer"></div>
</div>
</div>
</div>
<div class="clearer"></div>
</div>
<div class="related" role="navigation" aria-label="related navigation">
<h3>Navigation</h3>
<ul>
<li class="right" style="margin-right: 10px">
<a href="genindex.html" title="General Index"
>index</a></li>
<li class="right" >
<a href="BigEndianNEON.html" title="Using ARM NEON instructions in big endian mode"
>next</a> |</li>
<li class="right" >
<a href="AMDGPUDwarfExtensionsForHeterogeneousDebugging.html" title="DWARF Extensions For Heterogeneous Debugging"
>previous</a> |</li>
<li><a href="https://llvm.org/">LLVM Home</a> | </li>
<li><a href="index.html">Documentation</a>»</li>
<li class="nav-item nav-item-1"><a href="UserGuides.html" >User Guides</a> »</li>
<li class="nav-item nav-item-this"><a href="">Benchmarking tips</a></li>
</ul>
</div>
<div class="footer" role="contentinfo">
© Copyright 2003-2021, LLVM Project.
Last updated on 2021-09-18.
Created using <a href="https://www.sphinx-doc.org/">Sphinx</a> 3.5.4.
</div>
</body>
</html>
|
