Gitiles
Code Review Sign In
gerrit-public.fairphone.software / platform / external / llvm / 18ab2b91ff2526ec899b564f53a082a63770539c / . / docs / HowToSubmitABug.html
blob: 9d6c6e5f8f02485ed9fc0c1439917aa3be1cda7f [file] [log] [blame]
Chris Lattner8b6be362003-05-21 22:21:07 +0000[diff] [blame]1<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
2<html><head><title>How to submit an LLVM bug report</title></head>
3
4<body bgcolor=white>
5
6<table width="100%" bgcolor="#330077" border=0 cellpadding=4 cellspacing=0>
7<tr><td>&nbsp; <font size=+3 color="#EEEEFF" face="Georgia,Palatino,Times,Roman"><b>How to submit an LLVM bug report</b></font></td>
8</tr></table>
9
10<table border=0 width=100%>
11<tr><td valign=top>
12<p><font size=+1>
13<ol>
14 <li><a href="#introduction">Introduction - Got bugs?</a>
15 <li><a href="#crashers">Crashing Bugs</a>
16 <ul>
17 <li><a href="#front-end">Front-end bugs</a>
18 <li><a href="#gccas">GCCAS bugs</a>
19 <li><a href="#gccld">GCCLD bugs</a>
20 <li><a href="#passes">Bugs in LLVM passes</a>
21 </ul>
22 <li><a href="#miscompilations">Miscompilations</a>
Misha Brukmanf1d01fb2003-09-17 18:51:47 +0000[diff] [blame]23 <li><a href="#codegen">Incorrect code generation (JIT and LLC)</a>
Chris Lattner8b6be362003-05-21 22:21:07 +0000[diff] [blame]24
Misha Brukmanf1d01fb2003-09-17 18:51:47 +0000[diff] [blame]25 <p><b>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a> and
26 <a href="http://misha.brukman.net">Misha Brukman</a></b><p>
Chris Lattner8b6be362003-05-21 22:21:07 +0000[diff] [blame]27</ol><p></font>
28</td><td valign=top align=right>
29<img src="Debugging.gif" width=444 height=314>
30</td></tr>
31</table>
32
33
34<!-- *********************************************************************** -->
35<table width="100%" bgcolor="#330077" border=0 cellpadding=4 cellspacing=0>
36<tr><td align=center><font color="#EEEEFF" size=+2 face="Georgia,Palatino"><b>
37<a name="introduction">Introduction - Got bugs?
38</b></font></td></tr></table><ul>
39<!-- *********************************************************************** -->
40
41If you're working with LLVM and run into a bug, we definitely want to know about
42it. This document describes what you can do to increase the odds of getting it
43fixed quickly.<p>
44
45Basically you have to do two things at a minimum. First, decide whether the bug
46<a href="#crashers">crashes the compiler</a> (or an LLVM pass), or if the
47compiler is <a href="#miscompilations">miscompiling</a> the program. Based on
48what type of bug it is, follow the instructions in the linked section to narrow
49down the bug so that the person who fixes it will be able to find the problem
50more easily.<p>
51
52Once you have a reduced test-case, email information about the bug to: <a
53href="mailto:llvmbugs@cs.uiuc.edu">llvmbugs@cs.uiuc.edu</a>. This should
54include all of the information necessary to reproduce the problem, including
55where you got the LLVM tree from (if you're not working out of CVS).<p>
56
57Thanks for helping us make LLVM better!<p>
58
59
60<!-- *********************************************************************** -->
61</ul><table width="100%" bgcolor="#330077" border=0 cellpadding=4 cellspacing=0>
62<tr><td align=center><font color="#EEEEFF" size=+2 face="Georgia,Palatino"><b>
63<a name="crashers">Crashing Bugs
64</b></font></td></tr></table><ul>
65<!-- *********************************************************************** -->
66
67More often than not, bugs in the compiler cause it to crash - often due to an
68assertion failure of some sort. If you are running <tt><b>opt</b></tt> or
69<tt><b>analyze</b></tt> directly, and something crashes, jump to the section on
70<a href="#passes">bugs in LLVM passes</a>. Otherwise, the most important
71piece of the puzzle is to figure out if it is the GCC-based front-end that is
72buggy or if it's one of the LLVM tools that has problems.<p>
73
74To figure out which program is crashing (the front-end, <tt><b>gccas</b></tt>,
75or <tt><b>gccld</b></tt>), run the <tt><b>llvm-gcc</b></tt> command line as you
76were when the crash occurred, but add a <tt>-v</tt> option to the command line.
77The compiler will print out a bunch of stuff, and should end with telling you
78that one of <tt><b>cc1</b></tt>, <tt><b>gccas</b></tt>, or <tt><b>gccld</b></tt>
79crashed.<p>
80
81<ul>
82<li>If <tt><b>cc1</b></tt> crashed, you found a problem with the front-end.
83Jump ahead to the section on <a href="#front-end">front-end bugs</a>.
84<li>If <tt><b>gccas</b></tt> crashed, you found a bug in <a href="#gccas">one of
85the passes in <tt><b>gccas</b></tt></a>.
86<li>If <tt><b>gccld</b></tt> crashed, you found a bug in <a href="#gccld">one
87of the passes in <tt><b>gccld</b></tt></a>.
88<li>Otherwise, something really weird happened. Email the list with what you
89have at this point.
90</ul><p>
91
92
93<!-- ======================================================================= -->
94</ul><table width="100%" bgcolor="#441188" border=0 cellpadding=4 cellspacing=0>
95<tr><td>&nbsp;</td><td width="100%">&nbsp;
96<font color="#EEEEFF" face="Georgia,Palatino"><b>
97<a name="front-end">Front-end bugs
98</b></font></td></tr></table><ul>
99
Brian Gaekeeefe9792003-05-23 21:03:50 +0000[diff] [blame]100If the problem is in the front-end, you should re-run the same
101<tt>llvm-gcc</tt> command that resulted in the crash, but add the
102<tt>-save-temps</tt> option. The compiler will crash again, but it
103will leave behind a <tt><i>foo</i>.i</tt> file (containing preprocessed
104C source code) and possibly <tt><i>foo</i>.s</tt> (containing LLVM
105assembly code), for each compiled <tt><i>foo</i>.c</tt> file. Send us
106the <tt><i>foo</i>.i</tt> file, along with a brief description of the
107error it caused.<p>
Chris Lattner8b6be362003-05-21 22:21:07 +0000[diff] [blame]108
109<!-- ======================================================================= -->
110</ul><table width="100%" bgcolor="#441188" border=0 cellpadding=4 cellspacing=0>
111<tr><td>&nbsp;</td><td width="100%">&nbsp;
112<font color="#EEEEFF" face="Georgia,Palatino"><b>
113<a name="gccas">GCCAS bugs
114</b></font></td></tr></table><ul>
115
116If you find that a bug crashes in the <tt><b>gccas</b></tt> stage of
Brian Gaekeeefe9792003-05-23 21:03:50 +0000[diff] [blame]117compilation, compile your test-case to a <tt>.s</tt> file with the
118<tt>-save-temps</tt> option to <tt><b>llvm-gcc</b></tt>. Then run:<p>
Chris Lattner8b6be362003-05-21 22:21:07 +0000[diff] [blame]119
120<pre>
121 <b>gccas</b> -debug-pass=Arguments &lt; /dev/null -o - &gt; /dev/null
122</pre><p>
123
124... which will print a list of arguments, indicating the list of passes that
125<tt><b>gccas</b></tt> runs. Once you have the input file and the list of
126passes, go to the section on <a href="#passes">debugging bugs in LLVM
127passes</a>.<p>
128
129
130<!-- ======================================================================= -->
131</ul><table width="100%" bgcolor="#441188" border=0 cellpadding=4 cellspacing=0>
132<tr><td>&nbsp;</td><td width="100%">&nbsp;
133<font color="#EEEEFF" face="Georgia,Palatino"><b>
134<a name="gccld">GCCLD bugs
135</b></font></td></tr></table><ul>
136
137If you find that a bug crashes in the <tt><b>gccld</b></tt> stage of
138compilation, gather all of the <tt>.o</tt> bytecode files and libraries that are
139being linked together (the "<tt><b>llvm-gcc</b> -v</tt>" output should include
140the full list of objects linked). Then run:<p>
141
142<pre>
Misha Brukmanf1d01fb2003-09-17 18:51:47 +0000[diff] [blame]143 <b>llvm-as</b> &lt; /dev/null &gt; null.bc
Chris Lattnerddaf5002003-05-23 14:49:32 +0000[diff] [blame]144 <b>gccld</b> -debug-pass=Arguments null.bc
Chris Lattner8b6be362003-05-21 22:21:07 +0000[diff] [blame]145</pre><p>
146
147... which will print a list of arguments, indicating the list of passes that
148<tt><b>gccld</b></tt> runs. Once you have the input files and the list of
149passes, go to the section on <a href="#passes">debugging bugs in LLVM
150passes</a>.<p>
151
152<!-- ======================================================================= -->
153</ul><table width="100%" bgcolor="#441188" border=0 cellpadding=4 cellspacing=0>
154<tr><td>&nbsp;</td><td width="100%">&nbsp;
155<font color="#EEEEFF" face="Georgia,Palatino"><b>
156<a name="passes">Bugs in LLVM passes
157</b></font></td></tr></table><ul>
158
159At this point, you should have some number of LLVM assembly files or bytecode
160files and a list of passes which crash when run on the specified input. In
161order to reduce the list of passes (which is probably large) and the input to
162something tractable, use the <tt><b>bugpoint</b></tt> tool as follows:<p>
163
164<pre>
165 <b>bugpoint</b> &lt;input files&gt; &lt;list of passes&gt;
166</pre><p>
167
168<tt><b>bugpoint</b></tt> will print a bunch of output as it reduces the
169test-case, but it should eventually print something like this:<p>
170
171<pre>
172 ...
173 Emitted bytecode to 'bugpoint-reduced-simplified.bc'
174
175 *** You can reproduce the problem with: opt bugpoint-reduced-simplified.bc -licm
176</pre><p>
177
178Once you complete this, please send the LLVM bytecode file and the command line
179to reproduce the problem to the llvmbugs mailing list.<p>
180
181
182<!-- *********************************************************************** -->
183</ul><table width="100%" bgcolor="#330077" border=0 cellpadding=4 cellspacing=0>
184<tr><td align=center><font color="#EEEEFF" size=+2 face="Georgia,Palatino"><b>
Misha Brukmanf1d01fb2003-09-17 18:51:47 +0000[diff] [blame]185<a name="miscompilations">Miscompilations</a>
Chris Lattner8b6be362003-05-21 22:21:07 +0000[diff] [blame]186</b></font></td></tr></table><ul>
187<!-- *********************************************************************** -->
188
Misha Brukmanf1d01fb2003-09-17 18:51:47 +0000[diff] [blame]189A miscompilation occurs when a pass does not correctly transform a program, thus
190producing errors that are only noticed during execution. This is different from
191producing invalid LLVM code (i.e., code not in SSA form, using values before
192defining them, etc.) which the verifier will check for after a pass finishes its
193run.<p>
Chris Lattner8b6be362003-05-21 22:21:07 +0000[diff] [blame]194
Misha Brukmanf1d01fb2003-09-17 18:51:47 +0000[diff] [blame]195To debug a miscompilation, you should choose which program you wish to run the
196output through, e.g. C backend, the JIT, or LLC, and a selection of passes, one
197of which may be causing the error, and run, for example:
198
199<pre>
Chris Lattner18ab2b92003-10-14 20:58:07 +0000[diff] [blame^]200 <b>bugpoint</b> -run-cbe [... optimization passes ...] file-to-test.bc
Misha Brukmanf1d01fb2003-09-17 18:51:47 +0000[diff] [blame]201</pre>
202
203<tt>bugpoint</tt> will try to narrow down your list of passes to the one pass
204that causes an error, and simplify the bytecode file as much as it can to assist
205you. It will print a message letting you know how to reproduce the resulting
206error.
207
208<!-- *********************************************************************** -->
209</ul><table width="100%" bgcolor="#330077" border=0 cellpadding=4 cellspacing=0>
210<tr><td align=center><font color="#EEEEFF" size=+2 face="Georgia,Palatino"><b>
211<a name="codegen">Incorrect code generation</a>
212</b></font></td></tr></table><ul>
213<!-- *********************************************************************** -->
214
215Similarly to debugging incorrect compilation by mis-behaving passes, you can
216debug incorrect code generation by either LLC or the JIT, using
217<tt>bugpoint</tt>. The process <tt>bugpoint</tt> follows in this case is to try
218to narrow the code down to a function that is miscompiled by one or the other
219method, but since for correctness, the entire program must be run,
220<tt>bugpoint</tt> will compile the code it deems to not be affected with the C
221Backend, and then link in the shared object it generates.<p>
222
223To debug the JIT:
224<pre>
Chris Lattner18ab2b92003-10-14 20:58:07 +0000[diff] [blame^]225 <b>bugpoint</b> -run-jit -output=[correct output file] [bytecodefile]
Misha Brukmanf1d01fb2003-09-17 18:51:47 +0000[diff] [blame]226</pre>
227
228Similarly, to debug the LLC, one would run:
229<pre>
Chris Lattner18ab2b92003-10-14 20:58:07 +0000[diff] [blame^]230 <b>bugpoint</b> -run-llc -output=[correct output file] [bytecodefile]
Misha Brukmanf1d01fb2003-09-17 18:51:47 +0000[diff] [blame]231</pre>
232
233At the end of a successful <tt>bugpoint</tt> run, you will be presented
234with two bytecode files: a <em>safe</em> file which can be compiled with the C
235backend and the <em>test</em> file which either LLC or the JIT
236mis-codegenerates, and thus causes the error.<p>
237
238To reproduce the error that <tt>bugpoint</tt> found, it is sufficient to do the
239following:
240
241<ol>
242 <li>Regenerate the shared object from the safe bytecode file:<br>
243<pre>
244 <b>llvm-dis</b> -c safe.bc -o safe.c<br>
245 <b>gcc</b> -shared safe.c -o safe.so
246</pre></li>
247 <li>If debugging LLC, compile test bytecode native and link with the shared object:<br>
248<pre>
249 <b>llc</b> test.bc -o test.s -f<br>
250 gcc test.s safe.so -o test.llc<br>
251 ./test.llc [program options]
252</pre></li>
253 <p>
254 If debugging the JIT, load the shared object and supply the test bytecode:<br>
255<pre>
256 <b>lli</b> -load=safe.so test.bc [program options]
257</pre></li>
258</ol>
Chris Lattner8b6be362003-05-21 22:21:07 +0000[diff] [blame]259
260<!-- *********************************************************************** -->
261</ul>
262<!-- *********************************************************************** -->
263
264<hr><font size-1>
265<address><a href="mailto:sabre@nondot.org">Chris Lattner</a></address>
266<!-- Created: Tue Aug 6 15:00:33 CDT 2002 -->
267<!-- hhmts start -->
Chris Lattner18ab2b92003-10-14 20:58:07 +0000[diff] [blame^]268Last modified: Tue Oct 14 15:57:47 CDT 2003
Chris Lattner8b6be362003-05-21 22:21:07 +0000[diff] [blame]269<!-- hhmts end -->
270</font></body></html>