Gitiles
Code Review Sign In
gerrit-public.fairphone.software / platform / external / llvm / a9498a2ee63f11ca925497199b96abcefdd70582 / . / docs / Bugpoint.html
blob: 71f288d7720ef33ad5ff1c03dd10a64f1bee3ed2 [file] [log] [blame]
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]1<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]3<html>
Reid Spencerd206bc92004-11-01 21:55:46 +0000[diff] [blame]4<head>
NAKAMURA Takumi5c6e4df2011-10-31 11:21:59 +0000[diff] [blame]5 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]6 <title>LLVM bugpoint tool: design and usage</title>
Daniel Dunbaradea4972012-04-19 20:20:34 +0000[diff] [blame]7 <link rel="stylesheet" href="_static/llvm.css" type="text/css">
Reid Spencerd206bc92004-11-01 21:55:46 +0000[diff] [blame]8</head>
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]9
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]10<h1>
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]11 LLVM bugpoint tool: design and usage
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]12</h1>
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]13
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]14<ul>
15 <li><a href="#desc">Description</a></li>
16 <li><a href="#design">Design Philosophy</a>
17 <ul>
18 <li><a href="#autoselect">Automatic Debugger Selection</a></li>
19 <li><a href="#crashdebug">Crash debugger</a></li>
20 <li><a href="#codegendebug">Code generator debugger</a></li>
21 <li><a href="#miscompilationdebug">Miscompilation debugger</a></li>
22 </ul></li>
23 <li><a href="#advice">Advice for using <tt>bugpoint</tt></a></li>
Joel Jonesa9498a22012-04-20 18:20:24 +0000[diff] [blame^]24 <li><a href="#notEnough">What to do when <tt>bugpoint</tt> isn't enough</a></li>
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]25</ul>
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]26
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]27<div class="doc_author">
28<p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a></p>
29</div>
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]30
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]31<!-- *********************************************************************** -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]32<h2>
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]33<a name="desc">Description</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]34</h2>
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]35<!-- *********************************************************************** -->
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]36
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000[diff] [blame]37<div>
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]38
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]39<p><tt>bugpoint</tt> narrows down the source of problems in LLVM tools and
40passes. It can be used to debug three types of failures: optimizer crashes,
41miscompilations by optimizers, or bad native code generation (including problems
42in the static and JIT compilers). It aims to reduce large test cases to small,
Reid Spencer434262a2007-02-09 15:59:08 +0000[diff] [blame]43useful ones. For example, if <tt>opt</tt> crashes while optimizing a
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]44file, it will identify the optimization (or combination of optimizations) that
45causes the crash, and reduce the file down to a small example which triggers the
46crash.</p>
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]47
Michael J. Spencer75338092012-04-19 19:27:54 +0000[diff] [blame]48<p>For detailed case scenarios, such as debugging <tt>opt</tt>, or one of the
49LLVM code generators, see <a href="HowToSubmitABug.html">How To Submit a Bug
50Report document</a>.</p>
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]51
52</div>
53
54<!-- *********************************************************************** -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]55<h2>
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]56<a name="design">Design Philosophy</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]57</h2>
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]58<!-- *********************************************************************** -->
59
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000[diff] [blame]60<div>
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]61
62<p><tt>bugpoint</tt> is designed to be a useful tool without requiring any
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]63hooks into the LLVM infrastructure at all. It works with any and all LLVM
64passes and code generators, and does not need to "know" how they work. Because
65of this, it may appear to do stupid things or miss obvious
66simplifications. <tt>bugpoint</tt> is also designed to trade off programmer
67time for computer time in the compiler-debugging process; consequently, it may
68take a long period of (unattended) time to reduce a test case, but we feel it
69is still worth it. Note that <tt>bugpoint</tt> is generally very quick unless
70debugging a miscompilation where each test of the program (which requires
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]71executing it) takes a long time.</p>
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]72
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]73<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]74<h3>
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]75 <a name="autoselect">Automatic Debugger Selection</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]76</h3>
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]77
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000[diff] [blame]78<div>
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]79
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]80<p><tt>bugpoint</tt> reads each <tt>.bc</tt> or <tt>.ll</tt> file specified on
81the command line and links them together into a single module, called the test
82program. If any LLVM passes are specified on the command line, it runs these
83passes on the test program. If any of the passes crash, or if they produce
84malformed output (which causes the verifier to abort), <tt>bugpoint</tt> starts
85the <a href="#crashdebug">crash debugger</a>.</p>
86
87<p>Otherwise, if the <tt>-output</tt> option was not specified,
88<tt>bugpoint</tt> runs the test program with the C backend (which is assumed to
89generate good code) to generate a reference output. Once <tt>bugpoint</tt> has
90a reference output for the test program, it tries executing it with the
91selected code generator. If the selected code generator crashes,
92<tt>bugpoint</tt> starts the <a href="#crashdebug">crash debugger</a> on the
93code generator. Otherwise, if the resulting output differs from the reference
94output, it assumes the difference resulted from a code generator failure, and
95starts the <a href="#codegendebug">code generator debugger</a>.</p>
96
97<p>Finally, if the output of the selected code generator matches the reference
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]98output, <tt>bugpoint</tt> runs the test program after all of the LLVM passes
99have been applied to it. If its output differs from the reference output, it
100assumes the difference resulted from a failure in one of the LLVM passes, and
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]101enters the <a href="#miscompilationdebug">miscompilation debugger</a>.
102Otherwise, there is no problem <tt>bugpoint</tt> can debug.</p>
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]103
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]104</div>
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]105
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]106<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]107<h3>
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]108 <a name="crashdebug">Crash debugger</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]109</h3>
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]110
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000[diff] [blame]111<div>
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]112
113<p>If an optimizer or code generator crashes, <tt>bugpoint</tt> will try as hard
114as it can to reduce the list of passes (for optimizer crashes) and the size of
115the test program. First, <tt>bugpoint</tt> figures out which combination of
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]116optimizer passes triggers the bug. This is useful when debugging a problem
Reid Spencer434262a2007-02-09 15:59:08 +0000[diff] [blame]117exposed by <tt>opt</tt>, for example, because it runs over 38 passes.</p>
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]118
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]119<p>Next, <tt>bugpoint</tt> tries removing functions from the test program, to
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]120reduce its size. Usually it is able to reduce a test program to a single
121function, when debugging intraprocedural optimizations. Once the number of
122functions has been reduced, it attempts to delete various edges in the control
123flow graph, to reduce the size of the function as much as possible. Finally,
124<tt>bugpoint</tt> deletes any individual LLVM instructions whose absence does
125not eliminate the failure. At the end, <tt>bugpoint</tt> should tell you what
Gabor Greif04367bf2007-07-06 22:07:22 +0000[diff] [blame]126passes crash, give you a bitcode file, and give you instructions on how to
Reid Spencer84f82f72006-08-28 00:34:19 +0000[diff] [blame]127reproduce the failure with <tt>opt</tt> or <tt>llc</tt>.</p>
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]128
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]129</div>
130
131<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]132<h3>
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]133 <a name="codegendebug">Code generator debugger</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]134</h3>
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]135
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000[diff] [blame]136<div>
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]137
138<p>The code generator debugger attempts to narrow down the amount of code that
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]139is being miscompiled by the selected code generator. To do this, it takes the
140test program and partitions it into two pieces: one piece which it compiles
141with the C backend (into a shared object), and one piece which it runs with
142either the JIT or the static LLC compiler. It uses several techniques to
143reduce the amount of code pushed through the LLVM code generator, to reduce the
Gabor Greif04367bf2007-07-06 22:07:22 +0000[diff] [blame]144potential scope of the problem. After it is finished, it emits two bitcode
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]145files (called "test" [to be compiled with the code generator] and "safe" [to be
146compiled with the C backend], respectively), and instructions for reproducing
147the problem. The code generator debugger assumes that the C backend produces
148good code.</p>
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]149
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]150</div>
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]151
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]152<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]153<h3>
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]154 <a name="miscompilationdebug">Miscompilation debugger</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]155</h3>
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]156
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000[diff] [blame]157<div>
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]158
159<p>The miscompilation debugger works similarly to the code generator debugger.
160It works by splitting the test program into two pieces, running the
161optimizations specified on one piece, linking the two pieces back together, and
162then executing the result. It attempts to narrow down the list of passes to
163the one (or few) which are causing the miscompilation, then reduce the portion
164of the test program which is being miscompiled. The miscompilation debugger
165assumes that the selected code generator is working properly.</p>
166
167</div>
168
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000[diff] [blame]169</div>
170
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]171<!-- *********************************************************************** -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]172<h2>
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]173 <a name="advice">Advice for using bugpoint</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]174</h2>
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]175<!-- *********************************************************************** -->
176
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000[diff] [blame]177<div>
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]178
179<tt>bugpoint</tt> can be a remarkably useful tool, but it sometimes works in
180non-obvious ways. Here are some hints and tips:<p>
181
182<ol>
183<li>In the code generator and miscompilation debuggers, <tt>bugpoint</tt> only
184 works with programs that have deterministic output. Thus, if the program
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]185 outputs <tt>argv[0]</tt>, the date, time, or any other "random" data,
186 <tt>bugpoint</tt> may misinterpret differences in these data, when output,
187 as the result of a miscompilation. Programs should be temporarily modified
188 to disable outputs that are likely to vary from run to run.
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]189
190<li>In the code generator and miscompilation debuggers, debugging will go
191 faster if you manually modify the program or its inputs to reduce the
192 runtime, but still exhibit the problem.
193
194<li><tt>bugpoint</tt> is extremely useful when working on a new optimization:
195 it helps track down regressions quickly. To avoid having to relink
196 <tt>bugpoint</tt> every time you change your optimization however, have
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]197 <tt>bugpoint</tt> dynamically load your optimization with the
198 <tt>-load</tt> option.
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]199
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]200<li><p><tt>bugpoint</tt> can generate a lot of output and run for a long period
201 of time. It is often useful to capture the output of the program to file.
202 For example, in the C shell, you can run:</p>
203
204<div class="doc_code">
205<p><tt>bugpoint ... |&amp; tee bugpoint.log</tt></p>
206</div>
207
208 <p>to get a copy of <tt>bugpoint</tt>'s output in the file
209 <tt>bugpoint.log</tt>, as well as on your terminal.</p>
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]210
211<li><tt>bugpoint</tt> cannot debug problems with the LLVM linker. If
212 <tt>bugpoint</tt> crashes before you see its "All input ok" message,
213 you might try <tt>llvm-link -v</tt> on the same set of input files. If
214 that also crashes, you may be experiencing a linker bug.
215
Patrick Jenkins82681662006-08-15 17:03:17 +0000[diff] [blame]216<li><tt>bugpoint</tt> is useful for proactively finding bugs in LLVM.
217 Invoking <tt>bugpoint</tt> with the <tt>-find-bugs</tt> option will cause
218 the list of specified optimizations to be randomized and applied to the
Patrick Jenkins94f78502006-08-15 17:38:36 +0000[diff] [blame]219 program. This process will repeat until a bug is found or the user
Patrick Jenkins82681662006-08-15 17:03:17 +0000[diff] [blame]220 kills <tt>bugpoint</tt>.
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]221</ol>
222
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]223</div>
Joel Jones7605b292012-04-20 18:11:07 +0000[diff] [blame]224<!-- *********************************************************************** -->
225<h2>
226 <a name="notEnough">What to do when bugpoint isn't enough</a>
227</h2>
228<!-- *********************************************************************** -->
229
230<div>
231
232<p>Sometimes, <tt>bugpoint</tt> is not enough. In particular, InstCombine and
233TargetLowering both have visitor structured code with lots of potential
234transformations. If the process of using bugpoint has left you with
235still too much code to figure out and the problem seems
236to be in instcombine, the following steps may help. These same techniques
237are useful with TargetLowering as well.</p>
238
239<p>Turn on <tt>-debug-only=instcombine</tt> and see which transformations
240within instcombine are firing by selecting out lines with "<tt>IC</tt>"
241in them.</p>
242
243<p>At this point, you have a decision to make. Is the number
244of transformations small enough to step through them using a debugger?
245If so, then try that.</p>
246
247<p>If there are too many transformations, then a source modification
248approach may be helpful.
249In this approach, you can modify the source code of instcombine
250to disable just those transformations that are being performed on your
251test input and perform a binary search over the set of transformations.
252One set of places to modify are the "<tt>visit*</tt>" methods of
253<tt>InstCombiner</tt> (<I>e.g.</I> <tt>visitICmpInst</tt>) by adding a
254"<tt>return false</tt>" as the first line of the method.</p>
255
256<p>If that still doesn't remove enough, then change the caller of
257<tt>InstCombiner::DoOneIteration</tt>, <tt>InstCombiner::runOnFunction</tt>
258to limit the number of iterations.</p>
259
260<p>You may also find it useful to use "<tt>-stats</tt>" now to see what parts
261of instcombine are firing. This can guide where to put additional reporting
262code.</p>
263
264<p>At this point, if the amount of transformations is still too large, then
265inserting code to limit whether or not to execute the body of the code
266in the visit function can be helpful. Add a static counter which is
267incremented on every invocation of the function. Then add code which
268simply returns false on desired ranges. For example:</p>
269
270<div class="doc_code">
271<p><tt>static int calledCount = 0;</tt></p>
272<p><tt>calledCount++;</tt></p>
273<p><tt>DEBUG(if (calledCount &lt; 212) return false);</tt></p>
274<p><tt>DEBUG(if (calledCount &gt; 217) return false);</tt></p>
275<p><tt>DEBUG(if (calledCount == 213) return false);</tt></p>
276<p><tt>DEBUG(if (calledCount == 214) return false);</tt></p>
277<p><tt>DEBUG(if (calledCount == 215) return false);</tt></p>
278<p><tt>DEBUG(if (calledCount == 216) return false);</tt></p>
279<p><tt>DEBUG(dbgs() &lt;&lt; "visitXOR calledCount: " &lt;&lt; calledCount
280 &lt;&lt; "\n");</tt></p>
281<p><tt>DEBUG(dbgs() &lt;&lt; "I: "; I->dump());</tt></p>
282</div>
283
284<p>could be added to <tt>visitXOR</tt> to limit <tt>visitXor</tt> to being
285applied only to calls 212 and 217. This is from an actual test case and raises
286an important point---a simple binary search may not be sufficient, as
287transformations that interact may require isolating more than one call.
288In TargetLowering, use <tt>return SDNode();</tt> instead of
289<tt>return false;</tt>.</p>
290
291<p>Now that that the number of transformations is down to a manageable
292number, try examining the output to see if you can figure out which
293transformations are being done. If that can be figured out, then
294do the usual debugging. If which code corresponds to the transformation
295being performed isn't obvious, set a breakpoint after the call count
296based disabling and step through the code. Alternatively, you can use
297"printf" style debugging to report waypoints.</p>
298
299</div>
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]300
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]301<!-- *********************************************************************** -->
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]302
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]303<hr>
304<address>
305 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
Misha Brukman44408702008-12-11 17:34:48 +0000[diff] [blame]306 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]307 <a href="http://validator.w3.org/check/referer"><img
Misha Brukman44408702008-12-11 17:34:48 +0000[diff] [blame]308 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]309
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]310 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
NAKAMURA Takumib9a33632011-04-09 02:13:37 +0000[diff] [blame]311 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br>
Misha Brukmand97cb462004-12-09 20:26:20 +0000[diff] [blame]312 Last modified: $Date$
313</address>
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]314
Brian Gaeke74f470b2004-07-01 20:10:40 +0000[diff] [blame]315</body>
316</html>