blob: 69da4fafdbf2052f49f3c3c42859c70ffe451c73 [file] [log] [blame]
Misha Brukman94082392003-10-31 18:37:20 +00001<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
3<html>
4<head>
5 <title>How to submit an LLVM bug report</title>
6 <link rel="stylesheet" href="llvm.css" type="text/css">
7</head>
8<body>
Chris Lattner8b6be362003-05-21 22:21:07 +00009
Misha Brukman94082392003-10-31 18:37:20 +000010<div class="doc_title">
11 How to submit an LLVM bug report
12</div>
Chris Lattner8b6be362003-05-21 22:21:07 +000013
Misha Brukman94082392003-10-31 18:37:20 +000014<table border="0" width="100%">
15<tr>
16<td valign="top">
17
Chris Lattner8b6be362003-05-21 22:21:07 +000018<ol>
Misha Brukman94082392003-10-31 18:37:20 +000019 <li><a href="#introduction">Introduction - Got bugs?</a></li>
Chris Lattner8b6be362003-05-21 22:21:07 +000020 <li><a href="#crashers">Crashing Bugs</a>
21 <ul>
22 <li><a href="#front-end">Front-end bugs</a>
23 <li><a href="#gccas">GCCAS bugs</a>
24 <li><a href="#gccld">GCCLD bugs</a>
25 <li><a href="#passes">Bugs in LLVM passes</a>
Misha Brukman94082392003-10-31 18:37:20 +000026 </ul></li>
27 <li><a href="#miscompilations">Miscompilations</a></li>
28 <li><a href="#codegen">Incorrect code generation (JIT and LLC)</a></li>
Chris Lattner8b6be362003-05-21 22:21:07 +000029
Misha Brukman94082392003-10-31 18:37:20 +000030</ol>
31
Misha Brukman637cab02004-01-15 19:03:47 +000032<div class="doc_text">
33 <p><b>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a> and
34 <a href="http://misha.brukman.net">Misha Brukman</a></b></p>
35</div>
Misha Brukman94082392003-10-31 18:37:20 +000036
37</td>
38<td align="right">
Misha Brukman637cab02004-01-15 19:03:47 +000039 <img src="Debugging.gif" alt="Debugging" width="444" height="314">
Misha Brukman94082392003-10-31 18:37:20 +000040</td>
41</tr>
Chris Lattner8b6be362003-05-21 22:21:07 +000042</table>
43
Chris Lattner8b6be362003-05-21 22:21:07 +000044<!-- *********************************************************************** -->
Misha Brukman94082392003-10-31 18:37:20 +000045<div class="doc_section">
46 <a name="introduction">Introduction - Got bugs?</a>
47</div>
Chris Lattner8b6be362003-05-21 22:21:07 +000048<!-- *********************************************************************** -->
49
Misha Brukman94082392003-10-31 18:37:20 +000050<div class="doc_text">
Chris Lattner8b6be362003-05-21 22:21:07 +000051
Misha Brukman94082392003-10-31 18:37:20 +000052<p>If you're working with LLVM and run into a bug, we definitely want to know
53about it. This document describes what you can do to increase the odds of
54getting it fixed quickly.</p>
55
56<p>Basically you have to do two things at a minimum. First, decide whether the
57bug <a href="#crashers">crashes the compiler</a> (or an LLVM pass), or if the
Chris Lattner8b6be362003-05-21 22:21:07 +000058compiler is <a href="#miscompilations">miscompiling</a> the program. Based on
59what type of bug it is, follow the instructions in the linked section to narrow
60down the bug so that the person who fixes it will be able to find the problem
Misha Brukman94082392003-10-31 18:37:20 +000061more easily.</p>
Chris Lattner8b6be362003-05-21 22:21:07 +000062
Misha Brukman94082392003-10-31 18:37:20 +000063<p>Once you have a reduced test-case, go to <a
64href="http://llvm.cs.uiuc.edu/bugs/enter_bug.cgi">the LLVM Bug Tracking
65System</a>, select the category in which the bug falls, and fill out the form
66with the necessary details. The bug description should contain the following
67information:</p>
68
John Criswell4d37c7b2003-10-22 15:06:11 +000069<ul>
Misha Brukmanafac73e2003-10-22 17:01:44 +000070 <li>All information necessary to reproduce the problem.</li>
71 <li>The reduced test-case that triggers the bug.</li>
72 <li>The location where you obtained LLVM (if not from our CVS
73 repository).</li>
John Criswell4d37c7b2003-10-22 15:06:11 +000074</ul>
75
Misha Brukman94082392003-10-31 18:37:20 +000076<p>Thanks for helping us make LLVM better!</p>
Chris Lattner8b6be362003-05-21 22:21:07 +000077
Misha Brukman94082392003-10-31 18:37:20 +000078</div>
Chris Lattner8b6be362003-05-21 22:21:07 +000079
80<!-- *********************************************************************** -->
Misha Brukman94082392003-10-31 18:37:20 +000081<div class="doc_section">
82 <a name="crashers">Crashing Bugs</a>
83</div>
Chris Lattner8b6be362003-05-21 22:21:07 +000084<!-- *********************************************************************** -->
85
Misha Brukman94082392003-10-31 18:37:20 +000086<div class="doc_text">
87
88<p>More often than not, bugs in the compiler cause it to crash - often due to an
Chris Lattner8b6be362003-05-21 22:21:07 +000089assertion failure of some sort. If you are running <tt><b>opt</b></tt> or
90<tt><b>analyze</b></tt> directly, and something crashes, jump to the section on
91<a href="#passes">bugs in LLVM passes</a>. Otherwise, the most important
92piece of the puzzle is to figure out if it is the GCC-based front-end that is
Misha Brukman94082392003-10-31 18:37:20 +000093buggy or if it's one of the LLVM tools that has problems.</p>
Chris Lattner8b6be362003-05-21 22:21:07 +000094
Misha Brukman94082392003-10-31 18:37:20 +000095<p>To figure out which program is crashing (the front-end,
96<tt><b>gccas</b></tt>, or <tt><b>gccld</b></tt>), run the
97<tt><b>llvm-gcc</b></tt> command line as you were when the crash occurred, but
98add a <tt>-v</tt> option to the command line. The compiler will print out a
John Criswell79926912004-03-12 20:42:16 +000099bunch of stuff, and should end with telling you that one of
100<tt><b>cc1</b>/<b>cc1plus</b></tt>, <tt><b>gccas</b></tt>, or
101<tt><b>gccld</b></tt> crashed.</p>
Chris Lattner8b6be362003-05-21 22:21:07 +0000102
103<ul>
Chris Lattner8b6be362003-05-21 22:21:07 +0000104
John Criswell79926912004-03-12 20:42:16 +0000105 <li>If <tt><b>cc1</b></tt> or <tt><b>cc1plus</b></tt> crashed, you found a
106 problem with the front-end.
Misha Brukman94082392003-10-31 18:37:20 +0000107 Jump ahead to the section on <a href="#front-end">front-end bugs</a>.</li>
108
109 <li>If <tt><b>gccas</b></tt> crashed, you found a bug in <a href="#gccas">one
110 of the passes in <tt><b>gccas</b></tt></a>.</li>
111
112 <li>If <tt><b>gccld</b></tt> crashed, you found a bug in <a href="#gccld">one
113 of the passes in <tt><b>gccld</b></tt></a>.</li>
114
115 <li>Otherwise, something really weird happened. Email the list with what you
116 have at this point.</li>
117
118</ul>
119
120</div>
Chris Lattner8b6be362003-05-21 22:21:07 +0000121
122<!-- ======================================================================= -->
Misha Brukman94082392003-10-31 18:37:20 +0000123<div class="doc_subsection">
124 <a name="front-end">Front-end bugs</a>
125</div>
Chris Lattner8b6be362003-05-21 22:21:07 +0000126
Misha Brukman94082392003-10-31 18:37:20 +0000127<div class="doc_text">
128
129<p>If the problem is in the front-end, you should re-run the same
Brian Gaekeeefe9792003-05-23 21:03:50 +0000130<tt>llvm-gcc</tt> command that resulted in the crash, but add the
Misha Brukman94082392003-10-31 18:37:20 +0000131<tt>-save-temps</tt> option. The compiler will crash again, but it will leave
132behind a <tt><i>foo</i>.i</tt> file (containing preprocessed C source code) and
133possibly <tt><i>foo</i>.s</tt> (containing LLVM assembly code), for each
134compiled <tt><i>foo</i>.c</tt> file. Send us the <tt><i>foo</i>.i</tt> file,
135along with a brief description of the error it caused.</p>
136
137</div>
Chris Lattner8b6be362003-05-21 22:21:07 +0000138
139<!-- ======================================================================= -->
Misha Brukman94082392003-10-31 18:37:20 +0000140<div class="doc_subsection">
141 <a name="gccas">GCCAS bugs</a>
142</div>
Chris Lattner8b6be362003-05-21 22:21:07 +0000143
Misha Brukman94082392003-10-31 18:37:20 +0000144<div class="doc_text">
145
146<p>If you find that a bug crashes in the <tt><b>gccas</b></tt> stage of
Brian Gaekeeefe9792003-05-23 21:03:50 +0000147compilation, compile your test-case to a <tt>.s</tt> file with the
Misha Brukman94082392003-10-31 18:37:20 +0000148<tt>-save-temps</tt> option to <tt><b>llvm-gcc</b></tt>. Then run:</p>
Chris Lattner8b6be362003-05-21 22:21:07 +0000149
Misha Brukmanf4e4feb2004-04-15 20:49:32 +0000150<div class="doc_code">
151<p><tt><b>gccas</b> -debug-pass=Arguments &lt; /dev/null -o - &gt; /dev/null
152</tt></p>
153</div>
Chris Lattner8b6be362003-05-21 22:21:07 +0000154
Misha Brukman94082392003-10-31 18:37:20 +0000155<p>... which will print a list of arguments, indicating the list of passes that
Chris Lattner8b6be362003-05-21 22:21:07 +0000156<tt><b>gccas</b></tt> runs. Once you have the input file and the list of
157passes, go to the section on <a href="#passes">debugging bugs in LLVM
Misha Brukman94082392003-10-31 18:37:20 +0000158passes</a>.</p>
Chris Lattner8b6be362003-05-21 22:21:07 +0000159
Misha Brukman94082392003-10-31 18:37:20 +0000160</div>
Chris Lattner8b6be362003-05-21 22:21:07 +0000161
162<!-- ======================================================================= -->
Misha Brukman94082392003-10-31 18:37:20 +0000163<div class="doc_subsection">
164 <a name="gccld">GCCLD bugs</a>
165</div>
Chris Lattner8b6be362003-05-21 22:21:07 +0000166
Misha Brukman94082392003-10-31 18:37:20 +0000167<div class="doc_text">
168
169<p>If you find that a bug crashes in the <tt><b>gccld</b></tt> stage of
Chris Lattner8b6be362003-05-21 22:21:07 +0000170compilation, gather all of the <tt>.o</tt> bytecode files and libraries that are
171being linked together (the "<tt><b>llvm-gcc</b> -v</tt>" output should include
Misha Brukman94082392003-10-31 18:37:20 +0000172the full list of objects linked). Then run:</p>
Chris Lattner8b6be362003-05-21 22:21:07 +0000173
Misha Brukmanf4e4feb2004-04-15 20:49:32 +0000174<div class="doc_code">
175<p><tt><b>llvm-as</b> &lt; /dev/null &gt; null.bc
176 <b>gccld</b> -debug-pass=Arguments null.bc</tt></p>
177</div>
Chris Lattner8b6be362003-05-21 22:21:07 +0000178
Misha Brukman94082392003-10-31 18:37:20 +0000179<p>... which will print a list of arguments, indicating the list of passes that
Chris Lattner8b6be362003-05-21 22:21:07 +0000180<tt><b>gccld</b></tt> runs. Once you have the input files and the list of
181passes, go to the section on <a href="#passes">debugging bugs in LLVM
Misha Brukman94082392003-10-31 18:37:20 +0000182passes</a>.</p>
183
184</div>
Chris Lattner8b6be362003-05-21 22:21:07 +0000185
186<!-- ======================================================================= -->
Misha Brukman94082392003-10-31 18:37:20 +0000187<div class="doc_subsection">
188 <a name="passes">Bugs in LLVM passes</a>
189</div>
Chris Lattner8b6be362003-05-21 22:21:07 +0000190
Misha Brukman94082392003-10-31 18:37:20 +0000191<div class="doc_text">
192
193<p>At this point, you should have some number of LLVM assembly files or bytecode
Chris Lattner8b6be362003-05-21 22:21:07 +0000194files and a list of passes which crash when run on the specified input. In
195order to reduce the list of passes (which is probably large) and the input to
Misha Brukman94082392003-10-31 18:37:20 +0000196something tractable, use the <tt><b>bugpoint</b></tt> tool as follows:</p>
Chris Lattner8b6be362003-05-21 22:21:07 +0000197
Misha Brukmanf4e4feb2004-04-15 20:49:32 +0000198<div class="doc_code">
199<p><tt><b>bugpoint</b> &lt;input files&gt; &lt;list of passes&gt;</tt></p>
200</div>
Chris Lattner8b6be362003-05-21 22:21:07 +0000201
Misha Brukman94082392003-10-31 18:37:20 +0000202<p><tt><b>bugpoint</b></tt> will print a bunch of output as it reduces the
203test-case, but it should eventually print something like this:</p>
Chris Lattner8b6be362003-05-21 22:21:07 +0000204
Misha Brukmane8cad782004-04-19 03:07:29 +0000205<div class="doc_code">
Chris Lattner8b6be362003-05-21 22:21:07 +0000206<pre>
Misha Brukmane8cad782004-04-19 03:07:29 +0000207...
208Emitted bytecode to 'bugpoint-reduced-simplified.bc'
Chris Lattner8b6be362003-05-21 22:21:07 +0000209
Misha Brukmane8cad782004-04-19 03:07:29 +0000210*** You can reproduce the problem with: opt bugpoint-reduced-simplified.bc -licm
Misha Brukman94082392003-10-31 18:37:20 +0000211</pre>
Misha Brukmane8cad782004-04-19 03:07:29 +0000212</div>
Chris Lattner8b6be362003-05-21 22:21:07 +0000213
Misha Brukman94082392003-10-31 18:37:20 +0000214<p>Once you complete this, please send the LLVM bytecode file and the command
215line to reproduce the problem to the llvmbugs mailing list.</p>
Chris Lattner8b6be362003-05-21 22:21:07 +0000216
Misha Brukman94082392003-10-31 18:37:20 +0000217</div>
Chris Lattner8b6be362003-05-21 22:21:07 +0000218
219<!-- *********************************************************************** -->
Misha Brukman94082392003-10-31 18:37:20 +0000220<div class="doc_section">
221 <a name="miscompilations">Miscompilations</a>
222</div>
Chris Lattner8b6be362003-05-21 22:21:07 +0000223<!-- *********************************************************************** -->
224
Misha Brukman94082392003-10-31 18:37:20 +0000225<div class="doc_text">
Chris Lattner8b6be362003-05-21 22:21:07 +0000226
Misha Brukman94082392003-10-31 18:37:20 +0000227<p>A miscompilation occurs when a pass does not correctly transform a program,
228thus producing errors that are only noticed during execution. This is different
229from producing invalid LLVM code (i.e., code not in SSA form, using values
230before defining them, etc.) which the verifier will check for after a pass
231finishes its run.</p>
232
Chris Lattner8436c212004-05-11 19:34:27 +0000233<p>If it looks like the LLVM compiler is miscompiling a program, the very first
Misha Brukman300e74b2004-05-12 00:57:24 +0000234thing to check is to make sure it is not using undefined behavior. In
235particular, check to see if the program <a
236href="http://valgrind.kde.org/">valgrind</a>s clean, passes purify, or some
Chris Lattner8436c212004-05-11 19:34:27 +0000237other memory checker tool. Many of the "LLVM bugs" that we have chased down
238ended up being bugs in the program being compiled, not LLVM.</p>
239
240<p>Once you determine that the program itself is not buggy, you should choose
241which code generator you wish to compile the program with (e.g. C backend, the
242JIT, or LLC) and optionally a series of LLVM passes to run. For example:</p>
Misha Brukmanf1d01fb2003-09-17 18:51:47 +0000243
Misha Brukmanf4e4feb2004-04-15 20:49:32 +0000244<div class="doc_code">
245<p><tt><b>bugpoint</b> -run-cbe [... optimization passes ...] file-to-test.bc
246--args -- [program arguments]</tt></p>
247</div>
Misha Brukmanf1d01fb2003-09-17 18:51:47 +0000248
Misha Brukman94082392003-10-31 18:37:20 +0000249<p><tt>bugpoint</tt> will try to narrow down your list of passes to the one pass
Misha Brukmanf1d01fb2003-09-17 18:51:47 +0000250that causes an error, and simplify the bytecode file as much as it can to assist
251you. It will print a message letting you know how to reproduce the resulting
Misha Brukman94082392003-10-31 18:37:20 +0000252error.</p>
253
254</div>
Misha Brukmanf1d01fb2003-09-17 18:51:47 +0000255
256<!-- *********************************************************************** -->
Misha Brukman94082392003-10-31 18:37:20 +0000257<div class="doc_section">
258 <a name="codegen">Incorrect code generation</a>
259</div>
Misha Brukmanf1d01fb2003-09-17 18:51:47 +0000260<!-- *********************************************************************** -->
261
Misha Brukman94082392003-10-31 18:37:20 +0000262<div class="doc_text">
263
264<p>Similarly to debugging incorrect compilation by mis-behaving passes, you can
Misha Brukmanf1d01fb2003-09-17 18:51:47 +0000265debug incorrect code generation by either LLC or the JIT, using
266<tt>bugpoint</tt>. The process <tt>bugpoint</tt> follows in this case is to try
267to narrow the code down to a function that is miscompiled by one or the other
268method, but since for correctness, the entire program must be run,
269<tt>bugpoint</tt> will compile the code it deems to not be affected with the C
Misha Brukman94082392003-10-31 18:37:20 +0000270Backend, and then link in the shared object it generates.</p>
Misha Brukmanf1d01fb2003-09-17 18:51:47 +0000271
Misha Brukman94082392003-10-31 18:37:20 +0000272<p>To debug the JIT:</p>
273
Misha Brukmanf4e4feb2004-04-15 20:49:32 +0000274<div class="doc_code">
275<p><tt><b>bugpoint</b> -run-jit -output=[correct output file] [bytecode file]
276--args -- [program arguments]</tt></p>
277</div>
Misha Brukmanf1d01fb2003-09-17 18:51:47 +0000278
Misha Brukman94082392003-10-31 18:37:20 +0000279<p>Similarly, to debug the LLC, one would run:</p>
280
Misha Brukmanf4e4feb2004-04-15 20:49:32 +0000281<div class="doc_code">
282<p><tt><b>bugpoint</b> -run-llc -output=[correct output file] [bytecode file]
283--args -- [program arguments]</tt></p>
284</div>
Misha Brukmanf1d01fb2003-09-17 18:51:47 +0000285
Misha Brukman180e8af2004-04-15 21:01:21 +0000286<p><b>Special note:</b> if you are debugging MultiSource or SPEC tests that
287already exist in the <tt>llvm/test</tt> hierarchy, there is an easier way to
288debug the JIT, LLC, and CBE, using the pre-written Makefile targets, which
289will pass the program options specified in the Makefiles:</p>
290
291<div class="doc_code">
292<p><tt>cd llvm/test/../../program<br>
293 make bugpoint-jit</tt></p>
294</div>
295
Misha Brukman94082392003-10-31 18:37:20 +0000296<p>At the end of a successful <tt>bugpoint</tt> run, you will be presented
Misha Brukmanf1d01fb2003-09-17 18:51:47 +0000297with two bytecode files: a <em>safe</em> file which can be compiled with the C
298backend and the <em>test</em> file which either LLC or the JIT
Misha Brukman94082392003-10-31 18:37:20 +0000299mis-codegenerates, and thus causes the error.</p>
Misha Brukmanf1d01fb2003-09-17 18:51:47 +0000300
Misha Brukman94082392003-10-31 18:37:20 +0000301<p>To reproduce the error that <tt>bugpoint</tt> found, it is sufficient to do
302the following:</p>
Misha Brukmanf1d01fb2003-09-17 18:51:47 +0000303
304<ol>
Misha Brukman94082392003-10-31 18:37:20 +0000305
Misha Brukmanf4e4feb2004-04-15 20:49:32 +0000306 <li><p>Regenerate the shared object from the safe bytecode file:</p>
Misha Brukman94082392003-10-31 18:37:20 +0000307
Misha Brukmanf4e4feb2004-04-15 20:49:32 +0000308 <div class="doc_code">
309 <p><tt><b>llc</b> -march=c safe.bc -o safe.c<br>
310 <b>gcc</b> -shared safe.c -o safe.so</tt></p>
311 </div></li>
Misha Brukman94082392003-10-31 18:37:20 +0000312
Misha Brukmanf4e4feb2004-04-15 20:49:32 +0000313 <li><p>If debugging LLC, compile test bytecode native and link with the shared
314 object:</p>
Misha Brukman94082392003-10-31 18:37:20 +0000315
Misha Brukmanf4e4feb2004-04-15 20:49:32 +0000316 <div class="doc_code">
317 <p><tt><b>llc</b> test.bc -o test.s -f<br>
318 <b>gcc</b> test.s safe.so -o test.llc<br>
319 ./test.llc [program options]</tt></p>
320 </div></li>
Misha Brukman94082392003-10-31 18:37:20 +0000321
Misha Brukmanf4e4feb2004-04-15 20:49:32 +0000322 <li><p>If debugging the JIT, load the shared object and supply the test
323 bytecode:</p>
Misha Brukman94082392003-10-31 18:37:20 +0000324
Misha Brukmanf4e4feb2004-04-15 20:49:32 +0000325 <div class="doc_code">
326 <p><tt><b>lli</b> -load=safe.so test.bc [program options]</tt></p>
327 </div></li>
Misha Brukman94082392003-10-31 18:37:20 +0000328
Misha Brukmanf1d01fb2003-09-17 18:51:47 +0000329</ol>
Chris Lattner8b6be362003-05-21 22:21:07 +0000330
Misha Brukman94082392003-10-31 18:37:20 +0000331</div>
332
Chris Lattner8b6be362003-05-21 22:21:07 +0000333<!-- *********************************************************************** -->
Misha Brukman94082392003-10-31 18:37:20 +0000334<hr>
Misha Brukman637cab02004-01-15 19:03:47 +0000335<address>
336 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
337 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
338 <a href="http://validator.w3.org/check/referer"><img
339 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
340
341 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
Misha Brukman94082392003-10-31 18:37:20 +0000342 <a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a>
343 <br>
344 Last modified: $Date$
Misha Brukman637cab02004-01-15 19:03:47 +0000345</address>
Misha Brukman94082392003-10-31 18:37:20 +0000346
347</body>
348</html>