1<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN" 2 "http://www.w3.org/TR/html4/strict.dtd"> 3<html> 4<head> 5 <meta http-equiv="Content-Type" content="text/html; charset=utf-8"> 6 <title>LLVM bugpoint tool: design and usage</title> 7 <link rel="stylesheet" href="llvm.css" type="text/css"> 8</head> 9 10<h1> 11 LLVM bugpoint tool: design and usage 12</h1> 13 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> 24</ul> 25 26<div class="doc_author"> 27<p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a></p> 28</div> 29 30<!-- *********************************************************************** --> 31<h2> 32<a name="desc">Description</a> 33</h2> 34<!-- *********************************************************************** --> 35 36<div> 37 38<p><tt>bugpoint</tt> narrows down the source of problems in LLVM tools and 39passes. It can be used to debug three types of failures: optimizer crashes, 40miscompilations by optimizers, or bad native code generation (including problems 41in the static and JIT compilers). It aims to reduce large test cases to small, 42useful ones. For example, if <tt>opt</tt> crashes while optimizing a 43file, it will identify the optimization (or combination of optimizations) that 44causes the crash, and reduce the file down to a small example which triggers the 45crash.</p> 46 47<p>For detailed case scenarios, such as debugging <tt>opt</tt>, 48<tt>llvm-ld</tt>, or one of the LLVM code generators, see <a 49href="HowToSubmitABug.html">How To Submit a Bug Report document</a>.</p> 50 51</div> 52 53<!-- *********************************************************************** --> 54<h2> 55<a name="design">Design Philosophy</a> 56</h2> 57<!-- *********************************************************************** --> 58 59<div> 60 61<p><tt>bugpoint</tt> is designed to be a useful tool without requiring any 62hooks into the LLVM infrastructure at all. It works with any and all LLVM 63passes and code generators, and does not need to "know" how they work. Because 64of this, it may appear to do stupid things or miss obvious 65simplifications. <tt>bugpoint</tt> is also designed to trade off programmer 66time for computer time in the compiler-debugging process; consequently, it may 67take a long period of (unattended) time to reduce a test case, but we feel it 68is still worth it. Note that <tt>bugpoint</tt> is generally very quick unless 69debugging a miscompilation where each test of the program (which requires 70executing it) takes a long time.</p> 71 72<!-- ======================================================================= --> 73<h3> 74 <a name="autoselect">Automatic Debugger Selection</a> 75</h3> 76 77<div> 78 79<p><tt>bugpoint</tt> reads each <tt>.bc</tt> or <tt>.ll</tt> file specified on 80the command line and links them together into a single module, called the test 81program. If any LLVM passes are specified on the command line, it runs these 82passes on the test program. If any of the passes crash, or if they produce 83malformed output (which causes the verifier to abort), <tt>bugpoint</tt> starts 84the <a href="#crashdebug">crash debugger</a>.</p> 85 86<p>Otherwise, if the <tt>-output</tt> option was not specified, 87<tt>bugpoint</tt> runs the test program with the C backend (which is assumed to 88generate good code) to generate a reference output. Once <tt>bugpoint</tt> has 89a reference output for the test program, it tries executing it with the 90selected code generator. If the selected code generator crashes, 91<tt>bugpoint</tt> starts the <a href="#crashdebug">crash debugger</a> on the 92code generator. Otherwise, if the resulting output differs from the reference 93output, it assumes the difference resulted from a code generator failure, and 94starts the <a href="#codegendebug">code generator debugger</a>.</p> 95 96<p>Finally, if the output of the selected code generator matches the reference 97output, <tt>bugpoint</tt> runs the test program after all of the LLVM passes 98have been applied to it. If its output differs from the reference output, it 99assumes the difference resulted from a failure in one of the LLVM passes, and 100enters the <a href="#miscompilationdebug">miscompilation debugger</a>. 101Otherwise, there is no problem <tt>bugpoint</tt> can debug.</p> 102 103</div> 104 105<!-- ======================================================================= --> 106<h3> 107 <a name="crashdebug">Crash debugger</a> 108</h3> 109 110<div> 111 112<p>If an optimizer or code generator crashes, <tt>bugpoint</tt> will try as hard 113as it can to reduce the list of passes (for optimizer crashes) and the size of 114the test program. First, <tt>bugpoint</tt> figures out which combination of 115optimizer passes triggers the bug. This is useful when debugging a problem 116exposed by <tt>opt</tt>, for example, because it runs over 38 passes.</p> 117 118<p>Next, <tt>bugpoint</tt> tries removing functions from the test program, to 119reduce its size. Usually it is able to reduce a test program to a single 120function, when debugging intraprocedural optimizations. Once the number of 121functions has been reduced, it attempts to delete various edges in the control 122flow graph, to reduce the size of the function as much as possible. Finally, 123<tt>bugpoint</tt> deletes any individual LLVM instructions whose absence does 124not eliminate the failure. At the end, <tt>bugpoint</tt> should tell you what 125passes crash, give you a bitcode file, and give you instructions on how to 126reproduce the failure with <tt>opt</tt> or <tt>llc</tt>.</p> 127 128</div> 129 130<!-- ======================================================================= --> 131<h3> 132 <a name="codegendebug">Code generator debugger</a> 133</h3> 134 135<div> 136 137<p>The code generator debugger attempts to narrow down the amount of code that 138is being miscompiled by the selected code generator. To do this, it takes the 139test program and partitions it into two pieces: one piece which it compiles 140with the C backend (into a shared object), and one piece which it runs with 141either the JIT or the static LLC compiler. It uses several techniques to 142reduce the amount of code pushed through the LLVM code generator, to reduce the 143potential scope of the problem. After it is finished, it emits two bitcode 144files (called "test" [to be compiled with the code generator] and "safe" [to be 145compiled with the C backend], respectively), and instructions for reproducing 146the problem. The code generator debugger assumes that the C backend produces 147good code.</p> 148 149</div> 150 151<!-- ======================================================================= --> 152<h3> 153 <a name="miscompilationdebug">Miscompilation debugger</a> 154</h3> 155 156<div> 157 158<p>The miscompilation debugger works similarly to the code generator debugger. 159It works by splitting the test program into two pieces, running the 160optimizations specified on one piece, linking the two pieces back together, and 161then executing the result. It attempts to narrow down the list of passes to 162the one (or few) which are causing the miscompilation, then reduce the portion 163of the test program which is being miscompiled. The miscompilation debugger 164assumes that the selected code generator is working properly.</p> 165 166</div> 167 168</div> 169 170<!-- *********************************************************************** --> 171<h2> 172 <a name="advice">Advice for using bugpoint</a> 173</h2> 174<!-- *********************************************************************** --> 175 176<div> 177 178<tt>bugpoint</tt> can be a remarkably useful tool, but it sometimes works in 179non-obvious ways. Here are some hints and tips:<p> 180 181<ol> 182<li>In the code generator and miscompilation debuggers, <tt>bugpoint</tt> only 183 works with programs that have deterministic output. Thus, if the program 184 outputs <tt>argv[0]</tt>, the date, time, or any other "random" data, 185 <tt>bugpoint</tt> may misinterpret differences in these data, when output, 186 as the result of a miscompilation. Programs should be temporarily modified 187 to disable outputs that are likely to vary from run to run. 188 189<li>In the code generator and miscompilation debuggers, debugging will go 190 faster if you manually modify the program or its inputs to reduce the 191 runtime, but still exhibit the problem. 192 193<li><tt>bugpoint</tt> is extremely useful when working on a new optimization: 194 it helps track down regressions quickly. To avoid having to relink 195 <tt>bugpoint</tt> every time you change your optimization however, have 196 <tt>bugpoint</tt> dynamically load your optimization with the 197 <tt>-load</tt> option. 198 199<li><p><tt>bugpoint</tt> can generate a lot of output and run for a long period 200 of time. It is often useful to capture the output of the program to file. 201 For example, in the C shell, you can run:</p> 202 203<div class="doc_code"> 204<p><tt>bugpoint ... |& tee bugpoint.log</tt></p> 205</div> 206 207 <p>to get a copy of <tt>bugpoint</tt>'s output in the file 208 <tt>bugpoint.log</tt>, as well as on your terminal.</p> 209 210<li><tt>bugpoint</tt> cannot debug problems with the LLVM linker. If 211 <tt>bugpoint</tt> crashes before you see its "All input ok" message, 212 you might try <tt>llvm-link -v</tt> on the same set of input files. If 213 that also crashes, you may be experiencing a linker bug. 214 215<li><tt>bugpoint</tt> is useful for proactively finding bugs in LLVM. 216 Invoking <tt>bugpoint</tt> with the <tt>-find-bugs</tt> option will cause 217 the list of specified optimizations to be randomized and applied to the 218 program. This process will repeat until a bug is found or the user 219 kills <tt>bugpoint</tt>. 220</ol> 221 222</div> 223 224<!-- *********************************************************************** --> 225 226<hr> 227<address> 228 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img 229 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a> 230 <a href="http://validator.w3.org/check/referer"><img 231 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a> 232 233 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br> 234 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br> 235 Last modified: $Date$ 236</address> 237 238</body> 239</html> 240