1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
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">
11 <div class="doc_title">
16 <li><a href="#introduction">Introduction - What is a pass?</a></li>
17 <li><a href="#quickstart">Quick Start - Writing hello world</a>
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>
23 <li><a href="#passtype">Pass classes and requirements</a>
25 <li><a href="#ImmutablePass">The <tt>ImmutablePass</tt> class</a></li>
26 <li><a href="#ModulePass">The <tt>ModulePass</tt> class</a>
28 <li><a href="#runOnModule">The <tt>runOnModule</tt> method</a></li>
30 <li><a href="#CallGraphSCCPass">The <tt>CallGraphSCCPass</tt> class</a>
32 <li><a href="#doInitialization_scc">The <tt>doInitialization(CallGraph
33 &)</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 &)</tt> method</a></li>
38 <li><a href="#FunctionPass">The <tt>FunctionPass</tt> class</a>
40 <li><a href="#doInitialization_mod">The <tt>doInitialization(Module
41 &)</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 &)</tt> method</a></li>
46 <li><a href="#LoopPass">The <tt>LoopPass</tt> class</a>
48 <li><a href="#doInitialization_loop">The <tt>doInitialization(Loop *,
49 LPPassManager &)</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()
54 <li><a href="#BasicBlockPass">The <tt>BasicBlockPass</tt> class</a>
56 <li><a href="#doInitialization_fn">The <tt>doInitialization(Function
57 &)</tt> method</a></li>
58 <li><a href="#runOnBasicBlock">The <tt>runOnBasicBlock</tt>
60 <li><a href="#doFinalization_fn">The <tt>doFinalization(Function
61 &)</tt> method</a></li>
63 <li><a href="#MachineFunctionPass">The <tt>MachineFunctionPass</tt>
66 <li><a href="#runOnMachineFunction">The
67 <tt>runOnMachineFunction(MachineFunction &)</tt> method</a></li>
70 <li><a href="#registration">Pass Registration</a>
72 <li><a href="#print">The <tt>print</tt> method</a></li>
74 <li><a href="#interaction">Specifying interactions between passes</a>
76 <li><a href="#getAnalysisUsage">The <tt>getAnalysisUsage</tt>
78 <li><a href="#AU::addRequired">The <tt>AnalysisUsage::addRequired<></tt> and <tt>AnalysisUsage::addRequiredTransitive<></tt> methods</a></li>
79 <li><a href="#AU::addPreserved">The <tt>AnalysisUsage::addPreserved<></tt> method</a></li>
80 <li><a href="#AU::examples">Example implementations of <tt>getAnalysisUsage</tt></a></li>
81 <li><a href="#getAnalysis">The <tt>getAnalysis<></tt> and
82 <tt>getAnalysisIfAvailable<></tt> methods</a></li>
84 <li><a href="#analysisgroup">Implementing Analysis Groups</a>
86 <li><a href="#agconcepts">Analysis Group Concepts</a></li>
87 <li><a href="#registerag">Using <tt>RegisterAnalysisGroup</tt></a></li>
89 <li><a href="#passStatistics">Pass Statistics</a>
90 <li><a href="#passmanager">What PassManager does</a>
92 <li><a href="#releaseMemory">The <tt>releaseMemory</tt> method</a></li>
94 <li><a href="#registering">Registering dynamically loaded passes</a>
96 <li><a href="#registering_existing">Using existing registries</a></li>
97 <li><a href="#registering_new">Creating new registries</a></li>
99 <li><a href="#debughints">Using GDB with dynamically loaded passes</a>
101 <li><a href="#breakpoint">Setting a breakpoint in your pass</a></li>
102 <li><a href="#debugmisc">Miscellaneous Problems</a></li>
104 <li><a href="#future">Future extensions planned</a>
106 <li><a href="#SMP">Multithreaded LLVM</a></li>
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>
115 <!-- *********************************************************************** -->
116 <div class="doc_section">
117 <a name="introduction">Introduction - What is a pass?</a>
119 <!-- *********************************************************************** -->
121 <div class="doc_text">
123 <p>The LLVM Pass Framework is an important part of the LLVM system, because LLVM
124 passes are where most of the interesting parts of the compiler exist. Passes
125 perform the transformations and optimizations that make up the compiler, they
126 build the analysis results that are used by these transformations, and they are,
127 above all, a structuring technique for compiler code.</p>
129 <p>All LLVM passes are subclasses of the <tt><a
130 href="http://llvm.org/doxygen/classllvm_1_1Pass.html">Pass</a></tt>
131 class, which implement functionality by overriding virtual methods inherited
132 from <tt>Pass</tt>. Depending on how your pass works, you should inherit from
133 the <tt><a href="#ModulePass">ModulePass</a></tt>, <tt><a
134 href="#CallGraphSCCPass">CallGraphSCCPass</a></tt>, <tt><a
135 href="#FunctionPass">FunctionPass</a></tt>, or <tt><a
136 href="#LoopPass">LoopPass</a></tt>, or <tt><a
137 href="#BasicBlockPass">BasicBlockPass</a></tt> classes, which gives the system
138 more information about what your pass does, and how it can be combined with
139 other passes. One of the main features of the LLVM Pass Framework is that it
140 schedules passes to run in an efficient way based on the constraints that your
141 pass meets (which are indicated by which class they derive from).</p>
143 <p>We start by showing you how to construct a pass, everything from setting up
144 the code, to compiling, loading, and executing it. After the basics are down,
145 more advanced features are discussed.</p>
149 <!-- *********************************************************************** -->
150 <div class="doc_section">
151 <a name="quickstart">Quick Start - Writing hello world</a>
153 <!-- *********************************************************************** -->
155 <div class="doc_text">
157 <p>Here we describe how to write the "hello world" of passes. The "Hello" pass
158 is designed to simply print out the name of non-external functions that exist in
159 the program being compiled. It does not modify the program at all, it just
160 inspects it. The source code and files for this pass are available in the LLVM
161 source tree in the <tt>lib/Transforms/Hello</tt> directory.</p>
165 <!-- ======================================================================= -->
166 <div class="doc_subsection">
167 <a name="makefile">Setting up the build environment</a>
170 <div class="doc_text">
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>
179 <div class="doc_code"><pre>
180 # Makefile for hello pass
182 # Path to top level of LLVM heirarchy
185 # Name of the library to build
188 # Make the shared library become a loadable module so the tools can
189 # dlopen/dlsym on the resulting library.
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
195 LLVMLIBS = LLVMCore.a LLVMSupport.a LLVMSystem.a
197 # Include the makefile implementation stuff
198 include $(LEVEL)/Makefile.common
201 <p>This makefile specifies that all of the <tt>.cpp</tt> files in the current
202 directory are to be compiled and linked together into a
203 <tt>Debug/lib/Hello.so</tt> shared object that can be dynamically loaded by
204 the <tt>opt</tt> or <tt>bugpoint</tt> tools via their <tt>-load</tt> options.
205 If your operating system uses a suffix other than .so (such as windows or
206 Mac OS/X), the appropriate extension will be used.</p>
208 <p>Now that we have the build scripts set up, we just need to write the code for
213 <!-- ======================================================================= -->
214 <div class="doc_subsection">
215 <a name="basiccode">Basic code required</a>
218 <div class="doc_text">
220 <p>Now that we have a way to compile our new pass, we just have to write it.
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>"
228 <p>Which are needed because we are writing a <tt><a
229 href="http://llvm.org/doxygen/classllvm_1_1Pass.html">Pass</a></tt>, and
230 we are operating on <tt><a
231 href="http://llvm.org/doxygen/classllvm_1_1Function.html">Function</a></tt>'s.</p>
234 <div class="doc_code"><pre>
235 <b>using namespace llvm;</b>
237 <p>... which is required because the functions from the include files
238 live in the llvm namespace.
243 <div class="doc_code"><pre>
247 <p>... which starts out an anonymous namespace. Anonymous namespaces are to C++
248 what the "<tt>static</tt>" keyword is to C (at global scope). It makes the
249 things declared inside of the anonymous namespace only visible to the current
250 file. If you're not familiar with them, consult a decent C++ book for more
253 <p>Next, we declare our pass itself:</p>
255 <div class="doc_code"><pre>
256 <b>struct</b> Hello : <b>public</b> <a href="#FunctionPass">FunctionPass</a> {
259 <p>This declares a "<tt>Hello</tt>" class that is a subclass of <tt><a
260 href="http://llvm.org/doxygen/classllvm_1_1FunctionPass.html">FunctionPass</a></tt>.
261 The different builtin pass subclasses are described in detail <a
262 href="#passtype">later</a>, but for now, know that <a
263 href="#FunctionPass"><tt>FunctionPass</tt></a>'s operate a function at a
266 <div class="doc_code"><pre>
268 Hello() : FunctionPass(&ID) {}
271 <p> This declares pass identifier used by LLVM to identify pass. This allows LLVM to
272 avoid using expensive C++ runtime information.</p>
274 <div class="doc_code"><pre>
275 <b>virtual bool</b> <a href="#runOnFunction">runOnFunction</a>(Function &F) {
276 llvm::cerr << "<i>Hello: </i>" << F.getName() << "\n";
279 }; <i>// end of struct Hello</i>
282 <p>We declare a "<a href="#runOnFunction"><tt>runOnFunction</tt></a>" method,
283 which overloads an abstract virtual method inherited from <a
284 href="#FunctionPass"><tt>FunctionPass</tt></a>. This is where we are supposed
285 to do our thing, so we just print out our message with the name of each
288 <div class="doc_code"><pre>
292 <p> We initialize pass ID here. LLVM uses ID's address to identify pass so
293 initialization value is not important.</p>
295 <div class="doc_code"><pre>
296 RegisterPass<Hello> X("<i>hello</i>", "<i>Hello World Pass</i>",
297 false /* Only looks at CFG */,
298 false /* Analysis Pass */);
299 } <i>// end of anonymous namespace</i>
302 <p>Lastly, we <a href="#registration">register our class</a> <tt>Hello</tt>,
303 giving it a command line
304 argument "<tt>hello</tt>", and a name "<tt>Hello World Pass</tt>".
305 Last two RegisterPass arguments are optional. Their default value is false.
306 If a pass walks CFG without modifying it then third argument is set to true.
307 If a pass is an analysis pass, for example dominator tree pass, then true
308 is supplied as fourth argument. </p>
310 <p>As a whole, the <tt>.cpp</tt> file looks like:</p>
312 <div class="doc_code"><pre>
313 <b>#include</b> "<a href="http://llvm.org/doxygen/Pass_8h-source.html">llvm/Pass.h</a>"
314 <b>#include</b> "<a href="http://llvm.org/doxygen/Function_8h-source.html">llvm/Function.h</a>"
316 <b>using namespace llvm;</b>
319 <b>struct Hello</b> : <b>public</b> <a href="#FunctionPass">FunctionPass</a> {
322 Hello() : FunctionPass(&ID) {}
324 <b>virtual bool</b> <a href="#runOnFunction">runOnFunction</a>(Function &F) {
325 llvm::cerr << "<i>Hello: </i>" << F.getName() << "\n";
331 RegisterPass<Hello> X("<i>hello</i>", "<i>Hello World Pass</i>");
335 <p>Now that it's all together, compile the file with a simple "<tt>gmake</tt>"
336 command in the local directory and you should get a new
337 "<tt>Debug/lib/Hello.so</tt> file. Note that everything in this file is
338 contained in an anonymous namespace: this reflects the fact that passes are self
339 contained units that do not need external interfaces (although they can have
340 them) to be useful.</p>
344 <!-- ======================================================================= -->
345 <div class="doc_subsection">
346 <a name="running">Running a pass with <tt>opt</tt></a>
349 <div class="doc_text">
351 <p>Now that you have a brand new shiny shared object file, we can use the
352 <tt>opt</tt> command to run an LLVM program through your pass. Because you
353 registered your pass with the <tt>RegisterPass</tt> template, you will be able to
354 use the <tt>opt</tt> tool to access it, once loaded.</p>
356 <p>To test it, follow the example at the end of the <a
357 href="GettingStarted.html">Getting Started Guide</a> to compile "Hello World" to
358 LLVM. We can now run the bitcode file (<tt>hello.bc</tt>) for the program
359 through our transformation like this (or course, any bitcode file will
362 <div class="doc_code"><pre>
363 $ opt -load ../../../Debug/lib/Hello.so -hello < hello.bc > /dev/null
369 <p>The '<tt>-load</tt>' option specifies that '<tt>opt</tt>' should load your
370 pass as a shared object, which makes '<tt>-hello</tt>' a valid command line
371 argument (which is one reason you need to <a href="#registration">register your
372 pass</a>). Because the hello pass does not modify the program in any
373 interesting way, we just throw away the result of <tt>opt</tt> (sending it to
374 <tt>/dev/null</tt>).</p>
376 <p>To see what happened to the other string you registered, try running
377 <tt>opt</tt> with the <tt>--help</tt> option:</p>
379 <div class="doc_code"><pre>
380 $ opt -load ../../../Debug/lib/Hello.so --help
381 OVERVIEW: llvm .bc -> .bc modular optimizer
383 USAGE: opt [options] <input bitcode>
386 Optimizations available:
388 -funcresolve - Resolve Functions
389 -gcse - Global Common Subexpression Elimination
390 -globaldce - Dead Global Elimination
391 <b>-hello - Hello World Pass</b>
392 -indvars - Canonicalize Induction Variables
393 -inline - Function Integration/Inlining
394 -instcombine - Combine redundant instructions
398 <p>The pass name get added as the information string for your pass, giving some
399 documentation to users of <tt>opt</tt>. Now that you have a working pass, you
400 would go ahead and make it do the cool transformations you want. Once you get
401 it all working and tested, it may become useful to find out how fast your pass
402 is. The <a href="#passManager"><tt>PassManager</tt></a> provides a nice command
403 line option (<tt>--time-passes</tt>) that allows you to get information about
404 the execution time of your pass along with the other passes you queue up. For
407 <div class="doc_code"><pre>
408 $ opt -load ../../../Debug/lib/Hello.so -hello -time-passes < hello.bc > /dev/null
412 ===============================================================================
413 ... Pass execution timing report ...
414 ===============================================================================
415 Total Execution Time: 0.02 seconds (0.0479059 wall clock)
417 ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Pass Name ---
418 0.0100 (100.0%) 0.0000 ( 0.0%) 0.0100 ( 50.0%) 0.0402 ( 84.0%) Bitcode Writer
419 0.0000 ( 0.0%) 0.0100 (100.0%) 0.0100 ( 50.0%) 0.0031 ( 6.4%) Dominator Set Construction
420 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0013 ( 2.7%) Module Verifier
421 <b> 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0033 ( 6.9%) Hello World Pass</b>
422 0.0100 (100.0%) 0.0100 (100.0%) 0.0200 (100.0%) 0.0479 (100.0%) TOTAL
425 <p>As you can see, our implementation above is pretty fast :). The additional
426 passes listed are automatically inserted by the '<tt>opt</tt>' tool to verify
427 that the LLVM emitted by your pass is still valid and well formed LLVM, which
428 hasn't been broken somehow.</p>
430 <p>Now that you have seen the basics of the mechanics behind passes, we can talk
431 about some more details of how they work and how to use them.</p>
435 <!-- *********************************************************************** -->
436 <div class="doc_section">
437 <a name="passtype">Pass classes and requirements</a>
439 <!-- *********************************************************************** -->
441 <div class="doc_text">
443 <p>One of the first things that you should do when designing a new pass is to
444 decide what class you should subclass for your pass. The <a
445 href="#basiccode">Hello World</a> example uses the <tt><a
446 href="#FunctionPass">FunctionPass</a></tt> class for its implementation, but we
447 did not discuss why or when this should occur. Here we talk about the classes
448 available, from the most general to the most specific.</p>
450 <p>When choosing a superclass for your Pass, you should choose the <b>most
451 specific</b> class possible, while still being able to meet the requirements
452 listed. This gives the LLVM Pass Infrastructure information necessary to
453 optimize how passes are run, so that the resultant compiler isn't unneccesarily
458 <!-- ======================================================================= -->
459 <div class="doc_subsection">
460 <a name="ImmutablePass">The <tt>ImmutablePass</tt> class</a>
463 <div class="doc_text">
465 <p>The most plain and boring type of pass is the "<tt><a
466 href="http://llvm.org/doxygen/classllvm_1_1ImmutablePass.html">ImmutablePass</a></tt>"
467 class. This pass type is used for passes that do not have to be run, do not
468 change state, and never need to be updated. This is not a normal type of
469 transformation or analysis, but can provide information about the current
470 compiler configuration.</p>
472 <p>Although this pass class is very infrequently used, it is important for
473 providing information about the current target machine being compiled for, and
474 other static information that can affect the various transformations.</p>
476 <p><tt>ImmutablePass</tt>es never invalidate other transformations, are never
477 invalidated, and are never "run".</p>
481 <!-- ======================================================================= -->
482 <div class="doc_subsection">
483 <a name="ModulePass">The <tt>ModulePass</tt> class</a>
486 <div class="doc_text">
489 href="http://llvm.org/doxygen/classllvm_1_1ModulePass.html">ModulePass</a></tt>"
490 class is the most general of all superclasses that you can use. Deriving from
491 <tt>ModulePass</tt> indicates that your pass uses the entire program as a unit,
492 refering to function bodies in no predictable order, or adding and removing
493 functions. Because nothing is known about the behavior of <tt>ModulePass</tt>
494 subclasses, no optimization can be done for their execution. A module pass
495 can use function level passes (e.g. dominators) using getAnalysis interface
496 <tt> getAnalysis<DominatorTree>(Function)</tt>, if the function pass
497 does not require any module passes. </p>
499 <p>To write a correct <tt>ModulePass</tt> subclass, derive from
500 <tt>ModulePass</tt> and overload the <tt>runOnModule</tt> method with the
501 following signature:</p>
505 <!-- _______________________________________________________________________ -->
506 <div class="doc_subsubsection">
507 <a name="runOnModule">The <tt>runOnModule</tt> method</a>
510 <div class="doc_text">
512 <div class="doc_code"><pre>
513 <b>virtual bool</b> runOnModule(Module &M) = 0;
516 <p>The <tt>runOnModule</tt> method performs the interesting work of the pass.
517 It should return true if the module was modified by the transformation and
522 <!-- ======================================================================= -->
523 <div class="doc_subsection">
524 <a name="CallGraphSCCPass">The <tt>CallGraphSCCPass</tt> class</a>
527 <div class="doc_text">
530 href="http://llvm.org/doxygen/classllvm_1_1CallGraphSCCPass.html">CallGraphSCCPass</a></tt>"
531 is used by passes that need to traverse the program bottom-up on the call graph
532 (callees before callers). Deriving from CallGraphSCCPass provides some
533 mechanics for building and traversing the CallGraph, but also allows the system
534 to optimize execution of CallGraphSCCPass's. If your pass meets the
535 requirements outlined below, and doesn't meet the requirements of a <tt><a
536 href="#FunctionPass">FunctionPass</a></tt> or <tt><a
537 href="#BasicBlockPass">BasicBlockPass</a></tt>, you should derive from
538 <tt>CallGraphSCCPass</tt>.</p>
540 <p><b>TODO</b>: explain briefly what SCC, Tarjan's algo, and B-U mean.</p>
542 <p>To be explicit, <tt>CallGraphSCCPass</tt> subclasses are:</p>
546 <li>... <em>not allowed</em> to modify any <tt>Function</tt>s that are not in
547 the current SCC.</li>
549 <li>... <em>not allowed</em> to inspect any Function's other than those in the
550 current SCC and the direct callees of the SCC.</li>
552 <li>... <em>required</em> to preserve the current CallGraph object, updating it
553 to reflect any changes made to the program.</li>
555 <li>... <em>not allowed</em> to add or remove SCC's from the current Module,
556 though they may change the contents of an SCC.</li>
558 <li>... <em>allowed</em> to add or remove global variables from the current
561 <li>... <em>allowed</em> to maintain state across invocations of
562 <a href="#runOnSCC"><tt>runOnSCC</tt></a> (including global data).</li>
565 <p>Implementing a <tt>CallGraphSCCPass</tt> is slightly tricky in some cases
566 because it has to handle SCCs with more than one node in it. All of the virtual
567 methods described below should return true if they modified the program, or
568 false if they didn't.</p>
572 <!-- _______________________________________________________________________ -->
573 <div class="doc_subsubsection">
574 <a name="doInitialization_scc">The <tt>doInitialization(CallGraph &)</tt>
578 <div class="doc_text">
580 <div class="doc_code"><pre>
581 <b>virtual bool</b> doInitialization(CallGraph &CG);
584 <p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
585 <tt>CallGraphSCCPass</tt>'s are not allowed to do. They can add and remove
586 functions, get pointers to functions, etc. The <tt>doInitialization</tt> method
587 is designed to do simple initialization type of stuff that does not depend on
588 the SCCs being processed. The <tt>doInitialization</tt> method call is not
589 scheduled to overlap with any other pass executions (thus it should be very
594 <!-- _______________________________________________________________________ -->
595 <div class="doc_subsubsection">
596 <a name="runOnSCC">The <tt>runOnSCC</tt> method</a>
599 <div class="doc_text">
601 <div class="doc_code"><pre>
602 <b>virtual bool</b> runOnSCC(const std::vector<CallGraphNode *> &SCCM) = 0;
605 <p>The <tt>runOnSCC</tt> method performs the interesting work of the pass, and
606 should return true if the module was modified by the transformation, false
611 <!-- _______________________________________________________________________ -->
612 <div class="doc_subsubsection">
613 <a name="doFinalization_scc">The <tt>doFinalization(CallGraph
614 &)</tt> method</a>
617 <div class="doc_text">
619 <div class="doc_code"><pre>
620 <b>virtual bool</b> doFinalization(CallGraph &CG);
623 <p>The <tt>doFinalization</tt> method is an infrequently used method that is
624 called when the pass framework has finished calling <a
625 href="#runOnFunction"><tt>runOnFunction</tt></a> for every function in the
626 program being compiled.</p>
630 <!-- ======================================================================= -->
631 <div class="doc_subsection">
632 <a name="FunctionPass">The <tt>FunctionPass</tt> class</a>
635 <div class="doc_text">
637 <p>In contrast to <tt>ModulePass</tt> subclasses, <tt><a
638 href="http://llvm.org/doxygen/classllvm_1_1Pass.html">FunctionPass</a></tt>
639 subclasses do have a predictable, local behavior that can be expected by the
640 system. All <tt>FunctionPass</tt> execute on each function in the program
641 independent of all of the other functions in the program.
642 <tt>FunctionPass</tt>'s do not require that they are executed in a particular
643 order, and <tt>FunctionPass</tt>'s do not modify external functions.</p>
645 <p>To be explicit, <tt>FunctionPass</tt> subclasses are not allowed to:</p>
648 <li>Modify a Function other than the one currently being processed.</li>
649 <li>Add or remove Function's from the current Module.</li>
650 <li>Add or remove global variables from the current Module.</li>
651 <li>Maintain state across invocations of
652 <a href="#runOnFunction"><tt>runOnFunction</tt></a> (including global data)</li>
655 <p>Implementing a <tt>FunctionPass</tt> is usually straightforward (See the <a
656 href="#basiccode">Hello World</a> pass for example). <tt>FunctionPass</tt>'s
657 may overload three virtual methods to do their work. All of these methods
658 should return true if they modified the program, or false if they didn't.</p>
662 <!-- _______________________________________________________________________ -->
663 <div class="doc_subsubsection">
664 <a name="doInitialization_mod">The <tt>doInitialization(Module &)</tt>
668 <div class="doc_text">
670 <div class="doc_code"><pre>
671 <b>virtual bool</b> doInitialization(Module &M);
674 <p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
675 <tt>FunctionPass</tt>'s are not allowed to do. They can add and remove
676 functions, get pointers to functions, etc. The <tt>doInitialization</tt> method
677 is designed to do simple initialization type of stuff that does not depend on
678 the functions being processed. The <tt>doInitialization</tt> method call is not
679 scheduled to overlap with any other pass executions (thus it should be very
682 <p>A good example of how this method should be used is the <a
683 href="http://llvm.org/doxygen/LowerAllocations_8cpp-source.html">LowerAllocations</a>
684 pass. This pass converts <tt>malloc</tt> and <tt>free</tt> instructions into
685 platform dependent <tt>malloc()</tt> and <tt>free()</tt> function calls. It
686 uses the <tt>doInitialization</tt> method to get a reference to the malloc and
687 free functions that it needs, adding prototypes to the module if necessary.</p>
691 <!-- _______________________________________________________________________ -->
692 <div class="doc_subsubsection">
693 <a name="runOnFunction">The <tt>runOnFunction</tt> method</a>
696 <div class="doc_text">
698 <div class="doc_code"><pre>
699 <b>virtual bool</b> runOnFunction(Function &F) = 0;
702 <p>The <tt>runOnFunction</tt> method must be implemented by your subclass to do
703 the transformation or analysis work of your pass. As usual, a true value should
704 be returned if the function is modified.</p>
708 <!-- _______________________________________________________________________ -->
709 <div class="doc_subsubsection">
710 <a name="doFinalization_mod">The <tt>doFinalization(Module
711 &)</tt> method</a>
714 <div class="doc_text">
716 <div class="doc_code"><pre>
717 <b>virtual bool</b> doFinalization(Module &M);
720 <p>The <tt>doFinalization</tt> method is an infrequently used method that is
721 called when the pass framework has finished calling <a
722 href="#runOnFunction"><tt>runOnFunction</tt></a> for every function in the
723 program being compiled.</p>
727 <!-- ======================================================================= -->
728 <div class="doc_subsection">
729 <a name="LoopPass">The <tt>LoopPass</tt> class </a>
732 <div class="doc_text">
734 <p> All <tt>LoopPass</tt> execute on each loop in the function independent of
735 all of the other loops in the function. <tt>LoopPass</tt> processes loops in
736 loop nest order such that outer most loop is processed last. </p>
738 <p> <tt>LoopPass</tt> subclasses are allowed to update loop nest using
739 <tt>LPPassManager</tt> interface. Implementing a loop pass is usually
740 straightforward. <tt>Looppass</tt>'s may overload three virtual methods to
741 do their work. All these methods should return true if they modified the
742 program, or false if they didn't. </p>
745 <!-- _______________________________________________________________________ -->
746 <div class="doc_subsubsection">
747 <a name="doInitialization_loop">The <tt>doInitialization(Loop *,
748 LPPassManager &)</tt>
752 <div class="doc_text">
754 <div class="doc_code"><pre>
755 <b>virtual bool</b> doInitialization(Loop *, LPPassManager &LPM);
758 <p>The <tt>doInitialization</tt> method is designed to do simple initialization
759 type of stuff that does not depend on the functions being processed. The
760 <tt>doInitialization</tt> method call is not scheduled to overlap with any
761 other pass executions (thus it should be very fast). LPPassManager
762 interface should be used to access Function or Module level analysis
768 <!-- _______________________________________________________________________ -->
769 <div class="doc_subsubsection">
770 <a name="runOnLoop">The <tt>runOnLoop</tt> method</a>
773 <div class="doc_text">
775 <div class="doc_code"><pre>
776 <b>virtual bool</b> runOnLoop(Loop *, LPPassManager &LPM) = 0;
779 <p>The <tt>runOnLoop</tt> method must be implemented by your subclass to do
780 the transformation or analysis work of your pass. As usual, a true value should
781 be returned if the function is modified. <tt>LPPassManager</tt> interface
782 should be used to update loop nest.</p>
786 <!-- _______________________________________________________________________ -->
787 <div class="doc_subsubsection">
788 <a name="doFinalization_loop">The <tt>doFinalization()</tt> method</a>
791 <div class="doc_text">
793 <div class="doc_code"><pre>
794 <b>virtual bool</b> doFinalization();
797 <p>The <tt>doFinalization</tt> method is an infrequently used method that is
798 called when the pass framework has finished calling <a
799 href="#runOnLoop"><tt>runOnLoop</tt></a> for every loop in the
800 program being compiled. </p>
806 <!-- ======================================================================= -->
807 <div class="doc_subsection">
808 <a name="BasicBlockPass">The <tt>BasicBlockPass</tt> class</a>
811 <div class="doc_text">
813 <p><tt>BasicBlockPass</tt>'s are just like <a
814 href="#FunctionPass"><tt>FunctionPass</tt></a>'s, except that they must limit
815 their scope of inspection and modification to a single basic block at a time.
816 As such, they are <b>not</b> allowed to do any of the following:</p>
819 <li>Modify or inspect any basic blocks outside of the current one</li>
820 <li>Maintain state across invocations of
821 <a href="#runOnBasicBlock"><tt>runOnBasicBlock</tt></a></li>
822 <li>Modify the control flow graph (by altering terminator instructions)</li>
823 <li>Any of the things forbidden for
824 <a href="#FunctionPass"><tt>FunctionPass</tt></a>es.</li>
827 <p><tt>BasicBlockPass</tt>es are useful for traditional local and "peephole"
828 optimizations. They may override the same <a
829 href="#doInitialization_mod"><tt>doInitialization(Module &)</tt></a> and <a
830 href="#doFinalization_mod"><tt>doFinalization(Module &)</tt></a> methods that <a
831 href="#FunctionPass"><tt>FunctionPass</tt></a>'s have, but also have the following virtual methods that may also be implemented:</p>
835 <!-- _______________________________________________________________________ -->
836 <div class="doc_subsubsection">
837 <a name="doInitialization_fn">The <tt>doInitialization(Function
838 &)</tt> method</a>
841 <div class="doc_text">
843 <div class="doc_code"><pre>
844 <b>virtual bool</b> doInitialization(Function &F);
847 <p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
848 <tt>BasicBlockPass</tt>'s are not allowed to do, but that
849 <tt>FunctionPass</tt>'s can. The <tt>doInitialization</tt> method is designed
850 to do simple initialization that does not depend on the
851 BasicBlocks being processed. The <tt>doInitialization</tt> method call is not
852 scheduled to overlap with any other pass executions (thus it should be very
857 <!-- _______________________________________________________________________ -->
858 <div class="doc_subsubsection">
859 <a name="runOnBasicBlock">The <tt>runOnBasicBlock</tt> method</a>
862 <div class="doc_text">
864 <div class="doc_code"><pre>
865 <b>virtual bool</b> runOnBasicBlock(BasicBlock &BB) = 0;
868 <p>Override this function to do the work of the <tt>BasicBlockPass</tt>. This
869 function is not allowed to inspect or modify basic blocks other than the
870 parameter, and are not allowed to modify the CFG. A true value must be returned
871 if the basic block is modified.</p>
875 <!-- _______________________________________________________________________ -->
876 <div class="doc_subsubsection">
877 <a name="doFinalization_fn">The <tt>doFinalization(Function &)</tt>
881 <div class="doc_text">
883 <div class="doc_code"><pre>
884 <b>virtual bool</b> doFinalization(Function &F);
887 <p>The <tt>doFinalization</tt> method is an infrequently used method that is
888 called when the pass framework has finished calling <a
889 href="#runOnBasicBlock"><tt>runOnBasicBlock</tt></a> for every BasicBlock in the
890 program being compiled. This can be used to perform per-function
895 <!-- ======================================================================= -->
896 <div class="doc_subsection">
897 <a name="MachineFunctionPass">The <tt>MachineFunctionPass</tt> class</a>
900 <div class="doc_text">
902 <p>A <tt>MachineFunctionPass</tt> is a part of the LLVM code generator that
903 executes on the machine-dependent representation of each LLVM function in the
904 program. A <tt>MachineFunctionPass</tt> is also a <tt>FunctionPass</tt>, so all
905 the restrictions that apply to a <tt>FunctionPass</tt> also apply to it.
906 <tt>MachineFunctionPass</tt>es also have additional restrictions. In particular,
907 <tt>MachineFunctionPass</tt>es are not allowed to do any of the following:</p>
910 <li>Modify any LLVM Instructions, BasicBlocks or Functions.</li>
911 <li>Modify a MachineFunction other than the one currently being processed.</li>
912 <li>Add or remove MachineFunctions from the current Module.</li>
913 <li>Add or remove global variables from the current Module.</li>
914 <li>Maintain state across invocations of <a
915 href="#runOnMachineFunction"><tt>runOnMachineFunction</tt></a> (including global
921 <!-- _______________________________________________________________________ -->
922 <div class="doc_subsubsection">
923 <a name="runOnMachineFunction">The <tt>runOnMachineFunction(MachineFunction
924 &MF)</tt> method</a>
927 <div class="doc_text">
929 <div class="doc_code"><pre>
930 <b>virtual bool</b> runOnMachineFunction(MachineFunction &MF) = 0;
933 <p><tt>runOnMachineFunction</tt> can be considered the main entry point of a
934 <tt>MachineFunctionPass</tt>; that is, you should override this method to do the
935 work of your <tt>MachineFunctionPass</tt>.</p>
937 <p>The <tt>runOnMachineFunction</tt> method is called on every
938 <tt>MachineFunction</tt> in a <tt>Module</tt>, so that the
939 <tt>MachineFunctionPass</tt> may perform optimizations on the machine-dependent
940 representation of the function. If you want to get at the LLVM <tt>Function</tt>
941 for the <tt>MachineFunction</tt> you're working on, use
942 <tt>MachineFunction</tt>'s <tt>getFunction()</tt> accessor method -- but
943 remember, you may not modify the LLVM <tt>Function</tt> or its contents from a
944 <tt>MachineFunctionPass</tt>.</p>
948 <!-- *********************************************************************** -->
949 <div class="doc_section">
950 <a name="registration">Pass registration</a>
952 <!-- *********************************************************************** -->
954 <div class="doc_text">
956 <p>In the <a href="#basiccode">Hello World</a> example pass we illustrated how
957 pass registration works, and discussed some of the reasons that it is used and
958 what it does. Here we discuss how and why passes are registered.</p>
960 <p>As we saw above, passes are registered with the <b><tt>RegisterPass</tt></b>
961 template, which requires you to pass at least two
962 parameters. The first parameter is the name of the pass that is to be used on
963 the command line to specify that the pass should be added to a program (for
964 example, with <tt>opt</tt> or <tt>bugpoint</tt>). The second argument is the
965 name of the pass, which is to be used for the <tt>--help</tt> output of
967 well as for debug output generated by the <tt>--debug-pass</tt> option.</p>
969 <p>If you want your pass to be easily dumpable, you should
970 implement the virtual <tt>print</tt> method:</p>
974 <!-- _______________________________________________________________________ -->
975 <div class="doc_subsubsection">
976 <a name="print">The <tt>print</tt> method</a>
979 <div class="doc_text">
981 <div class="doc_code"><pre>
982 <b>virtual void</b> print(std::ostream &O, <b>const</b> Module *M) <b>const</b>;
985 <p>The <tt>print</tt> method must be implemented by "analyses" in order to print
986 a human readable version of the analysis results. This is useful for debugging
987 an analysis itself, as well as for other people to figure out how an analysis
988 works. Use the <tt>opt -analyze</tt> argument to invoke this method.</p>
990 <p>The <tt>llvm::OStream</tt> parameter specifies the stream to write the results on,
991 and the <tt>Module</tt> parameter gives a pointer to the top level module of the
992 program that has been analyzed. Note however that this pointer may be null in
993 certain circumstances (such as calling the <tt>Pass::dump()</tt> from a
994 debugger), so it should only be used to enhance debug output, it should not be
999 <!-- *********************************************************************** -->
1000 <div class="doc_section">
1001 <a name="interaction">Specifying interactions between passes</a>
1003 <!-- *********************************************************************** -->
1005 <div class="doc_text">
1007 <p>One of the main responsibilities of the <tt>PassManager</tt> is to make sure
1008 that passes interact with each other correctly. Because <tt>PassManager</tt>
1009 tries to <a href="#passmanager">optimize the execution of passes</a> it must
1010 know how the passes interact with each other and what dependencies exist between
1011 the various passes. To track this, each pass can declare the set of passes that
1012 are required to be executed before the current pass, and the passes which are
1013 invalidated by the current pass.</p>
1015 <p>Typically this functionality is used to require that analysis results are
1016 computed before your pass is run. Running arbitrary transformation passes can
1017 invalidate the computed analysis results, which is what the invalidation set
1018 specifies. If a pass does not implement the <tt><a
1019 href="#getAnalysisUsage">getAnalysisUsage</a></tt> method, it defaults to not
1020 having any prerequisite passes, and invalidating <b>all</b> other passes.</p>
1024 <!-- _______________________________________________________________________ -->
1025 <div class="doc_subsubsection">
1026 <a name="getAnalysisUsage">The <tt>getAnalysisUsage</tt> method</a>
1029 <div class="doc_text">
1031 <div class="doc_code"><pre>
1032 <b>virtual void</b> getAnalysisUsage(AnalysisUsage &Info) <b>const</b>;
1035 <p>By implementing the <tt>getAnalysisUsage</tt> method, the required and
1036 invalidated sets may be specified for your transformation. The implementation
1037 should fill in the <tt><a
1038 href="http://llvm.org/doxygen/classllvm_1_1AnalysisUsage.html">AnalysisUsage</a></tt>
1039 object with information about which passes are required and not invalidated. To
1040 do this, a pass may call any of the following methods on the AnalysisUsage
1044 <!-- _______________________________________________________________________ -->
1045 <div class="doc_subsubsection">
1046 <a name="AU::addRequired">The <tt>AnalysisUsage::addRequired<></tt> and <tt>AnalysisUsage::addRequiredTransitive<></tt> methods</a>
1049 <div class="doc_text">
1051 If your pass requires a previous pass to be executed (an analysis for example),
1052 it can use one of these methods to arrange for it to be run before your pass.
1053 LLVM has many different types of analyses and passes that can be required,
1054 spanning the range from <tt>DominatorSet</tt> to <tt>BreakCriticalEdges</tt>.
1055 Requiring <tt>BreakCriticalEdges</tt>, for example, guarantees that there will
1056 be no critical edges in the CFG when your pass has been run.
1060 Some analyses chain to other analyses to do their job. For example, an <a
1061 href="AliasAnalysis.html">AliasAnalysis</a> implementation is required to <a
1062 href="AliasAnalysis.html#chaining">chain</a> to other alias analysis passes. In
1063 cases where analyses chain, the <tt>addRequiredTransitive</tt> method should be
1064 used instead of the <tt>addRequired</tt> method. This informs the PassManager
1065 that the transitively required pass should be alive as long as the requiring
1070 <!-- _______________________________________________________________________ -->
1071 <div class="doc_subsubsection">
1072 <a name="AU::addPreserved">The <tt>AnalysisUsage::addPreserved<></tt> method</a>
1075 <div class="doc_text">
1077 One of the jobs of the PassManager is to optimize how and when analyses are run.
1078 In particular, it attempts to avoid recomputing data unless it needs to. For
1079 this reason, passes are allowed to declare that they preserve (i.e., they don't
1080 invalidate) an existing analysis if it's available. For example, a simple
1081 constant folding pass would not modify the CFG, so it can't possibly affect the
1082 results of dominator analysis. By default, all passes are assumed to invalidate
1087 The <tt>AnalysisUsage</tt> class provides several methods which are useful in
1088 certain circumstances that are related to <tt>addPreserved</tt>. In particular,
1089 the <tt>setPreservesAll</tt> method can be called to indicate that the pass does
1090 not modify the LLVM program at all (which is true for analyses), and the
1091 <tt>setPreservesCFG</tt> method can be used by transformations that change
1092 instructions in the program but do not modify the CFG or terminator instructions
1093 (note that this property is implicitly set for <a
1094 href="#BasicBlockPass">BasicBlockPass</a>'s).
1098 <tt>addPreserved</tt> is particularly useful for transformations like
1099 <tt>BreakCriticalEdges</tt>. This pass knows how to update a small set of loop
1100 and dominator related analyses if they exist, so it can preserve them, despite
1101 the fact that it hacks on the CFG.
1105 <!-- _______________________________________________________________________ -->
1106 <div class="doc_subsubsection">
1107 <a name="AU::examples">Example implementations of <tt>getAnalysisUsage</tt></a>
1110 <div class="doc_text">
1112 <div class="doc_code"><pre>
1113 <i>// This is an example implementation from an analysis, which does not modify
1114 // the program at all, yet has a prerequisite.</i>
1115 <b>void</b> <a href="http://llvm.org/doxygen/classllvm_1_1PostDominanceFrontier.html">PostDominanceFrontier</a>::getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
1116 AU.setPreservesAll();
1117 AU.addRequired<<a href="http://llvm.org/doxygen/classllvm_1_1PostDominatorTree.html">PostDominatorTree</a>>();
1123 <div class="doc_code"><pre>
1124 <i>// This example modifies the program, but does not modify the CFG</i>
1125 <b>void</b> <a href="http://llvm.org/doxygen/structLICM.html">LICM</a>::getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
1126 AU.setPreservesCFG();
1127 AU.addRequired<<a href="http://llvm.org/doxygen/classllvm_1_1LoopInfo.html">LoopInfo</a>>();
1133 <!-- _______________________________________________________________________ -->
1134 <div class="doc_subsubsection">
1135 <a name="getAnalysis">The <tt>getAnalysis<></tt> and
1136 <tt>getAnalysisIfAvailable<></tt> methods</a>
1139 <div class="doc_text">
1141 <p>The <tt>Pass::getAnalysis<></tt> method is automatically inherited by
1142 your class, providing you with access to the passes that you declared that you
1143 required with the <a href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a>
1144 method. It takes a single template argument that specifies which pass class you
1145 want, and returns a reference to that pass. For example:</p>
1147 <div class="doc_code"><pre>
1148 bool LICM::runOnFunction(Function &F) {
1149 LoopInfo &LI = getAnalysis<LoopInfo>();
1154 <p>This method call returns a reference to the pass desired. You may get a
1155 runtime assertion failure if you attempt to get an analysis that you did not
1156 declare as required in your <a
1157 href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> implementation. This
1158 method can be called by your <tt>run*</tt> method implementation, or by any
1159 other local method invoked by your <tt>run*</tt> method.
1161 A module level pass can use function level analysis info using this interface.
1164 <div class="doc_code"><pre>
1165 bool ModuleLevelPass::runOnModule(Module &M) {
1167 DominatorTree &DT = getAnalysis<DominatorTree>(Func);
1172 <p>In above example, runOnFunction for DominatorTree is called by pass manager
1173 before returning a reference to the desired pass.</p>
1176 If your pass is capable of updating analyses if they exist (e.g.,
1177 <tt>BreakCriticalEdges</tt>, as described above), you can use the
1178 <tt>getAnalysisIfAvailable</tt> method, which returns a pointer to the analysis
1179 if it is active. For example:</p>
1181 <div class="doc_code"><pre>
1183 if (DominatorSet *DS = getAnalysisIfAvailable<DominatorSet>()) {
1184 <i>// A DominatorSet is active. This code will update it.</i>
1191 <!-- *********************************************************************** -->
1192 <div class="doc_section">
1193 <a name="analysisgroup">Implementing Analysis Groups</a>
1195 <!-- *********************************************************************** -->
1197 <div class="doc_text">
1199 <p>Now that we understand the basics of how passes are defined, how they are
1200 used, and how they are required from other passes, it's time to get a little bit
1201 fancier. All of the pass relationships that we have seen so far are very
1202 simple: one pass depends on one other specific pass to be run before it can run.
1203 For many applications, this is great, for others, more flexibility is
1206 <p>In particular, some analyses are defined such that there is a single simple
1207 interface to the analysis results, but multiple ways of calculating them.
1208 Consider alias analysis for example. The most trivial alias analysis returns
1209 "may alias" for any alias query. The most sophisticated analysis a
1210 flow-sensitive, context-sensitive interprocedural analysis that can take a
1211 significant amount of time to execute (and obviously, there is a lot of room
1212 between these two extremes for other implementations). To cleanly support
1213 situations like this, the LLVM Pass Infrastructure supports the notion of
1214 Analysis Groups.</p>
1218 <!-- _______________________________________________________________________ -->
1219 <div class="doc_subsubsection">
1220 <a name="agconcepts">Analysis Group Concepts</a>
1223 <div class="doc_text">
1225 <p>An Analysis Group is a single simple interface that may be implemented by
1226 multiple different passes. Analysis Groups can be given human readable names
1227 just like passes, but unlike passes, they need not derive from the <tt>Pass</tt>
1228 class. An analysis group may have one or more implementations, one of which is
1229 the "default" implementation.</p>
1231 <p>Analysis groups are used by client passes just like other passes are: the
1232 <tt>AnalysisUsage::addRequired()</tt> and <tt>Pass::getAnalysis()</tt> methods.
1233 In order to resolve this requirement, the <a href="#passmanager">PassManager</a>
1234 scans the available passes to see if any implementations of the analysis group
1235 are available. If none is available, the default implementation is created for
1236 the pass to use. All standard rules for <A href="#interaction">interaction
1237 between passes</a> still apply.</p>
1239 <p>Although <a href="#registration">Pass Registration</a> is optional for normal
1240 passes, all analysis group implementations must be registered, and must use the
1241 <A href="#registerag"><tt>RegisterAnalysisGroup</tt></a> template to join the
1242 implementation pool. Also, a default implementation of the interface
1243 <b>must</b> be registered with <A
1244 href="#registerag"><tt>RegisterAnalysisGroup</tt></a>.</p>
1246 <p>As a concrete example of an Analysis Group in action, consider the <a
1247 href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>
1248 analysis group. The default implementation of the alias analysis interface (the
1250 href="http://llvm.org/doxygen/structBasicAliasAnalysis.html">basicaa</a></tt>
1251 pass) just does a few simple checks that don't require significant analysis to
1252 compute (such as: two different globals can never alias each other, etc).
1253 Passes that use the <tt><a
1254 href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a></tt>
1255 interface (for example the <tt><a
1256 href="http://llvm.org/doxygen/structGCSE.html">gcse</a></tt> pass), do
1257 not care which implementation of alias analysis is actually provided, they just
1258 use the designated interface.</p>
1260 <p>From the user's perspective, commands work just like normal. Issuing the
1261 command '<tt>opt -gcse ...</tt>' will cause the <tt>basicaa</tt> class to be
1262 instantiated and added to the pass sequence. Issuing the command '<tt>opt
1263 -somefancyaa -gcse ...</tt>' will cause the <tt>gcse</tt> pass to use the
1264 <tt>somefancyaa</tt> alias analysis (which doesn't actually exist, it's just a
1265 hypothetical example) instead.</p>
1269 <!-- _______________________________________________________________________ -->
1270 <div class="doc_subsubsection">
1271 <a name="registerag">Using <tt>RegisterAnalysisGroup</tt></a>
1274 <div class="doc_text">
1276 <p>The <tt>RegisterAnalysisGroup</tt> template is used to register the analysis
1277 group itself as well as add pass implementations to the analysis group. First,
1278 an analysis should be registered, with a human readable name provided for it.
1279 Unlike registration of passes, there is no command line argument to be specified
1280 for the Analysis Group Interface itself, because it is "abstract":</p>
1282 <div class="doc_code"><pre>
1283 <b>static</b> RegisterAnalysisGroup<<a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>> A("<i>Alias Analysis</i>");
1286 <p>Once the analysis is registered, passes can declare that they are valid
1287 implementations of the interface by using the following code:</p>
1289 <div class="doc_code"><pre>
1291 //<i> Analysis Group implementations <b>must</b> be registered normally...</i>
1292 RegisterPass<FancyAA>
1293 B("<i>somefancyaa</i>", "<i>A more complex alias analysis implementation</i>");
1295 //<i> Declare that we implement the AliasAnalysis interface</i>
1296 RegisterAnalysisGroup<<a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>> C(B);
1300 <p>This just shows a class <tt>FancyAA</tt> that is registered normally, then
1301 uses the <tt>RegisterAnalysisGroup</tt> template to "join" the <tt><a
1302 href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a></tt>
1303 analysis group. Every implementation of an analysis group should join using
1304 this template. A single pass may join multiple different analysis groups with
1307 <div class="doc_code"><pre>
1309 //<i> Analysis Group implementations <b>must</b> be registered normally...</i>
1310 RegisterPass<<a href="http://llvm.org/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a>>
1311 D("<i>basicaa</i>", "<i>Basic Alias Analysis (default AA impl)</i>");
1313 //<i> Declare that we implement the AliasAnalysis interface</i>
1314 RegisterAnalysisGroup<<a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>, <b>true</b>> E(D);
1318 <p>Here we show how the default implementation is specified (using the extra
1319 argument to the <tt>RegisterAnalysisGroup</tt> template). There must be exactly
1320 one default implementation available at all times for an Analysis Group to be
1321 used. Only default implementation can derive from <tt>ImmutablePass</tt>.
1322 Here we declare that the
1323 <tt><a href="http://llvm.org/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a></tt>
1324 pass is the default implementation for the interface.</p>
1328 <!-- *********************************************************************** -->
1329 <div class="doc_section">
1330 <a name="passStatistics">Pass Statistics</a>
1332 <!-- *********************************************************************** -->
1334 <div class="doc_text">
1336 href="http://llvm.org/doxygen/Statistic_8h-source.html"><tt>Statistic</tt></a>
1337 class is designed to be an easy way to expose various success
1338 metrics from passes. These statistics are printed at the end of a
1339 run, when the -stats command line option is enabled on the command
1340 line. See the <a href="http://llvm.org/docs/ProgrammersManual.html#Statistic">Statistics section</a> in the Programmer's Manual for details.
1345 <!-- *********************************************************************** -->
1346 <div class="doc_section">
1347 <a name="passmanager">What PassManager does</a>
1349 <!-- *********************************************************************** -->
1351 <div class="doc_text">
1354 href="http://llvm.org/doxygen/PassManager_8h-source.html"><tt>PassManager</tt></a>
1356 href="http://llvm.org/doxygen/classllvm_1_1PassManager.html">class</a>
1357 takes a list of passes, ensures their <a href="#interaction">prerequisites</a>
1358 are set up correctly, and then schedules passes to run efficiently. All of the
1359 LLVM tools that run passes use the <tt>PassManager</tt> for execution of these
1362 <p>The <tt>PassManager</tt> does two main things to try to reduce the execution
1363 time of a series of passes:</p>
1366 <li><b>Share analysis results</b> - The PassManager attempts to avoid
1367 recomputing analysis results as much as possible. This means keeping track of
1368 which analyses are available already, which analyses get invalidated, and which
1369 analyses are needed to be run for a pass. An important part of work is that the
1370 <tt>PassManager</tt> tracks the exact lifetime of all analysis results, allowing
1371 it to <a href="#releaseMemory">free memory</a> allocated to holding analysis
1372 results as soon as they are no longer needed.</li>
1374 <li><b>Pipeline the execution of passes on the program</b> - The
1375 <tt>PassManager</tt> attempts to get better cache and memory usage behavior out
1376 of a series of passes by pipelining the passes together. This means that, given
1377 a series of consequtive <a href="#FunctionPass"><tt>FunctionPass</tt></a>'s, it
1378 will execute all of the <a href="#FunctionPass"><tt>FunctionPass</tt></a>'s on
1379 the first function, then all of the <a
1380 href="#FunctionPass"><tt>FunctionPass</tt></a>es on the second function,
1381 etc... until the entire program has been run through the passes.
1383 <p>This improves the cache behavior of the compiler, because it is only touching
1384 the LLVM program representation for a single function at a time, instead of
1385 traversing the entire program. It reduces the memory consumption of compiler,
1386 because, for example, only one <a
1387 href="http://llvm.org/doxygen/classllvm_1_1DominatorSet.html"><tt>DominatorSet</tt></a>
1388 needs to be calculated at a time. This also makes it possible to implement
1390 href="#SMP">interesting enhancements</a> in the future.</p></li>
1394 <p>The effectiveness of the <tt>PassManager</tt> is influenced directly by how
1395 much information it has about the behaviors of the passes it is scheduling. For
1396 example, the "preserved" set is intentionally conservative in the face of an
1397 unimplemented <a href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> method.
1398 Not implementing when it should be implemented will have the effect of not
1399 allowing any analysis results to live across the execution of your pass.</p>
1401 <p>The <tt>PassManager</tt> class exposes a <tt>--debug-pass</tt> command line
1402 options that is useful for debugging pass execution, seeing how things work, and
1403 diagnosing when you should be preserving more analyses than you currently are
1404 (To get information about all of the variants of the <tt>--debug-pass</tt>
1405 option, just type '<tt>opt --help-hidden</tt>').</p>
1407 <p>By using the <tt>--debug-pass=Structure</tt> option, for example, we can see
1408 how our <a href="#basiccode">Hello World</a> pass interacts with other passes.
1409 Lets try it out with the <tt>gcse</tt> and <tt>licm</tt> passes:</p>
1411 <div class="doc_code"><pre>
1412 $ opt -load ../../../Debug/lib/Hello.so -gcse -licm --debug-pass=Structure < hello.bc > /dev/null
1414 Function Pass Manager
1415 Dominator Set Construction
1416 Immediate Dominators Construction
1417 Global Common Subexpression Elimination
1418 -- Immediate Dominators Construction
1419 -- Global Common Subexpression Elimination
1420 Natural Loop Construction
1421 Loop Invariant Code Motion
1422 -- Natural Loop Construction
1423 -- Loop Invariant Code Motion
1425 -- Dominator Set Construction
1431 <p>This output shows us when passes are constructed and when the analysis
1432 results are known to be dead (prefixed with '<tt>--</tt>'). Here we see that
1433 GCSE uses dominator and immediate dominator information to do its job. The LICM
1434 pass uses natural loop information, which uses dominator sets, but not immediate
1435 dominators. Because immediate dominators are no longer useful after the GCSE
1436 pass, it is immediately destroyed. The dominator sets are then reused to
1437 compute natural loop information, which is then used by the LICM pass.</p>
1439 <p>After the LICM pass, the module verifier runs (which is automatically added
1440 by the '<tt>opt</tt>' tool), which uses the dominator set to check that the
1441 resultant LLVM code is well formed. After it finishes, the dominator set
1442 information is destroyed, after being computed once, and shared by three
1445 <p>Lets see how this changes when we run the <a href="#basiccode">Hello
1446 World</a> pass in between the two passes:</p>
1448 <div class="doc_code"><pre>
1449 $ opt -load ../../../Debug/lib/Hello.so -gcse -hello -licm --debug-pass=Structure < hello.bc > /dev/null
1451 Function Pass Manager
1452 Dominator Set Construction
1453 Immediate Dominators Construction
1454 Global Common Subexpression Elimination
1455 <b>-- Dominator Set Construction</b>
1456 -- Immediate Dominators Construction
1457 -- Global Common Subexpression Elimination
1458 <b> Hello World Pass
1460 Dominator Set Construction</b>
1461 Natural Loop Construction
1462 Loop Invariant Code Motion
1463 -- Natural Loop Construction
1464 -- Loop Invariant Code Motion
1466 -- Dominator Set Construction
1475 <p>Here we see that the <a href="#basiccode">Hello World</a> pass has killed the
1476 Dominator Set pass, even though it doesn't modify the code at all! To fix this,
1477 we need to add the following <a
1478 href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> method to our pass:</p>
1480 <div class="doc_code"><pre>
1481 <i>// We don't modify the program, so we preserve all analyses</i>
1482 <b>virtual void</b> getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
1483 AU.setPreservesAll();
1487 <p>Now when we run our pass, we get this output:</p>
1489 <div class="doc_code"><pre>
1490 $ opt -load ../../../Debug/lib/Hello.so -gcse -hello -licm --debug-pass=Structure < hello.bc > /dev/null
1491 Pass Arguments: -gcse -hello -licm
1493 Function Pass Manager
1494 Dominator Set Construction
1495 Immediate Dominators Construction
1496 Global Common Subexpression Elimination
1497 -- Immediate Dominators Construction
1498 -- Global Common Subexpression Elimination
1501 Natural Loop Construction
1502 Loop Invariant Code Motion
1503 -- Loop Invariant Code Motion
1504 -- Natural Loop Construction
1506 -- Dominator Set Construction
1515 <p>Which shows that we don't accidentally invalidate dominator information
1516 anymore, and therefore do not have to compute it twice.</p>
1520 <!-- _______________________________________________________________________ -->
1521 <div class="doc_subsubsection">
1522 <a name="releaseMemory">The <tt>releaseMemory</tt> method</a>
1525 <div class="doc_text">
1527 <div class="doc_code"><pre>
1528 <b>virtual void</b> releaseMemory();
1531 <p>The <tt>PassManager</tt> automatically determines when to compute analysis
1532 results, and how long to keep them around for. Because the lifetime of the pass
1533 object itself is effectively the entire duration of the compilation process, we
1534 need some way to free analysis results when they are no longer useful. The
1535 <tt>releaseMemory</tt> virtual method is the way to do this.</p>
1537 <p>If you are writing an analysis or any other pass that retains a significant
1538 amount of state (for use by another pass which "requires" your pass and uses the
1539 <a href="#getAnalysis">getAnalysis</a> method) you should implement
1540 <tt>releaseMemory</tt> to, well, release the memory allocated to maintain this
1541 internal state. This method is called after the <tt>run*</tt> method for the
1542 class, before the next call of <tt>run*</tt> in your pass.</p>
1546 <!-- *********************************************************************** -->
1547 <div class="doc_section">
1548 <a name="registering">Registering dynamically loaded passes</a>
1550 <!-- *********************************************************************** -->
1552 <div class="doc_text">
1554 <p><i>Size matters</i> when constructing production quality tools using llvm,
1555 both for the purposes of distribution, and for regulating the resident code size
1556 when running on the target system. Therefore, it becomes desirable to
1557 selectively use some passes, while omitting others and maintain the flexibility
1558 to change configurations later on. You want to be able to do all this, and,
1559 provide feedback to the user. This is where pass registration comes into
1562 <p>The fundamental mechanisms for pass registration are the
1563 <tt>MachinePassRegistry</tt> class and subclasses of
1564 <tt>MachinePassRegistryNode</tt>.</p>
1566 <p>An instance of <tt>MachinePassRegistry</tt> is used to maintain a list of
1567 <tt>MachinePassRegistryNode</tt> objects. This instance maintains the list and
1568 communicates additions and deletions to the command line interface.</p>
1570 <p>An instance of <tt>MachinePassRegistryNode</tt> subclass is used to maintain
1571 information provided about a particular pass. This information includes the
1572 command line name, the command help string and the address of the function used
1573 to create an instance of the pass. A global static constructor of one of these
1574 instances <i>registers</i> with a corresponding <tt>MachinePassRegistry</tt>,
1575 the static destructor <i>unregisters</i>. Thus a pass that is statically linked
1576 in the tool will be registered at start up. A dynamically loaded pass will
1577 register on load and unregister at unload.</p>
1581 <!-- _______________________________________________________________________ -->
1582 <div class="doc_subsection">
1583 <a name="registering_existing">Using existing registries</a>
1586 <div class="doc_text">
1588 <p>There are predefined registries to track instruction scheduling
1589 (<tt>RegisterScheduler</tt>) and register allocation (<tt>RegisterRegAlloc</tt>)
1590 machine passes. Here we will describe how to <i>register</i> a register
1591 allocator machine pass.</p>
1593 <p>Implement your register allocator machine pass. In your register allocator
1594 .cpp file add the following include;</p>
1596 <div class="doc_code"><pre>
1597 #include "llvm/CodeGen/RegAllocRegistry.h"
1600 <p>Also in your register allocator .cpp file, define a creator function in the
1603 <div class="doc_code"><pre>
1604 FunctionPass *createMyRegisterAllocator() {
1605 return new MyRegisterAllocator();
1609 <p>Note that the signature of this function should match the type of
1610 <tt>RegisterRegAlloc::FunctionPassCtor</tt>. In the same file add the
1611 "installing" declaration, in the form;</p>
1613 <div class="doc_code"><pre>
1614 static RegisterRegAlloc myRegAlloc("myregalloc",
1615 " my register allocator help string",
1616 createMyRegisterAllocator);
1619 <p>Note the two spaces prior to the help string produces a tidy result on the
1622 <div class="doc_code"><pre>
1625 -regalloc - Register allocator to use: (default = linearscan)
1626 =linearscan - linear scan register allocator
1627 =local - local register allocator
1628 =simple - simple register allocator
1629 =myregalloc - my register allocator help string
1633 <p>And that's it. The user is now free to use <tt>-regalloc=myregalloc</tt> as
1634 an option. Registering instruction schedulers is similar except use the
1635 <tt>RegisterScheduler</tt> class. Note that the
1636 <tt>RegisterScheduler::FunctionPassCtor</tt> is significantly different from
1637 <tt>RegisterRegAlloc::FunctionPassCtor</tt>.</p>
1639 <p>To force the load/linking of your register allocator into the llc/lli tools,
1640 add your creator function's global declaration to "Passes.h" and add a "pseudo"
1641 call line to <tt>llvm/Codegen/LinkAllCodegenComponents.h</tt>.</p>
1646 <!-- _______________________________________________________________________ -->
1647 <div class="doc_subsection">
1648 <a name="registering_new">Creating new registries</a>
1651 <div class="doc_text">
1653 <p>The easiest way to get started is to clone one of the existing registries; we
1654 recommend <tt>llvm/CodeGen/RegAllocRegistry.h</tt>. The key things to modify
1655 are the class name and the <tt>FunctionPassCtor</tt> type.</p>
1657 <p>Then you need to declare the registry. Example: if your pass registry is
1658 <tt>RegisterMyPasses</tt> then define;</p>
1660 <div class="doc_code"><pre>
1661 MachinePassRegistry RegisterMyPasses::Registry;
1664 <p>And finally, declare the command line option for your passes. Example:</p>
1666 <div class="doc_code"><pre>
1667 cl::opt<RegisterMyPasses::FunctionPassCtor, false,
1668 RegisterPassParser<RegisterMyPasses> >
1670 cl::init(&createDefaultMyPass),
1671 cl::desc("my pass option help"));
1674 <p>Here the command option is "mypass", with createDefaultMyPass as the default
1679 <!-- *********************************************************************** -->
1680 <div class="doc_section">
1681 <a name="debughints">Using GDB with dynamically loaded passes</a>
1683 <!-- *********************************************************************** -->
1685 <div class="doc_text">
1687 <p>Unfortunately, using GDB with dynamically loaded passes is not as easy as it
1688 should be. First of all, you can't set a breakpoint in a shared object that has
1689 not been loaded yet, and second of all there are problems with inlined functions
1690 in shared objects. Here are some suggestions to debugging your pass with
1693 <p>For sake of discussion, I'm going to assume that you are debugging a
1694 transformation invoked by <tt>opt</tt>, although nothing described here depends
1699 <!-- _______________________________________________________________________ -->
1700 <div class="doc_subsubsection">
1701 <a name="breakpoint">Setting a breakpoint in your pass</a>
1704 <div class="doc_text">
1706 <p>First thing you do is start <tt>gdb</tt> on the <tt>opt</tt> process:</p>
1708 <div class="doc_code"><pre>
1711 Copyright 2000 Free Software Foundation, Inc.
1712 GDB is free software, covered by the GNU General Public License, and you are
1713 welcome to change it and/or distribute copies of it under certain conditions.
1714 Type "show copying" to see the conditions.
1715 There is absolutely no warranty for GDB. Type "show warranty" for details.
1716 This GDB was configured as "sparc-sun-solaris2.6"...
1720 <p>Note that <tt>opt</tt> has a lot of debugging information in it, so it takes
1721 time to load. Be patient. Since we cannot set a breakpoint in our pass yet
1722 (the shared object isn't loaded until runtime), we must execute the process, and
1723 have it stop before it invokes our pass, but after it has loaded the shared
1724 object. The most foolproof way of doing this is to set a breakpoint in
1725 <tt>PassManager::run</tt> and then run the process with the arguments you
1728 <div class="doc_code"><pre>
1729 (gdb) <b>break llvm::PassManager::run</b>
1730 Breakpoint 1 at 0x2413bc: file Pass.cpp, line 70.
1731 (gdb) <b>run test.bc -load $(LLVMTOP)/llvm/Debug/lib/[libname].so -[passoption]</b>
1732 Starting program: opt test.bc -load $(LLVMTOP)/llvm/Debug/lib/[libname].so -[passoption]
1733 Breakpoint 1, PassManager::run (this=0xffbef174, M=@0x70b298) at Pass.cpp:70
1734 70 bool PassManager::run(Module &M) { return PM->run(M); }
1738 <p>Once the <tt>opt</tt> stops in the <tt>PassManager::run</tt> method you are
1739 now free to set breakpoints in your pass so that you can trace through execution
1740 or do other standard debugging stuff.</p>
1744 <!-- _______________________________________________________________________ -->
1745 <div class="doc_subsubsection">
1746 <a name="debugmisc">Miscellaneous Problems</a>
1749 <div class="doc_text">
1751 <p>Once you have the basics down, there are a couple of problems that GDB has,
1752 some with solutions, some without.</p>
1755 <li>Inline functions have bogus stack information. In general, GDB does a
1756 pretty good job getting stack traces and stepping through inline functions.
1757 When a pass is dynamically loaded however, it somehow completely loses this
1758 capability. The only solution I know of is to de-inline a function (move it
1759 from the body of a class to a .cpp file).</li>
1761 <li>Restarting the program breaks breakpoints. After following the information
1762 above, you have succeeded in getting some breakpoints planted in your pass. Nex
1763 thing you know, you restart the program (i.e., you type '<tt>run</tt>' again),
1764 and you start getting errors about breakpoints being unsettable. The only way I
1765 have found to "fix" this problem is to <tt>delete</tt> the breakpoints that are
1766 already set in your pass, run the program, and re-set the breakpoints once
1767 execution stops in <tt>PassManager::run</tt>.</li>
1771 <p>Hopefully these tips will help with common case debugging situations. If
1772 you'd like to contribute some tips of your own, just contact <a
1773 href="mailto:sabre@nondot.org">Chris</a>.</p>
1777 <!-- *********************************************************************** -->
1778 <div class="doc_section">
1779 <a name="future">Future extensions planned</a>
1781 <!-- *********************************************************************** -->
1783 <div class="doc_text">
1785 <p>Although the LLVM Pass Infrastructure is very capable as it stands, and does
1786 some nifty stuff, there are things we'd like to add in the future. Here is
1787 where we are going:</p>
1791 <!-- _______________________________________________________________________ -->
1792 <div class="doc_subsubsection">
1793 <a name="SMP">Multithreaded LLVM</a>
1796 <div class="doc_text">
1798 <p>Multiple CPU machines are becoming more common and compilation can never be
1799 fast enough: obviously we should allow for a multithreaded compiler. Because of
1800 the semantics defined for passes above (specifically they cannot maintain state
1801 across invocations of their <tt>run*</tt> methods), a nice clean way to
1802 implement a multithreaded compiler would be for the <tt>PassManager</tt> class
1803 to create multiple instances of each pass object, and allow the separate
1804 instances to be hacking on different parts of the program at the same time.</p>
1806 <p>This implementation would prevent each of the passes from having to implement
1807 multithreaded constructs, requiring only the LLVM core to have locking in a few
1808 places (for global resources). Although this is a simple extension, we simply
1809 haven't had time (or multiprocessor machines, thus a reason) to implement this.
1810 Despite that, we have kept the LLVM passes SMP ready, and you should too.</p>
1814 <!-- *********************************************************************** -->
1817 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1818 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
1819 <a href="http://validator.w3.org/check/referer"><img
1820 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
1822 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
1823 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
1824 Last modified: $Date$