1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN" "http://www.w3.org/TR/html4/strict.dtd">
4 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
5 <title>LLVM: bugpoint tool</title>
6 <link rel="stylesheet" href="llvm.css" type="text/css">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="A tool for automatic test case reduction">
14 <center><h1>LLVM: <tt>bugpoint</tt> tool</h1></center>
21 <tt>bugpoint [options] [input LLVM ll/bc files] [LLVM passes] --args <program arguments>...</tt>
23 <img src="img/Debugging.gif" width=444 height=314 align=right>
26 The <tt>bugpoint</tt> tool narrows down the source of
27 problems in LLVM tools and passes. It can be used to debug three types of
28 failures: optimizer crashes, miscompilations by optimizers, or bad native
29 code generation (including problems in the static and JIT compilers). It aims
30 to reduce large test cases to small, useful ones. For example,
31 if <tt><a href="CommandGuide/html/gccas.html">gccas</a></tt> crashes while optimizing a file, it
32 will identify the optimization (or combination of optimizations) that causes the
33 crash, and reduce the file down to a small example which triggers the crash.<p>
35 <a name="designphilosophy">
36 <h4>Design Philosophy</h4>
38 <tt>bugpoint</tt> is designed to be a useful tool without requiring any
39 hooks into the LLVM infrastructure at all. It works with any and all LLVM
40 passes and code generators, and does not need to "know" how they work. Because
41 of this, it may appear to do stupid things or miss obvious
42 simplifications. <tt>bugpoint</tt> is also designed to trade off programmer
43 time for computer time in the compiler-debugging process; consequently, it may
44 take a long period of (unattended) time to reduce a test case, but we feel it
45 is still worth it. Note that <tt>bugpoint</tt> is generally very quick unless
46 debugging a miscompilation where each test of the program (which requires
47 executing it) takes a long time.<p>
49 <a name="automaticdebuggerselection">
50 <h4>Automatic Debugger Selection</h4>
52 <tt>bugpoint</tt> reads each <tt>.bc</tt> or <tt>.ll</tt> file
53 specified on the command line and links them together into a single module,
54 called the test program. If any LLVM passes are
55 specified on the command line, it runs these passes on the test program. If
56 any of the passes crash, or if they produce malformed output (which causes the
58 <tt>bugpoint</tt> starts the <a href="#crashdebug">crash debugger</a>.<p>
60 Otherwise, if the <a href="#opt_output"><tt>-output</tt></a> option was not
61 specified, <tt>bugpoint</tt> runs the test program with the C backend (which is
62 assumed to generate good code) to generate a reference output. Once
63 <tt>bugpoint</tt> has a reference output for the test program, it tries
64 executing it with the <a href="#opt_run-">selected</a> code generator. If the
65 selected code generator crashes, <tt>bugpoint</tt> starts the <a
66 href="#crashdebug">crash debugger</a> on the code generator. Otherwise, if the
67 resulting output differs from the reference output, it assumes the difference
68 resulted from a code generator failure, and starts the <a
69 href="#codegendebug">code generator debugger</a>.<p>
71 Finally, if the output of the selected code generator matches the reference
72 output, <tt>bugpoint</tt> runs the test program after all of the LLVM passes
73 have been applied to it. If its output differs from the reference output, it
74 assumes the difference resulted from a failure in one of the LLVM passes, and
75 enters the <a href="#miscompilationdebug">miscompilation
76 debugger</a>. Otherwise, there is no problem <tt>bugpoint</tt> can debug.<p>
79 <h4>Crash debugger</h4>
81 If an optimizer or code generator crashes, <tt>bugpoint</tt> will try as hard as
82 it can to reduce the list of passes (for optimizer crashes) and the size of the
83 test program. First, <tt>bugpoint</tt> figures out which combination of
84 optimizer passes triggers the bug. This is useful when debugging a problem
85 exposed by <tt>gccas</tt>, for example, because it runs over 38 passes.<p>
87 Next, <tt>bugpoint</tt> tries removing functions from the test program, to
88 reduce its size. Usually it is able to reduce a test program to a single
89 function, when debugging intraprocedural optimizations. Once the number of
90 functions has been reduced, it attempts to delete various edges in the control
91 flow graph, to reduce the size of the function as much as possible. Finally,
92 <tt>bugpoint</tt> deletes any individual LLVM instructions whose absence does
93 not eliminate the failure. At the end, <tt>bugpoint</tt> should tell you what
94 passes crash, give you a bytecode file, and give you instructions on how to
95 reproduce the failure with <tt><a href="CommandGuide/html/opt.html">opt</a></tt>, <tt><a
96 href="CommandGuide/html/analyze.html">analyze</a></tt>, or <tt><a href="CommandGuide/html/llc.html">llc</a></tt>.<p>
98 <a name="codegendebug">
99 <h4>Code generator debugger</h4>
101 <p>The code generator debugger attempts to narrow down the amount of code that
102 is being miscompiled by the <a href="#opt_run-">selected</a> code generator. To
103 do this, it takes the test program and partitions it into two pieces: one piece
104 which it compiles with the C backend (into a shared object), and one piece which
105 it runs with either the JIT or the static LLC compiler. It uses several
106 techniques to reduce the amount of code pushed through the LLVM code generator,
107 to reduce the potential scope of the problem. After it is finished, it emits
108 two bytecode files (called "test" [to be compiled with the code generator] and
109 "safe" [to be compiled with the C backend], respectively), and instructions for
110 reproducing the problem. The code generator debugger assumes that the C backend
111 produces good code.</p>
113 <a name="miscompilationdebug">
114 <h4>Miscompilation debugger</h4>
116 The miscompilation debugger works similarly to the code generator
117 debugger. It works by splitting the test program into two pieces, running the
118 optimizations specified on one piece, linking the two pieces back together,
119 and then executing the result.
120 It attempts to narrow down the list of passes to the one (or few) which are
121 causing the miscompilation, then reduce the portion of the test program which is
122 being miscompiled. The miscompilation debugger assumes that the selected
123 code generator is working properly.<p>
125 <a name="bugpoint notes">
126 <h4>Advice for using <tt>bugpoint</tt></h4>
128 <tt>bugpoint</tt> can be a remarkably useful tool, but it sometimes works in
129 non-obvious ways. Here are some hints and tips:<p>
132 <li>In the code generator and miscompilation debuggers, <tt>bugpoint</tt> only
133 works with programs that have deterministic output. Thus, if the program
134 outputs <tt>argv[0]</tt>, the date, time, or any other "random" data, <tt>bugpoint</tt> may
135 misinterpret differences in these data, when output, as the result of a
136 miscompilation. Programs should be temporarily modified to disable
137 outputs that are likely to vary from run to run.
139 <li>In the code generator and miscompilation debuggers, debugging will go
140 faster if you manually modify the program or its inputs to reduce the
141 runtime, but still exhibit the problem.
143 <li><tt>bugpoint</tt> is extremely useful when working on a new optimization:
144 it helps track down regressions quickly. To avoid having to relink
145 <tt>bugpoint</tt> every time you change your optimization however, have
146 <tt>bugpoint</tt> dynamically load your optimization with the <a
147 href="#opt_load"><tt>-load</tt></a> option.
149 <li><tt>bugpoint</tt> can generate a lot of output and run for a long period of
150 time. It is often useful to capture the output of the program to file. For
151 example, in the C shell, you can type:<br>
152 <tt>bugpoint ..... |& tee bugpoint.log</tt>
153 <br>to get a copy of <tt>bugpoint</tt>'s output in the file
154 <tt>bugpoint.log</tt>, as well as on your terminal.
156 <li><tt>bugpoint</tt> cannot debug problems with the LLVM linker. If
157 <tt>bugpoint</tt> crashes before you see its "All input ok" message,
158 you might try <tt>llvm-link -v</tt> on the same set of input files. If
159 that also crashes, you may be experiencing a linker bug.
161 <li>If your program is <b>supposed</b> to crash, <tt>bugpoint</tt> will be
162 confused. One way to deal with this is to cause bugpoint to ignore the exit
163 code from your program, by giving it the <tt>-check-exit-code=false</tt>
171 <li><tt>-additional-so <library></tt><br>
172 Load <tt><library></tt> into the test program whenever it is run.
173 This is useful if you are debugging programs which depend on non-LLVM
174 libraries (such as the X or curses libraries) to run.<p>
176 <li><tt>-args <program args></tt><br>
177 Pass all arguments specified after <tt>-args</tt> to the
178 test program whenever it runs. Note that if any of
179 the <tt><program args></tt> start with a '-', you should use:
181 <tt>bugpoint <bugpoint args> -args -- <program args></tt>
183 The "<tt>--</tt>" right after the <tt>-args</tt> option tells
184 <tt>bugpoint</tt> to consider any options starting with <tt>-</tt> to be
185 part of the <tt>-args</tt> option, not as options to <tt>bugpoint</tt>
188 <li><tt>-tool-args <tool args></tt><br>
189 Pass all arguments specified after <tt>-tool-args</tt> to the
190 LLVM tool under test (llc, lli, etc.) whenever it runs.
191 You should use this option in the following way:
193 <tt>bugpoint <bugpoint args> -tool-args -- <tool args></tt>
195 The "<tt>--</tt>" right after the <tt>-tool-args</tt> option tells
196 <tt>bugpoint</tt> to consider any options starting with <tt>-</tt> to be
197 part of the <tt>-tool-args</tt> option, not as options to
198 <tt>bugpoint</tt> itself. (See <tt>-args</tt>, above.)<p>
200 <li><tt>-check-exit-code={true,false}</tt><br>
201 Assume a non-zero exit code or core dump from the test program is
202 a failure. Defaults to true.<p>
204 <li><tt>-disable-{dce,simplifycfg}</tt><br>
205 Do not run the specified passes to clean up and reduce the size of the
206 test program. By default, <tt>bugpoint</tt> uses these passes internally
207 when attempting to reduce test programs. If you're trying to find
208 a bug in one of these passes, <tt>bugpoint</tt> may crash.<p>
210 <li> <tt>-help</tt><br>
211 Print a summary of command line options.<p>
213 <a name="opt_input"><li><tt>-input <filename></tt><br>
214 Open <tt><filename></tt> and redirect the standard input of the
215 test program, whenever it runs, to come from that file.
218 <a name="opt_load"><li> <tt>-load <plugin></tt><br>
219 Load the dynamic object <tt><plugin></tt> into <tt>bugpoint</tt>
220 itself. This object should register new
221 optimization passes. Once loaded, the object will add new command line
222 options to enable various optimizations. To see the new complete list
223 of optimizations, use the -help and -load options together:
225 <tt>bugpoint -load <plugin> -help</tt>
228 <a name="opt_output"><li><tt>-output <filename></tt><br>
229 Whenever the test program produces output on its standard output
230 stream, it should match the contents of <tt><filename></tt>
231 (the "reference output"). If you do not use this option,
232 <tt>bugpoint</tt> will attempt to generate a reference output by
233 compiling the program with the C backend and running it.<p>
235 <li><tt>-profile-info-file <filename></tt><br>
236 Profile file loaded by -profile-loader.<p>
238 <a name="opt_run-"><li><tt>-run-{int,jit,llc,cbe}</tt><br>
239 Whenever the test program is compiled, <tt>bugpoint</tt> should generate
240 code for it using the specified code generator. These options allow
241 you to choose the interpreter, the JIT compiler, the static native
242 code compiler, or the C backend, respectively.<p>
247 If <tt>bugpoint</tt> succeeds in finding a problem, it will exit with 0.
248 Otherwise, if an error occurs, it will exit with a non-zero value.
251 Maintained by the <a href="http://llvm.cs.uiuc.edu">LLVM Team</a>.