2 <title>LLVM: bugpoint tool</title>
6 <center><h1>LLVM: <tt>bugpoint</tt> tool</h1></center>
13 <tt>bugpoint [options] [input LLVM ll/bc files] [LLVM passes] --args <program arguments>...</tt>
15 <img src="../Debugging.gif" width=444 height=314 align=right>
18 The <tt>bugpoint</tt> tool narrows down the source of
19 problems in LLVM tools and passes. It can be used to debug three types of
20 failures: optimizer crashes, miscompilations by optimizers, or invalid native
21 code generation. It aims to reduce large test cases to small, useful ones.
23 if <tt><a href="gccas.html">gccas</a></tt> crashes while optimizing a file, it
24 will identify the optimization (or combination of optimizations) that causes the
25 crash, and reduce the file down to a small example which triggers the crash.<p>
27 <a name="designphilosophy">
28 <h4>Design Philosophy</h4>
30 <tt>bugpoint</tt> is designed to be a useful tool without requiring any
31 hooks into the LLVM infrastructure at all. It works with any and all LLVM
32 passes and code generators, and does not need to "know" how they work. Because
33 of this, it may appear to do a lot of stupid things or miss obvious
34 simplifications. <tt>bugpoint</tt> is also designed to trade off programmer
35 time for computer time in the compiler-debugging process; consequently, it may
36 take a long period of (unattended) time to reduce a test case, but we feel it
37 is still worth it. :-) <p>
39 <a name="automaticdebuggerselection">
40 <h4>Automatic Debugger Selection</h4>
42 <tt>bugpoint</tt> reads each <tt>.bc</tt> or <tt>.ll</tt> file
43 specified on the command line and links them together into a single module,
44 called the test program. If any LLVM passes are
45 specified on the command line, it runs these passes on the test program. If
46 any of the passes crash, or if they produce malformed output,
47 <tt>bugpoint</tt> starts the <a href="#crashdebug">crash debugger</a>.<p>
49 Otherwise, if the <a href="#opt_output"><tt>-output</tt></a> option was not
50 specified, <tt>bugpoint</tt> runs the test program with the C backend (which
51 is assumed to generate good code) to generate a reference output. Once
52 <tt>bugpoint</tt> has a reference output for the test program, it tries
54 with the <a href="#opt_run-">selected</a> code generator. If
55 the resulting output differs from the reference output, it assumes the
56 difference resulted from a code generator failure, and starts the
57 <a href="#codegendebug">code generator debugger</a>.<p>
59 Otherwise, <tt>bugpoint</tt> runs the test program after all of the LLVM passes
60 have been applied to it. If its output differs from the reference output,
61 it assumes the difference resulted from a failure in one of the LLVM passes,
63 <a href="#miscompilationdebug">miscompilation debugger</a>. Otherwise,
64 there is no problem <tt>bugpoint</tt> can debug.<p>
67 <h4>Crash debugger</h4>
69 If an optimizer crashes, <tt>bugpoint</tt> will try as hard as it can to
70 reduce the list of passes and the size of the test program. First,
71 <tt>bugpoint</tt> figures out which combination of passes triggers the bug. This
72 is useful when debugging a problem exposed by <tt>gccas</tt>, for example,
73 because it runs over 25 optimizations.<p>
75 Next, <tt>bugpoint</tt> tries removing functions from the test program, to
77 size. Usually it is able to reduce a test program
78 to a single function, when debugging intraprocedural optimizations. Once the
80 functions has been reduced, it attempts to delete various edges in the control
81 flow graph, to reduce the size of the function as much as possible. Finally,
82 <tt>bugpoint</tt> deletes any individual LLVM instructions whose absence does
83 not eliminate the failure. At the end, <tt>bugpoint</tt> should tell you what
84 passes crash, give you a bytecode file, and give you instructions on how to
85 reproduce the failure with <tt><a href="opt.html">opt</a></tt> or
86 <tt><a href="analyze.html">analyze</a></tt>.<p>
88 <a name="codegendebug">
89 <h4>Code generator debugger</h4>
91 The code generator debugger attempts to narrow down the amount of code that is
92 being miscompiled by the <a href="#opt_run-">selected</a> code generator. To do
93 this, it takes the test program and partitions it into two pieces: one piece
94 which it compiles with the C backend (into a shared object), and one piece which
95 it runs with either the JIT or the static LLC compiler. It uses several
96 techniques to reduce the amount of code pushed through the LLVM code generator,
97 to reduce the potential scope of the problem. After it is finished, it emits
98 two bytecode files (called "test" [to be compiled with the code generator] and
99 "safe" [to be compiled with the C backend] respectively), and instructions for
100 reproducing the problem. The code generator debugger assumes that the C
101 backend produces good code.<p>
103 If you are using the code generator debugger and get an error message that
104 says "Non-instruction
105 is using an external function!", try using the <tt>-run-llc</tt> option instead
106 of the <tt>-run-jit</tt> option. This is due to an unimplemented feature in the
107 code generator debugger.<p>
109 <a name="miscompilationdebug">
110 <h4>Miscompilation debugger</h4>
112 The miscompilation debugger works similarly to the code generator
113 debugger. It works by splitting the test program into two pieces, running the
114 optimizations specified on one piece, linking the two pieces back together,
115 and then executing the result.
116 It attempts to narrow down the list of passes to the one (or few) which are
117 causing the miscompilation, then reduce the portion of the test program which is
118 being miscompiled. The miscompilation debugger assumes that the selected
119 code generator is working properly.<p>
121 <a name="bugpoint notes">
122 <h4>Advice for using <tt>bugpoint</tt></h4>
124 <tt>bugpoint</tt> can be a remarkably useful tool, but it sometimes works in
125 non-obvious ways. Here are some hints and tips:<p>
128 <li>In the code generator and miscompilation debuggers, <tt>bugpoint</tt> only
129 works with programs that have deterministic output. Thus, if the program
130 outputs the date, time, or any other "random" data, <tt>bugpoint</tt> may
131 misinterpret differences in these data, when output, as the result of a
132 miscompilation. Programs should be temporarily modified to disable
133 outputs that are likely to vary from run to run.
135 <li>In the code generator and miscompilation debuggers, debugging will go
136 faster if you manually modify the program or its inputs to reduce the
137 runtime, but still exhibit the problem.
139 <li><tt>bugpoint</tt> is extremely useful when working on a new optimization:
140 it helps track down regressions quickly. To avoid having to relink
141 <tt>bugpoint</tt> every time you change your optimization however, have
142 <tt>bugpoint</tt> dynamically load your optimization with the <a
143 href="#opt_load"><tt>-load</tt></a> option.
145 <li><tt>bugpoint</tt> can generate a lot of output and run for a long period of
146 time. It is often useful to capture the output of the program to file. For
147 example, in the C shell, you can type:<br>
148 <tt>bugpoint ..... |& tee bugpoint.log</tt>
149 <br>to get a copy of <tt>bugpoint</tt>'s output in the file
150 <tt>bugpoint.log</tt>, as well as on your terminal.
152 <li><tt>bugpoint</tt> cannot debug problems with the linker. If
153 <tt>bugpoint</tt> crashes before you see its "All input ok" message,
154 you might try <tt>llvm-link -v</tt> on the same set of input files. If
155 that also crashes, you may be experiencing a linker bug.
162 <li><tt>-additional-so <library.so></tt><br>
163 Load <tt><library.so></tt> into the test program whenever it is run.
164 This is useful if you are debugging programs which depend on non-LLVM
165 libraries (such as the X or curses libraries) to run.<p>
167 <li><tt>-args <program args></tt><br>
168 Pass all arguments specified after <tt>-args</tt> to the
169 test program whenever it runs. Note that if any of
170 the <tt><program args></tt> start with a '-', you should use:
172 <tt>bugpoint <bugpoint args> -args -- <program args></tt>
174 The "<tt>--</tt>" right after the <tt>-args</tt> option tells
175 <tt>bugpoint</tt> to consider any options starting with <tt>-</tt> to be
176 part of the <tt>-args</tt> option, not as options to <tt>bugpoint</tt>
179 <li><tt>-disable-{adce,dce,final-cleanup,simplifycfg}</tt><br>
180 Do not run the specified passes to clean up and reduce the size of the
181 test program. By default, <tt>bugpoint</tt> uses these passes internally
182 when attempting to reduce test programs. If you're trying to find
183 a bug in one of these passes, <tt>bugpoint</tt> may crash.<p>
185 <li> <tt>-help</tt><br>
186 Print a summary of command line options.<p>
188 <a name="opt_input"><li><tt>-input <filename></tt><br>
189 Open <tt><filename></tt> and redirect the standard input of the
190 test program, whenever it runs, to come from that file.
193 <a name="opt_load"><li> <tt>-load <plugin.so></tt><br>
194 Load the dynamic object <tt><plugin.so></tt> into <tt>bugpoint</tt>
195 itself. This object should register new
196 optimization passes. Once loaded, the object will add new command line
197 options to enable various optimizations. To see the new complete list
198 of optimizations, use the -help and -load options together:
200 <tt>bugpoint -load <plugin.so> -help</tt>
203 <a name="opt_output"><li><tt>-output <filename></tt><br>
204 Whenever the test program produces output on its standard output
205 stream, it should match the contents of <tt><filename></tt>
206 (the "reference output"). If you do not use this option,
207 <tt>bugpoint</tt> will attempt to generate a reference output by
208 compiling the program with the C backend and running it.<p>
210 <a name="opt_run-"><li><tt>-run-{int,jit,llc,cbe}</tt><br>
211 Whenever the test program is compiled, <tt>bugpoint</tt> should generate
212 code for it using the specified code generator. These options allow
213 you to choose the interpreter, the JIT compiler, the static native
214 code compiler, or the C backend, respectively.<p>
219 If <tt>bugpoint</tt> succeeds in finding a problem, it will exit with 0.
220 Otherwise, if an error occurs, it will exit with a non-zero value.
223 <a href="opt.html"><tt>opt</tt></a>,
224 <a href="analyze.html"><tt>analyze</tt></a>
227 Maintained by the <a href="http://llvm.cs.uiuc.edu">LLVM Team</a>.