Gitiles
Code Review Sign In
gerrit-public.fairphone.software / fp2-dev / platform / external / llvm / 3c00f8936502465d080aa1a79a3b51d0fb89820c / . / docs / WritingAnLLVMPass.html
blob: 7dbaa9e6bdadad0799ec565b0dd13d6e16fcdfe4 [file] [log] [blame]
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]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>Writing an LLVM Pass</title>
7 <link rel="stylesheet" href="llvm.css" type="text/css">
8</head>
9<body>
10
11<div class="doc_title">
12 Writing an LLVM Pass
13</div>
14
15<ol>
16 <li><a href="#introduction">Introduction - What is a pass?</a></li>
17 <li><a href="#quickstart">Quick Start - Writing hello world</a>
18 <ul>
19 <li><a href="#makefile">Setting up the build environment</a></li>
20 <li><a href="#basiccode">Basic code required</a></li>
21 <li><a href="#running">Running a pass with <tt>opt</tt></a></li>
22 </ul></li>
23 <li><a href="#passtype">Pass classes and requirements</a>
24 <ul>
25 <li><a href="#ImmutablePass">The <tt>ImmutablePass</tt> class</a></li>
26 <li><a href="#ModulePass">The <tt>ModulePass</tt> class</a>
27 <ul>
28 <li><a href="#runOnModule">The <tt>runOnModule</tt> method</a></li>
29 </ul></li>
30 <li><a href="#CallGraphSCCPass">The <tt>CallGraphSCCPass</tt> class</a>
31 <ul>
32 <li><a href="#doInitialization_scc">The <tt>doInitialization(CallGraph
33 &amp;)</tt> method</a></li>
34 <li><a href="#runOnSCC">The <tt>runOnSCC</tt> method</a></li>
35 <li><a href="#doFinalization_scc">The <tt>doFinalization(CallGraph
36 &amp;)</tt> method</a></li>
37 </ul></li>
38 <li><a href="#FunctionPass">The <tt>FunctionPass</tt> class</a>
39 <ul>
40 <li><a href="#doInitialization_mod">The <tt>doInitialization(Module
41 &amp;)</tt> method</a></li>
42 <li><a href="#runOnFunction">The <tt>runOnFunction</tt> method</a></li>
43 <li><a href="#doFinalization_mod">The <tt>doFinalization(Module
44 &amp;)</tt> method</a></li>
45 </ul></li>
46 <li><a href="#LoopPass">The <tt>LoopPass</tt> class</a>
47 <ul>
48 <li><a href="#doInitialization_loop">The <tt>doInitialization(Loop *,
49 LPPassManager &amp;)</tt> method</a></li>
50 <li><a href="#runOnLoop">The <tt>runOnLoop</tt> method</a></li>
51 <li><a href="#doFinalization_loop">The <tt>doFinalization()
52 </tt> method</a></li>
53 </ul></li>
54 <li><a href="#BasicBlockPass">The <tt>BasicBlockPass</tt> class</a>
55 <ul>
56 <li><a href="#doInitialization_fn">The <tt>doInitialization(Function
57 &amp;)</tt> method</a></li>
58 <li><a href="#runOnBasicBlock">The <tt>runOnBasicBlock</tt>
59 method</a></li>
60 <li><a href="#doFinalization_fn">The <tt>doFinalization(Function
61 &amp;)</tt> method</a></li>
62 </ul></li>
63 <li><a href="#MachineFunctionPass">The <tt>MachineFunctionPass</tt>
64 class</a>
65 <ul>
66 <li><a href="#runOnMachineFunction">The
67 <tt>runOnMachineFunction(MachineFunction &amp;)</tt> method</a></li>
68 </ul></li>
69 </ul>
70 <li><a href="#registration">Pass Registration</a>
71 <ul>
72 <li><a href="#print">The <tt>print</tt> method</a></li>
73 </ul></li>
74 <li><a href="#interaction">Specifying interactions between passes</a>
75 <ul>
76 <li><a href="#getAnalysisUsage">The <tt>getAnalysisUsage</tt>
77 method</a></li>
78 <li><a href="#AU::addRequired">The <tt>AnalysisUsage::addRequired&lt;&gt;</tt> and <tt>AnalysisUsage::addRequiredTransitive&lt;&gt;</tt> methods</a></li>
79 <li><a href="#AU::addPreserved">The <tt>AnalysisUsage::addPreserved&lt;&gt;</tt> method</a></li>
80 <li><a href="#AU::examples">Example implementations of <tt>getAnalysisUsage</tt></a></li>
Duncan Sands4e0d6a72009-01-28 13:14:17 +0000[diff] [blame]81 <li><a href="#getAnalysis">The <tt>getAnalysis&lt;&gt;</tt> and
82<tt>getAnalysisIfAvailable&lt;&gt;</tt> methods</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]83 </ul></li>
84 <li><a href="#analysisgroup">Implementing Analysis Groups</a>
85 <ul>
86 <li><a href="#agconcepts">Analysis Group Concepts</a></li>
87 <li><a href="#registerag">Using <tt>RegisterAnalysisGroup</tt></a></li>
88 </ul></li>
89 <li><a href="#passStatistics">Pass Statistics</a>
90 <li><a href="#passmanager">What PassManager does</a>
91 <ul>
92 <li><a href="#releaseMemory">The <tt>releaseMemory</tt> method</a></li>
93 </ul></li>
94 <li><a href="#registering">Registering dynamically loaded passes</a>
95 <ul>
96 <li><a href="#registering_existing">Using existing registries</a></li>
97 <li><a href="#registering_new">Creating new registries</a></li>
98 </ul></li>
99 <li><a href="#debughints">Using GDB with dynamically loaded passes</a>
100 <ul>
101 <li><a href="#breakpoint">Setting a breakpoint in your pass</a></li>
102 <li><a href="#debugmisc">Miscellaneous Problems</a></li>
103 </ul></li>
104 <li><a href="#future">Future extensions planned</a>
105 <ul>
106 <li><a href="#SMP">Multithreaded LLVM</a></li>
107 </ul></li>
108</ol>
109
110<div class="doc_author">
111 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a> and
112 <a href="mailto:jlaskey@mac.com">Jim Laskey</a></p>
113</div>
114
115<!-- *********************************************************************** -->
116<div class="doc_section">
117 <a name="introduction">Introduction - What is a pass?</a>
118</div>
119<!-- *********************************************************************** -->
120
121<div class="doc_text">
122
123<p>The LLVM Pass Framework is an important part of the LLVM system, because LLVM
124passes are where most of the interesting parts of the compiler exist. Passes
125perform the transformations and optimizations that make up the compiler, they
126build the analysis results that are used by these transformations, and they are,
127above all, a structuring technique for compiler code.</p>
128
129<p>All LLVM passes are subclasses of the <tt><a
130href="http://llvm.org/doxygen/classllvm_1_1Pass.html">Pass</a></tt>
131class, which implement functionality by overriding virtual methods inherited
132from <tt>Pass</tt>. Depending on how your pass works, you should inherit from
133the <tt><a href="#ModulePass">ModulePass</a></tt>, <tt><a
134href="#CallGraphSCCPass">CallGraphSCCPass</a></tt>, <tt><a
135href="#FunctionPass">FunctionPass</a></tt>, or <tt><a
136href="#LoopPass">LoopPass</a></tt>, or <tt><a
137href="#BasicBlockPass">BasicBlockPass</a></tt> classes, which gives the system
138more information about what your pass does, and how it can be combined with
139other passes. One of the main features of the LLVM Pass Framework is that it
140schedules passes to run in an efficient way based on the constraints that your
141pass meets (which are indicated by which class they derive from).</p>
142
143<p>We start by showing you how to construct a pass, everything from setting up
144the code, to compiling, loading, and executing it. After the basics are down,
145more advanced features are discussed.</p>
146
147</div>
148
149<!-- *********************************************************************** -->
150<div class="doc_section">
151 <a name="quickstart">Quick Start - Writing hello world</a>
152</div>
153<!-- *********************************************************************** -->
154
155<div class="doc_text">
156
157<p>Here we describe how to write the "hello world" of passes. The "Hello" pass
158is designed to simply print out the name of non-external functions that exist in
159the program being compiled. It does not modify the program at all, it just
160inspects it. The source code and files for this pass are available in the LLVM
161source tree in the <tt>lib/Transforms/Hello</tt> directory.</p>
162
163</div>
164
165<!-- ======================================================================= -->
166<div class="doc_subsection">
167 <a name="makefile">Setting up the build environment</a>
168</div>
169
170<div class="doc_text">
171
172 <p>First, you need to create a new directory somewhere in the LLVM source
173 base. For this example, we'll assume that you made
174 <tt>lib/Transforms/Hello</tt>. Next, you must set up a build script
175 (Makefile) that will compile the source code for the new pass. To do this,
176 copy the following into <tt>Makefile</tt>:</p>
177 <hr/>
178
179<div class="doc_code"><pre>
180# Makefile for hello pass
181
Edwin Törökb26e2792009-10-12 13:37:29 +0000[diff] [blame]182# Path to top level of LLVM hierarchy
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]183LEVEL = ../../..
184
185# Name of the library to build
186LIBRARYNAME = Hello
187
188# Make the shared library become a loadable module so the tools can
189# dlopen/dlsym on the resulting library.
190LOADABLE_MODULE = 1
191
192# Tell the build system which LLVM libraries your pass needs. You'll probably
193# need at least LLVMSystem.a, LLVMSupport.a, LLVMCore.a but possibly several
194# others too.
195LLVMLIBS = LLVMCore.a LLVMSupport.a LLVMSystem.a
196
197# Include the makefile implementation stuff
198include $(LEVEL)/Makefile.common
199</pre></div>
200
201<p>This makefile specifies that all of the <tt>.cpp</tt> files in the current
202directory are to be compiled and linked together into a
203<tt>Debug/lib/Hello.so</tt> shared object that can be dynamically loaded by
204the <tt>opt</tt> or <tt>bugpoint</tt> tools via their <tt>-load</tt> options.
205If your operating system uses a suffix other than .so (such as windows or
206Mac OS/X), the appropriate extension will be used.</p>
207
208<p>Now that we have the build scripts set up, we just need to write the code for
209the pass itself.</p>
210
211</div>
212
213<!-- ======================================================================= -->
214<div class="doc_subsection">
215 <a name="basiccode">Basic code required</a>
216</div>
217
218<div class="doc_text">
219
220<p>Now that we have a way to compile our new pass, we just have to write it.
221Start out with:</p>
222
223<div class="doc_code"><pre>
224<b>#include</b> "<a href="http://llvm.org/doxygen/Pass_8h-source.html">llvm/Pass.h</a>"
225<b>#include</b> "<a href="http://llvm.org/doxygen/Function_8h-source.html">llvm/Function.h</a>"
Chris Lattner1efd4fd62009-09-08 05:14:44 +0000[diff] [blame]226<b>#include</b> "<a href="http://llvm.org/doxygen/raw__ostream_8h.html">llvm/Support/raw_ostream.h</a>"
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]227</pre></div>
228
229<p>Which are needed because we are writing a <tt><a
Chris Lattner1efd4fd62009-09-08 05:14:44 +0000[diff] [blame]230href="http://llvm.org/doxygen/classllvm_1_1Pass.html">Pass</a></tt>,
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]231we are operating on <tt><a
Chris Lattner1efd4fd62009-09-08 05:14:44 +0000[diff] [blame]232href="http://llvm.org/doxygen/classllvm_1_1Function.html">Function</a></tt>'s,
233and we will be doing some printing.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]234
235<p>Next we have:</p>
236<div class="doc_code"><pre>
237<b>using namespace llvm;</b>
238</pre></div>
239<p>... which is required because the functions from the include files
240live in the llvm namespace.
241</p>
242
243<p>Next we have:</p>
244
245<div class="doc_code"><pre>
246<b>namespace</b> {
247</pre></div>
248
249<p>... which starts out an anonymous namespace. Anonymous namespaces are to C++
250what the "<tt>static</tt>" keyword is to C (at global scope). It makes the
251things declared inside of the anonymous namespace only visible to the current
252file. If you're not familiar with them, consult a decent C++ book for more
253information.</p>
254
255<p>Next, we declare our pass itself:</p>
256
257<div class="doc_code"><pre>
258 <b>struct</b> Hello : <b>public</b> <a href="#FunctionPass">FunctionPass</a> {
259</pre></div><p>
260
261<p>This declares a "<tt>Hello</tt>" class that is a subclass of <tt><a
262href="http://llvm.org/doxygen/classllvm_1_1FunctionPass.html">FunctionPass</a></tt>.
263The different builtin pass subclasses are described in detail <a
264href="#passtype">later</a>, but for now, know that <a
265href="#FunctionPass"><tt>FunctionPass</tt></a>'s operate a function at a
266time.</p>
267
268<div class="doc_code"><pre>
269 static char ID;
Dan Gohmanc74a1972009-02-18 05:09:16 +0000[diff] [blame]270 Hello() : FunctionPass(&amp;ID) {}
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]271</pre></div><p>
272
273<p> This declares pass identifier used by LLVM to identify pass. This allows LLVM to
274avoid using expensive C++ runtime information.</p>
275
276<div class="doc_code"><pre>
277 <b>virtual bool</b> <a href="#runOnFunction">runOnFunction</a>(Function &amp;F) {
Chris Lattner1efd4fd62009-09-08 05:14:44 +0000[diff] [blame]278 errs() &lt;&lt; "<i>Hello: </i>" &lt;&lt; F.getName() &lt;&lt; "\n";
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]279 <b>return false</b>;
280 }
281 }; <i>// end of struct Hello</i>
282</pre></div>
283
284<p>We declare a "<a href="#runOnFunction"><tt>runOnFunction</tt></a>" method,
285which overloads an abstract virtual method inherited from <a
286href="#FunctionPass"><tt>FunctionPass</tt></a>. This is where we are supposed
287to do our thing, so we just print out our message with the name of each
288function.</p>
289
290<div class="doc_code"><pre>
291 char Hello::ID = 0;
292</pre></div>
293
294<p> We initialize pass ID here. LLVM uses ID's address to identify pass so
295initialization value is not important.</p>
296
297<div class="doc_code"><pre>
Devang Patel3aab76e2008-03-19 21:56:59 +0000[diff] [blame]298 RegisterPass&lt;Hello&gt; X("<i>hello</i>", "<i>Hello World Pass</i>",
299 false /* Only looks at CFG */,
300 false /* Analysis Pass */);
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]301} <i>// end of anonymous namespace</i>
302</pre></div>
303
304<p>Lastly, we <a href="#registration">register our class</a> <tt>Hello</tt>,
305giving it a command line
Devang Patel3aab76e2008-03-19 21:56:59 +0000[diff] [blame]306argument "<tt>hello</tt>", and a name "<tt>Hello World Pass</tt>".
307Last two RegisterPass arguments are optional. Their default value is false.
308If a pass walks CFG without modifying it then third argument is set to true.
309If a pass is an analysis pass, for example dominator tree pass, then true
310is supplied as fourth argument. </p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]311
312<p>As a whole, the <tt>.cpp</tt> file looks like:</p>
313
314<div class="doc_code"><pre>
315<b>#include</b> "<a href="http://llvm.org/doxygen/Pass_8h-source.html">llvm/Pass.h</a>"
316<b>#include</b> "<a href="http://llvm.org/doxygen/Function_8h-source.html">llvm/Function.h</a>"
Chris Lattner1efd4fd62009-09-08 05:14:44 +0000[diff] [blame]317<b>#include</b> "<a href="http://llvm.org/doxygen/raw__ostream_8h.html">llvm/Support/raw_ostream.h</a>"
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]318
319<b>using namespace llvm;</b>
320
321<b>namespace</b> {
322 <b>struct Hello</b> : <b>public</b> <a href="#FunctionPass">FunctionPass</a> {
323
324 static char ID;
Dan Gohmanc74a1972009-02-18 05:09:16 +0000[diff] [blame]325 Hello() : FunctionPass(&amp;ID) {}
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]326
327 <b>virtual bool</b> <a href="#runOnFunction">runOnFunction</a>(Function &amp;F) {
Chris Lattner1efd4fd62009-09-08 05:14:44 +0000[diff] [blame]328 errs() &lt;&lt; "<i>Hello: </i>" &lt;&lt; F.getName() &lt;&lt; "\n";
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]329 <b>return false</b>;
330 }
331 };
332
Devang Patel8e46f052007-07-25 21:05:39 +0000[diff] [blame]333 char Hello::ID = 0;
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]334 RegisterPass&lt;Hello&gt; X("<i>hello</i>", "<i>Hello World Pass</i>");
335}
336</pre></div>
337
338<p>Now that it's all together, compile the file with a simple "<tt>gmake</tt>"
339command in the local directory and you should get a new
340"<tt>Debug/lib/Hello.so</tt> file. Note that everything in this file is
341contained in an anonymous namespace: this reflects the fact that passes are self
342contained units that do not need external interfaces (although they can have
343them) to be useful.</p>
344
345</div>
346
347<!-- ======================================================================= -->
348<div class="doc_subsection">
349 <a name="running">Running a pass with <tt>opt</tt></a>
350</div>
351
352<div class="doc_text">
353
354<p>Now that you have a brand new shiny shared object file, we can use the
355<tt>opt</tt> command to run an LLVM program through your pass. Because you
356registered your pass with the <tt>RegisterPass</tt> template, you will be able to
357use the <tt>opt</tt> tool to access it, once loaded.</p>
358
359<p>To test it, follow the example at the end of the <a
360href="GettingStarted.html">Getting Started Guide</a> to compile "Hello World" to
361LLVM. We can now run the bitcode file (<tt>hello.bc</tt>) for the program
362through our transformation like this (or course, any bitcode file will
363work):</p>
364
365<div class="doc_code"><pre>
366$ opt -load ../../../Debug/lib/Hello.so -hello &lt; hello.bc &gt; /dev/null
367Hello: __main
368Hello: puts
369Hello: main
370</pre></div>
371
372<p>The '<tt>-load</tt>' option specifies that '<tt>opt</tt>' should load your
373pass as a shared object, which makes '<tt>-hello</tt>' a valid command line
374argument (which is one reason you need to <a href="#registration">register your
375pass</a>). Because the hello pass does not modify the program in any
376interesting way, we just throw away the result of <tt>opt</tt> (sending it to
377<tt>/dev/null</tt>).</p>
378
379<p>To see what happened to the other string you registered, try running
Duncan Sandsc5c38f02010-02-18 14:08:13 +0000[diff] [blame]380<tt>opt</tt> with the <tt>-help</tt> option:</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]381
382<div class="doc_code"><pre>
Duncan Sandsc5c38f02010-02-18 14:08:13 +0000[diff] [blame]383$ opt -load ../../../Debug/lib/Hello.so -help
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]384OVERVIEW: llvm .bc -&gt; .bc modular optimizer
385
386USAGE: opt [options] &lt;input bitcode&gt;
387
388OPTIONS:
389 Optimizations available:
390...
391 -funcresolve - Resolve Functions
392 -gcse - Global Common Subexpression Elimination
393 -globaldce - Dead Global Elimination
394 <b>-hello - Hello World Pass</b>
395 -indvars - Canonicalize Induction Variables
396 -inline - Function Integration/Inlining
397 -instcombine - Combine redundant instructions
398...
399</pre></div>
400
401<p>The pass name get added as the information string for your pass, giving some
402documentation to users of <tt>opt</tt>. Now that you have a working pass, you
403would go ahead and make it do the cool transformations you want. Once you get
404it all working and tested, it may become useful to find out how fast your pass
405is. The <a href="#passManager"><tt>PassManager</tt></a> provides a nice command
406line option (<tt>--time-passes</tt>) that allows you to get information about
407the execution time of your pass along with the other passes you queue up. For
408example:</p>
409
410<div class="doc_code"><pre>
411$ opt -load ../../../Debug/lib/Hello.so -hello -time-passes &lt; hello.bc &gt; /dev/null
412Hello: __main
413Hello: puts
414Hello: main
415===============================================================================
416 ... Pass execution timing report ...
417===============================================================================
418 Total Execution Time: 0.02 seconds (0.0479059 wall clock)
419
420 ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Pass Name ---
421 0.0100 (100.0%) 0.0000 ( 0.0%) 0.0100 ( 50.0%) 0.0402 ( 84.0%) Bitcode Writer
422 0.0000 ( 0.0%) 0.0100 (100.0%) 0.0100 ( 50.0%) 0.0031 ( 6.4%) Dominator Set Construction
423 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0013 ( 2.7%) Module Verifier
424 <b> 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0033 ( 6.9%) Hello World Pass</b>
425 0.0100 (100.0%) 0.0100 (100.0%) 0.0200 (100.0%) 0.0479 (100.0%) TOTAL
426</pre></div>
427
428<p>As you can see, our implementation above is pretty fast :). The additional
429passes listed are automatically inserted by the '<tt>opt</tt>' tool to verify
430that the LLVM emitted by your pass is still valid and well formed LLVM, which
431hasn't been broken somehow.</p>
432
433<p>Now that you have seen the basics of the mechanics behind passes, we can talk
434about some more details of how they work and how to use them.</p>
435
436</div>
437
438<!-- *********************************************************************** -->
439<div class="doc_section">
440 <a name="passtype">Pass classes and requirements</a>
441</div>
442<!-- *********************************************************************** -->
443
444<div class="doc_text">
445
446<p>One of the first things that you should do when designing a new pass is to
447decide what class you should subclass for your pass. The <a
448href="#basiccode">Hello World</a> example uses the <tt><a
449href="#FunctionPass">FunctionPass</a></tt> class for its implementation, but we
450did not discuss why or when this should occur. Here we talk about the classes
451available, from the most general to the most specific.</p>
452
453<p>When choosing a superclass for your Pass, you should choose the <b>most
454specific</b> class possible, while still being able to meet the requirements
455listed. This gives the LLVM Pass Infrastructure information necessary to
Benjamin Kramer5fb9d7e2009-10-12 14:46:08 +0000[diff] [blame]456optimize how passes are run, so that the resultant compiler isn't unnecessarily
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]457slow.</p>
458
459</div>
460
461<!-- ======================================================================= -->
462<div class="doc_subsection">
463 <a name="ImmutablePass">The <tt>ImmutablePass</tt> class</a>
464</div>
465
466<div class="doc_text">
467
468<p>The most plain and boring type of pass is the "<tt><a
469href="http://llvm.org/doxygen/classllvm_1_1ImmutablePass.html">ImmutablePass</a></tt>"
470class. This pass type is used for passes that do not have to be run, do not
471change state, and never need to be updated. This is not a normal type of
472transformation or analysis, but can provide information about the current
473compiler configuration.</p>
474
475<p>Although this pass class is very infrequently used, it is important for
476providing information about the current target machine being compiled for, and
477other static information that can affect the various transformations.</p>
478
479<p><tt>ImmutablePass</tt>es never invalidate other transformations, are never
480invalidated, and are never "run".</p>
481
482</div>
483
484<!-- ======================================================================= -->
485<div class="doc_subsection">
486 <a name="ModulePass">The <tt>ModulePass</tt> class</a>
487</div>
488
489<div class="doc_text">
490
491<p>The "<tt><a
492href="http://llvm.org/doxygen/classllvm_1_1ModulePass.html">ModulePass</a></tt>"
493class is the most general of all superclasses that you can use. Deriving from
494<tt>ModulePass</tt> indicates that your pass uses the entire program as a unit,
Benjamin Kramer5fb9d7e2009-10-12 14:46:08 +0000[diff] [blame]495referring to function bodies in no predictable order, or adding and removing
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]496functions. Because nothing is known about the behavior of <tt>ModulePass</tt>
Daniel Dunbarb7750c52009-07-01 23:38:44 +0000[diff] [blame]497subclasses, no optimization can be done for their execution.</p>
498
499<p>A module pass can use function level passes (e.g. dominators) using
500the getAnalysis interface
501<tt>getAnalysis&lt;DominatorTree&gt;(llvm::Function *)</tt> to provide the
502function to retrieve analysis result for, if the function pass does not require
Devang Patel80eb3972009-08-10 16:37:29 +0000[diff] [blame]503any module or immutable passes. Note that this can only be done for functions for which the
Daniel Dunbarb7750c52009-07-01 23:38:44 +0000[diff] [blame]504analysis ran, e.g. in the case of dominators you should only ask for the
505DominatorTree for function definitions, not declarations.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]506
507<p>To write a correct <tt>ModulePass</tt> subclass, derive from
508<tt>ModulePass</tt> and overload the <tt>runOnModule</tt> method with the
509following signature:</p>
510
511</div>
512
513<!-- _______________________________________________________________________ -->
514<div class="doc_subsubsection">
515 <a name="runOnModule">The <tt>runOnModule</tt> method</a>
516</div>
517
518<div class="doc_text">
519
520<div class="doc_code"><pre>
521 <b>virtual bool</b> runOnModule(Module &amp;M) = 0;
522</pre></div>
523
524<p>The <tt>runOnModule</tt> method performs the interesting work of the pass.
525It should return true if the module was modified by the transformation and
526false otherwise.</p>
527
528</div>
529
530<!-- ======================================================================= -->
531<div class="doc_subsection">
532 <a name="CallGraphSCCPass">The <tt>CallGraphSCCPass</tt> class</a>
533</div>
534
535<div class="doc_text">
536
537<p>The "<tt><a
538href="http://llvm.org/doxygen/classllvm_1_1CallGraphSCCPass.html">CallGraphSCCPass</a></tt>"
539is used by passes that need to traverse the program bottom-up on the call graph
540(callees before callers). Deriving from CallGraphSCCPass provides some
541mechanics for building and traversing the CallGraph, but also allows the system
542to optimize execution of CallGraphSCCPass's. If your pass meets the
543requirements outlined below, and doesn't meet the requirements of a <tt><a
544href="#FunctionPass">FunctionPass</a></tt> or <tt><a
545href="#BasicBlockPass">BasicBlockPass</a></tt>, you should derive from
546<tt>CallGraphSCCPass</tt>.</p>
547
548<p><b>TODO</b>: explain briefly what SCC, Tarjan's algo, and B-U mean.</p>
549
550<p>To be explicit, <tt>CallGraphSCCPass</tt> subclasses are:</p>
551
552<ol>
553
554<li>... <em>not allowed</em> to modify any <tt>Function</tt>s that are not in
555the current SCC.</li>
556
557<li>... <em>not allowed</em> to inspect any Function's other than those in the
558current SCC and the direct callees of the SCC.</li>
559
560<li>... <em>required</em> to preserve the current CallGraph object, updating it
561to reflect any changes made to the program.</li>
562
563<li>... <em>not allowed</em> to add or remove SCC's from the current Module,
564though they may change the contents of an SCC.</li>
565
566<li>... <em>allowed</em> to add or remove global variables from the current
567Module.</li>
568
569<li>... <em>allowed</em> to maintain state across invocations of
570 <a href="#runOnSCC"><tt>runOnSCC</tt></a> (including global data).</li>
571</ol>
572
573<p>Implementing a <tt>CallGraphSCCPass</tt> is slightly tricky in some cases
574because it has to handle SCCs with more than one node in it. All of the virtual
575methods described below should return true if they modified the program, or
576false if they didn't.</p>
577
578</div>
579
580<!-- _______________________________________________________________________ -->
581<div class="doc_subsubsection">
582 <a name="doInitialization_scc">The <tt>doInitialization(CallGraph &amp;)</tt>
583 method</a>
584</div>
585
586<div class="doc_text">
587
588<div class="doc_code"><pre>
589 <b>virtual bool</b> doInitialization(CallGraph &amp;CG);
590</pre></div>
591
592<p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
593<tt>CallGraphSCCPass</tt>'s are not allowed to do. They can add and remove
594functions, get pointers to functions, etc. The <tt>doInitialization</tt> method
595is designed to do simple initialization type of stuff that does not depend on
596the SCCs being processed. The <tt>doInitialization</tt> method call is not
597scheduled to overlap with any other pass executions (thus it should be very
598fast).</p>
599
600</div>
601
602<!-- _______________________________________________________________________ -->
603<div class="doc_subsubsection">
604 <a name="runOnSCC">The <tt>runOnSCC</tt> method</a>
605</div>
606
607<div class="doc_text">
608
609<div class="doc_code"><pre>
610 <b>virtual bool</b> runOnSCC(const std::vector&lt;CallGraphNode *&gt; &amp;SCCM) = 0;
611</pre></div>
612
613<p>The <tt>runOnSCC</tt> method performs the interesting work of the pass, and
614should return true if the module was modified by the transformation, false
615otherwise.</p>
616
617</div>
618
619<!-- _______________________________________________________________________ -->
620<div class="doc_subsubsection">
621 <a name="doFinalization_scc">The <tt>doFinalization(CallGraph
622 &amp;)</tt> method</a>
623</div>
624
625<div class="doc_text">
626
627<div class="doc_code"><pre>
628 <b>virtual bool</b> doFinalization(CallGraph &amp;CG);
629</pre></div>
630
631<p>The <tt>doFinalization</tt> method is an infrequently used method that is
632called when the pass framework has finished calling <a
633href="#runOnFunction"><tt>runOnFunction</tt></a> for every function in the
634program being compiled.</p>
635
636</div>
637
638<!-- ======================================================================= -->
639<div class="doc_subsection">
640 <a name="FunctionPass">The <tt>FunctionPass</tt> class</a>
641</div>
642
643<div class="doc_text">
644
645<p>In contrast to <tt>ModulePass</tt> subclasses, <tt><a
646href="http://llvm.org/doxygen/classllvm_1_1Pass.html">FunctionPass</a></tt>
647subclasses do have a predictable, local behavior that can be expected by the
648system. All <tt>FunctionPass</tt> execute on each function in the program
649independent of all of the other functions in the program.
650<tt>FunctionPass</tt>'s do not require that they are executed in a particular
651order, and <tt>FunctionPass</tt>'s do not modify external functions.</p>
652
653<p>To be explicit, <tt>FunctionPass</tt> subclasses are not allowed to:</p>
654
655<ol>
656<li>Modify a Function other than the one currently being processed.</li>
657<li>Add or remove Function's from the current Module.</li>
658<li>Add or remove global variables from the current Module.</li>
659<li>Maintain state across invocations of
660 <a href="#runOnFunction"><tt>runOnFunction</tt></a> (including global data)</li>
661</ol>
662
663<p>Implementing a <tt>FunctionPass</tt> is usually straightforward (See the <a
664href="#basiccode">Hello World</a> pass for example). <tt>FunctionPass</tt>'s
665may overload three virtual methods to do their work. All of these methods
666should return true if they modified the program, or false if they didn't.</p>
667
668</div>
669
670<!-- _______________________________________________________________________ -->
671<div class="doc_subsubsection">
672 <a name="doInitialization_mod">The <tt>doInitialization(Module &amp;)</tt>
673 method</a>
674</div>
675
676<div class="doc_text">
677
678<div class="doc_code"><pre>
679 <b>virtual bool</b> doInitialization(Module &amp;M);
680</pre></div>
681
682<p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
683<tt>FunctionPass</tt>'s are not allowed to do. They can add and remove
684functions, get pointers to functions, etc. The <tt>doInitialization</tt> method
685is designed to do simple initialization type of stuff that does not depend on
686the functions being processed. The <tt>doInitialization</tt> method call is not
687scheduled to overlap with any other pass executions (thus it should be very
688fast).</p>
689
690<p>A good example of how this method should be used is the <a
691href="http://llvm.org/doxygen/LowerAllocations_8cpp-source.html">LowerAllocations</a>
692pass. This pass converts <tt>malloc</tt> and <tt>free</tt> instructions into
693platform dependent <tt>malloc()</tt> and <tt>free()</tt> function calls. It
694uses the <tt>doInitialization</tt> method to get a reference to the malloc and
695free functions that it needs, adding prototypes to the module if necessary.</p>
696
697</div>
698
699<!-- _______________________________________________________________________ -->
700<div class="doc_subsubsection">
701 <a name="runOnFunction">The <tt>runOnFunction</tt> method</a>
702</div>
703
704<div class="doc_text">
705
706<div class="doc_code"><pre>
707 <b>virtual bool</b> runOnFunction(Function &amp;F) = 0;
708</pre></div><p>
709
710<p>The <tt>runOnFunction</tt> method must be implemented by your subclass to do
711the transformation or analysis work of your pass. As usual, a true value should
712be returned if the function is modified.</p>
713
714</div>
715
716<!-- _______________________________________________________________________ -->
717<div class="doc_subsubsection">
718 <a name="doFinalization_mod">The <tt>doFinalization(Module
719 &amp;)</tt> method</a>
720</div>
721
722<div class="doc_text">
723
724<div class="doc_code"><pre>
725 <b>virtual bool</b> doFinalization(Module &amp;M);
726</pre></div>
727
728<p>The <tt>doFinalization</tt> method is an infrequently used method that is
729called when the pass framework has finished calling <a
730href="#runOnFunction"><tt>runOnFunction</tt></a> for every function in the
731program being compiled.</p>
732
733</div>
734
735<!-- ======================================================================= -->
736<div class="doc_subsection">
737 <a name="LoopPass">The <tt>LoopPass</tt> class </a>
738</div>
739
740<div class="doc_text">
741
742<p> All <tt>LoopPass</tt> execute on each loop in the function independent of
743all of the other loops in the function. <tt>LoopPass</tt> processes loops in
744loop nest order such that outer most loop is processed last. </p>
745
746<p> <tt>LoopPass</tt> subclasses are allowed to update loop nest using
747<tt>LPPassManager</tt> interface. Implementing a loop pass is usually
748straightforward. <tt>Looppass</tt>'s may overload three virtual methods to
749do their work. All these methods should return true if they modified the
750program, or false if they didn't. </p>
751</div>
752
753<!-- _______________________________________________________________________ -->
754<div class="doc_subsubsection">
755 <a name="doInitialization_loop">The <tt>doInitialization(Loop *,
756 LPPassManager &amp;)</tt>
757 method</a>
758</div>
759
760<div class="doc_text">
761
762<div class="doc_code"><pre>
763 <b>virtual bool</b> doInitialization(Loop *, LPPassManager &amp;LPM);
764</pre></div>
765
766<p>The <tt>doInitialization</tt> method is designed to do simple initialization
767type of stuff that does not depend on the functions being processed. The
768<tt>doInitialization</tt> method call is not scheduled to overlap with any
769other pass executions (thus it should be very fast). LPPassManager
770interface should be used to access Function or Module level analysis
771information.</p>
772
773</div>
774
775
776<!-- _______________________________________________________________________ -->
777<div class="doc_subsubsection">
778 <a name="runOnLoop">The <tt>runOnLoop</tt> method</a>
779</div>
780
781<div class="doc_text">
782
783<div class="doc_code"><pre>
784 <b>virtual bool</b> runOnLoop(Loop *, LPPassManager &amp;LPM) = 0;
785</pre></div><p>
786
787<p>The <tt>runOnLoop</tt> method must be implemented by your subclass to do
788the transformation or analysis work of your pass. As usual, a true value should
789be returned if the function is modified. <tt>LPPassManager</tt> interface
790should be used to update loop nest.</p>
791
792</div>
793
794<!-- _______________________________________________________________________ -->
795<div class="doc_subsubsection">
796 <a name="doFinalization_loop">The <tt>doFinalization()</tt> method</a>
797</div>
798
799<div class="doc_text">
800
801<div class="doc_code"><pre>
802 <b>virtual bool</b> doFinalization();
803</pre></div>
804
805<p>The <tt>doFinalization</tt> method is an infrequently used method that is
806called when the pass framework has finished calling <a
807href="#runOnLoop"><tt>runOnLoop</tt></a> for every loop in the
808program being compiled. </p>
809
810</div>
811
812
813
814<!-- ======================================================================= -->
815<div class="doc_subsection">
816 <a name="BasicBlockPass">The <tt>BasicBlockPass</tt> class</a>
817</div>
818
819<div class="doc_text">
820
821<p><tt>BasicBlockPass</tt>'s are just like <a
822href="#FunctionPass"><tt>FunctionPass</tt></a>'s, except that they must limit
823their scope of inspection and modification to a single basic block at a time.
824As such, they are <b>not</b> allowed to do any of the following:</p>
825
826<ol>
827<li>Modify or inspect any basic blocks outside of the current one</li>
828<li>Maintain state across invocations of
829 <a href="#runOnBasicBlock"><tt>runOnBasicBlock</tt></a></li>
830<li>Modify the control flow graph (by altering terminator instructions)</li>
831<li>Any of the things forbidden for
832 <a href="#FunctionPass"><tt>FunctionPass</tt></a>es.</li>
833</ol>
834
835<p><tt>BasicBlockPass</tt>es are useful for traditional local and "peephole"
836optimizations. They may override the same <a
837href="#doInitialization_mod"><tt>doInitialization(Module &amp;)</tt></a> and <a
838href="#doFinalization_mod"><tt>doFinalization(Module &amp;)</tt></a> methods that <a
839href="#FunctionPass"><tt>FunctionPass</tt></a>'s have, but also have the following virtual methods that may also be implemented:</p>
840
841</div>
842
843<!-- _______________________________________________________________________ -->
844<div class="doc_subsubsection">
845 <a name="doInitialization_fn">The <tt>doInitialization(Function
846 &amp;)</tt> method</a>
847</div>
848
849<div class="doc_text">
850
851<div class="doc_code"><pre>
852 <b>virtual bool</b> doInitialization(Function &amp;F);
853</pre></div>
854
855<p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
856<tt>BasicBlockPass</tt>'s are not allowed to do, but that
857<tt>FunctionPass</tt>'s can. The <tt>doInitialization</tt> method is designed
858to do simple initialization that does not depend on the
859BasicBlocks being processed. The <tt>doInitialization</tt> method call is not
860scheduled to overlap with any other pass executions (thus it should be very
861fast).</p>
862
863</div>
864
865<!-- _______________________________________________________________________ -->
866<div class="doc_subsubsection">
867 <a name="runOnBasicBlock">The <tt>runOnBasicBlock</tt> method</a>
868</div>
869
870<div class="doc_text">
871
872<div class="doc_code"><pre>
873 <b>virtual bool</b> runOnBasicBlock(BasicBlock &amp;BB) = 0;
874</pre></div>
875
876<p>Override this function to do the work of the <tt>BasicBlockPass</tt>. This
877function is not allowed to inspect or modify basic blocks other than the
878parameter, and are not allowed to modify the CFG. A true value must be returned
879if the basic block is modified.</p>
880
881</div>
882
883<!-- _______________________________________________________________________ -->
884<div class="doc_subsubsection">
885 <a name="doFinalization_fn">The <tt>doFinalization(Function &amp;)</tt>
886 method</a>
887</div>
888
889<div class="doc_text">
890
891<div class="doc_code"><pre>
892 <b>virtual bool</b> doFinalization(Function &amp;F);
893</pre></div>
894
895<p>The <tt>doFinalization</tt> method is an infrequently used method that is
896called when the pass framework has finished calling <a
897href="#runOnBasicBlock"><tt>runOnBasicBlock</tt></a> for every BasicBlock in the
898program being compiled. This can be used to perform per-function
899finalization.</p>
900
901</div>
902
903<!-- ======================================================================= -->
904<div class="doc_subsection">
905 <a name="MachineFunctionPass">The <tt>MachineFunctionPass</tt> class</a>
906</div>
907
908<div class="doc_text">
909
910<p>A <tt>MachineFunctionPass</tt> is a part of the LLVM code generator that
911executes on the machine-dependent representation of each LLVM function in the
Dan Gohmanc23b1a32010-03-10 01:29:39 +0000[diff] [blame]912program.</p>
913
914<p>Code generator passes are registered and initialized specially by
915<tt>TargetMachine::addPassesToEmitFile</tt> and similar routines, so they
916cannot generally be run from the <tt>opt</tt> or <tt>bugpoint</tt>
917commands.</p>
918
919<p>A <tt>MachineFunctionPass</tt> is also a <tt>FunctionPass</tt>, so all
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]920the restrictions that apply to a <tt>FunctionPass</tt> also apply to it.
921<tt>MachineFunctionPass</tt>es also have additional restrictions. In particular,
922<tt>MachineFunctionPass</tt>es are not allowed to do any of the following:</p>
923
924<ol>
Dan Gohmanc23b1a32010-03-10 01:29:39 +0000[diff] [blame]925<li>Modify or create any LLVM IR Instructions, BasicBlocks, Arguments,
926 Functions, GlobalVariables, GlobalAliases, or Modules.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]927<li>Modify a MachineFunction other than the one currently being processed.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]928<li>Maintain state across invocations of <a
929href="#runOnMachineFunction"><tt>runOnMachineFunction</tt></a> (including global
930data)</li>
931</ol>
932
933</div>
934
935<!-- _______________________________________________________________________ -->
936<div class="doc_subsubsection">
937 <a name="runOnMachineFunction">The <tt>runOnMachineFunction(MachineFunction
938 &amp;MF)</tt> method</a>
939</div>
940
941<div class="doc_text">
942
943<div class="doc_code"><pre>
944 <b>virtual bool</b> runOnMachineFunction(MachineFunction &amp;MF) = 0;
945</pre></div>
946
947<p><tt>runOnMachineFunction</tt> can be considered the main entry point of a
948<tt>MachineFunctionPass</tt>; that is, you should override this method to do the
949work of your <tt>MachineFunctionPass</tt>.</p>
950
951<p>The <tt>runOnMachineFunction</tt> method is called on every
952<tt>MachineFunction</tt> in a <tt>Module</tt>, so that the
953<tt>MachineFunctionPass</tt> may perform optimizations on the machine-dependent
954representation of the function. If you want to get at the LLVM <tt>Function</tt>
955for the <tt>MachineFunction</tt> you're working on, use
956<tt>MachineFunction</tt>'s <tt>getFunction()</tt> accessor method -- but
957remember, you may not modify the LLVM <tt>Function</tt> or its contents from a
958<tt>MachineFunctionPass</tt>.</p>
959
960</div>
961
962<!-- *********************************************************************** -->
963<div class="doc_section">
964 <a name="registration">Pass registration</a>
965</div>
966<!-- *********************************************************************** -->
967
968<div class="doc_text">
969
970<p>In the <a href="#basiccode">Hello World</a> example pass we illustrated how
971pass registration works, and discussed some of the reasons that it is used and
972what it does. Here we discuss how and why passes are registered.</p>
973
974<p>As we saw above, passes are registered with the <b><tt>RegisterPass</tt></b>
975template, which requires you to pass at least two
976parameters. The first parameter is the name of the pass that is to be used on
977the command line to specify that the pass should be added to a program (for
978example, with <tt>opt</tt> or <tt>bugpoint</tt>). The second argument is the
Duncan Sandsc5c38f02010-02-18 14:08:13 +0000[diff] [blame]979name of the pass, which is to be used for the <tt>-help</tt> output of
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]980programs, as
981well as for debug output generated by the <tt>--debug-pass</tt> option.</p>
982
983<p>If you want your pass to be easily dumpable, you should
984implement the virtual <tt>print</tt> method:</p>
985
986</div>
987
988<!-- _______________________________________________________________________ -->
989<div class="doc_subsubsection">
990 <a name="print">The <tt>print</tt> method</a>
991</div>
992
993<div class="doc_text">
994
995<div class="doc_code"><pre>
Edwin Török72a8fd22008-10-28 17:29:23 +0000[diff] [blame]996 <b>virtual void</b> print(std::ostream &amp;O, <b>const</b> Module *M) <b>const</b>;
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]997</pre></div>
998
999<p>The <tt>print</tt> method must be implemented by "analyses" in order to print
1000a human readable version of the analysis results. This is useful for debugging
1001an analysis itself, as well as for other people to figure out how an analysis
1002works. Use the <tt>opt -analyze</tt> argument to invoke this method.</p>
1003
1004<p>The <tt>llvm::OStream</tt> parameter specifies the stream to write the results on,
1005and the <tt>Module</tt> parameter gives a pointer to the top level module of the
1006program that has been analyzed. Note however that this pointer may be null in
1007certain circumstances (such as calling the <tt>Pass::dump()</tt> from a
1008debugger), so it should only be used to enhance debug output, it should not be
1009depended on.</p>
1010
1011</div>
1012
1013<!-- *********************************************************************** -->
1014<div class="doc_section">
1015 <a name="interaction">Specifying interactions between passes</a>
1016</div>
1017<!-- *********************************************************************** -->
1018
1019<div class="doc_text">
1020
John Criswella99e43f2007-12-03 19:34:25 +0000[diff] [blame]1021<p>One of the main responsibilities of the <tt>PassManager</tt> is to make sure
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1022that passes interact with each other correctly. Because <tt>PassManager</tt>
1023tries to <a href="#passmanager">optimize the execution of passes</a> it must
1024know how the passes interact with each other and what dependencies exist between
1025the various passes. To track this, each pass can declare the set of passes that
1026are required to be executed before the current pass, and the passes which are
1027invalidated by the current pass.</p>
1028
1029<p>Typically this functionality is used to require that analysis results are
1030computed before your pass is run. Running arbitrary transformation passes can
1031invalidate the computed analysis results, which is what the invalidation set
1032specifies. If a pass does not implement the <tt><a
1033href="#getAnalysisUsage">getAnalysisUsage</a></tt> method, it defaults to not
1034having any prerequisite passes, and invalidating <b>all</b> other passes.</p>
1035
1036</div>
1037
1038<!-- _______________________________________________________________________ -->
1039<div class="doc_subsubsection">
1040 <a name="getAnalysisUsage">The <tt>getAnalysisUsage</tt> method</a>
1041</div>
1042
1043<div class="doc_text">
1044
1045<div class="doc_code"><pre>
1046 <b>virtual void</b> getAnalysisUsage(AnalysisUsage &amp;Info) <b>const</b>;
1047</pre></div>
1048
1049<p>By implementing the <tt>getAnalysisUsage</tt> method, the required and
1050invalidated sets may be specified for your transformation. The implementation
1051should fill in the <tt><a
1052href="http://llvm.org/doxygen/classllvm_1_1AnalysisUsage.html">AnalysisUsage</a></tt>
1053object with information about which passes are required and not invalidated. To
1054do this, a pass may call any of the following methods on the AnalysisUsage
1055object:</p>
1056</div>
1057
1058<!-- _______________________________________________________________________ -->
1059<div class="doc_subsubsection">
1060 <a name="AU::addRequired">The <tt>AnalysisUsage::addRequired&lt;&gt;</tt> and <tt>AnalysisUsage::addRequiredTransitive&lt;&gt;</tt> methods</a>
1061</div>
1062
1063<div class="doc_text">
1064<p>
1065If your pass requires a previous pass to be executed (an analysis for example),
1066it can use one of these methods to arrange for it to be run before your pass.
1067LLVM has many different types of analyses and passes that can be required,
1068spanning the range from <tt>DominatorSet</tt> to <tt>BreakCriticalEdges</tt>.
1069Requiring <tt>BreakCriticalEdges</tt>, for example, guarantees that there will
1070be no critical edges in the CFG when your pass has been run.
1071</p>
1072
1073<p>
1074Some analyses chain to other analyses to do their job. For example, an <a
1075href="AliasAnalysis.html">AliasAnalysis</a> implementation is required to <a
1076href="AliasAnalysis.html#chaining">chain</a> to other alias analysis passes. In
1077cases where analyses chain, the <tt>addRequiredTransitive</tt> method should be
1078used instead of the <tt>addRequired</tt> method. This informs the PassManager
1079that the transitively required pass should be alive as long as the requiring
1080pass is.
1081</p>
1082</div>
1083
1084<!-- _______________________________________________________________________ -->
1085<div class="doc_subsubsection">
1086 <a name="AU::addPreserved">The <tt>AnalysisUsage::addPreserved&lt;&gt;</tt> method</a>
1087</div>
1088
1089<div class="doc_text">
1090<p>
1091One of the jobs of the PassManager is to optimize how and when analyses are run.
1092In particular, it attempts to avoid recomputing data unless it needs to. For
1093this reason, passes are allowed to declare that they preserve (i.e., they don't
1094invalidate) an existing analysis if it's available. For example, a simple
1095constant folding pass would not modify the CFG, so it can't possibly affect the
1096results of dominator analysis. By default, all passes are assumed to invalidate
1097all others.
1098</p>
1099
1100<p>
1101The <tt>AnalysisUsage</tt> class provides several methods which are useful in
1102certain circumstances that are related to <tt>addPreserved</tt>. In particular,
1103the <tt>setPreservesAll</tt> method can be called to indicate that the pass does
1104not modify the LLVM program at all (which is true for analyses), and the
1105<tt>setPreservesCFG</tt> method can be used by transformations that change
1106instructions in the program but do not modify the CFG or terminator instructions
1107(note that this property is implicitly set for <a
1108href="#BasicBlockPass">BasicBlockPass</a>'s).
1109</p>
1110
1111<p>
1112<tt>addPreserved</tt> is particularly useful for transformations like
1113<tt>BreakCriticalEdges</tt>. This pass knows how to update a small set of loop
1114and dominator related analyses if they exist, so it can preserve them, despite
1115the fact that it hacks on the CFG.
1116</p>
1117</div>
1118
1119<!-- _______________________________________________________________________ -->
1120<div class="doc_subsubsection">
1121 <a name="AU::examples">Example implementations of <tt>getAnalysisUsage</tt></a>
1122</div>
1123
1124<div class="doc_text">
1125
1126<div class="doc_code"><pre>
1127 <i>// This is an example implementation from an analysis, which does not modify
1128 // the program at all, yet has a prerequisite.</i>
1129 <b>void</b> <a href="http://llvm.org/doxygen/classllvm_1_1PostDominanceFrontier.html">PostDominanceFrontier</a>::getAnalysisUsage(AnalysisUsage &amp;AU) <b>const</b> {
1130 AU.setPreservesAll();
1131 AU.addRequired&lt;<a href="http://llvm.org/doxygen/classllvm_1_1PostDominatorTree.html">PostDominatorTree</a>&gt;();
1132 }
1133</pre></div>
1134
1135<p>and:</p>
1136
1137<div class="doc_code"><pre>
1138 <i>// This example modifies the program, but does not modify the CFG</i>
1139 <b>void</b> <a href="http://llvm.org/doxygen/structLICM.html">LICM</a>::getAnalysisUsage(AnalysisUsage &amp;AU) <b>const</b> {
1140 AU.setPreservesCFG();
1141 AU.addRequired&lt;<a href="http://llvm.org/doxygen/classllvm_1_1LoopInfo.html">LoopInfo</a>&gt;();
1142 }
1143</pre></div>
1144
1145</div>
1146
1147<!-- _______________________________________________________________________ -->
1148<div class="doc_subsubsection">
Duncan Sands4e0d6a72009-01-28 13:14:17 +0000[diff] [blame]1149 <a name="getAnalysis">The <tt>getAnalysis&lt;&gt;</tt> and
1150<tt>getAnalysisIfAvailable&lt;&gt;</tt> methods</a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1151</div>
1152
1153<div class="doc_text">
1154
1155<p>The <tt>Pass::getAnalysis&lt;&gt;</tt> method is automatically inherited by
1156your class, providing you with access to the passes that you declared that you
1157required with the <a href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a>
1158method. It takes a single template argument that specifies which pass class you
1159want, and returns a reference to that pass. For example:</p>
1160
1161<div class="doc_code"><pre>
1162 bool LICM::runOnFunction(Function &amp;F) {
1163 LoopInfo &amp;LI = getAnalysis&lt;LoopInfo&gt;();
1164 ...
1165 }
1166</pre></div>
1167
1168<p>This method call returns a reference to the pass desired. You may get a
1169runtime assertion failure if you attempt to get an analysis that you did not
1170declare as required in your <a
1171href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> implementation. This
1172method can be called by your <tt>run*</tt> method implementation, or by any
1173other local method invoked by your <tt>run*</tt> method.
1174
1175A module level pass can use function level analysis info using this interface.
1176For example:</p>
1177
1178<div class="doc_code"><pre>
1179 bool ModuleLevelPass::runOnModule(Module &amp;M) {
1180 ...
1181 DominatorTree &amp;DT = getAnalysis&lt;DominatorTree&gt;(Func);
1182 ...
1183 }
1184</pre></div>
1185
1186<p>In above example, runOnFunction for DominatorTree is called by pass manager
1187before returning a reference to the desired pass.</p>
1188
1189<p>
1190If your pass is capable of updating analyses if they exist (e.g.,
1191<tt>BreakCriticalEdges</tt>, as described above), you can use the
Duncan Sands4e0d6a72009-01-28 13:14:17 +0000[diff] [blame]1192<tt>getAnalysisIfAvailable</tt> method, which returns a pointer to the analysis
1193if it is active. For example:</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1194
1195<div class="doc_code"><pre>
1196 ...
Duncan Sands4e0d6a72009-01-28 13:14:17 +0000[diff] [blame]1197 if (DominatorSet *DS = getAnalysisIfAvailable&lt;DominatorSet&gt;()) {
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1198 <i>// A DominatorSet is active. This code will update it.</i>
1199 }
1200 ...
1201</pre></div>
1202
1203</div>
1204
1205<!-- *********************************************************************** -->
1206<div class="doc_section">
1207 <a name="analysisgroup">Implementing Analysis Groups</a>
1208</div>
1209<!-- *********************************************************************** -->
1210
1211<div class="doc_text">
1212
Chris Lattner942c3952007-11-16 05:32:05 +0000[diff] [blame]1213<p>Now that we understand the basics of how passes are defined, how they are
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1214used, and how they are required from other passes, it's time to get a little bit
1215fancier. All of the pass relationships that we have seen so far are very
1216simple: one pass depends on one other specific pass to be run before it can run.
1217For many applications, this is great, for others, more flexibility is
1218required.</p>
1219
1220<p>In particular, some analyses are defined such that there is a single simple
1221interface to the analysis results, but multiple ways of calculating them.
1222Consider alias analysis for example. The most trivial alias analysis returns
1223"may alias" for any alias query. The most sophisticated analysis a
1224flow-sensitive, context-sensitive interprocedural analysis that can take a
1225significant amount of time to execute (and obviously, there is a lot of room
1226between these two extremes for other implementations). To cleanly support
1227situations like this, the LLVM Pass Infrastructure supports the notion of
1228Analysis Groups.</p>
1229
1230</div>
1231
1232<!-- _______________________________________________________________________ -->
1233<div class="doc_subsubsection">
1234 <a name="agconcepts">Analysis Group Concepts</a>
1235</div>
1236
1237<div class="doc_text">
1238
1239<p>An Analysis Group is a single simple interface that may be implemented by
1240multiple different passes. Analysis Groups can be given human readable names
1241just like passes, but unlike passes, they need not derive from the <tt>Pass</tt>
1242class. An analysis group may have one or more implementations, one of which is
1243the "default" implementation.</p>
1244
1245<p>Analysis groups are used by client passes just like other passes are: the
1246<tt>AnalysisUsage::addRequired()</tt> and <tt>Pass::getAnalysis()</tt> methods.
1247In order to resolve this requirement, the <a href="#passmanager">PassManager</a>
1248scans the available passes to see if any implementations of the analysis group
1249are available. If none is available, the default implementation is created for
1250the pass to use. All standard rules for <A href="#interaction">interaction
1251between passes</a> still apply.</p>
1252
1253<p>Although <a href="#registration">Pass Registration</a> is optional for normal
1254passes, all analysis group implementations must be registered, and must use the
1255<A href="#registerag"><tt>RegisterAnalysisGroup</tt></a> template to join the
1256implementation pool. Also, a default implementation of the interface
1257<b>must</b> be registered with <A
1258href="#registerag"><tt>RegisterAnalysisGroup</tt></a>.</p>
1259
1260<p>As a concrete example of an Analysis Group in action, consider the <a
1261href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>
1262analysis group. The default implementation of the alias analysis interface (the
1263<tt><a
1264href="http://llvm.org/doxygen/structBasicAliasAnalysis.html">basicaa</a></tt>
1265pass) just does a few simple checks that don't require significant analysis to
1266compute (such as: two different globals can never alias each other, etc).
1267Passes that use the <tt><a
1268href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a></tt>
1269interface (for example the <tt><a
1270href="http://llvm.org/doxygen/structGCSE.html">gcse</a></tt> pass), do
1271not care which implementation of alias analysis is actually provided, they just
1272use the designated interface.</p>
1273
1274<p>From the user's perspective, commands work just like normal. Issuing the
1275command '<tt>opt -gcse ...</tt>' will cause the <tt>basicaa</tt> class to be
1276instantiated and added to the pass sequence. Issuing the command '<tt>opt
1277-somefancyaa -gcse ...</tt>' will cause the <tt>gcse</tt> pass to use the
1278<tt>somefancyaa</tt> alias analysis (which doesn't actually exist, it's just a
1279hypothetical example) instead.</p>
1280
1281</div>
1282
1283<!-- _______________________________________________________________________ -->
1284<div class="doc_subsubsection">
1285 <a name="registerag">Using <tt>RegisterAnalysisGroup</tt></a>
1286</div>
1287
1288<div class="doc_text">
1289
1290<p>The <tt>RegisterAnalysisGroup</tt> template is used to register the analysis
1291group itself as well as add pass implementations to the analysis group. First,
1292an analysis should be registered, with a human readable name provided for it.
1293Unlike registration of passes, there is no command line argument to be specified
1294for the Analysis Group Interface itself, because it is "abstract":</p>
1295
1296<div class="doc_code"><pre>
1297 <b>static</b> RegisterAnalysisGroup&lt;<a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>&gt; A("<i>Alias Analysis</i>");
1298</pre></div>
1299
1300<p>Once the analysis is registered, passes can declare that they are valid
1301implementations of the interface by using the following code:</p>
1302
1303<div class="doc_code"><pre>
1304<b>namespace</b> {
1305 //<i> Analysis Group implementations <b>must</b> be registered normally...</i>
1306 RegisterPass&lt;FancyAA&gt;
1307 B("<i>somefancyaa</i>", "<i>A more complex alias analysis implementation</i>");
1308
1309 //<i> Declare that we implement the AliasAnalysis interface</i>
1310 RegisterAnalysisGroup&lt;<a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>&gt; C(B);
1311}
1312</pre></div>
1313
1314<p>This just shows a class <tt>FancyAA</tt> that is registered normally, then
1315uses the <tt>RegisterAnalysisGroup</tt> template to "join" the <tt><a
1316href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a></tt>
1317analysis group. Every implementation of an analysis group should join using
1318this template. A single pass may join multiple different analysis groups with
1319no problem.</p>
1320
1321<div class="doc_code"><pre>
1322<b>namespace</b> {
1323 //<i> Analysis Group implementations <b>must</b> be registered normally...</i>
1324 RegisterPass&lt;<a href="http://llvm.org/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a>&gt;
1325 D("<i>basicaa</i>", "<i>Basic Alias Analysis (default AA impl)</i>");
1326
1327 //<i> Declare that we implement the AliasAnalysis interface</i>
1328 RegisterAnalysisGroup&lt;<a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>, <b>true</b>&gt; E(D);
1329}
1330</pre></div>
1331
1332<p>Here we show how the default implementation is specified (using the extra
1333argument to the <tt>RegisterAnalysisGroup</tt> template). There must be exactly
1334one default implementation available at all times for an Analysis Group to be
1335used. Only default implementation can derive from <tt>ImmutablePass</tt>.
1336Here we declare that the
1337 <tt><a href="http://llvm.org/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a></tt>
1338pass is the default implementation for the interface.</p>
1339
1340</div>
1341
1342<!-- *********************************************************************** -->
1343<div class="doc_section">
1344 <a name="passStatistics">Pass Statistics</a>
1345</div>
1346<!-- *********************************************************************** -->
1347
1348<div class="doc_text">
1349<p>The <a
1350href="http://llvm.org/doxygen/Statistic_8h-source.html"><tt>Statistic</tt></a>
1351class is designed to be an easy way to expose various success
1352metrics from passes. These statistics are printed at the end of a
1353run, when the -stats command line option is enabled on the command
1354line. See the <a href="http://llvm.org/docs/ProgrammersManual.html#Statistic">Statistics section</a> in the Programmer's Manual for details.
1355
1356</div>
1357
1358
1359<!-- *********************************************************************** -->
1360<div class="doc_section">
1361 <a name="passmanager">What PassManager does</a>
1362</div>
1363<!-- *********************************************************************** -->
1364
1365<div class="doc_text">
1366
1367<p>The <a
1368href="http://llvm.org/doxygen/PassManager_8h-source.html"><tt>PassManager</tt></a>
1369<a
1370href="http://llvm.org/doxygen/classllvm_1_1PassManager.html">class</a>
1371takes a list of passes, ensures their <a href="#interaction">prerequisites</a>
1372are set up correctly, and then schedules passes to run efficiently. All of the
1373LLVM tools that run passes use the <tt>PassManager</tt> for execution of these
1374passes.</p>
1375
1376<p>The <tt>PassManager</tt> does two main things to try to reduce the execution
1377time of a series of passes:</p>
1378
1379<ol>
1380<li><b>Share analysis results</b> - The PassManager attempts to avoid
1381recomputing analysis results as much as possible. This means keeping track of
1382which analyses are available already, which analyses get invalidated, and which
1383analyses are needed to be run for a pass. An important part of work is that the
1384<tt>PassManager</tt> tracks the exact lifetime of all analysis results, allowing
1385it to <a href="#releaseMemory">free memory</a> allocated to holding analysis
1386results as soon as they are no longer needed.</li>
1387
1388<li><b>Pipeline the execution of passes on the program</b> - The
1389<tt>PassManager</tt> attempts to get better cache and memory usage behavior out
1390of a series of passes by pipelining the passes together. This means that, given
1391a series of consequtive <a href="#FunctionPass"><tt>FunctionPass</tt></a>'s, it
1392will execute all of the <a href="#FunctionPass"><tt>FunctionPass</tt></a>'s on
1393the first function, then all of the <a
1394href="#FunctionPass"><tt>FunctionPass</tt></a>es on the second function,
1395etc... until the entire program has been run through the passes.
1396
1397<p>This improves the cache behavior of the compiler, because it is only touching
1398the LLVM program representation for a single function at a time, instead of
1399traversing the entire program. It reduces the memory consumption of compiler,
1400because, for example, only one <a
1401href="http://llvm.org/doxygen/classllvm_1_1DominatorSet.html"><tt>DominatorSet</tt></a>
John Criswell8a726152007-12-10 20:26:29 +0000[diff] [blame]1402needs to be calculated at a time. This also makes it possible to implement
1403some <a
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1404href="#SMP">interesting enhancements</a> in the future.</p></li>
1405
1406</ol>
1407
1408<p>The effectiveness of the <tt>PassManager</tt> is influenced directly by how
1409much information it has about the behaviors of the passes it is scheduling. For
1410example, the "preserved" set is intentionally conservative in the face of an
1411unimplemented <a href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> method.
1412Not implementing when it should be implemented will have the effect of not
1413allowing any analysis results to live across the execution of your pass.</p>
1414
1415<p>The <tt>PassManager</tt> class exposes a <tt>--debug-pass</tt> command line
1416options that is useful for debugging pass execution, seeing how things work, and
1417diagnosing when you should be preserving more analyses than you currently are
1418(To get information about all of the variants of the <tt>--debug-pass</tt>
Duncan Sandsc5c38f02010-02-18 14:08:13 +0000[diff] [blame]1419option, just type '<tt>opt -help-hidden</tt>').</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1420
1421<p>By using the <tt>--debug-pass=Structure</tt> option, for example, we can see
1422how our <a href="#basiccode">Hello World</a> pass interacts with other passes.
1423Lets try it out with the <tt>gcse</tt> and <tt>licm</tt> passes:</p>
1424
1425<div class="doc_code"><pre>
1426$ opt -load ../../../Debug/lib/Hello.so -gcse -licm --debug-pass=Structure &lt; hello.bc &gt; /dev/null
1427Module Pass Manager
1428 Function Pass Manager
1429 Dominator Set Construction
1430 Immediate Dominators Construction
1431 Global Common Subexpression Elimination
1432-- Immediate Dominators Construction
1433-- Global Common Subexpression Elimination
1434 Natural Loop Construction
1435 Loop Invariant Code Motion
1436-- Natural Loop Construction
1437-- Loop Invariant Code Motion
1438 Module Verifier
1439-- Dominator Set Construction
1440-- Module Verifier
1441 Bitcode Writer
1442--Bitcode Writer
1443</pre></div>
1444
1445<p>This output shows us when passes are constructed and when the analysis
1446results are known to be dead (prefixed with '<tt>--</tt>'). Here we see that
1447GCSE uses dominator and immediate dominator information to do its job. The LICM
1448pass uses natural loop information, which uses dominator sets, but not immediate
1449dominators. Because immediate dominators are no longer useful after the GCSE
1450pass, it is immediately destroyed. The dominator sets are then reused to
1451compute natural loop information, which is then used by the LICM pass.</p>
1452
1453<p>After the LICM pass, the module verifier runs (which is automatically added
1454by the '<tt>opt</tt>' tool), which uses the dominator set to check that the
1455resultant LLVM code is well formed. After it finishes, the dominator set
1456information is destroyed, after being computed once, and shared by three
1457passes.</p>
1458
1459<p>Lets see how this changes when we run the <a href="#basiccode">Hello
1460World</a> pass in between the two passes:</p>
1461
1462<div class="doc_code"><pre>
1463$ opt -load ../../../Debug/lib/Hello.so -gcse -hello -licm --debug-pass=Structure &lt; hello.bc &gt; /dev/null
1464Module Pass Manager
1465 Function Pass Manager
1466 Dominator Set Construction
1467 Immediate Dominators Construction
1468 Global Common Subexpression Elimination
1469<b>-- Dominator Set Construction</b>
1470-- Immediate Dominators Construction
1471-- Global Common Subexpression Elimination
1472<b> Hello World Pass
1473-- Hello World Pass
1474 Dominator Set Construction</b>
1475 Natural Loop Construction
1476 Loop Invariant Code Motion
1477-- Natural Loop Construction
1478-- Loop Invariant Code Motion
1479 Module Verifier
1480-- Dominator Set Construction
1481-- Module Verifier
1482 Bitcode Writer
1483--Bitcode Writer
1484Hello: __main
1485Hello: puts
1486Hello: main
1487</pre></div>
1488
1489<p>Here we see that the <a href="#basiccode">Hello World</a> pass has killed the
1490Dominator Set pass, even though it doesn't modify the code at all! To fix this,
1491we need to add the following <a
1492href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> method to our pass:</p>
1493
1494<div class="doc_code"><pre>
1495 <i>// We don't modify the program, so we preserve all analyses</i>
1496 <b>virtual void</b> getAnalysisUsage(AnalysisUsage &amp;AU) <b>const</b> {
1497 AU.setPreservesAll();
1498 }
1499</pre></div>
1500
1501<p>Now when we run our pass, we get this output:</p>
1502
1503<div class="doc_code"><pre>
1504$ opt -load ../../../Debug/lib/Hello.so -gcse -hello -licm --debug-pass=Structure &lt; hello.bc &gt; /dev/null
1505Pass Arguments: -gcse -hello -licm
1506Module Pass Manager
1507 Function Pass Manager
1508 Dominator Set Construction
1509 Immediate Dominators Construction
1510 Global Common Subexpression Elimination
1511-- Immediate Dominators Construction
1512-- Global Common Subexpression Elimination
1513 Hello World Pass
1514-- Hello World Pass
1515 Natural Loop Construction
1516 Loop Invariant Code Motion
1517-- Loop Invariant Code Motion
1518-- Natural Loop Construction
1519 Module Verifier
1520-- Dominator Set Construction
1521-- Module Verifier
1522 Bitcode Writer
1523--Bitcode Writer
1524Hello: __main
1525Hello: puts
1526Hello: main
1527</pre></div>
1528
1529<p>Which shows that we don't accidentally invalidate dominator information
1530anymore, and therefore do not have to compute it twice.</p>
1531
1532</div>
1533
1534<!-- _______________________________________________________________________ -->
1535<div class="doc_subsubsection">
1536 <a name="releaseMemory">The <tt>releaseMemory</tt> method</a>
1537</div>
1538
1539<div class="doc_text">
1540
1541<div class="doc_code"><pre>
1542 <b>virtual void</b> releaseMemory();
1543</pre></div>
1544
1545<p>The <tt>PassManager</tt> automatically determines when to compute analysis
1546results, and how long to keep them around for. Because the lifetime of the pass
1547object itself is effectively the entire duration of the compilation process, we
1548need some way to free analysis results when they are no longer useful. The
1549<tt>releaseMemory</tt> virtual method is the way to do this.</p>
1550
1551<p>If you are writing an analysis or any other pass that retains a significant
1552amount of state (for use by another pass which "requires" your pass and uses the
1553<a href="#getAnalysis">getAnalysis</a> method) you should implement
Dan Gohmand72d8ea2009-06-15 18:22:49 +0000[diff] [blame]1554<tt>releaseMemory</tt> to, well, release the memory allocated to maintain this
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1555internal state. This method is called after the <tt>run*</tt> method for the
1556class, before the next call of <tt>run*</tt> in your pass.</p>
1557
1558</div>
1559
1560<!-- *********************************************************************** -->
1561<div class="doc_section">
1562 <a name="registering">Registering dynamically loaded passes</a>
1563</div>
1564<!-- *********************************************************************** -->
1565
1566<div class="doc_text">
1567
1568<p><i>Size matters</i> when constructing production quality tools using llvm,
1569both for the purposes of distribution, and for regulating the resident code size
1570when running on the target system. Therefore, it becomes desirable to
1571selectively use some passes, while omitting others and maintain the flexibility
1572to change configurations later on. You want to be able to do all this, and,
1573provide feedback to the user. This is where pass registration comes into
1574play.</p>
1575
1576<p>The fundamental mechanisms for pass registration are the
1577<tt>MachinePassRegistry</tt> class and subclasses of
1578<tt>MachinePassRegistryNode</tt>.</p>
1579
1580<p>An instance of <tt>MachinePassRegistry</tt> is used to maintain a list of
1581<tt>MachinePassRegistryNode</tt> objects. This instance maintains the list and
1582communicates additions and deletions to the command line interface.</p>
1583
1584<p>An instance of <tt>MachinePassRegistryNode</tt> subclass is used to maintain
1585information provided about a particular pass. This information includes the
1586command line name, the command help string and the address of the function used
1587to create an instance of the pass. A global static constructor of one of these
1588instances <i>registers</i> with a corresponding <tt>MachinePassRegistry</tt>,
1589the static destructor <i>unregisters</i>. Thus a pass that is statically linked
1590in the tool will be registered at start up. A dynamically loaded pass will
1591register on load and unregister at unload.</p>
1592
1593</div>
1594
1595<!-- _______________________________________________________________________ -->
1596<div class="doc_subsection">
1597 <a name="registering_existing">Using existing registries</a>
1598</div>
1599
1600<div class="doc_text">
1601
1602<p>There are predefined registries to track instruction scheduling
1603(<tt>RegisterScheduler</tt>) and register allocation (<tt>RegisterRegAlloc</tt>)
1604machine passes. Here we will describe how to <i>register</i> a register
1605allocator machine pass.</p>
1606
1607<p>Implement your register allocator machine pass. In your register allocator
1608.cpp file add the following include;</p>
1609
1610<div class="doc_code"><pre>
1611 #include "llvm/CodeGen/RegAllocRegistry.h"
1612</pre></div>
1613
1614<p>Also in your register allocator .cpp file, define a creator function in the
1615form; </p>
1616
1617<div class="doc_code"><pre>
1618 FunctionPass *createMyRegisterAllocator() {
1619 return new MyRegisterAllocator();
1620 }
1621</pre></div>
1622
1623<p>Note that the signature of this function should match the type of
1624<tt>RegisterRegAlloc::FunctionPassCtor</tt>. In the same file add the
1625"installing" declaration, in the form;</p>
1626
1627<div class="doc_code"><pre>
1628 static RegisterRegAlloc myRegAlloc("myregalloc",
1629 " my register allocator help string",
1630 createMyRegisterAllocator);
1631</pre></div>
1632
1633<p>Note the two spaces prior to the help string produces a tidy result on the
Duncan Sandsc5c38f02010-02-18 14:08:13 +0000[diff] [blame]1634-help query.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1635
1636<div class="doc_code"><pre>
Duncan Sandsc5c38f02010-02-18 14:08:13 +0000[diff] [blame]1637$ llc -help
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1638 ...
Duncan Sands82400092010-02-18 14:37:52 +0000[diff] [blame]1639 -regalloc - Register allocator to use (default=linearscan)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1640 =linearscan - linear scan register allocator
1641 =local - local register allocator
1642 =simple - simple register allocator
1643 =myregalloc - my register allocator help string
1644 ...
1645</pre></div>
1646
1647<p>And that's it. The user is now free to use <tt>-regalloc=myregalloc</tt> as
1648an option. Registering instruction schedulers is similar except use the
1649<tt>RegisterScheduler</tt> class. Note that the
1650<tt>RegisterScheduler::FunctionPassCtor</tt> is significantly different from
1651<tt>RegisterRegAlloc::FunctionPassCtor</tt>.</p>
1652
1653<p>To force the load/linking of your register allocator into the llc/lli tools,
1654add your creator function's global declaration to "Passes.h" and add a "pseudo"
1655call line to <tt>llvm/Codegen/LinkAllCodegenComponents.h</tt>.</p>
1656
1657</div>
1658
1659
1660<!-- _______________________________________________________________________ -->
1661<div class="doc_subsection">
1662 <a name="registering_new">Creating new registries</a>
1663</div>
1664
1665<div class="doc_text">
1666
1667<p>The easiest way to get started is to clone one of the existing registries; we
1668recommend <tt>llvm/CodeGen/RegAllocRegistry.h</tt>. The key things to modify
1669are the class name and the <tt>FunctionPassCtor</tt> type.</p>
1670
1671<p>Then you need to declare the registry. Example: if your pass registry is
1672<tt>RegisterMyPasses</tt> then define;</p>
1673
1674<div class="doc_code"><pre>
1675MachinePassRegistry RegisterMyPasses::Registry;
1676</pre></div>
1677
1678<p>And finally, declare the command line option for your passes. Example:</p>
1679
1680<div class="doc_code"><pre>
1681 cl::opt&lt;RegisterMyPasses::FunctionPassCtor, false,
Dan Gohman8e58bc52008-10-14 17:00:38 +0000[diff] [blame]1682 RegisterPassParser&lt;RegisterMyPasses&gt; &gt;
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1683 MyPassOpt("mypass",
1684 cl::init(&amp;createDefaultMyPass),
1685 cl::desc("my pass option help"));
1686</pre></div>
1687
1688<p>Here the command option is "mypass", with createDefaultMyPass as the default
1689creator.</p>
1690
1691</div>
1692
1693<!-- *********************************************************************** -->
1694<div class="doc_section">
1695 <a name="debughints">Using GDB with dynamically loaded passes</a>
1696</div>
1697<!-- *********************************************************************** -->
1698
1699<div class="doc_text">
1700
1701<p>Unfortunately, using GDB with dynamically loaded passes is not as easy as it
1702should be. First of all, you can't set a breakpoint in a shared object that has
1703not been loaded yet, and second of all there are problems with inlined functions
1704in shared objects. Here are some suggestions to debugging your pass with
1705GDB.</p>
1706
1707<p>For sake of discussion, I'm going to assume that you are debugging a
1708transformation invoked by <tt>opt</tt>, although nothing described here depends
1709on that.</p>
1710
1711</div>
1712
1713<!-- _______________________________________________________________________ -->
1714<div class="doc_subsubsection">
1715 <a name="breakpoint">Setting a breakpoint in your pass</a>
1716</div>
1717
1718<div class="doc_text">
1719
1720<p>First thing you do is start <tt>gdb</tt> on the <tt>opt</tt> process:</p>
1721
1722<div class="doc_code"><pre>
1723$ <b>gdb opt</b>
1724GNU gdb 5.0
1725Copyright 2000 Free Software Foundation, Inc.
1726GDB is free software, covered by the GNU General Public License, and you are
1727welcome to change it and/or distribute copies of it under certain conditions.
1728Type "show copying" to see the conditions.
1729There is absolutely no warranty for GDB. Type "show warranty" for details.
1730This GDB was configured as "sparc-sun-solaris2.6"...
1731(gdb)
1732</pre></div>
1733
1734<p>Note that <tt>opt</tt> has a lot of debugging information in it, so it takes
1735time to load. Be patient. Since we cannot set a breakpoint in our pass yet
1736(the shared object isn't loaded until runtime), we must execute the process, and
1737have it stop before it invokes our pass, but after it has loaded the shared
1738object. The most foolproof way of doing this is to set a breakpoint in
1739<tt>PassManager::run</tt> and then run the process with the arguments you
1740want:</p>
1741
1742<div class="doc_code"><pre>
1743(gdb) <b>break llvm::PassManager::run</b>
1744Breakpoint 1 at 0x2413bc: file Pass.cpp, line 70.
1745(gdb) <b>run test.bc -load $(LLVMTOP)/llvm/Debug/lib/[libname].so -[passoption]</b>
1746Starting program: opt test.bc -load $(LLVMTOP)/llvm/Debug/lib/[libname].so -[passoption]
1747Breakpoint 1, PassManager::run (this=0xffbef174, M=@0x70b298) at Pass.cpp:70
174870 bool PassManager::run(Module &amp;M) { return PM-&gt;run(M); }
1749(gdb)
1750</pre></div>
1751
1752<p>Once the <tt>opt</tt> stops in the <tt>PassManager::run</tt> method you are
1753now free to set breakpoints in your pass so that you can trace through execution
1754or do other standard debugging stuff.</p>
1755
1756</div>
1757
1758<!-- _______________________________________________________________________ -->
1759<div class="doc_subsubsection">
1760 <a name="debugmisc">Miscellaneous Problems</a>
1761</div>
1762
1763<div class="doc_text">
1764
1765<p>Once you have the basics down, there are a couple of problems that GDB has,
1766some with solutions, some without.</p>
1767
1768<ul>
1769<li>Inline functions have bogus stack information. In general, GDB does a
1770pretty good job getting stack traces and stepping through inline functions.
1771When a pass is dynamically loaded however, it somehow completely loses this
1772capability. The only solution I know of is to de-inline a function (move it
1773from the body of a class to a .cpp file).</li>
1774
1775<li>Restarting the program breaks breakpoints. After following the information
1776above, you have succeeded in getting some breakpoints planted in your pass. Nex
1777thing you know, you restart the program (i.e., you type '<tt>run</tt>' again),
1778and you start getting errors about breakpoints being unsettable. The only way I
1779have found to "fix" this problem is to <tt>delete</tt> the breakpoints that are
1780already set in your pass, run the program, and re-set the breakpoints once
1781execution stops in <tt>PassManager::run</tt>.</li>
1782
1783</ul>
1784
1785<p>Hopefully these tips will help with common case debugging situations. If
1786you'd like to contribute some tips of your own, just contact <a
1787href="mailto:sabre@nondot.org">Chris</a>.</p>
1788
1789</div>
1790
1791<!-- *********************************************************************** -->
1792<div class="doc_section">
1793 <a name="future">Future extensions planned</a>
1794</div>
1795<!-- *********************************************************************** -->
1796
1797<div class="doc_text">
1798
1799<p>Although the LLVM Pass Infrastructure is very capable as it stands, and does
1800some nifty stuff, there are things we'd like to add in the future. Here is
1801where we are going:</p>
1802
1803</div>
1804
1805<!-- _______________________________________________________________________ -->
1806<div class="doc_subsubsection">
1807 <a name="SMP">Multithreaded LLVM</a>
1808</div>
1809
1810<div class="doc_text">
1811
1812<p>Multiple CPU machines are becoming more common and compilation can never be
1813fast enough: obviously we should allow for a multithreaded compiler. Because of
1814the semantics defined for passes above (specifically they cannot maintain state
1815across invocations of their <tt>run*</tt> methods), a nice clean way to
1816implement a multithreaded compiler would be for the <tt>PassManager</tt> class
1817to create multiple instances of each pass object, and allow the separate
1818instances to be hacking on different parts of the program at the same time.</p>
1819
1820<p>This implementation would prevent each of the passes from having to implement
1821multithreaded constructs, requiring only the LLVM core to have locking in a few
1822places (for global resources). Although this is a simple extension, we simply
1823haven't had time (or multiprocessor machines, thus a reason) to implement this.
1824Despite that, we have kept the LLVM passes SMP ready, and you should too.</p>
1825
1826</div>
1827
1828<!-- *********************************************************************** -->
1829<hr>
1830<address>
1831 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
Misha Brukman947321d2008-12-11 17:34:48 +0000[diff] [blame]1832 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1833 <a href="http://validator.w3.org/check/referer"><img
Misha Brukman5c1cc642008-12-11 18:23:24 +0000[diff] [blame]1834 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1835
1836 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
1837 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
1838 Last modified: $Date$
1839</address>
1840
1841</body>
1842</html>